Bol. Soc. Mat. Mex.

DOI 10.1007/s40590-016-0139-1

ORIGINAL ARTICLE

( p, q)-Beta functions and applications in approximation

Gradimir V. Milovanović1,2 · Vijay Gupta3 ·

Neha Malik3

Received: 9 January 2016 / Accepted: 8 June 2016

© Sociedad Matemática Mexicana 2016

Abstract In the present paper, we consider ( p, q)-analogue of the beta operators and
using it, we propose the integral modification of the generalized Bernstein polyno-
mials. We estimate some direct results on local and global approximation. Also, we
illustrate some graphs for the convergence of ( p, q)-Bernstein–Durrmeyer operators
for different values of the parameters p and q using Mathematica package.

Keywords ( p, q)-Beta function · ( p, q)-Gamma function · Bernstein polynomial ·

Durrmeyer variant · Direct results

Mathematics Subject Classification Primary 33B15; Secondary 41A25

This paper was supported by the Serbian Ministry of Education, Science and Technological Development
(No. #OI 174015).

B Vijay Gupta
[email protected]
Gradimir V. Milovanović
[email protected]
Neha Malik
[email protected]

1 Serbian Academy of Sciences and Arts, Kneza Mihaila 35, 11000 Beograd, Serbia
2 State University of Novi Pazar, Novi Pazar, Serbia
3 Department of Mathematics, Netaji Subhas Institute of Technology, Sector 3 Dwarka, New Delhi
110078, India
 G. V. Milovanović et al.

1 Introduction

In the past decade, the generalizations of several operators to quantum variant have
been introduced and their approximation behavior has been discussed (see for instance
[1,5,8,10,12] etc). The further generalization of quantum calculus is the post-quantum
calculus, denoted by ( p, q)-calculus. Recently, some researchers started working in
this direction (cf. [2,6,13,16]). Some basic definitions and theorems, which are men-
tioned below, may be found in these papers and references therein.

pn − q n
[n] p,q := p n−1 + p n−2 q + p n−3 q 2 + · · · + pq n−2 + q n−1 = .
p−q
n
The ( p, q)-factorial is given by [n] p,q ! = k=1 [k] p,q , n ≥ 1, [0] p,q ! = 1. The
( p, q)-binomial coefficient satisfies
 
n [n] p,q !
= , 0 ≤ k ≤ n.
k p,q [n − k] p,q ! [k] p,q !

The ( p, q)-power basis is defined as

(x  a)np,q = (x − a)( px − qa)( p 2 x − q 2 a) · · · ( p n−1 x − q n−1 a).

The ( p, q)-derivative of the function f is defined as

f ( px) − f (q x)
D p,q f (x) = , x = 0.
( p − q) x

As a special case when p = 1, the ( p, q)-derivative reduces to the q-derivative.

The ( p, q)-derivative fulfils the following product rules:

D p,q ( f (x)g(x)) = f ( px)D p,q g(x) + g(q x)D p,q f (x),

D p,q ( f (x)g(x)) = g( px)D p,q f (x) + f (q x)D p,q g(x).

Obviously D p,q (x  a)0p,q = 0 and for n ≥ 1, we have

D p,q (x  a)np,q = [n] p,q ( px  a)n−1

p,q ,
D p,q (a  x)np,q = − [n] p,q (a  q x)n−1
p,q .

Let f be an arbitrary function and a ∈ R. The ( p, q)-integral of f (x) on [0, a]

(see [15]) is defined as

a ∞
  k 
pk p p
f (x) d p,q x = (q − p) a f a if < 1
q k+1 q k+1 q
0 k=0
( p, q)-Beta functions and applications in approximation

and
a ∞
  k 
qk q q
f (x) d p,q x = ( p − q) a k+1
f a if < 1.
p p k+1 p
0 k=0

The formula of ( p, q)-integration by part is given by

b
f ( px) D p,q g (x) d p,q x = f (b) g (b) − f (a) g (a)
a b
− g (q x) D p,q f (x) d p,q x. (1.1)
a

Very recently Gupta and Aral [7] proposed the ( p, q) analogue of usual Durrmeyer
operators by considering some other form of ( p, q) Beta functions, which is not
commutative. In the present article, we define different ( p, q)-variant of Beta function
of first kind and find an identity relation with ( p, q)-gamma functions. It is observed
that ( p, q)-beta functions may satisfy the commutative property, by multiplying the
appropriate factor while choosing ( p, q) Beta function. As far as the approximation is
concerned, order is important in post-quantum calculus. We propose a generalization
of Durrmeyer type operators and establish some direct results.

