Professional Documents
Culture Documents
Modular Spaces Revised
Modular Spaces Revised
1. Introduction
In this paper, we investigate the problem of statistical approximation
to a function f by means of positive linear operators de…ned on a modu-
lar space. So, we obtain various Korovkin-type approximation theorems in
modular spaces via the concept of A-statistical convergence, where A is a
non-negative regular summability matrix. With the help of the notions of
modular convergence and strong convergence, some approximation theorems
have recently been introduced by Bardaro and Mantellini [3]. However, in
this paper, we replace the strong (or modular) convergence by the more
general A-statistical convergence with respect to a …xed summability ma-
trix A: When A is the identity matrix, our results reduce to the results in
[3]. Also, we display an example satisfying all conditions of our statistical
approximation result in a modular space but not the classical one in [3].
We now recall some basic de…nitions and notations used in the paper.
Let
A := [ajn ] (j; n 2 N := f1; 2; 3; :::g)
be an in…nite summability matrix. For a given sequence x := fxn g, the
A-transform of x, denoted by
Ax := f(Ax)j g;
is given by
1
X
(Ax)j = ajn xn ;
n=1
Key words and phrases. Positive linear operators, modular space, regular matrix, sta-
tistical convergence, Korovkin theorem.
2000 Mathematics Subject Classi…cation. 41A25, 41A36.
1
2 SEVDA KARAKUŞ, KAMIL DEMIRCI AND OKTAY DUMAN
provided the series converges for each j 2 N. We say that A is regular (see
[12]) if
lim Ax = L whenever lim x = L:
Assume that A is a non-negative regular summability matrix. Then the
sequence x = fxn g is called A-statistically convergent to L provided that,
for every " > 0;
X
(1.1) lim ajn = 0:
j
n: jxn Lj "
We denote this limit as follows (cf. [8]; see also [5, 6, 13])
stA lim x = L:
Actually, this convergence method is based on the concept of A-density.
Recall the A-density of a subset K N, denoted by
A fKg;
is given by
1
X
A fKg = lim ajn K (n) ;
j
n=1
provided the limit exists, where K is the characteristic function of K; or
equivalently X
A fKg = lim ajn :
n
n2K
So, by (1.1), we easily see that
stA lim x = L i¤ A fn : jxn Lj "g = 0 for every " > 0:
We should note that if we take A = C1 := [cjn ], the Cesáro matrix de…ned
by 8
< 1
; if 1 n j;
cjn := j
: 0; otherwise,
then A-statistical convergence reduces to the concept of statistical conver-
gence (cf. [7]; see also [9, 10, 11]). In this case, we write
st lim x = L instead of stC1 lim x = L:
Further, taking A = I, the identity matrix, A-statistical convergence coin-
cides with the ordinary convergence, i.e.,
stI lim x = lim x = L:
Observe that every convergent sequence (in the usual sense) is A-statistically
convergent to the same value for any non-negative regular matrix A, but its
converse is not always true. Actually, in [13], Kolk proved that A-statistical
convergence is stronger than convergence when A = [ajn ] is a non-negative
regular summability matrix such that
lim maxfajn g = 0:
j n
STATISTICAL APPROXIMATION BY POSITIVE LINEAR OPERATORS 3
The concepts of statistical limit superior and limit inferior have been
introduced by Fridy and Orhan [11]. A-statistical analogs of these concepts
have been examined by Connor and Kline [5], and Demirci [6] as follows.
