Beruflich Dokumente
Kultur Dokumente
George A. Anastassiou
Inequalities Based
on Sobolev Representations
123
George A. Anastassiou
Department of Mathematical Sciences
University of Memphis
Memphis, TN 38152
USA
ganastss@memphis.edu
ISSN 2191-8198
e-ISSN 2191-8201
ISBN 978-1-4614-0200-8
e-ISBN 978-1-4614-0201-5
DOI 10.1007/978-1-4614-0201-5
Springer New York Dordrecht Heidelberg London
Library of Congress Control Number: 2011929870
Mathematics Subject Classification (2010): 26D10, 26D15, 26D20
c George A. Anastassiou
All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York,
NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in
connection with any form of information storage and retrieval, electronic adaptation, computer software,
or by similar or dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are
not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject
to proprietary rights.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
Preface
This brief monograph is the first one to deal exclusively with very general tight
integral inequalities of ChebyshevGruss, Ostrowski types, and of the comparison
of integral means. These rely on the well-known Sobolev integral representations
of functions. The inequalities engage ordinary and weak partial derivatives of the
involved functions. Applications of these developments are illustrated. On the way
to prove the main results we derive important estimates for the averaged Taylor
polynomials and remainders of Sobolev integral representations. The exposed
results expand to all possible directions. We examine both the univariate and
multivariate cases.
For the convenience of the reader, each chapter of this book is written in a selfcontained style.
This treatise relies on the authors last year of related research work.
Advanced courses and seminars can be taught out of this brief book. All
necessary background and motivations are given in each chapter. A related list of
references is also given at the end of each chapter. These results first appeared
in my articles that are mentioned in the references. The results are expected to
find applications in many subareas of mathematical analysis, inequalities, partial
differential equations, information theory, etc. As such this monograph is suitable
for researchers, graduate students, seminars of the above subjects, and also for all
science libraries.
The preparation of this booklet took place during 20102011 in Memphis, TN,
USA.
I thank my family for their dedication and love to me, which was the strongest
support during the writing of this book.
Department of Mathematical Sciences
The University of Memphis, TN, USA
March 5, 2011
George A. Anastassiou
vii
Contents
35
35
36
50
64
65
ix
Chapter 1
1.1 Introduction
This chapter is greatly motivated by the following theorems:
Theorem 1.1 (Chebychev, 1882, [7]). Let f; gW a; b ! R absolutely continuous
functions. If f 0 ; g 0 2 L1 .a; b/, then
! Z
!
Z b
1 Z b
b
1
f .x/ g .x/ dx
f
.x/
dx
g
.x/
dx
b a a
.b a/2
a
a
1
.b a/2 f 0 1 g 0 1 :
12
(1.1)
! Z
!
Z b
1 Z b
b
1
f .x/ g .x/ dx
f
.x/
dx
g
.x/
dx
b a a
.b a/
a
a
1
.M m/ . / :
4
In 1938, A. Ostrowski [14] proved.
(1.2)
"
2 #
1 Z b
x aCb
1
2
C
f .t/ dt f .x/
.b a/ f 0 1 ;
2
b a a
4
.b a/
for any x 2 a; b. The constant
1
4
(1.3)
See also [13] for related works that inspired as well this chapter.
In this chapter using the univariate Sobolev-type representation formulae, see
Theorems 1.10, 1.14 and also Corollaries 1.11, 1.12, we first estimate their
remainders and then the involved averaged Taylor polynomials.
Based on these estimates we derive lots of very tight inequalities on R: of
ChebyshevGruss type, Ostrowski type, for comparison of integral means and
Csiszars f -divergence with applications. The results involve ordinary and weak
derivatives and they go to all possible directions using various norms. All of our
tools come from the excellent monograph by V. Burenkov, [6].
1.2 Background
Here we follow [6].
For a measurable nonempty set Rn , n 2 N we shall denote by Lloc
p ./
(1 p 1) the set of functions defined on such that for each compact
K ; f 2 Lp .K/.
Definition 1.4. Let Rn be an open set, 2 ZnC , 0 and f; g 2 Lloc
1 ./.
The function g is a weak derivative of the function f of order on (briefly
g D Dw f ) if 8 ' 2 C01 ./ (i.e., ' 2 C 1 ./ compactly supported in )
Z
Z
jj
fD
'
dx
D
.1/
g' dx:
(1.4)
X
D f
w
jjl
Lp ./
(1.5)
.loc/
./ WD
Definition 1.6. For l 2 N, we define the Sobolev-type local space W1l
ff W ! R W f 2 L1loc ./ and all f -distributional partials of orders l belong
to L1loc ./g D ff 2 Lloc
1 ./ W for each open set G compactly embedded into ,
f 2 W1l .G/g:
1.2
Background
We use Definitions 1.41.6 on R. Next we mention Sobolevs integral representation from [6].
Definition 1.7 ([6], p. 82). Let 1 < a < b < 1;
Z
! 2 L1 .a; b/ ,
! .x/ dx D 1:
(1.6)
Define
( Ry
a y x b;
a R! .u/ du;
.x; y/ WD
b
y ! .u/ du; a x < y b:
(1.7)
(1.9)
and if ! 0, then
8 x; y 2 a; b , j .x; y/j .b; b/ D 1:
If ! is symmetric with respect to
aCb
,
2
then 8 y 2 a; b we have
a C b ; y 1 :
2
2
Examples of !:
! .x/ D
also
! .x/ D
1
;
ba
8x 2 .a; b/ ;
1
.a;aC 1 / C .b 1 ;b / ;
m
m
2m
P0
kD1
D 0:
(1.10)
Moreover, assume that the derivative f .l1/ exists and is locally absolutely continuous on .a; b/. Then 8 x 2 .a; b/
f .x/ D
Z
l1
X
1 b .k/
f .y/ .x y/k ! .y/ dy
k a
kD0
1
.l 1/
(1.11)
and
Z
l1
X
1 .k/
f .x/ D
f .y/ .x y/k ! .y/ dy
k
kD0
1
C
.l 1/
bx
(1.12)
ax
(1.13)
f .x/ D
l1
X
.1/k h
kD0
1
C
.l 1/
k
Z
bx
ax
.x y/ ! .y/
i.k/
y
!
f .y/ dy
(1.14)
1.2
Background
In particular, here
! ./ D ::: D ! .l2/ ./ D ! ./ D ::: D ! .l2/ ./ D 0:
(1.15)
Corollary 1.12 ([6], p. 86). Assume that l; m 2 N, m < l. Then for the same f
and ! as in Corollary 1.11, 8 x 2 .a; b/
Z
f
.m/
.x/ D
lm1
X
kD0
!
i.kCm/
.1/kCm h
k
.x y/ ! .y/
f .y/ dy
y
k
1
C
.l m 1/
bx
ax
Remark 1.13 ([6], p. 86). The first summand in (1.14) can take the form:
8 R P
l1
s .s/
.x
y/
!
.y/
f .y/ dy;
s
<
sD0
s P
ls1 s C k
.1/
:
: where s WD s
kDs
k
(1.17)
Similarly, we have for the first summand of (1.16) the following form:
8 R P
l1
sm .s/
:
: where s;m WD .sm/ kDs
k
(1.18)
We need
Theorem 1.14 ([6], p. 91). Let l 2 N , 1 a < < < b 1, ! satisfy
condition
Z
! 2 L1 .R/ ; sup p! ; ;
! .x/ dx D 1
(1.19)
R
loc
and f 2 W1l
.a; b/ : Then for almost every x 2 .a; b/
f .x/ D
Z
l1
X
1 .k/
f .y/ .x y/k ! .y/ dy
k w
kD0
1
C
.l 1/
where ax , bx as in Theorem 1.10.
We denote fw.0/ WD f:
bx
ax
(1.20)
Remark 1.15 ([6], p. 92). By Theorem 1.14 it follows that if in Corollaries 1.11,
loc
1.12 f 2 W1l
.a; b/ then equalities (1.14) and (1.16) hold almost everywhere
on .a; b/, if we substitute f .l/ , f .m/ by the weak derivatives fw.l/ ; fw.m/ ; respectively.
Next we estimate the remainders of the above mentioned Sobolev representations.
We make
Remark 1.16. Denote by f
m 2 ZC . We estimate
.k/
1
Rm;l f .x/ WD
.l m 1/
.x y/lm1 .x; y/ f
.l/
.y/ dy;
(1.21)
1
.l 1/
.x y/l1 .x; y/ f
.l/
.y/ dy:
(1.22)
Thus we derive
Z
f .y/ dy
.l m 1/
.l/
. /lm1
k!kL1 .a;b/ f
L1 .;/
;
D
.l m 1/
jRm;l f .x/j
x 2 .; / :
We also have
jRm;l f .x/j
If f
.l/
k!kL1 .a;b/
.l m 1/
.l/
jx yjlm1 f .y/ dy DW I1 :
2 L1 .; /, then
I1
.l/
k!kL1 .a;b/ f
L1 .;/
.l m 1/
jx yj
lm1
dy :
But
Z
Z
jx yj
lm1
dy D
Z
.x y/
lm1
dy C
x
. x/lm C .x /lm
:
l m
.y x/lm1 dy
(1.23)
1.2
Background
Therefore if f
.l/
2 L1 .; /, then
jRm;l f .x/j
.l/
k!kL1 .a;b/ f
L1 .;/
.l m/
. x/lm C .x /lm ;
(1.24)
x 2 .; / :
Let now p; q > 1 W
1
p
D 1: If f
1
q
k!kL1 .a;b/
I1
.l m 1/
.l/
2 Lp .; /, then
! q1
jx yj
q.lm1/
dy
.l/
f
Lp .;/
However
Z
jx yj
q.lm1/
dy D
.x y/
q.lm1/
Hence if f
.l/
dy C
.y x/q.lm1/ dy
x
.x /q.lm1/C1 C . x/q.lm1/C1
:
q .l m 1/ C 1
2 Lp .; /, then
jRm;l f .x/j
.l/
k!kL1 .a;b/ f
Lp .;/
.l m 1/
. x/q.lm1/C1 C .x /q.lm1/C1
q .l m 1/ C 1
! q1
;
(1.25)
x 2 .; / :
If sup p! ; , then
k!kL1 .a;b/ D k!kL1 .;/ :
(1.26)
(1.27)
We make
Remark 1.17. Here we estimate from the Taylors averaged polynomial, see (1.12)
and (1.20), that part
Q
l1
Z
l1
X
1 .k/
f .x/ WD
f .y/ .x y/k ! .y/ dy;
k
(1.28)
kD1
l1
X
1 .k/
k
Q f .x/
f .y/ jx yj j! .y/j dy
k
kD1
!
