approximation of unbounded functions with linear positive operators

172
APPROXIMATION OF UNBOUNDED FUNCTIONS WITH LINEAR POSITIVE OPERATORS R. K. S. RATHORE DELFTSE UNIVERSITAIRE PERS

Upload: iitk

Post on 14-May-2023

0 views

Category:

Documents


0 download

TRANSCRIPT

APPROXIMATION OF UNBOUNDED FUNCTIONS WITH LINEAR POSITIVE OPERATORS

R. K. S. RATHORE

DELFTSE UNIVERSITAIRE PERS

Approximation of unbounded functions with linear positive operators

111 III

1 mil

111 III III III III!

1

11 III!

nihil ;|ll!l i hll III If

mil 1

|:|;l!i|i hlllillll! Ill Hill n u l l

li! Ill ill

n U) o o o

mu XI

N) CD

BIBLIOTHEEK TU Delft

P 1835 6267

566306

Approximation of unbounded functions with linear positive operators

PROEFSCHRIFT ter verkrijging van de graad van doctor in de technische wetenschappen aan de Technische Hogeschool Delft, op gezag van de rector magnificus ir. H. B. Boerema, hoogleraar in de afdeling der elektrotechniek, voor een commissie aangewezen door het college van dekanen te verdedigen op woensdag 27 november 1974 te 14.00 uur door

RAM KISHORE SINGH RATHORE

Ph. D. in Mathematics (Indian Institute of Technology, Delhi) geboren te Lakhimpur-Kheri (U.P.) India

1974/Delftse Universitaire Pers

r

Dit proefschrift is goedgekeurd door de promotor PROF. DR. P. C. SIKKEMA

k

CONTENTS

INTRODUCTION ^ 1

CHAPTER I METHOD OF TEST FUNCTIONS 8

1.1 Approximation of functions having at the most

a polynomial growth when the variable tends to

+ 00 8

1.1.1 Definitions and notations 8

1.1.2 Convergence of the operators L 10

1.1.3 Asymptotic formulae for twice differentiable

functions I8

1.1.i* A class of linear positive operator sequences 32

1.1.5 Generalizations for functions of several vari­

ables Ul

1.1.6 On a generalized sequence of linear positive

operators 55

1.1.T A method of constructing operators for func­

tions of several variables 61

1.2 The trigonometric case 63

1.2.1 Asymptotic formulae for twice differentiable

functions 6h

1.2.2 Generalizations for functions of several va­

riables 79

1.3 Approximation of functions of an exponential

growth 86

1.3.1 Single variable case 87

1.3.2 Generalizations for functions of several va­

riables 91

V

CHAPTER 2 METHOD OF BOUNDING FUNCTIONS AND THE W-FUNCTIONS 97

2.1 Method of bounding functions 97

2.1.1 Basic convergence 97

2.1.2 Asymptotic estimates 99

2.1.3 Asymptotic formulae 1OU

2.2 Combining the techniques of bounding functions

and the W-functions 111

2.2.1 Asymptotic formulae 111

2.2.2 Asymptotic estimates 117

CHAPTER 3 APPROXIMATION OF UNBOUNDED FUNCTIONS BY OPERA­

TORS OF SUMMATION TYPE 121

3.1 A general outline 121

3.2 The Bernstein polynomials 122

CHAPTER k APPROXIMATION OF UNBOUNDED FUNCTION<=! BY OPERA­

TORS OF INTEGRAL TYPE 133

h.^ Approximation of \mbounded integrable functions 139

i+.l.l The general method 139

U.I.2 The generalized Jackson operators L lUo np-p

U.I.3 The Gamma operators G lU2

U.1.U Singular integrals W of Gauss-Weierstrass lU6

U.I.5 De La Vallee - Poussin integrals V lU8

REFERENCES 151

SAI.IENVATTING 16o

VI

INTRODUCTION

1. Schurer [ 59-62] , Hsu [ 17-18] , Wang [ 18] , Wood [ 12-14]

llttller [48] Eisenberg [l2-14]and several other

researchers have studied the possibility of

approximating a real (or complex) valued fiinction f(t),

defined on the real line or on a subset of it and

unbounded as t — + oo (or to some other point or points),

at its points of continuity by means of a suitable

sequence JL } (n=1,2,...) of linear operators ultimately

positive for the points (or a single point) of continuity.

In general, such a procedure assumes the convergence

L (g;x) — g(x) as n — oo, where x is a fixed point ot it

belongs to the set of points on which an approximation

is desired, for some test fvmctions g(t) and, in

addition, the uniform boundedness (uniform with respect

to n) or a kind of convergence as n -» oo ) of the

sequence JL (Q;X)J (n=1,2,...), where Q(t) is a suitably

chosen unbounded function. Under such assumptions we

arrive at a class (for which Q(t) is called a bounding

function) of unbounded functions f(t) which can be

aporoximated at a point t=x of continuity by the

senuence JL (f;x)j as n ->• oo . In this connection we

remark that almost all the work of above authors does

not use the full force of their assumptions and that

it is possible to enlar en the classes of unbounded

fljinctions for \jixich their results hold under the same or

even milder assumptions.

Recently l/alk [81] and Schmid [56] used the notion

of a socalled \/-function h(t), defined for t £ [0,oo )

and satisfying h(t) s 0, t 0 and h(t) /t-»ooas t-'oo,

and assuFiinj the uniform boundedness or a certain kind

1

of convergence of the sequence jL (h(|f|);x)j

(n=1,2,...) at a point x of continuity of the function

f(t), proved some converi^ence properties of the sequence

jL (f;x)j (n=1,2,...) to f(x) as n -* oo . This approach

does not require an explicit knowledge of the kind of

unboundedness of f(t), Kov/ever, v/hereas the use of a

bounding function gives rise to a v/iiole class of

ujibounded function.s for v/Lich the approximation takes

place, the use of a "W'-function h(t) requires computations

for the sequence {L (h(|f|);x) }for each f. Tne concept of

the bounding functions is certainly better than that

of the V/-functions since in any case h(|f|) is a

bounding function for the function f.

2_, \/hile developing a general theory of linear

combinations of linear positive oper-.tors, the author

[54] introduced a concept of local unsaturation of

linear positive operators on positive zero orders.

Besides linear combinations, this concept plays a very

useful role also in the theory of simulti.neous

approximation by linear positive oper .tors, linear

combinations of trieir iterates and the approximation of

unbjunded functions, Ilain results based on this concept

are described as follov/s-r

Let>(z) denote a class of non-nei^ative functions

F (t,x), (a ranging over an index set l) of two

variables t and x, \.'here each F (t,x) satisfies the a

follov/ing properties: (a) F (x,x) = 0, (b) for an arbitrary 6 > 0» F (t,x) has a positive lo\/er bound

a for all t satisfying | t-x | a 6 v/hile x retains a fixed

value, (c) keeping x fixed, P (t,x) is continuous at

t=x and (d) there exists a trcaisitive relation denoted

by the "greater tl.an" sign (> ) sue. t]iat c(,p £ I,a > P

2

imply

F^(t,x) (1) lim •• f,^\ = 0 (keeping x fixed)

t -. X ^p^^'^-'

and that given a positive nuxiber if there exists a

positive number M such that v/henever P. (t,x) g N

(keeping x fixed), there holds P (t,x) g MP (t,x), p ex

Let JL } (n=1,2,...) be a se'- uence of linear

positive operators such that for each a £ I and a

fixed X, the function P (t.x) £ D, the domain of

definition of the oijer-tor seouonce jL 1. Further

assume that for each a £ I there holds (2) lim L (P (t,x);x) = 0. \ / n^ a

n — CO

Let a,p £ I and a > p . There are the following

three possibilities (if necessary, restricting to

sub-sequences):

(i) L (F„;x) = o(L (F ;x))

(ii) L^/Fp;x) = 0(L^.^(P^;x)), and

(iii) L (F ;X) = o(L (P„;x))

as n — CO , Here the symbol 0 denotes a strict capital

order. The lollot/ing three theorems wore proved in [54]

i/hich cover the possibilities (i)-(iii).

TilSORSi I If a > p , tliere does not hold the relation

(3) L^(Fp;x) = o(L^(F^;x)) , (n -* oo )

provided the function P is uniformly bounded by P a

whenever P is unbounded (that is to say given an

arbitrary N > 0 there exists an k > 0 sucii that v/henever

P (t,x) S K, there holds F (y,x) g HF (t,x), and without

any condition of P is bounded. ..Iso (j) is false if

3

for some y > « > p, L ' Y''''' ^ ° ' n p'"" ' ( n--), without any further condition.

THEOREM II If L (P ;x) = e(-7--r) and L (P„;x) = 0( / v ) n^ ' ^cp(n)' n^ p 9(,n) '

where (p(n) is non-zero and tends to oo v;ith n, and

a > p. then for no y > a can there hold

(4) L (F ;x) = o(—7—r), (n -* 00 ).

TH' .REII III If for some a > p there holds

(5) \^'^^^^-) = o(L^(Pp;x)), (n - c- ).

Then for every |i < p ,

(6) LjPp;x) = o(LjF^;x)).

P^r convenience the functions F (a £ l) are called a

"positive zeros" and the subscripts a are called the

corresponding "zero-oraers". If P has a nigher zero-

ordor than F (i.e. a> p ), from the ^oint of view of r

approxima,tion it seems natural to expect that the

convergence of the sequence JL (p ;x)} to zero De faster

than that of the sequence jL (F ;x) j . Ile-.ce \/e say tnat

j L } is saturated at a(a £ l) if a higher oraer of

zero does not im ly a better approximation, that is to

say, for n O Y > o : ( Y £ I ) there holds

L (P ;x) = O(L (P ;x)), as n—00 , On the other hand if n^ Y n a ^" for some Y > a there holds L (P ;x) = 0 (L (P ;x)), v/e

' n^ Y ^ n^ a say that JL ! is unsaturated at a . '' ' n'

In case P (t,x) and F„(t,x) have comparable zero a p

orders of same pairty, i.e. F (t,x)

(7) lim -Tf (^ „ \ = P

where p is a positive number and further if given an

arbitrary positive II ttiere exists an II > 0 such that

F^(t,x) g II and Pp(t,x) g il respectively imply

P ( t , x ) g MF„( t , x ) and P „ ( t ; x ) S HP ( t , x ) , we v ; r i t e a p p a a ~ p . The fo l lov / ing theorem of [54] d e a l s v/ith t h e case

a ~ p.

TKE0REI4 IV I f a ~ P and {L j i s u n s a t u r a t e d a t p , t han

t h e r e h o l d s

F „ ( t , x ) L (F ;x )

^Q) 1^"^ F ( t , x ) = ^^"^ L (P =x) •

t - * x p ^ ' ^ n — oo n ^ p ' ^

A converse of theorem IV is also true. If for every

a ~ p the relation (o) holds and if there exists a Y > a

such that given an arbitrary N > 0 there exists an M > 0

such that F (t;x) g N implies P (t;x) g MP (t;x), then

JL I is unsaturated at p. This converse proposition is

not included in [54]» which, however, can be proved

rather easily.

The definition of unsaturati^-n of {L } at p implies

the existence of a Y > P such that L (F ;x) = 0

(L (p„;x)) as n -* oo, in this context it was remarked

in [54] that the functions F and F_ in theorem IV may

be replaced by more general functions f(t) and g(t)

which have the same order of zero at t=x, are positive

in a neighbourhood of this point (encluding the point x),

are bounded by a function P outside this neighbourhood where F is such that fL | are not saturated before YI

Y ' n' ' '

has a zero of an order higher than that of f or g and

has other properties of this notation,

An example of a class of the type Z ) is given by

the set of functions P = It-xI , a £ R , the positive

real line excluding zero where a > P has the usual

significance.

The theory of local unsaturation of positive linear

operators on positive zero orders (together with

5

theorems I-IV and the remark after theorem IV) provides

a great deal of insight into the study of asymptotic

approximation fcrnulae of Voi-onovskaja type (for tvj-ice

differentiable (at x) functions) and those containing

higher derivatives upto an even order. For instance a

necessary and sufficient condition that a sequence

jL } of linear positive operators possesses an asymptotic

formula for twice differentiable (at x) functions of a

certain growth (if unbounded) is that the sequence JL }

be unsaturated at a second order positive zero having

a similar growth. For a higher asymptotic expansion

conta.ining derivatives upto f (x) a similar necessary

and sufficient condition is local unsaturation at a th

2m ' oraer positive zero of an appropirate growth. Prom

this point of vie v various equivalent formulations of

xureckii's theorem ([10], p. 76) are all obvious.

Above discussion of local unsaturation is usefull

for having a macroscopic viev/ of various results obtained

in tills thesis. It will also help in obtaining various

equivalent foriaulations of tiie given results v/iiica may

be easier to deal \/itii a particular situti-ti-n,

5. Contents of the thesis.

Chapters 1-4 deal v;lth the deternination of classes

of unbounded functions, members of which may be a.pproxi-

mable v/ith the help of a given sequence of linear

positive operators.

Chapter 1 is an extension of Schurer's work [62],

Under the sane (or milder) hypothesis as in [62], we

obtain results v/hich are ap_.licable to functions having

an unboundedness of a higher order. These results can be

generalized for otlier test f-onctions forming a

6

Tchebychev system (extended, in case of asymptotic

formulae), To indicate some of the manipulations in­

volved we also include, in this chapter, some results

using the test functions 1, sin t, cos t,... and 1, + ott +2at (' -V n^ e— , e- ,... (a > 0;,

Chapter 2 generalizes some of the results of Kliller

[48] Hiiller and V/alk [50], Eisenberg and Wood [14] and

Sclimid [56] etc, based on the concept of bounding

functions and the \/-functions. Besides giving an

independent treatment based on the concept of bounding

functions we also make a unified treatment which

combines the two tecliniques of the bounding and the

W-functions, This unified treatment generalizes several

results of Sclimid [56].

Chapter 3 makes use of certain estimates connected

with the theory of local unsaturation of linear positive

operators (of summation type) on positive zero orders

and envisages its application in determining some

classes of unbounded functions which admit of an

approximation by means of these operators. We limit

ourselves to the case of the Bernstein polynomials only.

It v/ould, however, be clear that the method introduced

has a general applicability,

Chapter 4 deals v/ith the approximation of unbounded

but integrable functions by means of sequences of linear

positive operators of integral type. \/e limit our

attention to the generalized Jackson operators, the

Vallee-Poussin integrals, the Gamma operators of Iluller

and the singular integrals of of Gauss-Weierstra.ss,

Nevertheless, the method introduced seems to be of a

general applicability,

7

CHAPTER 1.

METHOD OP TEST FUNCTIONS.

The test functions used in this chapter are

1, t, t ,...; 1, s m t, cos t, ... and e (a > 0,

m = 0, + 1, + 2,..,). Parallel formulations of various

results proved can be obtained in terms of other test

functions (e.g. those forming Tchebycheff and extended

Tchebycheff systems). Some other useful test functions

are {exp a-r^, a g 0|, Jexp a(s ), ex £ E, the real

1 s s

line! f ll, t"'', t"^,...| andje"/"* , a g OJ.The first

three of these have been used by the author [54] in

connection with several operator sequences generated by

special functions. The last one provides a deeper

understanding of approximation properties of the Gamma

operators of Miiller introduced in [45]•

1,1. Approximation of functions having at the most a

polynomial growth when the variable tends to + QQ.

1,1,1.Definitions and notations

Let X, X be two given subsets of R and let

L ( X ) , D(X) be two linear spaces of real (or complex)

valued functions defined on X and X respectively (the

scalar field being the set of real (or complex)

numbers). In the complex case it is assumed that if

f £ D(X) then also its complex conjugate f £ D(x), Let

U be an unbounded index set of positive real numbers

and let {L , n £ Uj be a class of linear operators

8

mapping D(X) into D(X), The class {L , n £ Uj is said

to be ultimately positive on a point set X (by definition

X S.X) if to each f £ D(X) and satisfying f(t) g 0 for

all t £ X, there exists a positive number n , say, such

that for all n £ U with n > n there holds L (f;x) > 0 o n^ ' ~

for all X £ X*. It would be convenient to call f(x)

respectively L (f;x)as the function and the operator

(L ) value of f at the point x, ..'ith n g 0, \ic defrne H „: class of all rea,l (or' cor.plex)valued functions n, X f(t) defined on X, to each of \/Lich there exist constants A,B > 0 such that |f(t)| < A+3(t)^ for all t £ X.

H ,,(x): class of all f £ H ,. \;hich have an extension m,X^ ' m,X

^ •

1 on R wnich is continuous at the point x.

H^''4(x) : class of all f £ H ^ which have an extension m,A m,X

f on R which is twice differentiable at the point x. In

the sequel f is used to denote both f and f. Similarly

the derivatives f'(x), f"(x),,., are denoted by

f'(x), f"(x),,.., respectively, Ii this context we

remark that the possibility of several extensions f

will lead to no ambiguity in our results.

<a,b> : some open interval (c,d) containing the closed

interval [a,b] .

H^ I <a,b> : class of all f £ H ^ which have an m,A ^ m,A

extension f on R which is twice differentiable at each

point of <a,b> with f" continuous at each point of

[a,b].

Let S be a subset of R. By K „(S) we denote m,A

the class of all f C H ^ such that for each x £ S, m,A

f £ H Y ( X ) ,

^y «Q,X' « Q , x ( - ) ' « Q,X^^) ' «^ Q,X<^'^> ^^^

H v ( S ) we d e n o t e t h e c l a s s e s of a l l f i i nc t i ons f such

9

(2) that for some positive m, f £ H ,,, H v(x), H^ v(x), /„>, m,X' m,X ' m,X H^ ^<a,b> and H ^(.S) respectively, m,A m,A (o\

Thus the classes H(x) and H^ ' (x) as defined

by Schurer [62] are identical with our classes H rjCx) (2) '

and E\ n(x) respectively, 2fR

1,1,2. Convergence of the operators L .

The following theorem is a generalization of a

theorem of Schurer [62], valid for functions of the

classes H_ Y(^) where m is a positive integer.

Schurer's theorem is related to the class H„ ^ (x) and

makes an improvement over a result of Bohman [5] and

Korovkin [30] (dealing with the class H„ -,([a,b]))

under the same premises. For m=1 we obtain the result

of Schurer. Unless otherwise clear form the context,

X will denote a fixed point.

THEOREM 1 Let m be a positive integer and m' an odd

positive integer such that m' < 2m, Let JL , n £ U)

be a class of linear operators defined on a common

domain D(X) (X S R) of functions into a domain

L ( X ) ( X S R ) of functions and ultimately positive on a

set 5f^X, Assuming that 1, t , t £ D(X) and writing

L (1;x) = 1 + a (x) n^ ' ^ n^ '

(1) L^(t'"';x) = x^' + p^(x)

T ('j_2m \ 2m / \ n^ ;x} = X + Yj (x) ,

where x £ X and n £ U, i f and only i f t he r e h o l d

(2 ) l i m a (x) = l im p (x ) = lim Y ( X ) = 0, \ / n^ ' ^n^ ' 'n^ ' ' n - » o o n - * c o n - » o o

10

then for each f £ D(x) r> H„ y(^) ^^ have

(3) lim Ljf;x) = f(x). n — oo

Further, let S ^ X be a compact set. Then for

each f £ D(X) r\ H„ yC^) relation (3) holds uniformly

in X £ S if and only if (2) hold uniformly in x £ S,

Note. Since in order to construct L (f;x), n £ U the

function values f(t) for t ^ X are not required, for

an arbitrary f £ H„ -^{x) rs D(X) one may expect the cm, A

convergence L (f;x) -» f(x) as n -» 00 only for the

points X £ 5?, the closure of X (b^ virtue of the

continuity of f). Tnus, at first sight it may seem

strange that in the formulation of above theorem we do

not include the hypothesis that x be a point or a

cluster point of X. In fact, such an explicit inclusion

would be redundant as it is already implicit in the

relations (1) and (2). To prove this assume that (1)

and (2) hold without x £ X. We show a contradiction. Let

p(t) be the polynomial as defined in (4) below. We have

shown in the sequel that for an arbitrary 6 > 0, p(t)

has a positive lower bound, say m., on the set

jt : I t-x I g 6] , if X ) X, there exists a 6 > 0 such

that (x-6, X +6) r X = 0. Hence for all t £ X there

holds the inequality

0 g 1 g m""" p(t).

By the ultimate positivity of JL ,n £ U] at the point

X it follows that for all sufficiently large n £ U

we have

0 g Ljl;x) g m '' L^(p(t);x).

11

From this, using (l) and (2), we reach at

lim L^(l;x) = 0, n -> 00

which IS a contradiction,

A similar remark would be seen to be applicable

m all further results on the basic convergence of

operator values,

Proof of theorem 1. In both the assertions, the

necessity part of the theorem is trivial. To prove that

the conditions are suflicient consider the polynomial

p(t) defined by

/ , \ / i_ . .2m _ , m' 2m-m' / , , \ 2m (4) P('t) = in "t - 2mt X + (2m-m')x .

By Descartes' rule of signs it follo\7S that if x 4= 0

then p(t) has at the most two real zeros, multiple

zeros counted after their multi^l"cities and tnat they

have the same sign. Clearly t=x is a zero of p(t), ,/hen

x=0, t=0 is the only zero of p(t) and we have p(t) > 0

if t 4= 0, If X + 0, v;e have p'(x) = 0 so that t=x is a

double zero and tnerefore p(t) has no other zero and

since p(t) IS of an even degree we nave p(t) > 0 if

t 4= X, hence for every 5 > 0, p(t) has a positive lower

bound on the set jt : | t-x| > b],

Since the op'--ra,tors L , (n £ U), are linear

it IS sufficient to prove the result for real valued

functions f £ D(X) rs H_ v(^)* ' hen the functions ^m, A

L (f) are also real valued. (This remark snail 'oe made n '

use of m all suosequent results). By the continuity

of f(t) at t=x, given an arbitrary e > 0, there exists

a 6 > 0 such that

|f(t) - f(x) 1 < e , (lt-x|< 6, t C X).

12

Also there exists a jjositive real number A such that

|f(t) - f(x) I < A p(t), ([b-x|g 6 , t £ X),

3y non-negativity of p(t), therefore

|f(t) - f(x)| < e + A p(t), (t £ X).

The functions c + A p(t) + (f(t) - f(x)) belong to

D(X) n, H„ y(^)« Hence by the ultimate positivity of

|L , n £ Uj there exists a number N > 0 such that

L ( c + A p(t) + (f(t) - f(x)); X ) g 0, n > N^,

By the linearity of L we then have

|L^(f;x) - f(x)| g 1 L (1.;x) - l||f(x)|

+ e L^(l;x) + A L^(p(t);x),

Using (l) and (2) we can find an Np > 0 such that

|L^(1;X) - I| 1 f(x)| < z

L^(l;x) < 2 I n > N^ .

and A L^(p(t);x) < e

Let N = max JN , N j. Then

|L^(f;x) - f(x)| < 4£ , n > N,

By the arbitrariness of e > 0 it follows that

lim L^(f;x) = f(x), n — oo

proving the first assertion,

To prove the second assertion, assume on the

contrary that under the given assumptions (3) does not

hold uniformly in x £ S, Then given an arbitrary c > 0

13

there exists a sequence jn, , n, £ Uj of numbers and a

sequence jx, , x, £ Sjof points satisfying n, - oo ,

k — oo and lim x, =x for some x £ S such that k - oo

(5) |L (f;xj ) - f(x^)l > e , k=1,2,.,. .

On the domain D(x) of functions define at x

a sequence {M, j (k=1,2,,..) of operators by the relation

(6) M^(f;x) = L (r;xj^).

Then, writing

Mj^(l;x) E 1 + %(xj^) = 1 + cej (x)

(7) ^it'^'ix) = xj' + p^(x^) = x" ' + p'(x)

Mj (t ^;x) = x^ + Yj, (xj ) = X ™ + Yj (x) ,

(the primes are just a convenient notation and do not

represent derivatives) since a , p , Y "* uniformly

on S as n -• oo and since x, — x as k — oo , it follows

that

(8) lim aj[.(x) = lim P^(x) = lim y^M = 0. k-»oo k-'oo k-»oo

By the first part of the theorem and (6) it follows

that

(9) lim Mj^(f;x) = f(x). k -* oo

Also since f(t) is continuous at t=x, lira f(x, ) 1 k k — oo

= f(x). Hence there exists a natural number N, such

that

(10) l n, '""k " ^ ""k l < e , k > Nj,

14

Since (l0) contradicts (5), the second assertion of the

theorem follows. This completes the proof of the theorem,

Remark 1, In the statement of theorem 1, (2) can

equivalently be replaced by

(11) lim {L^(l;x) - If = lim L^( (t-x)^°'3x)- 0, n -* OO n -* oo

(assuming that (t-x) £ D(X)), To prove this fact we

replace the polynomial p(t) by the polynomial (t-x)

in the proof of theorem 1 and argue as before.

Remark 2. In the light of theorem 1 we point out to

an improvement on the following result of Hsu [17] •

THEOREM I Let JL j (n=1,2,,,,) be a sequence of linear

operators such that for all large n and every f(t)

belonging to the domain of definition of JL j and

non-negative for -oo < t < oo we have L (f;x) g 0 for

all X £ [-1,1]. Let {a j be a sequence of real numbers

increasing to + oo with n(a 4= O) and let the following

limit relation

(12) lim L ((a t)^; a"''x) = x^ \ / n ^ n ' ' n ^ n -> oo

exist and hold uniformly for all values of x in every

finite interval, where k = 0,1,2,m,m+1,m+2; and m is a

non-negative even integer. Then for every function f(t)

defined and continuous on (-00,00) and satisfying the

condition

(15) f ( t ) = o d t D , ( t ->±00)

we have the limit relation

(14) lim L^(f(a^t); a~^x) = f(x), n -• 00

15

(_oo < X < oo). Moreover, this relation holds uniformly

on every finite interval of x,

Our improvement on this theorem is as follows:

if m=2, condition (12) is superfluous for k=3,4. If

m g 4, condition (12) is superfluous for k=1,m+1,m+2,

Also in the latter case condition (l2) for k=1 can be

replaced by (12) for any odd positive integer k < m,

To prove this assertion, consider the sequence

(L*j(n=1,2,,..) of operators defined by

(15) L* (f;x) = L (f(a t); a~''x), n=1,2,... \ ^ I n ^ ' ^ n ^ ^ n ^ ' n ^ ' ' '

where the L are as in the above theorem. Asstime that n

(12) holds uniformly for all values of x in every finite

interval for k=0, m', m, where m' is an odd positive

integer less than m. (in particular we may take m'=l),

Replacing L by L* and 2m by m in theorem 1, the result

of Hsu follows,

Remark 3. If the class {L , n £ Uj of operators is

restricted to the non-negative real axis (i.e. to

construct L (f;x), (n £ U), the values f(t) for

t £ (- °°, 0) are not required or equivalently

X :^R E [0, 00)), then theorem 1 can be generalized

to the following,

THEOREM 2. Let m, m' be two pooitive numbers with

m' < m. Let JL , n £ UJ be a class of linear operators

defined on a common domain D(X) ( X ^ R ) of functions

into a domain D ( X ) ( X C R ) of functions and ultimately

positive on a set X X , Assuming that 1,t ,t £ D(Xj

and writing

16

L (l;x) = 1 + a (x) n^ n^ '

(16) L^(t"'';x) = x""' + p^(x)

L (t ;x) = X + Y (x), n^ ' 'n '

where x £ X and n £ U, if and only if there hold

(17) lim a^(x) = lim p^(x) = lim Y^CX) = 0, n->oo n — oo n-*oo

then we have for each f £ D(x) n. H ^(x) ^ m,X '

(18) lim L^(f;x) = f(x), n — oo

Further, let S ^ X be a compact set. Then for

each f £ D(X) ^ H ^(S) relation (l8) holds uniformly m, A

in x £ S if ond only if (17) hold uniformly in x £ S.

Proof. Consider the function

I ,\ ,,m ,m' m-m' / • \ m q(t) = m't - mt X + (m-m')x ,

for t g 0 and for a fixed x g 0. Then t=x is a zero

of q(t). If y g 0 (y 4 x) would we another zero of q(t), m' m'

then u = X ,y would be two distinet zeros of the

function

/ N , m/m' m-m' / . \ m v(uj = m'u - mux + (m-m'Jx ,

Applying Rolle's theorem, there would then exist a m' m'

positive number ^ lying bet'..een x and y , such that

• ^ - 1 ,/„N „ m' m-m' ^

v ( U = ' n ^ - m x = 0 ,

m' Clearly this implies that ^ = x , vAich is a

contradiction. Hence the only non-negative real zero

of q(t) is t=x. It is clear that because m > m' we have

q(t) > 0 if t 4= X and that q(t) has a positive lov/er

bound on the set ft :It-xI g 6} for each 6 > 0,

17

Replacing the polynomial p(t) by the function

q(t) in the proof of theorem 1 and proceeding analogously

we can complete the proof of theorem 2,

1,1,3. Asymptotic formulae for twice differentiable

functions.

Next we consider the existence of an asymptotic

formula giving rate of convergence of L (f;x) to f(,x)

for twice differentiable functions. In connection with

Theorem 1 of [62], Schurer remarked that the

corresponding theorem of Bohman-Korovkin, although 2

utilizing the unbounded functions t and t , gives the

convergence only for bounded functions. Interestingly

enough, his next result. Theorem 2 [62] utilizes the

function (t-x) for some positive integer m, which

has the unboundedness of the order t , 11 | - oo and

yet produces a result only for f(t) = 0(t ), |tj -* <»,

With the help of the following theorem, with no extra

assumptions we can obtain the result for f(t) = 0(t ),

I t I — oo,

THEOREM 1, Let m > 2 be a positive number. Let

JL , n £ uj be a class of linear operators defined

on a common domain D(X) ( X ^ R) of functions into a

domain L ( X ) ( X ' ^ R ) of functions and ultimately positive

on a set Xc^x. Assume that x £ X and that the functions

1,t,t^, It-xl" £ D(X), If and only if there hold

18

and

^o(x) ^ L ( l ; x ) = 1 + —7—r + o(—7—v)

n^ 9 (n) > C n ) ' '? (x )

(1 ) L ( t ; x ) = X + "'/ ^ + o( / \ )

2 2 2 "' ' 1 L ( t ; x ) = X + 7—r + o(—7—v)

(2) L^(|t-xr;x) = o ( ^ ) ,

as n -> 00, v/here (p(n) 4= 0, q)(n) -> 00 as n -* oo,

t h e n f o r a l l f £ D ( X ) r H^^^(x) t h e r e e x i s t s t h e m,X

asymptotic relation

(3) L^(f;x) -f(x) = - ^ [f(.),^(x)

+ f'(x){T/x) - xw^(x)j + ^ ^ {?2(x)

-2x?^(x)+x2l'^(x)j] + o ( ^ ^ ) , n -' 00 .

Further, if [a,b] is an interval ([a,b]^X)

such that for each x £ [a,b], the function |t-x| £ D(X)

then for all f £ L(X) rs H^^^<a,b>, the formula (3) ' m,X ' ' ^^'

holds uniformly in x £ [a,b] if and only if (1) and (2)

hold uniformly in x £ [a,b] , provided that the function

?2(x) - 2x'?.(x) + X V (x) is bounded for x £ [a,b], Remark 1. If x is not a cluster point of X, for an

(2) f £ D(X) r\ VL- Y(X) we can have many extensions of f

which will assign arbitrary values to f'(x) and f"(x)

(the value f(x) being an exception if x £ X without

being a limit point of X). However, this fact does not

lead to a contradiction in relation (3)» since in such

a case 'I'.(x) - x F (x) and 5' (x) - 2x1'(x) + x 'If (x) are

automatically both zero. To prove this v/e observe that,

in this case, there exist positive constantsA and B

19

such

and

that for all t £ X

t-x < ii t-x 1

/-J. ^ 2 , -o , im (t-xj < B t-x

Using (2) we have then

L^(t-x;x) = o( ; j | ^ ) , (n -> 00 ) n- ^fsf\

2 1 and L^(( t -x) ;x) = o(rnn") »(n -^ °° ) ^

from which the assertion follows.

