Rev. Anal. Num´er. Th´eor. Approx., vol. 38 (2009) no. 1, pp. 77–86 ictp.acad.ro/jnaat
SOME APPROXIMATION PROPERTIES OF MODIFIED SZASZ-MIRAKYAN-KANTOROVICH OPERATORS
GRZEGORZ NOWAK∗and ANETA SIKORSKA-NOWAK†
Abstract. In this paper we consider the modified Szasz-Mirakyan-Kantorovich operators for functionsfintegrable in the sense of Denjoy-Perron. Moreover, we estimate the rate of pointwise convergence ofMnf(x) at the Lebesgue-Denjoy pointsxoff.
MSC 2000. 41A25.
Keywords. Szasz-Mirakyan operator, rate of convergence, Lebesgue-Denjoy point.
1. PRELIMINARIES
In [6], Z. Walczak introduced and considered some approximation properties of the modifed Szasz-Mirakyan operators defined by
Mnf(x) =
∞
X
k=0
pn,k(x;an)f(k/bn) (x∈R0, n∈N),
where pn,k(x;an) = exp(−anx)(ank!x)k,R0 = [0;∞), N ={1,2, ...} and (an)∞1 , (bn)∞1 are given increasing and unbounded sequences of positive numbers, such that
(1) limn→∞ 1
bn = 0 and abn
n = 1 +O
1 bn
.
If an=bn =n, for all n∈N, then Mn become the classical Szasz-Mirakyan operator examined for continuous and bounded function in [4]. Some approx- imation properties of the operatorsMn, for continuous functionsf onR0, can be found e.g. in [1] and [6].
Denote byMn∗the Kantorovich type modification of operatorMnand define it for measurable functionf on R0 as:
(2) Mn∗f(x) =bn
∞
X
k=0
pn,k(x;an)
Z (k+1)/bn
k/bn
f(t)dt x∈R0, n∈N,
∗University of Marketing and Management, Ostroroga 9a 64-100 Leszno, Poland, e-mail:
†Faculty of Mathematics and Computer Science Adam Mickiewicz University, Umul- towska 87, 61-614 Pozna´n, Poland, e-mail: [email protected].
where the integral is taken in the sense of Lebesgue or Denjoy-Perron. Con- vergence Theorems for the classical Szasz-Mirakyan-Kantorovich operators (an = bn = n) are presented in [2], [3]. Some approximation properties of Mn∗f can be found in [7].
The aim of this paper is to examine the rate of the convergence of operators Mn∗f, mainly, at those pointsx∈R0 at which
limh→0 1 h
Z h 0
(f(x+t)−f(x)dt= 0.
The general estimate is expressed in terms of the quantity wx(δ;f) = sup0<|h|≤δ
1 h
Z h 0
(f(x+t)−f(x))dt
(δ >0).
Clearly, iff is locally integrable in the sense of Denjoy-Perron onR0, then limδ→0+wx(δ;f) = 0 for almost everyx.
In view of this property, we deduce that in some classes of functions, limn→∞Mn∗f(x) =f(x) almost everywhere.
Moreover, using some other properties of wx(δ;f), we present the estimate of the rate of Mn∗f(x) the pointwise convergence of in terms of the weighted moduli of continuity.
Throughout the paper, the symbol K(...), Kj(...), (j = 1,2, ...) will mean some positive constants, not necessarily the same at each occurrence, depend- ing only on the parameters indicated in parentheses.
2. AUXILIARY ESTIMATES
It is easy to see, that for every x∈R0 and for all integersn∈N , (3)
∞
X
k=0
pn,k(x;an) = 1,
(4)
∞
X
k=0
k bn −x
pn,k(x;an) =x
an
bn −1 ,
∞
X
k=0
k an −x
pn,k(x;an) = 0.
By properties (1), we have that exists a positive, absolute constants K, K1, such that
(5)
an
bn −1 ≤ Kb
n, a1
n ≤ bK
n and abn
n ≤K1
for all n∈N.
Forq ∈N0 =N ∪ {0},l∈N0,n∈N we define (6) Sq,l(n)(x;bn) =bn
∞
X
k=0
k bn −x
q
k an −x
l
pn,k(x;an).
