Monika Herzog
Approximation of functions of two variables by modified Szasz-Mirakyan operators
Abstract. In this paper we study approximative properties of modified Szasz-Mirakyan operators for functions of two variables from polynomial weight spaces. We present some direct theorems giving a degree of approximation for these operators.
2000 Mathematics Subject Classification: 41A36.
Key words and phrases: Linear positive operators, Bessel function, Modulus of con- tinuity, Degree of approximation.
1. Introduction. Let us denote by C(R0) a set of all real-valued functions continuous on R0 = [0; +∞) and let N0 = N ∪ {0}. In paper [1] we investigated operators of Szasz-Mirakyan type defined as follows
Aνn(f; x) =
( 1
Iν(nx)P∞
k=0 (nx2 )2k+ν
Γ(k+1)Γ(k+ν+1)f (2kn), x > 0;
f (0), x = 0,
where Γ is the Euler-gamma function and Iν the modified Bessel function defined by the formula ([6], p. 77)
Iν(z) = X∞ k=0
(z2)2k+ν Γ(k + 1)Γ(k + ν + 1). We studied the operators in polynomial weight spaces
Cp= {f ∈ C(R0) : wpf is uniformly continuous and bounded onR0}, where wp was the polynomial weight function defined as follows
wp(x) =
1, p = 0;
1+x1p, p∈ N,
for x ∈ R0.
In the present paper we will consider the bivariate version of the operator Aνn in appropriate weight spaces.
Let C(R20) be the set of all real-valued functions continuous onR20. Similarly as in [1] we define the polynomial weight space
(1) Cp,q = {f ∈ C(R20) : wp,qf is uniformly continuous and bounded onR20}, where wp,q is the polynomial weight function defined as follows
(2) wp,q(x, y) =
1, p = q = 0;
1+x1p, q = 0, p∈ N;
1+y1q, p = 0, q∈ N;
(1+xp)(1+y1 q), p, q∈ N, for (x, y) ∈ R20. The space Cp,q is a normed space with the norm (3) kfkp,q = sup{wp,q(x, y)|f(x, y)|; (x, y) ∈ R20}.
Moreover, we consider the modulus of continuity
(4) ω(f, Cp,q; t, s) = sup{k∆h,dfkp,q; h ∈ [0, t], d ∈ [0, s]}, where
∆h,df (x, y) = f (x + h, y + d)− f(x, y) for (x, y) ∈ R20, h, d ∈ R0.
The note was inspired by the results of [2]-[5]. We introduce the modified Szasz- Mirakyan operator for functions f ∈ Cp,q in the following way
(5) Aν,µn,m(f; x, y) =
1 Iν(nx) 1
Iµ(my)P∞ k=0P∞
j=0
(nx2)2k+ν Γ(k+1)Γ(k+ν+1)
(my2 )2j+µ
Γ(j+1)Γ(j+µ+1)f (2kn,2jm), x > 0, y > 0;
1 Iν(nx)P∞
k=0
(nx2 )2k+ν
Γ(k+1)Γ(k+ν+1)f (2kn, 0), x > 0, y = 0;
1 Iµ(mx)P∞
j=0 (mx2 )2j+ν
Γ(j+1)Γ(j+µ+1)f (0,2jm), y > 0, x = 0;
f (0, 0), x = y = 0.
where p, q ∈ N0, ν, µ ∈ R0, n, m ∈ N.
We shall present direct approximation theorems for these operators. The main re- sults of the paper are theorems giving a degree of approximation of function f ∈ Cp,q by operators Aν,µn,m.
2. Auxiliary results. In this section we will show some basic properties of the operator Aν,µn,m.
At the beginning we will recall preliminary results from paper [1] for the operator Aνn, which we shall apply to the proofs of the main theorems.
Lemma 2.1 For each n ∈ N, ν ∈ R0 and x ∈ R0
Aνn(1; x) = 1, Aνn(t; x) = xIν+1(nx) Iν(nx) , Aνn(t2; x) = x2Iν+2(nx)
Iν(nx) + x2 n
Iν+1(nx) Iν(nx) , Aνn(t − x; x) = x
Iν+1(nx) Iν(nx) −1
,
Aνn((t − x)2; x) = x2
Iν+2(nx)
Iν(nx) −2Iν+1(nx) Iν(nx) + 1
+ x2
n
Iν+1(nx) Iν(nx) .
Lemma 2.2 For each ν ∈ R0there exists a positive constant M(ν) such that for all n∈ N and x ∈ R0 we have Iν+1(nx)
Iν(nx)
¬ M(ν),
nx
Iν+1(nx) Iν(nx) −1
¬ M(ν).
Hence we immediately get
Lemma 2.3 For each ν ∈ R0there exists a positive constant M(ν) such that for all n∈ N and x ∈ R0 we have
|Aνn(t − x; x)| ¬ M (ν)
n , |Aνn((t − x)2; x)| ¬ M(ν)x n.