2 ( p, q)-Gamma and ( p, q)-beta functions

Definition 2.1 ([14]) Let n be a nonnegative integer; we define the ( p, q)-gamma

function as
( p  q)np,q
 p,q (n + 1) = = [n] p,q !, 0 < q < p.
( p − q)n

Definition 2.2 Let m, n ∈ N; we define ( p, q)-beta integral as

1
B p,q (m, n) = x m−1 (1  q x)n−1
p,q d p,q x. (2.1)
0

Theorem 2.3 The ( p, q)-gamma and ( p, q)-beta functions fulfil the following funda-
mental relation:
 p,q (m) p,q (n)
B p,q (m, n) = p (n−1)(2m+n−2)/2 , (2.2)
 p,q (m + n)

where m, n ∈ N.
 G. V. Milovanović et al.

Proof For any m, n ∈ N since

1
B p,q (m, n) = x m−1 (1  q x)n−1
p,q d p,q x,
0

using (1.1) for f (x) = (x/ p)m−1 and g(x) = −(1  x)np,q /[n] p,q with the equality
D p,q (1  x)n = −[n] p,q (1  q x)n−1 we have

1
[m − 1] p,q
B p,q (m, n) = m−1 x m−2 (1  q x)np,q d p,q x
p [n] p,q
0
[m − 1] p,q
= m−1 B p,q (m − 1, n + 1) . (2.3)
p [n] p,q

Also we can write for positive integer n

1
B p,q (m, n + 1) = x m−1 (1  q x)np,q d p,q x
0
1

= x m−1 (1  q x)n−1
p,q p n−1
− q n
x d p,q x
0
1 1
= p n−1
x m−1
(1  q x)n−1
p,q d p,q x −q n
x m (1  q x)n−1
p,q d p,q x
0 0
= p n−1
B p,q (m, n) − q B p,q (m + 1, n) .
n

Using (2.3), we have

[m] p,q
B p,q (m, n + 1) = p n−1 B p,q (m, n) − q n B p,q (m, n + 1) ,
p m [n] p,q

which implies that

pn − q n
B p,q (m, n + 1) = p n+m−1 B p,q (m, n) .
p n+m − q n+m

Further, by definition of ( p, q) integration

1
1
B p,q (m, 1) = x m−1 d p,q x = .
[m] p,q
0
( p, q)-Beta functions and applications in approximation

We immediately have

p n−1 − q n−1
B p,q (m, n) = p n+m−2 B p,q (m, n − 1)
p n+m−1 − q n+m−1
p n−1 −q n−1 p n−2 − q n−2
= p n+m−2 n+m−1 p n+m−3
B p,q (m, n−2)
p −q n+m−1 p n+m−2 −q n+m−2
p n−1 − q n−1 p n−2 − q n−2
= p n+m−2 n+m−1 p n+m−3 n+m−2 ···
p −q n+m−1 p − q n+m−2
p−q
× p m m+1 B p,q (m, 1)
p − q m+1
p n−1 − q n−1 p n−2 − q n−2
= p n+m−2 n+m−1 p n+m−3
···
p − q n+m−1 p n+m−2 − q n+m−2
p−q 1
× p m m+1
p − q m+1 [m] p,q
( p  q)n−1
p,q
= p m+(m+1)+···+(m+n−2)
( p − q)
( p m  q m )np,q
( p  q)n−1
= p (n−1)(2m+n−2)/2
p,q
( p − q) . (2.4)
( p m  q m )np,q

p,q = (a  b) p,q (ap  bq ) p,q ; thus (2.4) leads

Following [14], we have (a  b)n+m n n n m

to

(n−1)(2m+n−2) ( p  q)n−1
p,q
B p,q (m, n) = p 2 ( p − q)
( p m  q m )np,q
(n−1)(2m+n−2) ( p  q)n−1
p,q ( p  q)m−1
p,q
= p 2 ·
( p − q)n−1 ( p − q)m−1
( p − q)m−1 ( p − q)n−1
· ( p − q)
p,q ( p  q ) p,q
( p  q)m−1 m m n

(n−1)(2m+n−2) ( p  q)n−1
p,q ( p  q)m−1
p,q ( p − q)m+n−1
= p 2 · ·
( p − q)n−1 ( p − q)m−1 ( p  q)m+n−1
p,q
(n−1)(2m+n−2)  p,q (m)  p,q (n)
= p 2 .
 p,q (m + n)

This completes the proof of the theorem.



Remark 2.4 The following observations have been made for ( p, q) Beta functions:

• For m, n ∈ N, we have

B p,q (m, n + 1) = p n−1 B p,q (m, n) − q n B p,q (m + 1, n).

 G. V. Milovanović et al.

• The ( p, q)-beta integral defined by (2.1) is not commutative. In order to make

commutative, we may consider the following form:
1

B p,q (m, n) = p m(m−1)/2 x m−1 (1  q x)n−1
p,q d p,q x.
0

For this form, ( p, q)-gamma and ( p, q)-beta functions fulfill the following fun-
damental relation:
 p,q (m)  p,q (n)

B p,q (m, n) = p (2mn+m +n −3m−3n+2)/2
2 2
, (2.5)
 p,q (m + n)

where m, n ∈ N. Obviously for form (2.5), we get 

B p,q (m, n) = 
B p,q (n, m) .