The A-statistical limit superior of a number sequence x = fxn g, denoted by
stA lim sup x;
is de…ned by
sup Bx , if Bx 6= ;
stA lim sup x =
1, if Bx = ;
where Bx := fb 2 R : A fn : xn > bg = 6 0g and denotes the empty set. We
note that by A fKg = 6 0 we mean either A fKg > 0 or K fails to have
A-density. Similarly, the A-statistical limit inferior of fxn g; denoted by
stA lim inf x;
is de…ned by
inf Cx , if Cx 6= ;
stA lim inf x =
+1 , if Cx = ;
where Cx := fc 2 R : A fn : xn < cg = 6 0g. Of course, if we take A = C1 ;
then the above de…nitions reduce to the concepts of st lim sup x and st
lim inf x given in [11], respectively. As in the ordinary limit superior or
inferior, it was proved that
stA lim inf x stA lim sup x
and also that, for any sequence x = fxn g satisfying A fn : jxn j > M g = 0
for some M > 0;
stA lim x = L i¤ stA lim inf x = stA lim sup x = L:
We now focus on modular spaces.
Let I = [a; b] be a bounded interval of the real line R provided with the
Lebesgue measure. Then, by X (I) we denote the space of all real-valued
measurable functions on I provided with equality a.e. As usual, let C (I)
denote the space of all continuous real-valued functions, and C 1 (I) denote
the space of all in…nitely di¤erentiable functions on I. In this case, we say
that a functional : X (I) ! [0; +1] is a modular on X (I) provided that
the following conditions hold:
(i) (f ) = 0 if and only if f = 0 a.e. in I,
(ii) ( f ) = (f ) for every f 2 X (I),
(iii) ( f + g) (f )+ (g) for every f; g 2 X(I) and for any ; 0
with + = 1.
Recall that a modular is called N -quasi convex if the following is satis…ed:
there exists a constant N 1 such that
( f + g) N (N f ) + N (N g)
holds for every f; g 2 X (I), ; 0 with + = 1. In particular,
if N = 1; then is called convex.
4 SEVDA KARAKUŞ, KAMIL DEMIRCI AND OKTAY DUMAN
and
E (I) := ff 2 L (I) : ( f ) < +1 for all > 0g :
Here, L (I) is called the modular space generated by ; and also E (I) is
called the space of the …nite elements of L (I) : Observe that if is N -quasi
semiconvex, then the space
ff 2 X (I) : [ f ] < +1 for some > 0g
coincides with L (I). The notions about modulars are introduced in [18]
and widely discussed in [4] (see also [15, 17]).
Now we recall the convergence methods in modular spaces.
Let ffn g be a function sequence whose terms belong to L (I) : Then,
ffn g is modularly convergent to a function f 2 L (I) i¤
(1.2) lim ( 0 (fn f )) = 0 for some 0 > 0:
n
Also, ffn g is F -norm convergent (or, strongly convergent) to f i¤
(1.3) lim ( (fn f )) = 0 for every > 0:
n
It is known from [17] that (1.2) and (1.3) are equivalent if and only if the
modular satis…es the 2 -condition, i.e. there exists a constant M > 0
such that [2f ] M [f ] for every f 2 X (I).
In this paper, we will need the following assumptions on a modular :
if (f ) (g) for jf j jgj ; then is called monotone,
if the characteristic function I of the interval I belongs to L (I) ;
is called …nite,
if is …nite and, for every " > 0, > 0; there exists a > 0 such
that ( B ) < " for any measurable subset B I with jBj < ,
then is called absolutely …nite,
if I 2 E (I) ; then is called strongly …nite
is called absolutely continuous provided that there exists an > 0
such that, for every f 2 X (I) with (f ) < +1, the following
condition holds: for every " > 0 there is > 0 such that ( f B ) <
" whenever B is any measurable subset of I with jBj < .
STATISTICAL APPROXIMATION BY POSITIVE LINEAR OPERATORS 5
Observe now that (see [3]) if a modular is monotone and …nite, then
we have C(I) L (I) : In a similar manner, if is monotone and strongly
…nite, then C(I) E (I). Some important relations between the above
properties may be found in [2, 4, 16, 18].
ei (x) = xi (i = 0; 1; 2; :::):
Now let f be any function belonging to L (I) such that f g 2 XT for every
g 2 C 1 (I) : Then, we have
(2.4) stA lim ( (Tn g g)) = 0 for every g 2 C(I) and every > 0:
n
To see this assume that g belongs to C (I) and is any positive number.