A similar remark is applicable in all further

results on asymptotic formulae and estimates and

therefore any more reference to this point will be

omitted,

Remark 2. If v;e assume that the function 2

l'p(x) - 2x?^(x) + X ? (x) is unbounded on [a,b] but

that (1) and (2) hold uniformly in x £ [a,b] then an

application of Holder's inequality shows that V (x) 0

must be unbounded on [a,b] . Consequently for all

sufficiently large n the function L (l;x) is unbounded

and therefore discontinuous on [a,b]. Thus in a

practical approximation method the unboundedness of

^^(x) - 2x 1' (x) + X 1' (x) is unlikely to arise,

Remark 3' In the second part of tneorem 1, dealing

with the uniformity of relation (3) on an interval

[a,b], we have made the assumption (in the definition (2)

of the class H -5.<a,b>) that the seG>.nd derivative f" m, A

be continuou" ,",t each point of the interv::l ^a,b].

However, Suzuki [74] (Theorem B, p, 451) in his

formulation (wnich is given v/ithout a proof) of Mamedov

[36] and Scnurer's tneorem (Theorem 2, [o2]) does not

20

assume this continuity. His statement, which we

reproduce exo.ctly, is as follows: (The sequence {L j

of linear positive operators, occuring in the following

theorem, is assumed to be mapping C[a,b] into C [a,b],

where [a,b] is a finite interval).

THE OH EIM I Assume that the sequence of linear positive

operators {L (f;x)j has the property that

L^(l;x) = 1 , X £ [a,b] ,

L (t;x) = X + —7—V + o(—7—v)f uniformly over n^ ' 9(n) ^cp(n)" ^

' o(x) [a,b],

L^(t ;x) = X + /^\ + o(-7^). uniformly over

[a,b],

If there exists a positive integer m ( > 1) such that

Lj^((t-x) ";x) = °('^(7T)> uniformly over [a,bj ,

then for each f(x) £ J [a,b ], we have

2f'(x)Y (x)+f"(x){ Y (x)-2xT. (x) j L (f;x)-f(x)= ^-^-7-.; ^ + o(-7^) n^ ' ^ ^ ' 29(n) ^(n)^

uniformly on [a ,b ] , a < a < b < b, where C •^[a,b]

is the set of all real functions f(x) of which the

second derivatives f"(x) exist in [a,b] and are bounded.

In the same paper Suzuki used theorem I to prove

Proposition 1, p. 432; Proposition 4> P« 434;

Corollary 1, p. 437; Corollary 2, p. 437 and Theorem 1,

p. 438 (in which actually Poposition 1 is used).

In Theorem 1 the operators L considered are such that n

they preserve linear functions and '{'p(x) is boundea,

twice continuously differentiable and not equal to zero

21

on (a,b).

In fact the statement of tneorem I is not true.

We give an example of a sequence L j of linear positive

operators mapping C[-2,2] into C[-2,2]preserving

linear functions and satisfying the assumptions of

theorem I witn ^p(x) bounded, tv/ice continuously

differentiable and not equal to zero on (-2,2) for

which the asymptotic formula given by theorem I does

not hold unii .irnly on [-1,1] for a function

f £ C^2) [_2,2].

Let jL j (n=1,2,...) be the Sequence of linear

positive operators, miapping C[-2,2] into C[-2,2J,

defined in the follov/ing way:

i[„, Jisdi^Js^^ , ,(,. vSujUi!; (2n+ JTt

L (f;: n^

(2n+ -l)::

1 < |x

• )

< 2

^ [i(x+ (2n+ ^)n

1,2,... , We have

r(x-

L^(l;x) = 1, x £ [-2,2] ,

L v"fc;x) n^ ' e [-2,2] ,

(2n+ •^)n

T rx2 N 2 y (x) . 1 -, L^(t ;x) = X + — ^ + o(-2) ,

4rt "n n

uniformly over [-2,2j, where

f |X| S 1,

'P(x) = 1, i x| g 1

Also 1-(| x!-l)^, 1 < Ixl g 2 .

. ((t-x)'^'^;x) = o ( ^ ) , uniformly over [-2,2] , n j^^

22

where m > 1 is an arbitrary positive integer.

Assuming that theorem I

f £ c'-^^[-2,2] we would have

Assuming that theorem I is true, for each (2),

1 t^ \ ff -S ^(x)f"(x) ^ / I N

8-11 n n

uniformly on [a^,b.] where -2 < a < b. < 2, Taking

a =-1, b.=1, for an arbitrary e > 0 we can then find

an integer N > 0 such that for all n > N. and x£ [-1,1]

there holds

|m2{f(x-t^) + f(x-^) - 2f(x)j - f"(x)| < e

-1 where m = (2n+T-)7i and f is a fixed element of (2) 1 C "^[-2,2], Choosing x=0 and —, in turn, we have

I 2 f{^) + f(-^) - 2f(o)j - f"(o)| < e , and

f(^) + f(o) - 2f(-i)j - f"(^)| < £ ,

whenever n > N., It follows from these two inequalities

that

|m2|f(|) + 3f(o) - 3f(^) - fi-i)]-{f"ii) - f(o)}|<2e

whenever n > N., By L'Hospital's rule

lim m2(f(|) + 3f(o) - 3f(^) - f(-i)! m — 00

= lim f (2f(^) - 3f'(-) + f'(-^)} 2 ' ^m' ^m' ^ m'' m -* 00

= lim m!f'(|) - f'(o)j - lim -^{f' ( )-f' (o)j m -• 00 m -• 00

+ lim f (f(-l) - f'(o)j m -* 00

= 2f"(o) - I f"(o) - f"(o) = 0.

23

Hence there exists a positive integer Np such that

|m^{f(^) + 3f(o) - 3f(-) - f(--)j| < £

for all n > Np. Thus for all n > N ,Np we have

f "( \ ) - f"(o)| < 3e . (2n+2)Ti

Consider the following choice of the function f.

0, if x=0,

f(x) =

x^sin -, if X £ [-2,2] and x 4= 0.

It is easy to check that f(x) £ C^^[-2,2] and that on

the interval [-2,2] its second derivative f"(x) exists

and is given by

0 , if x=0,

f"(x) = j (12x^-1)sin - - 6x cos -,if x £[-2,2],x40.

Clearly we can choose a positive integer N, such that

for all n > N, we have 3

|12( ——)^sin(2n-4)7i - —^—cos(2n-4)Ti| < e. (2n-t )7i (2n4^)K

Thus for all n > N , Np, N, we have

I 1

|sin (2n+2)7i| < 4e •

As e > 0 is arbitrary, this gives us a desired

contradiction.

Proof of theorem 1. Writing

24

(4) f(t) = f(x) + (t-x)f'(x) + % ^ f"(x)

+ {(t-x)2 + It-xrjh^(t)

and putting h (x) = 0, the continuity of h (t) at t=x DC 3C

fellows. Thus given an arbitrary £ >0, there exists

a 6 > 0 such that (5) |h (t)I < e , for all t £ X with jt-xj < b .

Since there exist positive constants A and B such that

|f(t)| < A + Bltl"^ for ail t £ X, it follows from (4)

that there exists a positive constant M such that

(6) Ih (t)I < M, for all t £ X with |t-x| g b .

Obviously we can choose M so large that both of the

inequalities

(7) l\(t)| < M, and

ra\ lu ^ ^ I / Mlt-xT"^ (8) h (t) < e + —' hi— ^ ' ' x^ ^ ' ,m-2

o

are valid for all t £ X.

By linearity and the positivity of L , using

(7) and (8) we have the inequality

(9) |L^(f;x)-f(x)L^(l;x)-f'(x){L^(t;x)-xL^(l;x)

- ^^{L^(tSx)-2xL^(t;x)+x^L^(l;;

n'

;x;

= lLj{(t-x)2+|t-xrjh^(t);x)|

S e L^((t-x)^x)+M(l4^j;^)L^(|t-x|"';x),

6

valid for all sufficiently large n.

It follows from (I) that, for all sufficiently

large n,

25

"f^M ^ ^ ^^ ^^ ^^^ ^T^(x)-xy^(x)

9 (10) |r(,){-°_^+1 -L^(l;x)j+f.(x)i ^

-L^(t;x)+xL^(l;x)j+ n

^„(^) Y2(^)-2x^-,(x)+x^?o(x) 2 5 ^(^

- L (t ;x) + 2 x L (t;x) - x L (l;x)j| g -f-y n^ ' n^ ^ n^ ' 'I 9(n)

and also that

(11) L^((t-x)2;x) g ^ ( )

where K is a positive number independent of n and e,

Again by (2) for all sufficiently large n

(12) M(1 . - ^ ) L ^ ( l t - x r ; x ) g ^ .

Combining above inequalities, it follows that for all

sufficiently large n

(13) |9(n)iL^(f;x) - f(x) - (7y[i"(x)y (x)

+ f'(x)fY^(x) - xV^(x)j + ^ f"(x){'?2(x)-2x? (x)

+ x^Y^(x)j]j| g e (2+K).

Since e > 0 is arbitrary (3) follows.

For the uniform convergence part, by a mean value

theorem

(14) f(t)-f(x)-f'(x)(t-x) = -^^=1^ f"(0

for some E, lying between t and x, if f" exists at all (2) points between t and x. If f £ H /. <a,b>, applying m, A

(Lemma 1, p,12, Korovkin [30]) to the function f", we

find that given an arbitrary e > 0, there exists a 6 > 0,

independent of x, such that

26

(15) | f " ( t ) - f " ( x : S £

f o r a l l t £ X such t h a t | t - x | < 6 , x £ [ a , b ] . Hence

from (4) we have t h e i n e q u a l i t y

(16) 2 | h ^ ( t ) | j l + | t - x r - 2 j = | f " ( 0 - f " ( x ) | § £

for all t £ X and satisfying |t-x| < 6 ,x £ [a,b].

It also follows that there exists a constant M such that

(7) and (8) hold for all x £ [a,b]. By the boundedness

of the function Y (x)-2xy (x) + x 'F (x), K in (II) can

also be chosen to be independent of x. In this way the

right hand side of (13) becomes independent of x and

the uniform convergence follows.

As the necessity parts in both the assertions of

the theorem are trivial this completes the proof of the

theorem.

COROLLARY 1,1 Let m^ and mp be two positive numbers

where m > 2, Let (L , n £ UJ be a class of linear

operators defined on a common domain D(X) (X^=.R) of

functions into a domain D(x) (X & R ) of functions and

ultimately positive on a set X £^X, Assuming that ~ 2 I i™1 I * imi I |m2 . ,

X £ X and 1,t,t , |t-x| and |t-x| |t| £ D ( X ) , if

and only if (1) together with the conditions

m . (17) L (|t-x| x ) = o{-i-v) ^ ' n^' I ' ' 9(n) and

(18) L^(it-x| ^ t ) 2;x) = o ( ^ )

hold as n ^ 00, where 9(n) 4= 0, (p(n) -* 00 as n -* 00 , then

the asymptotic relation (3) holds for each

f £ D(X) r. H^2) u). m^+mp.X ^ ^^

Further, assuming that [a,b]^X,|t-x| and

27

|t-x|"^'^ h i ^ G D ( X ) f o r a l l x £ [ a , b ] and t h a t t h e p

function T„(x)--2x'l'. (x) + x '? (x) is bounded on [a,b], (?) °

for each f £ D(x) r, E^ i ^<a.,-b>, (3) holds uniformly m.+mp,A

in X £ [a,b] if and only (l), (17) and (18) hold

uniformly in x £ [a,b],

Proof. The necessity parts in both the assertions of

the corollary are trivially true. In order to prove the

sufficiency parts, we observe that two positive constants

A and B can be found such that for all real values of t

the following inequality holds

m.+m m m. m (19) | t - x | g A | t - x | + B | t - x | | t | ,

m +m2 f o r each x £ [ a , b ] . Hence, assuming t h a t | t - x |

£ D ( X ) , f o r a l l n s u f f i c i e n t l y l a r g e

m +mp m (20) L ^ ( l t - x l ; x ) g A L ^ ( | t - x | ' ; x )

m mp + B L ( I t -xl It l ;x)

by (17) and (I8), Thus, in case (17) and (I8) hold

uniformly in x £ [a,b], the o-term in (20) holds

uniformly in x £ [a,b]. Now with m=m^+m^, theorem 1 is

applicable and the sufficiency parts of the corollary

follow, m.+mp

In case | t-x | ^ I^(x), we go back to the proof

of theorem 1 and replace the function |t-x| everywhere

by the function A|t-x| + B|t-x| |t| and proceed

analogously. This completes the proof of the corollary.

COROLLARY 1.2 Let m g 4 Ie an even positive integer,

Let JL , n £ uj be a class of linear operators defined

28

on a common domain D ( X ) ( X £ R ) of functions into a

domain D ( X ) ( X ^ R ) of functions and ultimately positive r^ '- r^ 2 3 4

on a set X X . Assume that x £ X and 1,t,t ,t ,t , ^m-4^^m-3^^m-2^^m-1 ^^^ ^m ^ ^^^^^ ^^^ ^^^^^^ ^^^^^ ^^^

Tp(x) be some functions of x. Then a necessary and

sufficient condition for (3) to hold for each (2)

f £ L(X) r> H^ ^(x) is that it holds for the functions

i,u,,..,u ancL u ,,.., X .

Further, assuming that [a,bj^X and that the 2

function ?p(x)-2xT (x) + x ¥ (x) is bounded on [a,b],

for each f £ D(X) r^ H^^2<a,b>, (3) holds uniformly in m,A

X £ [a,b] if and only if it holds so for the functions 1 + +4 ^ +m-4 .m

Proof. The necessity parts in both the assertions of the

corollary are trivial. To prove the sufficiency parts, 2

we note that for f(t) = 1,t,t , the relation (3) is

identical with the respective relations in (1). Now,

since f,f',f" occur linearly in (3), choosing m.,=4 and

mp=m-4 we find that (17) and (I8) are satisfied. Now

corollary 1.1 is applicable, completining the proof of

the corollary 1,2,

COROLLARY 1,3 Let m g 4 te an arbitrary positive

integer (i.e. not necessarily even). Let {L ,n £ Uj be

a class of linear operators defined on a common domain

D(X) ( X ^ R " ^ ) of functions into a domain D(X)(X^R'*')

of functions and ultimately positive on a set X.^X,

Assume that x £ X and 1,t,,.., t and t ,...,t £ D ( X ) ,

Let Y (X),'? (x) and Tp(x) be some functions of x. Then

a necessary and sufficient condition for (3) to hold

for each f £ D(X) H^^|(x) is that if holds for the ^ X • -IX x4 •, xm-4 xin functions 1,t,...,t and t ,...,t .

29

Proof. Choose m.=4, mp=m-4. Then (19) with |t| mp '

replaced by t ' is valid for all t g 0. Rest of the

proof follows along the lines of the proofs of the

corollary 1,1 and the corollary 1,2,

Remark 1. In the special case when ?p(x)-2x'i' (x) 2

+ X f (x) = 0, relation (2) is already fulfilled with

m=2. The second derivative f"(x) then does not occur

in (3) and this formula reduces to

(21) L^(f;x) - f(x) = ^[f(x)^o(x)

+ f'(x){¥,,(x) - xf^(x)j] + o ( - ^ ) ,

In fact the existence of f"(x) is not necessary in this

case and it is sufficient to assume that, with an

extension of f, f'(t) exists for t belonging to some

neighbourhood, say (c,d), of the point x and that the

first divided differences of f'(t), at the point x, are

uniformly bounded for all sufficiently small step-

lengths h, say |h| < 6 where 6 > 0. To prove this we

proceed as follows. By the assumptions on f'(t), for a

sufficiently small & > 0, there exists a constant M > 0

such that

if'(x+h) - f'(x)i I h 1 ^ ^ ^ '

whenever |h| < 6 (in this and the following step we are

working with an extension of f and so we need not

specialize x+h,t and ^ to belong to X), By a mean value

theorem if S > 0 is sufficiently small and I t-x | < 6,

we have

f(t)-f(x)-(t-x)f'(x) = (t-x)(f'(0-f'(x))

30

where ^ is a point lying between t and x and therefore

|(t-x)(f'(0-f(x))| g(t-x)2|^l^lM^Iil|.

Now we can specialize t to belong to X and have

(22) |f(t)-f(x)-(t-x)f'(x)I g M(t-x)^

wheij.ever |t-x| < 6 and t £ X, Rest of the proof follows

on the lines of the proof of theorem 1,

Formulae of type (21) occur in certain

perturbations of the Baskakov sequences recently studied

by Sikkema [70].

Remark 2 In the case when (2) or similar conditions

do not hold or are not known to hold, the following [l 1 result may, nevertheless be applicable. By H^ ^(x) m, A

we denote the subclass of H „(x) consisting of the m,X

functions f which, with an extension, possess a first

derivative at each point in a neighbourhood say (c,d)

of the point x such that at the point x all the first

divided differences of f'(t) with sufficiently small

step-lengths h, say |h| < 6 where 6 > 0, are uniformly

bounded by M„(x), say, i.e. |f'(t)-f'(x)| < M^(x) |t-x|

for all t satisfying 0 < |t-x| < b for a sufficiently

small 6 > 0. If for each x £ [a,b], f £ H'-''-!(X) and if '- ' •'' m,X^ '

the set {M„(X), X £ [a,b]j of numbers is bounded, say by M > 0, then we write f £ H- -! [a,b]. It is clear

that such an M exists if for same 6 > 0, f' £ Lip,,

on the interval [a-6,b+6] ; also that it is necessary

to have f £ Lip. 1 on the interval [a,b]. f

THEOREM 2. Let m g 2 be an even integer. Let [L ,n £ Uj

31

be a class of linear operators defined on a common

domain D ( X ) ( X S - R ) of functions into a domain D(JJ)(X^R)

of functions and ultimately positive on a set S!^X.

Assume that x £ X and 1,t,t ,t ~ ,t ~ and t £ D(X),

Then for each f £ D(x) r^ H'-''J(X) there holds

(23) L^(f;x) - f(x) = O ( ^ ) ,

as n -* oo, where (p(n) 4= 0, q)(n) - ooas n — oo, if and

only if it holds for the functions 1,t,t ,t ,t

and t .

Further, assiuning that [a,b] X , for each

f £ D(X) HL''^[a,b], (23) holds uniformly in x £ [a,b] m,A 2 _._p

if and only if it holds so for the functions 1,t,t ,t , ,m-1 , ,m t and t .

/ +

Also, in the case when X,X R , the above asseri

assertions are true even when m is an odd integer g 2.

A proof of theorem 2 can be given in a way similar

to that of the proof of theorem 1 where instead of

using (4) we start with the inequality (22) of remark 1.

We omit the details.

1.1.4 A class of linear positive operator sequences

In tnis section we apply the results of sections

1.1.2-1.1.3 to the case of the Baskakov-eequences of

linear positive operators and thereby extend an earlier

study of them made by Schurer [62]. First we give a

brief resume of the results of Schurer.

Schurer [62] considered the sequence jL j(n=1,2,...)

of operators defined by

^ cp' '(x) X

(1) L (f;x) = s (-i)^-^Hn ^(zh\)' n k=0 k! '^^(K)

32

THEOREM III Let I9 (x)j be such that we have the

special case

x(n) = n, 'i'(n,x) = n,

'?(m(n),x) = m(n), m(n) = n+c,

(for all sufficiently large values of n), where c is an

integer indepenuent of n and a, (x) are independent XV, n

of k, say, a (x). If the a (x) possess the property

that at a fixed point x £ [o,b]

6) «n -) = - "(7(^)

v/here T(n) has the prop rties: I) T(n) 4= 0; 2)

lim T(n) = 00; 3) T(n) = o(n), n -* 00 , then we have n - oc , ,

for f £ Up ^+(x)

(7) L„(f,x) - r(,) . -P(' j;;(-) . o ( ^ ) .

Before obtaining convergence thev,rems for classes (2)

larger than the classes Kp _^(x) and Kp 4.(x) of

functions, we note that no particular purpose is served

by taking n to be an integer in the definiti -n of the

operators L . We assume, therefore, that n £ U where

U is an unbounded set of positive real numbers. With

this stipulation the numbers m(n) also need not be

integers. Further, it v/ould be sufficient to assume

that the properties i)-iii) hold for all sufi'iently

large values of n.

The follo\/ing theorem extends the result of

theorem 1 of Scaurer to the classes H p+(x) and

Ji„ -n+ -fl j wituviut any extra assumption.

ThliOREM 1 Let f £ H p+(x), x £ [o,b]. Then for the ' ' ' ' i.6 f -iL

35

where jx(n) } (n=1,2,...) is a sequence of positive

numbers increasing to infinity with n and the sequences

W (x)j(n=1,2,...) of functions possesses the following

properties on an interval [o,b] (o < b < °o) s

i) V^{°) = 1

ii) 9 (x) is infinitely differentiable and

(-l)^9^^\x) g 0, (k=0,1,2,...);

iii) there exists a positive integer m(n), not

depending on k, such that

- 9i")(x) = .(n,x)cpif;))(x) il.a,^,(x)},

k=1,2,..., where

iii) a, (x) converges to zero uniformly in k when

n -* oo, and

iii)p 'i'(n,x) satisfies the following properties:

(2) li, Ii^4- =1, and n -• oo A\ /

v\ -, • 'y(m(n) .X,

3) lim —^—^Y^—' n -* OO

Here and in theorems I-III it is assumed that f

belongs to a certain class of functions for which(l)

for n=1,2,... is meaningful. This depends on the nature

of the functions 9 (x).

Remark. It is known that a function 9(x) satisfying

(_l) (pV -'(x) g 0 (k=0,1 ,2,...; X £ [o,b]) has an

analytic continuation for |x-b| g b. Hence for 0 g a g b

the series

00 , vk ( k ) / N k „ (-1) 9'- aa)x

k' k=0

33

has a radius of convergence > a. It follows that for

m=U,1,2,..., the series

°° / ^ \k (k) / N k, m (-1 ) 9 (x) X k

k' k=0 ^•

is convergent for each x £ [o,b] . As a Cv.nsequence, if

f(t) is a function bounded on each bounded subset of R

and satisfies f(t) = 0(t"), (t — oo), for some m > 0,

then (l) for n=1,2,... is meaningful for sucn a function.

Schurer proved the follov/ing tneorems l-III:

THEOREM I If f £ H^ j +(x) and if the sequence 9 (x)

satisfies the conditions (i)-(iii) then the sequence

{L (f;X)}(n=1,2,...) defined m (1) converges to f(x)

when n — 00, if^ moreover, m (iii)-, a, (x) converges I iC, n

to zero uniformly in x on [o,b] and if the relations

(2) and (3) hold uniformly m x on [o,b], then the

sequence JL (f;x)j (n=1,2,...) converges uniformly on

[o,b] to f(x), assuming f(x) £ Hp j +(x) (O g x g b).

THEOHEIi II Let {9 (x)j be such that we have the special

case

x(n) = n, 1'(n,x) = n

'i'(m(n),x) = m(n), m(n) = n+c,

(for all sufficiently large values of n), i/htre c is an

integer independent of n and a, (x) are independent rC I n

of k, say a (x). If the a (x) possess the property that

at a fixed point x £ [o,b]

(4) a (x) = ^ + o(l) ' n^ ' n ^n^

t hen we have l o r f £ lip n+(x)

(5 ) L ( f ; x ) - f ( x ) = 2 x p ( x ) f ' ( x ) 4 - f " ( x ) ( x + c x ' ) ^ _^^1^ \ ^ / j ^ \ ; / \ / 2n ^n^

34

TIIEuREM III Let {9 (x) j be such that we have the

special case

x(n) = n, 'i'(n,x) = n,

?(m(n),x) = m(n), m(n) = n+c,

(for all sufficiently large values of n), v/here c is an

integer independent of n and a (x) are independent K, n

of k, say, a (x). If the a (x) possess the property

that at a fixed point x £ [o,b]

v/here T(n) has the properties: I) T(n) 4= 0; 2)

lim T(n) = °o; 3) x(n) = o(n), n -* 00 , then we have n - 00 , .

for f £ i4^^+(x)

(7) L„(f,., - f(x) = ^ ^ ^ i ^ ^ . o ( - ^ ) .

Before .ob ta in ing convergence the. . rems fo r c l a s s e s

(2)

larger than the clasoes K„ „x(x) and lu T)+{^) of

functions, we note that no particular purpose is served

by taking n to be an integer in the definition of the

operators L . We assume, therefore, that n £ U where

U is an unbounded set of positive real numbers. \/ith

this stipulation the numbers m(n) also need not be

integers. Further, it v/ould be sufficient to assume

that the properties i)-iii) hold for all suffiently

large values of n.

The follo\/ing theorem extends the result of

tneorem 1 of Scnurer to the classes H p+(x) and

H„ P+<a,b>, witiKiut any extra assumption. THE OREL: 1 Let f £ K ;+(x), x £ [o,b]. Then for the

35

operators L defined in (l), there holds L^(f;x) -* f(x)

as n — oo ,

Also, if a, (x) - 0 uniformly in x £ [a,c]

(O g a < c g b) and if in (iii)p, (2) and (3) hold

uniformly in x £ [a,c], then L (f;x), (n £ U), converges

uniformly on [a,c] to f(x) as n -« 00 for each

^ e \,R+<"'^>-

Proof. For k g m, where k and m are positive integers,

we have

. m-i m-1 a

, m , , m k = k! s J-. -T-yr

1 = 0 (k-i"+i)'

where Q ~ , i=0,1,,..,m-1 are the Stirling's numbers

of second kind ([22], § 58, p. I68-I73). Hence for an

arbitrary positive integer m \/e have for k g m the

relation

(k)/ V k „ / ., \k (k)/ X k m-i r.^ . . ^ k ! L ™ L k" m-1 G ^ O ^ ( n ) x a„ (8) (-1) — Y i ;r—=.^^ (k-m+i)

X (n) 1 = 0 ^ , \ "" / -\k-m (k-m)/ \ k-m

m__m „ ^ (-1) m (x) X a X m-1 . ^ im m ? (n). -_nMm-(n),x)(l+a^_ i )(x))} ^ n) 1=0 » \ / X (n)

m-1 m-1 2

+ ^S 'n Mm^(n),x)(l+a, . u ^(x))} . x"'(n) i=0 k-i,mi(n)^ ^ I

/ .\k-m+1 (k-m+l)/ N k-m+1 (-1 ) m . (x)x ^ ^ ^ m-1/ N^ '

m (n) • (k-m+1)! + ... ,

v/here m (n), 1 = 0,1,2,... are defined inductively by

o m

(n)=n, m (n)=m(m (n)), m (n)=m(m (n)), and so on.

In the following we will use the easily verifiable fact

that a = 1 and a = m(m-l)/2. Omitted terms correspond m m ^ ' ^

36

to the values of i < m-1.

For the values of k g m-1, the only important

case is when k=m-1. In tnis case there holds

m-1 o" - (m-i)'" = (m-1)! E jf-jy ,

1 = 1 ^ ^'

and as in (8) we have

/ .Nm-1 (m-l)/ \ m-1

^^^ (^iryi m. ^ ^ X (n)

m-1 m-1 _ (-1)% (°) (x) x° a X m - 2 . ^ ' Tiii_-i\ / m n {?(m^(n),x)(l+a . . ±. Jx))] ^

._„' ^ \ /» /\ m-1-i,m-^(nj^ "' 0! x"^(n) i=0

"T* • • • •

Using" this and (S) we have

(10) 2 ( - 1 ) '

( k ) / N k -^ ( x j X ,,m

i n k ! ^f \ k=0 X (n )

m m a X oo m - i = — 2 [ n W ( m ^ ( n ) , x ) , x ) ( l + ( x , . ±f s ( x ) ) i ,

m, s , n . r^ \ /» /» / \ m + k - i , m ^ ( n ) ^ ^ ' ' X (n) k=0 1=0 ' ^ ^ / . \k ( k ) / N k ( -1 ) 9^^ ( x ) x ^m-1 ^m-1 ^ ^_^

. ^ ] + 2 [ n ^' x'^Cn) k=0 1 = 0

{ ? ( m ^ ( n ) , x ) ( l + a . , . ±f N ( X ) ) J , ' \ '» / \ m-1+k-i,m-^(nj ' ' *

/ . s k ( k ) / N k

^-'^ ^ m-V , ^^ k! ] + • • • !

where the omitted terms contribute a o(l/ x(n))

quantity,

From (2) and (3) it follows (since by (3) we have

37

m (n ) — oo as n -* oo) t h a t f o r each f i x e d 1 = 0 , 1 , 2 , . . .

(11) n — oo

, i , K 4 ^ = 1 . xTnT

From this and (iii) it follows that given an arbitrary

£ > 0 we can choose a positive number N such that for

all n > N we have

and

m-2 n

i=0 (13) 1-£ < ""ii j('^("^VM))(i+a , ^ . u (x))j < 1+£, ^ ' •-n x(^) m-1+k-i,m-'-(n)^ ''

We can choose N so large that if n > N then the

contribution of omitted terms in (lO) does not exceed

e/x(n) in absolute value. Then it follows from i), (10),

(12) and (l3) that if n > N

m-1 m-1 a X

(14) L (t ;x)-x < £x + j—T— (1+e) + —7—r . ^ ^ ' n^ • ^ I x(n) x(n)

Since e > 0 is arbitrary, we have

(15) lim L^(t"';x) = x'". n -* 00

Noting that m in (14) is an arbitrary positive

integer and that L (l;x) = 1 for all n, from theorem

1.1.2,2 we have

(16) lim L^(f;x) = f(x) , (x £ [o,b]) n -* 00

for each f £ H p+(x) where m is any positive integer

g 2. This is equivalent to the first assertion of

38

theorem 1.

If the conditions in the second assertion of

theorem 1 are satisfied then for x £ [a,c], I' in (14)

c m be chosen to be independent of x. It follows from

(14) that (15) then holds uniformly in x £ [a,cl and

the second assertion of theorem 1 follows from the

sec md assertion of theorem 1.1.2,2. This completes the

proof of theorem 1,

In the next two results we improve upon theorems

Il-lll of Schurer, Again we note that no extra

assumptions are involved,

THEOREM 2 Let the operators L defined in (1) be such

that v/e have the special case

x(n) = n+p, m(n) = n+c, il'(n,x) = n,

where for all sufficiently large values of n, p and c

are constants independent of n. Further if for all

sufficiently large n, a, (x) are independent of k, say

a (x), and satisfy

(17) a (x) = - e ^ + o(l) , n ^ 00 ,

(2) at a fixed point x £ [o,b], then for all f £ H); p+(x)

(18) L (f;x)-f(x) = 2(p(x)-p)xf'(x)+x(cx+l)f"(x).,^(l^ 2n

as n

Further, if (17) holds uniformly in x £ [a,c]

(0 g a < c g b), where p(x) is bounded on [a,c], then (2)

(18) holds uniformly in x £ [a,c] for each f £ Hl R+<S-,C^ Q, R

Proof. If ollows from (1) that for an arbitrary

positive integer m, we have

39

m m , o x m-1

L (t"';x) = - S n [(n+ic)!l+ ^^4^ + o(l)j] n^ ' ^ / sm . „ '- '' n+ic ^n''-'

(n+p) 1=0

m-1 m-1 a X m-2 m

(n+p)'" 1=0 n [(n+ic)ji + - a ^ + o(^)j]+o(-i) „ '- ^ ' n+ic °^n^ 1J ^n'

m m ^ fn+ic+p(x)+o(l), a X n ^—^ ^—'-] m . r. n+p '

1 = 0 m-1 m-1 „

a X m - 2 . /• \ ^ ^ \ y, S n ,n+ic+p(x)+o(l), ^^U

n+p ._„ ' n+p ' ^n^

= o™ x"" (1 + ^i^-^)'^ + "i(p(^)-p) m ' 2n n

m-1 m-1 m /1 \

n 'n'

Thus, putting the values of a and a , for ' ram'

m=0,1,2,...

