SpringerBriefs in Mathematics

For further volumes:

http://www.springer.com/series/10030
George A. Anastassiou

Approximation by
Multivariate Singular
Integrals

123
George A. Anastassiou
Department of Mathematical Sciences
University of Memphis
Memphis, TN 38152
USA
[email protected]

ISSN 2191-8198 e-ISSN 2191-8201

ISBN 978-1-4614-0588-7 e-ISBN 978-1-4614-0589-4
DOI 10.1007/978-1-4614-0589-4
Springer New York Dordrecht Heidelberg London

Library of Congress Control Number: 2011933230

Mathematics Subject Classification (2010): 41A17, 41A25, 41A28, 41A35, 41A36, 41A60, 41A80

© George A. Anastassiou 2011

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)

To my wife Koula and my daughters
Angela and Peggy
Preface

This short monograph is the first to deal exclusively with the study of the
approximation of multivariate singular integrals to the identity-unit operator. Here
we study quantitatively the basic approximation properties of the general multivari-
ate singular integral operators, special cases of which are the multivariate Picard,
Gauss-Weierstrass, Poisson-Cauchy and trigonometric singular integral operators,
etc. These operators are not general positive linear operators. In particular we
study the rate of convergence of these operators to the unit operator, as well as
the related simultaneous approximation. These are given via inequalities and by
the use of multivariate higher order modulus of smoothness of the high order
partial derivatives of the involved function. Also we study the global smoothness
preservation properties of these operators. Some of these multivariate inequalities
are proved to be attained, that is sharp. Furthermore we give asymptotic expansions
of Voronovskaya type for the error of approximation. These properties are studied
with respect to Lp norm, 1  p  1: The last chapter presents a related Korovkin
type approximation theorem for functions of two variables. Plenty of examples
are given.
For the convenience of the reader, the chapters are self-contained.
This brief monograph relies on author’s last two years of related research work,
more precisely see author’s articles in the list of references of each chapter.
Advanced courses can be taught out of this short book. All necessary background
and motivations are given per chapter.
The presented results are expected to find applications in many areas of pure
and applied mathematics, such as mathematical analysis, probability, statistics and
partial differential equations, etc. As such this brief monograph is suitable for
researchers, graduate students, and seminars of the above subjects, also to be in
all science libraries.
The preparation of this book took place during 2010–2011 in Memphis,
Tennessee, USA.

vii
viii Preface

I would like to thank my family for their dedication and love to me, which was
the strongest support during the writing of thismonograph.

Department of Mathematical Sciences George A. Anastassiou

The University of Memphis, TN 38152, USA
Contents

1 Uniform Approximation by General Multivariate Singular

Integral Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 1
1.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 1
1.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 14
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 16
2 Lp -Approximation by General Multivariate Singular
Integral Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 19
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 19
2.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 19
2.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 30
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 32
3 Global Smoothness Preservation and Simultaneous
Approximation by Multivariate General Singular Integrals .. . . . . . . . . . . . 33
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 33
3.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 34
3.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 44
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 45
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 45
4 Multivariate Voronovskaya Asymptotic Expansions
for General Singular Integrals . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 47
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 47
4.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 47
4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 55
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 56

ix
x Contents

5 Simultaneous Approximation by Multivariate Complex

General Singular Integrals .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 57
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 57
5.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 57
5.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 67
5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 68
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 68
6 Approximation of Functions of Two Variables via Almost
Convergence of Double Sequences . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 69
6.1 Introduction and Preliminaries . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 69
6.2 Korovkin-Type Approximation Theorem .. . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 71
6.3 Some Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 76
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . 79
Chapter 1
Uniform Approximation by General
Multivariate Singular Integral Operators

In this chapter, we present the uniform approximation properties of general multi-

variate singular integral operators over RN , N  1. We give their convergence to the
unit operator with rates. The estimates are pointwise and uniform. The established
inequalities involve the multivariate higher order modulus of smoothness. We list the
multivariate Picard, Gauss-Weierstrass, Poisson Cauchy and trigonometric singular
integral operators where this theory can be applied directly. This chapter relies
on [2].

1.1 Introduction

The rate of convergence of univariate singular integral operators has been studied
in [1, 7–9, 11, 12], and these works motivate the current chapter. Here we consider
some very general multivariate singular integral operators over RN , N  1, and
we study the degree of approximation to the unit operator with rates over smooth
functions. We present related inequalities involving the multivariate higher modulus
of smoothness with respect to kk1 . The estimates are pointwise and uniform.
See Theorems 1.9, 1.11. We mention particular operators that fulfill our theory.
The discussed linear operators are not in general positive. Other motivation comes
from [4, 5].

1.2 Main Results

Here r 2 N; m 2 ZC , we define
8  
ˆ
ˆ .1/ rj r
j m ; if j D 1; 2; : : : ; r;
<
j
Œm
˛j;r WD   (1.1)
ˆ Pr r m
:̂ 1  .1/ri i ; if j D 0;
i D1 i

G.A. Anastassiou, Approximation by Multivariate Singular Integrals, 1

SpringerBriefs in Mathematics, DOI 10.1007/978-1-4614-0589-4 1,
© George A. Anastassiou 2011
2 1 Uniform Approximation by General Multivariate Singular Integral Operators

and
Œm
X
r
Œm
ık;r WD ˛j;r j k ; k D 1; 2; : : : ; m 2 N. (1.2)
j D1

See that
X
r
Œm
˛j;r D 1; (1.3)
j D0

and
X
r    
r r
 .1/rj D .1/r : (1.4)
j 0
j D1

Let n be a probability Borel measure on RN , N  1, n > 0, n 2 N.

We now define the multiple smooth singular integral operators
X
r Z
Œm
Œm
r;n .f I x1 ; : : : ; xN / WD ˛j;r f .x1 C s1 j; x2 C s2 j; : : : ; xN CsN j / dn .s/;
j D0 RN
(1.5)

where s WD .s1 ; : : : ; sN /, x WD .x1 ; : : : ; xN / 2 RN I n; r 2 Z, m 2 ZC , f W RN ! R

is a Borel measurable function, and also .n /n2N is a bounded sequence of positive
real numbers.
Œm
Remark 1.1. The operators r;n are not in general positive. For example, consider
PN 2
the function ' .u1 ; : : : ; uN / D i D1 ui and also take r D 2, m D 3; xi D 0,
i D 1; : : : ; N . See that '  0, however
0 1 !
X
2 Z X
N
.'I 0; 0; : : : ; 0/ D @ j 2 ˛j;2 A
Œ3 Œ3
2;n si2 dn .s/
j D1 RN i D1

!
 Z X
N
Œ3 Œ3
D ˛1;2 C 4˛2;2 si2 dn .s/
RN i D1

 Z !
1 X
N
D 2 C si2 dn .s/ < 0; (1.6)
2 RN i D1

 
R P
N
assuming that RN si2 dn .s/ < 1:
i D1
1.2 Main Results 3

Œm
Lemma 1.2. The operators r;n preserve the constant functions in N variables.
Proof. Let f .x1 ; : : : ; xN / D c, then

X
r Z
Œm
Œm
r;n .cI x1 ; : : : ; xn / D ˛j;r cdn .s/ D c:
j D0 Rn

t
u
We need
 
Definition 1.3. Let f 2 CB RN , the space of all bounded and continuous func-
tions on RN . Then, the rth multivariate modulus of smoothness of f is given by
(see, e.g. [6])
 r 
!r .f I h/ WD p sup  .f /1 < 1; h > 0; (1.7)
u1 ;u2 ;:::;uN
u21 C:::Cu2N h

where kk1 is the sup-norm and

ru f .x/ WD ru1 ;u2 ;:::;uN f .x1 ; : : : ; xN /

X
r  
r
D .1/rj f .x1 C j u1 ; x2 C j u2 ; : : : ; xN C j uN / : (1.8)
j
j D0

 
Let m 2 N and let f 2 C m RN :
Suppose that all partial derivatives of f of order m are bounded, i.e.
 m 
 @ f .; ; : : : ; / 
 
 @x ˛1 : : : @x ˛N  < 1; (1.9)
1 N 1

PN
for all ˛j 2 ZC , j D 1; : : : ; N I j D1 ˛j D m:
We make
Remark 1.4.˛ Let l D 0; 1; : : : ; m. The lth order partial derivative is denoted
by f˛ WD @@xf˛ , where ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC , i D 1; : : : ; N and j˛j WD
PN
i D1 ˛i D l:
Consider gz .t/ WD f .x0 C t .z  x0 //, t  0; x0 ; z 2 RN .
Then
2 !j 3
X N
@
gz.j / .t/ D 4 .zi  x0i / f 5.x01 C t .z1  x01 / ; : : : ; x0N Ct .zN x0N // ;
i D1
@x i

(1.10)
for all j D 0; 1; : : : ; m:
4 1 Uniform Approximation by General Multivariate Singular Integral Operators

We have the multivariate Taylor’s formula, [3]:

f .z1 ; : : : ; zN / D gz .1/
X Z
m .j /
gz .0/ 1 1  
D C .1  /m1 gz.m/ ./  gz.m/ .0/ d:
j D0
jŠ .m  1/Š 0

(1.11)

Notice gz .0/ D f .x0 /. Also for j D 0; 1; : : : ; m, we have

0 1
B C N !
X B jŠ C Y
B C ˛i
gz.j / .0/ D B N C .zi  x0i / f˛ .x0 / : (1.12)
BY C
˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; @ ˛ ŠA
i D1
P
N i
i D1;:::;N; j˛jWD ˛i Dj i D1
i D1

Furthermore,
0 1
B C N !
X B mŠ C Y
B C ˛i
gz.m/ ./ D B N C .zi  x0i / f˛ .x0 C  .z  x0 // ;
BY C
@ ŠA
˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; i D1
PN
˛ i
i D1;:::;N; j˛jWD ˛i Dm i D1
i D1
(1.13)
0    1:
We apply the above for

z D .z1 ; : : : ; zN / D .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / D x C sj;

and
x0 D .x01 ; : : : ; x0N / D .x1 ; x2 ; : : : ; xN / D x;
to obtain

X
m . jQ /
gxCsj .0/ 1
f .x1 C s1 j; : : : ; xN C sN j / D gxCsj .1/ D C
jŠQ .m  1/Š
jQD0
Z 1  
.m/ .m/
 .1  /m1 gxCsj ./  gxCsj .0/ d;
0
(1.14)

where gxCsj .t/ WD f .x C t .sj //:

Notice gxCsj .0/ D f .x/.
1.2 Main Results 5

Also for jQ D 0; 1; : : : ; m we have

0 1
B C N !
X B jQŠ C Y
.jQ/ B C
gxCsj .0/ D B N C .si j / ˛i
f˛ .x/: (1.15)
BY C
˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; @ ˛ Š A i D1
P
N i
i D1;:::;N; j˛jWD ˛i DjQ i D1
i D1

Furthermore, we get
0 1
B C N !
X B 1 C Y
.m/ B C
gxCsj ./=mŠ D B N C .si j / ˛i
f˛ .x C  .sj // ;
BY C
˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; @ ˛ Š A i D1
PN i
i D1;:::;N; j˛jWD ˛i Dm i D1
i D1
(1.16)
0    1:
jQ D 1; : : : ; m and ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 ZC ; i D 1; : : : ; N;
For P
j˛j WD N Q
i D1 ˛i D j , will be proved that

Z Y
N
c˛;n;jQ WD c˛;n WD si˛i dn .s1 ; : : : ; sN / 2 R, (1.17)
RN i D1

see (1.35).
Consequently, we derive

R .jQ/
X
m
RN gxCsj .0/dn .s/
jQŠ
jQD1

0 0 1 1
B B C C
B B 1 C C
X
m
B X B C C
jQ B
D j B B N C c˛;n f˛ .x/C
C: (1.18)
B ˛WD.˛ ;:::;˛ /; ˛ 2ZC ; BY C C
jQD1 @ @ ˛ A A
iŠ
1 N i
N P
i D1;:::;N; j˛jWD ˛i DjQ i D1
i D1
6 1 Uniform Approximation by General Multivariate Singular Integral Operators

Next we observe that

1
Z Z 1   
m1 .m/ .m/
.1  / gxCsj ./  gxCsj .0/ d dn .s/
.m  1/Š RN 0

X 1
D mj m !
0 1 Y
N

B C ˛i Š
B C
B C i D1
B C
B ˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; C
B C
@ P
N A
i D1;:::;N; j˛jWD ˛i Dm
i D1

Z Z ! !
Y
N 1
si˛i .1  / m1
Œf˛ .x C  .sj //  f˛ .x/ d dn .s/ :
RN i D1 0
(1.19)

We further make
Remark 1.5. We further notice that

X
r Z X
r
Œm Œm Œm
r;n .f I x/  f .x/ D ˛j;r .f .x C sj /  f .x// dn .s/ D ˛j;r 
j D0 RN j D0

Z " .jQ/ Z #
X
m
gxCsj .0/ 1 1  
m1 .m/ .m/
C .1  / gxCsj ./gxCsj .0/ d dn .s/:
RN jQŠ .m  1/Š 0
jQD1
(1.20)

That is
0 0 1 1
X
r Z X
m .jQ/
gxCsj .0/
˛j;r @ @ A dn .s/A
Œm
r;n .x/ WD r;n .f I x/  f .x/ 
Œm Œm

j D0 RN jQŠ
jQD1

X
r Z Z 1
1 Œm
D ˛j;r .1  /m1
.m  1/Š j D0 RN 0

 !
 
.m/ .m/
 gxCsj ./  gxCsj .0/ d dn .s/

DW Rr;n
Œm
: (1.21)
1.2 Main Results 7

We see that
0 1
B C
X
r X
m B X c f .x/ C
Œm jQ B ˛;n ˛ C
Œm
r;n .x/ D Œm
r;n .f I x/  f .x/  ˛j;r j B C
B YN C
j D1 jQD1 @˛1 ;:::;˛N 0W ˛Š A
j˛jDjQ i
i D1
0 1
0 1B C
X m Xr B X c f .x/ C
@ Q B C
˛j;r j A B
Œm ˛;n ˛
D r;n .f I x/  f .x/ 
Œm j
C
B YN C
@ ˛Š A
Q
j D1 j D1 ˛1 ;:::;˛N 0W
j˛jDjQ i
i D1
0 1
B C
X
m B X c˛;n f˛ .x/ C
Œm B C
D Œm
r;n .f I x/  f .x/  ıjQ;r B C: (1.22)
B YN C
jQD1 @ 1 N
˛ ;:::;˛ 0W
˛i Š A
j˛jDjQ
i D1

So we have proved that

0 1
B C
X
m B X c f .x/ C
Œm B ˛;n ˛ C
Œm
r;n .x/ D Œm
r;n .f I x/  f .x/  ıjQ;r B C: (1.23)
B C
@˛1 ;:::;˛N 0W Y ˛ Š A
N
jQD1
j˛jDjQ i
i D1

We also make
Remark 1.6. We observe that
0 1
B C
X
r X B 1 C
Œm B C
Œm
Rr;n Dm ˛j;r j m B N C
0 1BY C
@ ŠA
j D1
@ A ˛ i
˛1 ;:::;˛N 0W
j˛jDm i D1

Z Z ! !
Y
N 1
si˛i .1  / m1
.f˛ .x C  .sj //  f˛ .x// d dn .s/
RN i D1 0
(1.24)
8 1 Uniform Approximation by General Multivariate Singular Integral Operators

0 1
B C
X B 1 C
B C
Dm B N C
0 1 BY C
@ A
@ ˛ i ŠA
˛1 ;:::;˛N 0W
j˛jDm i D1

Z Y
N Z 1 X
r
Œm
si˛i .1  / m1
˛j;r j m
RN i D1 0 j D1
! !
 .f˛ .x C  .sj //  f˛ .x// d dn .s/
0 1
B C Z Z
X B 1 C Y
N 1
B C
Dm B N C si˛i .1  /m1 
0 1 BY C RN
@ ˛ ŠA
i D1 0
@ A i
˛1 ;:::;˛N 0W
j˛jDm i D1
"     # ! !
X
r
r r
.1/ rj
f˛ .x C  .sj // C .1/ r
f˛ .x/ d dn .s/
j 0
j D1
(1.25)
0 1
B C Z Z
X B 1 C Y
N 1
B C
Dm B N C si˛i .1  /m1 
BY C RN
@ ŠA
i D1 0
˛1 ;:::;˛N 0W
j˛jDm ˛ i
i D1
"   # ! !
X
r
r
.1/rj f˛ .xC .sj // d dn .s/ : (1.26)
j
j D0

We have proved that

0 1
B C
X B 1 C
B C
Œm
Rr;n Dm B N C
BY C
˛1 ;:::;˛N 0 @ ˛ ŠA
j˛jDm i
i D1
Z Z ! !
Y
N 1  
si˛i .1  / m1
rs f˛ .x/ d dn .s/ : (1.27)
RN i D1 0
1.2 Main Results 9

We further make
Remark 1.7. We observe that
0 1
B C
ˇ ˇ X B 1 C
ˇ Œm ˇ .1.27/ B C
ˇRr;n ˇ  m B N C
BY C
˛1 ;:::;˛N 0 @ ˛ ŠA
j˛jDm i
i D1

Z Z ! !
Y
N
˛i
1
jsi j .1  / m1
jrs f˛ .x/j d dn .s/ (1.28)
RN i D1 0

0 1
B C
X B 1 C
B C
m B N C
BY C
˛1 ;:::;˛N 0
j˛jDm
@ ˛ i ŠA
i D1

Z Z ! !
Y
N 1
jsi j ˛i
.1  / m1
krs f˛ k1 d dn .s/
RN i D1 0

0 1
B C
X B 1 C
B C
m B N C
BY C
˛1 ;:::;˛N 0
j˛jDm
@ ˛ i ŠA
i D1

Z Z ! !
Y
N
˛i
1
m1
jsi j .1  / !r .f˛ ;  ksk2 / d dn .s/
RN i D1 0

0 1
B C Z ! !
X B 1 C Y
N
B C ˛i
 B N C jsi j !r .f˛ ; ksk2 / dn .s/ :
BY C RN
@ ŠA
˛1 ;:::;˛N 0 i D1
j˛jDm ˛ i
i D1
(1.29)
10 1 Uniform Approximation by General Multivariate Singular Integral Operators

So far we have established

0 1
B C Z ! !
ˇ ˇ X B 1 C Y
N
ˇ Œm ˇ B C ˛i
ˇRr;n ˇ  B N C jsi j !r .f˛ ; ksk2 / dn .s/
BY C RN
˛1 ;:::;˛N 0 @ ˛ ŠA
i D1
j˛jDm i
i D1
0 1
B C Z  ! !
X B 1 C Y
N
ksk2
B C ˛i
D B N C jsi j !r f˛ ; n dn .s/
BY C RN n
@ ŠA
˛1 ;:::;˛N 0 i D1
j˛jDm ˛ i
i D1
(1.30)

(using the fact

!r .f I u/  .1 C /r !r .f I u/ ; (1.31)
; u > 0; we get)
0 1
B C ! r !
X B ! .f ;  / C Z Y
N
ksk2
B r ˛ n C ˛i
 B N C jsi j 1C dn .s/ :
B Y C RN n
@ ˛i Š A
0 1
i D1
B C
@˛ ;:::;˛ 0 A
1 N
j˛jDm i D1
(1.32)
We have proved that
0 1
B C ! r !
ˇ ˇ X B ! .f ;  / C Z Y
N
ksk2
ˇ Œm ˇ B r ˛ n C
ˇRr;n ˇ  B N C jsi j ˛i
1C dn .s/ :
B Y C RN n
0 1
@ ˛i Š A
i D1
B C
@˛ A
1 ;:::;˛N 0
j˛jDm i D1
(1.33)
We also make
Remark 1.8. Notice that for j˛j D m:
Z !
Y
N
˛i
jsi j dn .s/ (1.34)
RN i D1
Z ! r
Y
N
˛i ksk2
 jsi j 1C dn .s/ < 1;
RN i D1
n

by the assumption in Theorem 1.9, next.