Then, by the uniform continuity of g on I and by the linearity and positivity
of the operators Tn ; we easily see that (see, for instance, [1, 14]), for a given
" > 0; there exists a > 0 such that
where c := maxfjaj ; jbjg: So, the last inequality gives, for any > 0; that
jTn (g; x) g(x)j " + K fjTn (e0 ; x) e0 (x)j + jTn (e1 ; x) e1 (x)j
+ jTn (e2 ; x) e2 (x)jg;
2M c2 4M c 2M
where K := max " + M + 2 ; 2 ; 2 : Now, applying the modular
in the both-sides of the above inequality, since is monotone, we have
( (Tn g g)) ( " + K jTn e0 e0 j + K jTn e1 e1 j + K jTn e2 e2 j) :
Hence, we may write that
( (Tn g g)) (4 ") + (4 K (Tn e0 e0 )) + (4 K (Tn e1 e1 ))
+ (4 K (Tn e2 e2 ))
Since is N -quasi semiconvex and strongly …nite, we have, assuming 0 <
" 1
( (Tn g g)) N " (4 N ) + (4 K (Tn e0 e0 ))
+ (4 K (Tn e1 e1 )) + (4 K (Tn e2 e2 )) .
For a given r > 0; choose an " 2 (0; 1] such that N " (4 N ) < r: Now de…ne
the following sets:
G : = fn : ( (Tn g g)) rg
r N " (4 N )
G ;1 : = n : (4 K (Tn e0 e0 ))
3
r N " (4 N )
G ;2 : = n : (4 K (Tn e1 e1 ))
3
r N " (4 N )
G ;3 : = n : (4 K (Tn e2 e2 )) :
3
Then, it is easy to see that G G ;1 [ G ;2 [ G ;3 . So, we can write, for
all j 2 N, that
X X X X
(2.5) ajn ajn + ajn + ajn .
n2G n2G ;1 n2G ;2 n2G ;3
STATISTICAL APPROXIMATION BY POSITIVE LINEAR OPERATORS 7
which proves our claim (2.4). Observe that (2.4) also holds for every g 2
C 1 (I) because of C 1 (I) C(I): Now let f 2 L (I) satisfying f g 2 XT
for every g 2 C 1 (I). Since jIj < 1 and is strongly …nite and absolutely
continuous, we can see that is also absolutely …nite on X(I) (see [2]). Using
these properties of the modular ; it is known from [4, 16] that the space
C 1 (I) is modularly dense in L (I) ; i.e., there exists a sequence fgk g
C 1 (I) such that
lim [3 0 (gk f )] = 0 for some 0 > 0:
k
This means that, for every " > 0; there is a positive number k0 = k0 (") so
that
(2.6) [3 0 (gk f )] < " for every k k0 :
On the other hand, by the linearity and positivity of the operators Tn , we
may write that
0 jTn (f ; x) f (x)j 0 jTn (f gk0 ; x)j + 0 jTn (gk0 ; x) gk0 (x)j
+ 0 jgk0 (x) f (x)j
holds for every x 2 I and n 2 N. Applying the modular in the last
inequality and using the monotonicity of ; we have
( 0 (Tn f f )) (3 0 Tn (f gk0 )) + (3 0 (Tn gk0 gk0 ))
(2.7)
+ (3 0 (gk0 f )) .
