Computers and Mathematics with Applications 60 (2010) 2212–2218

Contents lists available at ScienceDirect

Computers and Mathematics with Applications

journal homepage: www.elsevier.com/locate/camwa

Equi-statistical σ -convergence of positive linear operators

Sevda Karakuş, Kamil Demirci ∗
Sinop University, Faculty of Sciences and Arts, Department of Mathematics, 57000 Sinop, Turkey

article info abstract

Article history: In this paper we introduce the notion of equi-statistical σ -convergence which is stronger
Received 21 April 2010 than the uniform convergence (in the ordinary sense), statistical uniform convergence
Accepted 3 August 2010 and statistical uniform σ -convergence. Then, we also give its use in the Korovkin-type
approximation theory. We also compute the rate of equi-statistical σ -convergence of a
Keywords: sequence of positive linear operators.
Statistical uniform σ -convergence
© 2010 Elsevier Ltd. All rights reserved.
Equi-statistical σ -convergence
Statistical pointwise σ -convergence
Positive linear operator
Korovkin-type approximation theorem

1. Introduction

The concept of convergence of a sequence of real numbers has been extended to statistical convergence by Fast [1]. Let
K be the subset of positive integers. Then the natural density of K is given by
1
δ (K ) = lim |Kn |
n→∞ n
if it exists, where Kn = {k ∈ K : k ≤ n} and the vertical bars denote the cardinality of the set Kn . A sequence x = {xk } of
numbers is statistically convergent to L provided that, for every ε > 0, δ {k : |xk − L| ≥ ε} = 0 holds. In this case we write
st − lim x = L.
Recently various kinds of statistical convergences which are stronger than the statistical convergence have been
introduced by Mursaleen and Edely [2]. We first recall this convergence method.
Let σ be a one-to-one mapping from the set N into itself. A continuous linear functional ϕ defined on the space l∞ of all
bounded sequences is called an invariant mean (or σ -mean) [3] if and only if
(i) ϕ(x) ≥ 0 when the sequence x = {xk } has xk ≥ 0 for all k,
(ii) ϕ(e) = 1, where e = (1, 1, . . .),
(iii) ϕ(x) = ϕ((xσ (n) )) for all x ∈ l∞ .
Thus, the σ -mean extends the limit functional on c of all convergent sequences in the sense that ϕ(x) = lim x for all x ∈ c [4].
Consequently, c ⊂ Vσ where Vσ is the set of bounded sequences all of whose σ -means are equal. It is known [5] that
 
Vσ = x ∈ l∞ : lim tpm (x) = L uniformly in m, L = σ − lim x
p→∞

where
xm + xσ (m) + xσ 2 (m) + · · · + xσ p (m)
tpm (x) := .
p+1

∗ Corresponding author.
E-mail addresses: [email protected] (S. Karakuş), [email protected], [email protected] (K. Demirci).

0898-1221/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.camwa.2010.08.010
 S. Karakuş, K. Demirci / Computers and Mathematics with Applications 60 (2010) 2212–2218 2213

We say that a bounded sequence x = {xk } is σ -convergent if and only if x ∈ Vσ . Let

 
s
Vσ = x ∈ l∞ : st − lim tpm (x) = L uniformly in m, L = σ − lim x .
p→∞

A sequence x = {xk } is said to be statistically σ -convergent to L if and only if x ∈ Vσs (see, for details, [2]). In this case we
write δ(σ ) − lim xk = L. That is,
1 
 p ≤ n : tpm (x) − L ≥ ε  = 0, uniformly in m.
  
lim
n→∞ n
Note that if x is statistically convergent then x is statistically σ -convergent. However, its converse is not always true. Such
an example was given in [2].
Let f and fn belong to C (X ), which is the space of all continuous real valued functions on a compact subset X of the real
numbers. The usual supremum norm on the spaces C (X ) is given by
‖f ‖C (X ) := sup |f (x)| for f ∈ C (X ) .
x∈X

In [6] a kind of convergence for a sequence lying between pointwise and uniform statistical convergence is presented.
We first recall these convergence methods.