1.2 Main Results 11

Hence
Z !
Y
N
jc˛;n j  jsi j ˛i
dn .s/ < 1; (1.35)
RN i D1
P
for all ˛ W .˛1 ; : : : ; ˛N / W j˛j WD ˛i D m, ˛i 2 ZC .
Hence also Z
jsi jm dn .s/ < 1; (1.36)
RN

for all i D 1; : : : ; N:
Let 1  jQ  m  1, then
Z Z  m  mjQ
jQ jQ jQ
jsi j dn .s/  jsi j dn .s/
RN RN
Z  mjQ
m
D jsi j dn .s/ < 1: (1.37)
RN

That is for 1  jQ  m  1, we obtain that

Z
Q
jsi jj dn .s/ < 1: (1.38)
RN

PN
Let jQ D 1; : : : ; m  1I ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC , i D 1; : : : ; N; j˛j WD i D1
P
˛i D jQ. That is N i D1
˛i
D 1. Hence
jQ

Z Y
N N 
Y  ˛Qi
˛i R  ˛i
 ˛jQ j
jsi j dn .s/  RN jsi j i dn .s/
RN i D1 i D1

N 
Y R  ˛Qi
jQ
D jsi j dn .s/ < 1:
j
RN (1.39)
i D1

Therefore, it holds
Z Y
N
jsi j˛i dn .s/ < 1: (1.40)
RN i D1

Based on the above, we present

 m 
 .;:::;/ 
Theorem 1.9. Let m2 N, f 2 C m .RN /, N 1, x2 RN . Assume  @x@ ˛f1 :::@x ˛N  < 1,
P 1 N 1
for all ˛j 2 ZC , j D 1; : : : ; N W j˛j WD j D1 ˛j D m:
N

Let n be a Borel probability measure on RN , for n > 0, .n /n2N bounded
sequence.
12 1 Uniform Approximation by General Multivariate Singular Integral Operators

C
PNSuppose that for all ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 Z , i D 1; : : : ; N; j˛j WD
i D1 ˛i D m we have that

Z ! r
Y
N
˛i ksk2
un WD jsi j 1C dn .s/ < 1: (1.41)
RN i D1
n

For jQ D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC , i D 1; : : : ; N; j˛j WD

P N Q
i D1 ˛i D j , call

Z Y
N
c˛;n WD c˛;n;jQ WD si˛i dn .s1 ; : : : ; sN / : (1.42)
RN i D1

Then
i)
ˇ 0 1ˇ
ˇ ˇ
ˇ ˇ
ˇ B C ˇ
ˇ B
X Œm B X c˛;n;jQ f˛ .x/ Cˇˇ
m C
ˇ Œm
Er;n .x/ WD ˇr;n .f I x/  f .x/ 
Œm
ıjQ;r B Cˇ
ˇ B Y
N Cˇ
ˇ @ ˇ
˛i Š A ˇ
Q
j D1 ˛ ;:::;˛ 0W
ˇ
1 N
j˛jDjQ
ˇ ˇ
i D1

Z !  !
X .!r .f˛ ; n // Y N
˛i ksk2 r
 ! jsi j 1C dn .s/ ;
0 1 YN
RN i D1
n
B C
@˛ ;:::;˛ 0 A
˛i Š
1 N
j˛jDm i D1

(1.43)

8 x 2 RN :
ii)  
 Œm 
Er;n   R:H:S: .1.43/: (1.44)
1

Given that n ! 0, as n ! 1, and un is uniformly bounded, then we derive

that Er;n
Œm 
1
! 0 with rates.
iii) It holds also that
0 1
B ˇ ˇ C
  m ˇ ˇB ˇ ˇ C
 Œm  X ˇ Œm ˇ B X ˇc˛;n;jQˇ kf˛ k1 C
r;n .f /  f   ˇıjQ;r ˇ B C C R:H:S: .1.43/;
1 B Y
N C
jQD1 @˛1 ;:::;˛N 0W ˛Š A
j˛jDjQ i
i D1
(1.45)
1.2 Main Results 13

given that kf˛ k1 < 1, for all ˛ W j˛j D jQ, jQ D 1; : : : ; m: Furthermore,

as n ! 0 when n ! 1, assuming that c˛;n;jQ ! 0, while un is uniformly
bounded, we conclude that
 
 Œm 
r;n .f /  f  ! 0 (1.46)
1

with rates.
Case m D 0:
 
Remark 1.10. Here f 2 CB RN (bounded and continuous functions). We notice
that
Œ0
r;n .f I x/  f .x/
X r Z
Œ0
D ˛j;r .f .x C js/  f .x// dn .s/
j D0 RN

0 0 1 1
Z X
r X
r
@ .x C js/  @ ˛j;r A f .x/A dn .s/
Œ0 Œ0
D ˛j;r f (1.47)
RN j D0 j D0

0 0 1 1
Z X
r X
r
D @ ˛j;r f .x C js/  @
Œ0
˛j;r A f .x/A dn .s/
Œ0
(1.48)
RN j D1 j D1

0 0 1 1
Z X
r   X
r  
r r A
D @ .1/rj f .x C js/  @ .1/rj f .x/A dn .s/
RN j j
j D1 j D1

0 1
Z X
r    
@ r r
D .1/rj f .x C js/ C .1/r f .x/A dn .s/
RN j 0
j D1

0 1
Z X
r   Z
r  r 
D @ .1/ rj A
f .x C js/ dn .s/ D s f .x/ dn .s/:
RN j RN
j D0
(1.49)

We established that
Z
 r 
Œ0
r;n .f I x/  f .x/ D s f .x/ dn .s/: (1.50)
RN
14 1 Uniform Approximation by General Multivariate Singular Integral Operators

Consequently, it holds
ˇ ˇ Z
ˇ Œ0 ˇ
ˇr;n .f I x/  f .x/ˇ  jrs f .x/j dn .s/
RN
Z Z
 krs f k1 dn .s/  !r .f; ksk2 / dn .s/
RN RN
Z  
ksk2
D !r f; n dn .s/
RN n
Z  
ksk2 r
 !r .f; n / 1C dn .s/: (1.51)
RN n

So we proved
ˇ ˇ Z  
ksk2 r

ˇ Œ0 ˇ
ˇr;n .f I x/  f .x/ˇ  1C dn .s/ !r .f; n / ; (1.52)
RN n

8 x 2 R:
Based on the above, we present
 
Theorem 1.11. Let f 2 CB RN , N  1: Then
  Z   
 Œ0  ksk2 r
r;n f  f   1C dn .s/ !r .f; n / ; (1.53)
1 RN n

under the assumption

Z  
ksk2 r
ˆn WD 1C dn .s/ < 1: (1.54)
RN n

As n ! 1 and n ! 0, given that ˆn are uniformly bounded, we derive

 
 Œ0 
r;n f  f  !0 (1.55)
1

with rates.

1.3 Applications

Let all entities as in Sect. 1.2. We define the following specific operators:
(a) The general multivariate Picard singular integral operators:
1.3 Applications 15

1 X
r
Œm
Œm
Pr;n .f I x1 ; : : : ; xN / WD ˛j;r  (1.56)
.2n /N j D0
!
P
N
Z jsi j
i D1
f .x1 Cs1 j; x2 Cs2 j; ::: ; xN CsN j /e n ds1 : : : dsN :
RN

Observe that !
P
N
Z jsi j
1 i D1

N
e n ds1 : : : dsN D 1; (1.57)
.2n / RN

see [1].
(b) The general multivariate Gauss–Weierstrass singular integral operators:

1 X
r
Œm
Œm
Wr;n .f I x1 ; : : : ; xN / WD p N ˛j;r  (1.58)
n j D0

!
P
N
Z si2
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

Notice that !
P
N
Z si2
1 i D1
p N e n ds1 : : : dsN D 1; (1.59)
RN
n

see [7].
(c) The general multivariate Poisson–Cauchy singular integral operators:

X
r
Œm
Œm
Ur;n .f I x1 ; : : : ; xN / WD WnN ˛j;r  (1.60)
j D0

Z Y
N
1
f .x1 C s1 j; : : : ; xN C sN j /  ˇ ds1 : : : dsN ;
RN i D1 si2˛ C n2˛

with ˛ 2 N, ˇ > 1
2˛
; and

2˛ˇ1
.ˇ/ ˛n
Wn WD    ; (1.61)
1
2˛ ˇ 1
2˛

see [8].
16 1 Uniform Approximation by General Multivariate Singular Integral Operators

Observe that
Z Y
N
1
WnN  ˇ ds1 : : : dsN D 1; (1.62)
RN 2˛
i D1 si C n2˛

see [8, 13], p. 397, formula 595.

(d) The general multivariate trigonometric singular integral operators:

X
r
.f I x1 ; : : : ; xN / WD N
Œm
Œm
Tr;n n ˛j;r  (1.63)
j D0
0   12ˇ
Z Y
N sin si
n
f .x1 Cs1 j; : : : ; xN CsN j / @ A ds1 : : : dsN ;
RN i D1
si

where ˇ 2 N, and

X̌ k 2ˇ1
n WD 2n12ˇ  .1/ˇ ˇ .1/k ; (1.64)
.ˇ  k/Š .ˇ C k/Š
kD1

see [9, 10], p. 210, item 1033.

Notice that
0   12ˇ
Z Y
N sin si
n
N
n
@ A ds1 : : : dsN D 1; (1.65)
RN i D1 si

see also [9, 10], p. 210, item 1033.

Œm Œm Œm Œm
One can apply Theorems 1.9 and 1.11 to operators Pr;n ; Wr;n ; Ur;n , Tr;n
Œm
(special cases of r;n ) and derive interesting results.

References

1. G. Anastassiou, Basic Convergence with rates of smooth Picard singular integral operators,
J. Comput. Anal. Appl, 8(2006), 313-334.
2. G. Anastassiou, General Uniform Approximation Theory by Multivariate Singular Integral
Operators, submitted, 2011.
3. G. Anastassiou and S.S. Dragomir, On some estimates of the remainder in Taylor’s formula,
J. of Math. Anal. and Appl., Vol 263, issue 1, pp. 246-263, (2001).
4. G. Anastassiou and O. Duman, Statistical Approximation by double Picard singular integral
operators, Studia Univ. Babes-Bolyai Math., 55(2010), 3-20.
References 17

5. G. Anastassiou and O. Duman, Uniform Approximation in statistical sense by double

Gauss-Weierstrass singular integral operators, Nonlinear Funct. Anal. Appl. (accepted for
publication, 2009).
6. G. Anastassiou and S. Gal, Approximation Theory, Birkhaüser, Boston, Basel, Berlin, 2000.
7. G. Anastassiou and R. Mezei, Uniform convergence with rates of smooth Gauss-Weierstrass
singular integral operators, Applicable Analysis, 88, 7(2009), 1015-1037.
8. G. Anastassiou and R. Mezei, Uniform convergence with rates of smooth Poisson-Cauchy type
singular integral operators, Mathematical and Computer Modelling, 50(2009), 1553-1570.
9. G. Anastassiou and R. Mezei, Uniform convergence with rates of general singular operators,
CUBO, accepted, 2011.
10. J. Edwards, A treatise on the integral calculus, Vol II, Chelsea, New York, 1954.
11. S.G. Gal, Remark on the degree of approximation of continuous functions by singular integrals,
Math. Nachr., 164(1993), 197-199.
12. R.N. Mohapatra and R.S. Rodriguez, On the rate of convergence of singular integrals for
Hölder continuous functions, Math. Nachr., 149(1990), 117-124.
13. D. Zwillinger, CRC standard Mathematical Tables and Formulae, 30th edition, Chapman &
Hall/CRC, Boca Raton, 1995.
Chapter 2
Lp -Approximation by General Multivariate
Singular Integral Operators

In this chapter, we present the Lp , 1  p < 1 approximation properties of

general multivariate singular integral operators over R N , N  1. We give their
convergence to the unit operator with rates. The established inequalities involve the
multivariate higher order modulus of smoothness. We list the multivariate Picard,
Gauss–Weierstrass, Poisson Cauchy and trigonometric singular integral operators
where this theory can be applied directly. This chapter is based on [3].

2.1 Introduction

The rate of Lp , 1  p < 1 convergence of univariate singular integral operators

has been studied in [1, 7–9], see also the related [11, 12], and these works motivate
the current chapter. Here, we consider some very general multivariate singular
integral operators over RN , N  1, and we study the degree of approximation to the
unit operator with rates over smooth functions. The derived related inequalities are
involving the multivariate higher modulus of smoothness with respect to kkp : See
Theorems 2.4, 2.6, 2.8, 2.10. We mention particular operators that fulfill this theory.
The discussed linear operators are not in general positive, see [2]. Other motivation
comes from [4, 5].

2.2 Main Results

Here r 2 N; m 2 ZC , we define
8  
ˆ
ˆ r
ˆ
< .1/rj
j m ; if j D 1; 2; : : : ; r;
j
Œm
˛j;r WD Xr   (2.1)
ˆ
ˆ r m
:̂ 1  .1/ri
i ; if j D 0;
i
i D1

G.A. Anastassiou, Approximation by Multivariate Singular Integrals, 19

SpringerBriefs in Mathematics, DOI 10.1007/978-1-4614-0589-4 2,
© George A. Anastassiou 2011
20 2 Lp -Approximation by General Multivariate Singular Integral Operators

and
Œm
X
r
Œm
ık;r WD ˛j;r j k ; k D 1; 2; : : : ; m 2 N. (2.2)
j D1

See that
X
r
Œm
˛j;r D 1; (2.3)
j D0

and
X
r    
r r
 .1/ rj
D .1/ r
: (2.4)
j 0
j D1

Let n be a probability Borel measure on RN , N  1, n > 0, n 2 N.

We now define the multiple smooth singular integral operators

X
r Z
Œm
Œm
r;n .f I x1 ; : : : ; xN / WD ˛j;r f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j /dn .s/ ;
j D0 RN
(2.5)

where s WD .s1 ; : : : ; sN /, x WD .x1 ; : : : ; xN / 2 RN I n; r 2 N, m 2 ZC , f W RN ! R

is a Borel measurable function, and also .n /n2N is a bounded sequence of positive
real numbers.
Œm
The operators r;n preserve constants, see [2].
   
Here, we deal with f 2 C m RN , m 2 ZC , with f˛ 2 Lp RN , j˛j D m 2 ZC ,
@jQ f .;:::;/
p  1I where f˛ denotes the mixed partial ˛
@x1 1 :::@xNN
˛ , ˛j 2 ZC , j D 1; : : : ; N W
P
j˛j WD N Q Q
j D1 ˛j D j , j D 1; : : : ; m:
We need
Definition 2.1 (see also [6]).
We call

ru f .x/ WD ru1 ;u2 ;:::;uN f .x1 ; : : : ; xN / (2.6)

Xr  
r
WD .1/rj f .x1 C j u1 ; x2 C j u2 ; : : : ; xN C j uN / :
j
j D0

Let p  1, the modulus of smoothness of order r is given by

 
!r .f I h/p WD sup ru .f /p ; (2.7)
kuk2 h

h > 0.
2.2 Main Results 21

I) First, we consider the case of m 2 N, p > 1:

We make
RemarkP2.2. For jQ D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC , i D 1; : : : ; N;
j˛j WD N Q
i D1 ˛i D j , under the assumption of Theorem 2.4,

Z Y
N
c˛;n;jQ WD c˛;n WD si˛i dn .s1 ; : : : ; sN / 2 R; (2.8)
RN i D1

see also [2].