l1
X
. /k
.k/
k!kL1 .R/ ;
f
L1 .;/
k
(1.29)
kD1
Q f .x/
l1
X
.k/
.k/
f
kD1
L1 .;/
k!kL1 .R/ Z
l1
X
. x/kC1 C .x /kC1
D
.k C 1/
kD1
jx yjk dy
.k/
f
L1 .;/
!
k!kL1 .R/ ;
(1.30)
x 2 .; /.
.k/
Let p; q > 1 W p1 C q1 D 1; f
2 Lp .; /, k D 1; :::; l 1; and again
k!kL1 .R/ < 1. Then
.k/
! q1
Z
l1
X f Lp .;/
l1
kq
Q f .x/
jx yj dy
k!kL1 .R/
k
kD1
.k/
0
1
! q1
f
l1
.kqC1/
.kqC1/
C .x /
Lp .;/ C
BX . x/
D@
A k!kL1 .R/ ;
kq C 1
k
kD1
(1.31)
x 2 .; / :
1.2
Background
.k/
2 L1 .; /, k D 1; :::; l 1; then
.k/
1
k
l1 f
.
/
X
l1
L1 .a;/
C
Q f .x/ B
@
A k!kL1 .R/ ;
k
0
(1.32)
kD1
x 2 .; /.
Suppose p; q > 1 W
then
1
C q1
p
.k/
D 1; f
2 Lp .; /, k D 1; :::; l 1I ! 2 Lq .; /,
!
l1
X
l1
. /k
.k/
Q f .x/
k!kLq .;/ ;
f
Lp .;/
k
(1.33)
kD1
x 2 .; /.
Suppose p; q; r > 1 W
! 2 Lq .; /, then
0
l1
Q f .x/ B
@
l1
X
1
p
1
q
D 1; f
1
r
.k/
f
Lp .;/
kD1
.k/
. x/
2 Lp .; /, k D 1; :::; l 1I
.krC1/
C .x /
.kr C 1/
.krC1/
k!kLq .;/ ;
! 1r
1
C
A
(1.34)
x 2 .; /:
We also make
R
Remark 1.18. Here l > 1, ! 2 C .l2/ .R/, sup p! ; , R ! .x/ dx D 1, and
the derivative ! .l2/ is absolutely continuous on a; b. Hence we have that
Ql1 f .x/ D
Z
l1
X
.1/k
kD1
h
.x y/k ! .y/
i.k/
y
f .y/ dy;
(1.35)
8 x 2 .; /.
And it holds
l1
X
l1
1
Q f .x/
k
kD1
8 x 2 .; /.
h
i.k/
(1.36)
10
Consequently, 8 x 2 .; /,
l1
Q f .x/
8
h
i.k/
Pl1 1
kf kL1 .;/ ;
kD1 k .x y/ ! .y/
y 1
if f 2 L1 .; /;
h
i.k/
P
l1
k
1
kf kL1 .;/ ;
kD1 k .x y/ ! .y/
<
y L .;/
1
if f 2 L1 .; /;
when
p;
q > 1 W p1 C q1 D 1, we have
h
i.k/
Pl1 1
k
kf kLp .;/ ;
kD1 k .x y/ ! .y/
y L .;/
:
if f 2 Lp .; /:
Let l; m 2 N, m < l, and f; ! as above, x 2 .; /.
We consider here
lm1
i.kCm/
X .1/kCm Z h
k
l1
.x y/ ! .y/
f .y/ dy:
Qm f .x/ WD
y
k
(1.37)
(1.38)
kD1
l1
When l D m C 1, then Qm
f .x/ WD 0:
Hence it holds
i.kCm/
X 1 Z h
l1
lm1
jf .y/j dy;
Q f .x/
.x y/k ! .y/
m
y
k
(1.39)
kD1
8 x 2 .; / :
Consequently, 8 x 2 .; /,
l1
Q f .x/
m
8
h
i.kCm/
Plm1 1
.x y/ ! .y/
kf kL1 .;/ ;
kD1
k
if f 2 L1 .; /;
h
i.kCm/
P
lm1
k
1
kf kL1 .;/ ;
kD1
k .x y/ ! .y/ y
<
L1 .;/
if f 2 L1 .; /;
when
p;
q > 1 W p1 C q1 D 1, we have
h
i.kCm/
Plm1 1
k
.x y/ ! .y/
kf kLp .;/ ;
kD1
k
Lq .;/
:
if f 2 Lp .; /:
(1.40)
1.3
Main Results
11
We also need
.k/
Remark 1.19. Here again f means either f .k/ or fw.k/ , k 2 N. We rewrite (1.12),
(1.14), and (1.20). For x 2 .; / we get
Z
f .x/ D
(1.41)
.m/
.x/ D .1/m
l1
f .y/ ! .m/ .y/ dy C Qm
f .x/ C Rm;l f .x/ :
(1.42)
.f; g/ WD
! .x/ f .x/ g .x/ dx
! .x/ f .x/ dx
! .x/ g .x/ dx
" Z
!
Z
l 1
l 1
!#
Z
Z
:
C
j! .x/j jg .x/j R0;l f .x/ dx C
j! .x/j jf .x/j R0;l g .x/ dx
(1.43)
f .x/ D
and
Z
g .x/ D
Hence
f .y/ ! .y/ dy
12
and
Z
g .y/ ! .y/ dy
! Z
! .x/ g .x/ dx
f .x/ ! .x/ dx
and
Z
! Z
! .x/ f .x/ dx
g .x/ ! .x/ dx
Consequently it holds
Z
! Z
! .x/ f .x/ dx
g .x/ ! .x/ dx
and
Z
! Z
! .x/ f .x/ dx
g .x/ ! .x/ dx
1.3
Main Results
13
! Z
! .x/ f .x/ dx
g .x/ ! .x/ dx
1
D
2
" Z
! .x/ g .x/ Q
l1
f .x/ dx C
! .x/ f .x/ Q
l1
g .x/ dx
!#
.k/
""
!
l1 f
X
. /k
k!k2L1 .R/
1;.;/
kgk1;.;/
2
k
kD1
!#
l1 g .k/
X
. /k
1;.;/
C kf k1;.;/
k
kD1
"
C
kgk1;.;/ f .l/
L1 .a;/
C kf k1;.;/ g .l/
L1 .;/
. /l1
.l 1/
##
:
(1.44)
"
(
!!
l1
f .k/
X
. /k
k!kL1 .R/
1;.;/
k!kL1 .R/
kgk1;.;/
2
k
kD1
14
.k/
!!)
l1
g
X
. /k
1;.;/
C kf k1;.;/
kD1
"
C k!k1;.;/
. /lC1
.l C 1/
kgk1;.;/ f .l/
1;.;/
##
C kf k1;.;/ g .l/
1;.;/
(1.45)
k!k2L1 .R/
("
kgkL1 .;/
l1
X
. /k
.k/
fw
L1 .;/
k
!!
kD1
C kf kL1 .;/
l1
X
. /k
.k/
gw
L1 .;/
k
!!#
kD1
"
C
kgkL1 .;/ fw.l/
L1 .;/
C kf kL1 .;/ gw.l/
L1 .;/
. /l
.l 1/
#)
:
(1.46)
k!k2L1 .R/
2
("
kgkL1 .;/
!
l1 f .k/
X
. /kC1
w
L1 .;/
kD1
1.3
Main Results
15
C kf kL1 .;/
"
C
!#
l1 g .k/
X
. /kC1
w
L1 .;/
k
kD1
kgkL1 .;/ fw.l/
L1 .;/
C kf kL1 .;/ gw.l/
L1 .;/
. /lC1
.l 1/
#)
: (1.47)
.f; g/ WD
! .x/ f .x/ g .x/ dx
! .x/ f .x/ dx
!
Z
! .x/ g .x/ dx
!
"(
1
l1
X
k!k2L1 .R/
. /kC1 p
.k/
kgkL1 .;/
fw
Lp .;/
2
k
kD1
!)
1
l1
X
. /kC1 p
.k/
C kf kL1 .;/
gw
Lp .;/
k
kD1
(
C
kgkL1 .;/ fw.l/
Lp .;/
C kf kL1 .;/ gw.l/
Lp .;/
. /lC1 p
.l 1/
)#
:
(1.48)
R
Theorem 1.27. Let l 2 N f1g; ! 2 C .l2/ .R/ ; sup p! ; , R ! .x/
dx D 1; and the derivative ! .l2/ is absolutely continuous on a; b RI .; /
loc
.l/
.a; b/, or f; g 2 C l .a; b/ : Here f denotes
.a; b/ : Here suppose f; g 2 W1l
either fw.l/ or f .l/ , and .f; g/ as in (1.43).