Then, it follows from (2.6) and (2.7) that
(2.8) ( 0 (Tn f f )) " + (3 0 Tn (f gk0 )) + (3 0 (Tn gk0 gk0 )) :
So, taking A-statistical limit superior as n ! 1 in the both-sides of (2.8)
and also using the facts that gk0 2 C 1 (I) and f gk0 2 XT ; we obtained
from (2.1) that
stA lim sup ( 0 (Tn f f )) " + P (3 0 (f gk0 ))
n
+stA lim sup (3 0 (Tn gk0 gk0 )) ;
n
which gives
(2.9)
stA lim sup ( 0 (Tn f f )) "(P +1)+stA lim sup (3 0 (Tn gk0 gk0 )) :
n n
By (2.4), since
stA lim (3 0 (Tn gk0 gk0 )) = 0;
n
we get
(2.10) stA lim sup (3 0 (Tn gk0 gk0 )) = 0:
n
8 SEVDA KARAKUŞ, KAMIL DEMIRCI AND OKTAY DUMAN
If the modular satis…es the 2 -condition, then one can get the following
result from Theorem 2.1 at once.
If one replaces the matrix A by the identity matrix, then the condition
(2.1) reduces to
3. Concluding Remarks
In this section, we display an example such that our Korovkin-type statis-
tical approximation results in modular spaces are stronger than the classical
ones in [3].
(3.2)
(k+1)=(n+1)
Z
n
X n k
Un (f ; x) := x (1 x)n k
(n + 1) f (t) dt for x 2 I:
k
k=0 k=(n+1)
Observe that the operators Un map the Orlicz space L' (I) into itself.
Moreover, the property (2.11) is satis…ed with the choice of XU := L' (I).
Then, by Corollary 2.3, we know that, for any function f 2 L' (I) such that
f g 2 XU for every g 2 C 1 (I), fUn f g is modularly convergent to f:
Now take A = C1 ; the Cesáro matrix of order one, and de…ne a sequence
fsn g by
0; if n is square
(3.3) sn =
1; otherwise.
In this case, we know that C1 -statistical convergence coincides with statis-
tical convergence, and its limit is denoted by st lim : Observe now that,
for the sequence fsn g given by (3.3),
st lim sn = 1:
n
However, the sequence fsn g does not convergent in the usual sense. Then,
using the operators Un , we de…ne the sequence of positive linear operators
V := fVn g on L' (I) as follows:
(3.4) Vn (f ; x) = sn Un (f ; x) for f 2 L' (I) , x 2 [0; 1] and n 2 N,
10 SEVDA KARAKUŞ, KAMIL DEMIRCI AND OKTAY DUMAN
where s = fsn g is the same as in (3.3). By Lemma 5.1 of [3], we get, for
every h 2 XV := L' (I); > 0 and for an absolute positive constant P; that
'
( Vn h) = ' ( sn Un h)
= sn ' ( Un h)
P sn ' ( h) :
' '
(3.5) st lim sup ( Vn f ) P ( f ):
n
Therefore, the condition (2.1) works for our operators Vn given by (3.4) with
the choice of XV = XU = L' (I):
We now claim that
'
(3.6) st lim ( (Vn ei ei )) = 0 for every > 0 and i = 0; 1; 2:
n
Vn (e0 ; x) = sn ;
nx 1
Vn (e1 ; x) = sn + ;
n + 1 2 (n + 1)
n (n 1) x2 2nx 1
Vn (e2 ; x) = sn 2 + 2 + :
(n + 1) (n + 1) 3 (n + 1)2
which implies
' '
( (Vn e0 e0 )) = ( (1 sn ))
Z1
= ' ( (1 sn )) dx
0
= ' ( (1 sn ))
= (1 sn )' ( )
which yields
' 15n + 4
(3.8) ( (Vn e2 e2 )) (1 sn )' (2 ) + sn ' 2
3(n + 1)2
Considering the continuity of '; it follows from (3.8) that
'
st lim ( (Vn e2 e2 )) = 0 for every > 0:
n
So, our claim (3.6) holds true for each i = 0; 1; 2 and for any > 0: Now,
from (3.5) and (3.6), we can say that our sequence V := fVn g de…ned by
(3.4) satisfy all assumptions of Theorem 2.1. Therefore, we conclude that
'
st lim ( 0 (Vn f f )) = 0 for some 0 >0
n
holds for any f 2 L' (I) such that f g 2 XV = L' (I) for every g 2 C 1 (I).