(19) L^(t'";x) = x'" + " ("'-1)( +I)x"'~

+ I^(P(^)-P)^'" + o(l).

From (19) and the first part of corollary 1.1.3.1

the rel:ition (I8) is immediate.

To prove the second part of tiie rem 2, it is

easily verified tnat under the given uniformity

conditions (l9) holds uniformly in x £ [a,c]. Thus the

second part follows from (19) and the second assertion

of corollary 1.1.3.1.3-[1 ] [1 ]

Let RL jj!(x) and Ht i[a,b] respectively denote the Si«,A y,A Til

totality of functions of the classes HL J ( X ) and •m,X'

40

[1 ] H ^[a,b], m varying over all positive numbers. m, A

V/e have the following generalization of theorem IH

THEOREM 3. If in the statement of theorem 2 the condi

condition (l7) is replaced by

(20) a (x) = -44 + oi-r^) , ^ ' n^ ' x(n) ^T(n)' '

where T(n) 4= 0, lim T(n) = <», T(n) = o(n), n - oo, n - oo , ,

then \/e have for each f £ Hl v+^-^^

(21) L (f;x) - f(x) = P lf'l" + o{-K), n oo, R~^ -" ^ TUT'

Further, if (2) holds uniformly in x £ [a,c]

(O g a < c g b) where (x) is bounded on [a,G], then

the convergence in (21) is uniform in x £ [a,c1 for (2)

each f £ H^^+<a,c>,

Also, in tne first assertion above, the class

n\ /}+{x) can be replaced by the larger class H|- j+(x) U,n / s Q,K

and in the second asserti.,n the class R \ pj.<a,c>

can be replaced by the class H^ J^[a,c].

The proof of theorem 3 proceeds on the lines

similar to those of the proof of theorem 2. After

pr. ving the first two assertions, we apply remark

1.1.3.1 made in connection with the relation (1.1.3.21)

to obtain the third assertion.

1.1.5 C-eneralizations for functions of several variables

In this section we obtain generalizations of some

of the results of earlier secti ns, suitable for

studying the approximation of certain classes of real

or complex valued functions defined on a subset of a

41

Euclidean m-space R by means of a seauence of linear m

positive operators. Proofs of the result of this section

can be carried out on the ba,sis of the notions

intr-^duoed in earlier sections and along the lines of

the proofs of results in ([62], chapter 2) and [59]. Definitions, conventions -.nd notations

Let the ra-tuples (n.,n^,...,n ) and (p. ,p,.,... ,p ), 1' 2 ' ' m ' \x--|Fx-2» '-^m '

where m is a natural number, be denoted by the symbols

II and p respectively. This notation is consisitent with

that of Schurer ([62], chapter 2). 'v/e shall always

assume that II,p £ R" , the first hyperquadrant of R .

The notation R — oo signifies that n. — oo, j = 1,2,...,m.

Let U d R be such that the set! min n.: II £ U } "•"•'" 1 g j gm - "'

is unbounded. Let lL.,,N £ U \ denote a class of linear ' N' m'

operators mapping a linear space D(X ) of real or complex valued functions defined on the set X d R

m — ra into a linear space D(X ) of real or complex valued functions defined on the set X ^E_ R . V/e assume that

m m if f £ D(X ) then f, the complex conjugate -jf f, also belongs to L(X ). The symbol X in D(X ) represents an ° ^ m' •' m ^ m' - assumption that for f £ D(X ), II £ U and

^ ^ m ' m X = X(x-,x„,...,x ) £ X we can construct L (f;X)

\ -|» 2' ' m' n^ ' ' provided the values f(H), 3 = zi^^ ,1^, . ..l^) £ X^

.are given. An operator L,., II £ U , is said to be _ N' m'

positive on a set JT cz. X if for each f £ D(X ) the m — m' ^ m^

assumption f(H) g 0 for each E £X leads to L,.(f;X) > 0 ^^' m 1\^ ' ' ~

for each X £ X . The class (L,-, K £ U j is said to be m ' ir ^ m'

ultimately positive on a set X S: X , if to each m m '

f £ D(X ) and satisfying f(E) s 0 for each S £ X , m ./ o \ / m there exists a natural ntunber n such that L (f;X) g 0

for each X £ jt whenever min n. > n. The set 7. 1 g j g m J

42

may consist of a single point X £ X or may contain

more than one point.

In the sequel we shall deal with the following

classes of functions:

H : the class of all real or complex valued functions P,A f(H) defined on X for each of which there exist ^ ' m

positive constants C,D, say, such that there holds

m p. f ( H ) g C + D 2 |^.|^, for all S £ X .

j = 1 '^ ^

K „ (H): subclass of H ,. consisting of the functions P , A P , A ' m • ' m

f(s) which, with an extension, are continuous at the

point 3 = X. (2)

H^ / (x): subclass of H consisting of the functions p.A P«X - ' m • ' m

f(3) which, with an extension, are twice differentiable

at the point 3 = X, in the sense tliat at the point X

all the first and second partial derivatives of an

extension of f exist and there holds m

f(3) - f(X) = 2 (l.-x.)f' (X) j-1 ^ ^ ""j

., m m p + j 2 (^-x )(^ -X )f" (x)+o( 2 ( .-x ) )

m p if 2 (E.-X.) tends to zero.

j = 1 ' '

Let S c=-E , Denote by H „ (S ) the class of all m — m '' p,X ^ m'

• ' m functions f £ H ^ which with an extension are

p,X " (2) continuous at all points X £ S . By H^ /. <S > we denote m p,X ra ^' m

the class of all functions f £ H ^ which with an p,X ^ ' m extension are unif-^rmly twice d i f f e r e n t i a b l e on S , i . e .

m' *

43

twice differentiable at all points X £ S such that m

the above twice differentiability relation holds

uniformly in X £ S . fl 1 ^

R^ 4 (X) : subclass of H ^ of functions f(H) which P tA PfA • ' m ' m with an extension possess all first order partial

derivatives at the point X and in addition satisfy

m m p f(H) - f(x) = 2 (^-x.)f' (X) + 0( 2 ( .-x ) ) ^ a = i ^ ^ ""o j = 1 ^ ^

if 2 (t.-x.) tends to zero. The class of all

functions f(3) £ H"- ^ (X) for each X £ S and for ^ ' p,X ^ ' m

' m which this relation holds uniformly in X £ S is

ril " m denoted by E^ 4 [S ].

p.A ^ m-' •^' m

If for some p, f £ H (X) we say that P , A

f £ E^^ (X). The classes E^^^ (x), H^^^ (x), ' m ' m ' m

H„ ^ (S ), E),^ <S > and H^''^ [s ] are defined in the Q,A m fi.A m Q,X '- m-' ' m ' m ' m

same way,

In the direction of approximation of functions of

many variables Schurer [62] (also see [59]) gave the

following two theorems.

THEOREM I Let H ( X ) denote the class of all real

functions f(x) which are defined in R and which have ^ ' m

the properties

1) f(X) is continuous for X=X ,

2) f(x) = 0( 2 X.) when |x.| - oo (j = 1 ,2,... ,m).

j = 1 ^ ^

Let f(x) £ H(X ) and let jL j (n=1,2,...) be a sequence

of linear and positive operators defined on H(X ). If

we write

44

L^(1;X) = 1 + a^(x),

L (t.;X) = x + p (X), (j=1 m; n=1,2,...) n J J _ 'i»J

Ul p 1" p

L ( 2 t;;X) = 2 x; + Y„(X), j = 1 '^ j = 1 "

and if a (X), p .(x) and Y (l ) have the property that n '' n, J ' n^ ^ ^ •'

lim a (X ) = lim P . (X ) = lim Y (X ) = 0 n^ o' n,j o' 'n^ o'

n-*oo n-'oo '" n->oo (j=1,...,m), then v/e have

lim L (f;X) = f(X ). n ' ^ o' n -* oo

TiIEOREM II Let H^^^(X ) denote the class of all real

functions f(X) £ H(X ) of which all the second

° (2)

derivatives exist at the point X . Let f(x) £ H^ (X )

and let {IT J te a sequence of linear and positive

operators defined on H(X ). If the operators L,. have

the pr.iperty that in a fixed point X L„(1;X ) = 1 + o( / 0 iN^'o' ^cp(n.)'

Y. -(X ) + '^ X + o(^^)

9(n.) >(n.)' L„(t.;X ) = X .

(k,j=1,...,m; k + j)

L(t2;x ) =x2. , ! 2 i i ^ ^ o ( - ^ ) N' j' o' oj 9(n.) >(n.)'

J 0

L,,(t, t. ;X ) = X X . ir k j' o' °k ''

-(XJ

^2-k i^^o) 9(n^)

2:i,k^ o' f 1 N t'\^^_^\ + y \ + o(—7 v) + o(~7 V) »

9(n^) >(nj)^ >(j^k)

where 9(n.) + 0 and 9(n.) -* oo when n. — oo (j = 1,...,m). J J J

If there exist positive integers p. (j=1,...,m) such J

that

45

then we have

[?, .f +4('i'o • •-2X .?, .)f" ] m L i;j X. 2^ 2;j,j oj 1;j^ x.x.J

L,,(f;X J-f(Xj= 2 ^ ^ ' ^-L_ N^ ' 0 ' ^ o' ^ 9(n.

m 'I'o.v .-x„, ..v ' 2;j,k~ o 1;j •^i 2 p;k,.1 o.i-1;k, '-> k '- ., ^,,(1 )

0 +k

where the values of all functions ¥, f and f" are

taken in X , o

Remark Theorem II of Schurer is not correct. It

requires a modification. In fact a mere assumption on

the existence of all the second derivatives at the

point X is not enough. Following is a counter example.

In Rp, let an operator sequence JL j, (n=1,2,...),

be defined as follows:

Vf;(x,y)) =lf(x+J, y+^) -Jf(x-f. y-J).

n=1,2,..., where a and b are two positive numbers,

We have

\{M{x,y)) = 1,

L (t;(x,y)) = X,

and

\is',ix,y)) = y, \ { i ;(x,y)) = X + (a/n) ,

L (s ;(x,y)) = y + (b/n) , 2 L^(ts;(x,y)) = xy + ab/n ,

V(t-x)2^-^2.(x,y)) = (a/n)2'"+^

L^((s-y)2^^S(x,y)) = (b/n)2^+2^

46

for an arbitrary positive integer m.

The asymptotic formula given by theorem II becomes

L^(f;(x,y)) - f(x,y)

= iKg)^f;;,(x,y).(^)2f;,(x,y)j

Consider the follov/ing function

0 , (x,y) = (0,0)

f(x,y) = \ (. 2 2 xy ^ 2^ 2 » otherwise,

X +y

defined on the xy-plane. At the point (0,0) all the (2)

second derivatives of f exist and f £ H^ ((0,0)). We

have

f(o,o) = f^(o,o) = f^(o,o) = f;^(o,o) = f^y(0,0)=0,

f;;y(0,0) = -1 and f^^(0,0) = 1,

If theorem II would be correct we should have then

L^(f;("0,0)) = o ( ^ ) . n

However, an actual calculation shows that

n (a +b )

For an arbitrary choice of a and b the two results

are clearly incompatible,

As we shall see in the sequel, a correction, to

render theorem II applicable, would be to assume

further that f is twice differentiable at the point X ,

In the above counter example the function f is clearly

not twice differentiable at the point (0,0).

47

The following thejrem is a generalization of

theorem 1.1.2.1, a theorem of Volkov [77] and theorem I

of Schurer. It gives the basic approximation result

for the classes H „ (x) and H ^ (S ). p,X p,X m ' ra ' m

THEOREM 1 Let p and p'(p! < p., j=1,2,...,m) be two J J

m-tuples of even and odd positive integers, respectively. Let{L„,N £ U j be a class of linear operators defined

' N' m' on a common domain D(X ) (X <- R ) of functions into a

^ m m — m domain D(X )(X c;- R ) of functions and ultimately

^ m' m — m' ^ -ni r '* P-;

positive on a set X i^X. Assuming that 1, E,.>J m p. J

(j = 1,2,... ,m) and 2 •'' £ D(X ) and writing j = 1 ^ "

L,,(1;X) = 1 + a.j(x)

p! p! (1) h {l/^',X) = x ^ + p (X), (j = 1,2,...,m), and

N Jj a i.,Pj m Pi ^ V-

L ( 2 ^ • ;X) = 2 X J + y (X), n j^., J j^., J iJ,P

where X £ X and N £ U , i f and only i f there hold m m

(2 ) l im a (X) = l im p (X) = l im Y:- ^ ( X ) = 0 , N - oo ' N - o= ^"'Pj N - c-- ^"'P

j=1,2,...,m, then for each f £ D(X ) , K „ (x) we have ^ ' ' ' ' ^ ra' p,X ^ '

'• ' m

(3) lira L (f;X) = f(z), N - oo

Further, let S ^ X be a compact set. Then for ' m m ^ each f £ D(X ) r H (S ) relation (3) holds uniformly

m p • A m ' m

in X £ S if and only if (2) hold uniformly in X £ S .

Remarks 1-3 made in connection with theorem 1.1.2.1

have the follov/ing analogues; m p.

Remark 1 Let 1, 2 (E,.-X.) ^ £ D(Z ). Then in tiie

48

statement of theorem 1 relations (2) can be replaced

by the following

m p. (4) lim {L (l;X)-1j = lim L ( 2 ( .-x.) ;X)=0.

N - o o " ' N - o o ' ' j = l J > J

Remark 2 In the case of functions of many variables

there is the following generalization of Hsu's theorem

[17] for a single variable.

THEOREM 2. Let la j be a set of positive real numbers ' n . '

J increasing to infinity with n., j=1,2,...,m, N £ U . Let II.., N £ U I be a class of linear operators

' N' m^ L». : D(X ) -» L(X ) and ultimately positive on X where N ^ ^ m' ^ m' "" m

X ,X S R and X •^ X . Let X contain an open m m m m m m

neighbourhood of the point zero in R . Let the p'. p. m

functions 1, ^.J, ^.0 £ D(X ), j=1,2,...,m, where p and p'(p'. < p., j = 1,2,...,m) are two m-tuples of even

J J and odd positive integers, respectively. Further let the limit relations

k. _ k. (5) lim L ((a ^ ) ^; aJ x ) = x ^, j=1,2,...,m,

N - 00 '' "j J "j J J

where k.=0, p!, p. (j=1,2,...,m), exist and hold J J J

uniformly for all values of X in every bounded sphere

of R . Let f be a real or complex valued ftuiction m

defined on R such tha t for each II £ U the function n m ^^V^l' V 2 "n^m)^'^(\)-

1 2 m Then, if f £ H „ (x) where X £ R , there holds

' p,X ^ ' m'

the limit relation

— 1 — 1 (6) lim L„(f(a ^.,...,a ^ );(a~ x.,...,a~ x )) ^ ' ,, N^ ^ n. 1' ' n ^m" n^ 1' ' n m^'

N — CO 1 m 1 m

f(x).

49

Also, if S is any corapact subset of R and

f £ H V (S ), then (6) holds uniformly in X £ S . p,X^ m m

Remark J In the special case when X c:- R , we have

the follov/ing generalization of theorem 1.

THEOREM 3 Let X , X ^^ R" and let p and p' be two ^ m' ra m

m-tuples such that 0 < p! < p., j=1,2,...,n. Let J J

fL.-,II £ U j be a class of linear operators ' H' m' ^ L, : D(X ) -* L ( X ) and ultimately positive on a set IN m m -n I

X SE. X . Let the functions 1, .J (j = 1,2,...,m) and m ra f •'j \o » » I /

m p. 2 .' £ L(X ). V/riting

0 = 1 J ' " '

L^j(l;X) = 1 + aj.(x),

p! p' . (7) Lj .( J;X) = x^ ^ + Pj^^p,(x), j = 1,2 m,

J m p. m p.

L ( 2 ?; J;X) = 2 x ^ + Y: p(X) j = 1 ^ j=1 ^ ^ 'P

where X £ X and II £ U , i f and . nly i f m m ' ''

(8) lira a (X) = lim p (x) = lim y „(x)=0, N - oo •"' II - oo - ' ' P j II - oo ">^

j = 1 , 2 , . . . , m , t h e n f o r each f £ PI .. (X) r D(X ) ' ' ' ' p , A ^ ' ^ m ' i_ ' m

we have

(9) lira L ( f ;X) = f ( x ) . II - oo '

F u r t h e r , l e t S S^ X be a c^mpo-ct s e t . Then f o r ' m m ^ each f £ Il .. (S ) / I)(X ) r e l a t i o n (9 ) nolds u n i i ^ r m l y p,A m" ^ m' ^ ' •' i n X £ S i f and only i f (8) ho ld u n i ; j r n l y i n X £ £ . m m

Theorem 4 be lo \ / i s a j ^ e n e r a l i z a t i o n of ti ieorem

1 . 1 . 5 . 1 . I t a l s o g e n e r a l i z e s and improves ti:iejrera I I

of L c h u r e r ,

50

THEORni 4 let p. > 2, j = 1,2,...,m and let {L.„iJ £ U j

be a class of linear operators L,. : D(X ) -- D(X ) and L ^ m' ^ m'

ultimately positive on X S^ X where X ,X ^. R . Assume m m m' m m

that the functions 1, •, • , (j,k=1,2,...,m) and m p. ^ ^ ^ 2 U--X.1 "^ (where X varies over X ) £ D(X ), With . I ' J J ' ^ m' ^ m' X £ X if md only if there hold the relations

m

+ S o ( — 7 — r ) , L (1 X) = 1 + D J r- + s o( 7 ! N • - i 9 - ( n . ) . . ^m. in. ,

1 = V 1 ^ l ' 1 = 1 ^ 1 ^ 1 '

m 7 . . (X) m

(10) m <!' , , ( X )

L„(^^;X) = x^ + 2 ^'^^ V + 2 o(—^ r"

L , ( , , , . ; X ) = x , x . + ^ 2 ^ ^ ^ ^ ^ E ^ o ( ^ ) ,

(k j ) ,

L ( 2 \i.-x \ ^;x) = s o(-Tr-y),

j , k = 1 , 2 , , . . , r a , a n d m p .

( 1 1 ) L,,( V Ir _^ I J

J

a s II — CO , w h e r e 9 . ( n . ) 4= 0 , cp. ( n . ) ^ 00 v/hen n.-* 00 ,

i = 1 , 2 , . . . , m , t h e n f o r e a c h i £ H ( 2 ) ( X ) ^ D(X ) t h e r e p, A m •^' m

holds the asymptotic formula m

(12) L.-(f;X) - f(X) = 2 ^ - ^ [^,i(x)f(X) 1=1 ^ 1 ^ 1'

+ 2 {(V,..(X) - x.Y .(X))f' (X) + ^(V..-(X) . H 1ji J 01^ '^ x.^ ' 2''''2.ii ' 0 = 1

^j'"1j: - 2x.'i', , . (X) + x^? .(X))f" (X)l

J Iji J 01 ^ x.x.^ ^ ' J J

51

+ ^ 2 ( Y^, ..(.X) - X, T, ..(X) - x.'?-, . (X) 2 _ ^ 2kji^ ^ k Iji^ ' J Iki^ '

k, 0 -1 k + j

m + X, X. ^ .(x))f" (X)] + 2 o ( — r),

k J 01^ ^ Xj x ^ J ^^^ ^9^(n^)^'

as N -* °o . Further, let S be a bounded subset of X ' ra m m and assume that t e function '?(X) = 2 {?p..(x)

i,j=1 ^^^

- 2x.Y,..(X) + x^T .(X)j is bounded on S , Then, in J Iji^ ^ 2 o±^ m '

order that (12) holds uniformly in X £ S for each (2) ™

f e H^ v^S > r\ D ( X ) , it is necessary and sufficient p.X m ^ m'' •' ^' m

that (1) and (II) h^Id uniformly in X £ S .

Remark 1 If p' = (p', Pp,..., p') is an m-tuple such

that 2 < p'. < p., j = 1,2,...,m, then in theorem 4, J J

relation (II) can be replaced by the follov/ing

m p! P • -P '• m (13) L .( 2 U-x M ( U j M .1);X) = _2 o ( ^ - ^ ) ,

J = 1 ^ >J >J 1 = 1 T^V -[_/

as N ^ 00 , Here we have asstimed that the function m p! P • -P '• 2 k.-x. I ( U. I "^ -^+1) £ L(X ) for the points X . ., ' J J ' ^ ' J ' ^ m^

under consideration,

COROLLARY 1 Let p. > 2, (j=1,2,...,m), be even J

positive integers. Let {L,., N £ U 1 be a class of ^ ° ' L' _ m' linear operators L, : D ( X ) — L ( X ) and ultimately

~ J^ ^ m' _ m' ^ A positive on X cz X where X ,X ^ R . Let 1,^.,E,.,^.,

m m m' m m j J J P-:-4 p.-3 p.-2 p.-l p

^1,^^,^^^ ,^-'^ fl^'^ fl-'' and l . \ (j,k=1,2,...,m),

£ L(X ) , If X £ X , then in order that (12) holds for m' /pN m'

each f £ H^ .;> (X) r\ D(X ) , if is necessary and p.X ^ ' ^ m " " •^' m

sufficient tnat it holds for the above mentioned

52

functions, Fti' thp'", i'" S i s J, b-undsl ub "t of X ' m -1

and the function

ra P ¥(X) = 2 {l' ..(X)-2x.1'. ..(X) + X ¥„. (X)!, ^ ' • --1 2ji^ ' J Iji^ ' J 01^ )»

1 , J - I

where I',,..,*!'... and ? . are as in (l2), is bounded on 2ji' Iji 01 ^ "

S , then in order that (l2) holds uniforraly in X £ S m' (2) "" for each f £ H^ .;( <S > r L(X ), it is necessary and p, X m ^ m " "'

•^' m

sufficient that it holds so for the functions mentioned

above.

COROLLARY 2 Let p . > 3 , ( j = 1 , 2 , . . . , m ) , be p o s i t i v e J /\

i n t e g e r s and l e t X , X ^ R+. Let JL^, N £ U jbe a c l a s s ^ m' m m '^N' m'

of l i n e a r o p e r a t o r s L, : D(A ) -• D(X ) and u l t i m a t e l y

p o s i t i v e on X ^ ^ X ^ ^ . Let 1, ^j , ^ ^ ^ ] , ^k^j . ^ j ^ . ^ j ^ . p.-2 p.-l p. ^^J , -i,^ and l ^ , (j,k=1,2,...,m), £ J(X^), If

X £ X , then in order that (l2) holds for each m' ^ ' (2)

f £Ii^ ' (x) /-^D(X ), it is necessary and suffient that p,X ^ ' ^ m'' " •^' m

it holds for the functions enumerated above. Further,

if S is a bounded subset of X and the function "(X), m ra \ / f

as defined previously, is bounded on S , then in order ' ra

that (12) holds uniformly in X £ S for each (2) "^

f £ H^ . <S > r> D(X ), it is necessary and sufficient p,A m ^ m " •' ^' m

that it holds so for the above enumerated functions,

Remark 2. Referring to theorem 4, let us consider the

special case when there holds the relation

m P

(14) V(X) = _ 2 1^2ji^^)-2''/iai^^^ + /oi^^M =0-

1, J = 1

In case the relations (IO) are satisfied, as a

consequence of (14), ne can easily show that

53

(15) ?. . . . ( X ) - X, V, . . (X) - x . T , , . (X) + X, X. V . (X)=0, ^ -^' ^KJl^ ' k I j l ^ ' J I k l ^ ' K J 01^ ' '

k + 0 , i , j , k = 1 , 2 , . . . , m . Thus t h e formula ( l 2 )

r e d u c e s to

m (16) Lj^j(f;X) - f(X) = 2 — r ^ [ ? ^ . ( X ) f ( X )

i=1 ^1^ ±' i l l H i ^

+ 2 ('?, . . ( X ) - x . ¥ . ( X ) ) f ' (X)] + 2 o ( — v) ^ I j i ' ^ J 01^ ^^ x ^ ^^^ > i (n j_ ) ' ' j=1 •" " " "J

il -* oo, in which the second order partial derivatives

of f do not occur,

Analogously, as in the case of a single variable

(remark 1 following corollary 1,1.3.1.3), here also

it is not necessary to assume the existence of the

second partial derivatives and it is sufficient to

[l 1 ass-ome that f £ H"- j (x) r\ D(X ). Also, in the case of

P'\ '^ uniform convergence of (l6) for X £ S , it is sufficient

Til "* to assume that f £ K'- ^ [S ] r D(X ),

p,X '- ra-" ^ m' ' m

Further, we note that if (14) and the first three relations of (l0) hold, then we have

m p m (17) Ljj( 2 (^-x ) ;X) = 2 o ( ^ - ^ ^ ) ,

j=1 "^ '^ 1 = 1 ^1^ 1^

and irrespective of whether (ll) holds or not, it can

be shown that (l6) holds for each f £ H'-]-' (x) r. D(X ), ' P ,X ^ ' ^ m^'

• ' m where p' = (2,2,,..,2). Also when (14) holds for each X £ S and the first three relations of (10) hold m ^ ' uniformly in X £ S , it can be shown that (I6) holds

^ Til uniformly in X £ S for each f £ HL ,J [S ] L(X ).

m p.X ^ m-* ^ m^ m

Lastly, it is to be noted that v/hen (14) holds,

in order to prove (16) it is not necessary to assume

that \ \ ^ t (k + j;j,k=1,2,...,m), £ L(X ) and the

54

fourth relation in (lO) can be discarded altogether.

THEORSI-1 5 Let p. g 2, (j = 1 , 2,... , m ) , be even positive J

integers. Let JL,., N £ U j be a class of linear ' ' IM' m^

operators L„ : D ( X ) -» D ( X ) and ultimately positive ^ N ^ m' ^ m' '' ^ on X where X , X d R and X ci. X . Assuming that

m* m — m m — m 2 Pi~2 Pj-I Pj , N / N

^»^y^y ^j . ^j and ^^^ (j = 1,2,...,), e D^xj and X £ X , if and only if there hold

m' k. k. m

(18) L„(^-'';X) =x.^ + 2 0(— ] v). 1 1 - ^ , ^ ^ N^ J ^ J i ., 9i(n^)^'

j=1,2,...,m; k. = 0,1,2,p.-2, p.-l, p., then for each r-t 1 o o J J

f £ HL'-! (X) ^ D(X ) p,X ^ ' ^ m' ' m

m (19) L (f;X) - f(x) = 2 0( 0 , N ^ - .

i=1 ^i^ ±^

Further,if S is a bounded subset of X , then (l9)

holds uniformly in X £ S for each f £ H'-''j [S 1 r^D(X„), "' m p.X I- m-J ^ m"

if and only if (18) hold uniformly in X £ S . m " +

Also, when X , X c^R , the above results remain ' m' m —• m'

valid when p. g 2, (j=1,2,...,m), are positive numbers, J

not necessarily even positive integers.

1.1,6. On a generalized sequence of linear positive

operators.

Analogous to the sequence of the operators L

defined in (1.1.4.1) for functions of a single variable,

Schurer [62] considered operators L„, for real valued

functions of several varaable, defined as follov/s: i-+...+i^ - i. i^

/ -\ 1 m I/v^ 1 ^ oo oo CO (-1) 9 | A X ) X ...X

(1) L (f;X) = 2 2 ... 2 . , . , ., , ^^ i,=0 i^ = 0 i =0 ^r ^2 ^m'

1 2 m

55

„/ 1 2 ra ^

•^^7;T^'^?V^ 71:^^' where for each function 9|,T(X) the follovvring properties

are satisfied in a domain K of the first hyperquadrant

of R . m 1) 9|^(X) can be expanded in a Taylor's series in a

closed region K , K is the union (when X runs through

K ) of the closed spheres with centre at X and radius |x|; 2) 9i (o) = 1;

i +...+i -, 5) (-1) 9i(x) S 0, (i 1 =0,1,2,..,;X £ K);

4) there exist positive integers n. (j=1,2,,..,m) not J

depending on i ,i2,,..,i , such that in the region K we have

I-E. (2) -9d(X) = ?.(n.;X)9,- ,'^ —>,„ (X), ^ ' ^N^ ' J J '^lM-(n.-n. )E. '

J J J

• ^'' " " i n ^^^^ ' ( j = 1 , , , . , m ) ,

where a) a. (x) ( j=1 , , . , , r a ) converges uniformly

in i . to zero in K when n. — 00 ; J J

b) there ex i s t the pos i t i ve functions x- (n- ) J J

( j = 1 , . . . , m ) (used in (1)) (monotonically) increas ing to i n f i n i t y as n. — 00 with the property tha t in K

J ¥ (n ;X) ? . ( ^ ; X )

(3) lim ' z* V = lim - V ^ S — = 1 , X•(n . ) V . (n . J

J J

J i ^ + i 2 + . . . + i ^

By 9jj we mean , , 9^ ' 1 2 , III

o x , 0X_ . . . o X 1 ^ m

56

I = (i-i ij^). 1-Ej = (i-,,...,ij-1,...,ijj^), and

N-(n^-ir)E^ = (n^,..., n^,...,nj^).

In above we shall, however, allov/ n., n., (j = 1,.» ,m) J J

to assume arbitrary positive values not necessarily

integers. Moreover, as we shall be concerned with an

asymptotic behaviour of {I'^T] it would be sufficient

to assume that the properties l)-4) are satisfied when

min n. is sufficiently large, 1 g j g m ^

Similar to theorem 1.1.4.1, in this case we have

the following basic convergence result, which

generalizes and improves the result of (Theorem 10,

Schurer [62]) under the same assiunptions.

THEOREM 1. If f £ H^ R+^^) "* ^ ^^^ functions f^{X) * m

satisfy the above conditions l)-4) then L^(f;X) defined in (1) converge to f(x) when N - oo and X £ K. If,

moreover, a. (X) (j=1,...,m) converges to zero

uniformly in X £ K' K where K' is a bounded and

closed domain, and if the relations in 4) h) hold

uniformly in the same domain then L (f;X) converge

uniformly in X £ K' to f(x) for each f £ H„ T,+(K'). ' m

The following tv/o theorems improve the result of

(Theorem 14» Schurer [62]). They also generalize

theorems 1,1.4.2 and 1.1.4.3.

THEOREI'I 2. If, for min n. sufficiently large, the 1 g j g m '^

operators L„ defined in (I) have the special properties

that

X.(n . ) = n + p , (p being a cons tan t ) , J u J J J

(4) n. = n. + c. , ( c . being a consteint), J <J J *}

57

'?(n ;X) = n , J J

P,(X) . (4) a._^_(X) = a^ (X) = ^ — + o ( - ) , n ^ e c ,

0 J 0 J J

(j=1,2,...,m), where p. are functions in A, then for ( 2)

each f £ 'A\ 1+ (X) and X £ K we have 0 , J

m m

(5) L„(f;X) - f(x) = Z — ((p.-p.)x.f' J ' J J

1 2 ' 1 + 77 (x.+c.x.)f" j + 2 o(-i-),

2 ' J J j' X X 1 ^ n ^ '

as N -• ooj v/here the values of all the functions occurin

in (5) are taken at the point X.