16
"
k!k1
2 kgkL1 .;/ kf kL1 .;/
.f; g/
2
!
h
l1
i.k/
X
1
k
sup .x y/ ! .y/
y 1
k x2.;/
kD1
.l/
C kgkL1 .;/ f
L1 .;/
Ckf kL1 .;/ g .l/
L1 .;/
#
. /l
k!k1 : (1.49)
.l 1/
.l/
Ckf kL1 .;/ g .l/
L1 .;/
. /lC1
k!k1
.l 1/
)#
:
(1.50)
(iii) Let p; q > 1 W
Then
1
p
k!k1
.f; g/
2
1
q
"(
.l/
; g .l/ 2 Lp .; / :
kgkL1 .;/ kf kLp .;/ C kf kL1 .;/ kgkLp .;/
!)
h
l1
i.k/
X
1
k
sup .x y/ ! .y/
y L .;/
k x2.;/
q
kD1
(
.l/
C
kgkL1 .;/ f
Lp .;/
C kf kL1 .;/ g .l/
Lp .;/
lC1 p1
. /
.l 1/
)#
k!k1
(1.51)
1.3
Main Results
17
l1
f .y/ ! .y/ dy Q f .x/
f .x/
.l/
k!kL1 .R/ f L1 .;/ . /l1
WD A1 :
(1.52)
.l 1/
If additionally we assume k!kL1 .R/ < 1, then 8 x 2 .; /, we obtain
f .y/ ! .y/ dy
f .x/
!
!
l1
X
. x/kC1 C .x /kC1
.k/
k!kL1 .R/
f
1
.k C 1/
kD1
k!kL1 .R/ f .l/ L1 .;/ . /l1
WD B1 .x/ :
C
(1.53)
.l 1/
Proof. By (1.41), (1.23), and (1.30).
Theorem 1.29. All as in Theorem 1.28. Assume f 2 C l .a; b/. Then 8 x 2
.; / ;
l1
f .y/ ! .y/ dy Q f .x/
f .x/
k!kL1 .R/ f .l/ 1 . x/l C .x /l
DW A2 .x/ :
(1.54)
l
If additionally we assume k!kL1 .R/ < 1; then 8 x 2 .; / ;
f .y/ ! .y/ dy
f .x/
!
!
l1
X
. x/kC1 C .x /kC1
.k/
k!kL1 .R/
f
1
.k C 1/
kD1
k!kL1 .R/ f .l/ 1 . x/l C .x /l
DW B2 .x/ :
C
l
(1.55)
18
loc
.a; b/ and rest as in
Theorem 1.30. Let all as in Theorem 1.28 or f 2 W1l
Theorem 1.28. Then 8 x 2 .; / (or almost every x 2 .; /, respectively),
we get
l1
E .f / .x/ WD f .x/
f .y/ ! .y/ dy Q f .x/
.l/
k!kL1 .a;b/ f
L1 .;/
. /l1
DW A3
.l 1/
(1.56)
f .y/ ! .y/ dy
.f / .x/ WD f .x/
!
l1
X
. /k
.k/
k!kL1 .R/
f
L1 .;/
k
kD1
.l/
k!kL1 .a;b/ f
L1 .;/
. /l1
.l 1/
DW B3 :
(1.57)
loc
Theorem 1.31. Let all as in Theorem 1.28 or f 2 W1l
.a; b/ and rest as in
.l/
E .f / .x/
.l/
k!kL1 .R/ f
L1 .;/
. x/l C .x /l
DW A4 .x/ :
(1.58)
1.3
Main Results
19
.k/
Additionally f
2 L1 .; /, k D 1; :::; l 1 and if k!kL1 .R/ < 1, then 8
x 2 .; / (or almost every x 2 .; /, respectively), we get
0
.f / .x/ @
l1
X
kD1
1
0
. x/kC1 C .x /kC1
.k/
@
A f
.k C 1/
1
A k!kL
1 .R/
L1 .;/
.l/
k!kL1 .R/ f
L1 .;/
. x/l C .x /l
DW B4 .x/ :
(1.59)
loc
.a; b/ and rest as in
Theorem 1.32. Let all as in Theorem 1.28 or f 2 W1l
.l/
E .f / .x/
.l/
k!kL1 .R/ f
Lp .;/
.l 1/
. x/q.l1/C1 C .x /q.l1/C1
q .l 1/ C 1
! q1
DW A5 .x/ : (1.60)
.k/
Additionally, if f
2 Lp .; /, k D 1; :::; l 1 and k!kL1 .R/ < 1, then 8
x 2 .; / (or almost every x 2 .; /, respectively), we get
.f / .x/
0
B
B
@
l1
X
kD1
1
1 q1
0
.k/
f
. x/.kqC1/ C .x /.kqC1/
Lp .;/ C
C
@
A
A
kq C 1
k
(1.61)
20
We further give
Theorem 1.33. Let all as in Theorem 1.31. Here assume ! 2 L1 .R/. Then 8
x 2 .; / (or almost every x 2 .; /, respectively), we get
.f / .x/ WD f .x/
f .y/ ! .y/ dy
0
1
.k/
l1 f
. /k
X
L1 .;/
B
C
@
A k!kL1 .R/ C A4 .x/
k
kD1
DW B6 .x/ :
(1.62)
loc
Theorem 1.35. Let all as in Theorem 1.28 or f 2 W1l
.a; b/ and rest as in The.k/
.k/
1
0
! 1r
f
l1
.krC1/
.krC1/
. x/
C .x /
Lp .;/
C
BX
@
A k!kLq .;/
k
.kr C 1/
kD1
DW .x/ :
(1.64)
.f / .x/ WD f .x/
f .y/ ! .y/ dy
1.3
Main Results
21
that:
(i) It holds
!
h
l1
i.k/
X
1
k
.x y/ ! .y/
.f / .x/
kf kL1 .;/
y
k
1
kD1
.l/
. /l1
k!kL1 .R/ f
L1 .;/
DW C1 .x/ : (1.65)
C
.l 1/
(ii) If f; f
.l/
2 L1 .; / ; then
!
h
l1
i.k/
X
1
k
.x y/ ! .y/
.f / .x/
kf kL1 .;/
y L .;/
k
1
kD1
.l/
. x/l C .x /l
k!kL1 .R/ f
L1 .;/
C
l
DW C2 .x/ :
(1.66)
(iii) Let p; q > 1 W
1
p
1
q
.l/
2 Lp .; / : Then
!
h
l1
i.k/
X
1
k
.f / .x/
.x y/ ! .y/
kf kLp .;/
y
k
Lq .;/
kD1
.l/
k!kL1 .R/ f
Lp .;/
C
.l 1/
! q1
. x/q.l1/C1 C .x /q.l1/C1
q .l 1/ C 1
DW C3 .x/ :
(1.67)
Z
.m/
m
.m/
l1
.x/ .1/
f .y/ ! .y/ dy Qm f .x/ ; (1.68)
E .f / .x/ WD f
22
and
Z
.m/
m
.m/
.x/ .1/
f .y/ ! .y/ dy
.f / .x/ WD f
(1.69)
that
(i) It holds
.l/
k!kL1 .R/ f
E .f / .x/
L1 .;/
. /lm1
DW E1 ;
.l m 1/
(1.70)
and
lm1
X
.f / .x/
kD1
!
h
i.kCm/
1
.x y/k ! .y/
kf kL1 .;/
y
k
1
CE1 DW G1 .x/ :
(ii) If f
.l/
(1.71)
2 L1 .; / ; then
.l/
k!kL1 .R/ f
L1 .;/
E .f / .x/
.l m/
. x/lm C .x /lm DW E2 .x/ ;
(1.72)
lm1
X
kD1
!
h
i.k/
1
.x y/k ! .y/
y L .;/
k
1
1
p
1
q
E .f / .x/
.l/
(1.73)
2 Lp .; / ; then
.l/
k!kL1 .R/ f
Lp .;/
.l m 1/
. x/q.lm1/C1 C .x /q.lm1/C1
q .l m 1/ C 1
DW E3 .x/ ;
! q1
(1.74)
1.3
Main Results
23
lm1
X
kD1
!
h
i.kCm/
1
.x y/k ! .y/
kf kLp .;/
y
k
Lq .;/
(1.75)
u .f / WD
f .x/ .x/ dx
f .y/ ! .y/ dy
Ql1 f .x/
1
1
A1 ;
.x/ dx
(1.77)
and
.x/ dx
f .y/ ! .y/ dy
!
l1
X
. /kC1
.k/
k!kL1 .R/
f
1
.k C 1/
kD1
k!kL1 .R/ f .l/ L1 .;/ . /l1
:
C
.l 1/
Z
1
f .x/
m .f / WD
1
(1.78)
24
Proof. By Remark 1.38, Theorem 1.28, and the fact that the functions . x/kC1
C .x /kC1 , k D 1; :::; l 1 are positive and convex with maximum
. /kC1 :
Theorem 1.40. All as in Theorem 1.29. Then
k!kL1 .R/ f .l/ 1 . /l
u .f /
;
l
(1.79)
and
!
l1
X
. /kC1
.k/
m .f /
k!kL1 .R/
f
1
.k C 1/
kD1
k!kL1 .R/ f .l/ 1 . /l
:
C
l
(1.80)
(1.81)
m .f / B3 :
(1.82)
and
Theorem 1.42. All as in Theorem 1.31. Then
u .f /
.l/
k!kL1 .R/ f
L1 .;/
. /l
;
(1.83)
and
!
l1
X
. /kC1
.k/
m .f /
k!kL1 .R/
f
L1 .;/
.k C 1/
kD1
.l/
k!kL1 .R/ f
L1 .;/
. /l
(1.84)
A5 .x/
1
.x/ dx;
(1.85)
1.3
Main Results
25
and
m .f /
B5 .x/
.x/ dx:
(1.86)
B6 .x/
.x/ dx;
(1.87)
.x/ dx;
(1.88)
B7 .x/
1
f .x/
1
Z
.x/ dx
f .y/ ! .y/ dy
.R0;l f .x//
1
.x/ dx
.x/
.x/ dx;
(1.89)
l1
X
1
m .f / @
sup
k x2 1 ;
kD1
1
h
i.k/
.x y/k ! .y/ A kf k .;/
L1
y
1
.l/
k!kL1 .R/ f
L1 .;/
.l 1/
. /l1
:
(1.90)
26
(ii) If f; f
.l/
2 L1 .; / ; then
Z
m .f /
C2 .x/
.x/ dx:
(1.91)
1
p
1
q
D 1I assume further f; f
Z
.l/
2 Lp .; / ; then
m .f /
C3 .x/
.x/ dx:
(1.92)
um .f / WD
f
.x/
1
Z
1
1
.x/ dx .1/m
l1
Qm
f .x/
.x/ dx ;
(1.93)
and
Z
1 .m/
m .f / WD
f
.x/
1
.x/ dx .1/
f .y/ !