However, for the same f ’s and for any > 0; since
' '
( (Vn f f )) = ( (sn Un f f ))
' ( f ); if n is square
= ' ( (U f
n f )) ; otherwise,
the sequence f ' ( (Vn f f ))g has two subsequences with di¤erent limit
points. In other words, there is not any > 0 such that the sequence
f ' ( (Vn f f ))g goes to zero as n ! 1: This clearly shows that fVn f g
is not modularly convergent to f: So, Corollary 2.3 does not work for the
sequence V := fVn g:
With this example, we can immediately say that our Theorem 2.1 is a non-
trivial generalization of the classical Korovkin-type results in the modular
spaces.
Acknowledgement. The authors would like to thank the referee for care-
fully reading the manuscript.
References
[1] F. Altomare and M. Campiti, Korovkin Type Approximation Theory and Its Appli-
cation, Walter de Gruyter Publ., Berlin, 1994.
[2] C. Bardaro and I. Mantellini, Approximation properties in abstract modular spaces
for a class of general sampling-type operators, Appl. Anal. 85 (2006), 383-413.
[3] C. Bardaro and I. Mantellini, Korovkin’s theorem in modular spaces, Commentationes
Math. 47 (2007), 239-253.
[4] C. Bardaro, J. Musielak and G. Vinti, Nonlinear Integral Operators and Applications,
de Gruyter Series in Nonlinear Analysis and Appl. Vol. 9, Walter de Gruyter Publ.,
Berlin, 2003.
[5] J.S. Connor and J. Kline, On statistical limit points and the consistency of statistical
convergence, J. Math. Anal. Appl. 197 (1996), 392-399.
[6] K. Demirci, A-statistical core of a sequence, Demonstratio Math. 33 (2006), 343-353.
[7] H. Fast, Sur la convergence statistique, Colloq, Math. 2 (1951), 241-244.
[8] A.R. Freedman and J.J. Sember, Densities and summability, Paci…c. J. Math. 95
(1981), 293-305.
[9] J.A. Fridy, On statistical convergence, Analysis 5 (1985), 301-313.
STATISTICAL APPROXIMATION BY POSITIVE LINEAR OPERATORS 13
[10] J.A. Fridy, Statistical limit points, Proc. Amer. Math. Soc. 118 (1993), 1187-1192.
[11] J.A. Fridy and C. Orhan, Statistical limit superior and limit inferior, Proc. Amer.
Math. Soc. 125 (1997), 3625-3631.
[12] G.H. Hardy, Divergent Series, Oxford Univ. Press, London, 1949.
[13] E. Kolk, Matrix summability of statistically convergent sequences, Analysis 13 (1993),
77-83.
[14] P.P. Korovkin, Linear Operators and Approximation Theory, Hindustan Publ. Corp.,
Delhi, 1960.
[15] W.M. Kozlowski, Modular Function Spaces, Pure Appl. Math., Vol. 122, Marcel
Dekker, Inc., New York, 1988.
[16] I. Mantellini, Generalized sampling operators in modular spaces, Commentationes
Math., 38 (1998), 77-92.
[17] J. Musielak, Orlicz Spaces and Modular Spaces, Lecture Notes in Mathematics, Vol.
1034, Springer-Verlag, Berlin, 1983.
[18] J. Musielak, Nonlinear approximation in some modular function spaces I, Math.
Japon. 38 (1993), 83-90.
Sevda Karakuş
Sinop University, Faculty of Arts and Sciences,
Department of Mathematics, 57000 Sinop, Turkey
E-Mail: skarakus@omu.edu.tr
Kamil Demirci
Sinop University, Faculty of Arts and Sciences,
Department of Mathematics, 57000 Sinop, Turkey
E-Mail: kamild@omu.edu.tr