3 ( p, q) Bernstein-type operators and moments

For n ∈ N and k ≥ 0, we have the following identity, which can be easily verified
using the principle of mathematical induction:
n  
n
p k(k−1)/2 x k (1  x)n−k
p,q = p
n(n−1)/2
. (3.1)
k p,q
k=0

Using the above identity, we consider the ( p, q)-analogue of Bernstein operators for
x ∈ [0, 1] and 0 < q < p ≤ 1 as


n
p,q p n−k [k] p,q
Bn, p,q ( f, x) = bn,k (1, x) f , (3.2)
[n] p,q
k=0

where the ( p, q)-Bernstein basis is defined as

 
n
p [k(k−1)−n(n−1)]/2 x k (1  x)n−k
p,q
bn,k (1, x) = p,q .
k p,q

Remark 3.1 Other form of the ( p, q)-analogue of Bernstein polynomials has been
recently considered by Mursaleen et al. [13].
Remark 3.2 Using the identity (3.1) and the following recurrence relation (for m ≥
p,q
1, Un,m (x) = Bn, p,q (em , x) = Bn, p,q (t m , x)):
p,q p,q p,q
[n] p,q Un,m+1 ( px) = p n x(1 − px)D p,q [Un,m (x)] + [n] p,q pxUn,m ( px),

the ( p, q)-Bernstein polynomial satisfies

p n−1 x(1 − x)
Bn, p,q (e0 , x) = 1, Bn, p,q (e1 , x) = x, Bn, p,q (e2 , x) = x 2 + ,
[n] p,q

where ei = t i , i = 0, 1, 2.
( p, q)-Beta functions and applications in approximation

Recently, Gupta and Wang (see [9]) discussed the q-variant of certain Bernstein–
Durrmeyer type operators. We now extend these studies and propose the following
( p, q)-Bernstein–Durrmeyer operators based on ( p, q)-beta function.
For x ∈ [0, 1] and 0 < q < p ≤ 1, the ( p, q)-analogue of Bernstein–Durrmeyer
operators is defined as


n
p −(n−k+1)(n+k)/2 bn,k (1, x)
p,q p,q
Dn ( f, x) = [n + 1] p,q
k=1
1
p,q p,q
× bn,k−1 (t) f (t)d p,q t + bn,0 (1, x) f (0), (3.3)
0

p,q
where bn,k (1, x) is defined by (3.2) and
 
p,q n
bn,k (t) = t k (1  qt)n−k
p,q .
k p,q

It may be remarked here that for p = 1, these operators will not reduce to the
q-Durrmeyer operators, but for p = q = 1, these will reduce to the Durrmeyer
operators.

Lemma 3.3 Let em = t m , m ∈ N ∪ {0}; then for x ∈ [0, 1] and 0 < q < p ≤ 1, we
have

p,q p,q p[n] p,q x

Dn (e0 , x) = 1, Dn (e1 , x) = ,
[n + 2] p,q
p,q ( p + q) p n+1 [n] p,q x ([n] p,q − p n−1 ) p 2 q[n] p,q x 2
Dn (e2 , x) = + .
[n + 2] p,q [n + 3] p,q [n + 2] p,q [n + 3] p,q

Proof Using (2.2) and (2.1) and Remark 3.2, we have


n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q p,q
Dn (e0 , x) = [n + 1] p,q
k=1
1  
n p,q
× t k−1 (1  qt)n+1−k d p,q t + bn,0 (1, x)
k−1 p,q
p,q
0

n  
n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q
= [n + 1] p,q
k−1
k=1
p,q
× B p,q (k, n − k + 2) + bn,0 (1, x)

n
[n] p,q ![k − 1] p,q !
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q
= [n + 1] p,q
[n + 1 − k] p,q !
k=1
 G. V. Milovanović et al.

[k − 1] p,q ![n − k + 1] p,q !

× p [(n+1−k)(n+k)/2]
p,q
+ bn,0 (1, x)
[n + 1] p,q !
= Bn, p,q (1, x) = 1.