.m/
.y/ dy :
(1.94)
(i) It holds
um .f / E1 ;
(1.95)
lm1
X
m .f / @
kD1
1
sup
k x21 ;
1
h
i.kCm/
.x y/k ! .y/
A kf k .;/ C E1 ;
L1
y
1
(1.96)
(ii) If f
.l/
2 L1 .; / ; then
um .f /
.l/
k!kL1 .R/ f
.l m/
L1 .;/
. /lm ;
(1.97)
1.3
Main Results
27
h
i.kCm/
.x y/k ! .y/
y
lm1
X
m .f / @
1
sup
k x21 ;
kD1
1
.l/
k!kL1 .R/ f
1
q
A kf kL
L1 .;/
L1 .;/
D 1; assume further f
Z
1 .;/
. /lm
.l m/
1
p
.l/
(1.98)
2 Lp .; /, then
um .f /
E3 .x/
.x/ dx;
(1.99)
G3 .x/
.x/ dx;
(1.100)
p
, a.e. on X and 1 :
q
The quantity
f . 1 ; 2 / D
q .x/ f
X
p .x/
d .x/ ;
q .x/
(1.101)
was introduced by I. Csiszar in 1967, see [9], and is called f -divergence of the
probability measures 1 and 2 . By Lemma 1.1 of [9], the integral (1.101) is welldefined and
f . 1 ; 2 / 0 with equality only when 1 D 2 : Furthermore
28
f . 2 ; 1 / D
f . 1 ; 2 /:
(1.103)
In information theory and statistics many other divergences are used which are
special cases of the above general Csiszar f -divergence, e.g., the Hellinger distance
DH , -distance D , Bhattacharyya distance DB , Harmonic distance DHa , Jeffreys
distance DJ , triangular discrimination D , for all these see, e.g., [5, 10]. The problem of finding and estimating the proper distance (or difference or discrimination)
of two probability distributions is one of the major ones in Probability Theory.
Here we provide a general probabilistic representation formula for
f . 1 ; 2 /.
Then we present tight estimates for the remainder involving a variety of norms of
the engaged functions. Also are implied some direct general approximations for the
Csiszars f -divergence. We give some applications.
We make
Remark 1.50. Here 0 < a < p.x/
q.x/ < b < C1, a e. on X and 1 .
.l1/
exists and is absolutely continuous on a; b, l 2 N.
Also suppose that f
Furthermore f is convex from .0; C1/
into R, strictly convex at 1 with f .1/ D 0.
R
Let ! 2 L1 .R/, sup p! ; , R ! .x/ dx D 1.
Then 8 x 2 .; / we get by Theorem 1.10, as in (1.41), that
Z
f .x/ D
Therefore
f
p .x/
q .x/
a.e. on X:
f .y/ ! .y/ dy C Q
l1
p .x/
q .x/
C R0;l f
p .x/
;
q .x/
1.3
Main Results
29
Hence
q .x/ f
p .x/
q .x/
Z
D q .x/
f .y/ ! .y/ dy
C q .x/ Q
l1
p .x/
q .x/
C q .x/ R0;l f
p .x/
;
q .x/
a.e. on X .
Therefore we get the representation of f -divergence of 1 and 2 ,
Z
p .x/
f . 1 ; 2 / D
d .x/ D
q .x/ f
f .y/ ! .y/ dy
q .x/
X
Z
p .x/
d .x/
C
q .x/ Ql1 f
q .x/
X
Z
p .x/
C
d .x/ :
(1.104)
q .x/ R0;l f
q .x/
X
Call
Q
WD
q .x/ Ql1 f
X
and
R
WD
q .x/ R0;l f
X
p .x/
d .x/ ;
q .x/
(1.105)
p .x/
d .x/ :
q .x/
(1.106)
We estimate Q
and R
.
If k!kL1 .R/ < 1, we get by (1.29) that
!
l1
X
. /k
.k/
jQ
j
k!kL1 .R/ :
f
L1 .;/
k
(1.107)
kD1
Z
X
l1
X
B
q .x/ @
p.x/
q.x/
L1 .;/
p.x/
q.x/
kC1
.k C 1/
kD1
f .k/
kC1
C
C
A d .x/A k!kL1 .R/ :
(1.108)
30
1
p
1
q
Z
X
0
l1
X
B
B
q .x/ B
@
@
p.x/
q.x/
.kqC1/
p.x/
q.x/
.kqC1/ 1 q1
kq C 1
kD1
C
A
1
1
.k/
f
C
Lp .;/ C
C d .x/ C k!k .R/ :
L1
A
A
k
(1.109)
jQ
j
kD1
If p; q > 1 W
1
p
1
q
k!kL1 .R/ :
(1.110)
!
l1
X
. /k
.k/
jQ
j
k!kLq .;/ :
f
Lp .;/
k
(1.111)
kD1
Assume p; q; r > 1 W
0
B
jQ
j @
1
p
1
q
1
r
l1 .k/
Lp .;/
BX f
q .x/ @
k
X
kD1
0
B
@
p.x/
q.x/
.krC1/
p.x/
q.x/
kr C 1
1
.krC1/ 1 1r 1
C
C C
C
A C
A d .x/A
k!kLq .;/ :
(1.112)
We make
Remark 1.51 (continuation
of Remark 1.50). Here l > 1, ! 2 C .l2/ .R/,
R
sup p! ; , R ! .x/ dx D 1, and ! .l2/ is absolutely continuous on a; b :
Then (by (1.35))
0
q .x/ @
Q
D
X
Z
l1
X
.1/k
kD1
0"
1
#.k/ 1
k
@ p .x/ y ! .y/
A f .y/ dy A d .x/ :
q .x/
y
(1.113)
1.3
Main Results
31
X
R
1
@
@
q
.x/
X
k
0kD1
l1
X
R
< @ q .x/ @
1
X
1
1
.k/
k
p.x/
A
d .x/A kf kL1 .;/ ,
q.x/ y ! .y/
y 1
1
1
.k/
k
p.x/
A d .x/A kf kL .;/ ;
q.x/ y ! .y/
1
y
kD1
L1 .;/
1
1
when
p;
q
>
1
W
C
D
1,
we
have
p
0
0
1
1
q
.k/
l1
k
X
R
p.x/
1
A d .x/A kf kL .;/ :
@ q .x/ @
y ! .y/
k
p
: X
q.x/
y
Lq .;/
kD1
(1.114)
We also make
Remark 1.52 (another continuation of Remark 1.50). Here we estimate the remainder R
of (1.104). By (1.23) and (1.106), we obtain
jR
j
k!kL1 .a;b/ f .l/ L1 .;/ . /l1
.l 1/
jR
j
1
p
1
q
(1.116)
k!kL1 .a;b/ f .l/ Lp .;/
q .x/
p .x/
C
q .x/
.q.l1/C1/
p .x/ .q.l1/C1/
q .x/
.q .l 1/ C 1/ q .l 1/
(1.115)
! q1
1
d .x/A :
(1.117)
f . 1 ; 2 /
f .y/ ! .y/ dy D Q
C R
;
(1.118)
32
and
T WD
f . 1 ; 2 /
f .y/ ! .y/ dy jQ
j C jR
j :
(1.119)
1.4 Applications
Example 1.53. Let V WD fx 2 R W jx x0 j < g, x0 2 R, and
' .x/ WD
8
<
:
1
e
0,
.xx0 /2
1
; if jx x0 j < ;
if jx x0 j :
2
(1.120)
R
Call c WD R ' .x/ dx > 0, then .x/ WD 1c ' .x/ 2 C01 .R/ (space of
continuously infinitely
R 1 many times differentiable functions of compact support) with
sup p D V and 1 .x/ dx D 1 and max jj cons tan t 1 : We call a
cut-off function.
1
, etc.,
One for this chapters results by choosing ! .x/ D .x/ or ! .x/ D 2
can give lots of applications. Due to lack of space we avoid it.
Instead, selectively, we give some special cases inequalities. We start with
ChebyshevGruss-type inequalities.
. /2
k!kL1 .R/ k!k1;.;/
2
0
kgk1;.;/ f 1;.;/ C kf k1;.;/ g 0 1;.;/ :
(1.121)
If f D g, then
!2
Z
Z
! .x/ f 2 .x/ dx
! .x/ f .x/ dx
k!kL1 .R/ k!k1;.;/ . /2 kf k1;.;/ f 0 1;.;/ :
(1.122)
1.4
Applications
33
loc
Corollary 1.55 (to Theorem 1.23). Let f 2 W11
.a; b/ I a; b 2 RI .; /
1
for x 2 ; , and zero elsewhere. Then
.a; b/ ; ! .x/ WD
Z
1
1
f 2 .x/ dx
.