Lemma 2.4 For all p ∈ N0and ν ∈ R0there exists a positive constant M(p, ν) such that for each n ∈ N we have
kAνn(1/wp; ·)kp ¬ M(p, ν).
Theorem 2.5 For all p ∈ N0 and ν ∈ R0 there exists a positive constant M(p, ν) such that for each n ∈ N we have
kAνn(f; ·)kp¬ M(p, ν)kfkp.
Lemma 2.6 For all p ∈ N0and ν ∈ R0there exists a positive constant M(p, ν) such that for all x ∈ R0 and n ∈ N we have
wp(x)Aνn
(t − x)2 wp(t) ; x
¬ M(p, ν)x + 1 n .
The definition of the operator Aν,µn,m implies
(6) Aν,µn,m(f; x, y) = Aνn(f1; x)Aµm(f2; y) for all functions f ∈ Cp,q of the form f(x, y) = f1(x)f2(y).
In particular we get
Aν,µn,m(1; x, y) = 1,
Aν,µn,m(1/wp,q; x, y) = Aνn(1/wp; x)Aµm(1/wq; y).
From the above facts and Lemma 2.4 we derive
Lemma 2.7 For all p, q ∈ N0 and ν, µ ∈ R0 there exists a positive constant M (p, q, ν, µ) such that for all n, m ∈ N we have
kAν,µn,m(1/wp,q; ·)kp,q ¬ M(p, q, ν, µ).
Theorem 2.8 For all p, q ∈ N0 and ν, µ ∈ R0 there exists a positive constant M (p, q, ν, µ) such that for any f ∈ Cp,q and n, m ∈ N we have
kAν,µn,m(f; ·)kp,q¬ M(p, q, ν, µ)kfkp,q. Proof Applying equation (6) and definition (3) we get
wp,q(x, y)|Aν,µn,m(f(t, s); x, y)| ¬ wp,q(x, y)Aν,µn,m(|f(t, s)|; x, y) =
wp,q(x, y)Aν,µn,m
wp,q(t, s)f(t, s) 1
wp,q(t, s); x, y
¬
kfkp,qwp(x)wq(y)Aνn
1
wp(t); x
Aµq
1
wq(s); y
¬ M(p, ν)M(q, µ)kfkp,q. Hence the operator Aν,µn,mtransforms the space Cp,q into Cp,q.
3. Approximation theorems.
Theorem 3.1 For all p, q ∈ N0, ν, µ ∈ R0 and for each function g ∈ Cp,q1 = {f ∈ Cp,q : f0 ∈ Cp,q} there exists a positive constant M(p, q, ν, µ) such that for all (x, y) ∈ R20 and n, m ∈ N we have
wp,q(x, y)|Aν,µn,m(g; x, y) − g(x, y)| ¬
M (p, q, ν, µ) kgx0kp,q
x + 1 n
12
+ kgy0kp,q
y + 1 m
12! .
Proof Pick (x, y) ∈ R20. For (t, s) ∈ R20 and g ∈ Cp,q1 we can write
g(t, s)− g(x, y) = Z t
x
gk0(k, s) dk +Z s y
gu0(x, u) du.
By the linearity of the operator Aν,µn,mwe obtain
|Aν,µn,m(g(t, s); x, y) − g(x, y)| = |Aν,µn,m(g(t, s) − g(x, y); x, y)| =
Aν,µn,m
Z t x
gk0(k, s) dk +Z s y
g0u(x, u) du; x, y ¬ Aν,µn,m
Z t x
g0k(k, s) dk ; x, y
+ Aν,µn,m Z s
y
gu0(x, u) du ; x, y
. Since
Z t x
gk0(k, s) dk
¬ kg0xkp,q
Z t x
1 wp,q(k, s)dk
¬ kg0xkp,q
1
wp,q(t, s)+ 1 wp,q(x, s)
|t − x|, we get
wp,q(x, y)Aν,µn,m
Z t x
g0k(k, s) dk ; x, y
¬
kgx0kp,qwp,q(x, y)
Aν,µn,m
|t − x|
wp,q(t, s); x, y
+ Aν,µn,m
|t − x|
wp,q(x, s); x, y
=
kg0xkp,qwp,q(x, y)Aµm
1
wq(s); y
Aνn
|t − x|
wp(t); x
+ 1
wp(x)Aνn(|t − x|; x)
=
kg0xkp,qwq(y)Aµm
1
wq(s); y
wp(x)Aνn
|t − x|
wp(t); x
+ Aνn(|t − x|; x)
. Using the H¨older inequality and Lemmas 2.3 - 2.5 we obtain
wp,q(x, y)Aν,µn,m Z t
x
g0k(k, s) dk ; x, y
¬
M (p, q, ν, µ)kgx0kp,q
x + 1 n
12 . Analogously we can write the following estimation
wp,q(x, y)Aν,µn,m Z s
y
g0u(x, u) du ; x, y
¬
M (p, q, ν, µ)kg0ykp,q
y + 1 m
12 ,
which completes the proof.
