General Mathematics

Vol. 6 /1998

ROGER
1.06{3.06.1998, Sibiu-Romania

On a problem of A. Lupas
Octavian Agratini
1 Introduction
At the International Dortmund Meeting held in Witten (Germany, March, 1995), A. Lupas [4] formulated the following problem. " Starting with the identity
1

()k k
a ; jaj < 1;
=
(1 a) k=0 k!
let  = nx; x  0, and consider the linear positive operators
(1)

(Lnf )(x) = (1

1
X

0 1

1 (nx)k k k
X
a)nx
a f @ A;
n
k=0 k !

x  0;

with f : [0; 1) ! R. If we impose that Lne = e we nd

a = 1=2. Therefore
1 (nx)k 0 k 1
X
nx
f @ A ; x  0:
(2)
(Lnf )(x) = 2
k
2 k! n
k
1

=0

On a problem of A. Lupas

This Ln-operator has a form very similar with Szasz-Mirakyan

operators. We have Lnh = h; h 2  and
1

nlim
!1(Lne2)(x)

= e2(x):

Find other properties of Lnf ."

The focus of this note is to investigate these operators. An
asymptotic formula and some quantitative estimates for the rate
of convergence are given. By using a probabilistic method, this
sequence is reobtained. Also two modied sequences are constructed.

2 The results
Firstly, we recall the common notation
()0 = 1; ()k = ( + 1) : : : ( + k 1); k  1:
For any real x  0 and integer r  0 we set
er (x) := xr ;

x;r (t)

:= (t x)r (t  0);

n;r (x) := (Ln

x;r )(x):

At this point, it has been proved that Lner = er ; r 2 f0; 1g.

Octavian Agratini

Remarks. 1) We can consider that Ln; n  1, are dened

on E where
E=

[
a>0

Ea

and Ea is the subspace of all real valued continuous functions f

on [0; 1) such as
e(f ; a) := sup(exp( ax)jf (x)j) < 1:
x0

The space Ea is endowed with the norm kf ka = e(f ; a) with

respect to which it becomes a Banach lattice.
2) In our investigations we also need to consider the Banach
lattice CB [0; 1) of all real-valued bounded continuous functions
on [0; 1) endowed with the sup-norm k
k1. The operator Ln
maps CB [0; 1) into itself, it is continuous with respect to the
sup-norm and kLnk = kLne0k1 = 1.

Lemma 1. If Ln is dened by (2) then, for each x  0, the

following identities are valid
(Lne2)(x) = x2 +

2x
2x
; n;2(x) = :
n
n

On a problem of A. Lupas

Theorem 1. If Ln is dened by (2) then one has

nlim
!1 Lnf

= f uniformly on [0; b];

for any b > 0.

Theorem 2. Let Ln be dened by (2) and b > 0. One has


p  0B 1 1C
j(Lnf )(x) f (x)j  1 + 2b ! @f ; pn A ; x 2 [0; b]:
1

If f has a continuous derivative on [0; b] then

p 0
1
2b + 2b B@ 0 1 CA
j(Lnf )(x) f (x)j  p ! f ; p ; x 2 [0; b]:

n 1
n
In the following we are going to prove another estimate by

involving the second order modulus of smoothness. In fact, our

estimate will be based upon a more general theorem which is
due to Gonska ([3], Theorem 4.1, page 331).

Theorem 3. Let

Ln be dened by (2) and b > 0. The

following inequality

j(Lnf )(x)
x 2 [0; b]; holds.

1
11 0
0
0
1
b
f (x)j  @3 + 2b max @1; AA !2 B@f ; p CA ;

Also we establish a Voronovskaja-type formula.

Octavian Agratini

Theorem 4. Let f 2 C [0; 1) be twice dierentiable at some

point x > 0 and let us assume that f (t) = O(t ) as t ! 1.
2

If the operators Ln are dened by (2) then

' (x) 0
(3)
f (x);
lim
n
((
L
f
)(
x
)
f
(
x
))
=
n
n!1
2
p
holds, where ' is dened by '(x) = 2x, x  0.
Actually, ' represents the step weight functions of the Lupas
operators and it controls their rate of convergence.
It is known that by using some concepts of the probability
theory have been obtained several classical positive and linear
operators. Pioneers in this eld to be mentioned here are W.
Feller [2] and D.D. Stancu [6].
Let (Xj;x)j  be a sequence of independent random variables
identically distributed
(x)
(4)
P (Xj;x = k) = 2 x k k ; k  0;
k!
where x is a positive real parameter. Denoting by  the common
characteristic function of these random variables, the identity (1)
implies
2