From [2], we obtain
0 1
B C
X
m B X c f .x/ C
Œm B ˛;n ˛ C
Œm
Er;n .x/ WD Œm
r;n .f I x/  f .x/  ıjQ;r B C (2.9)
B YN C
jQD1 @j˛jDjQ ˛i Š A
i D1
0 1
B C Z Z
X B B 1
C
C Y
N 1
Dm B N C si˛i .1  /m1
BY C RN
j˛jDm @
ŠA
i D1 0
˛ i
i D1
! !
 
 rs f˛ .x/ d dn .s/

DW Rr;n
Œm
.x/ ; 8 x 2 RN : (2.10)

Let p; q > 1 W 1
p
C 1
q
D 1. Then
ˇ ˇp ˇ ˇp
ˇ Œm ˇ ˇ Œm ˇ
ˇEr;n .x/ˇ D ˇRr;n .x/ˇ
0 0 0 11p 1
B B B CC C
B B X B 1 CC C
B B B CC C
Bset c1 WD Bm B N CC C
B B B CC C
@ @ j˛jDm @ Y ˛ Š AA A
i
i D1
ˇZ Z ! ˇp
ˇ Y
N 1  r  ˇ
ˇ ˇ
D c1 ˇ si˛i .1  / m1
s f˛ .x/ d dn .s/ˇ
ˇ RN 0 ˇ
i D1
Z Z ! !p
Y
N
˛i
1
 c1 jsi j .1  / m1
jrs f˛ .x/j d dn .s/ :
RN i D1 0
(2.11)
22 2 Lp -Approximation by General Multivariate Singular Integral Operators

Hence, we have
Z ˇ ˇp
ˇ Œm ˇ
I1 WD ˇEr;n .x/ˇ dx (2.12)
RN

Z Z Z ! !p
Y
N 1
 c1 jsi j ˛i
.1  / m1
jrs f˛ .x/j d dn .s/ dx
RN RN i D1 0

Z !
Y
N 1
Call 0   .s; x/ WD jsi j ˛i
.1  / m1
jrs f˛ .x/j d
i D1 0

Z Z p
D c1  .s; x/ dn .s/ dx DW I2 : (2.13)
RN RN

Therefore, it holds
Z Z 
I 2  c1  .s; x/ dn .s/ dx DW I3 :
p
(2.14)
RN RN

But we have

Y
N Z 1  q  q1 Z 1  p1
p
 .s; x/  jsi j ˛i
.1  / m1
d jrs f˛ .x/j d (2.15)
i D1 0 0

Y
N Z 1  p1
1 p
D jsi j ˛i
1 jrs f˛ .x/j d :
i D1 .q .m  1/ C 1/ q 0

Hence, we obtain

Y
N Z 1 
1
 p .s; x/  jsi j˛i p p jrs f˛ .x/jp d : (2.16)
i D1 .q .m  1/ C 1/ q 0

Thus, we get

Z Z Z  !
Y
N 1
I3  c2 jsi j˛i p jrs f˛ .x/jp d dn .s/ dx (2.17)
RN RN i D1 0
2.2 Main Results 23

!
1
set c2 WD c1  p
.q .m  1/ C 1/ q
Z Z Z ! !
1 Y
N
p
˛i p
D c2 jsi j jrs f˛ .x/j dx d dn .s/
RN 0 RN i D1

Z Y
N Z 1 Z  
˛i p
D c2 jsi j jrs f˛ .x/jp dx d dn .s/ DW I4 : (2.18)
RN i D1 0 RN

Consequently, we derive
Z Y
N Z 1  p 
I4  c2 jsi j˛i p !r .f˛ I  ksk2 /p d dn .s/
RN i D1 0

Z !
Y
N
 c2 jsi j˛i p !r .f˛ I ksk2 /pp dn .s/
RN i D1

Z !  
Y
N
˛i p ksk2 p
D c2 jsi j !r f˛ I n dn .s/ (2.19)
RN n p
i D1
 
by !r .f; h/p  .1 C /r !r .f; h/p ; for any h;  > 0; p  1
Z ! rp !
Y N
ksk
 c2 !r .f˛ I n /pp jsi j˛i p 1C 2
dn .s/ : (2.20)
RN i D1 n

We have proved that

20 0 11 3p

Z 6B B CC !7
ˇ ˇp 6B X B 1 CC 7
ˇ Œm ˇ 6B B CC m 7
ˇEr;n .x/ˇ dx  6B B N CC  7 
6B B CC .q .m  1/ C 1/ 7
4@j˛jDm @ Y ˛
1
RN
Š AA
q
5
i
i D1

Z " !  #p !
Y
N
ksk2 r
˛i
!r .f˛ I n /pp jsi j 1C dn .s/ : (2.21)
RN i D1
n
24 2 Lp -Approximation by General Multivariate Singular Integral Operators

Thus, we derive (p > 1)

0 0 11
!B B CC
  BX B 1 CC
 Œm  m B B CC
Er;n .x/  B B N CC 
B B CC
@j˛jDm @ Y ˛
1
p
.q .m  1/ C 1/
Š AA
q
i
i D1

"Z " ! r #p # p1
Y
N
˛i ksk2
jsi j 1C dn .s/ !r .f˛ I n /p : (2.22)
RN i D1
n

We make
Remark 2.3. Notice that (p > 1)

Z ! r
Y
N
ksk2
jsi j ˛i
1C dn .s/ (2.23)
RN i D1
n

"Z ! r !p # p1
Y
N
ksk2
 jsi j˛i 1C dn .s/ < 1,
RN i D1
n

by assumption of Theorem 2.4.

As in [2] then we get that
Z Y
N
jsi j˛i dn .s/ < 1: (2.24)
RN i D1

Hence, c˛;n;jQ 2 R.
From the above, we have proved
   
Theorem 2.4. Let f 2 C m RN , m 2 N, N  1, with f˛ 2 Lp RN ; j˛j D m,
x 2 RN : Let p; q > 1 W p1 C q1 D 1. Here, n is a Borel probability measure on RN
C
PN sequence. Assume for all ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 Z ,
for n > 0, .n /n2N bounded
i D 1; : : : ; N; j˛j WD i D1 ˛i D m that we have

Z ! r !p
Y
N
ksk2
˛i
jsi j 1C dn .s/ < 1: (2.25)
RN i D1
n
2.2 Main Results 25

For jQ D 1; : : : ; m; and ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 ZC , i D 1; : : : ; N; j˛j WD

P N Q
i D1 ˛i D j , call
Z Y N
c˛;n;jQ WD si˛i dn .s/ : (2.26)
RN i D1

Then
 0 1
 
 
 B C
   X m B X c Qf˛ .x/ C
 Œm   Œm Œm B ˛;n;j C
Er;n  D r;n .f I x/  f .x/  ıjQ;r B ! C (2.27)
 B C
@j˛jDjQ Y ˛ Š A
p N
 jQD1
 i 
 
i D1 p;x
0 1
!B C
BX C
m B 1 C
 B C
.q .m  1/ C 1/
1
B Y
N C
@j˛jDm ŠA
q
˛ i
i D1
"Z " ! r #p # p1
Y
N
˛i ksk2
jsi j 1C dn .s/ !r .f˛ ; n /p :
RN i D1
n
 Œm 
As n ! 1 and n ! 0, by (2.27), we obtain that Er;n  ! 0 with rates.
p
One also finds by (2.27) that
 
 Œm 
r;n .f I x/  f .x/ 
p;x
0 1
m ˇ ˇ ˇ
ˇB C
X ˇ Œm ˇ B X
ˇ ˇ
ˇc˛;n;jQ ˇ C
ˇıjQ;r ˇ B kf˛ kp C C R:H:S: .2.27/;
@ Y
N
A
jQD1 j˛jDjQ ˛i Š
i D1

given that kf˛ kp < 1; j˛j D jQ, jQ D 1; : : : ; m:

 Œm 
Assuming that c˛;n;jQ ! 0, n ! 0; as n ! 1, we get r;n .f /  f p ! 0,
Œm
that is r;n ! I the unit operator, in Lp norm, with rates.
II) Case of m D 0, p > 1.
We make
Remark 2.5. In [2] we proved that
Z
 
Œ0
r;n .f I x/  f .x/ D rs f .x/ dn .s/ : (2.28)
RN
26 2 Lp -Approximation by General Multivariate Singular Integral Operators

Let p; q > 1 W 1
p
C 1
q
D 1. Hence

ˇ ˇp Z p
ˇ Œ0 ˇ
ˇr;n .f I x/  f .x/ ˇ  jrs f .x/j dn .s/ (2.29)
RN
Z
 jrs f .x/jp dn .s/ :
RN

And it holds
Z ˇ ˇp
ˇ Œ0 ˇ
ˇr;n .f I x/  f .x/ˇ dx (2.30)
RN
Z Z 
p
 jrs f .x/j dn .s/ dx
RN RN
Z Z 
p
D jrs f .x/j dx dn .s/ (2.31)
RN RN
Z
 !r .f; ksk2 /pp dn .s/
RN
Z  
ksk2 p
D !r f; n dn .s/ (2.32)
RN n p
Z  
ksk2 rp
 !r .f; n /p
p
1C dn .s/ :
RN n

Therefore, we obtain
 
 Œ0 
r;n .f /  f  (2.33)
p

Z    p1
ksk2 rp
 1C dn .s/ !r .f; n /p :
RN n

We proved
    
Theorem 2.6. Let f 2 C RN \ Lp RN ; N  1; p; q > 1 W p1 C q1 D 1.
Assume n probability Borel measures on RN , .n /n2N > 0 and bounded. Also
suppose
Z  
ksk2 rp
1C dn .s/ < 1: (2.34)
RN n
2.2 Main Results 27

Then
 
 Œ0 
r;n .f /  f  (2.35)
p
Z    p1
ksk2 rp
 1C dn .s/ !r .f; n /p :
RN n
 Œ0 
As n ! 0, when n ! 1, we derive r;n .f /  f p ! 0, i.e. r;n
Œ0
! I , the unit
operator, in Lp norm.
III) Next follows the case m D 0, p D 1.
We make
Remark 2.7. As before we have
ˇ ˇ Z
ˇ Œ0 ˇ
ˇr;n .f I x/  f .x/ˇ  jrs f .x/j dn .s/ : (2.36)
RN

Hence
  Z ˇ ˇ
 Œ0  ˇ Œ0 ˇ
r;n .f /  f  D ˇr;n .f I x/  f .x/ˇ dx (2.37)
1 RN

Z Z 
 jrs f .x/j dn .s/ dx (2.38)
RN RN

Z Z 
D jrs f .x/j dx dn .s/
RN RN

Z
 !r .f; ksk2 /1 dn .s/
RN

Z  
ksk2
D !r f; n dn .s/ (2.39)
RN n 1
Z  
ksk2 r
 !r .f; n /1 1C dn .s/ :
RN n

We have proved
    
Theorem 2.8. Let f 2 C RN \ L1 RN , N  1: Assume n probability
Borel measures on RN , .n /n2N > 0 and bounded. Also suppose
Z  
ksk2 r
1C dn .s/ < 1: (2.40)
RN n
28 2 Lp -Approximation by General Multivariate Singular Integral Operators

Then
 
 Œ0 
r;n .f /  f  (2.41)
1
Z   
ksk2 r
 1C dn .s/ !r .f; n /1 :
RN n

As n ! 0, we get r;n
Œ0
! I in L1 norm.
IV) Case of m 2 N, p D 1.
We make
Remark 2.9. We have
0 1

Z B C
  ˇ ˇ X B C
 Œm  ˇ Œm ˇ B 1 C
Er;n  D ˇEr;n .x/ˇ dx  m B N C
1 RN BY C
j˛jDm @
˛ i ŠA
i D1

Z Z Z ! !
Y
N
˛i
1
m1
jsi j .1  / jrs f˛ .x/j d dn .s/ dx
RN RN i D1 0
(2.42)

0 0 11
B B CC
B X B CC
B B 1 CC
Bset c1 WD m B N CC
B BY CC
@ j˛jDm @
˛ ŠA A
i
i D1

Z Y
N Z 1 Z  
˛i
D c1 jsi j .1  / m1
jrs f˛ .x/j dx d dn .s/
RN i D1 0 RN

Z ! Z 
Y
N
˛i
1
 c1 jsi j .1  /m1 !r .f˛ ;  ksk2 /1 d dn .s/
RN i D1 0

(2.43)
Z !
c1 Y
N
 jsi j ˛i
!r .f˛ ; ksk2 /1 dn .s/
m RN i D1
2.2 Main Results 29

0 1
B CZ !  
X B C Y
N
ksk2
B 1 C ˛i
D B N C jsi j !r f˛ ; n dn .s/ (2.44)
BY C RN n 1
j˛jDm @
ŠA
i D1
˛ i
i D1
0 1
B C Z ! r
X B C Y
N
ksk2
B 1 C ˛i
 B N C !r .f˛ ; n /1 jsi j 1C dn .s/ :
BY C RN n
j˛jDm @
ŠA
i D1
˛ i
i D1
(2.45)

We have proved
   
Theorem 2.10. Let f 2 C m RN , m; N 2 N, with f˛ 2 L1 RN ; j˛j D m,
x 2 RN : Here, n is a Borel probability measure on RN for n > 0, .n /n2N is
C
P sequence. Suppose for all ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 Z , i D 1; : : : ; N;
a bounded
j˛j WD N ˛
i D1 i D m that we have

Z ! r !
Y
N
˛i ksk2
jsi j 1C dn .s/ < 1: (2.46)
RN i D1
n

Q C
PN j D 1; : : : ; m; and ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 Z , i D 1; : : : ; N; j˛j WD
For
Q
i D1 ˛i D j , call
Z Y N
c˛;n;jQ WD si˛i dn .s/ : (2.47)
RN i D1

Then
 0 1
 
 
 B C
   m B
X Œm B X c˛;n;jQ f˛ .x/ C C
 Œm   Œm 
Er;n  D r;n .f I x/  f .x/  ıjQ;r B C (2.48)
1  B Y N C
 @j˛jDjQ 
 jQD1
˛i Š A
 
i D1 1;x
0 1
B C Z ! r
X B B 1
C
C Y
N
˛i ksk2
 B N C !r .f˛ ; n /1 jsi j 1C dn .s/ :
BY C RN n
j˛jDm @
ŠA
i D1
˛ i
i D1

 Œm 
As n ! 0, we get Er;n  ! 0 with rates.
1
30 2 Lp -Approximation by General Multivariate Singular Integral Operators

From (2.48) we get

 
 Œm 
r;n f  f 
1
0 1
B ˇ ˇ C
m ˇ ˇB ˇ ˇ C
X ˇ Œm ˇ B X ˇc˛;n;jQ ˇ C
 ˇıjQ;r ˇ B kf˛ k1 C C R:H:S: .2:48/;
B C
@j˛jDjQ Y ˛ Š
N
jQD1 A
i
i D1

given that kf˛ k1 < 1; j˛j D jQ, jQ D 1; : : : ; m:  Œm 

As n ! 1, assuming n ! 0 and c˛;n;jQ ! 0, we obtain r;n f  f 1 ! 0,
Œm
that is r;n ! I in L1 norm, with rates.

2.3 Applications

Let all entities as in Sect. 2.2. We define the following specific operators:
(a) The general multivariate Picard singular integral operators:

1 X
r
Œm
Œm
Pr;n .f I x1 ; : : : ; xN / WD ˛j;r  (2.49)
.2n /N j D0
!
P
N
Z jsi j
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

Notice that !
P
N
Z jsi j
1 i D1

N
e n ds1 : : : dsN D 1; (2.50)
.2n / RN

see [1].
(b) The general multivariate Gauss–Weierstrass singular integral operators:

1 X
r
Œm
Œm
Wr;n .f I x1 ; : : : ; xN / WD p N ˛j;r  (2.51)
n j D0

!
P
N
Z si2
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN
2.3 Applications 31

Observe that !
P
N
Z si2
1 i D1
p N e n ds1 : : : dsN D 1; (2.52)
RN
n

see [7].
(c) The general multivariate Poisson–Cauchy singular integral operators:

X
r
Œm
Œm
Ur;n .f I x1 ; : : : ; xN / WD WnN ˛j;r  (2.53)
j D0

Z Y
N
1
f .x1 C s1 j; : : : ; xN C sN j /  ˇ ds1 : : : dsN ;
RN i D1 si2˛ C n2˛

with ˛ 2 N, ˇ > 1
2˛
; and

2˛ˇ1
.ˇ/ ˛n
Wn WD    ; (2.54)
1
2˛ ˇ 1
2˛

see [8].
Notice that
Z Y
N
1
WnN  ˇ ds1 : : : dsN D 1; (2.55)
RN i D1 si2˛ C n2˛

see [8, 13], p. 397, formula 595.