If, further, (4) hold uniformly in X on a bounded

domain K ' S K with p.(x) bounded on K', then for each (2) s J

f £ H^^+<K'>, (5) holds uniformly in X £ K'. ' m

THEOREM 3» If in place of the last condition in (4) we have

P.(X) ., (6) an.(x) = - V ^ + o(—/—r) ,

'^ T^(n^) ^T^(n^)^ '

where T.(n.) =1= 0, T.(n.) -'oo, T.(n.) = o(n.), n. -«=

(j=1,...,m) then v/ith the remaining conditions of (4)

intact, there holds for each f £ E'- J+(x), X £ K, the -, , . ' m

relation x.p.(x)f' (X)

m j^j^ ^ n.^ ' m

(7) L ,-(f;x) - f(x) = z ^ ( y + .\-(Tih J=1 G J J=1 r J

a s II -> 00 ,

Further, if p.(x) (j=1,2,...,m), is bounded on a

boujided domain K' K and (6), (j = 1, 2,... ,m), holds

uniformly in X £ K', then (7) holds uniformly in X £ K'

for each f £ RL''j^. K' ,

58

Applying these theorems for the three types of

operators given below: i . - n . X . i .

oo oo m n .^ e "^ ^x."^ (8 ) L ( f ; X ) = 2 , . . 2 H -^ r—, ^ ,

i^=o i^=o j= i y i-, i

f (— —) ^n- ' ' n 1 ra

oo oo m n . ( n . + l ) . . , ( n . + i . - 1 ) (9) L ( f ; X ) = 2 . . . 2 n - J - ^ ^ . , ' ' ' ' .

i^=o i ^ = o j = i -y

- ( n . + i . ) i . i . i . (1+x ) J J x^ f ( - l - f ) , and

" "J 1 m

oo oo (n+a)(n+a+1 ) . . . ( n + a + i + . . . + i -l) (10) L (f,.X) = 2 , , . 2 T-, T -7 - J S —

^ i , = o i =0 ^^• • •* "-m-1 m

f, ^ -(n+a+i^+,.,+i^) i i i i ,(l+x^+,.,+x^) x ..,x^ ^ ( T ' - ' — ' '

(a is a positive integer or zero), which are defined

for X. g 0, (j=1,,..,m), we have, respectively, the

relations

x.f" (X) m J X ,x. ' m

(11) Lj^(f;X) - f(x) = 2 ^ + 2 o(^),N ->.x>, j=1 j j=1 j

x.(l+x.)f" (X) j^ 3^ x.x "• '

u u 2n. • „ "•

J=1 0 J=1 J

m y 0' i^i ^ 1 (12) L^(f;X)-f(X)= 2 2n. +.^ °(^)»^ ~'

and

1 " (15) L„(f;X)-f(x) = - [ 2 X. {2af' (x) + f" (x)i

J=1 J J J m

• . J , j^k x.x^^^)] +°(i)' ^-*-' J,k=1 J k

valid for f £ YS^IPC), x. g 0, (j = 1,...,m), and that ' "m

59

they hold uniformly in X on any bounded region K'^ R" (2) ^

for each f £ HJJ^^+<K'>. ' ^ (2)

Remark 1. For the classes H^ (X^) (as defined in

theorem I.I.5.II) the formulae (11)-(13) were obtained

by Schurer [62]. For a correct interpretation we refer

back to the remark made at the end of theorem 1.1.5.II.

Remark 2. Let us note that in the case of operators

(10) a direct application of theorem 2 is not possible.

It is due to the fact that in this case condition 4)

is not satisfied, as not only there is a change in the

j-th component n. but that all the components suffer a J

change. The essential effect of it is an addition of the quantity x,x./n to the function f^v • °^ theorem

1.1.5*4 and consequently the quantity

. m x, X. 1 k i 77 2 "• f" must be added on the right hand side 2 , . . n X, X. ^ k,j=1 k J

k + j

of (5) in theorem 2 to obtain the correct expression,

ignoring the non-satisfaction of 4). Thus we arrive at

(13).

Remark 3. In fact in the same way as in remark 2 above,

applying theorem 1.1.5.4 we can prove results similar

to those of theorems 2-3 for the operators in (1) for

which more than one component of N may suffer a change

in 4)» Thus instead of (2), if we have

I—E (14) -9j(X) = j(n.,X)9 I ^ (x).(l+ai_^_(X)),

k=1 j

where the functions ?.(n.,X) and a. (x) satisfy the J J

conditions a) and b) as before, then if for all N, with

60

min n. sufficiently large, we have n, =m,+C , 1 = j = m ' j j

where C are fixed constants (j,k=1,...,m), with all j

other evaluation of L„(t.t, ;X) shows that

C. C. (15) ^ ^k ,. , , V ^ ' ~r "^ ~r ' (J.k=1,...,m),

Thus the components suffering a change are related

(and in particular if all the components do change, we

can take n. = n.(n), j=1,...,m, and the operators

L„ can be thought of as some operators L of a single

index). Now, the required asymptotic formula will be

obtained by increasing the right hand side of (5) by the

quantity C.

1 " ^k •k 2 — - X, X. f" 2 , . . n. k J X, X. k,j = 1 J ' k J k 4= j

and replacing C. in (5) by C. (since the role of C. is

now taken over by the constant C. , j=1,,..,m),

The result of theorem 3, however, remains valid,

without any change, for the operators (1) with (14)

instead of (2) and n, = n, + C, , (j ,k=1,... ,m), j j

and the other conditions remaining the same.

1,1.7 A method of constructing operators for functions

of several variables.

Theorems 1.1.5.1-1.1.5.4 not only enable us to

study convergence properties of a knov/n sequence of

linear positive operators but they also provide us with

a method of constructing such operators for functions

of several variables with the help of those for ftmctions

61

of a single or fewer variables.

Thus, for instance, let {L^j,{L^ },..., |L^j be m

sequences of linear positive operators each defined

for functions of a single variable. Renaming these by

L (-;x ), L (-;x ),...,L (-;x ), consider the "l 2 m operator L^(-;X) constructed as follows: Let

f = f(H), 3 = ( ,^2 ^m)* Operate on f by L^

th ^ treating all except the m coordinate ^ as constants, Thus we get a function L (f;x ), Now operate on it by

^^ " th L . treating all but the m-1 coordinate t ..as n .' m-1 m-1

constants. Carry on this process upto an operation by

L . We define L„ to be the resultant operator. Thus n^ N

(1) L^(f;X) = L (L (...L (L^ (f ;xj ix^_^),., ;x2;x ) 1 2 m—1 ra

Now it follows from theorem 1,1,5.1 that if L i

separately have convergence property (L (f;x. ) ->• f(x. ) i

as n. — oo) for the classes

H . (x. ) or uniformly for the classes H „ (S.)), Pi,X(.) 1 H'\±)' ^"'

i=1,2,..., m, then the operators L„ have the convergence

property for the classes H (x) (uniformly for the P 1 A

' m classes H ^ (S ), in the second case), where X and S

P,X^ m ' m m

are the Cartesian products of X/.x, 1=1,2,..., m and

S/.\, 1=1,2,..., m, respectively.

Similarly, it follows from theorem 1.1.5,4 that

if L , 1=1,2,..., m, possess the asymptotic formulae i

of the type (1.1,3.3) given by

62

(2) L^_(r;x.)-f(x.)=-^[f(x.)f^(x.)+f'(x.)i^.(x,) 9i.-i

f"(x.) -x.f^.(x.)j + - ^ — i T2i(^i)-2^i^u(^i)

2 1 + X. ? . (x- ) 1] + o(—7 r) , n.- oo ,

1 01 i''-' ^9.(n.j' ' 1 '

for f £ H^^^ (x.) (or uniformly for f £ H ^ . <a. ,b.> p . , X / . N ^ i' •' p.,X/.s x'1 1 (i) i' (i)

(p. > 2), i=1,2,..., m, then for f £ H^^^ (X) (or •"- P' m

(2) " uniformly for f £ H 4- <S >, in the second case, where '' p,X m ' '

•^' m S is the Cartesian product of the intervals [a.,b.l, m 1- 3_» x J '

i=1,2,,.., m) we have the following asymptotic formula

for the operators L^

m . (3) L^(f;x) -f(x) = 2 ;-f^[foi(-i) f(^)

1=1 ^1^ i'

+ JT, .(x.) - x.'P .(x. )if' (X) + U„.(x.) ' 1x^ i' 1 01^ i') X. 2 ' 2i^ i'

1

P m - 2x:>F, .(x.) + xff .(x.)l f" ] + s o( / \)lI-oo.

1 1l^ l' 1 01^ l' ' X.X.-i . . m.(n.)^ I X 1 = 1 ^1^ 1

V/e notice that terms in the mixed derivatives

f" (kj j; kj j = 1,2,..., ra) are absent from (3). k J

In fact in this case it is possible to show that (3)

holds for a more general class of functions.

1.2 The trigonometric case.

In this section we shall be concerned mainly with

the asymptotic approximation of real or complex valued

2K-peri()dic functions by means of sequences of linear

positive operators. The test functions 1, sin t, cos t,...

63

being bounded on R, in the general setting the results

are applicable only to bounded functions. However, for

many particular operator sequences these results are

helpful in determining the asymptotic approximation of

unbounded functions as well. In the sequel we assume

that if a function f belongs to the domain of definiticn

of an operator L then so does J, the complex conjugate

of f.

1.2.1 Asymptotic formulae for twice differentiable

functions

Regarding an asymptotic formula for tv/ice

differentiable 27t-periodic functions Schurer ([62], p.

22) gave the follov/ing theorem.

THEOREI'I I Let f(t) be an arbitrary hounded 2Ti-periodic

function v/hich is tv/ice differentiable at a point

t=x £ [-''i,'']. Suppose that on the lata i"'/'/-l [-7t,7i] a

sequence (L j (n=1,2,...) of linear positive operators

L applies to all such functions f and possesses the

properties

L (l;x) = 1 + o(-jrUr) n^ ' ^9(n)'

^1 k "" 1 (1) L (sin kt;x) = sin kx + — S — r - + o(—7—r) ^ ' n^ ' ' 9(n) 9(n)^

'^2 k "" 1 L (cos kt;x) = cos kx + —*-?—r- + o(—7—^ ), n ' 9 (n) 9 (n) ''

k=1,2, where 9(n)j^ and 9(n)-* 00 when n-*co.

If there exists a positive integer m such that for the

point X, L applies to (t-x) and there Iiolds

(2) L^((t-x)2^-+2^x) = o ( - ^ ) , (n - ex.),

then we have

64

(3) L^(f;x)-f(x) = [4f(x) {cos x f-| -,(x)

-sin x'P2 ^(x)}-f"(x){cos 2x Yp 2^^'^'^^^^ ^x f a^^^H

: 4?^n) + o ( ^ ) .

(The formula (3) as it is given in [62]contains 2

instead of 4 as the coefficient of f'(n) on the right

hand side. However, it can be verified that the

correct value is 4 as is put in above),

Another theorem in the same direction, but for

a more special class of operators, was given by

Korovkin ([29], p. 99).

THE0REI4 II On the space T of bounded 2K-periodic

functions f integrable on [-7t,7i] let (L j (n=1,2,...)

be a sequence of trigonometric polynomial operators

defined by

n n (4) L^(f;x) = j f(x+t){| + 2 p "" cos kt jdt ,

-71

where (p^ ') (k,n=1,2,...) is a matrix of constants such

that

^ + 2 p^^^ cos kt g 0 ^ k=1 ^

for all t £ [-7i,n] and n=1,2,... and x £ R.

Then in order that for all functions f of T which

are twice differentiable at a point x, the relation

(5) L^(f;x)-f(x) = (l^5^b f"(x) + o(l-p^(")),n ^ o. ,

be valid, a necessary and sufficient condition is that

65

1_p(n)

(6) lira — ^ = 4. n - oo i_p^^ ^

Remark. Let us note that in theorem II the condition

(6) together v/ith the non-negativity of the kernels

1 ^ ( )

•p + 2 P, cos kt, (n=1,2,...), implies that for each k=1

f i x e d k = 1 , 2 , . . . , v/e h a v e

n • ( n ) l i m P^ ^ = 1 ,

n - CO k

and in particular for k=1 this limit relation asserts

that (5) is a meaningful p^symptotic approximation

formula and that the se' uence j L j of operators defines n'

an approximation process for a function f £ T at its

points of continuity.

For a proof 01 this assertion v/e notice that each

of the functions

J (1 + cos kt)!^ + 2 p^"'' cos jt !, (n,k=1,2,...)

j = 1 -^

is non-negative on L-' ,' ] and assumes 8, positive lower

bound on a subset of positive measure. Integrating

these functions beti een the limits -ft and n v/e therefore

have

-1 < P "" < 1 , (n,k=1,2,...).

Thus, in particular the boundedness of the sequence

{p\^^] and (6^ imply that

T • ( - (n 1 , (n), lira i^ + p ^ - - 4 plj 'j = .

n -* 00

Since this is tantamount to

-, . - f • A t-x N

lim 1. "in -2-;x, n -» 00

66

^ ^hich in turn by Gauchy's inequa l i ty im.plies tha t

lim L ( s in —r—> x) = 0 n ^ 2 ' ' n -, oo

and as L (l;x) = 1, (n=1,2,...), for all x, if follov/s

by Korovkin's theorem [30] tnat if f £ T and is

continuous at a point x, then

lim L (f;x) = f(x). n^ ' \ y

n — oo

In particular chuosing f = cos kt, (k=1,2,...) and x-0

the above assertion follows.

As regards theorem I of Schurer, condition (2)

fits very -./ell as long as the operators L explicitly

depend on the functi n values in the interval [-•rt,n]

(in fact it can be shown that in this case (2) is

necessary as v/ell). Iio\."ever, in other cases (2) may be

disadvantageous. To take an . .rtificial example, define

an operator L by L(f;x) = f(x), x ^ 0, x £ [-•n;,!:] and

L(I;O) = f(27i:). Clearly (2) is not satisfied even though

L(f;x) = f(x) for each 2Tt-periodic f. Also the function

(t-x)'"' being an algebraic one, the determination of

L ((t-x)''" ;x) itself may present difficulties, . s v/e

are interested in the apci-oximatio.x --f 27t-pGriodic

functi -ns, it seems reo.sono,ble to replace (2) by another

oultfiblfc condition involving 27:-periodic functions.

w'e aim at establishing the following results:

(l) In theoi-em I relation (l) corresponding to k=2 are

entirely superfluous, that is to say, the asymptotic

formula can be obtained in terms of "f, and ^, . only. 1,1 ^,1

Very recently T-Liamernans [ 75J already proved this fs,ct,

ti.oujjh along different lines, (ii) If, however, (2) is

satisfied and the opex-ators have a meaning for functions

67

(2) of the class Hp ' y(x), m a positive integer, then the

asymptotic formula is valid for functions of this class,

(iii) Condition (2) can be replaced by a necessary and

suffi^KN; condition which can be tested v-/ith the help

of the functions L (sin 2t; n) and L (cos 2t:x) n^ ' n^ '

(n=1,2,..,). Thus the relations (l), if known are

suffient to determine the existence and the form of

the asymptotic formula, (iv) The results that we obtain

generalize theorem II of Koiovkin, which can be obtained

as an easy corollary, \.e also consider the uniform

convergence of the asymptotic formulae in a closed

interval and obtain necessary and sufficient conditions

for this for certain classes of functions. Lastly we

give applications of our results and their genralizations

suitable for applications to the operators v/hich are

defined for functions of several variables.

(') By KX'~J^{x), (x £ R, X ^ R ) , we denote the class c. Tl, A

of a l l bounded functions f defined on X v/hich possess a. 27t-periodic extension f on R \\/hich i s tv/ice

(2) d i f f e r e n t i a b l e at the point x, Q \ <a,b> , 2Tt , A

(-00 < a < b < oo), denotes the class of all bounded

functions f defined on X which possess a 2Tt-periodic

extension f on R which ist-Afice differentiable in an

open interval containing the interval [a,b] such that

the second derivative is continuous at each x £ [a,b] .

In the sequel both f and f are denoted by the common

symbol f,

THEOREI i 1 Let j L , n £ UJ be a class of linear

operators defined on a common domain D(X) ( X ^ R) of

functions into a domain D(X) ( X ^ R) cf functions

and ultimately positive on a set X ^ X. Let 1,

68

sin kt, cos kt, (k=1,2), £ D(X) and let x £ 5', Then in (2)

order that for each f £ D(X) r Q.X ^(x) there holds ^71, A

t h e a s y m p t o t i c r e l a t i o n

( 7 ) L ^ ( f ; x ) - f ( x ) = - ^ [ f ( x ) v ^ ( x ) + f ' ( x ) ( c o s x^^ (x )

- s in X ' i '2(^)l ~ ^ " ( ^ ) {'^°^ ^ "^2^^^ " ^^"- ^ ^ i ( ^ )

-^^-)!^ - ° ^ ^ ) ' where 9(n) ^ 0, 9 ( n ) -' oo as n -* oo , i t i s n e c e s s a r y

and s u f f i c i e n t t h a t t h e r e ho ld t h e c o n d i t i o n s

^ ^ ( x ) ., L ( l ; x ) = 1 + 7—r + o(—7—v)

^^ ' ^ (p (n) ^9 ( n ) ' n ^ ' - ( • ( x )

1 ( 8 ) L ( s i n t ; x ) = s i n x + —T—r + o(—T—v) ^ •' n^ 9 ( n ) 9 ( n } ' ^ ^ ( x ) .,

L ( c o s t ; x ) = cos x + —T—r + o(-7—v—) n^ 9 ( n ) 9 ( n ) '

and

(9) L ^ ( s i n 4 i ^ ; x ) = o ( ^ - ^ ) ,

Further, if [a,b]^ X and if the function

cos xy^/(x) + ;ln x; f. (x) - T, (x) io bounded on [i,b],

then a necei^oary :.nd sr.ffioiant c:adltIo.i that for (2)

each f £ D(X) r^ H\ V < a,b > the relation (7) holds <i , A

uniformly in x £ [a,b] is that (8) and (9) hold

uniformly in x £ [a,b].

Before giving a proof of this theorem, we remark

that Schurer [62] only indicated a proof of theorem I

and that in proving asymptotic approximation formula

for twice differentiable functions for the De La Vallee-

Poussin's integrals, Ilatanson ([51], PP. 212-214)

utilizes the 'relation

69

f(x+2t) = f(x) + f'(x)sin 2t + ^f"(x)sin^ 2t

+ a(x+2t)sin^ 2t

which is inconsistent with the strict 2Ti-periodicity of

f, since putting t = •n:/2 and t = -TC/2 we get

f(x+'n:) = f(x) = f(x-n) which implies that f is TI-

periodic. Thus in the neighbourhood of points t= + TI/2,

in tnis relation, the function a(x+2t) may not be

bounded and therefore the proof requires a modification,

Indeed, if we assume f to be 7i-periodic, Ilatanson's

proof, as it is, is correct.

Proof of theorem 1 Let f £ D(x) r '^[^^{x) , V /riting

(10) f(t) - f(x) - f'(x) sin(t-x) - 2f"(x) sin^ ^

, ,,x . 2 t-x = h^(t) s m - 2 - ,

it is easily verified that v/ith h (x+2mTi)=0, (ra=0,+1,.,),

h (t) is continu us at t=2m7i+x, (m=0,+1,... ), and is

bounded on X. Further h (t) is 2:1-periodic.

Let 6 be a number satisfying 0 < 6 < 2Tt, Then with

\/^ \{t) defined by

0 , if |t-x+2m7t I < 6 for

(11) ^(^ \(t) = I some m=0,+1,...,

1 , othervi/ise,

for an arbitrary e > 0 we have a 6 ( 0 < 6 < 7 ^ such that

(12) |h^(t)| g = + ^'\6,x)^^^' for all ^ £ X,

for some constant M > 0, Clearly

, .. < . 2 t-X/ . 2 5

^(6,x)(^) - ^^" -r^^^^ - . so tha t for a l l n su f f i c i en t l y l a rge

70

(13) L^(h^( t )s in2 ^ ;x) s eL^(sin2 ^ ; x )

+ M L^(sin'^ • ^ ; x ) / s i n ^ 6 / 2 .

I t follows from ( 8 ) - ( l 0 ) and (13) tha t

| L ^ ( f ; x ) - f ( x ) - ^ ^ [ f (x)Y^(x)+f ' (x)Scos x f^(x)

- s i n X ' t '2 (^) ) - f" (x) {cos x1'2(x)+sinx¥ (x ) - ? (x) j ] |

V (x) cosxTp(x)+sinx?'-(x)-Ti (x)

9 X^ - 2 9 ( n )

where lim v (x) = 0. Hence there exxsts a positive n^ / ^ n -» 00

integer II such that for all n g II the left hand side

of this inequality is smaller thane A/9(n) where

A = max (1, I cos x 'H Ax) + sin x Y.(x)-^ (x)| ).

Since £ is arbitrary positive (7) follows. The

necessity of (8)-(9) being a direct verification, this

completes the proof of the first assertion of the

theorem.

To prove the uniformity of (7) in the second

assertion of the theorem, as in the case of similar

results of earlier sections, it is sufficient to show

that M and 6, occuring in (12), can be chosen

independently of x £ [ a,b].

Applying Rolle's theorem to the function

F(0 = f(t) - f(.0 - sin (t-i) f'(0

_ f(t)-f(x)-sin(t-x)f'(x) 2 t:^ . 2 t-x ^^^ 2

sin -g—

where x £ [ a,b] , t £ < a,b> , t / x, it follows that

there exists a lying between t and x such that

71

(14) f(t)-f(x)-sin(t-x)f'(x) ^ j.,(^)^^,(^)^^^ t ^ ^ r-, . c- t-x c,

2 s m p

Applying lemma 1, p. 12, [30], to f", for an arbitrary

E > 0 we can choose a 6 > 0 independent of x such that

(15)

|f"(t) - f"(x)| < J

M. Itan - ^ 1 < f , M, = max |f'(x)| , 1 2 ' 2 ' 1 I - ^ ^ , 1

X £ La,b J whenever | t - x | < 6 f o r a l l x £ [ a , b ] . By (14) and (15)

we have

f ( t ) - f ( x ) - s i n ( t - x ) f ' ( x ) _ ^,,^^-j o • 2 t - x 2 s m —p—

£ I f " ( ^ ) - f " ( x ) | + | f ' ( O l Itan ^ I < e ,

for all t with | t-x | < 6 and x £ [a,b] , where 6 is

independent of x. Having shov/n this, it is clear

that M can also be chosen independently of x. The

second assertion of the theorem then follows.

COROLLARY 1 Let |L , n £ U j be as in theorem 1 and

let (8) be satisfied. Then a necessary and sufficient

condition for (9) to hold is that (7) holds for the

functions f = sin 2t and cos 2t. Also when (s) hold

uniformly in x £ [a,b]^X v/ith the functiwn

cos X ? (x) + sir. -. 'P.(x) - I* (x) being bounded on [a,b]

then a necessary a,nd sufficient condition for (9) to

hold uniformly in x £ [a,b] is that (7) holds uniformly

in x £ [ a,b] for the functions f = sin 2t and cos 2t.

Proof Since the functions sin 2t,

cos 2t £ D(X) r. ^r^\[,x) and L(X) r (i '*„<a,b> both, by ^71,A ^71,A

theorem 1, (9) implies the required satisfaction of (7)

for these functions. The converse proposition follows

72

from the linearity of the situation, t=x being a 4 t-x

fourth order zero of the function sin p •- , This

completes the proof of the corollary.

We omit an analogue of theorem 1.1.3.2 in the

trigonometric case.

Utilising (2) in full and combining the proofs of

theorem 1 and theorem 1.1.3.1, starting from the

relation

(16) f(t)-f(x)-f'(x) sin(t-x)-2f"(x)sin^ ^

, fj.\< - 2 t-x , /, \2m+2 , = h^(t) ! sm -2- + (t-x) } ,

we arrive at the following result.

THEOREM 2 Let { L , n £ Uj be a class of linear

operators defined on a common domain D(X) (X £^R) of

functions into a domain D(X) ( X ^ R ) of functions and

ultimately positive on a set 5?'^X. Let 1,sin t, cos t,

t,t , ..., t £ D ( X ) , where m is a positive integer,

and let x £ "X. Then in order that for each

f £ D(X) r ^2^1.2 X^^^ ^^^ relation (7) holds, it is

necessary and sufficient that (2) and (8) hold,

Further, assuming that the function

cos X p(x) + sin X ¥ (x) - T (x) is bounded on

[a,b]^X, a necessary and sufficient condition for (7)

to hold uniformly in x £ [a,b] for each

f £ D(X) r, H^^lg X <^»^> ^^ ^^^^^ ^2) and (8) hold

uniformly in x £ [a,b].

It is clear that the result of theorem 2 can be (2) (2)

extended to classes H^ -;,(x) and H^ i <a,b>(m > 2 and a m,X m,X

positive number) if we reformulate the theorem in terras

of the expression L (|t-x| ;x) in the manner of theorem

1.1.3.1.

73

Next v/e consider a class of sequences of linear

positive operators generated in the following v/ay,

Let (p, ) and (6^ '), n,k=1,2,..., be two oo x oo

raatrices of real numbers satisfying the follov/ing

properties

(i) ^ + 2 (pf' -'cos kt + bl^^ sin kt) is ^ k=1 ' ^

uniformily convergent on the interval [-TI, TC] for each

n=1,2,... .

(ii) 4- + 2 {ry-' cos kt + b}^^ sin kt) S 0,t £ [-7i,Tt], ^ k=1 ^ ^

n=1,2,... ,

Along the lines of the remark to theorem II, let

us note that (i) and (ii) imply that -1 < p^. ,

6 " ^ < 1 for n,k=1,2

Putting

U^(t) = - + 2 (p^^^ cos kt + b^' sin kt), k=1

n=1,2,..., on the space T (as defined in theorem II) of

functions define a sequence JL j(n=1,2,...) of

operators oj

(7) L (f;x) = 7 f f(x+t) u^(t)dt ,

-oo < X < oo ; n = 1 , 2 , . . . .

Applying Korovkin's remark to Theorem 4, p. 18,

[ 30] , v/e have

THEOREM 3 Let {L^j(n=1,2,...) be the sequence of

operators defined in (17) such that (i) and (ii) hold.

Then in order that for each f £ T which is continuous

74

a t '- p o i n t X £ R v/e h a v e

( 1 8 ) l i m L ^ ( i ; x ) = f ( x ) , n -* CO

it is necessary and sufficient that

(19) lira p ^ ^ = 1. n — "

Further, (19) is also a necessary and sufficient

Condition th-.,t for each function f £ T v/hich is

continuous at each point of a compact set S R, (l8)

holds uniformly in x £ S.

COROLLARY 1 If (l9) holds, then for each fixed k

v/e iiave

(20) lim '.<i^^ = 0 , (k=1,2,...), n -> CD

and

(21) lira i^l""^ = 1 , (k=2,3,...). n - c

Proof Let (l9) hold. By theorem 3 for each fixed

k(k=1,2,...) we have

lim L (sin kt;0) = 0 n^ / n -' vo

and

lim L (cos kt;O) = 1, n^ y

n -- c-

As L (sin kt;0) " -)f ^ and L (cos kt;0) = f^\ (20)

and (21) follov/, Tnis completes the proof of the

corollary,

In the next theorem v/e determine a necessary and

sufficient condition for the existence of an asymptotic

formula for twice differentiable functions,

TIir-ORFlI 4 Let L (n=1,2,...) be the sequence of

operators defined in (l7) such that (i) and (ii) hold.

Then for each f ' T v/hich is tv'/ice differentiable at a

75

point X G R, there holds

(22) L^( f ;x ) - f (x ) = •.5'^)f'(x) + ( l - p 5 " ) ) f " ( x ) + o ( l - p 5 ' ' ) ) ,

as n -* oo if and only if

. (n) 1-Pp '

n -> oo 1 _ p; '

Further, (23) is also a necessary and sufficient

condition that for each f £ T, which is twice

differentiable in an open interval containing the (2)

closed interval [a,b] (-or, < a < b < 00) with f (x)

continuous at each x £ [a,b], (22) holds uniformly in

X £ [a,b] ,

Note As in the remark to theorem II, here also we find

that (23) implies (19). Thus by the corollary to

theorem 3 it follows that (22) is a meaningful

asymptotic approximation formula.

Proof of theorem 4» We have for all x

L^(i;x) = 1 T / • _L \ (n) . , (n) L (sin t;x) = p^ 'sxn x + 6 'cos x L (cos t;x) = p!; -'cos x - b\ ' sin x

l-o '" , / . 2 t-x N 1 n^^^^ ~T"' ^' " — 2

and L^(sin^ ^f^jx) = J (| + - 2p ^ ) ,

Moreover,

L^(sin(t-x);x) = b^f^ .

Let us take 9(n) = (l-pij ) . Then condition (9) of

theorem 1 becomes

| - ^ ^ - 2 p S - ) = o(l-pW) .

76

Since -1 < plj < 1, dividing by l-plj^'^ t h i s i s seen

to be equivalent to (23) . As the terms

{cos X .' ( x ) - s i n J (x)j/c[)(n) and

-{cos X ;'2(x) + s in x J^ix) - ^(x) / (n) in (7) are

j u s t asymptotic evaluat ions of L ( s i n ( t - x ) ; x ) and

L (2 sin ~^i x) r e spec t i ve ly , the formula (7) i s

rc.'-dily seen to be equivalent to (22) in t h i s case .

Having observed t h i s theorem 4 follows from theorem 1,

Theorem I I of Korovkin i s a p a r t i c u l a r case of

tneorem 4 v/hen ' ^^ = 0 (n ,k=1 ,2 , . . . ) and -J"^ = 0

( k = n + 1 , n + 2 , . . . ; n = 1 , 2 , . . . ) ,

Applications of above theorems eas i ly give

asymptotic formulae for the operators A ( f ; x ) of

Korovkin ( 30 , p . 74)» the De La Val lee-Pouss in ' s

operators V ( f ; x ) ( 51 »[62 ) , the Jackson 's operators

Lo o (e .g . r62l) and t h e i r genera l i za t ions L 2n-2 ^ " - J' ° np-p s tudied by Schurer ( [ 6 2 ] , chapter 3) and o the r s .

Jn a l l these cases the asymptotic formulae hold (2)

unif irmly in x £ [a,b] for f £ tA ^ <a,b> and integrable

on [-71, Ti], where [a,b] is any bounded :ind closed

interval on the real line. The Pejer operators L _.,([62])

and the ? dsson operators L (f;x) ([30],p, 192) do not

satisfy the conditi m (9) 01 theorom 1 (or the

condition (6) or (23) of theorems II :nd 5, respectivelj^

and therefore do not possess such asymptotic formulae.

A convenient reference in v/hich evaluations (1) for

most of these operators can be found is [62].

Before proceeding to the next section we give one

more result for the determination of an asymptotic

formula. This result depends on the fact that for 2 2

k=^,3,..> the function cos kt - k cos t + k - 1 has a fourth order zero at t=0 and is positive at all

77

other points of the interval [-71,7:],

THBOREI-I 5 Let { L , n £ Uj be a class of linear operators

defined on a common domain D(-X) (X S- R ) of functionc

i-ito a domain D(X)(X ^ R ) of functions and ultimately

positive on a set X 3=X, Let D(X; contain all trigono­

metric polynomials. Let x £ X, Then in order that for (2)

each f £ D(X) ^ Q^ v(^) there holds the asymptotic £- 71, A

relation

(24) L^(f;x) - f(x) = f(x) { L^(l;x)-1 j

+ f(x) L^(sin(t-x);x)+2f"(x)L^(sin^ ^ ; x )

+ o(L^ (sin - ^ ;x)), n - c. ,

it is necessary and sufficient that for some k

(k=2,3,...) we have

. / . 2 k(t-x) ^ L (sxn - ^ <- ; x) p

(25) l i - ; . , 2 t - x , — - ^ ' n - 03 L (sxn —^T" 5 x) n^ 2 ' '

Further, a necessary and sufficient condition (2) 2 " 2

the relation (24) holds uniformly in x £ a,b is that

(25), for some k(k=2,3,...), holds uniformly in

X - _ a,b ,

Proof From (25) it follows that

, / . 2 k(t-a) N , 2 ., , . 2 t-x ^ L^(sxn -^2 ^?^) = k L * ^ ^ ~ 2 ~ ''^'

+ o(L^(sin -~; x)), n - co

which is equivalent to

(26) L (cos k ( t - x ) - k ' ^ c o s ( t - x ) + k 2 - 1 ; x )

= o(L„ ( s i n - ^ ; x ) ) , n - oo.

t h a t f o r each f £ D(x) r^QlX ' < a , b > ( ' a , b ' ^ X )

78

If f £ D(X) r Q^2) (x) , for all t £ X 271,X^ '

we have

(27) f(t)-f(x)-f'(x)sin (t-x)-2f"(x)sin2 ^

, / N . 2 t-x = h^(t) sxn -g- ,

where h (t) is such that given an arbitrary e > 0

there exists a 6 > 0 and a positive number M. , say,

such that for all t £ X there holds

/'nn^ ll /j_\l • 2 t—X ^ . 2 t—X (28) h (t) sxn -T;-^ e sxn —r-\ / I ^\ /1 2 2

2 2 + M jcos k(t-x)-k cos(t-x)+k -1j.

