Professional Documents
Culture Documents
Abstract
The purpose of this paper is to construct a bivariate generalization of new family of Kan-
ϕ
torovich type sampling operators (Kw f )w>0 . First, we give the pointwise convergence theorem
and a Voronovskaja type theorem for these Kantorovich generalized sampling series. Further,
we obtain the degree of approximation by means of modulus of continuity and quantitative ver-
ϕ
sion of Voronovskaja type theorem for the family (Kw f )w>0 . Finally, we give some examples of
kernels such as box spline kernels and Bochner-Riesz kernel to which the theory can be applied.
Keywords. Bivariate Kantorovich sampling operators Voronovskaja type formula rate of con-
vergence modulus of smoothness.
1 Introduction
In recent years, theory of generalized sampling series is an attractive topic in approximation
theory due to its wide range of applications, especially in signal and image processing. The
theory of generalized sampling series was first initiated by P. L. Butzer and his school [14] and
[13]. After this many researchers have studied in this direction and obtained the convergence
properties of these operators (cf. [7, 28, 29, 32] etc.).
Also, the Kantorovich type generalizations of the generalized sampling operators were intro-
duced by P. L. Butzer and his school (see e.g. [14, 17, 23, 35] etc). In [10], the authors introduced
the sampling Kantorovich operators and studied their rate of convergence in the general settings
∗
Department of Mathematics, Visvesvaraya National Institute of Technology, Nagpur, Nagpur-440010, India.
E-mail: mathsatish9@gmail.com
†
Department of Mathematics, Indian Institute of Science Education and Research, Thiruvananthapuram
E-mail: devarajp@iisertvm.ac.in
1
of Orlicz spaces. Danilo and Vinti [21] obtained the rate of approximation for the family of sam-
pling Kantorovich operators in the uniform norm for uniformly continuous bounded functions
belonging to Lipschitz classes and for functions in Orlicz spaces. Also, the nonlinear version of
sampling Kantorovich operators have been studied in [20] and [36].
Let ϕ : R2 → R be a suitable kernel function. Then, the two dimensional generalized sam-
pling series of a function f : R2 → R is defined by
∞ ∞
ϕ k j k j
(Tw f )(x, y) = ∑ ∑ ϕ w x − ,w y− f , ,
k=−∞ j=−∞ w w w w
where w ∈ N and (x, y) ∈ R2 . These operators have great importance in the development of
models for signal recovery. These type of operators have been studied by many authors (cf.
[9, 15, 16, 25] etc.).
Altomare and Leonessa [4] considered a new sequence of positive linear operators acting
on the space of Lebesgue-integrable functions on the unit interval. Such operators include the
Kantorovich operators as a particular case. Later, in order to obtain an approximation process for
spaces of locally integrable functions on unbounded intervals, Altomare et. al. introduced and
studied the generalized Szász-Mirakjan-Kantorovich operators in [5]. Also in [19], the authors
obtained some qualitative properties and an asymptotic formula for such a sequence of operators.
Motivated by the above works, we consider the bivariate generalized Kantorovich sampling
series. Let {ak }k∈Z , {bk }k∈Z , {c j } j∈Z and {d j } j∈Z be sequences of real numbers such that for
every k ∈ Z, ak < bk and for every j ∈ Z, c j < d j . In this paper, we analyse the approximation
properties of the following type of bivariate generalized Kantorovich sampling series: For f ∈
C(R2 ) (The class of all uniformly continuous and bounded functions on R2 ),
∞ ∞ Z k+bk Z j+d j
w2 w w
(Kwϕ f )(x, y) = ∑ ∑ ϕ (wx − k, wy − j) j+c j
f (u, v)dvdu,
k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
2 Preliminaries
Let us denote by C0 := C0 (R2 ) = C(R2 ) the space of all uniformly continuous and bounded
functions f : R2 → R with the usual supnorm k f k∞ and that by Cn := Cn (R2 ) the space of all
n-times continuously differentiable functions for which
∂ |k| f
∈ C0 ,
∂ xk1 ∂ yk2
2
for k = (k1 , k2 ) with |k| = k1 + k2 = n ≥ 1.
For δ > 0 and (x, y) ∈ R2 , Uδ (x, y) denotes the ball of radius δ centered at (x, y). i.e.,
Uδ (x, y) = {(u, v) ∈ R2 : (x − u)2 + (y − v)2 ≤ δ 2 }.
Let ϕ ∈ C0 be fixed. For any ν ∈ N0 = N ∪ {0}, h = (h1 , h2 ) ∈ N20 with |h| = h1 + h2 = ν ,
we define the algebraic moments as
∞ ∞
mν(h1 ,h2 ) (ϕ ) := ∑ ∑ ϕ (x − k, y − j)(k − x)h1 ( j − y)h2
k=−∞ j=−∞
and
ν
Mν (ϕ ) := max M(h1 ,h2 )
(ϕ ).
|h|=v
Note that for µ , ν ∈ N0 with µ < ν , Mν (ϕ ) < +∞ implies Mµ (ϕ ) < +∞. Indeed for µ < ν ,
we have
∞ ∞
∑ ∑ |ϕ (x − k, y − j)||k − x|h1 | j − y|h2
k=−∞ j=−∞
When ϕ has compact support, we immediately have that Mν (ϕ ) < +∞ for every ν ∈ N0 .