(d) The general multivariate trigonometric singular integral operators:

X
r
.f I x1 ; : : : ; xN / WD N
Œm
Œm
Tr;n n ˛j;r  (2.56)
j D0
0   12ˇ
Z Y
N sin si
n
f .x1 C s1 j; : : : ; xN C sN j / @ A ds1 : : : dsN ;
RN i D1
si

where ˇ 2 N, and

X̌ k 2ˇ1
n WD 2n12ˇ .1/ˇ ˇ .1/k ; (2.57)
.ˇ  k/Š .ˇ C k/Š
kD1

see [9, 10], p. 210, item 1033.

32 2 Lp -Approximation by General Multivariate Singular Integral Operators

Observe that
0   12ˇ
Z Y
N sin si
n
N
n
@ A ds1 : : : dsN D 1; (2.58)
RN i D1 si

see also [9, 10], p. 210, item 1033.

Œm Œm
One can apply Theorems 2.4, 2.6, 2.8, and 2.10 to operators Pr;n ; Wr;n ;
Œm Œm Œm
Ur;n , Tr;n (special cases of r;n ) and derive interesting results.

References

1. G. Anastassiou, Lp convergence with rates of smooth Picard singular operators, Differential

and difference equations and applications, Hindawi Publ. Corp., New York, (2006), 31-45.
2. G. Anastassiou, General uniform Approximation Theory by Multivariate singular integral
operators, submitted, 2011.
3. G. Anastassiou, Lp -general Approximations by Multivariate Singular Integral Operators,
submitted, 2011.
4. G. Anastassiou and O. Duman, Statistical Lp -approximation by double Gauss-Weierstrass
singular integral operators, Comput. Math. Appl. 59(2010), 1985-1999.
5. G. Anastassiou and O. Duman, Statistical Lp -convergence of double smooth Picard singular
integral operators, J. Comput. Anal. Appl., Vol. 13, No. 1(2011), 167-187.
6. G. Anastassiou and S. Gal, Approximation Theory, Birkhaüser, Boston, Basel, Berlin, 2000.
7. G. Anastassiou and R. Mezei, Lp convergence with rates of smooth Gauss-Weierstrass singular
operators, Nonlinear Studies, accepted, 2011.
8. G. Anastassiou and R. Mezei, Lp convergence with rates of smooth Poisson-Cauchy type
singular operators, International Journal of Mathematical Sciences, accepted, 2010.
9. G. Anastassiou and R. Mezei, Lp convergence with rates of general singular integral operators,
submitted, 2010.
10. J. Edwards, A treatise on the integral calculus, Vol II, Chelsea, New York, 1954.
11. S.G. Gal, Remark on the degree of approximation of continuous functions by singular integrals,
Math. Nachr., 164(1993), 197-199.
12. R.N. Mohapatra and R.S. Rodriguez, On the rate of convergence of singular integrals for
Hölder continuous functions, Math. Nachr., 149(1990), 117-124.
13. D. Zwillinger, CRC standard Mathematical Tables and Formulae, 30th edition, Chapman &
Hall/CRC, Boca Raton, 1995.
Chapter 3
Global Smoothness Preservation
and Simultaneous Approximation
by Multivariate General Singular Integrals

In this chapter, we continue with the study of multivariate smooth general singular
integral operators over RN , N  1, regarding their simultaneous global smoothness
preservation property with respect to the Lp norm, 1  p  1, by involving
multivariate higher order moduli of smoothness. Also, we present their multivariate
simultaneous approximation to the unit operator with rates. The derived multivariate
Jackson-type inequalities are almost sharp containing elegant constants, and they
reflect the high order of differentiability of the engaged function. In the uniform case
of global smoothness we prove optimality. At the end we list the multivariate Picard,
Gauss–Weierstrass, Poisson–Cauchy and Trigonometric singular integral operators
as applicators of this general theory. This chapter relies on [4].

3.1 Introduction

The main motivation for this chapter comes from [1, 5, 6]. We give here the
multivariate simultaneous global smoothness preservation property of multivariate
general smooth singular integral operators. We study also the simultaneous Lp ,
1  p  1, approximation of these operators to the unit operator with rates.
See Theorems 3.2, 3.4, Proposition 3.5, Theorem 3.8, Corollary 3.9 and Theorems
3.10–3.15. At the end, we list specific operators that fulfill our theory. One can find
many interesting convergence properties based on these results.

G.A. Anastassiou, Approximation by Multivariate Singular Integrals, 33

SpringerBriefs in Mathematics, DOI 10.1007/978-1-4614-0589-4 3,
© George A. Anastassiou 2011
34 3 Global Smoothness and Simultaneous Approximation by Singular Integrals

3.2 Main Results

Here r 2 N; m 2 ZC , we define
8  
ˆ
ˆ r
ˆ
< .1/
rj
j m ; if j D 1; 2; : : : ; r;
j
˛j;r
Œm
WD X r   (3.1)
ˆ
ˆ1  r m
:̂ .1/ri i ; if j D 0;
i
i D1

and
Œm
X
r
Œm
ık;r WD ˛j;r j k ; k D 1; 2; : : : ; m 2 N. (3.2)
j D1

See that
X
r
Œm
˛j;r D 1; (3.3)
j D0

and
X
r    
r r
 .1/rj D .1/r : (3.4)
j 0
j D1

Let n be a probability Borel measure on RN , N  1, n > 0, n 2 N.

We now define the multiple smooth singular integral operators

X
r Z
Œm
Œm
r;n .f I x1 ; : : : ; xN / WD ˛j;r f .x1 C s1 j; x2 C s2 j; : : : ;
j D0 RN

xN C sN j / dn .s/ ; (3.5)

where s WD .s1 ; : : : ; sN /, x WD .x1 ; : : : ; xN / 2 RN I n; r 2 N, m 2 ZC , f W RN ! R

is a Borel measurable function, and also .n /n2N is a bounded sequence of positive
real numbers.
Œm
Above operators r;n are not in general positive operators and they preserve
constants, see [2].
 
Definition 3.1. Let f 2 C RN , N  1, m 2 N, the mth modulus of smoothness
for 1  p  1, is given by

!m .f I h/p WD sup km

t .f /kp;x ; (3.6)
kt k2 h
3.2 Main Results 35

h > 0, where
X
m  
mj m
m
t f .x/ WD .1/ f .x C jt/ : (3.7)
j
j D0

Denote
!m .f I h/1 D !m .f; h/ : (3.8)
Above, x; t 2 RN :
We present the general global smoothness preservation result
Q
 
Œm
Theorem 3.2. We assume r;n .f I x/ 2 R, 8 x 2 R. Let h > 0, f 2 C RN ,
N  1.
i) Assume !m .f; h/ < 1. Then
0 1
  r ˇ
X ˇ
ˇ ˇ
!m r;nQ f; h  @ ˇ˛jQ;r ˇA !m .f; h/ :
Q
Œm  Œ m 
(3.9)
jQD0
    
ii) Assume f 2 C RN \ L1 RN : Then
0 1
  r ˇ
X ˇ
ˇ ˇ
Q
f; h  @ ˇ˛jQ;r ˇA !m .f; h/1 :
Q
Œm Œ m 
!m r;n (3.10)
1
jQD0
    
iii) Assume f 2 C RN \ Lp RN , p > 1: Then
0 1
  r ˇ
X ˇ
ˇ Œ mQ  ˇA
ŒmQ
!m r;n f; h  @ ˇ˛jQ;r ˇ !m .f; h/p : (3.11)
p
jQD0

Proof. We recall (x 2 RN , N  1)
X
r Z
 
f x C s jQ dn .s/ :
Q ΠQ
Œm
r;n .f I x/ D ˛jQm
;r
RN
jQD0

We see (t 2 RN )
  Xm  
Q m Q
m
t  Œm
r;n .f I x/ D .1/mj Œm
r;n .f I x C jt/ (3.12)
j
j D0
0 1
X
m  Z Xr
m  
@ f x C jt C jQk A dn.k/
ΠQ
D .1/mj ˛jQm
;r
j RN
j D0 jQD0
(3.13)
36 3 Global Smoothness and Simultaneous Approximation by Singular Integrals

0 1
X
r Z X
m  
m  
D
Œ
˛jQmQ  @ .1/mj f xCjtCjQk A dn .k/
;r j
RN j D0
jQD0

X Z
r
  
D
Œ
˛jQmQ  Q
t f x C j k dn .k/ :
m (3.14)
;r
RN
jQD0

That is we find
  Xr Z
 m  
Q Q
t f x C jQk dn .k/ :
Œm
m
t r;n .f I x/ D
Œm
˛jQ;r (3.15)
RN
jQD0

Therefore
ˇ  ˇ X r ˇ ˇZ ˇ m  ˇ
ˇ m Œ mQ  ˇ ˇ Œ mQ  ˇ ˇ f x C jQk ˇ d .k/ :
ˇt r;n .f I x/ ˇ  ˇ˛jQ;r ˇ t n (3.16)
RN
jQD0

(a) From the last (3.16), we derive

  Xr ˇ ˇZ
Q ˇ Œ mQ  ˇ
Œm
!m r;n f; h  ˇ˛jQ;r ˇ !m .f; h/ dn .k/ (3.17)
RN
jQD0
0 1
r ˇ
X ˇ
ˇ Q ˇ
D@ ˇ˛jQ;r ˇA !m .f; h/ ;
Œm

jQD0

proving the claim (3.9).

(b) We see that

Z ˇ  ˇ r ˇ
X ˇZ Z 
ˇ m Œ mQ  ˇ ˇ Œ mQ  ˇ ˇ m  ˇ
ˇt r;n .f I x/ ˇ dx  ˇ˛jQ;r ˇ ˇ f x C jQk ˇ dx
t
RN RN RN
jQD0

 dn .k/
r ˇ
X ˇZ
ˇ Œ mQ  ˇ
D ˇ˛jQ;r ˇ t f k1 dn .k/
km
RN
jQD0
0 1
r ˇ
X ˇ
ˇ Œ mQ  ˇA m
D@ ˇ˛jQ;r ˇ kt f k1 : (3.18)
jQD0
3.2 Main Results 37

So we derive
0 1
   r ˇ
X ˇ
 m Œ mQ   ˇ Œm
Q ˇ
t r;n .f /   @ ˇ˛jQ;r ˇA km
t f k1 ; (3.19)
1
jQD0

which implies
0 1
  r ˇ
X ˇ
ˇ Œ mQ  ˇA
ŒmQ
!m r;n f; h  @ ˇ˛jQ;r ˇ !m .f; h/1 ; (3.20)
1
jQD0

proving the claim (3.10).

(c) Let p; q > 1 W p1 C q1 D 1. Then

   r ˇ ˇ Z 
 m Œ mQ   X ˇ Œ mQ  ˇ  ˇ m  ˇ 
t r;n .f I x/   ˇ˛jQ;r ˇ  ˇ f x C jQk ˇ d .k/
p;x  t n 
RN p;x
jQD0
(3.21)
r ˇ
X ˇ Z Z
ˇ m  ˇ
ˇ Œ mQ  ˇ ˇ f x C jQk ˇ
D ˇ˛jQ;r ˇ t
RN RN
jQD0

p  p1
 dn .k/ dx (3.22)

r ˇ
X ˇ Z Z
ˇ m  ˇ
ˇ Œ mQ  ˇ ˇ f x C jQk ˇp
 ˇ˛jQ;r ˇ t
RN RN
jQD0

  p1
 dn .k/ dx

r ˇ
X ˇ Z Z 
ˇ Œ mQ  ˇ ˇ m  ˇ
D ˇ˛jQ;r ˇ ˇ f x C jQk ˇp dx
t
RN RN
jQD0

 p1
 dn .k/

r ˇ
X ˇ Z  p1
ˇ Œ mQ  ˇ
D ˇ˛jQ;r ˇ km
t f kpp dn .k/
RN
jQD0
0 1
r ˇ
X ˇ
ˇ Q ˇ
D@ ˇ˛jQ;r ˇA km
Πm 
t f kp : (3.23)
jQD0
38 3 Global Smoothness and Simultaneous Approximation by Singular Integrals

That is 0 1
   r ˇ
X ˇ
 m Œ mQ   ˇ Œm
Q ˇ
t r;n .f /   @ ˇ˛jQ;r ˇA km
t f kp : (3.24)
p
jQD0

Now clearly (3.24) proves (3.11). t

u
Œm Œm
Remark 3.3. Let r D 1, then ˛0;1 D 0, ˛1;1 D 1. Hence
Z
1;nmQ .f I x/ D
Π
f .x C s/ dn .s/ DW n .f I x/ : (3.25)
RN

By Theorem 3.2, we get

 
Theorem 3.4. We suppose n .f I x/ 2 R, 8 x 2 R. Let h > 0, f 2 C RN ,
N  1.
i) Assume !m .f; h/ < 1. Then

!m .n f; h/  !m .f; h/ : (3.26)

    
ii) Assume f 2 C RN \ L1 RN : Then

!m .n f; h/1  !m .f; h/1 : (3.27)

    
iii) Assume f 2 C RN \ Lp RN , p > 1: Then

!m .n f; h/p  !m .f; h/p : (3.28)

Next, we get an optimality result

Proposition 3.5. Above inequality (3.26):

!m .n f; h/  !m .f; h/

is sharp, namely it is attained by any

 
fj .x/ D xjm ; j D 1; : : : ; N; x D x1 ; : : : ; xj ; : : : ; xN 2 RN : (3.29)

Proof. We observe that (for x; t 2 RN )

X
m  
 m  m
t fj .x/ D
m .1/mi xj C i tj (3.30)
i
i D0

D mŠtjm ; j D 1; : : : ; N:
3.2 Main Results 39

So that   ˇ ˇm ˇ ˇm
!m fj ; h D sup mŠ ˇtj ˇ D mŠ sup ˇtj ˇ D mŠhm : (3.31)
kt k2 h kt k2 h

Also
   Z  
 
m
t n fj I x D m f
t j .x C s/ dn .s/ (3.32)
RN
Z
D mŠtjm dn .s/ D mŠtjm , j D 1; : : : ; N; etc,
RN

proving the claim. t

u
We need
 
Theorem 3.6. Let f 2 C l RN , l; N 2 N. Here, n is a Borel probability
measure on RN ; n > 0, .n /n2N P a bounded sequence. Let ˇ WD .ˇ1 ; : : : ; ˇN /,
ˇi 2 ZC , i D 1; : : : ; N I jˇj WD N
i D1 ˇi D l: Here f .x C sj /, x; s 2 R , is
N

n -integrable wrt s, for j D 1; : : : ; r: There exist n -integrable functions hi1 ;j ;
hˇ1 ;i2 ;j ; hˇ1 ;ˇ2 ;i3 ;j ; : : : ; hˇ1 ;ˇ2 ;:::;ˇN 1 ;iN ;j  0 (j D 1; : : : ; r) on RN such that
ˇ ˇ
ˇ @i1 f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hi1 ;j .s/ ; i1 D 1; : : : ; ˇ1 ; (3.33)
ˇ i1
@x1 ˇ
ˇ ˇ
ˇ @ˇ1 Ci2 f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hˇ1 ;i2 ;j .s/ ; i2 D 1; : : : ; ˇ2 ;
ˇ i2
@x @x
ˇ1 ˇ
2 1

::
:
ˇ ˇ
ˇ @ˇ1 Cˇ2 C:::CˇN 1 CiN f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hˇ1 ;ˇ2 ;:::;ˇN 1 ;iN ;j .s/ ; iN D 1; : : : ; ˇN ;
ˇ @x iN @x ˇN 1 : : : @x ˇ2 @x ˇ1 ˇ
N N 1 2 1

8x; s 2 RN .
Then, both of the next exist and
   
Q Q
Œm
r;n .f I x/ D r;n
Œm
fˇ I x : (3.34)
ˇ

Proof. By H. Bauer [7], pp. 103–104. t

u
Corollary 3.7. (to Theorem 3.6, r D 1) It holds
 
.n .f I x//ˇ D n fˇ I x : (3.35)

We present simultaneous global smoothness results.

40 3 Global Smoothness and Simultaneous Approximation by Singular Integrals

Theorem 3.8. Let h > 0 and assumptions of Theorem 3.6 are valid. Here  D 0; ˇ,
(0 D .0; : : : ; 0/).
 
i) Assume !m f ; h < 1. Then
0 1
   r ˇ
X ˇ
ˇ ˇ  
!m r;nQ .f / ; h  @ ˇ˛jQ;r ˇA !m f ; h :
Q
Œm  Œ m 
(3.36)

jQD0

 
ii) Additionally suppose f 2 L1 RN : Then
0 1
   r ˇ
X ˇ
ˇ Œ mQ  ˇA  
!m ŒmQ
r;n .f / ;h  @ ˇ˛jQ;r ˇ !m f ; h 1 : (3.37)
 1
jQD0

 
iii) Additionally suppose f 2 Lp RN , p > 1: Then
0 1
   r ˇ
X ˇ
ˇ ˇ  
Q
.f / ; h  @ ˇ˛jQ;r ˇA !m f ; h p :
Q
Œm Œm
!m r;n (3.38)
 p
jQD0

It follows
Corollary 3.9. Let h > 0 and assumptions of Corollary 3.7 are valid. Here
 D 0; ˇ.
 
i) Assume !m f ; h < 1. Then
   
!m .n .f // ; h  !m f ; h : (3.39)
 
ii) Additionally assume f 2 L1 RN : Then
   
!m .n .f // ; h 1  !m f ; h 1 : (3.40)
 
iii) Additionally assume f 2 Lp RN , p > 1: Then
   
!m .n .f // ; h p  !m f ; h p : (3.41)

Next comes multi-simultaneous approximation.