Definition 1 ([6]). {fn } is said to be statistically pointwise convergent to f on X if st − limn→∞ fn (x) = f (x) for each x ∈ X ,
i.e., for every ε > 0 and for each x ∈ X ,
|{k ≤ n : |fk (x) − f (x)| ≥ ε}|
lim = 0.
n→∞ n
In this case, we denote this limit by fn → f (stat ) on X .

Definition 2 ([6]). {fn } is said to be equi-statistically convergent to f on X if for every ε > 0,

|{k ≤ n : |fk (x) − f (x)| ≥ ε}|
 
lim sup = 0.
n→∞ x∈X n
In this case, we denote this limit by fn → f (equi-stat) on X .

Definition 3 ([6]). {fn } is said to be statistically uniformly convergent to f on X if st − limn→∞ ‖fn − f ‖C (X ) = 0, or

 k ≤ n : ‖f k − f ‖C ( X ) ≥ ε 
 
lim = 0.
n→∞ n
In this case, we denote this limit by fn ⇒ f (stat ) on X .

Now we give the following definitions:

f (x) := {fn (x)} is said to be statistically pointwise σ -convergent to f on X if

Definition 4. 

f (x) = f (x)
 
st − lim tpm  uniformly in m
p→∞

for each x ∈ X , i.e. for every ε > 0 and for each x ∈ X

f (x) − f (x) ≥ ε 
     
 p ≤ n : tpm 
lim = 0 uniformly in m
n→∞ n
where
fm (x) + fσ (m) (x) + fσ 2 (m) (x) + · · · + fσ p (m) (x)
f (x) := .
 
tpm 
p+1
Then, this is denoted by fn → f (stat (σ )) on X .

f (x) := {fn (x)} is said to be equi-statistically σ -convergent to f on X if for every ε > 0

Definition 5. 
  
f (x) − f (x) ≥ ε 
    
 p ≤ n : tpm 
lim sup = 0 uniformly in m.
n→∞ x∈X n

This limit is denoted by fn → f (equi-stat (σ )) on X .

2214 S. Karakuş, K. Demirci / Computers and Mathematics with Applications 60 (2010) 2212–2218

f (x) := {fn (x)} is said to be statistically uniformly σ -convergent to f on X if for every ε > 0
Definition 6. 
 
f (x) − f (x)C (X ) ≥ ε 
   
 p ≤ n : tpm 
 
lim = 0 uniformly in m.
n→∞ n
This limit is denoted by fn ⇒ f (stat (σ )) on X .

Using the above definitions, the next result follows immediately.

Lemma 7. fn ⇒ f on X (in the ordinary sense) implies fn ⇒ f (stat ) on X , which also implies fn ⇒ f (stat (σ )) on X . Furthermore
fn ⇒ f (stat (σ )) on X implies fn → f (equi-stat (σ )) on X and fn → f (equi-stat (σ )) on X implies fn → f (stat (σ )) on X . Also,
fn → f on X (in the ordinary sense) implies fn → f (stat ) on X , which also implies fn → f (stat (σ )) on X and fn → f (equi-stat )
on X implies fn → f (equi-stat (σ )) on X .

However one can construct an example which guarantees that the converses of Lemma 7 are not always true. Such an
example is the following:

Example 8. Consider the case σ (n) = n + 1. Let g (x) = 0, x ∈ [0, 1], and define continuous functions gk : [0, 1] → R,
k ∈ N, by the formula
  2 [ ]
2(k+1) 1 1 1 1
2 x− k , if x ∈ k , k−1 − k+1 ,




 2 2 2 2
gk (x) =
2
(1)
 [ ]
2(k+1) 1 1 1 1
2 x − , if x ∈ − ,
2k−1 2k−1 2k+1 2k−1



0, otherwise.


Then observe that gk → g = 0 (equi-stat (σ )) on [0, 1]; however {gk } is not statistically uniformly σ -convergent to the
function g = 0 on the interval [0, 1]. Also, {gk } is not uniformly convergent (in the ordinary sense) and statistically uniformly
convergent to the function g = 0 on [0, 1] .