3
(ii) M2 (ϕ ) < +∞ and there holds
3 Main Results
First, we give a pointwise convergence theorem at continuity points of the function f for the
bivariate operators.
Theorem 3.1 For f ∈ L∞ (R2 ), lim (Kwϕ f )(x, y) = f (x, y) at every point (x, y) of continuity of f .
w→∞
Moreover, if the function is uniformly continuous and bounded on R2 , then lim kKwϕ f − f k = 0.
w→∞
Proof 1 Let ε > 0 be fixed. By the pcontinuity of f at the point (x, y), there exists δ > 0 such that
| f (u, v) − f (x, y)| ≤ ε , whenever (u − x)2 + (v − y)2 ≤ δ .
Now, we have
ϕ
|(Kw f )(x, y) − f (x, y)|
Z k+bk Z j+d j
w2 w w
≤ ∑ ∑ ∆ak ∆c j
|ϕ (wx − k, wy − j)| k+ak j+c j
| f (u, v) − f (x, y)|dvdu
( wk , wj )∈U δ (x,y) w w
2
Z k+bk Z j+d j
w2 w w
+∑ ∑ |ϕ (wx − k, wy − j)| j+c j
| f (u, v) − f (x, y)|dvdu
∆ak ∆c j k+ak
( wk , wj )∈U
/ δ (x,y) w w
2
= I1 + I2 .
4
By the continuity of f at the point (x, y), we get
Il ≤ ε ∑ ∑ |ϕ (wx − k, wy − j)| ≤ ε .M0 (ϕ ),
( wk , wj )∈U δ (x,y)
2
p
for sufficiently large w > 0. Let δ > 0 be fixed. If ( wk , wj ) ∈
/ U δ (x, y) then (wx − k)2 + (wy − j)2 >
2
wδ
2 . Hence, we get
8k f k∞
|ϕ (wx − k, wy − j)|(wx − k)2 + (wy − j)2 → 0 as w → ∞
w2 δ 2 ∑ ∑
I2 ≤
( wk , wj )∈U
/ δ (x,y)
and hence the first part of the theorem follows. The second part can be proved similarly.
Remark 3.1 Since functions in L∞ (R2 ) are locally Lebesgue integrable, the generalized Kan-
torovich sampling series used in the above theorem are well defined for f ∈ L∞ (R2 ).
Next, we obtain a Voronovskaja type theorem for the bivariate Kantorovich sampling opera-
tors.
Theorem 3.2 Let f ∈ L∞ (R2 ) be such that ∂∂ xf , ∂∂ yf exist at (x, y) and let {ak }k∈Z , {bk }k∈Z , {c j } j∈Z
and {d j } j∈Z be bounded sequences such that ak + bk = α , c j + d j = β , bk − ak ≥ l0 > 0,
c j − d j ≥ s0 > 0 and sup(k, j) {|ak |, |bk |, |ck |, |dk |} ≤ K for some K > 0. Then, we have
ϕ 1 ∂f ∂f
lim w[(Kw f )(x, y) − f (x, y)] = α (x, y) + β (x, y) .
w→∞ 2 ∂x ∂y
Proof 2 Let K = sup{|ak |, |bk |, |c j |, |d j |}. Let (x, y) ∈ R2 . By Taylor’s theorem, we have
(k, j)
∂f ∂f
f (u, v) = f (x, y) + (x, y)(u − x) + (x, y)(v − y) + h(u − x, v − y))((u − x) + (v − y)),
∂x ∂y
for some bounded function h such that h(u, v) → 0 as (u, v) → (0, 0).
Thus, we have
ϕ
(Kw f )(x, y) − f (x, y)
∞ ∞ Z k+bk Z j+d j
∂f w2 w w
= (x, y) ∑ ∑ ϕ (wx − k, wy − j) j+c j
(u − x)dvdu
∂y k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∞ ∞ Z k+bk Z j+d j
∂f w2 w w
+ (x, y) ∑ ∑ ϕ (wx − k, wy − j) j+c j
(v − y)dvdu
∂y k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∞ ∞
w2
+ ∑ ∑ ϕ (wx − k, wy − j)
k=−∞ j=−∞ ∆ak ∆c j
Z k+bk Z j+d j
w w
× k+ak j+c j
h(u − x, v − y)((u − x) + (v − y))dvdu
w w
= I1 + I2 + I3 , (say).
5
First, we obtain I1 .