/2
!2
kf kL1 .;/ fw.1/ L1 .;/
:
f .x/ dx
. /
(1.123)
Z
0
L1 .;/
(1.124)
f .x/
.x/ dx
0
L1 .;/
(1.125)
Proof. By (1.124).
We finish with an application of f -divergence.
Remark 1.58. All here as in Background 1.49 and Remark 1.50. Case of l D 1. By
(1.104) we get
Z
f . 1 ; 2 / D
Z
f .y/ ! .y/ dy C
q .x/ R0;1 f
X
That is here
Z
R
D
q .x/ R0;1 f
X
p .x/
d .x/ :
q .x/
p .x/
d .x/ :
q .x/
(1.126)
(1.127)
(1.128)
(1.129)
34
1
p
1
q
1
jR
j k!kL1 .a;b/ f 0 Lp .;/ . / q :
(1.130)
K WD
f . 1 ; 2 /
f .y/ ! .y/ dy D R
;
(1.131)
(l D 1 case).
So the estimates (1.128)(1.130) are also estimates for K:
References
1. G.A. Anastassiou, Quantitative Approximations, Chapman & Hall/CRC, Boca Raton, New
York, 2000.
2. G.A. Anastassiou, Probabilistic Inequalities, World Scientific, Singapore, New Jersey, 2010.
3. G.A. Anastassiou, Advanced Inequalities, World Scientific, Singapore, New Jersey, 2011.
4. G.A. Anastassiou, Univariate Inequalities based on Sobolev Representations, Studia
Mathematica-Babes Bolyai, accepted 2011.
5. N.S. Barnett, P. Cerone, S.S. Dragomir, A. Sofo, Approximating Csiszars f -divergence by the
use of Taylors formula with integral remainder, (paper #10, pp. 16), Inequalities for Csiszars
f -Divergence in Information Theory, S.S. Dragomir (ed.), Victoria University, Melbourne,
Australia, 2000. On line: http://rgmia.vu.edu.au
6. V. Burenkov, Sobolev spaces and domains, B.G. Teubner, Stuttgart, Leipzig, 1998.
7. P.L. Chebyshev, Sur les expressions approximatives des integrales definies par les autres prises
entre les memes limites, Proc. Math. Soc. Charkov, 2(1882), 93-98.
8. I. Csiszar, Eine Informationstheoretische Ungleichung und ihre Anwendung auf den Beweis
der Ergodizitat von Markoffschen Ketten, Magyar Tud. Akad. Mat. Kutato Int. Kozl. 8 (1963),
85-108.
9. I. Csiszar, Information-type measures of difference of probability distributions and indirect
observations, Studia Math. Hungarica 2 (1967), 299-318.
10. S.S. Dragomir (ed.), Inequalities for Csiszar f -Divergence in Information Theory, Victoria
University, Melbourne, Australia, 2000. On-line: http://rgmia.vu.edu.au
h
Rb
1
Chapter 2
2.1 Introduction
This chapter is greatly motivated by the following theorems:
Theorem A (Chebychev, 1882, [7]). Let f; g W a; b ! R absolutely continuous
functions. If f 0 ; g 0 2 L1 .a; b/, then
! Z
!
Z b
1 Z b
b
1
f .x/ g .x/ dx
f
.x/
dx
g
.x/
dx
2
b a a
.b a/
a
a
1
.b a/2 f 0 1 g 0 1 :
12
! Z
!
Z b
1 Z b
b
1
f .x/ g .x/ dx
f
.x/
dx
g
.x/
dx
b a a
.b a/2
a
a
1
.M m/ . / :
4
35
36
"
2 #
1 Z b
x aCb
1
2
.b a/ f 0 1 ;
C
f .t/ dt f .x/
2
b a a
4
.b a/
for any x 2 a; b. The constant
1
4
See also [13] for related works that inspired as well this chapter.
In this chapter using the Sobolev-type representation formulae, see Theorems
2.6, 2.8, 2.11 and 2.23, also Corollaries 2.12 and 2.13, we estimate first their
remainders and then the involved averaged Taylor polynomials.
Based on these estimates we establish lots of very tight inequalities on Rn ,
n 2 N, of ChebyshevGruss type, Ostrowski type and of Comparison of integral
means with applications. The results involve ordinary and weak partial derivatives
and they go to all possible directions using various norms. All of our machinery
comes from the excellent monograph by V. Burenkov, [6].
2.2 Background
Here we follow [6].
For a measurable nonempty set Rn , n 2 N we shall denote by Lloc
p ./
(1 p 1) - the set of functions defined on such that for each compact
K , f 2 Lp .K/.
Definition 2.1. Let Rn be an open set, 2 ZnC , 0 and f; g 2 Lloc
1 ./.
The function g is a weak derivative of the function f of order on (briefly
g D Dw f ) if 8 ' 2 C01 ./ (i.e., ' 2 C 1 ./ compactly supported in )
Z
Z
fD 'dx D .1/jj
g' dx:
(2.1)
jjl
.loc/
Definition 2.3. For l 2 N, we define the Sobolev-type local space W1l
./ WD
ff W ! R W f 2 L1loc ./ and all f -distributional partials of orders l belong
to L1loc ./g D ff 2 Lloc
1 ./ W for each open set G compactly embedded into ,
f 2 W1l .G/g:
2.2
Background
37
X .D f / .x0 /
X
.x x0 / C l
jj<l
.x x0 /
jjDl
.1 t/l1 .D f / .x0 C t .x x0 // dt
(2.3)
(here we mean x0 Ct .x xP
0 / D .x01 C t .x1 x01 / ; : : : ; x0n C t .xn x0n //, D
. 1 ; : : : ; n / 2 ZnC , jj D niD1 i , D 1 : : : n , .x x0 / D .x1 x01 /1 : : :
qP
n
2
.xn x0n / n ). Here jj stands for the Euclidean norm: jxj D
i D1 xi , x WD
.x1 ; : : : ; xn / :
Next we mention the Sobolev representations.
Theorem 2.6. Let Rn be a domain star-shaped with respect to the open
n
ball
R B D B .x0 ; r/ such that B l, ! 2 L1 .R /, the support supp ! B,
!
.x/
dx
D
1,
l
2
N
and
f
2
C
./.
Then
for
every
x2
Rn
f .x/ D
X 1 Z
X 1 Z
.D f / .y/ .x y/ ! .y/ dy C l
.x y/ ! .y/
B
B
jj<l
jjDl
Z
(2.4)
Proof. We write (2.3) for x; x0 D y, multiply it both sides by ! .y/ and integrate
on B with respect to y.
Call kD f kmax
1;l;B WD max kD f k1;B , where kk1;B is the supremum norm
jjDl
on B, d WD diameter of B:
38
(2.5)
Z 1
Z
X 1
l
.x y/ ! .y/
.1 t/l1 .D f / .y C t .x y// dt dy
0
jjDl B
l
Z 1
X 1 Z
.1t/l1 j.D f / .yCt .xy//j dt dy
j.xy/ j j! .y/j
B
0
jjDl
0
kD f
kmax
1;l;B
1
X 1 Z
@
j.x y/ j j! .y/j dy A
B
jjDl
kD f kmax
1;l;B
1
X 1 Z
dl @
j! .y/j dy A
B
jjDl
0
1
max
l
X
f
d
k!kL1 .B/
kD k1;l;B
l A
@
D
l
jjDl
kD f
kmax
1;l;B
.d n/ k!kL1 .B/
:
l
l
From [6], p. 104, we mention
Theorem 2.8. Let Rn be a domain star-shaped with respect to the open ball
B D B .x0 ; r/ such that B ,
Z
n
! 2 L1 .R / ; supp ! B,
! .x/ dx D 1;
(2.6)
Rn
XZ
jjDl
.D f / .y/
Vx
jx yjnl
w .x; y/ dy;
(2.7)
2.2
Background
39
where for x; y 2 Rn , x y;
jj .x y/
w .x; y/ ;
jx yjjj
(2.8)
yx
! xC
n1 d ;
jy xj
jxyj
(2.9)
w .x; y/ WD
and
Z
w .x; y/ WD
l1
l
.x y/ ! .y/
.1 t/ .D f / .y C t .x y// dt dy
B
0
jjDl
XZ
jjDl
.D f / .y/
Vx
jx yjnl
w .x; y/ dy:
(2.10)
X Z
:
w
.x;
y/
dy
nl
l
jjDl Vx jx yj
(2.11)
(2.12)
(2.13)
Notice kw .x; y/kC .Rn Rn / k!kL1 .Rn / d n and kw .x; y/kC .Rn Rn /
k!kL1 .Rn / nd n , if D B: Hence, if ! is bounded, then for bounded the
functions w and w are bounded on Rn Rn . Also by [6], if is unbounded, then
w, w are bounded on K Rn for each compact K:
If ! 2 C 1 .Rn /, then w .x; y/, w .x; y/ have continuous derivatives of all
orders 8 x; y 2 Rn W x y and at the points .x; x/, where x B they are
discontinuous, see [6].