For a sequence {Ln } of positive linear operators on C (X ), Korovkin [7] first introduced the sufficient conditions for the
uniform convergence of Ln (f ) to a function f by using the test function ei defined by ei (x) = xi (i = 0, 1, 2). Later many
researchers investigated these conditions for various operators defined on different spaces. Furthermore, in recent years,
with the help of the concept of uniform statistical convergence, which is a regular (non-matrix) summability transformation,
various statistical approximation results have been proved [8–14]. Then, it was demonstrated that those results are more
powerful than the classical Korovkin theorem.
In this paper, we obtain the Korovkin-type approximation theorem by means of the concept of equi-statistical
σ -convergence in Definition 5.

2. A Korovkin-type approximation theorem

In this section, using equi-statistical σ -convergence, we prove the following Korovkin-type approximation theorem.

Theorem 9. Let X be a compact subset of the real numbers and let L := {Ln } be a sequence of positive linear operators acting
from C (X ) into itself. Then for all f ∈ C (X ),

Ln (f ) → f (equi-stat (σ )) on X (2)

if and only if

Ln (ei ) → ei (equi-stat (σ )) on X (3)

with ei (x) = x , i = 0, 1, 2.
i

Proof. Since each ei ∈ C (X ), i = 0, 1, 2, the implication (2) ⇒ (3) is obvious.

Now suppose that (3) holds. Let f ∈ C (X ) and x ∈ X be fixed. By the continuity of the point x, we may write that for
every ε > 0, there exists a number δ > 0 such that |f (y) − f (x)| < ε for all y ∈ X satisfying |y − x| < δ . Since

|f (y) − f (x)| = |f (y) − f (x)| χXδ (y) + |f (y) − f (x)| χX \Xδ (y)
where Xδ = [x − δ, x + δ ] ∩ X and χA denotes the characteristic function of the set A, then we have

(y − x)2
|f (y) − f (x)| ≤ ε + 2M (4)
δ2
 S. Karakuş, K. Demirci / Computers and Mathematics with Applications 60 (2010) 2212–2218 2215

for all y ∈ X , where M := ‖f ‖C (X ) . Using the positivity and monotonicity of Ln , we obtain from (4)

tpm (L (f ; x)) − f (x) =  Lm (f ; x) + Lσ (m) (f ; x) + · · · + Lσ (m) (f ; x) − f (x)

 
   p 
 p+1 
 Lm (f ; x) + Lσ (m) (f ; x) + · · · + Lσ p (m) (f ; x) Lm (e0 ; x) + Lσ (m) (e0 ; x) + · · · + Lσ p (m) (e0 ; x)

=  − f ( x)
p+1 p+1
Lm (e0 ; x) + Lσ (m) (e0 ; x) + · · · + Lσ p (m) (e0 ; x)

f (x) − f (x)

+
p+1
Lm (|f (y) − f (x)| ; x) + Lσ (m) (|f (y) − f (x)| ; x) + · · · + Lσ p (m) (|f (y) − f (x)| ; x)
≤
p+1
+ |f (x)| tpm (L (e0 ; x)) − e0 (x)
 
 
2M
≤ ε + ε + M + 2 ‖e2 ‖C (X ) tpm (L (e0 ; x)) − e0 (x)
 
δ
4M  2M 
+ 2 ‖e1 ‖C (X ) tpm (L (e1 ; x)) − e1 (x) + 2 tpm (L (e2 ; x)) − e2 (x) .
 
δ δ
Then, we get

tpm (L (f ; x)) − f (x) ≤ ε + K tpm (L (e0 ; x)) − e0 (x) + tpm (L (e1 ; x)) − e1 (x) + tpm (L (e2 ; x)) − e2 (x)
       
(5)

where K := ε + M + 2M ‖e2 ‖C (X ) + 2 ‖e1 ‖C (X ) + 1 . Now, for a given r > 0, choose ε > 0 such that ε < r. Then, it follows
 
δ2
from (5) that

 p ≤ n : tpm (L (f ; x)) − f (x) ≥ r  ≤  p ≤ n : tpm (L (e0 ; x)) − e0 (x) ≥ r − ε 

 
       
 3K 
 r −ε 
 
+  p ≤ n : tpm (L (e1 ; x)) − e1 (x) ≥
 
   
3K 
 r −ε 
 
+  p ≤ n : tpm (L (e2 ; x)) − e2 (x) ≥
 

 3K
which gives
 p ≤ n : tpm (L (f ; x)) − f (x) ≥ r   p ≤ n : tpm (L (e0 ; x)) − e0 (x) ≥
       r −ε