Lemma 1. Suppose that q ∈ N and (an)∞1 ,(bn)∞1 are fixed. Then, for n∈N, x∈R0 we have
(7) S(n)q,0(x;bn) =xan q−2
X
j=0 q−1
j
1
bq−jn
Sj,0(n)(x;bn) +x an
bn −1
Sq−1,0(n) (x;bn),
(8) Sq,1(n)(x;bn) =x
q−1
X
j=0 q j
1
bq−jn
Sj,0(n)(x;bn),
Sq,2(n)(x;bn) = ax
n
q
X
j=0 q j
1
bq−jn Sj,0(n)(x;bn) (9)
+x2
q−1
X
j=1 q j
1
bq−jn
j−1
X
i=0 j i
1
bj−in
Si,0(n)(x;bn).
(Note that the symbol P−1
j=0...denotes zero.) Proof. In view of the definition (6)
Sq,0(n)(x;bn) =
∞
X
k=0
exp(−anx)(ank!x)k k
bn −x q
= abnx
n
∞
X
k=1
exp(−anx)(a(k−1)!nx)k−1 k
bn −x q−1
−xSq−1,0(n) (x;bn)
= abnx
n
∞
X
k=0
exp(−anx)(ank!x)k
k+1
bn −xq−1
−xSq−1,0(n) (x;bn)
=
q−1
X
j=0 q−1
j
anx
bq−1n
∞
X
k=0
pn,k(x;an) k
bn −x j
−xSq−1,0(n) (x;bn)
=xan
q−2
X
j=0 q−1
j
1
bq−jn
Sj,0(n)(x;bn) +x
an
bn −1
Sq−1,0(n) (x;bn).
This mean that (7) is true.
Next, assuming (6), we have Sq,1(n)(x;bn) =
∞
X
k=0
exp(−anx)(ank!x)k
k
bn −xq
k an −x
=
∞
X
k=0
exp(−anx)(anx)k!k+1a1
n
k+1 bn −x
q
−xSq,0(n)(x;bn)
=x
q
X
j=0 q j
1
bq−jn
Sj,0(n)(x;bn)−xSq,0(n)(x;bn)
=x
q−1
X
j=0 q j
1
bq−jn Sj,0(n)(x;bn).
So (8) is true. Hence we get Sq,2(n)(x;bn) =
∞
X
k=0
exp(−anx)(ank!x)k k
bn −x q
k an −x
2
=x
∞
X
k=0
exp(−anx)(ank!x)k
k+1
bn −xq
k+1 an −x
−xSq,1(n)(x;bn)
=x
q
X
j=0 q j
1
bq−jn
Sj,1(n)(x;bn) +a1
nSj,0(n)(x;bn)
−xSq,1(n)(x;bn)
=x
q−1
X
j=0 q j
1
bq−jn
Sj,1(n)(x;bn) +a1
nSj,0(n)(x;bn)
+ax
nSq,0(n)(x;bn).
Applying (8) the desired equality (9) follows.
Lemma 2. For every n∈N, q ∈N and q ∈R0 the following inequality is true
(10) |Sq,2(n)(x;bn)| ≤K(q)
x2(1 +x)q−1 1
b[(q+3)/2]n
+x(x+ 1) 1
bq+1n
,
where (an)∞1 ,(bn)∞1 are fixed.
Proof. Using (4), (7) and the method of induction one can easily verify that for all n∈N,q∈N,x∈R0 there holds
(11) |Sq,0(n)(x;bn)| ≤K(q)x(1 +x)q−1 1
b[(q+1)/2]n
Simple calculation, (3), (4) give us S0,2(n)(x;bn) =
∞
X
k=0
exp(−anx)(ank!x)k
k an −x2
= a1
n
∞
X
k=1
exp(−anx)(a(k−1)!nx)k
k an −x
−xS0,1(n)(x;bn)
=x
∞
X
k=0
exp(−anx)(ank!x)k k+1
an −x
−xS(n)0,1(x;bn)
= ax
nS0,0(n)(x;bn) = ax
n,
and
S1,2(n)(x;bn) =
∞
X
k=0
exp(−anx)(ank!x)k
k
an −x2
k bn −x
= b1
n
∞
X
k=1
exp(−anx)(a(k−1)!nx)k
k
an −x2
−xS0,2(n)(x;bn)
= xabn
n
∞
X
k=0
exp(−anx)(ank!x)k
k an −x
2
+a12 n
+ a2
n
k
an −x
−xS0,2(n)(x;bn)
= xabn
n
S(n)0,2(x;bn) +a12
nS0,0(n)(x;bn) + a2
nS0,1(n)(x;bn)
−xS0,2(n)(x;bn)
= xabn
n
x an +a12
n
−xax
n = ax2
n
an
bn −1 + ax
nbn. Applying (5) we obtained (10) for q= 1. Hence
S2,2(n)(x;bn) = ax
n
2
X
j=0 2 j
1
b2−jn
Sj,0(n)(x;bn) +x2 2b
n
1
bnS0,0(n)(x;bn)
= ax
n
1
b2nS0,0(n)(x;bn) +b2
nS1,0(n)(x;bn) +S2,0(n)(x;bn)
+2xb22
n S0,0(n)(x;bn).