 
Theorem 3.10. Let f 2 C mCl RN , m; l;N 2 N. The assumptions of Theorem
3.6 are valid. Call  D 0; ˇ. Assume f C˛ 1 < 1 and

Z ! r
Y
N
˛i ksk2
jsi j 1C dn .s/ < 1;
RN i D1
n
3.2 Main Results 41

PN
for all ˛j 2 ZC , j D 1; : : : ; N; j˛j WD j D1 ˛j D m, where n is a Borel
probability measure on R , for n > 0, .n /n2N bounded sequence.
N
Q C
PNFor j D 1; : : : ; m; and ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 Z , i D 1; : : : ; N; j˛j WD
Q
i D1 ˛i D j , call
Z Y N
c˛;n;jQ WD si˛i dn .s/ : (3.42)
RN i D1

Then
 0 1
 
 
 B C 

 Œm
 X Œm B X c˛;n;jQ f C˛ ./ C
m B C

 r;n .f I /  f ./  ıjQ;r B C (3.43)
  B Y N C
 Q
j D1 @ ˛ ;:::;˛ 0W A 
 
1 N
j˛jDjQ ˛i Š
 
i D1 1
   Z ! r !
X !r f C˛ ; n Y ˛
N
ksk2
 ! jsi j i
1C dn .s/ :
0 1 Y
N
RN i D1 n
@ A ˛i Š
˛1 ;:::;˛N 0
j˛jDm i D1

Proof. Based on Theorems 6 and 9 of [2]. t

u
 N
Theorem 3.11. Let f 2 CB R , l; N 2 N (functions l-times continuously
l

differentiable and bounded). The assumptions of Theorem 3.6 are valid. Call
 D 0; ˇ. Suppose
Z  
ksk2 r
1C dn .s/ < 1: (3.44)
RN n
Then
  Z   
 Œ0   ksk2 r  
  f  f   1C dn .s/ !r f ; n : (3.45)
 r;n   n
1 RN

Proof. By Theorems 6 and 11 of [2]. t

u
We continue with
 
Theorem 3.12. Let f 2 C mCl RN ; m; l; N 2 N: The assumptions of Theorem
3.6 are valid. Call  D 0; ˇ. Let f. C˛/ 2 Lp RN , j˛j D m, x 2 RN , and p; q >
1 W p1 C q1 D 1: Here, n is a Borel probability measure on RN for n > 0, .n /n2N
bounded Psequence. Assume for all ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC , i D 1; : : : ; N;
N
j˛j WD i D1 ˛i D m we have that

Z ! r !p
Y
N
˛i ksk2
jsi j 1C dn .s/ < 1: (3.46)
RN i D1
n
42 3 Global Smoothness and Simultaneous Approximation by Singular Integrals

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 Z , i D 1; : : : ; N; j˛j WD
For
Q
i D1 ˛i D j , call
Z Y N
c˛;n;jQ WD si˛i dn .s/ : (3.47)
RN i D1

Then
 0 1
 
 
 B C
  X m B X c Q f C˛ .x/ C
 Œm Œm B ˛;n;j C
 r;n .f I x/  f .x/  ıjQ;r B C
  B Y N C
 jQD1 @j˛jDjQ A
 ˛i Š 
 
i D1 p;x
0 1
!B C
BX C
mB 1 C
 B ! C (3.48)
.q .m  1/ C 1/
1
B Y
N C
@j˛jDm ˛i Š A
q

i D1
"Z " ! r #p # p1
Y
N
˛i ksk2  
jsi j 1C dn .s/  !r f C˛ ; n p :
RN i D1
n

Proof. By Theorems 6 and 4 of [3]. t

u
We give also
 
Theorem 3.13. Let f 2 C l RN ; l; N 2 N: The assumptions of Theorem 3.6 are
 
valid. Call  D 0; ˇ. Let f 2 Lp RN , x 2 RN ; p; q > 1 W p1 C q1 D 1: Assume
n probability Borel measures on RN ; .n /n2N > 0 and bounded. Also suppose
Z  
ksk2 rp
1C dn .s/ < 1:
RN n

Then
  Z    p1
 Œ0   ksk2 rp  
  f  f   1C dn .s/ !r f ; n p : (3.49)
 r;n   n
p RN

Proof. By Theorems 6 and 6 of [3]. t

u
 
Theorem 3.14. Let f 2 C l RN ; l; N 2 N: The assumptions of Theorem 3.6 are
valid. Call  D 0; ˇ. Let f 2 L1 RN , x 2 RN : Assume n probability Borel
measures on RN ; .n /n2N > 0 and bounded. Also assume
3.2 Main Results 43

Z  
ksk2 r
1C dn .s/ < 1:
RN n

Then
   Z   
 Œ0  ksk2 r  
  .f /  f   1C dn .s/ !r f ; n 1 : (3.50)
 r;n 
 n
1 RN

Proof. By Theorems 6 and 8 of [3]. t

u
 
Theorem 3.15. Let f 2 C mCl RN , m; l; N 2 N.  The assumptions of Theorem
3.6 are valid. Call  D 0; ˇ. Let f. C˛/ 2 L1 RN ; j˛j D m, x 2 RN . Here, n
is a Borel probability measure on RN for n > 0, .n /n2N is a bounded P sequence.
Suppose for all ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC , i D 1; : : : ; N; j˛j WD N
i D1 ˛i D m;
we have that
Z ! r !
Y N
˛i ksk2
jsi j 1C dn .s/ < 1: (3.51)
RN i D1
n

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 Z , i D 1; : : : ; N; j˛j WD
For
Q
i D1 ˛i D j , call
Z Y N
c˛;n;jQ WD si˛i dn .s/ : (3.52)
RN i D1

Then
 0 1
 
 
 B C 
  X m B X c Q f. C˛/ .x/ C
 Œm Œm B ˛;n;j C
 r;n .f I x/  f .x/  ıjQ;r B C (3.53)
  B Y N C
 jQD1 @j˛jDjQ A
 ˛i Š 
  i D1 1;x
0 1
B C Z ! r
XB C   Y
N
ksk2
B 1 C ˛i
 B N C !r f C˛ ; n 1 jsi j 1C dn .s/ :
BY C RN n
j˛jDm @
ŠA
i D1
˛ i
i D1

Proof. Based on Theorem 6 and Theorem 10 of [3]. t

u
44 3 Global Smoothness and Simultaneous Approximation by Singular Integrals

3.3 Applications

Let all entities as in Sect. 3.2. We define the following specific operators:
(a) The general multivariate Picard singular integral operators:

1 X
r
Œm
Œm
Pr;n .f I x1 ; : : : ; xN / WD ˛j;r  (3.54)
.2n /N j D0
!
P
N
Z jsi j
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

(b) The general multivariate Gauss–Weierstrass singular integral operators:

1 X
r
Œm
Œm
Wr;n .f I x1 ; : : : ; xN / WD p N ˛j;r  (3.55)
n j D0

!
P
N
Z si2
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

(c) The general multivariate Poisson–Cauchy singular integral operators:

X
r
Œm
Œm
Ur;n .f I x1 ; : : : ; xN / WD WnN ˛j;r  (3.56)
j D0

Z Y
N
1
f .x1 C s1 j; : : : ; xN C sN j /  ˇ ds1 : : : dsN ;
RN i D1 si2˛ C n2˛

with ˛ 2 N, ˇ > 1
2˛
; and
2˛ˇ1
.ˇ/ ˛n
Wn WD    : (3.57)
1
2˛ ˇ 1
2˛

(d) The general multivariate trigonometric singular integral operators:

X
r
N Œm
Œm
Tr;n .f I x1 ; : : : ; xN / WD n ˛j;r  (3.58)
j D0

0   12ˇ
Z Y
N sin si
n
f .x1 C s1 j; : : : ; xN C sN j / @ A ds1 : : : dsN ;
RN i D1
si
References 45

where ˇ 2 N, and

X̌ k 2ˇ1
n WD 2n12ˇ  .1/ˇ ˇ .1/k : (3.59)
.ˇ  k/Š .ˇ C k/Š
kD1

Œm Œm Œm Œm

One can apply the results of this chapter to the operators Pr;n ; Wr;n ; Ur;n , Tr;n
Œm
(special cases of r;n ) and obtain interesting results.

3.4 Conclusion

The approximation results here imply important convergence properties of operators

Œm
r;n to the unit operator.

References

1. G. Anastassiou, Global smoothness and uniform convergence of smooth Picard singular

operators, Comput. Math. Appl. 50(2005), no. 10-12, 1755-1766.
2. G. Anastassiou, General uniform Approximation theory by multivariate singular integral
operators, submitted, 2011.
3. G. Anastassiou, Lp -general approximations by multivariate singular integral operators, sub-
mitted, 2011.
4. G. Anastassiou, Global Smoothness Preservation and Simultaneous Approximation for Multi-
variate General Singular Integral Operators, Appl. Math. Letters, 24(2011), 1009-1016.
5. G. Anastassiou, R. Mezei, Global smoothness and uniform convergence of smooth Gauss-
Weierstrass singular operators, Mathematical and Computer Modelling, 50(2009), 984-998.
6. G. Anastassiou, R. Mezei, Global smoothness and uniform convergence of smooth
Poisson-Cauchy type singular operators, Applied Mathematics and Computation, 215(2009),
1718-1731.
7. H. Bauer, M˛ˇ-und Integrations theorie, de Gruyter, Berlin, 1990.
Chapter 4
Multivariate Voronovskaya Asymptotic
Expansions for General Singular Integrals

In this chapter, we continue with the study of approximation properties of smooth

general singular integral operators over RN , N  1. We present multivariate
Voronovskaya asymptotic type results and give quantitative results regarding the
rate of convergence of multivariate singular integral operators to unit operator.
We list specific multivariate singular integral operators that fulfill this theory. This
chapter is based on [2].

4.1 Introduction

The main motivation for this chapter comes from [3–5]. We give here multivariate
Voronovskaya-type asymptotic expansions regarding the multivariate singular inte-
gral operators, see Theorem 4.2 and Corollaries 4.3, 4.4. In Theorem 4.6, we present
the simultaneous corresponding Voronovskaya asymptotic expansion for these
operators. The expansions give also the rate of convergence of multivariate general
singular integral operators to unit operator. In Sect. 4.3, we list the multivariate
singular Picard, Gauss–Weierstrass, Poisson–Cauchy and Trigonometric operators
that fulfill the main results.

4.2 Main Results

Here r 2 N; m 2 ZC , we define
8  
ˆ
ˆ r
ˆ
< .1/
rj
j m ; if j D 1; 2; : : : ; r;
j
Œm
˛j;r WD X r   (4.1)
ˆ
ˆ1  r m
:̂ .1/ri i ; if j D 0;
i
i D1

G.A. Anastassiou, Approximation by Multivariate Singular Integrals, 47

SpringerBriefs in Mathematics, DOI 10.1007/978-1-4614-0589-4 4,
© George A. Anastassiou 2011
48 4 Voronovskaya Asymptotic Expansions for General Singular Integrals

and
Œm
X
r
Œm
ık;r WD ˛j;r j k ; k D 1; 2; : : : ; m 2 N. (4.2)
j D1

Observe that
X
r
Œm
˛j;r D 1; (4.3)
j D0

and
X
r    
r r
 .1/rj D .1/r : (4.4)
j 0
j D1

Let n be a probability Borel measure on RN , N  1, n > 0, n 2 N.

We now define the multiple smooth singular integral operators

X
r Z
Œm
Œm
r;n .f I x1 ; : : : ; xN / WD ˛j;r f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j /
j D0 RN

 dn .s/ ; (4.5)

where s WD .s1 ; : : : ; sN /, x WD .x1 ; : : : ; xN / 2 RN I n; r 2 N, m 2 ZC , f W RN ! R

is a Borel measurable function, and also .n /n2N is a bounded sequence of positive
real numbers.
Œm
The above r;n are not in general positive operators and they preserve constants,
see [1].
We make
 
Remark 4.1. Here f 2 C m RN , m; N 2 N. Let l D 0; 1; : : : ; m. The lth order
˛
partial derivative is denoted by f˛ WD @@xf˛ , where ˛ WD .˛1 ; : : : ; ˛N /, ˛i 2 ZC ,
P
i D 1; : : : ; N and j˛j WD N i D1 ˛i D l:
Consider gz .t/ WD f .x0 C t .z  x0 //, t  0; x0 ; z 2 RN .
Then
2 !j 3
X N
@
gz.j / .t/ D 4 .zi  x0i / f 5 .x01 C t .z1  x01 / ; : : : ; x0N
i D1
@xi

C t.zN  x0N // ; (4.6)

for all j D 0; 1; : : : ; m:
In particular, we choose

z D .z1 ; : : : ; zN / D .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / D x C sj;

4.2 Main Results 49

and
x0 D .x01 ; : : : ; x0N / D .x1 ; x2 ; : : : ; xN / D x;
to get gxCsj .t/ WD f .x C t .sj // :
Notice gxCsj .0/ D f .x/ :
Also for jQ D 0; 1; : : : ; m  1 we get
0 1
B C N !
X B jQŠ C Y
.jQ/ B C ˛i
gxCsj .0/ D B N C .si j / f˛ .x/ : (4.7)
BY C
˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; @ ˛ Š A i D1
X
N i
i D1;:::;N; j˛jWD ˛i DjQ i D1
i D1

Furthermore, we obtain
0 1
B C N !
.m/
gxCsj ./ X B 1 C Y
B C
D B N C .si j / ˛i
f˛ .x C  .sj // ;
mŠ BY C
@ ŠA
˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; i D1
˛
X
N i
i D1;:::;N; j˛jWD ˛i Dm i D1
i D1
(4.8)
0    1:
jQ D 1; : : : ; m  1; and ˛ WD .˛1 ; : : : ; ˛N / ; ˛i 2 ZC ; i D 1; : : : ; N;
For P
j˛j WD N Q
i D1 ˛i D j , we define

Z Y
N
c˛;n;jQ WD c˛;n WD si˛i dn .s1 ; : : : ; sN / : (4.9)
RN i D1

Consequently, we derive

R .jQ/
X
m
RN gxCsj .0/ dn .s/
jQŠ
jQD1
0 1
0 1
B C
B C
B B C C
X QB
m1 X B 1 C C
j B B C C
D j B B N C c˛;n f˛ .x/C : (4.10)
B BY C C
Q
j D1 B ˛WD.˛1 ;:::;˛N /; ˛i 2ZC ; @ A C
B X
˛ iŠ C
@ N
i D1
A
i D1;:::;N; j˛jWD ˛i DjQ
i D1
50 4 Voronovskaya Asymptotic Expansions for General Singular Integrals

Next, we observe by multivariate Taylor’s formula that

X
m1 . jQ / .m/
gxCjs .0/ gxCjs ./
f .x C js/ D gxCjs .1/ D C ; (4.11)
jQŠ mŠ
jQD0

where  2 .0; 1/, which leads to

Z
.f .x C sj /  f .x// dn .s/ (4.12)
RN
0 1
B C
X
m1 BX C
jQ B 1 C
D j B ! c Q f˛ .x/ C
B Y
N ˛;n;j
C
jQD1 @j˛jDjQ ˛i Š A
i D1
0 1
B CZ
X B C Y N
B 1 C
Cj m
B N C si˛i f˛ .x C sj / dn .s/ :
BY C RN
j˛jDm @
ŠA
i D1
˛ i
i D1

Hence

X
r Z
Œm
Œm
r;n .f I x/  f .x/ D ˛j;r .f .x C sj /  f .x// dn .s/
j D0 RN
0 1
B C
XX
m1 r BX C
Œm Q B 1 C
D ˛j;r j j B ! c Q f˛ .x/C (4.13)
B Y
N ˛;n;j
C
jQD1 j D1 @j˛jDjQ ˛i Š A
i D1
0 1
B CZ
X
r X B C Y N
Œm B 1 C
C ˛j;r j m B N C si˛i
BY C RN
j˛jDm @
ŠA
j D1 i D1
˛ i
i D1

 f˛ .x C sj / dn .s/

4.2 Main Results 51

0 1
B C
X
m1 B X c Q f˛ .x/ C
Œm B ˛;n;j C
D ıjQ;r B ! C (4.14)
B YN C
jQD1 @j˛jDjQ ˛Š A i
i D1
0 1
B CZ !
X B C Y
N
B 1 C
C B N C si˛i
BY C RN
j˛jDm @
ŠA
i D1
˛ i
i D1
0 1
Xr  
r
@ .1/rj f˛ .x C sj /A dn .s/ :
j
j D1

Thus, we have
0 1
B C
X
m1B X c Q f˛ .x/ C
Œm B ˛;n;j C
WD r;n
Œm
.f I x/  f .x/  ıjQ;r B ! C (4.15)
B C
@j˛jDjQ Y ˛ Š A
N
jQD1
i
i D1
0 1
B CZ !0 r 1
X B C Y
N X  
B 1 C r
D B N C si˛i @ .1/rj f˛ .x C sj /A dn .s/ :
BY C RN j
j˛jDm @
ŠA
i D1 j D1
˛ i
i D1
(4.16)

Call  
X
r
r
˛ .x; s/ WD .1/rj f˛ .x C sj / : (4.17)
j
j D1

Thus,
0 1
B CZ !
X B C Y
N
B 1 C
D B N C si˛i ˛ .x; s/ dn .s/ : (4.18)
BY C RN
j˛jDm @
ŠA
i D1
˛ i
i D1
52 4 Voronovskaya Asymptotic Expansions for General Singular Integrals

Consider
n WD : (4.19)
nm

Suppose f˛ is bounded for all ˛ W j˛j D m; by M > 0. That is kf˛ k1  M .