3K
sup ≤ sup
x∈X n x∈X n
 p ≤ n : tpm (L (e1 ; x)) − e1 (x) ≥
   r −ε


3K
+ sup
x∈X n
 p ≤ n : tpm (L (e2 ; x)) − e2 (x) ≥
   r −ε


+ sup 3K
. (6)
x∈X n
Then using the hypothesis (3) and considering Definition 5 in (6),
  
 p ≤ n : tpm (L (f ; x)) − f (x) ≥ r 
  
lim sup = 0 uniformly in m
n→∞ x∈X n

for every r > 0. The proof is complete. 

Remark 10. Now we show that our result Theorem 9 is stronger than the classical Korovkin theorem and the statistical
Korovkin theorem.
Let X = [0, 1] and consider the following Bernstein–Kantorovich operators:
(k+1∫
)/(n+1)
n  
− n
Un (f ; x) = (n + 1) x (1 − x)
k n −k
f (t ) dt
k=0
k
k/(n+1)

where x ∈ [0, 1], f ∈ C [0, 1] and n ∈ N. Using these polynomials, we introduce the following positive linear operators on
C [0, 1]:

Tn (f ; x) = (1 + gn (x)) Un (f ; x) , x ∈ [0, 1] , f ∈ C [0, 1] (7)

2216 S. Karakuş, K. Demirci / Computers and Mathematics with Applications 60 (2010) 2212–2218

where gn (x) is given by (1). Then observe that

Tn (e0 ; x) = (1 + gn (x)) e0 (x)
 
n 1
Tn (e1 ; x) = (1 + gn (x)) e1 (x) +
n+1 2 (n + 1)
n (n − 1)
[ ]
2nx 1
Tn (e2 ; x) = (1 + gn (x)) e2 (x) + +
(n + 1)2 (n + 1)2 3 (n + 1)2
where ei (x) = xi , i = 0, 1, 2. Consider the case σ (n) = n + 1. Since gn → g = 0 (equi-stat (σ )) on [0, 1], we conclude that
Tn (ei ) → ei (equi-stat (σ )) on [0, 1] for each i = 0, 1, 2.
So the {Tn } satisfy all hypotheses of Theorem 9 and we immediately see that
Tn (f ) → f (equi-stat (σ )) on [0, 1] for all f ∈ C [0, 1] .
However, since {gn } is not uniformly convergent (in the ordinary sense) to the function g = 0 on [0, 1], we can say that
Tn (ei ) is not uniformly convergent to ei for i = 0, 1, 2. Also since {gn } is not statistically uniformly convergent to the function
g = 0 on [0, 1], we can say that Tn (ei ) is not statistically uniformly convergent to ei on [0, 1] for i = 0, 1, 2. So the classical
Korovkin theorem and statistical Korovin theorem do not work for our operators defined by (7). Therefore, this application
clearly shows that our Theorem 9 is a non-trivial generalization of the classical and the statistical cases of the Korovkin-type
results introduced in [7,13], respectively.

3. Rate of equi-statistical σ -convergence

In this section, we study the rates of equi-statistical σ -convergence of positive linear operators defined on C (X ) with the
help of the modulus of continuity.

f (x) := {fn (x)} is equi-statistically σ -convergent to a function f with the rate of β ∈ (0, 1) if for
Definition 11. A sequence 
every ε > 0,
  
f (x) − f (x) ≥ ε 
    
 p ≤ n : tpm 
lim sup = 0 uniformly in m
n→∞ x∈X n1−β

where
fm (x) + fσ (m) (x) + fσ 2 (m) (x) + · · · + fσ p (m) (x)
f (x) := .
 
tpm 
p+1
In this case, it is denoted by
fn − f = o(n−β ) (equi-stat (σ )) on X .

Using this definition, we obtain the following auxiliary result.