∞ ∞ Z k+bk Z j+d j
∂f w2 w w
I1 = (x, y) ∑ ∑ ϕ (wx − k, wy − j) k+a j+c j
(u − x)dvdu
∂x k=−∞ j=−∞ ∆ak ∆c j w
k
w
∞ ∞ 2 2
∂f w k + bk k + ak
= (x, y) ∑ ∑ ϕ (wx − k, wy − j) −x − −x
∂x k=−∞ j=−∞ 2∆ak w w
∂f 1 ∞ ∞
=
∂x
(x, y) ∑ ∑
2w k=−∞ j=−∞
ϕ (wx − k, wy − j)[(bk + ak ) + 2(k − wx)]
∂f α ∞ ∞
=
∂x
(x, y) ∑ ∑ ϕ (wx − k, wy − j)
2w k=−∞ j=−∞
∂f 1 ∞ ∞
+
∂x
(x, y) ∑ ∑ ϕ (wx − k, wy − j)(k − wx)
w k=−∞ j=−∞
∂f α 1∂f α ∂f
= (x, y) + (x, y)m1(1,0) (ϕ , wx, wy) = (x, y).
∂x 2w w ∂ x 2w ∂ x
β ∂f
Similarly, we obtain I2 = (x, y).
2w ∂ y
√ we estimate I3 . Let ε > 0 be fixed. Then, there exists δ > 0 such that |h(t, s)| ≤ ε for
Now,
every t 2 + s2 ≤ δ . Now, we have
Z k+bk Z j+d j
w2 w w
|I3 | ≤ ∑ ∑ ∆ak ∆c j
|ϕ (wx − k, wy − j)| k+ak j+c j
|h(u − x, v − y)|(|u − x| + |v − y|)dvdu
( wk , wj )∈U δ (x,y) w w
2
Z k+bk Z j+d j
w2 w w
+∑ ∑ |ϕ (wx − k, wy − j)| j+c j
|h(u − x, v − y)|(|u − x| + |v − y|)dvdu
∆ak ∆c j k+ak
( wk , wj )∈U
/ δ (x,y) w w
2
= I3′ + I3′′ .
Z k+bk Z j+d j
w2 w w
|I3′ | ≤ ε∑ ∑ |ϕ (wx − k, wy − j)| j+c j
|u − x|dvdu
∆ak ∆c j k+ak
( wk , wj )∈U δ (x,y) w w
2
Z k+bk Z j+d j
w2 w w
+ε ∑ ∑ |ϕ (wx − k, wy − j)| j+c j
|v − y|dvdu
∆ak ∆c j k+ak
( wk , wj )∈U δ (x,y) w w
2
:= J1 + J2 .
2 2
ε w k + bk k + ak
2∑ ∑
J1 ≤ |ϕ (wx − k, wy − j)| −x + −x
∆ak w w
( wk , wj )∈U δ (x,y)
2
6
ε
2wl0 ∑ ∑
≤ |ϕ (wx − k, wy − j)|
( wk , wj )∈U δ (x,y)
2
× (a2k + b2k ) + 2(ak + bk )|k − wx| + 2(k − wx)2
ε 2 1 2
≤ 2K M0 (ϕ ) + 2|α |M(1,0) (ϕ ) + 2M(2,0) (ϕ ) .
2wl0
Similarly, we get
ε
J2 ≤ 1
2K 2 M0 (ϕ ) + 2|β |M(0,1) 2
(ϕ ) + 2M(0,2) (ϕ ) .
2ws0
Therefore, we have
α M 1 (ϕ ) β M 1 (ϕ )
ε 1 1 (1,0) (0,1)
|I3′ | ≤ 2
K M0 (ϕ ) + + +
w l0 s0 l0 s0
2 2
M(2,0) (ϕ ) M(0,2) (ϕ )
+ + .
l0 s0
Let M > 0 be a constant such that |h(t, s)| ≤ M. Let δ > 0 be fixed. If ( wk , wj ) ∈
/ U δ (x, y), then
s 2 2
2
k j δ
−x + − y > and by assumption (ii), we have
w w 2
′′ ((k − wx)2 + ( j − wy)2 )
|I3 | ≤ 4M ∑ ∑ |ϕ (wx − k, wy − j)|
k j δ w2
( w , w )∈U
/ δ (x,y)
2
for sufficiently large w. Hence, we get lim wI3 = 0. So the assertion follows.
w→∞
Remark 3.2 The boundedness assumption on f in the previous theorem can be relaxed by as-
suming that there are two positive constants a, b such that | f (x, y)| ≤ a + b(x2 + y2 ), for every
(x, y) ∈ R2 .