Next, applying Remark 3.2, we have


n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q p,q
Dn (e1 , x) = [n + 1] p,q
k=1
1  
n
× t k (1  qt)n+1−k d p,q t
k−1 p,q
p,q
0

n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q
= [n + 1] p,q
k=1
 
n
× B p,q (k + 1, n − k + 2)
k−1 p,q

n  
−[(n+1−k)(n+k)/2] p,q n
= [n + 1] p,q p bn,k (1, x)
k−1 p,q
k=1
[k] p,q ![n − k + 1] p,q !
× p (n+1−k)(n+k+2)/2
[n + 2] p,q !

n
p,q [k] p,q
= p n−k+1 bn,k (1, x)
[n + 2] p,q
k=1

p[n] p,q  p,q

n
p n−k [k] p,q p[n] p,q x
= bn,k (1, x) = .
[n + 2] p,q [n] p,q [n + 2] p,q
k=1

Further, using the identity [k + 1] p,q = p k + q[k] p,q and by Remark 3.2, we get

n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q p,q
Dn (e2 , x) = [n + 1] p,q
k=1
1  
n
× t k+1 (1  qt)n+1−k d p,q t
k−1 p,q
p,q
0

n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q
= [n + 1] p,q
k=1
 
n
× B p,q (k + 2, n − k + 2)
k−1 p,q

n  
n
p −[(n+1−k)(n+k)/2] bn,k (1, x)
p,q
= [n + 1] p,q
k−1 p,q
k=1
( p, q)-Beta functions and applications in approximation

[k + 1] p,q ![n − k + 1] p,q !

× p (n+1−k)(n+k+4)/2
[n + 3] p,q !

n
p,q [k] p,q [k + 1] p,q
= p 2(n−k+1) bn,k (1, x) ;
[n + 2] p,q [n + 3] p,q
k=1

i.e.,

p,q

n
p,q [k] p,q ( p k + q[k] p,q )
Dn (e2 , x) = p 2(n−k+1) bn,k (1, x)
[n + 2] p,q [n + 3] p,q
k=1

p n+2 [n] p,q  n

p,q p n−k [k] p,q
= bn,k (1, x)
[n + 2] p,q [n + 3] p,q [n] p,q
k=1
2
p 2 q[n]2p,q n
p,q p n−k [k] p,q
+ bn,k (1, x)
[n + 2] p,q [n + 3] p,q [n] p,q
k=1
 
p n+2 [n] p,q x p 2 q[n]2p,q p n−1 x(1−x)
= + x2+
[n + 2] p,q [n + 3] p,q [n + 2] p,q [n + 3] p,q [n] p,q
p n+2 [n] p,q x p 2 q[n]2p,q x 2 p n+1 q[n] p,q x(1 − x)
= + +
[n+2] p,q [n+3] p,q [n+2] p,q [n + 3] p,q [n + 2] p,q [n + 3] p,q
( p + q) p n+1 [n] p,q x ([n] p,q − p n−1 ) p 2 q[n] p,q x 2
= + .
[n + 2] p,q [n + 3] p,q [n + 2] p,q [n + 3] p,q


Remark 3.4 Using above lemma, we can obtain the following central moments:
( p[n] p,q − [n + 2] p,q )x
1◦ Dn
p,q
((t − x), x) =
[n + 2] p,q

 ( p + q) p n+1 [n] x
2 ◦ Dn
p,q p,q
(t − x)2 , x =
[n + 2] p,q [n + 3] p,q

[([n] p,q − p n−1 ) p 2 q[n] p,q −2 p[n] p,q [n + 3] p,q +[n + 2] p,q [n + 3] p,q ]x 2
+ .
[n + 2] p,q [n + 3] p,q

Lemma 3.5 Let n be a given natural number; then

 
p,q 6 1
Dn ((t − x) , x) ≤
2
ϕ (x) +
2
,
[n + 2] p,q [n + 2] p,q

where ϕ 2 (x) = x(1 − x), x ∈ [0, 1].

Proof In view of Lemma 3.3, we obtain

p,q ( p + q) p n+1 [n] p,q x

Dn ((t − x)2 , x) =
[n + 2] p,q [n + 3] p,q
 G. V. Milovanović et al.

[([n] p,q − p n−1 ) p 2 q[n] p,q − 2 p[n] p,q [n + 3] p,q + [n + 2] p,q [n + 3] p,q ]x 2
+ .
[n + 2] p,q [n + 3] p,q

By direct computations, using the definition of the ( p, q) -numbers, we get

( p + q) p n+1 [n] p,q = p n+1 ( p + q)( p n−1 + p n−2 q + p n−3 q 2

+ · · · + pq n−2 + q n−1 ) > 0

for every q ∈ (q0 , 1).

Furthermore, the expression

( p + q) p n+1 [n] p,q + ([n] p,q − p n−1 ) p 2 q[n] p,q − 2 p[n] p,q [n + 3] p,q
+ [n + 2] p,q [n + 3] p,q

is equal to

( p + q) p n+1 [n] p,q + p 2 q[n]2p,q − p n+1 q[n] p,q

− 2 p[n] p,q ( p n+2 + qp n+1 + q 2 p n + q 3 [n] p,q )
+ ( p n+1 + qp n + q 2 [n] p,q )( p n+2 + qp n+1 + q 2 p n + q 3 [n] p,q ) ≤ 6.