40
loc
./. Then for almost every x 2
l 2 N and f 2 W1l
X 1 Z
Dw f .y/ .x y/ ! .y/ dy
f .x/ D
B
jj<l
Dw f .y/
XZ
Vx
jjDl
jx yjnl
w .x; y/ dy:
(2.15)
loc
Then 8 f 2 C l ./ for every x 2 and 8 f 2 W1l
./ for almost every
x2
1
0
Z
X .1/jj
@
Dy .x y/ ! .y/A f .y/ dy
f .x/ D
B
jj<l
XZ
jjDl
.D f / .y/
jx yjnl
Vx
w .x; y/ dy
(2.17)
loc
with Dw f replacing D f in the case of f 2 W1l
./ :
Next means i i ; i D 1; :::; n and 2 ZC :
Corollary 2.13 ([6]). Under the assumptions of Corollary 2.12, let 2 ZnC and
loc
./ for
0 < jj < l. Then 8 f 2 C l ./ for every x 2 and 8 f 2 W1l
almost every x 2
0
1
Z
X .1/jjCjj
@
DyC .x y/ ! .y/A f .y/ dy
D f .x/ D
B
jj<ljj
X
jjDl;
.D f / .y/
Vx
jx yjnlCjj
w .x; y/ dy
loc
./ :
with Dw f replacing D f and Dw f replacing D f if f 2 W1l
(2.18)
2.2
Background
41
X Z
D f .y/
w .x; y/ dy
R2 WD
nl
jjDl Vx jx yj
XZ
jjDl
Vx
XZ
@
jjDl
Vx
(2.19)
if D < 1:
Next we assume x 2 B, l n, then we retake D B, i.e., d D D, etc., and
thus
X Z
R2
D f .y/ dy k!kL1 .Rn / nd l
jjDl
X
D@
D f
L1 .B/
jjDl
A k!kL
1 .R
n/
nd l :
Z D f .y/
X
X
A k!kL .Rn / nd l :
w .x; y/ dy @
D f
1
nl
.B/
L
1
yj
jx
B
jjDl
jjDl
(2.20)
< 1 for all W jj D l (which
Again we assume x 2 B, l n and D f
L1 .B/
X
R2 @
D f
jjDl
L1 .B/
1 .R
n/
nd l DW .
/ :
We know that
n
Vol .B/ D
2
dn
2
n
rn D n
n;
2 C1
2 C1 2
where
is the gamma function.
42
Therefore
0
X
.
/ D @
D f
jjDl
L1 .B/
A k!kL
.Rn /
So we have proved if l n, x 2 B, and D f
n 2
d lCn :
n
2
n2 C 1
L1 .B/
Z D f .y/
X
w .x; y/ dy
nl
jjDl B jx yj
0
1
X
A k!kL .Rn /
@
D f
1
jjDl
L1 .B/
n 2
d lCn :
n
2
n2 C 1
(2.21)
m pn
jx zjmn jf .z/j dz cp d
kf kLp . / ;
(2.22)
Remark 2.16 (continuing from Remark 2.14). We assume now that D f 2 Lp .B/,
jj D l, 1 < p < 1, l > pn . Then by (2.22),
Z
B
n
jx yjln D f .y/ dy cp d l p D f
; x 2 B:
(2.23)
Z D f .y/
w
.x;
y/
dy
nl
jjDl B jx yj
0
1
X
n
A k!kL .Rn / ncp d l p Cn ; x 2 B:
@
D f
1
(2.24)
Lp .B/
Consequently, we derive
jjDl
Lp .B/
2.2
Background
43
Also we make
Remark 2.17. Again here d D diamB, assume ! 2 C01 ./, supp ! B, i.e.,
R
Z
D f .y/
w
.x; y/ dy
nlCjj
jjDl; B jx yj
0
1
X
A k!k1 nd ljj :
@
D f
Also for l n C jj, D f
(2.25)
L1 .B/
jjDl;
L1 .B/
< 1, jj D l, , x 2 B, we get
X Z
D f .y/
w
.x;
y/
dy
nlCjj
jjDl; B jx yj
0
1
n
X
n 2
A k!k1
n
d ljjCn :
@
D f
L1 .B/
2n
2 C 1
(2.26)
jjDl;
Z
D f .y/
X
w
.x;
y/
dy
nlCjj
jjDl; B jx yj
0
1
X
n
A k!k1 ncp d ljj p Cn :
@
D f
jjDl;
Lp .B/
n
p
C jj,
(2.27)
We make
X
1jjl1
Z
44
X 1 Z
l1
Q f .x/ @
D f .y/ j.x y/ j dy A k!kL1 .Rn /
B
0
1jjl1
@
1jjl1
1
Z
jj
D f .y/ jx yj dy A k!kL1 .Rn /
B
DW .
/ :
We notice that
(2.29)
/ @
d jj
1jjl1
8
<
jj
1jjl1
!1
Z
D f
L1 .B/
!9
=
;
k!kL1 .Rn / :
Q f .x/
k!kL1 .Rn / :
D f
:
L1 .B/ ;
(2.30)
1jjl1
Also, when D f
L1 .B/
0
@
jj
1jjl1
!1
Z
D f .y/ dy A k!kL1 .Rn /
@
1jjl1
0
D@
1jjl1
!1
d jj
A Vol .B/ k!kL .Rn /
D f
1
L1 .B/
!1
n
d jj
dn
2
A
n:
k!kL1 .Rn / n
D f
L1 .B/
2 C1 2
So we proved, when D f
L1 .B/
l1
Q f .x/ @
X
1jjl1
!1
n
k!kL1 .Rn / 2
d .nCjj/
A
n
: (2.31)
D f
L1 .B/
2n
2 C 1
2.2
Background
45
We need
Lemma 2.19. Let be a region of Rn of finite diameter d > 0 and f 2
Lp . /, 1 < p; q < 1 W p1 C q1 D 1 and m 2 N, then
Z
jx zj jf .z/j dz cq;m;n d
m
mC nq
kf kLp . / ; 8 x 2 :
(2.32)
Z
jx zjm jf .z/j dz
q1
jx zjmq dz kf kLp . /
d
Cn
! q1
kf kLp . /
r mqCn1 dr
0
mqCn
q
D cq;m;n d
kf kLp . / D
mC nq
cq;m;n d
kf kLp . / :
0
l1
Q f .x/ @
0
@
1jjl1
X
1jjl1
1
q
1
Z
jj
D f .y/ jx yj dy A k!kL1 .Rn /
B
1
cq;jj;n d
jjC nq
D f
Lp .B/
1
A k!kL
1 .R
n/
That is
0
l1
B
Q f .x/ cq;l;n k!k .Rn / B
L1
@
1jjl1
x 2 B:
11
0
jjC nq
f
d
D
B
CC
Lp .B/
B
CC ;
@
AA
(2.33)
l
./. Here D denotes any of
Rn ! .x/ dx D 1; f 2 C ./ or f 2 W1
D , Dw : We also consider
46
Ql1 f .x/ D
B
X
1jjl1
X
1jjl1
.1/jj
Hence
l1
Q f .x/
1jjl1
1
.1/jj
Dy .x y/ ! .y/A f .y/ dy
Z
B
Dy .x y/ ! .y/ jf .y/j dy:
1jjljj1
(2.35)
We derive 8 x 2 B,
1
80
X
1
Dy
1
01jjl1
X
1
A kf kL .B/ ;
@
.x y/ ! .y/
Dy
1
L1 .B/
<
l1
1jjl1
Q f .x/
if f 2 L1 .B/ ;
1
1
0
1
1
@
Dy .x y/ ! .y/L .B/ kf kLp .B/ ;
1jjl1
:
if f 2 Lp .B/ :
(2.34)
(2.36)
1
.1/jjCjj C
Dy .x y/ ! .y/A f .y/ dy
.1/jjCjj
Z
B
DyC .x y/ ! .y/
f .y/ dy:
(2.37)
l1
Q f .x/
X
1jjljj1
C
Dy .x y/ ! .y/ jf .y/j dy: (2.38)
B
2.2
Background
47
We obtain 8 x 2 B;
1
80
C
1
@
Dy
1jjljj1
1
0
1 C
A kf kL .B/ ;
@
.x y/ ! .y/
Dy
1
L1 .B/
< 1jjljj1
l1
Q f .x/
if f 2 L1 .B/ ;
1
1
1
0
1 C
A kf kL .B/ ;
@
.x y/ ! .y/
Dy
p
Lq .B/
1jjljj1
:
if f 2 Lp .B/ :
(2.39)
The final remark follows.
f .x/ D
(2.40)
where
Rl f .x/ WD
XZ
jjDl
D f .y/
Vx
jx yjnl
w .x; y/ dy
(2.41)
where
Rl f .x/ WD
X
jjDl;
Z
Vx
D f .y/
jx yjnlCjj
w .x; y/ dy:
(2.43)
48
Additionally we give
Theorem 2.23. Let Rn be a domain star-shaped with respect
R to the open ball
B D B .x0 ; r/ such that B , ! 2 L1 .Rn /, supp ! B, Rn ! .x/ dx D 1,
loc
l 2 N and f 2 W1l
./. Then for almost every x 2
X 1 Z
Dw f .y/ .x y/ ! .y/ dy
f .x/ D
B
jj<l
Cl
X 1 Z
.x y/ ! .y/
B
jjDl
Z
.1 t/
l1
Dw f
.y C t .x y// dt dy:
(2.44)
Proof. From the assumptions of the theorem, we get for almost every x 2 that
f .x/
X 1 Z
Dw f .y/ .x y/ ! .y/ dy
B
jj<l
XZ
jjDl
Vx
Dw f .y/
jx yjnl
w .x; y/ dy;
R .D f /.y/
implying that Vx w nl w .x; y/ dy is finite for almost every x 2 .
jxyj
From [6], p. 105, (3.41) there, we know that 8 x 2 Rn
sup py w .x; y/ D sup py w .x; y/ Kx ;
where Kx is the cone in Rn related to Vx , see again [6], pp. 93100.