∞ ∞ Z k+bk Z j+d j
w2 w w
≤ ∑ ∑ |ϕ (wx − k, wy − j)| k+a j+c j
(a + b(u2 + v2 ))dvdu
∆ ∆
k=−∞ j=−∞ ak c j w
k
w
2 2 b 2bK 2
≤ M0 (ϕ ) a + b(x + y ) + (|α ||x| + |β ||y|) + 2
w w
2b 1 1
b
1 1
+ |x|M(1,0) (ϕ ) + |y|M(0,1) (ϕ ) + 2 |α |M(1,0) (ϕ ) + |β |M(0,1) (ϕ )
w w
b 2 2
+ 2 M(2,0) (ϕ ) + M(0,2) (ϕ )
w
7
ϕ
and hence the series Kw f is absolutely convergent for every (x, y) ∈ R2 . Moreover, for a fixed
(x, y) ∈ R2 ,
∂f ∂f
P1 (u, v) = f (x, y) + (x, y)(u − x) + (x, y)(v − y),
∂x ∂y
the Taylor’s polynomial of first order centered at the point (x, y) by the Taylor’s formula, we can
write
f (u, v) − P1 (u, v)
= h(u − x, v − y),
|u − x| + |v − y|
where h is a function such that lim h(u, v) = 0. Then h is bounded on a neighbourhood of
(u,v)→0
(x, y), say Uδ (x, y) > 0. For (u, v) ∈
/ Uδ (x, y), we have
| f (u, v)| | f (x, y)| ∂f |u − x| ∂f |v − y|
|h(u − x, v − y)| ≤ + + + ,
|u − x| + |v − y| |u − x| + |v − y| ∂ x |u − x| + |v − y| ∂ y |u − x| + |v − y|
and all the terms on the right-hand side of the above inequality are bounded for (u, v) ∈
/ Uδ .
Hence, h(. − x, . − y) is bounded on R2 . The same Voronovskaya formula can be obtained by
writing the proof along the lines of the proof of Theorem 3.2.
Remark 3.3 One can easily verify that the following set of sequences satisfy the conditions of
1 1 1 1
the above theorem: ak = |k|+1 , bk = 2 − |k|+1 , c j = − | j|+1 and d j = 1 + | j|+1 . Moreover, several
sequences can be constructed in a similar way.
For f ∈ C(R2 ), the modulus of continuity for the bivariate case is defined as follows:
2
ω ( f , δ1 , δ2 ) := sup | f (t, s) − f (x, y)| : (t, s), (x, y) ∈ R and |t − x| ≤ δ1 , |s − y| ≤ δ2 ,
|t − x| |s − y|
(b) | f (t, s) − f (x, y)| ≤ ω ( f , δ1 , δ2 ) 1 + 1+ ;
δ1 δ2
The details of the modulus of continuity for the bivariate case can be found in [6].
In what follows, we shall define the following K-functional:
0 (1) ∂g ∂g (1)
K( f , δ ,C ,C ) := inf k f − gk∞ + δ max , ,g ∈ C ,
∂x ∞ ∂y ∞
where f ∈ C0 and δ > 0. For every f ∈ C0 there holds
1
K( f , δ /2,C0 ,C1 ) = ω ( f , δ ), (3.1)
2
where ω ( f , .) denotes the least concave majorant of ω ( f , .) (see e.g. [6]).
ϕ
Now, we give the estimate of the rate of convergence of the bivariate operators (Kw f ).
8
Theorem 3.3 Let f ∈ C(R2 ) and sup{ak , bk , c j , d j } ≤ K for some K > 0. Then, for all (x, y) ∈
k, j
R2 , we have
KM0 (ϕ ) 1 1 K
|(Kwϕ f )(x, y) − f (x, y)| ≤ ω ( f , δ1 , δ2 ) M0 (ϕ ) + + +
w δ1 δ2 w δ1 δ2
K 1 1
1 1 1
+ M(1,0) + M(0,1) + +
w δ1 δ2 δ1 δ2 w
for any w > 0, and δ1 > 0, δ2 > 0.
Proof 4 Using the linearity and positivity of the operators and taking into account the property
(b) above, we have
ϕ
|(Kw f )(x, y) − f (x, y)|
∞ ∞ Z k+bk Z j+d j
w2 w w
≤ ∑ ∑ |ϕ (wx − k, wy − j)| j+c j
| f (u, v) − f (x, y)|dvdu
k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∞ ∞
w2
≤ ω ( f , δ1 , δ2 ) ∑ ∑ |ϕ (wx − k, wy − j)|
k=−∞ j=−∞ ∆ak ∆c j
Z k+bk Z j+d j
w w |u − x| |v − y|
× j+c j
1+ 1+ dvdu := I1 + I2 + I3 + I4 .
k+ak
w w
δ1 δ2
It is easy to see that I1 = ω ( f , δ1 , δ2 )M0 (ϕ ).
k + ak k + bk
Now, we estimate I2 . For every u ∈ , , we have
w w
k + ak k + ak k max{|ak |, |bk |} k K
|u − x| ≤ u − + −x ≤ x− + ≤ x− + .
w w w w w w
Then, we obtain
Z k+bk Z j+d j
ω ( f , δ1 , δ2 ) ∞ ∞
w2 w w
I2 ≤
δ1 ∑ ∑ |ϕ (wx − k, wy − j)| k+a j+c j
|u − x|dvdu
k=−∞ j=−∞ ∆ak ∆c j
k
w w
ω ( f , δ1 , δ2 ) ∞ ∞
≤
δ1 w ∑ ∑ |ϕ (wx − k, wy − j)| |k − wx| + K
k=−∞ j=−∞
ω ( f , δ1 , δ2 ) 1
≤ KM0 (ϕ ) + M(0,1) (ϕ ) .