Lemma 12. Let  f (x) := {fn (x)} and g (x) := {gn (x)} be function sequences belonging to C (X ). Assume that fn − f =
o(n−β1 ) (equi-stat (σ )) on X and gn − g = o(n−β2 ) (equi-stat (σ )) on X . Let β := min {β1 , β2 }. Then we have
(i) (fn − f ) ∓ (gn − g ) = o(n−β ) (equi-stat (σ )) on X ,
(ii) λ(fn − f ) = o(n−β1 ) (equi-stat (σ )) on X , for any real number λ.
Proof. (i) Assume that fn − f = o(n−β1 ) (equi-stat (σ )) on X and gn − g = o(n−β2 ) (equi-stat (σ )) on X . Then, since
β := min {β1 , β2 } for ε > 0, observe that
g (x)) − g (x)  ≥ ε 
f (x) − f (x) ∓ tpm (
       
 p ≤ n :  tpm 
n 1−β

f (x) − f (x) ≥ 2ε  +  p ≤ n : tpm (

g (x)) − g (x) ≥ 2ε 
         
 p ≤ n : tpm 
≤
n1−β
f (x) − f (x) ≥ 2ε  g (x)) − g (x) ≥ ε 
 p ≤ n : tpm (
         
 p ≤ n : tpm 
≤ + 2
. (8)
n1−β1 n1−β2
Now, by taking the limit as n → ∞ in (8) and using the hypotheses, we conclude that
  
g (x)) − g (x)  ≥ ε 
f (x) − f (x) ∓ tpm (
      
 p ≤ n :  tpm 
lim sup = 0, uniformly in m,
n→∞ x∈X n1−β

which completes the proof of (i). Since the proof of (ii) is similar, we omit it. 
 S. Karakuş, K. Demirci / Computers and Mathematics with Applications 60 (2010) 2212–2218 2217

Now we recall that the modulus of continuity of a function f ∈ C (X ) is defined by

ω (f ; δ) = sup |f (y) − f (x)| (δ > 0) .
|y−x|≤δ,x,y∈X

It is also well know that, for any λ > 0 and for all f ∈ C (X ),
ω (f ; λδ) ≤ (1 + [λ]) ω (f ; δ)
where [λ] is defined to be the greatest integer less than or equal to λ.
Then we have the following result.

Theorem 13. Let L := {Ln } be a sequence of positive linear operators acting from C (X ) into itself. Assume that the following
conditions hold:

Ln (e0 ) − e0 = o(n−β1 ) (equi-stat (σ )) on X (9)

and
ω f ; γpm = o(n−β2 ) (equi-stat (σ )) on X
 
(10)

where γpm := tpm (L (ϕ; x)) with ϕ (y) = (y − x)2 . Then we have for all f ∈ C (X )


Ln (f ) − f = o(n−β ) (equi-stat (σ )) on X
where β := min {β1 , β2 } .
Proof. Let f ∈ C (X ) and x ∈ X be fixed. Using the properties of ω and the positivity and monotonicity of Ln , we get, for any
δ > 0, that

tpm (L (f ; x)) − f (x) =  Lm (f ; x) + Lσ (m) (f ; x) + · · · + Lσ (m) (f ; x) − f (x)

 
   p 
 p+1 
Lm (|f (y) − f (x)| ; x) + Lσ (m) (|f (y) − f (x)| ; x) + · · · + Lσ (m) (|f (y) − f (x)| ; x)
p
≤
p+1
+ |f (x)| tpm (L (e0 ; x)) − e0 (x)
 
           
Lm ω f ; δ δ ; x + Lσ (m) ω f ; δ |y−δ x| ; x + · · · + Lσ p (m) ω f ; δ |y−δ x| ; x
|y−x|

≤
p+1
+ |f (x)| tpm (L (e0 ; x)) − e0 (x)
 
 
           
|y−x|
Lm 1 + δ
; x + Lσ (m) 1 + |y−δ x| ; x + · · · + Lσ p (m) 1 + |y−δ x| ; x
≤ ω (f ; δ)
p+1
+ |f (x)| tpm (L (e0 ; x)) − e0 (x)
 
     
(y−x)2 (y−x)2 (y−x)2
Lm 1 + δ 2 ; x + Lσ (m) 1 + δ 2 ; x + · · · + Lσ p (m) 1 + δ 2 ; x
≤ ω (f ; δ)
p+1
+ |f (x)| tpm (L (e0 ; x)) − e0 (x)
 

≤ ω (f ; δ) + tpm (L (e0 ; x)) − e0 (x) ω (f ; δ)

 

Lm (ϕ; x) + Lσ (m) (ϕ; x) + · · · + Lσ p (m) (ϕ; x) ω (f ; δ)

+ |f (x)| tpm (L (e0 ; x)) − e0 (x)
 
+
p+1 δ 2

ω (f ; δ)
≤ ω (f ; δ) + tpm (L (e0 ; x)) − e0 (x) ω (f ; δ) + tpm (L(ϕ; x)) + |f (x)| tpm (L (e0 ; x)) − e0 (x) .
   