A proof of (28) utilizes L'Hospital's rule and the 2 2

property of the function cos kt-k cos t+k -1 stated

at the beginning of theorem 5. Using (26)-(28) the proof

of (24) follows, (2)

In case f £ D(X) n. KX „ <a,b>, 6 and M in above ^ 7t, A 0

can be chosen to be independent of x £ [a,b]. Then

uniformity .of (25) for x £ [ a,b] implies the uniformity

of (26) for x£ [a,b] v/hich in conjunction with (27) and

(28) shov/s that (24) holds uniform-ly in x £ [ a,b] ,

Taking f(t) = sin ^p~—^ the necessity parts of

the theorem are evident. This completes the proof of

the theorem.

1,2.2 Generalizations for functions of several

variables.

V/e state without proofs some generalizations of

results of section 1,2,1 to the case of operators

defined for functions of finitely many variables. The

proofs may be given along the lines parallel to those

in [59], [62] and in the previous results of this

thesis. We follow the conventions and notations of

79

sections 1.1,5 and 1,2.1,

By Q i X ( )' ^^ ^ \ » \ - \ ) ' '' ' ''° ^ * ^ ' m

class of all bounded functions f(?,. ,..., ^ ), defined

on X , which with an extension on R are 27i -periodic m m -

in each of their arg-uments and are tv/ice differentiable

(in the sense of section 1.1.5; at the point X.

Qoiv < S > , (S ,X R ), denotes the class of all 27f X ra ' ra' m — m^'

' m bounded functions f(E^,...,r ), defined on X , v/hich

1 ' m'' ra' with an extension on R a-re 2it-periodic in each of

m ^ their arguments and are uniforraly tv/ice differentiable on S (in the sense of section 1.1.5).

m THEOREM 1 Let { L , N £ U jbe a class of linear

operators defined on a common domain D(X )(X C: R ) of ^ m' m — m'

functions into a domain D(X ) (X "S:R ) of functions ^ ra^ m — m'

and ultimately positive on a set X '—.X , Let 1,sin t., "" ^ m m ' J' cos t., sin 2t., cos 2t., sin t. sin t, , sin t. cos t, ,

O' j' j' 0 k' 0 k' cos t. sin t, , cos t. cos t, (j / k; j,k=1,..., ra)£D(X ).

J k' J ^ .^ , , ... , - , , - . ^

id only if the op

that at a point X £ X

If and only if the operators L.. possess the property

m „, m ^-- "^

(1)

L„(1;X) = 1 + 2 - ^ r + 2 o(— ] r) , N • ., 9 • (n. ) . . ^9. (n. )^ '

1 = 1 ^x^ x^ x = 1 ^x^ J.'

L,.(sin t.;X) = sin x. + 2 — ' ^ ^ \ + 2 o(—7 r),

m ?2ii '^ 1 L„(cos t.;X) = cos x.+ 2 — j r + 2 o( 7 r) , N y J ^^^ 9i(n^) ^^^ >^(n_^)^'

ra Y, .,. m L„(sin (t.-x.)sin(t, -X, );X)= 2 ''y"\+ 2 o ( — r ^ ) ,

(j / k; j,k=1,,.., m),

and

80

t.-x. m . (2) L (sin^ ly-^ ; X) = 2 o(—-J—r), (j = 1,,,,,m),

as N-* oo , where 9 . (n. ) j 0, 9. (n. ) — 00 , n. — 00 , and

?'s are functions of X, then for each

f £ D(X ) r QA^iv (X) there holds ' m

m ( 3 ) L.,(f;X) - f ( x ) = 2 —7 r [ T . f + ^ ^ N ' ^ ^ ^ ^^^ (f^{n^) '- ox

m 2 (^-, . . cos X. - Yo • • s i n x . ) f '

j ^ 1 101 J 2jx 0^ X.

-, m ra + r- 2 ? , . , . f" + 2 ( f . - 'P^ . . cos X.

2 k , j = 1 5jkx Xj^x. .^^^ ox 2jx 0 k ^ j

m - " i j i ^ '^^a^^x .x . l - . ^^^ ° ( ^ T ^ ) ' ( ^ — ) .

where va lue s of t h e f u n c t i o n s o c c u r i n g i n (3 ) a r e t a k e n

a t t he p o i n t X,

F u r t h e r , l e t S be a bounded s u b s e t of 5 and ' m ra

assume that the functions f . - f„.. cos x. - v-.. sin x. ox 2ox J Mjx J

(i,j=1,...,m) are bounded on S , Then for each (2) "

f £ D(X ) r Q, i V (S ), (3) holds uniformly in x £ S m 2n*,X m m ' m if and only if (1) and (2) hold uniformly in X £ S .

Also, the above statements remain true when (2) (2 )

i s r e p l a c e d by ( l . 1 , 5 . 1 l ) a l o n g w i t h Ol „L y (X) and c- 71 .A

(2) (2) ™ Q^^; X ' m' being replaced by H '' (x) and (2)' ^

H ; .;: < S >(p > 2, j = 1,...,m) respectively. p,A^ m J

COROLLARY 1 In the statement of theorem 1, if (1) hold,

then (2) are equivalent to the condition that (3) holds

for the functions sin 2t. and cos 2t., (j=1,...,m). In

81

the uniformity part of theorem 1, if (l) hold

uniformly in X £ S , then the unifornity of (2) is

equivalent to the uniformity of (3) for the functions

sin 2 t. and cos 2t., (j=1,...,m), for X £ S .

An analogue and generalization of theorem 1.2.1.4

is the following (n.) (n )

THEOREM 2 Let (p , " ), (&/ ), (n . ,k=1, 2,... ;

j = 1,..., m), be 2m, 00 x 00 matrices of real numbers

satisfying the properties

-I °° (n.) (n ) (i) ^ + 2 (pi ^ ) cos kt + 6, ^ sin kt) is uniformly

^ k=1 ^ '

convergent on the interval [-71,71:] for n. = 1,2,...; J

j = 1 , 2 , . . . , m,

1 °° "^^i) "^" i^ ( i i ) T5- + 2 (pi cos k t + 6, 'J s i n k t ) = 0, t £ [-7i ,7.],

2 k=1 ^ ^

:iie

n , = 1 , 2 , . . . ; j = 1 , 2 , . . . , m. On t h e space T of a l l

f u n c t i o n s f ( t - , . . . , t ) , 27 i -por iod ic i n each of 1

argument t . , j = 1 , . . . , m) , bounded i n R and i n t e g r a b l e

o n [ - 7 i , 7 i ] v/ithi r e s p e c t t c t . , ( j = 1 , . . . , m ) , v/hile o the r

arQ-'am.ents assume a r b i t r a r y but f i x e d v a l u e s , d e f i n e

t he o p e r a t o r s L,. by

7: Ti;

(4) Ly(f ;X) - -^ r . . . [ f(X+T) .

- 7 1 - 7 1

ra 00 ( n . ) ^ V

n i j + 2 ( p ^ cos k t + 6|.^^Gin k t . ) j j = 1 "^ k=1 ^ J - J

d t . . . . d t , 1 m'

X £ R ; n , n , . . . , n =^^1,2, . . . , '..iiere X+T =

( x . + t . , . . . , A +t ) . Ihen i n orJ.. • t i n t x. .:• each ^ 1 I ' ' n v/

82

i £ T ^ 1; i. -, (x), (X £ R ), there holds the relation m 71*,iv ^ " ^ n "

m m (n ) (n )

(;) L,(f;x)-i(x) = 2 &-, 5^ f" (X) j,k=1 j k

' ^ k m (n.) m (n.)

+ 2 6 J f^ (X) + 2 (1-p^ J )f^ (X) + j=1 j j=1 j j

m (n ) + 2 o(l-p J ), j=1

(E -• oo), a necessary and sufficient condition is that

(n,) J

,,, '-P2

\o) li-i — [ v r y = 4 , j=1,2,..., m.

Purthor, (6) is also a necessary and sufficient

condition tnat (5) holds unifoj>mly in X £ S for (2) ^

occn f £ T r\ <X „ n <S > v/here S is an arbitrary m 2 71 *, n m. m

ri

subet wi R . n

Remark 1. In theurera 2 -e note that (i) and (ii) imply (n ) (n.)

thct -1< Pv ^ ,6k ' < 1 i^°^"j'^=1'2^A-)' (n.) j = 1,2,..., m. Tnus the expressions (l-p^ )/'\'i-p^ ),

j = 1,L,...,m, in (6) 8,re meaningful. Moreover, v/e observe

that (6) hold only if ("i) xim pj "J = 1, j = 1,2,..., m, n .-» 00

(nj In v,-hic";i case automatically lim 6., = 0 ,

n . -• 00 J

j-^1,2,... ,m, showing that (5) is o- meaningful asymptotic

relati .m.

Remark 2. Let jpi j,{6, j, (l:=1, 2,...), be tv/o sequences

o'' real numbers such th t tne secies

83

(7) S(t) = • + 2 (pi COS kt + 6, sin kt) ^ k=1 ^ ^

is uniformly convergent on [-TI, 7t]. Further assume that

(8) S(t) go, t £ [-71:,T:].

Then there holds

1-Pp

To prove (9), we observe that for all real t

(10) sin' I ^ sin^ I .

Since there must exist a set A [ -7i;,7i] of positive

measure on which strict inequality holds both in (8) -1

and (10), multiplying (l0) by — S(t) and integrating

from -K to 7t we have

|(|-^T-2Pi) =f (1-Pi) where a is some positive constant.

1-Pp

,4- = 4(1-«)

It follows that

from which (9) is obvious,

Using (9) it follov/s that the conditions (6) are

equivalent to each of the following two conditions

^\) m l-pp

(11 ) lim 2 7-—Y = 4m, II - oo k=1 , ' k

1-p,

^ 1~Pp p„ (12) lira n 7—r = 2 ^ ,

N - oo k=1 , " k^ 1-P1

Remark 3 Inequality (9) can be generalized to the

84

follov/ing inequa l i t y

^ ' '"Pk 2 (13) ^ — ^ < k^ , k = 2 , 3 , . . . ,

Pi

under the same assumptions (i.e. S(t) is uniformly

convergent on [-71,71] and that (S) holds).

A proof of (13) is as follov/s.

From the elementary inequality (k=2,3,...)

(14) cos kt - k^cos t + k -1 a 0, t £ [-TI:,7I],

v/hich v/e have used in the proof of theorem 1.2.1.5,

and the fact that strict inequalities hold in (8) and

(14) n a subset of [-7i,n] of positive measure, as

before, \/e have the inequality

Pk - ^^Pl + ^^-^ > °

which is tantamount to (13). Incidentally, (13)

improves on the following inequality of Stark [ 73],

p. 24,

1-P-l 2

In the light of theorem 1.2.1.4 it follows from

(9) and (13) that the operator sequences (I.2.I.I7) in

theorem 1.2.1.4 possessing asymptotic formulae for

tv/ice differentiable functions are precisely those

which attain the upper bounds in (9) and (13) in a

limiting sense.

V/e omit an obvious generalizatio.i of theorem

1.2,1.5 for the many vari;.',ble case,

As Indicated in section 1.1.7, we can construct

operators for functions of several variables v/ith the

85

help of those for functions of fev/er variables. Results

of this section are readily applicable in studying their

approximation properties. In fact the operators in (4)

are constructed precisely in the same manner.

Also we remark that some of the results of earlier

sections, though intended for algebraic functions, may

in cases furnish us with asymptotic formulae for periodic

functions more conveniently than the results of this or

the preceding section (e.g. for the operators L ,

p > 1, Schurer utilizes Theorem 2, [62] to obtain the

asymptotic formulae).

1.3 Approximation of functions of an exponential

growth.

In this section we require the following extra

notations:

E „ : class of all real (or complex) valued functions cc, A

f(t) defined on X ( S R ) , to each of v/hich there exists

a positive constant A such that |f(t)| < A e ' ' for all

t £ X. Here a is assumed to be a fixed positive real

number. E „(x): class of all f £ E „ which v/ith an a,X^ (X,X

extension on R are continuous at the point x(£ R ) , (2)

E^ -^(X): class of all f £ E which with an extension 0£,A a,A

on R are twice differentiable at the point x.

Let S be a subset of R. By E -^{S) v/e denote the ex, A

class of all f £ S ^ which v/ith an extension on R are a,X

continuous at each point of S. Also let <a,b> denote

some open interval (c,d) containing the closed interval

[a,b]. By E^ I <a,b> we denote the class of all f £ E „ ^ -' a,X a,X

which with an extension on R are twice differentiable

at each point of <a,b> v/ith the second derivative

86

continuous at each point of [a,b],

1,3.1 Single variable case

After having considered the periodic case in

section 1.2.1, proofs of various convergence results

for functions of exponential classes follov/ nearly

the same pattern v/ith the difference that here v/e use

hyperbolic functions instead of circular ones.

Following is the basic convergence theorem for

the classes E . (x) and E ^(S). (X fX 0£ , A

THEOREI'I 1, Let j L , n £ U j be a class of linear

operators defined on a common domain D(X) (X cr R) of

functions into a domain D ( X ) ( X S ' R ) of functions and

ultimately positive on a set X .S-X, Let a > 0 and 1,

e"^, e"""* £ D ( X ) . Let x £ X, Then, if and only if the

operators L possess the property that in

L (l;x) = 1 + a (x),

(1) L^(e'^;x) = e"^ +P^(x), and

T / -at >, -ax , / >, L (e ;x) = e + y ix) n^ ' ' 'n^ '

there hold

(2) lim a (x) = lim (3 (x) = lim y (x) = 0, n — oo n-'oo n-*oo

then for all f £ D(x) r E „(x)

(3) lira L^(f;x) = f(x), n — oo

Furtiier, let S S. X be a compact set. Then for each

f £ L(X) E „(S), (3) holds uniformly in x £ S if and

only if (2) hold uniformly in x £ S,

A proof of theorem 1 is obtained by using the

following: v/hen f(t) £ E y{x), there exists to each a, A

87

e > 0 a positive number A, depending also on x, but not

on t, such that for all t £ X

(4) |f(t)-f(x) I < e + A sinh^ ^ ^ % ^ ,

In case f £ E ^(S) (S compact), A can be chosen a, A

independently of x £ S.

A criteria, for asymptotic formula for the classes (2) (2)

E^ ^(x) and E ) <a,b> (p > O) is given in the following

theorem, V/e take (3 to be of the form ma v/here m is

a positive integer (m = 2),

THEOREI'I 2 Let {L , n £ U j be a class of linear operators

defined on a common domain L(X)(X R) of functions into

a domain D ( X ) ( X S R ) of functions and ultimately

positive on a set X C x. Let a > 0 and assume that the

functions 1, c " (k=+1, +2,..., +m) £ D(X) where "n g 2

is a positive integer. Let x £ }<f. If and only if the operators L satisfy the conditions

n 'i'o(x) .,

? (x) (5) L^(e ;x) = e + - ^ ^ + o ( ^ ,

W (x) T / -gt \ -ax , -g^ /'_1_\ L (e ;x) = e + —TTTT + ^(TT;^) » n'

and •*" 9(n) " °>(n)'

(6) L^(sinh ^ a(-=^);x) = o ( - ^ ) ,

as n -• 00 , where 9(n) / 0, 9 (n) -• 00 as n — 00, then

for each f £ D(X) ,- E^ ^(x) there holds the asymptotic ma,X^ ' •' ^

formula

(7) L^(f;x) - f(x) = —^ [2a2f(x)? (x) 2a 9(n)

+ af'(x)i e-"\(x) - e"^^_^(x)j

88

+ f"(x) {e-«"Y^(x) + e " X j x ) - 2Y^(x) j]

Further, a necessary and sufficient condition that

for each f £ D(X) r^ E^^^ <a,b> ([a,b] ^X), (7) holds

uniformly in x £ [a,b] (we assume that the function

e""^^ (x) + e"' ? (x)-2T (x) remains bounded on [a,b]). a —a o L » J'

A proof of theorem 2 is based on the relation

/Q^ s.f^\ fl \ , f(x) . , /, s , 2f"(x) . ,2 a(t-x) (8) f(t) = f(x) + —^—^ sxnh a(t-x) + ^—'- sxnh ^ p ' a

, 1. /'x ( • 1.2 a (t-x) , . , 2m a (t-x), + h (t)( sxnh —^ '- + sxnh —^—5—'•j

(2) where with f £ E^ Y(^) and h (x) = 0, the function

g'

h (t) has properties similar to the ones encountered in

the algebraic and the periodic case. We omit the details.

For the uniformity part we apply Rolle's theorem

to the function

P(0 = f(t) - fix) - ^ ^ sinh a(t-0

f(t)-f(x)- ^ 1 1 ^ sinh a(t-x) P ." . '- sinh^ ^^j^

with X £ [a,b], t £ <a,b> (t ^ x) and f £ E^^'^^<a,b>, a

Here we are taking f as extended on the whole of R and

so t is not necessarily restricted to X only and

assumes values on R. A restriction of t to X will be

required only in the final steps. It follows that for

some % lying between t and x

f(t)-f(x)- ^'^^^ sinh g(t-x)

2 -,,2 0 —^ sxnh — 2 g

^ ^ ^ = f"(0+af'(£)°°^^°/^"^h^ (t-x) ^ u;+ai U^si^^^,^)

89

Hence we have

h,(t) = ^ [f"(o-f(x)+af.(o :°:iiiii-)']. a

(. . , 2ra-2 a ( t - x ) )-1 , i1 + sxnh ^ 2 ' ]

The function cosh a(t-^)-1 has a second order zero at

t=^. Hence applying Lemma 1, p. 12 [30], to f", it is

clear that given an arbitrary e > 0 v\re can choose

a 6 > 0, independent of x £ [a,b] such that with t

restricted to X

|h^(t)| < e , if |t-x| < 6 , X £ [a,b], t £ <a,b>,

Having chosen 6 in above manner, it is clear from the

expression

f(t)-f(x)- ^ ^ s i n h a(t-x)- ^ ^ ^ sinh^^a^^

h (t) X

a

3inh2 - ^ + sinh2- ^ % ^

that there exists a positive number M, independent of

X £ [a,b], such that |h (t)| < M whenever |t-x| g 6 and

t £ X. Now, b and M having been chosen independently of

X £ [a,b_|, the uniformity part follov/s. The necessity of

various conditions is obvious,

COROLLARY 1 In the statement of theorem 2 the conditions

(5)-(6) are equivalent to the requirement that (7) holds (Ykt

for the functions f = 1, e (k= +1,..., +m). Further,

assuming that the function e""^? (x) + e"^? (x) - 2? (x) a -a o^

is bounded on [a,b], the uniformity of (5)-(6) for

X £ [a,b] is equivalent to the uniformity of (7) for the

above mentioned functi. ns. Assuming that for certain values of the parameter p

an evaluation of L (e' ;x) can easily be obtained, the

following corollary gives tlie most convenient method of

90

testing and an immediate determination of the asymptotic (2)

formula for functions of the classes E^ \r(x) and (p) ma,X'

E^^\<a,b>, ma,X

COROLLARY 2. Let jL , n £ Uj be a class of linear

operators defined on a common domain D(X) (X R) of

functions into a domain D(X)(X R ) of functions and

ultimately positive on X £ X, Let a > 0 and m g 2 be a ka positive integer and assume that the functions 1, e

(k=+1,...,+m) £ D ( X ) , Let x £ "X, A necessary and

sufficient criteria for the existence of an asymptotic

relation of the type (7) for all f £ D(X) r> E^^\(x) is ma, A

that

(a) L (e^Sx) = eP^ + e^^ ^^^ ''H^'^^ " g^v(x) ^ n^ ' ' 9(n)

for p= ka; k = 0,+1,,.. ,+m, where x(x), |i(x) and v(x) are

functions in x independent of p, The asymptotic formula

is given by

(10) L^(f;x) - f(x) = [f(x)x(x) + f'(x)j,(x)

+ f"(x)v(x)] + o(^l^). n -«, ,

Further, assuming that v(x) is bounded on [a,b] c. X,

a necessary and sufficient condition that (IO) holds uni-(2)

formly in xe[a,b]for each f£D(X)r E < ,ii>is that (9)holds

uniformly in x £ [a,b] for the values of p specified

above.

1.3.2. Generalizations for functions of several variables

In this section v/e give some generalizations of the

results of section 1.3.1. Proofs of the results are

omitted, which, however, can be obtained along the lines

91

indicated in section 1.1.5. In addition to the

conventions and notations of section 1.1.5 we record

the following.

By a we denote the m-tuple (a-,ap,...,a ) and we

take a.(i=1,...,m) to be positive numbers, E : class m

of all real (or complex) valued functions

f(t-,to,..., t ) defined on X («S.R ) to each of which ^ 1 ' 2' ' m^ m^ m" there exists a positive number A such that

m a. 11. I |f(t.,t t )| < A 2 e "- ^

i = 1

for all (t.,t„,...,t ) £ X . ^ 1' 2' ' m' m

E „ (x): class of all f £ E „ which v/ith an extension a,X a,X on R are continuous at the point X (£ R ). /p-vm ^ m'

E i (x): class of all f £ E ^ which with an extension a,X ' a,X ' m m on R are twice d i f f e r e n t i a b l e ( in the sense of sect ion m ^ 1.1.5) a t the point X.

Let S be a subset of R . By E „ (S ) v/e denote the m m " a,X m'

' m class of all f £ E ^ which with an extension on R

a,X ra m .2)

are continuous at each point of S . Also by E^ J. (S ) m " a,X m'

' m we denote the class of all f £ E ^ which with an

a,X ' m

extension on R are uniformly twice differentiable on S m ' m

(in the sense of section 1.1»5).

THE OREM 1 Let iLvrjN £ U j be a class of linear operators

defined on a comnon domain L(x)(X ^ R ) of functions m m'

in to a domain D(X ) of functions and u l t imate ly pos i t ive TTl 4 - 0 .

~ ^ " o^i^'i ~ g i ' ' i on a set X •:= X . Let the functions 1, e " " , e " " m m ' ' ( j = 1 ,2 , . . . ,m) £ D(X) and l e t X £ X . i /r i t ing m

Lj^(l;X) = 1 + a^^iX),

92

m a. t. m a .x. L ( 2 e ^ ;X) = 2 e +p (X), and

j = 1 j = 1 ^ '

m -a . t. m -a . x . L ( e J J;X) = 2 e " + v„ „(X) ,

j=1 j=1 ^'"

if and only if there hold

(2) lim a^ix) = lim p (x) = lim v (X) = 0 JN-»co XJ-.00 ' N-*oo '

t hen f o r a l l f £ D(X ) r^ E „ (x ) v/e have ^ m' a,X ^ ' ' m

(3 ) lira L ( f ; X ) = f ( X ) . N - oo

Further, let S ^ X be a compact set. Then for ' m m ^ each f £ D(X ) / E ^ (S ), (3) holds uniformly in

m X £ S if and only if (2) hold uniformly in X £ S .

THEOREM _2 Let {l TjN £ U j be a class of linear operators

defined on a common domain D(X ) (X "=. R ) of functions m' ^ m m

i n t o a domain L ( X ) ( X " ^ R ) of f u n c t i o n s and m m m u l t i m a t e l y p o s i t i v e on a s e t X ^ X . Let t h e f u n c t i o n s

,1 1 , 4. , X m m

+ k . a . t . ± g - t . + a , t ,

1, e J ^ ^ e J J ^ ^ ( j / k ) , (k = 1 , 2 , . . . , p . ;

j , k = 1 , 2 , . . . , m) £ D(X ) v/here p . g 2 , j = 1 , 2 , . . . , m, a r e

p o s i t i v e i n t e g e r s . Le t X £ X . Then i f and only i f

t h e r e ho ld ra ? . m

1 = 1 ^ 1 ^ l ' 1 = 1 ^ 1 ^ 1 ^

a . t . a . x . m T,, . . m

1 = 1 ^ 1 ^ l ' X = 1 ^ 1 ^ 1 ^

- a . t . - a . x . m f m ^

L^(e ^•^;X)=e ^ ^ + _. 2 ^ . _, o(-irhn^' ^ / \ + 2 O 7 y, . . cp. (n . ) . ^ ^cp. (n . j • 1=1 ^ 1 ^ 1^ 1=1 ^ 1 ^ 1^

(5 ) Lj j (s inh aj^(tj^-xj^)sinh a ^ ( t ^ - x ^ ) ; X )

93

m ? ra

1=1 ^ 1 ^ 1^ 1=1 ^ x ^ x^ and

2 p . a . ( t . - X . ) m

(6) L (sinh J - i ' ' ) = .\ °^v:ijr)^ ' 1 = 1 ^X^ l '

a s H - oo , w h e r e j , k = 1 , 2 m; 9 j_ (n^ ) / 0 , 9 ^ ( n ^ ) - «=

(7)

a s n . -» oo , 1 = 1 , 2 , . . . , m , t h e n f o r e a c h

f £ L(X^) E^2^^ ( ) ^p^ (p^^^^ Ppap,.... P^aJ) • ' m

there holds the relation m -

L„(f;X) - f(x) = 2 / \ [Y^. f

m . -a.x. a.x. + 2 ^ (e ^ - ip, .. - e ^ - f .. )f' j=1 2«j 1J- 2ox> X.

m + 2 -^r-^— T,, .. f" k,j=i 2«j,gj 5kjx xj x.

k / j m ^ -a.x. a.x.

+ 2 — 5 — (e '^ ^ -a... + e -^ • '? ..-4Y • )f" ] . . . 2 1ix 2ii 01' x.x.-' J = 1 4aj ' ^ J J

m

1 = 1 ^ 1 ^ 1 ^

where values of various functions occuring in (5) and (7) are taken at the point X.

Further, assuming that the functions -a.x. a.x. e ^ J V,, . + e ^ ^Ypji - 4^Q. (i,j = 1,..., m) are

bounded on S C_ X , where S is a bounded set, a m •— m' m ' necessary and sufficient condition for (7) to hold

unif.-rmly in X £ S for each f £ D ( X ) r^ 'E^'^K. (S ) that ^ m ^ m^ pa,-v m' ' ra

(4)-(6) hold uniformly in X £ S .

COROLLARY 1 In the statement of theorem 2 assume that

(4)-(5) hold. Then a necessary and sufiieint conxition

94

for (6) to hold is that (7) holds for the functions a .k.t. e " • ', j = 1,..., m; k = +2,..., +p . In case (4)-(5)

J J

hold uniformly in X £ S and the functions -a.x. a -x. e ^ " f, .. + e ' ' ¥„..-4f • (i, j = 1 ,... ,m) are bounded Iji 2ji ox '' ' ' ' on S , a necessary and sufficient condition for (6) to

m' " ^ ' hold uniformly in X £ S is that (7) holds uniformly in

m X £ S for the functions mentioned above. m COROLLARY 2. Let f L.,, II £ U j be a class of linear

operators defined on a common dom.ain D(X )(X S . H ) ^ ^ m'^ m m

of functions into a domain D(X ) of functions and ultimately positive on a set X X . Let the functions

+k .a .t. +a .t .4a, t, 1, e ' ^ \ e ^ ^ ^ (j / k ) , (k.=1,2,...,p.(g 2);

j , k = 1 ,2 , . . . , m) £ L(X ) and l e t X £ "X . Then the

conditions

(8) L„ (sinh a, (t, -x, ) sinh a . ( t . -x . ) ; X) \ / j j \ k ^ k k ' O J J m Y m

= a,a. 2 -2fi^s + 2 o(—7 y) , k J i^i 9i(n~) i^^ > i ( " i ) '

(j / k; j , k=1 ,

a n d \, a, t,

9) Ve ^ Sx) ^k«k^k ^k«k^k Z 1

= e + e 2 —7 r 1=1 ^ i^" i -2 2„. , !! , 1

^^oi + Vk^iki + Vk^2ki) + . ^ ° ( 7 : T ^ ) ' x=1 ^ 1 ^ x^

(^l. = 0> ± 1 f - . IPjjJ k = 1 , . . . , m),

v/here 9.(n.) / 0, 9 . (n . ) —00 as n. -*ooj 1 = 1,...,m,

are necessary'' and sufficient in order that for each f £ D(X ) n. E -* ^ (X) there holds

m ap,X^'

m m (10) L-,(f;X)-f(x) = 2 — r - T [ y .f + 2 ¥, . . f

N i^l 9i(nj_)'- ox ^^^ I j i x^

95

m m m 2 ? , . . f" + 2 ? ' . v - f " ] + 2 o ( — 7 Y ) , I I - > O O .

. ^ 2 j x x . x . , . . 3k J X X, X.- ' . . c p . l n . K ' ] = 1 J J k , j = 1 - ^ k J x = 1 't'x^ 1^

m + 2 ?

' k7 j F u r t h e r , assui.iing t h a t the f u n c t i o n s >?„ . .

2 j x ( j , i = 1 , 2 , . . . , m ) a r e bounded on a bounded s u b s e t S

~ (2) ^^' of X , f o r e a c h f £ D ( X ) ^ E ^ ^ ^ (S ) , ( 1 0 ) h o l d s

m m' a p , X m' ' ^ ^ m

uniformly in X £ S if and only if (8)-(9) hold

uniformly in X £ S . m

96

CHAPTER 2

METHODS OF BOUITDIMG FUIICTIONS AND THE W-PUNCTIONS

Taking into account the unboundedness of given

test functi-ns, in this chapter, v/e extend the

applicability of several earlier theorems ([14],[48],

[50],[56]) on the approximation of unbounded functions

by linear positive operators, based on the methods

of bounding functions and the V/-functions. In accordance

with the common practice, the results are formulated in 2

terms of the test functions 1, t, t ,... . However, it

would be obvious from the proofs that using the same

ideas one can reformulate the results in terms of other

unbounded test functijns as v/ell. All the functions

considered in this chapter are real or complex VEilued.

2.1 Method of bounding functions.

2.1.1 Basic convergence

Let X R and let Q(t) be a non-negative and

unbounded function defined on X. Such a function Q(t) is

called a bounding function. Let I be an index set and

let Q(I,X) = JQ , a £ Ij be a collection of bounding

functions Q defined on X. Let D„(Q,I) denote the class a A^ '

of all functions f defined on X to each of which there

exist an a £ I and two positive numbers A and B such

that |f(t)l < A+BQ (t) for all t £ X, By D°(Q,I) we

denote the class of all functions f £ D (Q,I) to each of

which there corresponds an « £ I such that given an

arbitrary e > 0 there exists a positive number C such

that for all t £ X, |f(t)/Q (t)| < e whenever |f(t)| > C.

Thus D (Q,I) consists of functions having an unboundedne^

97

of a lov/er order than that of some element of Q ( I , X ) .

Similarly D (Q,I) consists of functions v/hose A

unboundedness is at the most of the same order as that

of some member of Q ( I , X ) . Clearly D°(Q,I) C^L^(Q,I)

and also Q(I,X) D (fi,l).

Let I'Y(X) denote the class of all functions defined

on X v/hich with an extension on R are continuous at the

point X £ R. The class of all functions defined on X

which with an extension on R are continuous at each

point of a set S SiR is denoted by L (s),

The follov/ing theorem gives the basic convergence

result for a class of linear operators ultimately

positive on a set of points.