δ1 w
ω ( f , δ1 , δ2 ) 0 1
Similarly, we obtain I3 ≤ KM(0,0) (ϕ ) + M(1,0)(ϕ ) .
δ1 w
Now,
ω ( f , δ1 , δ2 ) 1 ∞ ∞
I4 ≤
δ1 δ2 ∑ ∑ |ϕ (wx − k, wy − j)| |k − wx| + K | j − wy| + K
w2 k=−∞ j=−∞
ω ( f , δ1 , δ2 ) 1 2 0 1 1 2
≤ K M (ϕ ) + K M(1,0) (ϕ ) + M(0,1) (ϕ ) + M(1,1) (ϕ ) .
δ1 δ2 w2
Combining the estimates I1 − I4 , we get the desired result.
9
Using an estimate of the remainder in two dimensional Taylor formula and a technique de-
veloped in [26] by Gonska et al., we obtain a quantitative version of the Voronovskaja formula
for bivariate sampling series. We remark that quantitative Voronovskaja formulae have important
links with the theory of semi-groups of operators (see [3, 11]). The strict connections between
the two theories were described in [2] and recently developed in [1, 18, 30].
Now, we have the following quantitative version of the Voronovskaja formula for bivariate
sampling series.
Lemma 3.1 Let f ∈ C1 and (x, y) ∈ R2 , (x0 , y0 ) ∈ R2 . Then for the remainder R1 ( f ) := R1 ( f , (x0 , y0 ), (x, y))
in the Taylor formula, we have
∂f ∂f
|R1 ( f )| ≤ |x − x0 |ω , |x − x0 | + |y − y0 |ω , |y − y0 | .
∂x ∂y
Proof 5 The proof follows from the Taylor formula of the first order using the same analysis as
of Lemma 1 in [9].
Theorem 3.4 Let f ∈ C1 (R2 ) and let ak , bk , c j , d j satisfy the conditions of theorem 3.2. Then,
for any (x, y) ∈ R2 , we have
ϕ 1 ∂f ∂f
w[(Kw f )(x, y) − f (x, y)] − α +β
2 ∂x ∂y
∂f B ∂f D
≤ Aω , +Cω , ,
∂ x ∞ Aw ∂ y ∞ Cw
where A = l10 2K 2 M0 (ϕ )+2α M(1,0) 1 2
(ϕ )+2M(2,0) (ϕ ) , B = K 2 M0 (ϕ )+ α M(1,0)
1 2 (ϕ ) ,
(ϕ )+M(2,0)
1 2 1 2 2 1 2
C = s0 2K M0 (ϕ )+2β M(0,1) (ϕ )+2M(0,2) (ϕ ) and D = K M0 (ϕ )+ β M(0,1)(ϕ )+M(0,2) (ϕ ) .
∞ ∞ Z k+bk Z j+d j
∂f w2 w w
= (x, y) ∑ ∑ ϕ (wx − k, wy − j) j+c j
(u − x)dvdu
∂x k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∞ ∞ Z k+bk Z j+d j
∂f w2 w w
+ (x, y) ∑ ∑ ϕ (wx − k, wy − j) j+c j
(v − y)dvdu
∂y k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∞ ∞
w2
+ ∑ ∑ ϕ (wx − k, wy − j)
k=−∞ j=−∞ ∆ak ∆c j
Z k+bk Z j+d j
w w 1 ∂f ∂f
× j+c j
h(u − x, v − y)((u − x) + (v − y))dudv − α +β
k+ak
w w
2 ∂x ∂y
∞ ∞
w2
≤ ∑ ∑ |ϕ (wx − k, wy − j)|
k=−∞ j=−∞ ∆ak ∆c j
Z k+bk Z j+d j
w w
× k+ak j+c j
|h(u − x, v − y)|(|u − x| + |v − y|)dvdu := I1 .
w w
10
Using the relation (3.1) and Lemma 3.1, we obtain
∞ ∞ Z k+bk Z
j+d j
w2 ∂f
w w
I1 ≤ ∑ ∑ |ϕ (wx − k, wy − j)| j+c j
|u − x|ω
∂x
, |u − x| dvdu
k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∞ ∞ Z k+bk Z j+d j
w2 w w ∂f
+ ∑ ∑ |ϕ (wx − k, wy − j)| k+a j+c j
|v − y|ω , |v − y| dvdu
k=−∞ j=−∞ ∆ak ∆c j w
k
w
∂y
:= I1′ + I1′′ .