In conclusion, for x ∈ [0, 1], we have

p,q 6
Dn ((t − x)2 , x) ≤ δ 2 (x), (3.4)
[n + 2] p,q n

which was to be proved.



4 Local and global estimates

In this section, we estimate some direct results, viz., local and global approximation
in terms of modulus of continuity.
Our first main result is a local
 theorem. 
For this, we denote W 2 = g ∈ C[0, 1] : g  ∈ C[0, 1] ; for δ > 0, K -functional
is defined as  
K 2 ( f, δ) = inf f − g + η g  : g ∈ W 2 ,
where norm- . denotes the uniform norm on C[0, 1]. Following the well-known
inequality due to DeVore and Lorentz [3], there exists an absolute constant C > 0
such that √
K 2 ( f, δ) ≤ Cω2 ( f, δ), , (4.1)
where the second-order modulus of smoothness for f ∈ C[0, 1] is defined as
√
ω2 ( f, δ) = sup√ sup | f (x + h) − f (x)|.
0<h≤ δ x,x+h∈[0,1]
( p, q)-Beta functions and applications in approximation

The usual modulus of continuity for f ∈ C[0, 1] is defined as

ω( f, δ) = sup sup | f (x + h) − f (x)|.

0<h≤δ x,x+h∈[0,1]

Theorem 4.1 Let n > 3 be a natural number and let 0 < q < p ≤ 1, q0 = q0 (n) ∈
(0, p) be defined as in Lemma 3.5. Then, there exists an absolute constant C > 0 such
that

  
p,q −1/2 2x
|Dn ( f, x) − f (x)| ≤ C ω2 f, [n + 2] p,q δn (x) + ω f, ,
[n + 2] p,q

where f ∈ C[0, 1], δn2 (x) = ϕ 2 (x) + 1

[n+2] p,q , ϕ 2 (x) = x(1 − x), x ∈ [0, 1] and
q ∈ (q0 , 1).
Proof For f ∈ C[0, 1] we define
 
p[n] p,q x
np,q ( f, x) = Dnp,q ( f, x) + f (x) − f
D .
[n + 2] p,q

Then, by Lemma 3.3, we immediately observe that

p,q p,q
D̃n (1, x) = Dn (1, x) = 1

and

p,q p,q p[n] p,q x

D̃n (t, x) = Dn (t, x) + x − = x.
[n + 2] p,q

By applying Taylor’s formula

t

g(t) = g(x) + (t − x)g (x) + (t − u)g  (u) du,
x

we get
⎛ ⎞
t
np,q (g, x) = g(x) + D
D np,q ⎝ (t − u)g  (u) du, x ⎠
x
⎛ ⎞
t
= g(x) +
p,q
Dn ⎝ (t − u)g  (u) du, x ⎠
x
p[n] p,q x

[n+2] p,q  
p[n] p,q x
− − u g  (u) du.
[n + 2] p,q
x
 G. V. Milovanović et al.

np,q (g, x) − g(x)|

Thus, | D

⎛ t ⎞ p[n] p,q x
 
[n+2] p,q
p,q ⎝
, x ⎠ +
p[n] p,q x


|g (u)|du


≤ Dn |t − u| · |g  (u)| du [n + 2] −u
p,q
x x
 2
p[n] p,q x
≤ Dn ((t − x)2 , x) g  + g  .
p,q
−x (4.2)
[n + 2] p,q

Also, we have

 2
p,q p[n] p,q x
Dn ((t − x)2 , x) + −x
[n + 2] p,q
   
6 1 ( p[n] p,q − [n + 2] p,q )x 2
≤ ϕ (x) +
2
+
[n + 2] p,q [n + 2] p,q [n + 2] p,q
 
10 1
≤ ϕ 2 (x) + . (4.3)
[n + 2] p,q [n + 2] p,q

Hence, by (4.2) and with the condition n > 3 and x ∈ [0, 1], we have

n (g, x) − g(x)| ≤ 10
δ 2 (x) g  .
p,q
|D (4.4)
[n + 2] p,q n

p,q
Furthermore, for f ∈ C[0, 1] we have ||Dn ( f, x)|| ≤ || f ||; thus
 
p[n] p,q x
 p,q p,q
| Dn ( f, x)| ≤ |Dn ( f, x)| + | f (x)|+ f ≤ 3 f (4.5)
[n + 2] p,q

for all f ∈ C[0, 1].