THEOREI'I 1 Let (L , n £ Uj be a class of linear operators

defined on a common domain D(X) (X — R ) of functions

into a domain D ( X ) ( X ^ R ) of functions and ultimately

positive on a set X ^ X , Let Q ( I , X ) . ^ D ( X ) be a 2

collection of bounding functions and let 1,t,t £

D(X) r\ D ( Q , I ) , Let X £ X and assume that for each a £ I A

(I) sup L (Q ;x) < oo ,

n £ U "

Then, in order that for each f £ D(X) r D°(Q,I) \(^)

v/e have

(2) lim L^(f;x) = f(x), n -' oo

it is necessary and suffioient that for 1=0,1,2

(3) lira L^(t ;x) = X . n — oo

Further, if S X is a compact set and (l) holds

uniformly in x £ S for each a £ I, then in order that

for each f £ D(X) r-. J^'^{Q,1) r \ i s ) > (2) holds

uniformly in x £ S, it is necessary and sufficient that

98

for 1=0,1,2, (3) holds uniforraly in X £ S.

Moreover, if Q ( I , X ) S D ( X ) , (X £ X), then in order

that for each f £ L(X) rs D (Q,I) r^ D (x), (2) be satis­

fied, it is necessary and sufficient that

(4) lim L (Q ;X) = Q (x) ^ ' n^ a a '

n — 00

for each a £ I and that (3) holds for 1=0,1,2. Also

if Q ( I , X ) S D ( S ) , ( S ^ X is a compact set), then in A

order that for each f £ L(X) r^ D (Q,I) rs D (S), (2)

holds uniformly in x £ S, it is necessary and sufficient

that (5) for 1=0,1,2 and (4) for each a £ I hold

uniformly in x £ S.

Proof. If f £ D(X) D°(Q,I) r Djr(x), it is easily

shown that there exists an a £ I such that given an

arbitrary e > 0 we can find a positive constant A such

that for all t £ X (5) |f(t)-f(x)| < e + A(t-x)^ + cQ|;t).

Using (1), (3) and (5), rest of the proof of the first

assertion of the theorem follov/s along the lines similar

to the proof of theorem 1.1.2.1,

To prove the second assertion, if f £ D(X) r\ D (Q,I) A

r D (x) V/e can show that there exists an a £ I such that

given an arbitrary c > 0 we can find positive constants

A,B such that for all t £ X (6) |f(t)-f(x)| < e + A(t-x)^ + B ( Q (t)-Q (x)),

Using (3), (4) and (6), rest of the proof 16 easily

completed on the lines similar to the proof of theorem

1,1.2.1.

2,1,2 Asymptotic estimates

Eisenberg and V/ood [14] stated the follov/ing

99

theorem referring to the preprint of Muller and V/ "• k

[50],

THEOREM I Let JL j denote a sequence of linear

operators which are positive on (-00,00) and have a

common domain D of functions defined on (-00,00),

Let 1,t,t £ L and f £ C(-oo , 00 ) L, Let

-00 < a < x < b < c n a n d L ( l ; x ) = 1 , L e t T ( t ) = ( t - x ) ^

and u (x) = [L (^ ;x)]^ . If there exists a number p > 1 n^ ' '- n X '-I ^

and a positive increasing (with |t|) function Q(t)

such that Q-P £ D and f(t) = 0(Q(t)) (|t| -' 00), then

for n=1,2,...

(1) |f(x)-L^(f;x)I s 2a)(f,u^) + ra;^|f(x)| u^

+ C(L (QP;x))^/Pm-2/P' u^/?' , \ -Q\ » / / X n '

where l/p + l/p' = 1, u(f,6) = maxj |f(u)-f(v)| :

|u-v| g 6 and u,v £ [a,b]j, C is a constant depending

only on f, and m = min{ | a-x | , |b-x|j.

Taking into account the unboundedness of the 2

function t we obtain the follov/ing improvement on

theorem I.

Theorem 1 Let jL j be a sequence of linear operators

which are positive on (-00,00) and have a common domain

D of functions defined on (-00 ,00), Let 1,t,t £ D

a n d f £ C(-oo ,co ) ^ D, L e t - o o < a < x < b < o o a n d

L ( l ; x ) = 1 . L e t ¥ ( t ) = ( t - x ) ^ a n d u ( x ) = [L {^ ; x ) ]^. j-^\ > / x^ ' ^ ^ n^ ' L n^ X ' ->

If there exists a number p > 1 and a positive increasing

(viith |t|) function Q(t) such that Q- £ D and f(t) =

0(Q(t)|t|^/P' + t^)(|tl -00) where l/p + l/p' = 1, then

for n=1,2,...

(2) lf(x)-L^(f;x)| g 2(o(f,u^)+m-V(|f(x)|+C^)

- Cp(L ( Q P ; X ) ) V P ,,-2/P' 2/P' ,

100

where u(f,6) = max j|f(n)-f(v)l : |u-vl § b ,

u,v £ (a,b)j, m = minj |a-x|,|b-x|j, C and Cp are

constants depending only on f such that C =0 if

|t|^/P = 0(Q(t)) (|t| - «=) and C2 = 0 if Q(t ) = 0( 111 /^)

(|t| - oo).

Proof Define the function \ by

(5) x^ 0 , t £ (a,b)

1 , o ther \ / i se .

It is clear that there exist constants C. and Cp

satisfying the conditions of the theorem and such that

(4) |f(t)| ,^ s C^(i^)2 , Cp Q(t) \ ^ '/'' X X

for all t £ (-00,00) and x £ (a,b). Also a usual

manipulation with the modulus of continuity gives

(5) (1-x^) |f(t)-f(x)| g { 1 + (i^)2j^(r,u^),

n

for all t £ (-00,00) and x £ (a,b). Finally, since

(6) Xjf(x)| § |f(x)| (^)2 , X

for all t £ (-0°, <») and x £(a,b) , the proof of (2)

follows by combining (4)-(6) and an application of

HSlder's inequality,

All the remarks of Eisenberg and Wood [14] made in

connection with theorem I are also valid for theorem 1.

For the sake of completeness with appropriate

modifications we reproduce them here,

Remark 1 If sup jL (Q-P;X) : n=1,2,...j < 00 , then

(7) |f(x)-L^(f;x)| g 2a)(f,u^)+m;;V(|f(x)|+C.,)

^ 2/p' -2/p' + Mu ' - m ' ^ ,

n X

101

where M is a constant depending on x but otherwise

satisfying the properties of Cp in the statement of

theorem 1.

Remark 2 If in addition L reproduces all linear n -

functions, i.e. L (t;x) = x, then u may be replaced P n 2 i n

by t = (L (t ;x)-x )^. Also in this case the modulus " n ^ n^ ' '

of contituity co*(f,6) (see [14] and the reference cited

therein) which is zero if f is linear, may be substituted

for u(f,6). The modulus of continuity u*(f,6) is

defined by w*(f,6) = inf co(f (x)-cx; 6 ). c

Remark 3 If on (a,b), f £ Lip a , we may replace u(f,u ) with Tu". ^ ' n n

Remark 4 An analogous result holds when JL j is defined

on a common domain D(X) where X is not necessarily the

whole of R.

Remark 5 The factor |t| and the term t in the

order 0(Q(t)|t| '-^ +t ) could be included in the 0-

symbol because of the unboundedness of the test function 2

t . In a similar manner using other test functions and

bounding functi jns Q(t) their unboundedness (not

necessarily when |t| -* 00) can also be taken into account.

The criteria for using other test functions is that the

basic convergence result for them implies the basic

convergence result at least f.-r all bounded and 2

continuous functions. Thus the test functions 1,t,t

may be replaced by a unisolvent system or a Tchebychev

system of an order two or more.

Remark 6 The multiplier enlargement may be employed

to obtain analogous results on (- 00, 00) and [0,oo ) in

case the opert'tors L are positive only on a finite

interval. In particular we have

THEOREM 2 Let jL j be s sequence of linear operators,

102

defined on a common domain D of functions defined on

[ 0, oo), which are positive on (0,c), 0 < c < oo and let

{a j be a sequence of positive numbers strictly

increasing to <» . Let 0 g a < x < b < oo, f £ c[0, oo)

and 1,t,t^, f(a t) (n=1,2,...) £ D. Let L (l;x) = 1, If

there exist a number p > 1 and a positive increasing

(with t) function Q(t) such that for n=1,2,...,

Q-P(a t) £ D and f(t) = 0(Q(t) t^/^'+t^)(t - oo) where

l/p + 1/P' = 1> then for n g N(x) v/here N(x) is a

suitably large positive number,

1/p

(8) |f(x)-L^(f(a^t); a~^x) | g 2a)(f,p^)

+ ni-2 P^ (|f(x)|+C^) + C2[L^QP(a^t);a;^x)]

-2/p' 2/p' . m ' ^ P , X n '

where 1 (t) = (a t-x)^, p (x) = [L (? ;a~^x)]'^ , nx^ n ' n^ ' ^ n^ nx n '-^ '

0. and C„ are constants depending only on f such that 9 /

0 =0 if t '^ = o(Q(t)) (t - oo) and C =0 if O I -r\

Q(t) = 0(t ) (t -* oo) and co and ra are as in theorem 1,

Theorem 2 is an improvement over the follov/ing

theorem of Eisenberg and V/ood [l4]«

THE'lHEM II Let JL } be a sequence of linear operators

which are positive on [0,c] , 0 < c < oo , with common

domain D, and let ja j be a sequence of positive numbers

strictly increasing to oo. Let O g a < x < b < o o ,

f £ D o C[0, ex.), and 1,t,t^ £ D. Suppose L (f(a t); a-1x) involves only values of f(a t) for 0 g « t < oo and n ' '' ^ n ' n

L (l;x) = 1. If there exist a number p > 1 and a positive

increasing function Q such that Q £ L and f(t)=0(Q(t))

(t -^ oo), then, for n g N(x),

(9) |f(x)-L^(f(a^t); a;^x)| g 2a)(f,p^)

103

+ m - 2 l f ( x ) | p 2 + C j L {Q\a t ) ; a - \ ) ] ' ^ V ^ / P ' p ^ / P ' , x ' ' n I ' - n ^ ^ n ^ n ^ •' x n

where l / p + l / p ' = 1 , '^^^.(t) = ( a ^ t - x ) S p ^ ( x ) =

[ L ( T ;a x ) ] ^ and C. i s a c o n s t a n t depending on ly on f. ' - n ^ n x n ' - ' 1 •" ° ''

2 . 1 . 3 As.ymptotic f^.rmulae

I n our n e x t theorem v/e improve t h e follovi/ing r e s u l t

of E i s e n b e r g and \/ood [ l 4 ] .

THEOREM I Let [ L j be a setiuence of l i n e a r o p e r a t o r s

which a r e p o s i t i v e on ( - oo, oo), have a common domain p

L, and s a t i s f y L ( l ; x ) E 1. Let 1 , t , t , . . . £ L, Let f,

f £ L and be d e f i n e d on ( - oo, oo) vjlth d e r i v a t i v e s upto

and i n c l u d i n g t h e 2m-th o r d e r a t t h e p o i n t x £ ( - 0 0 , 0 0 ) .

Suppose numbers p . , p > 1 and p o s i t i v e i n c r e a s i n g

( w i t h | t | ) f u n c t i o n s Q , Q e x i s t such t h a t Q.,Qp £ D

and f ( t ) = 0 ( Q ^ ( t ) ) , ? ( t ) = 0 ( Q 2 ( t ) ) ( | t | -> ~ ) , I f

(1) sup i L ^ ( Q . i ; x ) : n = 1 , 2 , . . . j < c« , 1=1,2 ;

(2 ) sup{L ( ( t - x ) ' "^ ;x) : n = 1 , 2 , . . . j < 00 , p=max(p ,p );

[L ( ( t - x ) 2 - + 2 j . ^ ) ^ l / p .

(3) l im -^ — = 0 n - 00 L ^ ( ( t - x ) ;x )

f o r a t l e a s t one v a l u e of j ( j = 1 , 2 , . . . ) , where

1/P+1/P'=1> then

2m-1 /, \ , L ( f ; x ) - f ( x ) - 2 ( f ^ ^ ^ ( x ) / k l ) L ^ ( ( t - x ) ^ ; x )

/ , N , . k=1 (4 ) 1 , ^ 2m-1 .^s —

" ^ ~ L ^ ( ^ ; x ) - Y ( n ) - 2 ( f ^ ^ ^ ( x ) / k ! ) L ^ ( ( t - x ) ^ ; x ) k=1

= f^^ '^^(x)

Note There is a slight error in the statement of above

theorem and its proof given in [14]. In order to render

104

the result correct l/p' in (3) must be replaced by

min(l/p.|,l/p') where l/p^+l/p^ = 1, 1 = 1,2 .

With this change, the correct proof is a minor modifi­

cation of the given one.

THEOREM 1 Let {L j be a sequence of linear operators

ultimately positive at a point x £ (- oo, oo) and defined

on a comraon domain D ( - OO, OO). Let m be a positive

integer and j a positive number and let 1,t,t ,...,t ,

|t-x| ™' '' % II(- ~, °°). let L (l;x) = 1, n=1,2,..., and

let ¥,f £ D and possess derivatives upto and including

the 2m-th order at the point x. Let Q,(t), Qp(t) be

positive functions increasing with |t| and let numbers p.

P-iiPp > 1 exist such that Q.- £ L, 1 = 1,2, and let ^2m+2j 2ra+2j

(5) f(t),g(t) = 0[Q^(t)|t| P-1 +Q2(t)|t| P^ +|t|2 -'2j]

as |t| - OO where l/p. + l/p? = 1, 1 = 1,2 . If

(6) n;;: L^((t-x)2'";x) < c« , n — oo

P-i (7) lira L^(fi^^;x) < ^ , 1=1,2,

n -• oo

and

[L^(|t-x| 2-2j^^)^l/p.

(8) lim - ^ ^ = 0 n - OO L^((t-x) ;x)

where l/p + l/p' = 1, p=min (p..,Pp), then

L^(f;x)-f(x)-T\f(^)(x)/k!)L ((t-x)^;x) k£!

( 2m-1 /, N ,

""^'^ L (f;x)-Y(x)- 2 (f^^\x)/k!)L^((t-x)^;x) k=1 "

f(^")(x) - ^(2m)^^)

(9) lim

105

if ?^2ni)(^) Q

Remark 1 Since (t-x) ™ < 1 + |t-x|™'*', v > 1, the

condition (6) is much less severe than the condition (2)

in theorem I. Also (5) extends the applicability of (9)

to a larger class of functions than given by theorem I.

Remark 2 It is interesting (though in an entirely

different direction) to note that the statement of

theorem I remains true, as it is, if v/e replace (2) by

the follov/ing one

(10) lim L (t-x)^'";x) = oo , n — CO

However, then the unboundedness of Q and Q„ is of

little use. A similar assertion holds for theorem 1

as well, if v/e replace p by max (p,.,Pp) and replace

(6) by (10).

Remark 3 It is possible to replace {Q-,Qpj by a set

{ Q. , i £ Ij where I may be vacuous or may consist of

an arbitrary finite number of elements. Then in place

of (5) and (7), respectively, v/e must have

2m+2.i

(11) f(t),f(t) = 0( 2 Q.(t)|t| ^i +|t|2'"+2j) (|t|-.oo) i £ I

and P-i

(12) lim L (Q. ;x) < 00 , i £ I, n — 00

P i w h e r e p=min { p . , i £ l j , p . > 1 , i £ I a n d Q. £ D(-OO,OO).

If I is empty, theorem 1 reduces to a generalization of

a result of Mamedov [36].

Remark 4 Remarks 2.7-2.9, [14], p. 269, made by

Eisenberg and V/ood at the end of the proof of theorem I

remain valid for theorem 1 as well, V/e also refer to a

remark (Rathore [54], P. 145) made in connection with

106

theorem I, in the light of local unsaturation on zero

orders, generalizing the above limit ratio phenomena of

(4).

Proof of theorem 1 Borrowing the notations used in the

proof of theorem I ([14]), we have, given an arbitrary

£ > 0 a 6 > 0 such that

-e(t-x)2°' g |)(t)-Kx)-A(t-x)2'° g e(t-x)^"'

for a l l t s a t i s fy ing | t - x | < 6. For | t - x | g 6 we have

a constant C(6) such that 2m+2.1

|^(t)-(^(x)| g C(6) { 2 Q.(t)|t-x| ^^ +|t-x|2'"+2jj,

1=1

Thus

|4(t)-(l)(x)-A(t-x)2'"| 2m+2,i

g C(6){ 2 Q.(t)|t-x| ^^ + |t-x|2'"+2^| i = 1

^ U ^2m , J A I 14- |2m+2j + E(t-x) + -hrf It-xI 6 ^

Rest of the proof can be completed by an application of

HSlder's inequality and using (6)-(8).

Next we come to the follov/ing two theorems of

Muller [48], in which no direct assumption is made on

the order of the growth of f but only a local assumption

on the L -images of |f| is required. In the rest of this

section we use Mtiller's notations in which Q denotes a

non-void (bounded or unbounded) domain of the real line,

D(S2) the linear space of all complex-valued functions f

defined on Q and Dp (Q;X) the linear sub.space of those

functions f £ D(Q) which possess derivatives upto and

including the 2p-th order (p=1,2,...) at a fixed point

X £ Q.

107

THEOREM II Let f € D2(Q;X) and let (L^j (n=1,2,...) be

a sequence of linear positive operators mapping D(Q)

into D(Q) which satisfies the conditions

L (l;x) = 1 + O(-TU-) n ' ' 9(n)' -,(x) ^

(13) L (t;x) = X + / \ + o(—7—v) ^ ' n^ ' ^ 9(n) ^9(n)^

2 2 2 ^^' 1 L (t ;x) = X + —7—r- + o(—7—r) n^ ' 9(n) ^9(n)'

and

(14) lim L (|f|-' ;x) < 00 (v > 1 a real constant). n ^ 00

If there exists a natural number m such that

(15) x[''"^^^(x) = Lj(t-x)2-+2.^^ __ o ( _ ^ - ^ ) ,

[9(n)]^

l/r+l/r' = 1, for n — 00, then the local degree of

approximation to f by the sequence {L f j(n=1,2,...)

is given by the asymptotic formula

2T.,(x)f'(x) + {?2(x)-2xY^(x)jf"(x) (16) L^(f;x)-f(x)

n ~^U^

(rTTTT) . + 01 / s >(n)'

THEOREM III Let f £ l'2m+2 ' '' "^ •'"®* I ^ 5 "''' 2'* * *

be a sequence of linear positive operators mapping D(Q)

into D(Q) and satisfying the condition (14) with a real

constant r > 1. If there exists a natural number j such

that

T [2m+2j+2], ^,1/r' n ^^'^ „ 1 1 •17) lim ^ — 7 7 ^ = 0 , - + -L = 1

• ' 2m+2 \ i \ > -£, -^i » n - ' T^ -i(x) n

then

.,,. ,. V^2m+1'") f(2" 2)(x) ( ) 1" [2m+2], s = 7 2 ^ ^

108

where

2m+1 ^(k)/ N

(19) R2m+l(^) - ' ^ ' ^ - ' , ^ ^ ( t - x ) ^

k=0

An improvement over theorem II is as follows.

THEOREM 2 In the statement of theorem II, (14) can be

replaced by the more general condition 2m+2

(20) li^ L^(|f(t)/(l + jt| ' )|'';x)<oo, n -' 00

Also, if f is bounded on all bounded subsets of Q, then

in (15) the o-term can be replaced by a corresponding

O-term, the other hypotheses of theorem II reraaing

unchanged.

In a similar way, theorem III can be improved to

give

TIIEOREM 3 In the statement of theorem III, (14) can be

replaced by the more general condition

2m+2j+2

(21) Ti^ L^(|f(t) /(l+|t| ^' )|^;x)<o^. n - 00

Also, if f is> bounded on all bounded subsets of Q, then (17) can be replaced by

/^-^ r-r— r2m+2j+2]/ N,l/r'/ [2m+2]/ v (22) lim T- ^ J(x)j / T^ •J(x) < CO,

n -* cc

provided

(23) lim T^2'"+2^(x) = 0, n -• 00

the other hypotheses of theorem III remaining unchanged.

Remark 1 It is obvious that (14) implies both (20)

and (21) but that the converse is not true in general.

Remark 2 Replacing (t-x) by |t-x| , m in

theorems II and 2 can be taken to be an arbitrary

109

positive number not necessarily an integer. Analogous

result holds for j occuring in theorems III and 3.

Proof of theorem 2 Let us start from the relation

,2 (24) f(t)-f(x)-f'(x)(t-x)-f"(x)

(t-x)-2

2m+2

= g^(t)(t-x)2+h^(t) - i - ^ |t-x| r-

1+|t| ^'

/ 4. 14. I 2m+2 + P^(t) |t-x|

It is easy to check that the unknown functions g (t),

h (t) and p (t) can be chosen so that given an

arbitrary e > 0 there exist positive constants A(e) and

B(e) so that |g^(t)| < e,|h^(t)| < A(e) and |p^(t)|< B(e)

hold for all t £ Q . Rest of the proof of the first

assertion consists in an application of HSlder's

inequality and some routine manipulations.

The proof of the second assertion is an easy

modification of the proof of the first assertion and is

therefore omitted.

Proof of theorem 3 Writing

„(2m+2)/ \ o ,o

(25) ^2m+l(^) = ( x ) - (2m+2ST^) (^-)'"^'' 2m+2j+2

|f(t)| It-x 3:' o ,o- o + h (t) ' "p p. , + p (t)|t-x|2'"+2a+2

x^ ' 2m+2j+2 ^x^ ' ^ I '

1 + |t| r'

as in the proof of theorem 2, one easily verifies that

in this case as well, the functions g (t), h (t) and

p (t) can be chosen so as to satisfy similar constraints,

i.e., for an arbitrary e > 0, |g^(t)| < e ,|h^(t)|< A(E )

and |p (t)| < B(G) for all t £ Q, where A(e) and B(e) are

positive constants. Thus using Holder's inequality we 110

have

(20 |Ln( 2m+l( )'-) - T^^SyH \(i^-^f'-'';-)\

g eL^((t-x)2"'+2.x) + A(e).

[ nU 'i:!j,.2i^^-)]^^^tvi^-i^^^^w^' 1+|t| ^'

+ B(e) L^ (|t-x|2^+2j+2.^)^

Dividing the inequality (26) by T"- •' and taking

limit as n -• oo, (18) follows as a consequence of the

arbitrariness of e >0. The proof of the second assertion

is an easy modification of above proof and is therefore

omitted.

2.2 Combinin,-? the techniques of bounding functions

and the W-functions

In this section we generalize and improve upon

several results of Schmid [56], based on the concept

of a V/-function, by recasting them in terms of bounding

functions and taking into account the unboundedness of

given test functions.

2.2.1 Asymptotic formulae

In the sequel the functions denoted by Q(t), Q (t)

(n £ U) etc. are taken to be arbitrary non-negative and

unbounded bounding functions defined on a subset X of R.

In particular, the follovring may be convenient choices

for the functions Q (t)(n £ U):

(a) Q^(t) = ej h( |f(t) | ) , where h is a fixed V/-

function, f is the function to be approximated and

e -• 0 as n -» oo. n

(h) Q (t) =e Q (t), where Q(t) is a fixed bounding

111

function and e -* 0 as n -* oo, n

(c) Q (t) = Q(t), where Q(t) is a fixed bounding

function (depending on the points at which the functions

are to be approximated) such that it has zeros of

suitably high orders at the points where the approxi­

mation is desired.

THEOREM 1 Let {L , n £ UJ be a class of linear

operators defined on a comraon domain D ( X ) ( X S R ) of

functions into a domain D(X)(X £r R) of functions and

ultimately positive on a set X X. Let{Q (t),n £ Uj Sr

D(X) be a set of bounding functions. Let m be a positive

number and let 1,t,t ,|t-x| £ D(X) for all x £ X. Let

X £ 5! and f £ D(X) such that with an extension on R, f

is tv/ice differentiable at the point x. If there hold

the conditions

^o(x) ^ L (l;x) = 1 + —7—Y + o( V v) n^ ' 9(n) ^<5)\n)'

^ (x) (1) L (t;x) = X + "*/ V + o(—r-r) ^ ' n^ ' 9(n) 9(n)'

2 2 2 1 L (t ;x) = X + —7—V + o(—7—r) , n^ 9(n) ^9(n)' '

(2) L (Q ;X) = © ( - T W ) , and ^ ' n^ n ^9(n)''

(3) C^(2m+2,a)42-2]^^) __ ^_^^ ^

as n -• OO, where 9 ( n ) / 0, 9 ( n ) -» oo as n -• oo, Y. ( x ) ,

1 = 0 , 1 , 2 , a r e f u n c t i o n s i n x , a i s a n o n - n e g a t i v e r e a l

c o n s t a n t ,

| f ( t ) | - Q ^ ( t ) (4) C^(2m+2,a) = sup { _ — _ _ ^ j , and

t £ X 1+ aft I

(5) 42ra+2]^^) =L^(|t-x|2--2.x) ,

112

then there holds the formula

(6) L^(f;x)-f(x) = ^ [f(x)^^(x)

+ f ( x ) {Y^(x)-xY^(x)) + ^ ^ { f2(x) - 2x^P^(x) + xh^{x)]] + o ( ^ ) , n ^ oo .

Proof. For an arbitrary 6 >0 define for all t £ R

0 , |t-x| < 6

.,(t) = \

1 , |t-x| g 6 .

Then we have for all t £ X

|f(t)|,.^(t) g C^(2m+2,g)(l+a|t|2°^-*-2)^^(t) + Q^(t).

Clearly there exists a constant A > 0, depending on x,

such that for all t £ R /., |,|2m+2N /,^ ^ ,1, |2m+2 (1+a|t I )(ij (t) g A I t-x I

Hence for all t £ X

|f(t)|„^(t) g A C^(2m+2,a)|t-x|2^+2 ^ ^^^^

Rest of the proof is evident from (l)-(3) along the

lines of the proof of theorem 1.1.3.1.

Remark 1 Putting Q (t) = e h([f(t)|), a=0 and taking

m to be a positive integer, Satz 1.1, p.16, [56] of

Schmid follows.

Remark 2 Theorem 1.1.3.1 can be obtained as a corollary

of theorem 1 .

Remark 3 Por continuously twice differentiable

functions, uniformity of (6) can easily be incorporated

in the statement of theorem 1,

THEOREM 2 Let {L , n £ U} be a class of linear operators

defined on a common domain D ( X ) ( X ^ R ) of functions into

a domain D(J(!)(X^R)of functions and ultimately positive

on a set l ^ t . Let {Q (t), n £ U}^D(x) be a set of

bounding functions, m a positive number, p a positive

113

integer and let 1,t,t ,...,t •P,|t-x| •^ £ D(X) for all

X £ X. Let X £ X and f £ D(X) such that with an

extension on R, f is 2p-times differentiable at the

point X. If there hold the conditions

/r.\ -I • L ( Q ;X)

(7) Ixra n^ n ^ n-oo;j2i7( ^ 0 , and • -Ifx)

n

C (2ra+2p,a)42in+2p](^)

(O ^ - [2p], ; = ° ' n-*oo T ' - - ' ( X )

where

42P](x) = L^((t-x)2P;x) ,

^[2-2p]^,^(,,_^,2m+2^^)^ and

|f(t)|-fl^(t) C (2m+2p,a) = sup { 2m+2p ^ '

t £ X 1+a|t| P

where o is a positive number, then there holds

n ^ '

where

2p-1 M(^ ^ (10) R ( t ) = f ( t ) - 2 S r r ^ ( t-x)^.

P~' k=0 ^•

for all t £ X.

Proof. For an arbitrary 6 >0 let n. (t) be defined as

in the proof of theorem 1. We then have for all t £ X

I^-Z^^I^'S^*) = Cj,(2m+2p,a)(l+a|tl2"'+2P)^^(,)^.Q^(t).

As we can find a positive constant A such that for all

t £ R

(l+altl^'^^^p)^^^^^ g^|^_^|2m+2p^

we have for all t £ X

114

(11) |R2p_i(t)|Hj,(t) g A C^(2ra+2p,a)|t-x|2'^+2p ^ ^^^^^^

Having shown this, since given an arbitrary e > 0 we can

find 6 > 0 such that for all t £ X and satisfying

It-x] < b

( 2) l%-i(0-%iy]^^^'"^(-)l <^(t-x)2p, the proof of (9) is obvious from (7), (8), (ll), (l2),

the arbitrariness of e > 0 and the positivity of the

operators.

Remark 1 Putting Q = e h(|f|), a = 0 and taking m

to be a positive integer, Satz 3.1, p. 35> [56] of

Schmid is obtained as a corollary to theorem 2.

In case, v /ith an extension of f on R, only the

left and the right derivatives f^^^^ (x) and f^^^^x) •— +

exist without necessarily being equal, we have

THEOREM 3 Under the remaning hypotheses of theorem 2

there holds ^n(^2p V") f(2P)(x)+f(2p)(^)

(13) lim [2P5" . = 2-(2^l n — oo x'- -'(x) ^ ^ ' n

where Rp _ is given by

2p-1 „(k) (2LI (^_^\^ (14) R* .(t) = f(t) - 2 ^ - T ^ it-x)'

P J Q

r(^P)(.)-f(^'')(,) I,. ,,,.,,2P-,

2 (2p)!

for all t £ X.

Proof. Using the knov/n i d e n t i t y (see p . 42, Schmid [56])

('^) , - i g ^ - M a p ) , -

115

The proof of theorem 3 is essentially the same as that

of theorem 2,

Remark 1 Putting Q =e h(|f|), o=0 and taking ra to be

a positive integer, Satz 3.2, p.415 [56] of Schmid

follov/s as a corollary to theorem 3»

In case v rhen with an extension of f on R, the right

and the left derivatives f (x) and f (x)

exist v/e have the follov/ing result.

THEOHEIi 4 Let jL , n £ Uj be a class of linear operators

defined on a common domain D ( X ) ( X £ R) of functions into

a domain D(X)(XS: R ) of functions and ultimately

positive on a set 'X S X. Let JQ (t), n £ Uj D(X) be a

set of bounding functions, mi-a positive number, p a

positive integer and let 1,t,t ,..., t ,|t-x| ,

I _^ I 2m+2p+1 ^ ^ - ^^_^ ^-^-^ X £ J.. Let X £ 'X and f £ D(X)

such that with an extension on R, f is 2p-times

differentiable at the point x and that the derivatives

f(2p+l)^^^ and f[^^^^\x) exist. If there hold the

conditions

(16) lim rn^,<-i = 0, and n^oo xL2P+^J(x)

C (2m+2p+1,a)xt2--^2p+1](^^) (17) lira -^

n -«• 00 ^[2P-1](,) n ^ '

v/here

?[2P-1](,) =L(lt-x|2P-^ n ' ' n~

;x, I

J2m+2p+1]/ > . /u |2m.+2p+1 s , Ti; -"(x) = 14 ( t-x ;x), and

n

r i

Jf( t ) | -Q^(t) (2m+2p+1,a; = sup

-^ ^ '"^ t £ X S + a | t | 2 - 2 p . 1

where a i s a pos i t i ve number, then v.-e have

116

L^(R ;x ) f ( 2 P - ^ ^ ) ( x ) - f ( 2 P - ^ ) ( x )

(^Q) 1^^ . [ 2 p + l ] , . = 2 . ( 2 p + l ) ! ^ n -» oo x"- (x) ^ ' n ^ '

where Rp i s g iven by

2P 4.(k) ( 1 9 ) R p j t ) = f ( t ) - 2 ^ . i^^ ( t - x ) ^

• k=0 „ ( 2 p + l ) / X .o(2p+l) / \ o ., f;; ^ H x ) + f ^ ^(x) (t-oc}fP+]_

2 (2p+1) !

f o r a l l t e X.

P r o o f . To p rove theorem 4 we use t h e knovm r e s u l t

( p . 44, Schmid [ 5 6 ] )

H2p( t ) f i 2 P + ^ ^ ( x ) - f p P - ^ ) ( x ) ( 2 0 ) l i m — ' ^—T;—7 = ;;—r;;—rv; ^ t ^ x ( t - x ) 2 P + 1 2 . ( 2 p + l ) !

and proceed in a vay analogous to that of the proof of

theorem 2.

Remark 1 Putting Q =e h(|f|), o=0 and choosing m

sttch that 2m+1 is an even positive integer, Satz 3.3,

p. 42, [56], of Schmid follows as a corollary to theorem

4.