∞ ∞ Z k+bk Z j+d j
w2 w w ∂ f |u − x|
|I1′ | ≤ 2 ∑ ∑ |ϕ (wx − k, wy − j)| j+c j
|u − x|K , dvdu
k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
∂x 2
∞ ∞
2w2
≤ ∑ ∑ |ϕ (wx − k, wy − j)|
k=−∞ j=−∞ ∆ak ∆c j
Z k+bk Z j+d j
w w ∂ ( f − g) |u − x| ∂g ∂g
× k+a j+c j
|u − x| + max , dvdu
w
k
w
∂x ∞ 2 ∂x ∞ ∂y ∞
∞ ∞
∂ ( f − g) 2w2
≤
∂x ∑ ∑ |ϕ (wx − k, wy − j)|
∞ k=−∞ j=−∞ ∆ak ∆c j
Z k+bk Z j+d j
∞ ∞
w w w2
× j+c j
|u − x|dvdu + ∑ ∑ |ϕ (wx − k, wy − j)|
k=−∞ j=−∞ ∆ak ∆c j
k+ak
w w
Z k+bk Z j+d j
w w
2 ∂g ∂g
× j+c j
(u − x) max , dvdu := J1 + J2 .
k+ak
w w
∂x ∞ ∂y ∞
w ∂ ( f −g)
∞ ∞ 2 2
∂x ∞ k + bk k + ak
|J1 | ≤
∆ak ∑ ∑ |ϕ (wx − k, wy − j)| w − x + w − x
k=−∞ j=−∞
∂ ( f − g) 1 2 1 2
≤ 2K M0 (ϕ ) + 2α M(1,0) (ϕ ) + 2M(2,0) (ϕ )
∂x ∞ wl0
Now,
∞ ∞ Z k+bk Z j+d j
∂g ∂g w2 w w
J2 ≤ max , ∑ ∑ ∆a ∆c |ϕ (wx − k, wy − j)| j+c j
(u − x)2 dvdu
∂x ∞ ∂y ∞ k=−∞ j=−∞ k j
k+ak
w w
∞ ∞ 3 3 !
∂g
w ∂g k + bk k + ak
≤ max
∂x ∑ ∑ |ϕ (wx − k, wy − j)|
,
∂y
−x − −x
∞ 3∆ak k=−∞ j=−∞
∞ w w
∂g ∂g 1 2 1 2
≤ max , K M0 (ϕ ) + α M(1,0) (ϕ ) + M(2,0)(ϕ ) .
∂ x ∞ ∂ y ∞ w2
11
Thus, we have
∂ ( f − g) 1B ∂g ∂g
wI1′ ≤A + max , .
∂x ∞ wA ∂x ∞ ∂y ∞
Similarly, we obtain
∂ ( f − g) D
wI1′′ ≤ Cω , .
∂y ∞ Cw
Combining the estimates I1′ and I1′′ , we get the required result.
where
1, if − 21 ≤ x ≤ 1
2
χ[− 1 , 1 ] (x) =
2 2 0, otherwise
and ∗ denotes the convolution.
The Fourier transform of the functions βd1 (x) and βd2 (y) are given by
sin u/2 d1 +1
βbd1 (u) = , u ∈ R,
u/2
b sin v/2 d2 +1
βd2 (v) = , v ∈ R.
v/2
(see [13] and [33]). Given real numbers ε0 , ε1 , ε0′ , ε1′ with ε0 < ε1 , ε0′ < ε1′ and define ϕ (x, y) =
ϕ1 (x)ϕ2 (y), where
ϕ1 (x) = a0 βd1 (x − ε0 ) + a1 βd1 (x − ε1 )
ϕ2 (y) = b0 βd2 (y − ε0′ ) + b1 βd2 (y − ε1′ )
12
The Fourier transform of ϕ is ϕb(u, v) = ϕ c2 (v), where
c1 (u)ϕ
′v ′v
c1 (u) = a0 e
ϕ −iε0 u
+ a1 e−iε1 u b c2 (v) = b0 e 0 + b1 e 1 βbd2 (v).
βd1 (u) and ϕ −iε −iε
we obtain
∞ ∞
∑ ϕ1 (u − k) = ∑ c1 (2π k)ei2π ku .
ϕ
k=−∞ k=−∞
We have
d1 +1
sin(π k) 1, if k = 0
βc
d1 (2π k) = =
πk 0, if k 6= 0
and hence
a0 + a1 , if k = 0
c1 (2π k) =
ϕ
0, if k 6= 0
Thus
∞
∑ ϕ1 (u − k) = a0 + a1 .
k=−∞
Also, we have
c1 ′ (u) = (−iε0 a0 e−iε0 u − iε1 a1 e−iε1 u )βbd1 (u) + (a0 e−iε0 u + a1 e−iε1 u )βbd′ 1 (u).
ϕ
13
4.2 Translates of Box Splines
Boor and Devore [24], initiated the theory of bivariate splines and extensively studied by several
researchers (see, e.g., [22] and [27]). First, we give some basic definitions here.