Theorem 3.2 For all p, q ∈ N0, ν, µ ∈ R0 and for each f ∈ Cp,q there exists a positive constant M(p, q, ν, µ) such that for all (x, y) ∈ R20 and n, m ∈ N we have
wp,q(x, y)|Aν,µn,m(f; x, y)−f(x, y)| ¬ M(p, q, ν, µ)ω f, Cp,q,
x + 1 n
12 ,
y + 1 m
12! .
Proof Pick (x, y) ∈ R20and h, d ∈ R+. We define the Steklov mean of f ∈ Cp,q as follows
fh,d(x, y) = 1 hd
Z h 0
Z d
0 f (x + u, y + v) du dv fh,d(x, y) = 1
hd Z x+h
x
Z y+d y
f (u, v) du dv.
Observe that
(fh,d)0x(x, y) = 1 hd
Z y+d y
f (x + h, v) dv− Z y+d
y
f (x, v) dv
!
=
1 hd
Z y+d
y (f(x + h, v) − f(x, v)) dv = 1 hd
Z d
0 (f(x + h, y + v) − f(x, y + v)) dv = 1
hd Z d
0
(∆h,vf (x, y)− ∆0,vf (x, y)) dv.
Similarly we can show that (fh,d(x, y))0y= 1
hd Z h
0
(∆u,df (x, y)− ∆u,0f (x, y)) du.
Therefore (fh,d)0x, (fh,d)0y ∈ Cp,q. Moreover,
fh,d(x, y) − f(x, y) = 1 hd
Z h 0
Z d
0 ∆u,vf (x, y) du dv, hence and definitons (3), (4) we get
kfh,d− fkp,q ¬ ω(f, Cp,q, h, d), k(fh,d)0xkp,q ¬ 1
h(k∆h,vfkp,q+ k∆0,vfkp,q) ¬ 2
hω(f, Cp,q, h, d), k(fh,d)0ykp,q ¬ 2
dω(f, Cp,q, h, d).
By the linearity of Aν,µn,m, Theorems 2.2 and 3.1 we obtain wp,q(x, y)|Aν,µn,m(f; x, y) − f(x, y)| ¬
wp,q(x, y)(Aν,µn,m(|f − fh,d|; x, y) + |Aν,µn,m(fh,d; x, y) − fh,d(x, y)| + |fh,d(x, y) − f(x, y)|) ¬ M(p, q, ν, µ)(kf−fh,dkp,q+k(fh,d)0xkp,qx+ 1
n
12
+k(fh,d)0ykp,qy+ 1 m
12
+kfh,d−fkp,q) ¬
M(p, q, ν, µ)ω(f, Cp,q, h, d)
2 +2
h
x+ 1 n
12 +2
d
y+ 1 m
12 ,
for (x, y) ∈ R20, n, m ∈ N and h, d ∈ R+. Setting, for (x, y) ∈ R20 and n, m ∈ N, h = x+1n 12 and d = y+1m 12 we get the desired estimation.
Theorem 3.2 implies the following corollaries.
Corollary 3.3 If ν, µ ∈ R0 and f ∈ Cp,q with some p, q ∈ N0, then for all (x, y) ∈ R20
n,m→∞lim Aν,µn,m(f; x, y) = f(x, y).
Moreover, the above convergence is uniform on every set [x1, x2] × [y1, y2] with 0 ¬ x1< x2, 0 ¬ y1< y2.
Corollary 3.4 For all α, β ∈ (0; 1], p, q ∈ N0 and for each f ∈ Lip(Cp,q, α, β)) = {f ∈ Cp,q: ω(f, Cp,q; t, s) = O(tα+sβ)} there exists a positive constant M(p, q, α, β) such that for all (x, y) ∈ R20 and n, m ∈ N we have
wp,q(x, y)|Aν,µn,m(f; x, y) − f(x, y)| ¬ M(p, q, α, β)
x + 1 n
α2 +
y + 1 m
β2! .
References
[1] M. Herzog, Approximation theorems for modified Szasz-Mirakjan operators in polynomial we- ight spaces, Le Matematiche54.1 (1999), 77-90.
[2] L. Rempulska, M. Skorupka, On some operators in weighted spaces of functions of two varia- bles, Ricerche Mat.48.1 (1999), 1-20.
[3] E. Wachnicki, Approximation by bivariate Mazhar-Totik operators, Comment. Math.50.2 (2010), 141-153.
[4] Z. Walczak, On certain modified Szasz-Mirakjan operators for functions of two variables, Demonstratio Math.33.1 (2000), 91-100.
[5] Z. Walczak, Approximation of functions of two variables by modified Szasz-Mirakyan opera- tors, Fasc. Math.34 (2004), 129-140.
[6] G. N. Watson, Theory of Bessel functions, Cambridge Univ. Press, Cambridge, 1966 Monika Herzog
Inistitute of Mathematics, Cracow University of Technology Warszawska 24, 31-155 Cracow, Poland
E-mail: mherzog@pk.edu.pl
(Received: 16.09.2011)