So acting similarly to [6], p. 107 and working backwards we derive
XZ
jjDl
Vx
Dw f .y/
jx yj
nl
w .x; y/ dy D l
X 1
J ;
jjDl
where
Z
J D
.x z/
Dw f .z/
jx zjn
Rn
Replacing by
jxzj
,
1t
Z
J D
Rn
Z
zx
! xC
n1 d dz:
jz xj
jxzj
1
we obtain
Dw f .z/ .x z/
Z
!
0
z tx
1t
dt
.1 t/nC1
dz:
2.2
Background
49
Z
.1 t/l1
Rn
Dw f .y C t .x y// .x y/ ! .y/ dy dt
Z
.x y/ ! .y/
D
Rn
Z
.x y/ ! .y/
D
B
.1 t/l1 Dw f .y C t .x y// dt dy
.1 t/l1 Dw f .y C t .x y// dt dy:
XZ
jjDl
Dw f .y/
jx yjnl
Vx
w .x; y/ dy
Z 1
X 1 Z
l1
Dw f .y C t .x y// dt dy;
Dl
.x y/ ! .y/
.1 t/
B
0
jjDl
Proposition 2.24. Same assumptions as in Theorem 2.23. Then for almost every
x 2 B we get
jRemainder (2.44)j ld l k!kL1 .Rn /
X 1
D f .y C t .x y//
:
w
L1 .B0;1/
jjDl
(2.45)
In (2.45) we assume for all W jj D l that
D f .y C t .x y//
< 1:
w
L1 .B0;1/
Proof. We have that
X
Z 1
Z
1
l
.x y/ ! .y/
.1 t/l1 Dw f .y C t .x y// dt dy
jjDl B
Z
X 1
l
j.x y/ j j! .y/j
B
jjDl
Z
.1 t/
0
l1
D f .y C t .x y// dt dy
w
50
X 1 Z Z 1
l1
Dw f .y C t .x y// dt dy
.1 t/
B 0
ld k!kL1 .Rn /
l
jjDl
ld l k!kL1 .Rn /
X 1 Z Z 1
D f .y C t .x y// dt dy
w
B 0
jjDl
X 1
D f .y C t .x y//
;
w
L1 .B0;1/
D ld l k!kL1 .Rn /
jjDl
! .x/ f .x/ dx
! .x/ g .x/ dx
.f; g/ WD ! .x/ f .x/ g .x/ dx
B
1
2
Z
Z
C
Z
:
and
Z
g .y/ ! .y/ dy C Ql1 g .x/ C Rl g .x/ :
g .x/ D
B
(2.46)
2.3
Main Results
51
Hence
! .x/ f .x/ g .x/
Z
D ! .x/ g .x/ f .y/ ! .y/ dy C! .x/ g .x/ Ql1 f .x/C! .x/ g .x/ Rl f .x/;
B
and
! .x/ f .x/ g .x/
D ! .x/ f .x/
Z
g .y/ ! .y/ dy C! .x/ f .x/ Ql1 g .x/C! .x/ f .x/ Rl g .x/ :
B
Therefore
Z
Z
Z
! .x/ f .x/ g .x/ dx D
! .x/ g .x/ dx
f .y/ ! .y/ dy
B
Z
C
and
Z
Z
! .x/ f .x/ dx
g .y/ ! .y/ dy
Z
C
Z
! .x/ g .x/ dx
B
f .x/ ! .x/ dx
B
and
Z
Z
g .x/ ! .x/ dx
B
! .x/ f .x/ dx
52
1
D
2
Z
Z
B
B
B
22
1
0
X d jj kD f k1;B
k!k2L1 .Rn /
44kgk1;B @
A
2
1jjl1
C kf k1;B @
1jjl1
13
d jj kD gk1;B
A5
#3
.nd /l
max
5;
C
kgk1;B kD f kmax
1;l;B C kf k1;B kD gk1;l;B
l
"
(2.47)
where d is the diameter of B.
When l D 1 the sums in (2.47) collapse.
Proof. One also in general obtains (x 2 B)
l1
Q f .x/
X
1jjl1
X
1jjl1
X
1jjl1
D f .y/ j.x y/ j j! .y/j dy
B
Z
j! .y/j dy d jj D f
B
1
:
k!kL1 .B/ d jj D f
L1 .B/
L1 .B/
2.3
Main Results
53
So for D f
L1 .B/
0
X
l1
L1 .B/ C
A;
1jjl1
(2.48)
for x 2 B.
By (2.46), Proposition 2.7 and (2.48) we obtain
2
0
2
14
.f; g/
k!k2L1 .B/ 4kgk1;B @
2
0
1jjl1
C kf k1;B @
1jjl1
"
C k!kL1 .B/ kgk1;B
C kf k1;B
k!k2L1 .Rn /
2
22
1
d jj kD f k1;B
A
13
d jj kD gk1;B
A5
44kgk1;B @
X
1jjl1
0
C kf k1;B @
#3
5
1
d jj kD f k1;B
A
X
1jjl1
13
d jj kD gk1;B
A5
#3
.nd /l
max
5;
C
kgk1;B kD f kmax
1;l;B C kf k1;B kD gk1;l;B
l
"
54
Z
Z
Z
k!k2L1 .Rn /
2
0
28
<
4 kgkL .B/ @
1
:
!1
d jj
D f
A
w
L1 .B/
1jjl1
! 19
=
d jj
D g
A
w
L1 .B/
;
C kf kL1 .B/ @
1jjl1
1
X
D f
A
C .nd / 4kgkL1 .B/ @
w
L1 .B/
l
jjDl
133
X
D g
A55 : (2.49)
C kf kL1 .B/ @
w
L1 .B/
jjDl
.f; g/
k!k2L1 .Rn /
2
28
0
<
4 kgkL .B/ @
1
:
1jjl1
!1
d jj
D f
A
w
L1 .B/
C kf kL1 .B/ @
1jjl1
!19
=
d jj
D g
A
w
L1 .B/
;
82
0
1
<
X
D f
A
C 4kgkL1 .B/ @
w
L1 .B/
:
jjDl
93
13
=
X
l 5
D g
A
5
C kf kL1 .B/ @
nd
;
w
L1 .B/
;
0
jjDl
2.3
Main Results
55
R
loc
l 2 N and f; g 2 W1l
./ : Furthermore assume
Rn ! .x/
dx D
1;
D f
D g
,
<
1
for
all
W
1
l; l n. Then
jj
w
w
L1 .B/
L1 .B/
Z
Z
Z
B
B
B
82
0
!1
n
nCjj
X
2 k!k2L1 .Rn / <
d
D f
A
4kgkL1 .B/ @
nC1 n
w
L1 .B/
2
2 C1 :
1jjl1
C kf kL1 .B/ @
1jjl1
!13
d nCjj
D g
A5
w
L1 .B/
1
X
D f
A
C 4kgkL1 .B/ @
w
L1 .B/
2
jjDl
9
13
=
X
lCn
D g
A
5
C kf kL1 .B/ @
nd
: (2.50)
w
L1 .B/
;
0
jjDl
!
.x/
f
.x/
dx
!
.x/
g
.x/
dx
B
B
B
82
0
1
X
k!k2L1 .Rn / <
D f
A
4kgkL .B/ @
w
1
Lp .B/
:
2
jjDl
13
X
n
D g
A5 ncp d l p Cn
C kf kL1 .B/ @
w
Lp .B/
2
jjDl
X
1jjl1
0
C kf kL1 .B/ @
0
11
jjC nq
D f
w
Lp .B/ d
@
AA
X
1jjl1
0
1139
jjC nq
D g
=
w
Lp .B/ d
@
AA5 :
;
(2.51)
56
Remark 2.30. When f; g 2 C l ./ the Theorems 2.27, 2.28, 2.29 are again valid.
In this case we replace Dw by Dw in all inequalities (2.49), (2.50), and (2.51).
We give
to the open ball
Theorem 2.31. Let Rn be a domain star-shaped with respect
R
B D B .x0 ; r/ such that B , ! 2 C01 ./, supp ! B, Rn ! .x/ dx D 1: Let
loc
k!k1
2
28
<
4 2 kgkL .B/ kf kL .B/
1
1
:
0
@
1jjl1
19
=
1
A
Dy .x y/ w .y/
1;B 2 ;
X
C k!k1 nd l 4kgkL1 .B/ @
D f
jjDl
X
C kf kL1 .B/ @
D g
jjDl
L1 .B/
1
A
L1 .B/
133
A55 :
(2.52)
k!k1
kf kL1 .B/ kgkL1 .B/ C kf kL1 .B/ kgkL1 .B/
2
0
1
X
1
A
@
sup Dy .x y/ w .y/
L1 .B/
x2B
1jjl1
0
1
2
n
X
nd lCn 2 k!k1 4
A
C n n
kgkL1 .B/ @
D f
L1 .B/
2
2 C1
jjDl
0
139
=
X
A5 :
C kf kL1 .B/ @
(2.53)
D g
;
L1 .B/
jjDl
2.3
Main Results
57
1
p
1
q
D 1; l >
n
;
p
for all W jj D l; D f , D g; f; g 2
2
k!k1 4
.f; g/
kgkL1 .B/ kf kLp .B/ C kf kL1 .B/ kgkLp .B/
2
0
1jjl1
1
1
A
sup Dy .x y/ w .y/
Lq .B/
x2B
2
1jjl1
0
C kf kL1 .B/ @
1jjl1
0
11
D f
Lp .B/ jjC n
q
@
AA
d
0
1133
D g
Lp .B/ jjC n
q AA55 :
@
d
(2.54)
D f
A
4 kgkL1 .B/ @
.f; g/
w
L1 .B/
:
2
1jjl1
C kf kL1 .B/ @
2
1jjl1
!19
=
d jj
D g
A
w
L1 .B/
;
1
X 1
D f .y C t .x y//
A
Cld l 4kgkL1 .B/ @
w
L1 .B0;1/
jjDl
13 3
X 1
D g .y C t .x y//
A5 5
C kf kL1 .B/ @
w
L1 .B0;1/
jjDl
:
Proof. By Theorem 2.27 and Proposition 2.24.