Let A = (aν ,µ ) be a 2 × m matrix with m ≥ 3 and column vectors Aµ ∈ Z2 \{(0, 0)}, µ =
1, 2, . . . , m be such that rank(A)=2. The box spline BA is defined via the following:
Z Z
BA (x, y)g(x, y)dxdy = g(At)dt (g ∈ C(R2 )),
R2 Qm
where Qm = [− 12 , 12 ]m and t = (t1 ,t2 ...,tm ) ∈ Qm . It is well known that BA (x, y) ≥ 0, (x, y) ∈ R2
and Supp(BA ) = AQm (see [22]). Moreover, if ρ = ρ (A) is the greatest integer such that all
submatrices generated from A by deleting ρ columns, have rank 2, then BA ∈ C(ρ −1) (R2 ) and
the Fourier transform of BA is given by
1 m 1
BbA (u, v) = , (u, v) ∈ R2
2π µ∏
sinc (ua1µ + va2µ )
=1 2π
∂ |s| 1
(−1)⌊|s|/2⌋ ∑ bhr hs11 hs22 = (0, 0),
|h|≤r0
∂ u 1 ∂ v 2 BbA
s s
where s = (s1 , s2 ) ∈ N0 2 , |s| = s1 + s2 ≤ r0 and [α ] denotes the largest integer but not bigger
than α , then
1 1
ϕA,r (x, y) =
2π
b0r BA (x, y) + ∑ bhr (BA (x + h1, y + h2 ) + BA(x − h1 , y − h2))
2 0<|h|≤r
0
∞ ∞
(c) ∑ ∑ ϕA,r (x − k, y − j)(x − k)s1 (y − j)s2 = 0, for all (x, y) ∈ R2 , s = (s1 , s2 ) ∈ N0 2
k=−∞ j=−∞
with 0 < |s| < r.
14
The condition (c) of Proposition 4.1 is equivalent to the following one which uses the Fourier
transform of ϕA,r . For every s = (s1 , s2 ) with 0 ≤ |s| < r,
∂ |s| ϕbA,r 0, if (k, j) 6= (0, 0)
(2kπ , 2 jπ ) =
s
∂u ∂v1 s 2 δs0 , if (k, j) = (0, 0),
where
1, if (s1 , s2 ) = (0, 0)
δs0 =
0, otherwise.
Now, we construct the linear combinations of translates of ϕA,r for r ≥ 3 in such a way that all
the assumptions are satisfied for given values of the constants S1 , S2 and S3 . Let S1 , S2 and S3
be real numbers such that (S1 , S2 , S3 ) 6= (0, 0, 0). For given constants a, b, c, we define the kernel
function ϕ by
d
ϕ̂ (u, v) = ϕ A,r (u, v)(C1 + 2C2 cos(a(u + v)) + 2C3 cos(bu) + 2C4 cos(cv)).
for every w > 0. For the generalized sampling series (4.2), we can obtain the Voronovskaja
formula of Theorem 3.1 and the corresponding quantitative estimate.
15
4.3 Radial Kernels: The Bochner-Riesz Kernel
Next, we consider the two dimensional Bochner-Riesz Kernel bγ (x, y)(see, e.g., [12, 15, 16, 31,
34]). It is defined by
2γ Γ(γ + 1) p 2 −1−γ p
ϕ (x, y) ≡ bγ (x, y) = x + y2 J1+γ x2 + y2
2π
for γ > 1/2, where Jλ is the Bessel function of order λ . It is well known that
∞ ∞
∑ ∑ bγ (x − k, y − j) = 1, (x, y) ∈ R2
k=−∞ j=−∞
and so M2 (bγ ) < +∞. The following proposition shows that the kernel function bγ (x, y) satisfies
condition (ii).
Proposition 4.2 [8]. For the function bγ , γ > 5/2 there holds
As a consequences of the following results proved in [15], Lemma 3.2 and Lemma 2 in [9], we
obtain that bγ satisfies condition (iii).
Proposition 4.3 Let ϕ : R2 → R be a continuous and bounded function such that Mr (ϕ ) < +∞
for some r ∈ N0 , h = (h1 , h2 ) ∈ N0 2 with |h| = h1 + h2 ≤ r, the double series in the definition of
Mr (ϕ ) being uniformly convergent on each compact subset of R2 for each h such that |h| = r.
The following two assertions are equivalent for c ∈ R :
∞ ∞
(a) ∑ ∑ ϕ (x − k, y − j)(k − x)h1 (y − j)h2 = c, ∀(x, y) ∈ R2
k=−∞ j=−∞
∂ |h| ϕ̂ (−1)|h| 2cπ , if (k, j) = (0, 0)
(b) (2kπ , 2 jπ ) =
∂ uh1 ∂ vh2 0, if (k, j) 6= (0, 0).