(t) =

1 itk
X
e P (X

k=0

j;x

= k) = (2 eit) x:

On a problem of A. Lupas

If we set

1 Xn
Yn;x :=
X ; n  1;
n j =1 j
then the characteristic function of Yn;x will be n(t) = n(t=n)
which corresponds to the following distribution:
P (Yn;x = k=n) = ln;k (x)

where
(5)

ln;k (x) := 2

nx (nx)k ;
2k k!

k  0:

Furthermore, for every n  1 and every f 2 E we consider the

function Lnf : [0; 1) ! R dened by (Lnf )(x) := M (f  Yn;x)
where M (Z ) represents the mathematical expectation of Z . This
way we obtain the Lupas operators.
As a matter of fact, all those approximation processes (Pn)n1
of probabilistic type which are associated with a random scheme
1 Xn
Zn;x =
X
(n  1; x 2 I; Xk;x i.i.d.)
n k=1 k;x
satisfy the formula
2(x) 0
nlim
!1 n((Pnf )(x) f (x)) = 2 f (x)
for every f 2 CB2 (I ), see [1, page 368]. Here 2(x) = Var(Xk;x)
represents the variance of Xk;x. For the variables Xn;k dened

Octavian Agratini

by (4), we obtain
2
M (Xk;x) = x; M (Xk;x
) = x2 + 2x; Var(Xk;x) = 2x:

So, in the particular case when f

(3).

Theorem 5. If n  1;

 0;

2 CB [0; 1) we come across

2

2 (0; 1) and ln;k (x) is

dened by (5) then the following relations hold true

nx + k
l (x).
i) ln;k (x) =
2(k + 1)0 n;k
1
kX
0
ii) ln;k (x) = nln;k (x) @ (nx + i)
log 2A ; (k 6= 0),
+1

i=0

Z1

k
X

( 1)k isk;ii!(log 2) i ,
iii) ln;k (x) =
i
4(2x + 3x + 2)
pnx .
iv) ln;k (x) <
Here sk;i represents the Stirling numbers of the rst kind.
(n2k k!)

=0

3 Extensions
In order to obtain an approximation process in spaces of integrable functions, we introduce two integral modications of these
operators, Kantorovich-type operators
(Knf )(x) = n

1
X
k=0

ln;k (x)

k Z =n

( +1)

k=n

f (t)dt;

On a problem of A. Lupas

10

respectively Durrmeyer-type operators

(Dnf )(x) =

1
X
k=0

Z1

cn;k ln;k (x) ln;k (u)f (u)du:

0

The coecients cn;k are dened as follows

cn;k =
1

Z1
0

ln;k (x)dx:

In fact, this guarantees the relation Dne0 = e0. Also, we easily

obtain
1
(Kne0)(x) = 1; (Kne1)(x) = x + ;
n
3x 1
(Kne2)(x) = x2 + + 2 :
n 3n
As regards these integral operators we raise the problem to
investigate their convergence in Lp-spaces.

References
[1] F. ALTOMARE and M. CAMPITI: Korovkin-Type Approximation
Theory and its Applications. de Gruyter Series Studies in Mathematics, Vol.17, Walter de Gruyter, Berlin, New-York, 1994.
[2] W. FELLER: An Introduction to Probability Theory and its Applications I,II. John Wiley, New-York, 1957 resp. 1966.
[3] H.H. GONSKA and J. MEIER: Quantitative theorems on approximation by Bernstein-Stancu operators, Calcolo 21(1984), 4, 317335.

Octavian Agratini

11

[4] A. LUPAS: The approximation by some positive linear operators.

In: Proceedings of the International Dortmund Meeting on Approximation Theory (M.W. Muller et al., eds.) Akademie Verlag,
Berlin, 1995, pp.201-229.
[5] M.M. RAO: Probability Theory with Applications. Academic Press,
New York, 1984.
[6] D.D. STANCU: Use of probabilistic methods in the theory of uniform
approximation of continuous functions, Rev. Roum. Math. Pures et
Appl. 14(1969), 5, 673-691.