2.2.2 Asymptotic estimates 2

Utilizing the unboundedness of the function t

(as |t| -•00) we now obtain upper bounds of the error

in the approximation of continuous and continuously

differentiable functions by a class of linear positive

operators.

THEOREI l 1 Let {L , n £ U} be a class of linear

operators defined on a common domain D ( X ) ( X ^ R ) of

functions into a domain D(x) of functions and positive

on a set Sf SiX. Let I = (a,b) be an open interval

contained in K and let the functions 1,t,t ,f(t), Q(t)

117

£ D(X) where f(t) is bounded and continuous on I and

Q(t) is a bounding function. For x £ I let

L (l;x) = 1 + a (x) , n^ no^ ' '

(1) L^(t;x) = X + ot .](x),

L^(t ;x) = X + g^2^^''

where a .(x) — 0 as n-oo, 1 = 0,1,2; x £ I. Then for ni '

arbitrary e,6 > 0, for each x £ I v/e have

(2) |L^(f;x) - f(x)| g M(6,n,x) (0j(f;6)

+ CL^(Q;X) + |f(x)|(m;2xt2](^) + |a^^(x)|)

+ m-2 C(2,a;e)T[^^(x) [l + a(lx| +mj^]

where

M(6,n,x) = l+|a^^(x)|+min{6-^[x[^^(x)]*,6-2xt2](^)|^

m^ = min {|a-x|,|b-x|j,

' i -' ) = L^((t-x)^x) = g^2(^)-2xan1^^)+^^«no(^)'

C(2,a;E) = sup {Mtlb_eQ(tij^ ^^^^^ ^ g Q

t £ X l+gt

and (i)-p(f;6) is the local modulus of continuity of f on I

defined by

u,^(f;6) = sup(lf(t^)-f(t2)|: t^,t2 £ I, jt^-tpl g 6}.

Proof Let Xj denote the characteristic function of the

interval I and put i-j- = 1-\-j.. Por all t £ X, x £ I v/e

have then

(3) |f(t)-f(x)| g (Oj(f;|t-x|)\^+(|f(t)| + |f(x)|)nj.

Now

(4) a)j(f; lt-x|)\j g [l+minjs"^ |t-x|,6~^(t-x)^}]a)j(f;6),

118

(5) |f(t)|jij g (ij(l+at2) C(2,a;e) + £fi(t)

g m~2(t-x)2{l+a(|x|+m^)^}c(2,a;E)+efl(t),

and

(6) |f(x)| nj Sm-2(t-x)2|f(x)|.

Combining (3)-(6) by the linearity and positivity

of L , n £ U, the inequality (2) is immediate.

Remark 1 Taking g=0, Q(t) = h(|f(t)|) where h is a

W-function Satz 4.1, p.46, [56], of Schmid follows. It

is clear that if f(t), Q(t) have unboundedness as

|t| -» oo then a positive choice of a will keep G(2,a;E)

relatively smaller than in the case of a=0. Thus in

such c^ses (2) vi/ill result in a sharper estimate than

given by the result of Schmid.

Remark 2 In his result Schmid takes I to be the closed

interval [a,b]. However, then the function m~ is

undefined for x=a,b. For this reason vre take I to be the

open interval (a,b).

If f is differentiable on (a,b) and if f is

continuous and bounded there, then, with (o-p(f';6) to be

the modulus of continuity of f on (a,b) v/e have

THEOREM 2 Under the assumptions and notations of

theorem 1, for arbitrary e,& > 0, a = 0 and x £ I, in

this case we have

(7) |Ljf;x)-f(x)| g {[ ^[^^(x)]Kt-\l^\x)]o^^if';t)

+ eL^(Q;x) + |f (x) | (m;^^^]^^) +|ano^^)|)

+ m;2 C(2,a;s)4^](x)[l+a(|x|+m^)2]

+ |f'(x)| {ml\[^'^M + \a^^(x)-xa^^(x)\}.

The proof of theorem 2 is an obvious adaptation of

the proof of theorem 1 and so is omitted.

119

Taking g=0, Q(t) = h(|f(t)|), where h is a W-function,

we obtain Satz 4.2, p. 54, [56] of Schmid. The choice

g > 0 has the same advantages as given in remark 1

following theorem 1.

In connection with Satz 5«1> P. 57 and Satz 5.2,

p. 60, [56] of Schmid (concerning error estimates for

multiplier enlargement) we remark that they do not

require a separate proof. A.t a first reflection it is

clear that these are simply re-statements of the

earlier Satz 4.1 and 4.2 respectively, if we replace

JL { by the operators {L }. In this light v/e omit error n

estimates for the multiplier enlargement technique; they

can easily be derived from theorems 1-2.

120

CHAPTER 3

APPROXIMATION OP UNBOUNDED FUNCTIONS BY OPERATORS

OF SUCTIATION TYPE

In this chapter we indicate an application of

certain estimates closely connected with the local

unsaturation [54] of linear positive operators on

positive zero orders, mentioned in the introduction of

the thesis, to determine the approximation of certain

unbounded functions by operators of summation type.

3.1 A general outline

In the sequel we limit our treatment to the

Bernstein polyn'jmials and obtain certain classes of

unbounded functions which can be approximated by the

sequence of these polynomials at all points of contituity

of the functions. Here also we limit ourselves to proving

only the basic convergence result for these classes and

do not go into obvious details by means of which various

results on asymptotic estimates, asymptotic formulae,

simultaneous approximation of derivatives of functions

by the derivatives of the polynomials, linear

combinations and linear com.binations of iterates etc.

can be extended to the same classes of unbounded

functijns.

The tecnnique used serves as a prototype fir all

sequences of linear positive operators of summation type

for which asympt -tie estimates of the local unsaturation

order (see the Intr duction) or even those of certain

moments or of approximation of certain zero orders are

available. Por the Szasz-Mirakyan-Hille operators

vari us results as v/ell as their proofs are almost word

121

to word the same as those for the Bernstein polynomials

(with the difference of the intervals [0,l] and [ 0, oo)

on which they are respectively defined). As an

application of a mere knowledge of the asymptotic

behaviour of various moments, we remark, for instance,

that the Laguerre-Bernstein series of Cheney and Sherma

([9],[54]) can approximate functions having an

unboundedness of order t (b > O) as t -* 0 (for special

subsequences a similar result nolds for functions

unbounded in the neighbourhood of an interior point in

[0,1]).

3,2. The Bernstein polynomials

Let f be a function defined on the interval [0,l].

The Bernstein polynomials B (x), n=1,2,..., of the

function f are given by

(1) B^(x) = 2 p^^(x) f(J), X £ [0,1], v = 0

where

(2) p^^(x) = C^)x''il-x)''-\ (n=1,2,...; v = 0,1 n).

The follov/ing result which is a modification of

Theorem 1.5.3, PP. 18-19, Lorentz [33], gives a basic

estimate v/hich v/e require in the sequel.

THEOREM 1 Let 6 > 0. If for all x £ [0,l] there holds

the inequality

(3) S p^ (x) SA(6) e-*<"''(°',

IS-1 * n=1,2,..., v/liere A(6) and 3(6) are positive functions of

6, independent of n and x, then if 9(n) ^ 0, n-1,2,...,

the sequence [(p(n)/nj , n=1, _,..., is bounded. Also, if

122

0 g X g 1 and 0 g 6 < ^, then there holds the inequality

.2 (4) 2 p^ (x) g 2e"° " , n=1,2,...

F-x| 6 'n '

Proof. First v/e prove the second assertion. Proceeding

as in the proof of theorem 1.5.3, PP. 18-19,[33] , if

|u| g 3/2 and x £ [0,1] , then

U ( 1 -X) / . N -ux ^ . u

xe ^ + (1-x)e g 1 + -7- .

Thus A ( \ def _, u(v-nx) , N 0) (u,x) = = 2 e ^ ^p (x) 'n ' ' „ - nv

v=0

= [xe^^^-^)+(l-x)e-^]''

2 ^ nu g exp 4 •

If

n I 1 ,„ / N r, u v-nx / \ ? (u,x) = 2 e 1^ Ip (x), n^ ' „ m v v = 0

then for 0 g u g 3/2, 2

j (u,x) g i)^(u,x) + (| (-u,x) g 2 exp ~ .

Hence for u,c ^ 0

2 Pj (x) g ^ , exp(u|v-nx|) g c? (u,x) ^

which implies that

2 2 P (x) g - . ( \ M -. o nu"^ nv c ,

exp(u|v-nx|J g 2c exp —— / 6^n

Putting c =- ~|- e a,nd u=26 , the second assertion of

the theorem follov/s.

Por proving tho first assertion of the theorem,

assume that for a subsequence {n } , say, of the

sequence fnjof natural numbers, there holds 123

9(n^) lim

Then for an arbitrary positive 6 n^ n^-B(6)9(n^)

2 P^ ..(x)e g A(6)e

— 0, as n —

uniformly in x £ [0,1] . However

Pn /^)® = "* x(l-x) n* , "#-^ "*

•X-

n„x n, = — {(l-x)ej

-• oo , as n -* oo ,

for each x £ (0,1-e ). This contradiction proves the

first assertion, and the proof of the tneorem is

complete.

The follov/ing theorem generalizes Theorem 1.9.2, p

28, Lorentz [33].

THEOREM 2 Given an arbitrary x > 0, there exists a set

C S[0,1] of measure 1 such that if f is any function

defined on [0,1 ]such that (a)

(5) f(t) = 0 (exp |t-c|"' ' ' '' ), as t - c,

for every c £ C and (b) if jt, j, k=1,2,..., is any

sequence of numbers such that the sequence {f(t, )j is

unbounded then Jt, } has a limit point c £ C, then there

holds the relation

(6) lim B^(x) = f(x) n — oo

where x i s any point of con t inu i ty of f. Moreover, i f f

i s continuous at each x £ [ a , b ] ^ [ 0 , l ] , then (6) holds

uniformly in x £ [ a , b ] .

Hov/ever, i f B(v) i s a pos i t i ve and rionotonic

124

function defined for v > 0 and diverging to infinity as

v-»oo,then ther-e exists a set C * ^ [0,l] of the power of the

ontinuum such that for any non-negative f defined on

[0,1] if there holds

(7) Edt-c*!""") = 0(f(t)), t - a*(t / c,*),

where c* is some point of C*, then the sequence JB (x)}

is unbounded for every x £ (0,l).

To prove theorem 2 v/e require some basic results.

First we prove a general lemma. Por Bernstein

polynomials v/e require the result of this lemma

restricted to the interval [0,l].

LEMl iA 1 Let C be a subset of the extended rea-1 line

R* obtained by adding the points + oo to R and let to

each point c £ C there correspond a positive function

g (t) defined on R such that g (t) is continuous for

t / c and g (t) — oo as t -• c. Let f be a function

defined on R and possessing the following properties

(a) f(t) = 0(g^(t)), as t - c (t / c),

for every c £ C, and

(b) if Jt, } is any sequence of real numbers such that

|f(t, ) I -• oo as k -• oothen J t, } has a limit point c £ C.

Then there exists a finite and bounded subset C of tho

set C such that given any two positive numbers A and 6

there exists a positive number A., say, such that o

|f(t)| < A^

for all t satisfying both |t| g A and |t-c | g b for

each c° £ 0 ° . \ '

Proof Let C* be the subset of C consisting of all finite

c such that the set Jf(t) : t £ (c-6,c+6)j is unbounded

for every positive 6. If C* is finite then with C = C*

125

the assertion of the lemma is satisfied. For if not,

then for some A, 6 > 0 the set Jf(t):|t| g A and

|t-c I g 6 for each c £ C } is unbounded. Let Jf(t, )j

be a sequence of members of this set such that

|f(t^^)| -• oo as k — oo. Then Jt, j has a limit point c £ C.

Clearly c £ C . But, since the set Jt:|t| g A and

|t-c I g b for all c £ C j = A , say, is compact,

also c £ A . A s G ^ A = 0 , this gives a contradiction.

Next let C* be infinite, then C* has a limit point,

say, c* £ R*. If |c*| < oo , then for each sufficiently

small 6 > 0 there exists a positive number A , say, o 0

such that

|f(t)| < A g^^(t), for all t £ [c*-6^,cJ+6J. o o

By continuity of g „(t) for t / c* , it follov/s that C* 0

o on every closed subset of the set [c*-6 ,c*+6]-}c*},

' - O O O ' O '

f(t) is bounded. Also there exists a c*£ C* (c* / c*)

such that c* £ (c*--g-6, c*-t -6). Hence there exists a

closed interval [a,b] contained in one of the intervals

(c*-6 ,c*) or (G*,C*+6) such that c* £ (a,b). But then 0 0 0 0 0 \ r /

f(t) must be unbotinded on [a,b], which is a contradiction. It follov/s that c* coincides with one of the points + oo. o —

Let c* = +00. Then if A is a sufficiently large positive

number, there exists a constant B > 0 such that |f(t)| < B g^^(t), for all t g A.

o

Also it is clear that for some positive number k, there

exists a c*£ C* such that A < c* < kA. Consequently f(t)

must be unbounded on the interval [A,kA]. However,

the continuity of g ^(t) implies other\/ise. This contra-0

d ic t ion shov/s tha t c* / + oo. In a s i m i l a r \/ay we can

126

shov/ that c* / - oo. Hence C cannot be infinite. This o '

completes the pr,of of the lemma.

DEFINITION 1 (p. 28, [33]). Let A(V) be a positive and

monotonic function, defined for v > 0, and such that

A(V) -» 0 as V -• OO . Then a real number c is said to

admit the approximation A ( V ) , if there is an infinity

of positive integers n such that |c-v/n| g A(n) is

fulfilled for some integer v. In case we restrict n to

belong to a fixed set I, say, and if the above property

is satisfied with n £ I, we say that c admits the

approxiraation A(v) through I.

DEFINITION 2 Let A(v) be a function as above. The

auxiliary function A*(t) for t > 0 is defined by

(8) A*(t) = [1+inf {v : 1/A(V) > tj]""".

We also need a result of Koksma [25] (see Lorentz

[33], p. 28).

THEOREM I Alraost all real c do not admit the approxima­

tion A(V) = V~ , k > 2. Also, for any A(V) -* 0 for

V -• 00, there is a set of the c of the pov/er of the

continuum which admit the approximation A(v).

Proof of theorem 2 Let 0 < x < T. By Koksma's theorera

alraost all real c do not admit the approximation

A(v) = v~ ' ' o . Let C be the set of all such c £ [0,l].

With this C we prove the first assertion of the theorem.

First we show that there exists a positive integer

m and c.,c ,...,c £ C such that for an arbitrary 6 > 0

we can find a positive number A such that

|f(t) I g A^, for all t £ A-,,

vAere A . j = [ 0 , l ] ~ J t : | t - c . | < 6 f o r some i = 1 , 2 , . . . , r a j .

To each c £ C d e f i n e

127

.expj|t-c|~^2"^'') ) , t £ [0,1] , t / c, (9) g (t) =

M , t = c.

Restricting to the interval [0,1]identify the set C and

the functions g (t), c £ C, defined above v/ith the

corresponding notations in lemma 1. If f satisfies the

conditions of the first assertion of theorem 2, it

follov/s from lemma 1 that there exists a finite subset

C S C such that given an arbitrary 6 > 0 there exists

a positive number A such that f(t)| g A., for all o

t £ [0,l] and satisfying |t-c | g 6 for each c £ C .

Designating the elements of C by c ,c , . .. , c , say,

our assertion follov/s.

Now, v/e can choose a 6 > 0 so small that there

holds |f(t)| g A ^ expj|t-cJ-(2+x)-^j

for all t satisfying |t-c | < 6 , t / c , p=1,2,..,,m,

v/here A , p=1,2,...,m, are suitable positive numbers

independent of 6. Let M = max |f(c )| and put 1 g p g ra P

A = max (A,, M,A.,A„,..., A 1. ' 6 ' ' 1' 2' ' m'

Then v/e have

m (10) |f(t)| g A 2 g^ (t) , t £ [0,1] .

p=1 p

Since c.,c ,,...,c do not admit the approximation 1' /2' N ' m

A(v) = v~^ "^o^ it follows frora (lO) that for all n

sufficiently large v/e have

2-^-0 n -5-—

lf(J)| g ra A e ^^^ , v = 0,1,2 n.

3 By theorem 1, t h e r e f o r e , v/ith b <. -f , we have

128

H Pn,(-)|f(i)l ^ 2 m A e"^'-^-'^' ,

F-x| g 6 'n '

for all X £ [0,l] and for all n sufficiently large.

Hence for an arbitrary 6 > 0

(11) lim 2 p^^(x) f(J) = 0

^^°° |^x| g 6

uniformly in x £ [0,l]. From (ll), the first assertion

of theorem 2 is obvious.

To prove the second assertion of theorem 2, we

utilize the second part of theorem I. It is clear that

if B(V) satisfies the conditions of theorem 2, then

there exists a positive and monotonic function g(u)

defined for u > 0 such that g(u) -» oo as u -•ooand 2

B(g(u)) g e" , u > 0.

Put A(V) = l/g(v), V > 0. Then A(v) - 0 as v - oo and

so by Koksma's theorera there exists a set C* o [0,l],

say, of the j)ower of the continuum each member of which

admits the approximation -'i(v). Without loss of

generality we can assume C* to consist of only irrational

points. With this set C* v/e prove the second assertion

of the theorom.

Let f be a function satisfying the conditions of the

assertion of the theorem and let x £ (0,1). Let c* £ C*

be such that (7) holds. Let [(v »n. ) } be an infinxte

sequence of distinct pairs of integers satisfying

,x c l g A(n„) .

Since B(|t-c*| '') = 0(f(t)) as t - o*, there exists a

129

positive number A such that for all n sufficiently large V V

n„ , n, f(-^)p (x) g A" B(|—^ - c*r^)p (x)

^B(|^-c.|-^

where q i s t h e s m a l l e r of t h e numbers x and 1-x. Nov/, Vn n2

B ( | — ^ -c*\~^) g B ( g ( n ^ ) ) g e * . *

I t fo l lov/s t h a t f (v / n ^ ) p (x) — oo as n^ — oo . n^' * X v n *

Thus the sequence jB (x)} is unbounded for every x £(0,l)

This completes the proof of theorem 2.

THEOREM 3 Let I be an unbounded set of positive

integers denoted by n^, let C be a set of numbers

c £ [0,l] and to each c £ C let there correspond a

function A (v) satisfying the conditions of definition 1

such that c does not admit the approximation A (v)

through I. Let f be a function defined on [0,l] such

that if Jt, j is any sequence of real numbers such that

|f(t, )| -• oo as k -» oo then Jt, } has a lim.it point c £ C

and that for each c £ C and every a > 0

(12) f(t) = 0(expj r-i) , as t - c (t / c).

A*(|t-cr^)

Then f o r each x £ [ 0 , l ] where f i s con t inuous t h e r e holds

(13) lira B^ (x) = f ( x ) . n„ n -» oo •

Further (13) holds uniformly in x £ [ a , b ] ^ [0,l] if f

is continuous at each x £ [a,b] .

Proof. In order to prove theorem 3 it is sufficient to

show that, under the given hypotheses, for an arbitrary

6 > 0

130

lim 2 p (x) f ( ^ = 0 m„v n„^

'n '

uniformly in x £ [0,l].

Along the lines similar to the proof of the first

assertion of theorem 2, it can be shov>m that, under the

hypotheses of theorera 3, there exists a finite subset

Jc ,c ,...,c }, say, of C such that for an arbitrary

6 > 0 there exists a positive number A. such that o

|f(t)| g A for all t satisfying |t-c | g 6 ,p=1,2,...,r

Choosing 6 sufficiently small, if a is an arbitrary

positive number, one can choose a positive number

A(a), say, such that for all p=1,2,...,m and all t e [0,1]

f(t) g A(a) expj f -^ } A* (t-c b Cp^l pi

whenever |t-c | < 6 ( t / c ) .

Let A = max JA^, f(c^), f(c2),..., fioj] .

By theorem 1, for all values of v(v=0,1,..., n) such

that |f(v/n^)| g A we have

(14) 2^ Pnv^")^^I^)| IV I . * * | ^ x | g 6

^ r^ A —6 n ^

g 2 A e — 0, as n^ -* oo , -1

uniformly in x £ [0,l], where 2 denotes that the

summation is restricted only to such values of v. 2

Also, if 2 denotes the sumraation restricted to the

remaining values of v, v/e have

131

2^ V (x) f(-^)l

* ,2 m

gA(a) 2 p^_^(x)J 2 exp _ - _ — _ }

/ ^ I r-i T> 1 ' I V ^ -X g 6 ^ • "c ^'n. p' •n^ I P *

m gA(a) 2^ p^ . I2 exp ^~

""*^ p=1 A* (A-^n,)) I — -, - -'n^ ' P P

•X g 6 ^ c ' c

for all n^ sufficiently large (using the property that

for p=1,2,...,m, c does not admit the approximation

A (v) through l). P

Using the monotonicity of A (v) and the definition °P

of A* (t), p=1,2,...,m, and choosing g > 0 to be °P

sufficiently small we have

m (15) A(a) ly p (x) J 2 exp -^ 1

V *• P=1 A* (A^ (n, \—-x\ g 6 ^ • "c ^"c ^"*' 'n^ I " P P

g(n^+i)

g m A(a) e ^ Pn v "*

g(n^+l)-62S g 2m A.'a) e

-• 0, as n^ -* oo ,

uniformly in x £ [0,l]. Thus (14) and (15) together

imply that

I 2 Pn v "" ^^^^1 ^ °' a^ n* ^°° '

uniformly in x £ [0,l]. This completes the proof of

theorem 3.

THEOREI'I 4 Let C be a collection of rational points c

132

lying in [0,l] and let Jp } , n=1,2,..., be a

monotonically increasing sequence of positive integers

diverging to infinity with n, such that for each c £ C

all but a finite number of cp , n=1,2,..., are integers,

Let f be a function defined on [0,l] possessing the

property that if Jt, } is any sequence of real numbers

such that lf(t, )| -* oo as k -» oo then Jt, } has a limit

point c £ C and that for each c £ C and a > 0

(16) f(t) = 0(exp - p ^ ), as t - c, (t ^ c).

Then for each x £ [0,l] where f is continuous there

holds

(17) lim B^ (x) = f(x). n -• oo - n

Also (ll) holds uniformly in x £ [a,b] S. [0,1 ], if f

is continuous at each x £ [a,b],

On the other hand, if f is a non-negative function

such that for some c £ C and for some g > 0 there holds

(18) exp - r ^ = 0(f(t))

as t -• c, (t / c), then

(19) B (x) - 00 , as n - 00 , n

for every x satisfying One of the inequalities

(20) c e~" < X < c

or

(21) c < X < 1 - e~°'(l-c).

Proof Under the given assumptions, for each c £ C, there

holds

inf I - cl g -^ ' p ' p

0 g V g P n - n 1 n

V f cp

133

provided n is sufficiently large. Thus if v/e forbid v

to assume +- e -"-alues cp (n=1,2,...) and define

A(V) = l/v(v > O) then c does not adi.iit the approximation

A(V) through Jp ). Keeping this in mind, rest of the

proof of the first assertion of tneorem 4 follov/s along

the lines of the proof of theorem 3.

To prove the second assertion, it is sufficient

to show that for one of the values of v satisfying

v/p -c| = l/p , v/p e (0,1) and for each x

in the asserted range the sequence

(22) J(^^")x^1-x)^^-^e^^"j , (n=1,2,...),

diverges to i n f i n i t y with n .

The case when c=0 or 1 presents no d i f f i c u l t y and

can be proved r a t h e r e a s i l y . So v/e consider the case

v/hen c £ ( 0 , l ) . Using S t i r l i n g ' s formula

P +i -P Pn. ~ y ^ Pn^ e ^

' ^ , v ^ , - v ^ ^ ^ ^ _ ^ ) - n - v ^ - , - ^ - n - v )

Pn^^ Pn

,r^ v+-i f .Pn-^^S-• n

p -V - 1 ^ ) ^ i - : ^ ) " )

V2iiv(1-cJ ' ~Pn' ' Pn

Hence P P -V P a

C) x ^ 1 - x ) ' - \ ' -V

Pn Pn

134

Let X satisfy one of the conditions

c > X > c e~" or c < X < 1-e~"(l-c).

Then in the first case

1 > J> e-« a n d ^ > 1,

and in the second case

X . . , 1-x V -a

- > 1 and yi^ > e .

Thus, in each case, if n is sufficiently large, we have P p -V P a /• nx v/-, \ n - n (_. ) x''(l-x) e

1-E o(p -maxlp -v,vl)

V27ic(1-c)p^

l-Ej g max Jp^-v,v}

e V'27ic(1-c)p

whore e is a positive number depending on n such that

e -•O asn-*oo. Since n

_i a max Jcp^,(1-c)p^} p " e •^n

as n -• oo , the required divergence of (22) follo\/s.

This completes the pr.. of of theorem 4.

THE0REI4 5 Let I be an unbounded set of positive

integers, A(v) a function as in definition 1 and let

c £ [0,1] be such that it admits the approximation A(v)

through I. Let f,g be tv/o non-negative functions

defined on [0,l] such that for some a > 0

(23) exp f - T - = 0(f(t)), as t - c, (t / c), A*(|t-c|-^)

and

(24) exp - r ^ = 0(g(t)), as t - c, (t / c).

135

Then for each x satisfying one of the inequalities

c e"" < X < c

or c < X < 1 - e""(l-c) ,

at least one of the sets

jB^(x), n £ 1} and JBJ(X), n £ Ij

are unbounded. Moreover, if c is irrational then for all

X satisfying one of the above inequalities

JB ( X ) , n £ Ij is alv/ays unbounded.

Proof. Since c admits the approximation A(V) through I,

there exists an infinite sequence I*, say, consisting

of elements of I such that for each n £ I* there

exists a positive integer v such that

|c - 1 g A(n).

Let I. be the subsequence of I* such that for each n £ I

there holds

0 < |c - Jl g A(n)

for some positive integer v. Also let I denote the

subsequence of I* such that for each n £ I there holds o

c = ^

n

for some integer v •

It is clear that at least one of I and I. is an o 1

infinite sequence. Moreover, if c is an irrational then

I must be infinite.

Consider the case v/hen I is infinite. V/hen c 0

coincides with one of the points 0 or 1 the proof is

relatively easy. So we consider the case when c £ (0,1).

To each n £ I , choose v such that c = v/n. To shov/ the

unboundedness of jB^(x), n £ Ij it is sufficient to prove 136

that - ^

def ( n )^v+1^^_^)n-(v+l) ^ n

as n -» oo through I . We have

1-c X wn! x^ (1 --X) ~^ .gn^ 'n ^c+1/n 1-x' ^ v!(n-v)'

^oc^ w /• T"C X ^/n! x u-x; gnx (25) W^ = (—77- j—){ ..,/ 1 \ / e ).

Frora the proof of the second assertion of theorem 4

it follows that the second factor on the right hand side

of (25) diverges to infinity v/ith n for each x lying

in the given range. Since the first factor of the same

expression converges to a positive limit, the required

divergence of V/ follows,

When I- is an infinite sequence, to each n £ I.

let V he a positive integer such that

0 < Ic - -1 s A(n). ' n' ^

I t follov/s t h a t

A*(|c- jr^) g A*(A-^n)) = ^ .

Thus

exp 7— g exp g(n+l).

A*(ic-jr') Hence to shov/ that jB (x), n £ Ij is unbounded, it is

sufficient to prove that for the given range of x

„ def /nx \),. \n-v a(n+1) V = ( ) X (1-x) e ^ ^- 00 n v

as n -• 00 . This again is evident from the proof of the

Second assertion of theorera 4.

This completes the pr -of of theorem 5.

137

CHAPTER 4

APPROXIMATION OF UNBOUNDED FUNCTIONS BY OPERATORS OF

INTEGRAL TYPE

For linear positive operators of integral type, i.e.

those defined by means of an integral such as

L(f;x) = I f(t) K(t,x)dt ,

L

where K(t,x) is a non-negative kernel and D denotes

the range of integration, integrability of the function

f is demanded by the nature of the definition of L.

For sequences of linear positive operators of integral

type (e.g. the Gamma operators) unboundedness at the

extremities of the range of integration is often seen

to be permissible for the functions to be approximated at

interior points of continuity. Hov/ever, in such cases it is

generally assumed that the functions are bounded in the

intermediate range (collection of all compact subsets

contained in the interior of the closure of the range

of integration). In the sequal we show that, with an

asymptotic analysis of the kernels of a sequence of

linear positive operat^-rs, generally it is possible to

drop down the assumption on_the boundedness of the

functions in the intermediate range. V/e limit our

attention to four operator sequences, viz. the generelized

Jackson operators and the Vallee-Poussin's integrals,

the Gamma operators of Muller and the singular integrals

of Gauss-Weierstrass v/hich provide approximations for

functions defined on a finite interval, a semi-infinite

interval and the whole real line, respectively.

138

4.1 Approximation of unbounded integrable functions

4.1.1 The general method

Let JL , n £ uj be a class of linear positive

operators of integral type and let JK (t,x), n £ Uj

be the class of the corresponding kernels where t and

X range over the subsets X and X of R, respectively.

Assuming that for each fixed n and x, K (t,x) regarded

as a function of t is essentially bounded and measurable

on X the operators L are defined for all functions n

integrable on X. The general method in the following

four cases consists in showing that

(1) Tim sup jK ( t , x ) } = 0

n -»oo t £ X ~ J s : | s -x | < 6 }

for each 6 > 0 and x under consideration. V/hen (l)

holds the basic convergence result can be shov/n to hold

for the unbounded functions provided it holds for the

bounded functions. When we pass on to asymptotic

estimates and asymptotic formulae etc. we also require

an estimate of the speed of convergence of (l).

V/e remark that v/hile dealing with the operators

of integral type it is not necessary to assume the

"strict definiti-/ns" of continuity and differentiability

etc. of the functions to be approximated. Thus, for

instance, if given an arbitrary e > 0 it is possible

to find a 6 > 0 such that

|f(t) - f(x)| < e

for "almost all" t £ (x-6, x+b), then for our purposes

the function f is "continuous" at the point x. The

results in the sequel may be considered in the light

of this slight generality.

139

4.1.2 The generalized Jackson operators L _

For n,p positive integers and x £ R, the operators

L are defined by np-p

(1) L (f;x)=^-^ r f(x+t) (^iM^)2P dt, ^ ' np-p^ ' k \ ^sm i-t ' ' ^ ^ np-p - "

—71

where

71

A = f (^il^)2P dt , np-p j s m gt'

-71

and f is a real or complex valued function integrable

on the interval [x-7t, x+7t]. For the functions f which

are 27t-periodic and integrable on [-71,71] , L (fjx)

reduces to a trigonometric polynomial of degree atmost

np-p. Por bounded functions which are 27i-periodic and

integrable on [-7[,7i]approximation properties of the operators L have been extensively studied. Some

np-p

pertinent references are ([62],[63],[39],[26] and [55])«

LEIH'IA 1 If f is a 271-periodic function integrable on

[-71,7t] , then for an arbitrary positive 6 < TI

. nt

(2) ^ ^ — f |f(x+t)| { - ^ ^ )2P dt

P-P 6 g hi g 7. "^^2

g A n^"^P sin "2P I , (n,p=1,2,...)

v/here A is a positive number depending on f but indepen­

dent of n,x and 6.

Proof V/e have

nt n

r |f(x+t)| (^^^^)2P dt g sin-2P I r|f(t)|dt. 6g'^|t|g7; ^^^2 -\

140

Putting A = A ||f(t)|dt, where A is a number

-71

satisfying

71 . nt s m (3) r ( f )2P dt g A n^P-^ ,(n=1,2,...),

• sin — -7t 2

the inequality (3) follov/s. A proof of the existence of

a number A satisfying (3) is given in [62], p. 51*

THEOREM 1 Let f be a 27t-periodic function integrable

on [-71,7t] such that at a point x both f(x+) and f(x-)

e x i s t . Then f o r p = 1 , 2 , . . .

r A\ n- T /• ^ \ f (x+) + f ( x - ) (4 ) Ixm L ( f ; x ) = ^ ^ '

n — oo

Further, if f is continuous at each x £ [a,b]

S:[-7i,7t] then

(5) lim L (f;x) = f(x) ^ ' np-p^ ' ^ ' n -* CO

uniformly in x £ [a,b] .