16
Thus, Bochner-Riesz Kernel bγ (x, y) satisfies all required the conditions. Moreover, the bivariate
generalized Kantorovich sampling series now takes the form
q −1−γ
2γ Γ(γ + 1) ∞ ∞
w2
(Kwϕ f )(x, y) = ∑ ∑ 2
(wx − k) + (wy − j) 2
2π k=−∞ j=−∞ ∆ak ∆c j
q Z k+bk Z j+d j
w w
×J1+γ 2
(wx − k) + (wy − j) 2 f (u, v)dvdu, (4.3)
k+a j+c j
k
w w
for every w > 0. For the generalized sampling series (4.3), we can obtain the Voronovskaja
formula as in Theorem 3. 1 and the corresponding quantitative estimate.
17
References
References
[1] F. Altomare, Asymptotic formulae for Bernstein-Schnabl operators and smoothness, Boll.
Unione Mat. Ital. 9 (2) (2009) 135-150.
[6] G.A. Anastassiou, S.G. Gal, Approximation Theory. Moduli of Continuity and Global
Smoothness Preservation, Birkhäuser, Boston (2000).
[7] C. Bardaro, I. Mantellini, A Voronovskaya type theorem for a general class of discrete
operators, Rocky Mountain J. Math. 39 (5) (2009) 1411-1442.
[8] C. Bardaro, I. Mantellini, Asymptotic formulae for bivariate discrete operators. Applica-
tions to generalized sampling series and Szász-Mirakjan type operators, Panamer. Math. J.
20 (1) (2010) 1-21.
[10] C. Bardaro, G. Vinti, P.L. Butzer, R.L. Stens, Kantorovich-type generalized sampling series
in the setting of Orlicz spaces, Sampl. Theor. Signal Image Process. 6 (1) (2007) 29-52.
[11] P.L. Butzer, H. Berens, Semi-Groups of Operators and Approximation. Die Grundlehren
der mathematischen Wissenschaften 145. Springer-Verlag New York Inc., New York
(1967).
[12] P.L. Butzer, R.J. Nessel, Fourier analysis and approximation. Volume 1: One-dimensional
theory. Pure and Applied Mathematics, Vol. 40. Academic Press, New York-London, 1971.
[13] P.L. Butzer, R.L. Stens, Linear prediction by samples from the past, In: Advanced Topics in
Shannon Sampling and Interpolation Theory (Ed. R. J. Marks II) Springer Texts Electrical
Eng., Springer, New York (1993).
[14] P.L. Butzer, R.L. Stens, Sampling theory for not necessarily band-limited functions: a
historical overview, SIAM Rev. 34 (1) (1992) 40-53.
18
[15] P.L. Butzer, A. Fischer, R.L. Stens, Generalized sampling approximation of multivariate
signals; general theory, Atti Sem. Mat. Fis. Univ. Modena 41 (1) (1993) 17-37.
[16] P.L. Butzer, A. Fischer, R.L. Stens, Generalized sampling approximation of multivariate
signals; theory and some applications, Note Mat. 10 (1990) 173-191.
[17] P.L. Butzer, S. Ries, R.L. Stens, Approximation of continuous and discontinuous functions
by generalized sampling series, J. Approx. Theory. 50 (1) (1987) 25-39.
[19] M.M. Cappelletti, V. Leonessa, Approximation of some Feller semigroups associated with
a modification of Szász-Mirakjan-Kantorovich operators, Acta Math. Hungar. 139 (3)
(2013) 255-275.
[24] C. De Boor, R. DeVore, Approximation by smooth multivariate splines, Trans. Amer. Math.
Soc. 276 (2) (1983) 775-788.
[25] A. Fischer, R.L. Stens, Generalized sampling approximation of multivariate signals; in-
verse approximation theorems, Colloq. Math. Soc. János Bolyai (58) (1990) 275286.
[26] H.H. Gonska, P. Piţul, I. Raşa, On Peano’s form of the Taylor remainder, Voronovskaja’s
theorem and the commutator of positive linear operators, Proc. of the Int.Conf. on Numer-
ical Analysis and Approximation Theory, Cluj-Napoca, (2006) 55-80.
[27] K. Hollig, Box splines. In: Approximation Theory V. Proc. Conf. Texas A and M Univ.
(C.K.Chui et al., ed.), Academic Press, Boston, (1986) 71-95.
[29] I. Mantellini, Generalized sampling operators in modular spaces, Comment. Math. (Prace
Mat.) 38 (1998) 77-92.
19
[30] E.M. Mangino, I. Raşa, A quantitative version of Trotter’s approximation theorem, J. Ap-
prox. Theory. 146 (2) (2007) 149-156.
[31] R.J. Nessel, Contributions to the theory of saturation for singular integrals in several vari-
ables, III Radial Kernels. Nederl. Akad. Wetensch. Proc. Ser. 29 (1967) 65-73.
[32] S. Ries, R.L. Stens, Approximation by generalized sampling series, In: Constructive The-
ory of Functions (Ed-s: Bl. Sendov, P. Petrushev, R. Maalev, S. Tashev), Publ. House
Bulgarian Academy of Sciences, Sofia. (1984) 746-756.
[35] G. Vinti, A general approximation result for nonlinear integral operators and applications
to signal processing, Appl. Anal. 79 (1-2) (2001) 217-238.
20