Now, for f ∈ C[0, 1] and g ∈ W 2 , we obtain

p,q
|Dn ( f, x) − f (x)|
 
p,q p[n] p,q x

= Dn ( f, x) − f (x) + f − f (x)
[n + 2] p,q
 p,q  p,q
≤ | Dn ( f − g, x)| + | Dn (g, x) − g(x)| + |g(x) − f (x)|
 
p[n] p,q x
+ f − f (x)
[n + 2] p,q
 
10 ( p[n] p,q − [n + 2] p,q )x
≤ 4 f − g + δn2 (x) g  + ω f, ,

[n + 2] p,q [n + 2] p,q
(4.6)
( p, q)-Beta functions and applications in approximation

where we have used (4.4) and (4.5). Taking the infimum on the right-hand side over
all g ∈ W 2 , we obtain at once
   
p,q 1 2x
|Dn ( f, x) − f (x)| ≤ 10K 2 f, δn2 (x) + ω f, .
[n + 2] p,q [n + 2] p,q

Finally, in view of (4.1), we find

  
p,q −1/2 2x
|Dn ( f, x) − f (x)| ≤ C ω2 f, [n + 2] p,q δn (x) +ω f, .
[n + 2] p,q

This completes the proof of the theorem.


The weighted modulus of continuity of second order is defined as
ϕ √
ω2 ( f, δ) = sup√ sup | f (x + hϕ(x)) − 2 f (x) + f (x − hϕ(x))|,
0<h≤ δ x,x±hϕ∈[0,1]

√
where ϕ(x) = x(1 − x). The corresponding K -functional is defined by
 
K 2,ϕ ( f, δ) = inf f − g + δ ϕ 2 g  + δ 2 g  : g ∈ W 2 (ϕ) ,

where  
W 2 (ϕ) = g ∈ C[0, 1] : g  ∈ ACloc [0, 1], ϕ 2 g  ∈ C[0, 1]
and g  ∈ ACloc [0, 1] means that g is differentiable and g  is absolutely continuous on
every closed interval [a, b] ⊂ [0, 1]. It is well known due to Ditzian–Totik (see [4, p.
24, Theorem 1.3.1]) that √
ϕ
K 2,ϕ ( f, δ) ≤ Cω2 ( f, δ) (4.7)
for some absolute constant C > 0. Moreover, the Ditzian–Totik moduli of first order
is given by

ωψ ( f, δ) = sup sup | f (x + hψ(x)) − f (x)|,

0<h≤δ x,x±hψ(x)∈[0,1]

where ψ is an admissible step-weight function on [0, 1].

Now, we state our next main result, i.e., the global estimate.
Theorem 4.2 Let n > 3 be a natural number and let 0 < q < p ≤ 1, q0 = q0 (n) ∈
(0, p) be defined as in Lemma 3.5. Then, there exists an absolute constant C > 0 such
that
ϕ −1/2
Dn f − f ≤ Cω2 ( f, [n + 2]q ) + ωψ ( f, [n + 2]q−1 ),
p,q

where f ∈ C[0, 1], q ∈ (q0 , 1), and ψ(x) = x, x ∈ [0, 1].

Proof We again consider
 
p[n] p,q x
np,q ( f, x) = Dnp,q ( f, x) + f (x) − f
D ,
[n + 2] p,q
 G. V. Milovanović et al.

where f ∈ C[0, 1]. Now, using Taylor’s formula, we have

t

g(t) = g(x) + (t − x) g (x) + (t − u) g  (u) du.
x

Using (3.3), we obtain

⎛ ⎞
t
np,q (g, x)
D = g(x) +
p,q
Dn ⎝ (t − u) g  (u) du, x ⎠
x
p[n] p,q x

[n+2] p,q  
p[n] p,q x
− − u g  (u) du.
[n + 2] p,q
x

Thus,
⎛ ⎞
t

p,q
| D̃n (g, x) − g(x)| ≤
p,q
Dn ⎝ |t − u||g  (u)| du , x ⎠

x
p[n] p,q x

[n+2] p,q
p[n] p,q x 
+
[n + 2] −u |g (u)| du

.
(4.8)
p,q
x

Since δn2 is concave on [0, 1], for u = t + τ (x − t), τ ∈ [0, 1], the following
estimate holds:
|t − u| τ |x − t| τ |x − t| |t − x|
= 2 ≤ 2 ≤ 2 .
δn2 (u) δn (t + τ (x − t)) δn (t) + τ (δn2 (x) − δn2 (t)) δn (x)

Thus, using (4.8), we obtain

p,q
| D̃n (g, x) − g(x)|
p[n] p,q x
⎛ ⎞
t  p,q
[n+2] p[n] p,q x
p,q ⎝ |t − u| [n+2] p,q − u
≤ Dn du , x ⎠ δ 2 g  + du δn2 g 
δn2 (u) n δn2 (u)
x x
 2
1 1 p[n] p,q x
Dn ((t − x)2 , x) δn2 g  + 2 δn2 g  .
p,q
≤ 2 −x
δn (x) δn (x) [n + 2] p,q