(2.55)
58
l1
f .x/
f .y/ ! .y/ dy Q f .x/
(2.56)
f .x/
f
.y/
!
.y/
dy
B
0
!1
n
nCjj
X
k!kL1 .Rn / 2
d
@
A
kD f k1
2n
n2 C 1
1jjl1
(2.57)
l1
X
@
D f
jjDl
1
L1 .B/
A k!kL
1 .R
n/
nd l DW A2 :
(2.58)
Also it holds
.f / .x/ WD f .x/
f .y/ ! .y/ dy
B
28
!9
=
< X
jj
d
4
D f
:
L1 .B/ ;
1jjl1
X
C@
D f
jjDl
L1 .B/
A nd l 5 k!kL
1 .R
n/
DW B2 : (2.59)
2.3
Main Results
59
Proof. Use of Theorems 2.8, 2.11; (2.40), (2.20), (2.28), and (2.30).
We give
Theorem
2.35.
All as in Theorems 2.8, 2.11. Suppose l n, k!kL1 .Rn / < 1;
< 1 for all W jj D l. Then for every x 2 B (almost every
and D f
L1 .B/
x 2 B, respectively), it holds
0
X
E .f / .x/ @
D f
L1 .B/
jjDl
1 .R
Additionally, assume that D f
20
.f / .x/ 4@
L1 .B/
X
1jjl1
A k!kL
n/
n 2
d lCn DW A3 : (2.60)
n
2
n2 C 1
!1
d nCjj
A
D f
L1 .B/
X
C@
D f
L1 .B/
jjDl
A nd lCn 5
2n
2
n
2
C1
k!kL1 .Rn /
DW B3 :
(2.61)
1
1
n
p C q D 1, l > p ; D f 2 Lp .B/ for jj D l. Then for every x 2 B (almost every
x 2 B, respectively), it holds
0
1
X
n
A k!kL .Rn / ncp d l p Cn DW A4 :
E .f / .x/ @
(2.62)
D f
1
Lp .B/
jjDl
1jjl1
X
C@
D f
jjDl
DW B4 :
Lp .B/
1
A ncp d
3
l pn Cn
5 k!kL
1 .R
n/
(2.63)
60
0
B
.f / .x/ k!kL1 .B/ @
L1 .B/ C
1jjl1
1
0
n
nd lCn 2 k!kL1 .Rn / X
A
@
C
D f
L1 .B/
2n
n2 C 1
jjDl
DW B5 :
(2.64)
f .y/ ! .y/ dy
.f / .x/ WD f .x/
B
0
1
X
1
@
Dy .x y/ ! .y/ A kf kL1 .B/
1
1jjl1
X
C@
D f
jjDl
L1 .B/
A k!kL
1 .R
n/
nd l :
(2.65)
Theorem 2.39. Here all as in Corollary 2.12. Suppose ln and D f
L1 .B/
<1
1jjl1
1
1
A kf kL .B/
Dy .x y/ ! .y/
1
L1 .B/
1
0
n
nd lCn 2 k!kL1 .Rn / X
A:
@
C
D f
L1 .B/
2n
n2 C 1
jjDl
(2.66)
2.3
Main Results
61
1
p
D 1,
1
q
l>
D f 2 Lp .B/ for jj D l and f 2 Lp .B/. Then for every x 2 B (almost
every x 2 B, respectively), we derive
1
0
X
1
A kf kL .B/
.f / .x/ @
Dy .x y/ ! .y/
p
Lq .B/
n
;
p
1jjl1
C ncp d
l pn Cn
X
k!kL1 .Rn / @
D f
Lp .B/
jjDl
1
A:
(2.67)
Z
D ! .y/ f .y/ dy Ql1 f .x/
E .f / .x/ WD D f .x/ .1/jj
B
0
1
X
A k!k1 nd ljj DW A5 :
@
(2.68)
D f
L1 .B/
jjDl;
Also, we derive
Z
jj
D ! .y/ f .y/ dy
.f / .x/ WD D f .x/ .1/
B
1
0
X
1
C
@
Dy .x y/ ! .y/ A kf kL1 .B/
1
1jjljj1
C@
D f
L1 .B/
jjDl;
A k!k1 nd ljj :
(2.69)
1
L1 .B/
A k!k1
n 2
d ljjCn DW A6 :
n
2
n2 C 1
(2.70)
62
.f / .x/ @
1jjljj1
D f
C@
1
1
C
A kf kL .B/
Dy .x y/ ! .y/
1
L1 .B/
jjDl;
2
A k!k1 n d ljjCn :
n
L1 .B/
2n
2 C 1
(2.71)
D f
1
q
D 1,
1
A k!k1 ncp d ljj p Cn DW A7 :
n
Lp .B/
jjDl;
n
p
1
p
(2.72)
.f / .x/ @
1jjljj1
C@
X
jjDl;
1
1
C
A kf kL .B/
Dy .x y/ ! .y/
p
Lq .B/
D f
1
A k!k1 ncp d ljj p Cn :
n
Lp .B/
(2.73)
2.3
Main Results
63
M .f / WD
f .x/ .x/ dx
f .x/ ! .x/ dx
B1
.x/ dx A1 ;
(2.74)
f .x/ ! .x/ dx B1 :
(2.75)
l1
f .x/
B1
and
m .f / WD
Z
f .x/
.x/ dx
B1
(2.76)
m .f / B2 :
(2.77)
and
(2.78)
m .f / B3 :
(2.79)
and
(2.80)
m .f / B4 :
(2.81)
(2.82)
D ! .x/ f .x/ dx
.x/ D f .x/ dx .1/jj
M .f / WD
B1
B1
(2.83)
64
(2.84)
We finish with
Theorem 2.52. All as in Theorem 2.43. Then
M .f / A7 :
(2.85)
2.4 Applications
Example 2.53 (see also [5], p. 93). Let B WD fx 2 Rn W jx x0 j < g, and
' .x/ WD
<
2 !1
jxx0 j
1
: 0,
; if jx x0 j < ;
if jx x0 j :
(2.86)
R
CallRc WD Rn ' .x/ dx > 0, then .x/ WD 1c ' .x/ 2 C01 .Rn / with supp D B
and Rn .x/ dx D 1 and max jj cons tan t n :
We call a cut-off function.
1
One for this chapters results by choosing ! .x/ D .x/ or ! .x/ D Vol.B/
, etc.,
can give lots of applications.
Here, selectively we give some special cases inequalities. We start with
ChebyshevGruss-type inequalities.
Corollary 2.54 (to Theorem 2.26). All assumptions as in Theorem 2.26. Case of
l D 1. Then
Z
Z
Z
nd
k!k2L1 .Rn /
2
max
kgk1;B kD f kmax
1;1;B C kf k1;B kD gk1;1;B : (2.87)
If f D g, then
Z
Z
2
2
! .x/ f .x/ dx
! .x/ f .x/ dx
B
B
nd k!k2L1 .Rn / kf k1;B kD f kmax
1;1;B :
(2.88)
References
65
Corollary 2.55 (to Theorem 2.27). All assumptions as in Theorem 2.27. Case of
1
f D g, l D n and ! .x/ WD Vol.B/
; for all x 2 B, ! .x/ WD 0 on Rn B. Then
Z
n
Z
2
n
2
C
1
2
f .x/ dx
f 2 .x/ dx
n
n
B
2
d
B
8
0
!19
n
2<
=
n
jjn
X
2
2 C1
d
D f
4 kf kL .B/ @
A
n
w
1
.B/
L1
:
;
2
0
1jjn1
113
X
D f
AA5 :
C n @kf kL1 .B/ @
w
L1 .B/
(2.89)
jjDn
f .x/
f .y/ ! .y/ dy nd k!kL1 .Rn / kD f kmax
1;1;B WD Z1 :
(2.90)
Z
Z
f .x/ .x/ dx
f .y/ ! .y/ dy Z1 :
(2.91)
B1
References
1. G.A. Anastassiou, Quantitative Approximations, Chapman & Hall/CRC, Boca Raton, New
York, 2000.
2. G.A. Anastassiou, Probabilistic Inequalities, World Scientific, Singapore, New Jersey, 2010.
3. G.A. Anastassiou, Advanced Inequalities, World Scientific, Singapore, New Jersey, 2010.
4. G.A. Anastassiou, Multivariate Inequalities based on Sobolev Representations, Applicable
Analysis, accepted 2011.
5. S. Brenner and L.R. Scott, The mathematical theory of finite element methods, Springer, N. York,
2008.
6. V. Burenkov, Sobolev spaces and domains, B.G. Teubner, Stuttgart, Leipzig, 1998.
7. P.L. Chebyshev, Sur les expressions approximatives des integrales definies par les autres prises
entre les memes limites, Proc. Math. Soc. Charkov, 2(1882), 93-98. h
1 Rb
8. G. Gruss, Uber
das Maximum des absoluten Betrages von
a f .x/ g .x/ dx
ba
i
Rb
Rb
1
.ba/2 a f .x/ dx a g .x/ dx , Math. Z. 39 (1935), pp. 215-226.
9. A. Ostrowski, Uber die Absolutabweichung einer differentiabaren Funcktion von ihrem Integralmittelwert, Comment. Math. Helv. 10(1938), 226-227.