Proof. If f is bounded the results (4)-(5) follow from

a theorem of Schurer [62], a theorem of Korovkin [30]

and the symmetry of the positive kernels (sin -p-/sin -xj'

The general case is easily proved from this particular

case and an application of lerama 1 with 6 sufficiently

small.

In a similar way we obtain the followihg

generalization of a theorem of Schurer [62].

THEOREM 2 Let p g 2. If f is a 27[-periodic function

integrable on [-71,71] and tv/ice differentiable at a point

X, then

141

(np-p)

(6) L (f;x)-f(x) = (1 - -J ^)f"(x) + o ( ^ ) , ^ ' np-p^ ) / \ / X (np-p) ^ ' V 2 "

^o ( ^ \

(n - ooj

where the coefficients y P P and p are given by

the identity . nt

/7^ /^^" ~ N 2 P 1 (np-p) 1 "P-P (np-p) (7) (—T") - 6 Po + 3 ,^ p ^ ^^ cos kt,

sxn — k=1

valid for all t.

Further, if f"(x) exists in an open interval

containing the interval [a,b] and is continuous at each

X £ [a,b], then (6) holds uniformly in x £ [a,b].

V/e remark that the theory of linear combinations of the opero-tors L given in [551 can also be extended ^ np-p ' L^-^j

by lemma 1. Thus all the results of [55] are valid for

functions v/hich are integrable but not necessarily

bounded. A similar remark holds for various other known

results on the speed of conver;-;ence of L (f;x) to • ^ np-p^ ' '

f(x).

4.1.3 The Gairima operators G

Let A be the space of all complex-valued functions

f v/hich are measurable in (0,oo ), bounded in any interval

[Y,R] , 0 < Y < R < <», and such that f(t) = 0(e^'^) as

t — 0 and f(t) = 0(t ) as t — oo for some positive

constants a and b. The n-th Gamma operator for f £ A is

defined by ([45])

-1 oo

n+1 r. (1) G (f;x) = 2 L _ ^ ^-^ f(Il)du , X > 0, \ / n^ ' n! J ^u' ' '

0

n=1,?,..., which for a fixed x exists at least for

142

n g max J[J] + 1, [b]j.

Approximation properties of the operators G are studied

in [45],[46],[47][32] and [35]; very recently the

author has obta-ined Nikolski constants, general linear

combinations and general simultaneous approximation

properties of the operators G .

Dropping the boundedness assumption, we define

A* to be the class of all complex-valued functions f

which are measurable in (0,oo ) 3,nd are such that

f(t) = 0{e^''^) as t - 0 and f(t) = O(t^) as t - oo for

some positive constants a and b.

To extend the study of the operators G to the

class A* of ftmctions we require some preliminary

estimations. Ve ha,ve

(2) G (f;x) = ^^r"" f^" e-"^ f(- )du 0

vn+1 ^nx)

n+ - -n _ 0

XI - n u x „/i\-, T u e f(—Jdu -__ n+ - -n ^u^

V?Jt n "e

I ? " vl X / \n f n -nux r.A\.,

= V 2 ^ ^"^) J ^^u)'^^' 0

by an applica,tion of the Stirling's formula for the

gamma functions.

Por x,t > 0 let p(t,x) be the ftmction defined by

(?) p(t,x) = t e - e x .

We keep x fixed and regard p(t,x) as a function of t.

By elementary calculus one easily shows that p(t,x)

is non-negs.tive and assumes its minimum value only at

the point t=x. Thus for an a-rbitrary 6 > 0, the ftmction

143

p(t,x) has a positive lower bound A ( 6 ) , say, for all

t > 0 and satisfying |t-x| g 6 . Hence for |u -x| g 6 ,

where u > 0, we have

xu e_

Thus

g ex + A ( 6 ) . u ^

(4) (^)''S (ex + A(6))-^ , (n=1,2,...). e

Let 6 be a sufficiently small positive number and let

A be a sufficiently large positive number.

Put

1 1

^6,0 = t6o»Aj ~ (^^ , ),

where 6 is an arbitrary number satisfying 0 < 6 < x.

Then if f £ A*, we have

\ { ^ (ex)° / u V ° - l(l)au| I^ 6,0

(5)

-D 1 X / ex \n

= o(Xj)

n

for an arbitrary k > 0. It is readily verified that if

0 < a < b < oo, then (5) holds uniformly in x £ [a,b].

From the known properties of the Gamma operators

we know that if f £ A, then f(t) being majorized by the

functions e and t as t -* 0 and t -» 00, respectively,

for some positive constants a and b, we have for each

fixed X > 0

144

(6) ^ ^ ^ r u V ^ - ^ f(^)du = o(-L), and A

/ \n+1 r /„x Qnx) I n -nux „/1\-, r ^ \ (7) ^ - t ^ e f(-)du = o(—)

for arbitrary fixed k > 0, provided A > 0 is

sufficiently large and 6 > 0 is sufficiently small.

Also , i f 0 < 6 < a < b < A < oothen (6) and (7) ho ld

uniformly in x £ [a,b]

Since (6) and (7) depend only on the fact that

f(t) is majorized by the functions e and t in the

above mentioned sense, if follov/s that A and 6 can be ' o o

chosen so that for a given f £ A*, (6) and (7) hold for

any given x £ (0,oo ). Further 6 and A can be chosen

so tnat for a given f £ A* and a,b satisfying

0 < a < b < 00, (6) and (7) hold uniformly in x £ [a,b].

Now put

6 = f°'~)~ ^' i ) ' 0< <-)• Combining ( 5 )» (6 ) and ( 7 ) we have

LEI-MA 1 Let f £ A* and let 6 be a positive number. Let

X £ (0, 00) be a fixed point such that 6 < x. Then

/ xn+l p ^ A toN (nx) f n -nux „/1\, / 1 ^ (8) i -^^ u e f(-)du = o ( — ) , as n - 00 ,

• i for an arbitrary k > 0. Also i f O < 6 < a < b < o o , then

(8) holds uniformly in x £ [a,b].

Thus in lemma 1 we have shown that the contribution

of f(t) (f £ A*), for t > 0 satisfying ]t-x| g 6 > 0,

to the integral in (2) is of order o(n ) for an

arbitrary k > 0. Hence the asymptotic behaviour of

145

G (f;x) depends only on the behaviour of f in a small

neighbourhood of the point x. Generalizations of various

knov/n results on G (f;x) for f £ A are thus immediate n^ ' '

for f £ A*. To mention a few, we have the following

results.

THE0REI4 1 I f f £ A*, x > 0 and f i s con t inuous a t t h e

p o i n t X, t h e n

(9) lira G^( f ;x ) = f ( x ) , n - • oo

F u r t h e r , i f O < a < b < o o and i f f £ A* i s con t inuous

a t each x £ [ a , b ] , t hen (9) h o l d s un i fo rmly i n

X £ [ a , b ] .

THEOREM 2 If f £ A*, x > 0 and f"(x) exists, then

(10) lira (n-l)[G^(f;x) - f(x)] =^x^f"ix), n -• oo

Further, if 0 < a < b < o o and f"(x) exists in an open

interval containing [a,b] and is continuous at each

X £ [a,b] , then (lO) holds uniformly in x £ [a,b].

For f £ A, (9) and (lO) were proved by Muller [45].

Vi/e mention that results on linear combinations and the

simultaneous approximation property of the Gamma

operators obtained by the author hold for f £ A*. There

are to be published elsev/here.

4.1.4 Singular integrals V/ of Gauss-V/eierstrass

These are defined by

2 o oo n /, x2

n r —T-(t-x) (1) V/^(f;x) = ^ j f(t; e dt,

for X £ (-00 , oo ), n > 0 and for all those functions f

defined on R for v/hich the integral in (l) exists.

Approximation properties of the integrals V/ are well

knovm for the functions f bounded on R. For the class

146

E^ n of functions these properties (and also those of U,R the linear conbinations of W ) are studied in [54].

Let W_ „ denote the class of all functions f each bJ,R

of which is integrable on [-YIY] f°^ each y > 0 and

satisfies the growth restriction ,,2

(2) f(t) = 0(e" ) , as |t| - oo ,

for some A > 0. It is clear that for all sufficiently

large n, W (f;x) exists for all x belonging to a given

bounded subset of R for each f £ W^ „. Q,ii

LEMI-IA 1 Let f £ W^ .„ and let 6 be a positive number. y, K

If X is a fixed point on R, then n ,'+ 2 n r •-p-(t-x)

(3) ^ J f(t) e 2 dt = o ( ^ ) ,

|t-x| g 6 "

as n — oo, for each fixed k > 0. Further if B is any

fixed bounded subset of R, then (3) holds uniformly

in X £ B, for each fixed k > 0.

Proof. For all n sufficiently large, we have

2 o 2,2 n /, N2 n t -^-(t-x) |f(t)|e 2 <i = J l^(^+*)

|t| g 6

(^ -2A)6^ r p.,2 J|f(t+x)| e-2A* dt,

e ^ dt

|t-x| g 6 |t| g 6 -2 o

2 oo — _ V a f s e

where A is a positive number satisfying (2). Now, for

each fixed x £ R (and also uniforraly for each x £ B,

a fixed bounded subset of R) the integral

/ |f(t+x)| e"^** dt

147

is bounded. Since

-i^2A)b' _ ,,) e = o(n ^ ' ) , as n — oo ,

for each fixed k > 0, the lemma follows.

It is evident from lemma 1 that to an order o(n ),

where k is an arbitrary fixed positive number, the

asymptotic behaviour of W (f;x) for f £ W depends

only on the behaviour of f(t) for t lying in an

arbitrarily small neigbourhood of the point x. In

particular from the corresponding well known results

for bounded functions, we deduce.

THEOREM 1 If f £ W^ „, X £ R and f is continuous at x, Ufti

then

(4) lim V/^(f;x) = f(x). n -• oo

Further, if f is continuous at each x £ [a,b]

(-00 < a < b < oo), then (4) holds uniformly in x £ [a,b].

THEOREM 2 If f £ W. _ , x € R and f is twice differen-

tiable at x, then

(5) lim n2[w^(f;x) - f(x)] = ^ . n -* 00

Further, if f"(x) exists in an open interval containing

[a,b](-oo < a < b < 00) and is continuous at each x £ [a,b]

then (5) holds uniformly in x £ [a,b].

In a similar way lemma 1 extends other known results

on the asymptotic estimates of the quantity V/ (f;x) and

the linear combinations ([54]) of W etc.

4.1.5 De La Vallle-Poussin integrals V

These are defined by 71

(^) V " ' " ) = 2,;(2n-JJM / f ( x + t ) c o s 2 n I d t ,

148

n=1,2,..., for all functions f and points x € R for

which the integral in (l) exists. In particular, if f

is integrable on [-71,71] and is a 27i -periodic function

then V (f;x) exists for all x and n and defines a n^ '

trigonometric polynomial of degree g n.

LEMMA 1 Let f be a 27t-periodic function integrable

on [-71,7i] . If 6 < 71 is an arbitrary fixed positive

number, then

(2) 2 i { i ^ [ If(-+*)! -^'" I ^^ = °(i^)' K g |t| g 6

as n -• 00, for each fixed k > 0, uniformly for all x £ R.

Proof. We have 7t

r if(x+t)i cos^'' I dt g cos^" I r i f ( t ) |d t . TC g | t | g 6 - n

Since

and

( 2 n ) ! ! „/ -i-N (2n- l ) ! ! = °(^ ) ' "

2n 6 „ / -(k+4-)N cos r- = o ( n '^') , n -* 00

for an arbitrary fixed k > 0, (2) follows.

Using lemma 1, from the corresponding well knov/n

results on the approximation of bounded functions by the

integrals V ([51], [62]), in particular, we deduce

THEOREM 1 Let f be a 27i-periodic function integrable

on [-71,7i] . If f is continuous at a point x £ R, then

(5) lira V^(f;x) = f(x). n -• 00

Further, if f is continuous at each point x of an inter­

val [a,b], then (3) holds uniforraly in x £ [a,b].

149

THEOREM 2 Let f be a 27t-periodic function integrable

on [-71,Tt]. If f"(x) exists at a point x £ R, then

(4) lim n[V^(f;x) - f(x)] = f"(x). n -» oo

Further, if f"(x) exists at each x belonging to an

open interval containing an interval [a,b] and is

continuous at each x £ [a,b], then (4) holds uniformly

in X £ [a,b].

150

SFERENCES

[l] V.A. Baskakov, An example of a sequence of linear

positive operators in the space of continuous

functions, Dokl.Akad.Nauk. 113(1957), 249-251

(in Russian).

[2] , A generalization of the Bernstein

polynomials, Izv. Vyss. USebn.Zaved.Matematika,

i960. No. 3(16), 48-53 (in Russian)

[3] , On a construction of positive linear

operators for functions of two variables, Izv.

Vyss.USebn.Zaved. . Matematika, I963, No. 4(35)

7-14 (in Russian).

[4] , The degree of approximation of

differentiable functions by certain positive

linear operators, (Russian), Mat.Sb.(N.S.),1968,

No. 76(118), 344-361.

[5] H. Bohman, On approxiraation of continuous and of

analytic functions, Ark.Mat. 2(l952), 45-57.

[6] T.J. Bromv/ich (l926). An Introduction to the

Theory of Infinite Series, Cambridge 1926.

[7] P.L. Butzer, Linear combination of Bernstein

polynomials. Can. J.Math., 5(l953), IO7-II3.

[8] , On the extension of Bernstein polyno­

mials to the infinite interval, Proc, Amer.Math.

Soc. 5(1954), 547-553.

[9] E.V/. Cheney and A.Sharraa, Bernstein power series,

Can. J.Math 16(1964), 241-252.

10] R.A. De Vore, The Approximation of Continuous

Functions by Positive Linear Operators, Springer

Verlag, Berlin-New York, 1972.

[11] S. Eisenberg, Moment sequences and the Bernstein

polynomials. Can.Math.Bull, 1969, No. 12(4),

401-411.

[12] S. Eisenberg and B. V/ood, Approximating unbounded

functions with linear operators generated by

moment se^iUences, Studia Mathematica, 35(l970),

299-304.

[13] , Approximation of analytic functions by

Bernstein-type operators, Jr. Approx. Th, 6(l972),

242-248.

[14] , On the order of approximation of unbounded

functions by positive linear operators, SIAM

J.Numer.Anal. 9(l972), 266-276.

[15] G. Freud, On approximation by positive linear

methods, Studia Scientiarum Mathematicarum

Hungarica, 2(1967), 63-66.

[16] P.R. Halmos, Measure Thewry, D. Van Nostrand

Company, New York, 1950.

[17] L.C. Hsu, Approximation of non-bounded continuous

functions by certain sequences of linear positive

operators or polynomials, Studia Mathematica

21(1961), 37-43.

[18] L.C. Hsu and J.H. Wang, General "increasing

multiplier" methods and a,pproximation of unbounded

functions by certain concrete polynomial operators,

Dokl.Akad.Nauk SSSR 156(1964), 264-267.

[19] D. Jackson, The theory of Approximation, Amer.Math.

Soc. Colloquium Publications, vol. 11, New York,

1930.

[20] A. Jakimovski and D. Leviatan, Generalized

Bernstein polynomials. Math. Zeitschr. 93(l966),

416-426.

[21] , Generalized Szasz operators for the

approximation in the infinite interval, Mathema­

tica (Cluj) 1969 , 11(34), 97-103.

[22] C.Jordan, Calculus of Finite Difference, New York

i960.

[23] S. Karlin, Total Positivity, Stanford University

press, 1968.

[24] S. Karlin and './.J. Studden,Tchebyclieff Systems:

With Applications in Analysis and Statistics,

Interscience Publishers, New York, 1966.

[25] J.F. Koksma, Diophantische Approximationen

(Berlin, 1936), Ergebnisse der Math.Wiss.,vol. 4.

[26] E.A. Koraleva, On an asyraptotic property of

positive summation methods of Fourier series,

Izv.Vyss.U^ebn. ZaVed. Matematika 4(l959),

89-93, (in Russian).

[27] P.P. Korovkin, Convergence of linear positive

operators in the space of continuous functions,

D.A.N. 90(1955), 961-964 (in Russian).

[28] , On the order of approximation of the

functions of linear positive operators, D.A.N.

114, No. 6, (1957), 1158-1161 (in Russian).

[29] , On an asymptotic property of positive

summation methods of Fourier series and on the

153

best approximation of methods £ Zp by linear

positive polynomial operators, Uspehi Mat.Nauk.,

13, No. 6 (84), 1958, 99-103 (in Russian)

[ 50] , Linear Operators and Approximation Theory,

Hindustan Publ. Corp., Delhi, I960.

[31] D. Leviatan, On the remainder in the approximation

of functions by Bernstein type operators,

J. Approx.Th. 2(1969), 400-409.

[32] D. Leviatan and M. Muller, Some applications of

the Gamma operators, Arch.Math, 20(1969), 638-647.

[33] G.G. Lorentz, Bernstein Polynomials, University

of Toronto Press, Toronto, 1953.

[34] , Approximation of Functions, Holt,Rinehart

and Vi/inston, Inc., New York, 1966,

[35] A. Lupaj and M. Miiller, Approximationseigenschaften

der Gam.maoperatoren, Piath. Zeitschr. 98(1967),

208-226.

[36] R.G. Mamedov, Asymptotic approximation of

differentiable functions with linear positive

operators, D.A.N. 128(1959), 471-474 (in Russian)

[37] , On the order and asymptotic value of the

approximation of functions by positive linear

operators (Russian Azerbaijani summary), Izv.

Akad.Nauk.Azerbaidzan. SSR. Ser. Piz.-Mat.Teh.

Nauk, no. 2(l96l), 27-39.

[38] R. Martini, On the approximation of functions

together with their derivatives by certain

linear positive operators, Indag. Math. 31fNo. 5

(1969), 473-481.

154

[39] Y. Matsuoka, On the degree of approxiraation of

functions by some positive linear operators,

Sol. Rep. Kagoshiraa Univ. 9(l960), 11-16.

[40] , Asymptotic formula for Valine Poussin's

singular integrals, Sci.Rep. Kagoshiraa Univ.

9(1960), 25-34.

[41] P.K. Menon, A class of linear positive operators,

J. Indian Math. Soc. 26(l962), 77-80.

[42] V/. Meyer-Konig and K. Zeller, Bernsteinsche

Potenzreihen, Studia Math. 19(l960), 89-94.

[43] C.A. Micchelli, Saturation Classes and Iterates

of Operators, Ph.D. Thesis, Stanford University

1969, 100 p.

[44] B. Mond and O.Shiaha, On the approximation of

functions of several variables. J. Res. Nat.Bur.

Standards Sect. B, 70 B(1966), 211-218.

[45] M« Muller, Die Folge der Gammaoperatoren, Doctoral

Thesis, Technische Hochschule Stuttgart, 1967,

87 p.

[46] , Einige Approximationseigenschaften der

Gararaaoperatoren, Mathematica (ciuj) 10(33)»2,

(1968), 303-310.

[47] ,Tiber die Approximation durch Gammaoperatoren,

Abstract Spaces and Approximation, Proceedings of

the Conference at Oberv/olfach, ISNM 10(1969),

Birkhauser, Basel.

[43] , On asymptotic approximation theorems for

sequences of linear positive operators. Approxi­

mation Theory (Proc. Sympos. Lancaster, 1969),

155

315-320, Academic Press, London, 1970.

[49] , Approximation durch lineare positive

Operatoren bei gemischter Norm, Habilltations-

schrift, UniversitSt Stuttgart, 1970, 119 p.

[50] M.V/. Muller and H. Walk, Konvergenz- und Giite-

aussagen fur die Approximation durch Folgen

linearer positiver Operatoren,221-253,

Constructive Function Theory (Proc.Conf.Golden

Sands (Varna), May 19-25, 1970), Publishing

House of the Bulgarian Acad.of Sci., Sofia, 1972.

[51] I.P. Natanson, Constructive Function Theory,

vol. 1, Frederick Unger Publ. Co., Nev/ York, 1964.

[52] G. Pdlya and I.J. Schoenberg, Remarks on de la

Valine Poussin means and convex conformal maps

of tiie circle, Pac. J. Math., 8, No. 2(l958),

295-334.

[53] E.D. Rainville, Special Functions, The Mac-Millan

Company, New York, 196O.

[54] R.K.S. Rathore, Linear Combinations of Linear

Positive Operators and Generating Relations in

Special Functions, Ph.D. Thesis, Indian Institute

of Technology Delhi, 1973, 23O p.

[55 ] , On a sequence of linear trigonometric

polynomial oper tors, (to appear in SIAM J. Math.

Anal. Vol. 5, No.6, November 1974).

[56 ] G. Schmid, Approximation unbeschrankter Funktionen,

Doctoral Thesis, Universitat Stuttgart, 1972,

79 p.

[57] I.J. Schoenberg, On variation diminishing

156

approximation methods, 249-274, On Numerical

Approximation (ed. R.E. Langer), Madison, 1959.

[58] , On spline functions, 255-291, Inequalities,

Academic Press Inc. Nev/ York, I967.

[59] F- Schurer, On the approximation of functions of

many variables v/ith linear positive operators,

Indag. Math. 25, No. 3, 1963.

[60] , Some remarks on the approximation of

functions by some positive linear ojjerators,

Monatshefte fiir Mathemr.tik, Bd 67/4, 1963,355-358.

[61] , On linear positive operators, 190-199,

On Approximation Theory, ISM, Vol. 5(1964), Birk­

hauser Verlag, Basel/Stuttt^art.

[62] , On Linear Positive Operators in Approxima­

tion Theory, Doctoral Thesis, Technicche Hogeschool

Delft, 1965, 79 P .

[63] P. Schurer and P.V/. Steutel, On linear positive

operators of the Jackson type. Mathematical

Communications, Technological University of Tv/ente,

no. 1, October I966,

[64] 0. Shisha and B. Mond, The degree of convergence of

sequences of linear positive operators, Proc.Nat.

Acad. Sci. U.S.A. 60(l968), 1196-1200.

[65] P.C. Sikkema, Der V/ert einiger Konstantcn in der

Theorie der Approximation mit Bernstein-Polynomen,

HumeribChe Mathematik, 3(1I6I),107-116.

[66] , Uber Potenzen von verallgemeinerten Bern-

stein-Operatoren, Mathematica (Cluj) vol. 8(31),1,

(1966), 173-180.

157

[67] , On some linear positive operators, Indag.

Math., 32, Ho. 4, 1970, 327-337.

[68] , On the asymptotic approximation with

operators of Meyer-Kbnig and Zeller, Indag.Math.

32, No. 5, 1970, 428-440.

[69] , On some research in linear positive

operators in approximation theory, Nieuv/ Archief

voor Wiskunde (3), 18(l970), 36-6O.

[70] , Uber die Schurerschen linearen positiven

Operatoren (to appear in Indagationes Mathematicae).

[71] D.D. Stancu, A new class of uniform approximating

polynomial operators in two and several variables,

443-455, Proceedings of the Conf. on Constructive

Theory of Functions, Budapest, August 24-Sept.3,

1969.

[72] , Probabilistic methods in the theory of

approxim.ation of functions of several variables

by linear positive operators, 329-342, Approximation

Theory (Proc. Syrapos., Lanc8,ster, 1969), Academic

Press, London, 1970.

[73] E. Stark, Uber trigonometrische singulare Faltungs-

integrale mit Kernen endlicher Oszillation,

Doctoral Thesis, Technische Hochschule Aachen,1970,

85 p.

[74] Y. Suzuki, Sa,turation of local approximation by

linear positive operr.tors of Bernstein type,

Tohoku Math. Journ. 19(4), (1967), 429-453.

[75] C.A. Timmermans, A generalization of a theorem of

Mamedov, Constructive Function Tneory (to appear),

158

(proc. conf. Cluj, September 6-12. 1973)'

[76] E.G. Titchmarsh, The Theory of Functions, 2nd.

ed., Oxford Univ. Press, 1939.

[77] V.I. Volkov, Convergence of sequences of linear

positive operators in the space of continuous

functions of two variables, D.A.N., 115(1957),

I7-I9, (in Russian).

[78] , Some sequences of linear positive

operators in the space of continuous functions of

two variables, Kalinin. Gos.Ped.Inst.Pic.Zap. 26

(1958), 11-26, (in Russian).

[79] , Conditions for convergence of a sequence

of pointv/ise linear operators in the space of

continuous functions of tv/o variables, Kalinin,

Gos.Ped.Inst.Mc.Zap. 26(1958), 27-4O.

[80] H. V/alk, Approximation durch folgen linearer

positiver Operatoren, Arch.Math. 20(l969),398-404.

[81] , Approximation unbeschrankter Funktionen

durch lineare positive Operatoren, Habilitations-

schrift, Universitat Stuttgart, 1970, 110 p.

[82] B. V/ood, Convergence and almost convergence of

certain sequences of positive linear operators,

Studia Math. 34(2),(197"),113-119.(MR 4I =# 4086)

[83] , Generalized Szasz Operators for the

approximation in the complex domain, SIAM J. Appl.

Math. 17(1969), 790-801.

[84] , Graphic behaviour of positive linear

operators, SIAM J. .A.ppl. Math. 20 (l97l) 329-335.

159

Samenvatting

Deze dissertatie behandelt de bepaling van klassen van

onbegrensde functies, die met behulp van lineaire posi-

tieve operatoren, behorende tot een gegeven verzameling,

kunnen worden geapproximeerd. Zulk een bepaling wordt

mogelijk gemaakt door het gedrag van deze operatoren

te onderzoeken t.o.v. zekere functies die tot de verza­

meling van testfuncties of begrenzende functies behoren.

Een andere toegang tot dit probleem is het analyseren

van de kernen van de operatoren, indien informatie viae-

nige begrensde functies resulteert in een bepaling van

approximeerba,re onbegrensde functies.

Hoofdstuk 1 bespreekt de methode der testfuncties

en heeft het doel hun onbegrensdheid optimaal te ge-

bruiken teneinde approximeerbare onbegrensde functies

te bepalen, die een vergelijkbare orde van onbegrensdheid

bezitten. De fundamentele convergentie- en asymptotische

formules worden afgeleid; zij geven een nauwkeurige

graad van approximatie.

In hoofdstuk 2 wordt de methode der begrensde func­

ties bestudeerd; het doel is meer omvattende klassen van

approximeerbare functies te bepalen, dan door anderen

verden verkregen. Hier bestaat de techniek eveneens uit

het benutten van de voile omvang van het onbegrensd

zijn van de testfuncties.

Hoofdstuk 3 behandelt de analyse van de kernen van

operatoren van het sommatietype. In het geval van de

Bernstein operatoren worden zekere klassen van onbe­

grensde functies verkregen, die met behulp van deze

operatoren of deelwijze daarvan approximeerbaar zijn.

Hoofdstuk 4 analyseert op overeenkomstige wijze de

160

de kernen van operatoren van het integraaltype. Er

wordt aangetoond dat met velen daarvan niet-begrensde

functies in een punt, waar zij continu zijn, kunnen

worden geapproximeerd. Wij beperken ons daarbij tot de

generaliseerde operatoren van Jackson, de integraal-

operatoren van De La Vallee-Poussin, de operatoren,

gevormd door raiddel van de singuliere integralen van

Gauss-Weierstrass en de Gammaoperatoren van Muller.

161

Stellingen

Let JL j(n=1,2,...) be a sequence of variation

diminishing linear operators mapping P -» P and /• '\ f -r>\ m ra

C ' [a,b] - C [a,b] (m,y-0,1, 2,...), v/here P is the space of all algebraic polynoraials of degree g m and

C ' [a,b] i"' the space of all r-times continuously differentiable functions defined on [a,b].

F o r each f £ C^''^[ a , b ] , | | L ^ ^ ^ f ) - f < '' | |(j[-a,b]-^0 " n -• oo i f and on ly i f | | L ( t ; x ) - x M r -i -• 0 a s n -* oo

n (_/[_ a , DJ

for 1=0,1,2. Here r takes any value 0,1,2,... .

For an arbitray j;*-0,1 ,2, .. . for each f £ C ' [a,b] and

X £ [a,b] such that f " (x) exists v/e have

L^^)(f;x) - f^^)(x) = p4^J(r-l)(r-2)a^-2.?(r-2)a2

+ r(r-l)a4 f ' (x) + J2x[(r-l)a^-(2r-l)a2+ra^]

+ a,+ra^}f^^"^ - (x) + Jx (a^-2a2+a^)+x(a -2aj)

+ a^]f^''^^hx)] + o ( ^ ) , (n-~)

where (p(n) j^ 0, (fin) -*oo, n-*oo and a., 1 = 1,...,6 are

constants if and only if a. ^

L (l;x) = 1 + -7-^- + o(-7-^)

L^(t;x) = X + ^/,/ + o{^j^) (n - oo )

„ „ a X +acX+a^ . T f J.2 ^ 2 n 5 6 ^ 1 • L (t ;x) = X + 7^ + (—7~v n ' ' 9(n) l)(n)'

and for some p o s i t i v e i n t e g e r m

>2m+2 ^ /• 1 _ I [ t - x n L„( ( t -x) ^ ;x) = o ( ^ ^ ) , ( n - o o )

I I

A linear operator L has total variation diminishing

property for all f, i.e. T[L(f)] g T[f], if and only

if it has such ^.joperty for 'all monotone functions.

I l l

Let A,B be subgroups of a finite group. Let A v B denote

the subgroup generated by all the elements of A a-nd B.

Let the symbol [G] denote the order of a group G. Then

if and only if AB = BA.

IV

Let 0 < a < p < oo and let L be a linear operator

defined on a d imain of continuous functions defined on

an interval. Given a function g(t) define

g(a) , t < a

&-^(t) = I g(t) , a g t g p g(p) , t > p .

Por each f there holds

T[L(f)P] g T[fP] a-' •- a-

if and only if L is variation diminishing and L(l)=1.

The "if" part of this assertion is a result of I.J.

Schoenberg,

V

Let L be a non-trivial (i.e. not of the form L(f)

= ()i(f)T(x) v/here (|) is a functional and f a fixed

function) linear operator. L is said to be VD (variation m ^

diminishing of order m( if v(L(f)) g v(f) v/henever v(f) g m, (v(f) denoting the number of sign changes of the function f). L is said to be VD on a function m

' m ^

if v(L(f)-(p) g v(f-(p) v/henever v(f-(p) g m. L is VD on q)

if and only if L is VD and (p is a fixed point of L.

4

VI

For p=1,2,... let (L,p) be the summability method

defined by the matrix (a (p)) (n,v=0,1 , 2,,..) v/here

a (p)=1, a (p)=0, v=1,2,... and for n=1,2,... and 00 Ov

v=0,1,2,... a_(p) = c_(p)/^,(^'P)

where

and

nv nv 0

,,(n,p) _ ^~ ( -,\Jr2P I'np+p-nj-ls

^ " jfo j 2p-i ^ • (L,P) is a strongly regular summation method for

p=1,2,.... If f is a 27t-periodic integrable function

possessing the r derivatives f '^(x+), f (x-)

(r=0,1,2,...) at a point x, then S^ (x) the r

differentiated Fourier series of the function f is

(L,p) summable to the value (f ^Hx+)+f ^^x-) )/2 if r+2

p g """p™. Here, as is usual a derivative of order zero is synonymus with the function.

VII

Let f be a complex valued function defined on (O, oo)

and satisfying

|f(t)| < A(e«* + e«/^) , t £ (0,00 )

where A and a are two positive numbers. Let I (z) be th ^

the n modified Bessel function z" izZ2£ -2

f J x ) = f(]j (f)= + 1 + | ) , X £ (-00 ,oo )

Let

and define

Jjf;x) = exp J-n(x+^)j I i (2n)x-fJ^) v=-<»