For x ∈ [0, 1], in view of (4.3) and

1 1
δn2 (x) |g  (x)| = |ϕ 2 (x) g  (x)| + |g  (x)| ≤ ϕ 2 g  + g  ,
[n + 2] p,q [n + 2] p,q
( p, q)-Beta functions and applications in approximation

we have
 
n (g, x) − g(x)| ≤ 5 1
ϕ 2 g  + g  .
p,q
|D (4.9)
[n + 2] p,q [n + 2] p,q

Using the fact that [n] p,q ≤ [n + 2] p,q , (4.5) and (4.9), for f ∈ C[0, 1], we get

p,q p,q p,q

|Dn ( f, x) − f (x)| ≤ | D̃n ( f − g, x)| + | D̃n (g, x) − g(x)| + |g(x) − f (x)|
 
p[n] p,q x
+ f − f (x)
[n + 2] p,q
10 10
≤ 4 f − g + ϕ 2 g  + g 
[n + 2] p,q [n + 2] p,q
 
p[n] p,q x
+ f − f (x) .
[n + 2] p,q

On taking the infimum on the right-hand side over all g ∈ W 2 (ϕ), we obtain
   
1 p[n] p,q x
+ f − f (x) .
p,q
|Dn ( f, x) − f (x)| ≤ 10K 2,ϕ f,
[n + 2] p,q [n + 2] p,q
(4.10)
Moreover,
   
p[n] p,q x x p[n] p,q − [n + 2] p,q
f
− f (x) = f x + ψ(x) − f (x)
[n + 2] p,q [n + 2] p,q ψ(x)

 
x p[n] p,q − [n + 2] p,q
≤ sup f t + ψ(t) − f (t)
[n + 2] p,q ψ(x)
x ( p[n] p,q −[n+2] p,q )
t,t+ψ(t) [n+2] p,q ψ(x) ∈[0,1]

  
|x p[n] p,q − [n + 2] p,q |
≤ ωψ f,
[n + 2] p,q ψ(x)
   
x 1
≤ ωψ f, = ωψ f, .
[n + 2] p,q ψ(x) [n + 2] p,q

Hence, by (4.10) and (4.7), we finally get

ϕ −1/2
f − f ≤ C ω2 ( f, [n + 2] p,q ) + ωψ ( f, [n + 2]−1
p,q
Dn p,q ).

This completes the proof of the theorem.



Remark 4.3 For q ∈ (0, 1) and p ∈ (q, 1] it is obvious that limn→∞ [n] p,q = p−q .
1

An example of such choice for p, q depending on n is recently given in [7].

 G. V. Milovanović et al.

Fig. 1 Graphics of the difference x  → Dn0.5,0.4 ( f, x) − f (x) for x ∈ [0, 1], when f (x) = 9x 2 − 4x + 5
and n = 5, 10, 15 and n = 100

Fig. 2 Graphics of the difference x  → Dn0.5,0.4 ( f, x) − f (x) for x ∈ [0, 1], when f (x) = (x +
1)2 sin(10π x/3) and n = 5, 10, 15 and n = 100

Example Now, we show comparisons and some illustrative graphs for the convergence
p,q
of ( p, q)-analogue of Bernstein–Durrmeyer operators Dn ( f, x) for different values
of the parameters p and q, such that 0 < q < p ≤ 1.
For x ∈ [0, 1], p = 0.5 and q = 0.4, the convergence of the difference of the
p,q
operators Dn ( f, x) to the function f , where f (x) = 9x 2 − 4x + 5, for different
values of n, is illustrated in Fig. 1.
p,q
The convergence of the difference of the operators Dn ( f, x) to the function f ,
where f (x) = (x + 1) sin( 3 π x) for different values of n and x ∈ [0, 1] is illustrated
2 10

in Fig. 2.

Example For the function f (x) = 9x 2 − 4x + 5 (and p = 0.5, q = 0.4), the limit of
Dn0.5,0.4 ( f, x), when n → +∞, is f ∗ (x) = 5 + (124/25)x + (576/25)x 2 . Graphics
of Dn0.5,0.4 ( f, x) − f ∗ (x), for n = 10, 15, 20, 50, are presented in Fig. 3.
( p, q)-Beta functions and applications in approximation

Fig. 3 Graphics of Dn0.5,0.4 ( f, x) − f ∗ (x), when f (x) = 9x 2 − 4x + 5 and f ∗ (x) = 5 + (124/25)x +

(576/25)x 2 , for n = 10, 15, 20, 50

5 Better approximation

About a decade ago, King [11] proposed a technique to obtain better approximation for
the well-known Bernstein polynomials. In this technique, these operators approximate
each continuous function f ∈ [0, 1], while preserving the function e2 (x) = x 2 . These
were basically compared with estimates of approximation by Bernstein polynomials.
Various standard linear positive operators preserve e0 and e1 , i.e., preserve constant
and linear functions, but this approach helps in reproducing the quadratic functions as
well.
So, using King’s technique, we modify the operators (3.3) as follows:


n
∗
p −(n−k+1)(n+k)/2 bn,k (1, rn (x))
p,q
Dn, p,q ( f, x) = [n + 1] p,q
k=1
1
p,q p,q
× bn,k−1 (t) f (t) d p,q t + bn,0 (1, rn (x)) f (0),
0