Therefore,
0 1
r  
X r A M D .2r  1/ M:
j˛ .x; s/j  @ (4.20)
j
j D1

Consequently,
0 0 1 1
B B CZ ! C
ˇ ˇ BX B 1 C Y
N C
ˇn ˇ  .2  1/ M
r
B B C C
B B N C jsi j˛i dn .s/C : (4.21)
B B C RN C
@j˛jDm @ Y ˛
nm
ŠA
i D1 A
i
i D1

Suppose for j˛j D m that

Z !
Y
N
nm jsi j ˛i
dn .s/  ; for any .n /n2N : (4.22)
RN i D1

Therefore,
0 1
B C
ˇ ˇ BX C
B
ˇ ˇ  .2r  1/ M B 1 C
! C DW , > 0: (4.23)
B C
@j˛jDm Y ˛ Š A
n N

i
i D1

Hence
j j j j n
 and  n ! 0, (4.24)
nm nm

where 0 <  1, as n ! 0C.

That is
j j
m ! 0, as n ! 0C; (4.25)
n
 m 
which means D 0 n :
4.2 Main Results 53

We established
 
Theorem 4.2. Let f 2 C m RN , m; N 2 N,with all kf˛ k1  M; M > 0, all
˛ W j˛j D m. Let n > 0, .n /n2N bounded sequence, n probability Borel measures
on RN . R QN ˛i 
Call c˛;n;jQ D RN i D1 si dn .s/ ; all j˛j D jQ D 1; : : : ; m  1: Suppose
R Q 
nm RN N
i D1 jsi j
˛i
dn .s/  , all ˛ W j˛j D m;  > 0, for any such .n /n2N :
Also 0 <  1, x 2 RN : Then
0 1
B C
X
m1 B X c Q f˛ .x/ C  
Œm B ˛;n;j C
Œm
r;n .f I x/  f .x/ D ıjQ;r B ! C C 0 nm : (4.26)
B C
@j˛jDjQ Y ˛ Š A
N
jQD1
i
i D1

When m D 1, the sum collapses.

Above we assume r;n Œm
.f I x/ 2 R, 8 x 2 RN :
   
 @f 
Corollary 4.3. Let f 2 C 1 RN , N  1,with all  @x   M; M > 0, i D
i 1
1; : : : ; N . Let n > 0, .n /n2N bounded sequence, n probability Borel measures
on RN . Suppose
Z
n1 jsi j dn .s/  ; all i D 1; : : : ; N; (4.27)
RN

 > 0, for any such .n /n2N : Also 0 <  1, x 2 RN : Then

 
Œ1
r;n .f I x/  f .x/ D 0 n1 : (4.28)

Œ1
Above we assume r;n .f I x/ 2 R, 8 x 2 RN :
   2   2   2 
     f 
Corollary 4.4. Let f 2 C 2 R2 , with all  @@xf2  ;  @@xf2  ;  @x@1 @x   M; M
1 1 2 1 2 1
> 0. Let n > 0, .n /n2N bounded sequence, n probability Borel measures on R2 .
Call
Z Z
c1 D s1 dn .s/ ; c2 D s2 dn .s/ : (4.29)
R2 R2
Suppose
Z Z Z
n2 s12 dn .s/ ; n2 s22 dn .s/ ; n2 js1 j js2 j dn .s/  ;
R2 R2 R2
54 4 Voronovskaya Asymptotic Expansions for General Singular Integrals

 > 0, for any such .n /n2N : Also 0 <  1, x 2 R2 : Then

0 1
X r  
@f @f  
.f I x/  f .x/ D @ ˛j;r j A c1
Œ2
Œ2
r;n .x/ C c2 .x/ C 0 n2 : (4.30)
j D1
@x1 @x2

We continue with
 
Theorem 4.5. Let f 2 C l RN , l; N 2 N. Here, n is a Borel probability
measure on RN ; n > 0, .n /n2N P a bounded sequence. Let ˇ WD .ˇ1 ; : : : ; ˇN /,
ˇi 2 ZC , i D 1; : : : ; N I jˇj WD N
i D1 ˇi D l: Here f .x C sj /, x; s 2 R , is
N

n -integrable wrt s, for j D 1; : : : ; r: There exist n -integrable functions hi1 ;j ;
hˇ1 ;i2 ;j ; hˇ1 ;ˇ2 ;i3 ;j ; : : : ; hˇ1 ;ˇ2 ;:::;ˇN 1 ;iN ;j  0 (j D 1; : : : ; r) on RN such that
ˇ ˇ
ˇ @i1 f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hi1 ;j .s/ ; i1 D 1; : : : ; ˇ1 ;
ˇ i1
@x1 ˇ
ˇ ˇ
ˇ @ˇ1 Ci2 f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hˇ1 ;i2 ;j .s/ ; i2 D 1; : : : ; ˇ2 ;
ˇ ˇ
@x i2 @x 1 ˇ
2 1

::
:
ˇ ˇ
ˇ @ˇ1 Cˇ2 C:::CˇN 1 CiN f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hˇ1 ;ˇ2 ;:::;ˇN 1 ;iN ;j .s/ ; iN D 1; : : : ; ˇN ;
ˇ @x iN @x ˇN 1 : : : @x ˇ2 @x ˇ1 ˇ
N N 1 2 1
(4.31)
8 x; s 2 R . N

Then, both of the next exist and

   
Œm
r;n .f I x/ D r;n
Œm
fˇ I x : (4.32)
ˇ

Proof. By H. Bauer [6], pp. 103–104. t

u
We finish this chapter with
 
Theorem 4.6. Let f 2 C mCl RN , m; l; N 2 N. Assumptions of Theorem 4.5 are
valid. Call D 0; ˇ. Suppose f C˛ 1  M; M > 0, for all ˛ W j˛j D m.
Let n > 0, .n /n2N bounded sequence, n probability Borel measures on RN .
R Q N ˛i
Call c˛;n;jQ D RN i D1 si dn .s/ ; all j˛j D jQ D 1; : : : ; m  1: Assume
R  QN 
nm RN i D1 jsi j
˛i
dn .s/  , all ˛ W j˛j D m;  > 0, for any such .n /n2N :
Also 0 <  1, x 2 RN : Then
4.3 Applications 55

0 1
B C
  X Œm B C
B X c˛;n;jQ f C˛ .x/ C
m1
 
r;n .f I x/  f .x/ D
Œm
ıjQ;r B ! C C 0 nm : (4.33)
B YN C
jQD1 @j˛jDjQ ˛Š A
i
i D1

When m D 1, the sum collapses.

4.3 Applications

Let all entities as in Sect. 4.2. We define the following specific operators:
(a) The general multivariate Picard singular integral operators:

1 X
r
Œm
Œm
Pr;n .f I x1 ; : : : ; xN / WD ˛j;r  (4.34)
.2n /N j D0
0 1
B
X
N
C
B jsi jC
Z @ A
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

(b) The general multivariate Gauss–Weierstrass singular integral operators:

1 X
r
Œm
Œm
Wr;n .f I x1 ; : : : ; xN / WD p N ˛j;r  (4.35)
n j D0

0 1
B
X
N
C
B si2 C
Z @
i D1
A

f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :

RN

(c) The general multivariate Poisson–Cauchy singular integral operators:

X
r
Œm
Œm
Ur;n .f I x1 ; : : : ; xN / WD WnN ˛j;r  (4.36)
j D0

Z Y
N
1
f .x1 C s1 j; : : : ; xN C sN j /  ˇ ds1 : : : dsN ;
RN 2˛
i D1 si C n2˛
56 4 Voronovskaya Asymptotic Expansions for General Singular Integrals

with ˛ 2 N, ˇ > 1
2˛
; and

2˛ˇ1
.ˇ/ ˛n
Wn WD    : (4.37)
1
2˛
ˇ 1
2˛

(d) The general multivariate trigonometric singular integral operators:

X
r
N Œm
Œm
Tr;n .f I x1 ; : : : ; xN / WD n ˛j;r  (4.38)
j D0
0   12ˇ
Z Y
N sin si
n
f .x1 C s1 j; : : : ; xN C sN j / @ A ds1 : : : dsN ;
RN i D1
si

where ˇ 2 N, and

X̌ k 2ˇ1
n WD 2n12ˇ .1/ˇ ˇ .1/k : (4.39)
.ˇ  k/Š .ˇ C k/Š
kD1

Œm Œm Œm Œm

One can apply the results of this chapter to the operators Pr;n ; Wr;n ; Ur;n , Tr;n
Œm
(special cases of r;n ) and obtain interesting results.

References

1. G. Anastassiou, General uniform Approximation theory by multivariate singular integral

operators, submitted, 2011.
2. G. Anastassiou, Multivariate Voronovskaya asymptotic expansions for general singular opera-
tors, submitted, 2011.
3. G. Anastassiou, R. Mezei, A Voronovskaya type theorem for Poisson-Cauchy type singular
operators, J. Math. Anal. Appl. 366(2010), 525-529.
4. G. Anastassiou, R. Mezei, Voronovskaya type theorems for smooth Picard and Gauss-
Weierstrass singular operators, Studia Mathematica, Babes-Bolyai, Vol. LV, No. 4 (2010),
175-184.
5. G. Anastassiou, R. Mezei, A Voronovskaya type theorem for general singular operators with
applications, Nonlinear Functional Analysis & Appl., accepted, 2011.
6. H. Bauer, M˛ˇ-und Integrations theorie, de Gruyter, Berlin, 1990.
Chapter 5
Simultaneous Approximation by Multivariate
Complex General Singular Integrals

Here, we present complex multivariate simultaneous approximation for general

smooth singular integral operators converging with rates to the unit operator. The
associated and presented inequalities are in kkp , 1  p  1 norm and they
involve multivariate related moduli of smoothness. At the end, we list as this theory’s
applicators the special cases of multivariate complex Picard, Gauss–Weierstrass,
Poisson–Cauchy and Trigonometric singular integral operators. This chapter relies
on [2].

5.1 Introduction

Here, we are motivated by [1, 3, 4] and expand these works to complex valued
functions. We present simultaneous approximation in kkp , 1  p  1, of
multivariate general smooth singular integral operators to the unit operator with
rates. At the end, we list specific operators where our theory can be applied. From
our approximation results, one can derive interesting convergence properties of
these general operators. The expansion to complex case is based on basic properties
of complex numbers and complex valued functions.

5.2 Main Results

Here r 2 N; m 2 ZC , we define
8  
ˆ
ˆ r
ˆ
< .1/rj
j m ; if j D 1; 2; : : : ; r;
j
Œm
˛j;r WD Xr   (5.1)
ˆ
ˆ r m
:̂ 1  .1/ri
i ; if j D 0;
i
i D1

G.A. Anastassiou, Approximation by Multivariate Singular Integrals, 57

SpringerBriefs in Mathematics, DOI 10.1007/978-1-4614-0589-4 5,
© George A. Anastassiou 2011
58 5 Simultaneous Approximation by Multivariate Complex General Singular Integrals

and
Œm
X
r
Œm
ık;r WD ˛j;r j k ; k D 1; 2; : : : ; m 2 N. (5.2)
j D1

See that
X
r
Œm
˛j;r D 1; (5.3)
j D0

and    
X
r
r r
 .1/rj D .1/r : (5.4)
j 0
j D1

Let n be a probability Borel measure on RN , N  1, n > 0, n 2 N.

We now define the real multiple smooth singular integral operators

X
r Z
Œm
Œm
r;n .f I x1 ; : : : ; xN / WD ˛j;r f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j /
j D0 RN

dn .s/ ; (5.5)

where s WD .s1 ; : : : ; sN /, x WD .x1 ; : : : ; xN / 2 RN I n; r 2 N, m 2 ZC , f W RN !

R is a Borel measurable function, and also .n /n2N is a bounded sequence of positive
real numbers.
Œm
Above operators r;n are not in general positive operators and they preserve
constants, see [3].
 
Definition 5.1. Let f 2 C RN , N  1; m 2 N, the mth modulus of smoothness
for 1  p  1, is given by

!m .f I h/p WD sup km

t f .x/kp;x ; (5.6)
kt k2 h

h > 0, where
X
m  
m
m
t f .x/ WD .1/ mj
f .x C jt/ : (5.7)
j
j D0

Denote
!m .f I h/1 D !m .f; h/ : (5.8)
Above, x; t 2 RN :
We make
Remark 5.2. We consider here complex valued p Borel measurable functions f W
RN ! C such that f D f1 C if2 , i D 1, where f1 ; f2 W RN ! R are
implied to be real valued Borel measurable functions.
5.2 Main Results 59

We define the complex singular operators

Œm
r;n .f I x/ WD r;n
Œm
.f1 I x/ C ir;n
Œm
.f2 I x/ ; x 2 RN : (5.9)
 
Œm
We assume that r;n fj I x 2 R, 8 x 2 RN , j D 1; 2:
One notices easily that
ˇ ˇ
ˇ Œm ˇ
ˇr;n .f I x/  f .x/ˇ (5.10)
ˇ ˇ ˇ ˇ
ˇ Œm ˇ ˇ Œm ˇ
 ˇr;n .f1 I x/  f1 .x/ˇ C ˇr;n .f2 I x/  f2 .x/ˇ

also
 
 Œm 
r;n .f I x/  f .x/ (5.11)
1;x
   
 Œm   Œm 
 r;n .f1 I x/  f1 .x/ C r;n .f2 I x/  f2 .x/
1;x 1;x

and
 
 Œm 
r;n .f /  f  (5.12)
p
   
 Œm   Œm 
 r;n .f1 /  f1  C r;n .f2 /  f2  ; p  1:
p p

Furthermore, it holds
f˛ .x/ D f1;˛ .x/ C if2;˛ .x/ ; (5.13)
where ˛ denotes a partial derivative of any order and arrangement.
Here based on Theorem 9 of [3], we obtain
Theorem 5.3. Let f W RN ! C, N  1, such thatf D f1 C if2 , j D 1; 2.
Here m 2 N, fj 2 C m RN , x 2 RN : Assume fj;˛ 1 < 1, for all ˛k 2 ZC ,
P
k D 1; : : : ; N W j˛j D N kD1 ˛k D mI j D 1; 2: Let n be a Borel probability
measure on RN , for n > 0, .n /n2N bounded sequence. PN Assume that for all
˛ WD .˛1 ; : : : ; ˛N / ; ˛k 2 ZC , k D 1; : : : ; N; j˛j WD kD1 ˛k D m we have
that
Z !  
YN
˛k ksk2 r
jsk j 1C dn .s/ < 1: (5.14)
RN n
kD1

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z , k D 1; : : : ; N; j˛j WD
For
Q
kD1 ˛k D j , call
Z Y
N
c˛;n;jQ WD sk˛k dn .s1 ; : : : ; sN / : (5.15)
RN kD1
60 5 Simultaneous Approximation by Multivariate Complex General Singular Integrals

Then
 0 1
 
 
 B C 
 B
X Œm B X c˛;n;jQ f˛ .x/ C
m C
 Œm 
r;n .f I x/  f .x/  ıjQ;r B C
 B Y
N C
 @˛1 ;:::;˛N 0W 
 jQD1
j˛jDjQ ˛k Š A
 
kD1 1;x
X .!r .f1;˛ ; n / C !r .f2;˛ ; n //
 !  (5.16)
0 1 YN

@ A ˛k Š
˛1 ;:::;˛N 0
j˛jDm kD1

Z ! r !
Y
N
˛k ksk2
jsk j 1C dn .s/ ; x 2 RN :
RN n
kD1

The m D 0 case follows

Corollary 5.4. Let f W RN ! C W f D f1 C if2 , N  1. Here j D 1; 2. Let
fj 2 CB RN (continuous and bounded functions). Then
  Z   
 Œ0  ksk2 r
r;n f  f   1C dn .s/  (5.17)
1 RN n
.!r .f1 ; n / C !r .f2 ; n // ;

by assuming
Z  
ksk2 r
1C dn .s/ < 1: (5.18)
RN n

Proof. By Theorem 11 of [3]. t

u
 
Theorem 5.5 ([1]). Let f 2 C l RN , l; N 2 N. Here, n is a Borel probability
measure on RN ; n > 0, .n /n2NP a bounded sequence. Let ˇ WD .ˇ1 ; : : : ; ˇN /,
ˇi 2 ZC , i D 1; : : : ; N I jˇj WD N i D1 ˇi D l: Here, f .x C sj /, x; s 2 R , is
N

n -integrable wrt s, for j D 1; : : : ; r: There exist n -integrable functions hi1 ;j ;
hˇ1 ;i2 ;j ; hˇ1 ;ˇ2 ;i3 ;j ; : : : ; hˇ1 ;ˇ2 ;:::;ˇN 1 ;iN ;j  0 (j D 1; : : : ; r) on RN such that
ˇ ˇ
ˇ @i1 f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hi1 ;j .s/ ; i1 D 1; : : : ; ˇ1 ; (5.19)
ˇ @x1 i1 ˇ
ˇ ˇ
ˇ @ˇ1 Ci2 f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hˇ1 ;i2 ;j .s/ ; i2 D 1; : : : ; ˇ2 ;
ˇ i2
@x2 @x1
ˇ1 ˇ
::
:
5.2 Main Results 61