δ2
Then we have
tpm (L (f ; x)) − f (x) ≤ 2ω (f ; δ) + tpm (L (e0 ; x)) − e0 (x) ω (f ; δ) + M tpm (L (e0 ; x)) − e0 (x)
     
(11)

where δ := γpm := tpm (L (ϕ; x)) and M := ‖f ‖C (X ) . Hence, given ε > 0, it follows from (11) and Lemma 12 that


 p ≤ n : tpm (L (f ; x)) − f (x) ≥ ε 

   

n 1−β

 p ≤ n : ω (f ; δ) ≥ ε   p ≤ n : tpm (L (e0 ; x)) − e0 (x) ≥ ε 

      
6 3
≤ +
n1−β2 n1−β1
2218 S. Karakuş, K. Demirci / Computers and Mathematics with Applications 60 (2010) 2212–2218

 p ≤ n : ω (f ; δ) ≥ ε   p ≤ n : tpm (L (e0 ; x)) − e0 (x) ≥ ε

      

3 3M
+ + (12)
n1−β2 n1−β1
where β := min {β1 , β2 }. Letting n → ∞ in (12), we conclude from (9) and (10) that
  
 p ≤ n : tpm (L (f ; x)) − f (x) ≥ ε 
  
lim sup = 0, uniformly in m,
n→∞ x∈X n1−β

which means
Ln (f ) − f = o(n−β ) (equi-stat (σ )) on X . 

References

[1] H. Fast, Sur la convergence statistique, Colloq. Math. 2 (1951) 241–244.

[2] M. Mursaleen, O.H.H. Edely, On the invariant mean and statistical convergence, Appl. Math. Lett. 22 (2009) 1700–1704.
[3] R.A. Raimi, Invariant means and invariant matrix methods of summability, Duke Math. J. 30 (1963) 81–94.
[4] M. Mursaleen, On some new invariant matrix methods of summability, Q. J. Math. 34 (1983) 77–86.
[5] P. Schaefer, Infinite matrices and invariant means, Proc. Amer. Math. Soc. 36 (1972) 104–110.
[6] M. Balcerzak, K. Dems, A. Komisarski, Statistical convergence and ideal convergence for sequences of functions, J. Math. Anal. Appl. 328 (2007) 715–729.
[7] P.P. Korovkin, Linear Operators and Approximation Theory, Hindustan Publ. Co., Delhi, 1960.
[8] G.A. Anastassiou, O. Duman, A Baskakov type generalization of statistical Korovkin theory, J. Math. Anal. Appl. 340 (2008) 476–486.
[9] G.A. Anastassiou, O. Duman, Statistical fuzzy approximation by fuzzy positive linear operators, Comput. Math. Appl. 55 (2008) 573–580.
[10] O. Duman, M.K. Khan, C. Orhan, A-statistical convergence of approximating operators, Math. Inequal. Appl. 4 (2003) 689–699.
[11] O. Duman, E. Erkuş, V. Gupta, Statistical rates on the multivariate approximation theory, Math. Comput. Modelling 44 (2006) 763–770.
[12] E. Erkuş, O. Duman, H.M. Srivastava, Statistical approximation of certain positive linear operators constructed by means of the Chan–Chyan–Srivastava
polynomials, Appl. Math. Comput. 182 (2006) 213–222.
[13] A.D. Gadjiev, C. Orhan, Some approximation theorems via statistical convergence, Rocky Mountain J. Math. 32 (2002) 129–138.
[14] S. Karakuş, K. Demirci, O. Duman, Equi-statistical convergence of positive linear operators, J. Math. Anal. Appl. 339 (2008) 1065–1072.