[n+2] p,q x [n+2] p,q

where rn (x) = p[n] p,q and x ∈ In, p,q = [0, p[n] p,q ]. Then, we have

∗ ∗
Dn, p,q (e0 , x) = 1, Dn, p,q (e1 , x) = x,

∗ ( p + q) p n x ([n] p,q − p n−1 ) q [n + 2] p,q x 2

Dn, p,q (e2 , x) = + .
[n + 3] p,q [n] p,q [n + 3] p,q

Now, Theorem 4.1 can be modified as

Theorem 5.1 Let n > 3 be a natural number and let 0 < q < p ≤ 1, q0 = q0 (n) ∈
(0, p) be defined as in Lemma 3.5. Then, there exists an absolute constant C > 0 such
 G. V. Milovanović et al.

∗
 
Fig. 4 Graphics of the difference x  → Dn,0.5,0.4 ( f, x) − f (x) for x ∈ 0, [n + 2] p,q / p[n] p,q , when
f (x) = 9x 2 − 4x + 5 and n = 5, 10, 15 and n = 100

that
  
∗ p,q
|Dn, p,q ( f, x) − f (x)| ≤ C ω2 f, δn (x) ,

[n+2] p,q
where x ∈ In, p,q = [0, p[n] p,q ], q ∈ (q0 , 1), and

p,q ∗
δn (x) = Dn, p,q ((t − x) , x)
2

( p + q) p n x {([n] p,q − p n−1 ) q [n + 2] p,q − [n] p,q [n + 3] p,q }x 2

= + .
[n + 3] p,q [n] p,q [n + 3] p,q

The proof is on similar lines, so we omit the details.

Example We compare the convergence of ( p, q)-analogue of Bernstein–Durrmeyer

p,q
operators Dn ( f, x) with the operators Dn, ∗
p,q ( f, x). We have considered the same
function as in the previous example.
[n+2] p,q
For x ∈ 0, p[n] p,q , p = 0.5 and q = 0.4, the convergence of the difference of
∗
p,q ( f, x) to the function f , where f (x) = 9x − 4x + 5, for different
the operators Dn, 2

values of n is illustrated in Fig. 4.

References
1. Aral, A., Gupta, V., Agarwal, R.P.: Applications of q Calculus in Operator Theory. Springer, New York
(2013)
2. Acar, T.: ( p, q)-Generalization of Szász-Mirakyan Operators. Operators. Math. Meth. Appl. Sci.
39(10), 2685–2695 (2016)
3. DeVore, R.A., Lorentz, G.G.: Constructive Approximation. Springer, Berlin (1993)
4. Ditzian, Z., Totik, V.: Moduli of Smoothness. Springer, New York (1987)
5. Gupta, V.: Some approximation properties on q-Durrmeyer operators. Appl. Math. Comput. 197(1),
172–178 (2008)
( p, q)-Beta functions and applications in approximation

6. Gupta, V.: ( p, q)-genuine Bernstein Durrmeyer operators. Boll. Unione Mat. Italiana. doi:10.1007/
s40574-016-0054-4 (in press)
7. Gupta, V., Aral, A.: Bernstein Durrmeyer operators based on two parameters. Facta Univ. Ser. Math.
Inform. 31(1), 79–95 (2016)
8. Gupta, V., Finta, Z.: On certain q Durrmeyer operators. Appl. Math. Comput. 209, 415–420 (2009)
9. Gupta, V., Wang, H.: The rate of convergence of q-Durrmeyer operators for 0 < q < 1. Math. Meth.
Appl. Sci. 31(16), 1946–1955 (2008)
10. Gupta, V., Agarwal, R.P.: Convergence Estimates in Approximation Theory. Springer, Cham (2014)
11. King, J.P.: Positive linear operators which preserves x 2 . Acta Math. Hung. 99, 203–208 (2003)
12. Milovanović, G.V., Cvetković, A.S.: An application of little 1/q-Jacobi polynomials to summation of
certain series. Facta Univ. Ser. Math. Inform. 18, 31–46 (2003)
13. Mursaleen, M., Ansari, K.J., Khan, A.: On ( p, q)-analogue of Bernstein operators. Appl. Math. Com-
put. 266, 874–882 (2015)
14. Sadjang, P. N.: On the ( p, q)-Gamma and the ( p, q)-Beta functions (2015). arXiv:1506.07394v1
15. Sadjang, P. N.: On the fundamental theorem of ( p, q)-calculus and some ( p, q)-Taylor formulas.
arXiv:1309.3934 [math.QA]
16. Sahai, V., Yadav, S.: Representations of two parameter quantum algebras and p, q-special functions.
J. Math. Anal. Appl. 335, 268–279 (2007)