ˇ ˇ
ˇ @ˇ1 Cˇ2 C:::CˇN 1 CiN f .x C sj / ˇ
ˇ ˇ
ˇ ˇ  hˇ1 ;ˇ2 ;:::;ˇN 1 ;iN ;j .s/ ; iN D 1; : : : ; ˇN ;
ˇ @x iN @x ˇN 1 : : : @x ˇ2 @x ˇ1 ˇ
N N 1 2 1
(5.1)
8 x; s 2 R .
N

Then, bothof the next exist and

   
Q Q
Œm
r;n .f I x/ D r;n
Œm
fˇ I x : (5.20)
ˇ

We give

 f W R ! C such that f D f1 C if2 : Here j D 1; 2. Let

N
Theorem 5.6. NLet
fj 2 C mCl
R , m; l;N 2 N: For fj , the assumptions of Theorem 5.5 are valid.
Call  D 0; ˇ: Assume f. C˛/ 1 < 1 and

Z ! r
Y
N
ksk2
jsk j ˛k
1C dn .s/ < 1; (5.21)
RN n
kD1

where n is a Borel probability measure on RN , for n > 0, .n /n2N is bounded
P
sequence; for all ˛k 2 ZC , k D 1; : : : ; N W j˛j D N kD1 ˛k D m:
For jQ D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 ZC , k D 1; : : : ; N; j˛j WD
PN
Q
kD1 ˛k D j , call
Z Y N
c˛;n;jQ WD sk˛k dn .s/ :
RN kD1

Then
 0 1
 
 
 B C 
  X m B X c Q f C˛ ./ C
 Œm Œm B ˛;n;j C
 r;n .f I /  f ./  ıjQ;r B C
  B YN C
 @˛1 ;:::;˛N 0W 
 jQD1
˛k Š A
 j˛jDjQ 
kD1 1
    
X !r f1; C˛ ; n C !r f2; C˛ ; n
 ! 
0 1 YN

@ A ˛k Š
˛1 ;:::;˛N 0
j˛jDm kD1

Z ! r !
Y
N
ksk2
˛k
jsk j 1C dn .s/ : (5.22)
RN n
kD1

Proof. By Theorem 10 of [1]. t

u
62 5 Simultaneous Approximation by Multivariate Complex General Singular Integrals

Also, we have
Theorem5.7.  Let f W RN ! C such that f D f1 C if2 : Here j D 1; 2. Let
fj 2 CBl RN , l; N 2 N: The assumptions of Theorem 5.5 are valid for fj . Call
 D 0; ˇ: Assume Z  
ksk2 r
1C dn .s/ < 1:
RN n
Then
  Z   
 Œ0   ksk2 r
  f  f   1C dn .s/  (5.23)
 r;n   n
1 RN
    
!r f1; ; n C !r f2; ; n :

Proof. By Theorem 11 of [1]. t

u
By Theorem 4 of [4], we get

 N  5.8. Let f W R ! C W f D  f1 C if2 : Here j D 1; 2. Let fj 2

N
Theorem
C R , m 2 N, N  1; with fj;˛ 2 Lp RN ; j˛j D m, x 2 RN . Let p; q > 1 W
m

p
1
C q1 D 1: Here n is a Borel probability measure on RN for n > 0, .n /n2N is
C
PN sequence. Assume for all ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z , k D 1; : : : ; N;
a bounded
j˛j WD kD1 ˛k D m, we have that
Z ! r !p
Y
N
˛k ksk2
jsk j 1C dn .s/ < 1: (5.24)
RN n
kD1

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z , k D 1; : : : ; N; j˛j WD
For
Q
kD1 ˛k D j , call
Z Y N
c˛;n;jQ WD sk˛k dn .s/ : (5.25)
RN kD1

Then
 0 1
 
 
 B C
 B
X Œm B X c˛;n;jQ f˛ .x/ C
m C
 Œm 
r;n .f I x/  f .x/  ıjQ;r B C
 B Y
N C
 @j˛jDjQ 
 jQD1
˛k Š A
 
kD1 p;x
0 1
!B C
BX C
m B 1 C
 B ! C
.q .m  1/ C 1/ q B Y C
1 N
@j˛jDm ˛k Š A
kD1
5.2 Main Results 63

"Z " ! r #p #p 1
Y
N
ksk
jsk j˛k 1C 2
dn .s/ 
RN n
kD1
 
!r .f1;˛ ; n /p C !r .f2;˛ ; n /p : (5.26)

We further get

 W R ! C W f D f1 C if2 ; j D 1; 2. Let fj 2
N
  N  5.9. LetN f
Theorem
C R \ Lp R , N  1; p; q > 1 W p1 C q1 D 1: Assume n probability
Borel measures on RN , .n /n2N > 0 and bounded. Also suppose
Z  
ksk2 rp
1C dn .s/ < 1: (5.27)
RN n

Then

  Z    p1
 Œ0  ksk2 rp
r;n .f /  f   1C dn .s/  (5.28)
p RN n
 
!r .f1 ; n /p C !r .f2 ; n /p :

Proof. By Theorem 6 of [4]. t

u
Based on Theorem 8 of [4], we get

f W R ! C; f D f1 C if2 ; j D 1; 2. Let fj N2

N
Theorem
  N  5.10. Let
C R \ L1 R , N  1: Assume n probability Borel measures on R ,
N

.n /n2N > 0 and bounded. Also suppose

Z  
ksk2 r
1C dn .s/ < 1: (5.29)
RN n

Then
  Z   
 Œ0  ksk2 r
r;n .f /  f   1C dn .s/  (5.30)
1 RN n
.!r .f1 ; n /1 C !r .f2 ; n /1 / :

Based on Theorem 10 of [4], we get

 
Theorem 5.11. Let f W RN !  C; f D f1 C if2 ; j D 1; 2. Let fj 2 C m RN ,
m; N 2 N; with fj;˛ 2 L1 RN ; j˛j D m, x 2 RN . Here, n is a Borel probability
measure on RN for n > 0, .n /n2N is a bounded sequence. PN Assume for all
˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 ZC , k D 1; : : : ; N; j˛j WD kD1 k D m that we
˛
have
64 5 Simultaneous Approximation by Multivariate Complex General Singular Integrals

Z ! r !
Y
N
ksk2
jsk j˛k 1C dn .s/ < 1: (5.31)
RN n
kD1

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z , k D 1; : : : ; N; j˛j WD
For
Q
kD1 ˛k D j , call
Z Y N
c˛;n;jQ WD sk˛k dn .s/ : (5.32)
RN kD1

Then
 0 1
 
 
 B C 
 B
X Œm B X c˛;n;jQ f˛ .x/ C
m C
 Œm 
r;n .f I x/  f .x/  ıjQ;r B C
 B Y
N C
 @j˛jDjQ 
 jQD1
˛k Š A
 
kD1 1;x
0 1
B C
XB C 
B 1 C
 B N C !r .f1;˛ ; n /1 C !r .f2;˛ ; n /1  (5.33)
BY C
j˛jDm @
˛ k ŠA
kD1

Z ! r
Y
N
˛k ksk2
jsk j 1C dn .s/ : (5.2)
RN n
kD1

Basedon Theorem 12 of [1], we get

 
Theorem 5.12. Let f W RN ! C; f D f1 C if2 I j D 1; 2, with fj 2 C mCl RN ;
m; l; N 2 N: The  assumptions
 of Theorem 5.5 are valid for fj . Call  D 0; ˇ. Let
fj;. C˛/ 2 Lp RN , j˛j D m, x 2 RN , p; q > 1 W p1 C q1 D 1: Here, n is a Borel
probability measure on RN for n > 0, .n /n2N is a bounded P sequence. Assume for
all ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 ZC , k D 1; : : : ; N; j˛j WD N kD1 ˛k D m we have that
Z !   ! p
Y
N
ksk2 r
jsk j˛k 1C dn .s/ < 1: (5.34)
RN n
kD1

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z , k D 1; : : : ; N; j˛j WD
For
Q
kD1 ˛k D j , call
Z Y N
c˛;n;jQ WD sk˛k dn .s/ :
RN kD1
5.2 Main Results 65

Then
 0 1
 
 
 B C
  X m B X c Q f C˛ .x/ C
 Œm Œm B ˛;n;j C
 r;n .f I x/  f .x/  ıjQ;r B C (5.35)
  B YN C
 jQD1 @j˛jDjQ A
 ˛k Š 
 
kD1 p;x
0 1
!B C
BX C
m B 1 C
 B ! C
.q .m  1/ C 1/
1
B Y
N C
q
@j˛jDm ˛k Š A
kD1
"Z " ! r # p # p1
Y
N
ksk2
˛k
jsk j 1C dn .s/ 
RN n
kD1
     
!r f1; C˛ ; n p C !r f2; C˛ ; n p :

Based on Theorem 13 of [1] we get

 
Theorem 5.13. Let f W RN ! C; f D f1 C if2 ; j D 1; 2: Let fj 2 C l RN ;
l; N 2 N: The assumptions of Theorem 5.5 are valid. Call  D 0; ˇ. Let fj; 2
Lp RN , x 2 RN , p; q > 1 W p1 C q1 D 1: Assume n probability Borel measures
on RN ; .n /n2N > 0 and bounded. Also suppose
Z  
ksk2 rp
1C dn .s/ < 1:
RN n
Then
  Z    p1
 Œ0   ksk2 rp
  f  f   1C dn .s/  (5.36)
 r;n 
 n
p RN
     
!r f1; ; n p C !r f2; ; n p :

By Theorem 14 of [1], we get

 
Theorem 5.14. Let f W RN ! C; f D f1 C if2 ; j D 1; 2: Let fj 2 C l RN ;
l; N 2 N: The assumptions of Theorem 5.5 are valid for fj . Call  D 0; ˇ. Let
fj; 2 L1 RN , x 2 RN : Assume n probability Borel measures on RN ; .n /n2N >
0 and bounded. Also suppose
Z  
ksk2 r
1C dn .s/ < 1:
RN n
66 5 Simultaneous Approximation by Multivariate Complex General Singular Integrals

Then
  Z   
 Œ0   ksk2 r
  f  f   1C dn .s/  (5.37)
 r;n   n
1 RN
     
!r f1; ; n 1 C !r f2; ; n 1 :

Finally, we have
 
Theorem 5.15. Let f W RN ! C; f D f1 C if2 ; j D 1; 2, with fj 2 C mCl RN ,
m; l; N 2 N. The  assumptions
 of Theorem 5.5 are valid for fj . Call  D 0; ˇ. Let
fj;. C˛/ 2 L1 RN ; j˛j D m, x 2 RN . Here n is a Borel probability measure on
C
RN for n > 0, .n /n2N PNis bounded. Assume for all ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z ,
k D 1; : : : ; N; j˛j WD kD1 ˛k D m; we have that

Z ! r !
Y
N
ksk2
jsk j ˛k
1C dn .s/ < 1: (5.38)
RN n
kD1

Q C
PN j D 1; : : : ; m, and ˛ WD .˛1 ; : : : ; ˛N /, ˛k 2 Z , k D 1; : : : ; N; j˛j WD
For
Q
kD1 ˛k D j , call
Z Y N
c˛;n;jQ WD sk˛k dn .s/ : (5.39)
RN kD1

Then
 0 1
 
 
 B C
  m B
X Œm B X c˛;n;jQf. C˛/ .x/ C
C
 Œm 
 r;n .f I x/  f .x/  ıjQ;r B C (5.40)
  B Y
N C
 jQD1 @j˛jDjQ A
 ˛k Š 
 
kD1 1;x
0 1
B C
XB C    
B 1 C
 B N C !r f1; C˛ ; n 1 C !r f2; C˛ ; n 1 
BY C
j˛jDm @
˛ k ŠA
kD1
Z ! r
Y
N
˛k ksk2
jsk j 1C dn .s/ :
RN n
kD1

Proof. By Theorem 15 of [1]. t

u
5.3 Applications 67

5.3 Applications

Let all entities as in Sect. 5.2. We define the following specific operators for f W
RN ! C:
(a) The general multivariate Picard singular integral operators:

1 X
r
Œm
Œm
Pr;n .f I x1 ; : : : ; xN / WD N
˛j;r  (5.41)
.2n / j D0
0 1
B
X
N
C
B jsi jC
Z @ A
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

(b) The general multivariate Gauss–Weierstrass singular integral operators:

1 X r
Œm
Œm
Wr;n .f I x1 ; : : : ; xN / WD p N ˛j;r  (5.42)
n j D0

0 1
B
X
N
C
B si2 C
Z @ A
i D1
f .x1 C s1 j; x2 C s2 j; : : : ; xN C sN j / e n ds1 : : : dsN :
RN

(c) The general multivariate Poisson–Cauchy singular integral operators:

X
r
Œm
Œm
Ur;n .f I x1 ; : : : ; xN / WD WnN ˛j;r  (5.43)
j D0

Z Y
N
1
f .x1 C s1 j; : : : ; xN C sN j /  ˇ ds1 : : : dsN ;
RN 2˛
i D1 si C n2˛

with ˛ 2 N, ˇ > 1
2˛
; and

2˛ˇ1
.ˇ/ ˛n
Wn WD    : (5.44)
1
2˛ ˇ 1
2˛

(d) The general multivariate trigonometric singular integral operators:

X
r
N Œm
Œm
Tr;n .f I x1 ; : : : ; xN / WD n ˛j;r  (5.45)
j D0
68 5 Simultaneous Approximation by Multivariate Complex General Singular Integrals

0   12ˇ
Z Y
N sin si
n
f .x1 C s1 j; : : : ; xN C sN j / @ A ds1 : : : dsN ;
RN i D1
si

where ˇ 2 N, and

X̌ k 2ˇ1
n WD 2n12ˇ  .1/ˇ ˇ .1/k : (5.46)
.ˇ  k/Š .ˇ C k/Š
kD1

Œm Œm Œm Œm

One can apply the results of this chapter to the operators Pr;n ; Wr;n ; Ur;n , Tr;n
Œm
(special cases of r;n ) and derive interesting results.

5.4 Conclusion

Our approximation results here imply important convergence properties of operators

Œm
r;n to the unit operator.

References

1. G. Anastassiou, Global smoothness preservation and simultaneous approximation for multivari-

ate general singular integral operators, Applied Math. Letters, 24(2011), 1009-1016.
2. G. Anastassiou, Multivariate Complex General Singular Integral Operators Simultaneous
Approximation, Journal of Concrete and Applicable Mathematics, accepted, 2010.
3. G. Anastassiou, General uniform Approximation theory by multivariate singular integral
operators, submitted, 2011.
4. G. Anastassiou, Lp -general approximations by multivariate singular integral operators, sub-
mitted, 2011.
Chapter 6
Approximation of Functions of Two Variables
via Almost Convergence of Double Sequences

The idea of almost convergence for double sequences was introduced by Moricz
and Rhoades [6]. In this chapter, we use this concept to prove a Korovkin-type
approximation theorem for functions of two variables and we give an example.
Furthermore, we present the consequences of the main theorem. This chapter is
based on [1].

6.1 Introduction and Preliminaries

Let c and l1 denote the spaces of all convergent and bounded sequences,
respectively, and note that c  l1 . In the theory of sequence spaces, a beautiful
application of the well-known Hahn–Banach Extension Theorem gave rise to the
concept of the Banach limit. That is, the lim functional defined on c can be extended
to the whole of l1 and this extended functional is known as the Banach limit [2],
which was used by Lorentz [5] to define a new type of convergence, known as the
almost convergence.
A double sequence x D .xjk / of real or complex numbers is said to be bounded
if kxk1 D supj;k jxjk j < 1: The space of all bounded double sequences is denoted
by Mu .
A double sequence x D .xjk / is said to converge to the limit L in Pringsheim’s
sense (shortly, p-convergent to L) [10] if for every " > 0 there exists an integer N
such that jxjk  Lj < " whenever j; k > N . In this case L is called the p-limit
of x. If in addition x 2 Mu , then x is said to be boundedly convergent to L in
Pringsheim’s sense (shortly, bp-convergent to L).

G.A. Anastassiou, Approximation by Multivariate Singular Integrals, 69

SpringerBriefs in Mathematics, DOI 10.1007/978-1-4614-0589-4 6,
© George A. Anastassiou 2011
70 6 Approximation of Functions of Two Variables

Let  denote the vector space of all double sequences with the vector space op-
erations defined coordinatewise. Vector subspaces of  are called double sequence
spaces. In addition to the above-mentioned double sequence spaces, we consider the
double sequence space
8 9
< X =
Lu WD x 2  j kxk1 WD jxjk j < 1
: ;
j;k

of all absolutely summable double sequences.

All considered double sequence spaces are supposed to contain

ˆ WD span fejk j j; k 2g;

where 
jk 1 I if .j; k/ D .i; `/;
eil D
0 I otherwise:
P P P
We denote the pointwise sums j;k ejk , j ejk .k 2/, and k ejk .j 2/ by e, ek
and ej respectively.
The idea of almost convergence for double sequences was introduced and studied
by Moricz and Rhoades [6].
A double sequence x D .xjk / of real numbers is said to be almost convergent to
a limit L if ˇ ˇ
ˇ X nCq1 X ˇ
ˇ 1 mCp1 ˇ
lim sup ˇˇ xjk  Lˇˇ D 0 ()
p;q!1 m;n>0 ˇ pq ˇ
j Dm kDn

In this case, L is called the F2 -limit of x and we shall denote by F2 the space of all
almost convergent double sequences.
Note that a convergent double sequence need not be almost convergent. However,
every bounded convergent double sequence is almost convergent and every almost
convergent double sequence is bounded.
Example 6.1. The double sequence z D .zmn / defined by
8
ˆ
ˆ 1 if m D n odd;
<
zmn D 1 if m D n even; (6.1)
ˆ
:̂
0 .m 6D n/I

is almost convergent to zero but not convergent.

For recent developments on almost convergent double sequences and matrix
transformations, we refer to [7–9, 11].
If m D n D 1 in ./, then we get .C; 1; 1/-convergence, and in this case we write
xjk ! `.C; 1; 1/I where ` D .C; 1; 1/-lim x.
6.2 Korovkin-Type Approximation Theorem 71

Let C Œa; b be the space of all functions f continuous on Œa; b. We know that
C Œa; b is a Banach space with norm

kf k1 WD sup jf .x/j; f 2 C Œa; b:

x2Œa;b

The classical Korovkin approximation theorem states as follows [4]:

Let .Tn / be a sequence of positive linear operators from C Œa; b into
C Œa; b: Then limn kTn .f; x/  f .x/k1 D 0, for all f 2 C Œa; b if and only if
limn kTn .fi ; x/  fi .x/k1 D 0, for i D 0; 1; 2, where f0 .x/ D 1; f1 .x/ D x and
f2 .x/ D x 2 :
Quite recently, such type of approximation theorems are proved in [3] for
functions of two variables by using statistical convergence. In this chapter, we use
the notion of almost convergence to prove approximation theorems for functions of
two variables.

6.2 Korovkin-Type Approximation Theorem

The following is the F2 -version of the classical Korovkin approximation theorem

followed by an example to show its importance.
Let C.I 2 / be the space of all two dimensional continuous functions on I  I ,
where I D Œa; b. Suppose that Tm;n W C.I 2 / ! C.I 2 /. We write Tm;n .f I x; y/ for
Tm;n .f .s; t/I x; y/I and we say that T is a positive operator if T .f I x; y/  0 for
all f .x; y/  0.
Theorem 6.2. Let .Tj;k / be a double sequence of positive linear operators from
PmCp1 PnCq1
C.I 2 / into C.I 2 / and Dm;n;p;q .f I x; y/ D pq
1
j Dm kDn Tj;k .f I x; y/:
Then for all f 2 C.I 2 /
 
 

F2 - lim Tj;k .f I x; y/  f .x; y/
 D 0; i:e:
j;k!1 1
 
 
lim 
 Dm;n;p;q .f I x; y/  f .x; y/ 
 D 0; uniformly in m; n: (6.2)
p;q!1
1

if and only if
 
 

lim Dm;n;p;q .1I x; y/  1
 D 0 uniformly in m; n; (6.3)
p;q!1
1
 
 

lim Dm;n;p;q .sI x; y/  x  D 0 uniformly in m; n; (6.4)
p;q!1 
1
72 6 Approximation of Functions of Two Variables

 
 

lim Dm;n;p;q .tI x; y/  y  D 0 uniformly in m; n: (6.5)
p;q!1 
1
 
 
 2 2 2 2 
lim 4Dm;n;p;q .s C t I x; y/  .x C y / D 0 uniformly in m; n: (6.6)
p;q!1
1

Proof. Since each 1; x; y; x 2 C y 2 belongs to C.I 2 /; conditions (6.3)–(6.6) follow

immediately from (6.2). By the continuity of f on I 2 , we can write jf .x; y/j 
M;  1 < x; y < 1; where M D kf k1 : Therefore,

jf .s; t/  f .x; y/j  2M;  1 < s; t; x; y < 1: (6.7)

Also, since f 2 C.I 2 /; for every  > 0, there is ı > 0 such that

jf .s; t/  f .x; y/j < ; 8 js  xj < ı and jt  yj < ı: (6.8)

Using (6.7), (6.8), putting 1 D 1 .s; x/ D .s  x/2 and 2 D 2 .t; y/ D .t  y/2 ,

we get

2M
jf .s; t/  f .x; y/j <  C . 1 C 2 /; 8 js  xj < ı and jt  yj < ı:
ı2
This is,

2M 2M
  . 1 C 2/ < f .s; t/  f .x; y/ <  C . 1 C 2 /:
ı2 ı2

Now, operating Tj;k .1I x; y/ to this inequality since Tj;k .f I x; y/ is monotone and
linear. We obtain
 
2M
Tj;k .1I x; y/   2 . 1 C 2 / < Tj;k .1I x; y/.f .s; t/  f .x; y//
ı
 
2M
< Tj;k .1I x; y/  C 2 . 1 C 2 / :
ı

Note that x and y are fixed and so f .x; y/ is constant number. Therefore

2M
Tj;k .1I x; y/  Tj;k . 1 C 2 I x; y/ < Tj;k .f I x; y/  f .x; y/Tj;k .1I x; y/
ı2
2M
< Tj;k .1Ix;y/C Tj;k . 1 C 2 Ix;y/:
ı2
(6.9)
6.2 Korovkin-Type Approximation Theorem 73

But

Tj;k .f I x; y/  f .x; y/ D Tj;k .f I x; y/  f .x; y/Tj;k .1I x; y/

C f .x; y/Tj;k .1I x; y/  f .x; y/
D ŒTj;k .f I x; y/  f .x; y/Tj;k .1I x; y/
C f .x; y/ŒTj;k .1I x; y/  1: (6.10)

Using (6.9) and (6.10), we have

2M
Tj;k .f I x; y/  f .x; y/ < Tj;k .1I x; y/ C Tj;k . 1 C 2 I x; y/
ı2
C f .x; y/.Tj;k .1I x; y/  1/: (6.11)

Now

Tj;k . 1 C 2 I x; y/ D Tj;k ..s  x/2 C .t  y/2 I x; y/

D Tj;k .s 2  2sx C x 2 C t 2  2ty C y 2 I x; y/
D Tj;k .s 2 C t 2 I x; y/  2xTj;k .sI x; y/  2yTj;k .tI x; y/
C .x 2 C y 2 /Tj;k .1I x; y/
D ŒTj;k .s 2 C t 2 I x; y/  .x 2 C y 2 /  2xŒTj;k .sI x; y/  x
 2yŒTj;k .tI x; y/  y C .x 2 C y 2 /ŒTj;k .1I x; y/  1:

Using (6.11), we obtain

2M
Tj;k .f I x; y/  f .x; y/ < Tj;k .1I x; y/ C fŒTj;k .s 2 C t 2 I x; y/  .x 2 C y 2 /
ı2
 2xŒTj;k .sI x; y/  x  2yŒTj;k .tI x; y/  y
C .x 2 C y 2 /ŒTj;k .1I x; y/  1g
C f .x; y/.Tj;k .1I x; y/  1/
2M
D ŒTj;k .1I x; y/  1 C  C fŒTj;k .s 2 C t 2 I x; y/
ı2
 .x 2 C y 2 /  2xŒTj;k .sI x; y/  x
 2yŒTj;k .tI x; y/  y C .x 2 C y 2 /ŒTj;k .1I x; y/  1g
C f .x; y/.Tj;k .1I x; y/  1/:
74 6 Approximation of Functions of Two Variables

Since  is arbitrary, we can write

2M
Tj;k .f I x; y/  f .x; y/  ŒTj;k .1I x; y/  1 C fŒTj;k .s 2 C t 2 I x; y/
ı2

 .x 2 C y 2 /  2xŒTj;k .sI x; y/  x
 2yŒTj;k .tI x; y/  y C .x 2 C y 2 /ŒTj;k .1I x; y/  1g

C f .x; y/.Tj;k .1I x; y/  1/:

Similarly,

Dm;n;p;q .f I x; y/  f .x; y/  ŒDm;n;p;q .1I x; y/  1

2M ˚
C ŒDm;n;p;q .s 2 C t 2 I x; y/  .x 2 C y 2 /
ı2
2xŒDm;n;p;q .sI x; y/  x

2yŒDm;n;p;q .tI x; y/  y

C.x 2 C y 2 /ŒDm;n;p;q .1I x; y/  1

Cf .x; y/.Dm;n;p;q .1I x; y/  1/

and therefore
   
 
Dm;n;p;q .f I x; y/  f .x; y/   C 2M.a C b / C M
2 2

  ı2
1
 
 
 Dm;n;p;q .1I x; y/  1


1
 
4M a  
 2 Dm;n;p;q .sI x; t/  x 


ı 1
 
4M b  
 2  Dm;n;p;q .tI x; y/  y 

ı  1
 
2M 


2 
C 2 Dm;n;p;q .s C t I x; y/  .x C y / :
2 2 2
ı 1
6.2 Korovkin-Type Approximation Theorem 75

Letting p; q ! 1 and using (6.3), (6.4), (6.5), and (6.6), we get

 
 
lim 
Dm;n;p;q .f I x; y/  f .x; y/  D 0; uniformly in m; n:

p;q!1
1

This completes the proof of the theorem. t

u
In the following, we give an example of a double sequence of positive linear op-
erators satisfying the conditions of Theorem 6.2 but does not satisfy the conditions
of the Korovkin theorem.
Example 6.3. Consider the sequence of classical Bernstein polynomials of two
variables

X m Xn   ! !
j k m n j
Bm;n .f I x; y/ WD f ; x .1  x/mj
j D0
m n j k
kD0

 y k .1  y/nk I 0  x; y  1:

Let Pm;n W C.I 2 / ! C.I 2 / be defined by

Pm;n .f I x; y/ D .1 C zmn /Bm;n .f I x; y/;

where .zmn / is a double sequence defined as above. Then

Bm;n .1I x; y/ D 1;

Bm;n .sI x; y/ D x;

Bm;n .tI x; y/ D y;

x  x2 y  y2
Bm;n .s 2 C t 2 I x; y/ D x 2 C y 2 C C ;
m n

and a double sequence .Pm;n / satisfies the conditions (6.3), (6.4), (6.5) and (6.6).
Hence, we have

F2 - lim kPm;n .f I x; y/  f .x; y/k1 D 0:

m;n!1

On the other hand, we get Pm;n .f I 0; 0/ D .1 C zmn /f .0; 0/, since Bm;n .f I 0; 0/ D
f .0; 0/, and hence

kPm;n .f I x; y/  f .x; y/k1  jPm;n .f I 0; 0/  f .0; 0/j D zmn jf .0; 0/j:

We see that .Pm;n / does not satisfy the classical Korovkin theorem, since
lim zmn does not exist.
m;n!1
76 6 Approximation of Functions of Two Variables

6.3 Some Consequences

Now we present here some consequences of Theorem 6.2.

Theorem 6.4. Let .Tm;n / be a double sequence of positive linear operators on
C.I 2 / such that

lim kTmC1;nC1  Tm;nC1  TmC1;n C Tm;n k D 0: (6.12)

m;n

If
F2 - lim kTm;n .t I x; y/  t k1 D 0 . D 0; 1; 2; 3/; (6.13)
m;n

where t0 .x; y/ D 1, t1 .x; y/ D x, t2 .x; y/ D y and t3 .x; y/ D x 2 C y 2 . Then for

any function f 2 C.I 2 /, we have

lim kTm;n .f I x; y/  f .x; y/k1 D 0: (6.14)

m;n

Proof. From Theorem 6.2, we have that if (6.13) holds then

lim kDm;n;p;q .f I x; y/  f .x; y/k1 D 0; uniformly in m; n: (6.15)

p;q

We have the following inequality

kTm;n .f I x; y/  f .x; y/k1

 kDm;n;p;q .f I x; y/  f .x; y/k1
!
1 X X X X
mCp1 nCq1 j k
C kT˛;ˇ  T˛1;ˇ  T˛;ˇ1 C T˛1;ˇ1 k
pq
j DmC1 kDnC1 ˛DmC1 ˇDnC1

 kDm;n;p;q .f I x; y/  f .x; y/k1

 
p1q1
C sup kTj;k  Tj 1;k  Tj;k1 C Tj 1;k1 k : (6.16)
2 2 j m;kn

Hence using (6.12) and (6.15), we get (6.14).

This completes the proof of the theorem. t
u
We know that double almost convergence implies .C; 1; 1/ convergence. This
motivates us to further generalize our main result by weakening the hypothesis or to
add some condition to get more general result.
Theorem 6.5. Let .Tm;n / be a double sequence of positive linear operators on
C.I 2 / such that

.C; 1; 1/  lim kTm;n .t ; x/  t k1 D 0 . D 0; 1; 2; 3/; (6.17)

m;n
6.3 Some Consequences 77

and
   
mn  
lim sup mCp1;nCq1 .f I x; y/  m1;n1 .f I x; y/ D 0; (6.18)
p;q mp;nq pq  
1

where
1 XXm n
m;n .f I x; y/ D Tj;k .f I x; y/:
.m C 1/.n C 1/
j D0 kD0

Then for any function f 2 C.I /, we have2

 
 

F2 - lim Tm;n .f I x; y/  f .x; y/
 D 0:
m;n!1
1

Proof. For m  p  1I n  q  1, it is easy to show that

mn
Dm;n;p;q .f I x; y/ D mCp1;nCq1 .f I x; y/ C .mCp1;nCq1 .f I x; y/
pq
m1;n1 .f I x; y//;

which implies
 
 

sup Dm;n;p;q .f I x; y/  mCp1;nCq1 .f I x; y/

mp;nq 1
 
mn  
mCp1;nCq1 .f I x; y/  m1;n1 .f I x; y// :
D sup   (6.19)
mp;nq pq 1

Also by Theorem 6.2, Condition (6.17) implies that

 
 
.C; 1; 1/- lim Tm;n .f I x; y/  f .x; y/

 D 0: (6.20)
m;n!1
1

Using (6.17)–(6.20) and the fact that almost convergence implies .C; 1; 1/ conver-
gence, we get the desired result.
This completes the proof of the theorem. t
u
Theorem 6.6. Let .Tm;n / be a double sequence of positive linear operators on
C.I 2 / such that

sCm1 t Cn1
1 X X
lim sup kTm;n  Tj;k k D 0:
m;n s;t mn j Ds
kDt
78 6 Approximation of Functions of Two Variables

If
F2 - lim kTm;n .t ; x/  t k1 D 0 . D 0; 1; 2; 3/: (6.21)
m;n

Then for any function f 2 C.I 2 /, we have

lim kTm;n .f I x; y/  f .x; y/k1 D 0: (6.22)

m;n

Proof. From Theorem 6.2, we have that if (6.21) holds then

F2 - lim kTm;n .f I x; y/  f .x; y/k1 D 0;

m;n

which is equivalent to

lim sup kDs;t;m;n .f I x; y/  f .x; y/k1 D 0: (6.23)

m;n s;t

Now
sCm1 t Cn1
1 X X
Tm;n  Ds;t;m;n D Tm;n  Tj;k
mn j Ds
kDt

sCm1 t Cn1
1 X X
D .Tm;n  Tj;k /:
mn j Ds
kDt

Therefore,
sCm1 t Cn1
1 X X
sup kTm;n  Ds;t;m;n k1  sup kTm;n  Tj;k k:
s;t s;t mn j Ds
kDt

Now, by using the hypothesis we get

lim sup kTm;n .f I x; y/  Ds;t;m;n .f I x; y/k1 D 0: (6.24)

m;n s;t

By the triangle inequality, we have

kTm;n .f I x; y/  f .x; y/k1  kTm;n .f I x; y/  Ds;t;m;n .f I x; y/k1

CkDs;t;m;n .f I x; y/  f .x; y/k1 ;

and hence from (6.23) and (6.24), we get

lim kTm;n .f I x; y/  f .x; y/k1 D 0;

m;n

that is (6.22) holds.

This completes the proof of the theorem. t
u
References 79

References

1. G. Anastassiou, M. Mursaleen and S. Mohiuddine, Some Approximation Theorems for

Functions of Two Variables through Almost Convergence of Double Sequences, Journal of
Computational Analysis and Applications, Vol. 13, No. 1(2011), 37-46.
2. S. Banach, Théorie des Operations Lineaires, Warszawa, 1932.
3. F. Dirik and K. Demirci, Korovkin type approximation theorem for functions of two variables
in statistical sense, Turk. J. Math. 33(2009)1-11.
4. A.D. Gadz̆iev, The convergence problems for a sequence of positive linear operators on
unbounded sets, and theorems analogous to that of P.P. Korovkin, Soviet Math. Dokl.
15(1974)1433-1436.
5. G.G. Lorentz, A contribution to theory of divergent sequences, Acta Math. 80(1948)167-190.
6. F. Moricz and B.E. Rhoades, Almost convergence of double sequences and strong regularity
of summability matrices, Math. Proc. Camb. Phil. Soc. 104(1988)283-294.
7. M. Mursaleen, Almost strongly regular matrices and a core theorem for double sequences,
J. Math. Anal. Appl. 293(2004)523-531.
8. M. Mursaleen and O.H.H. Edely, Almost convergence and a core theorem for double
sequences, J. Math. Anal. Appl. 293(2004)532-540.
9. M. Mursaleen and S.A. Mohiuddine, Almost bounded variation of double sequences and some
four dimensional summability matrices, Publ. Math. Debrecen 75(2009)495-508.
10. A. Pringsheim, Zur theorie der zweifach unendlichen Zahlenfolgen, Math. Z. 53(1900)
289-321.
11. M. Zeltser, M. Mursaleen and S.A. Mohiuddine, On almost conservative matrix methods for
double sequence spaces, Publ. Math. Debrecen 75(2009)387-399.