Clearly, in view of (3), (5) and (11)
|S(n)2,2(x;bn)| ≤Kax
n
1 b2n +bx2
n +x(1 +x)b1
n
+2xb22
n
≤K1n
x2(1 +x)b1
n
1 an + b1
n
+x(1+x)a
nb2n
o
≤K2n
x2(1 +x)b12
n +x(1+x)b3 n
o .
Hence for q= 2 the inequality (10) is true.
Analogously we have (10) forq = 3 S3,2(n)(x;bn) =
=
x an
3
X
j=0 3 j
1
b3−jn
Sj,0(n)(x;bn) +x2
2
X
j=1 3 j
1
b3−jn
j−1
X
i=0 j i
1
bj−in
Si,0(n)(x;bn)
=
x an
1
b3nS0,0(n)(x;bn) + b32
nS1,0(n)(x;bn) +b3
nS2,0(n)(x;bn) +S3,0(n)(x;bn)
+x2 3b2 n
1
bnS0,0(n)(x;bn) +x2 3b
n
1
b2nS0,0(n)(x;bn) +b2
nS1,0(n)(x;bn)
≤ ax
n
1
b3n +Kxb13
n +x(1 +x)b12
n +x(1 +x)2 1b2 n
+K
x2 1b3
n +x3 1b3 n
≤K1x2(1+x)2
b3n +x(1+x)b4 n
.
Ifq ≥4 then, in view of (9),
|Sq,2(n)(x;bn)| ≤
≤ ax
n
q
X
j=0 q j
1
bq−jn
|Sj,0(n)(x;bn)|+x2
q−1
X
j=1
v 1
bq−jn
j−1
X
i=0 j i
1
bj−in
|Si,0(n)(x;bn)|
= ax
n
b1q
n|S0,0(n)(x;bn)|+ q
bq−1n
|S1,0(n)(x;bn)|+
q
X
j=2 q j
1
bq−jn
|Sj,0(n)(x;bn)|
+x2
q bq−1n
1
bn|S0,0(n)(x;bn)|+ q2 1
bq−2n
1
b2n|S0,0(n)(x;bn)|+b2
n|S1,0(n)(x;bn)|
+
q−1
X
j=3 q j
1
bq−jn
j−1
X
i=2 j i
1
bj−in
|Si,0(n)(x;bn)|
+1
bjn
|S0,0(n)(x;bn)|+j 1
bj−1n
|S1,0(n)(x;bn)|
≤ ax
n
b1q n + q
bq−1n
x
an
bn −1 +
q
X
j=2 q j
1
bq−jn
K(j)x(1 +x)j−1 1
b[(j+1)/2]n
+x2K1(q)
1 bqn + 1
bq−1n
x
an
bn −1
+
q−1
X
j=3 1 bq−jn
j−1
X
i=2 1 bj−1n
x(1 +x)i−1 1
b[(i+1)/2]n
+ 1
bjn
+x 1
bj−1n
an
bn −1
!
.
Consequently, from (5)
|Sq,2(n)(x;bn)| ≤
≤K(q)ax
n
1 bqn +b1q
nx+x(1 +x)q−1 1
b[(q+1)/2]n
+x2K(q)
1
bqn +b1q nx+
q−1
X
j=3 1 bq−jn
x(1 +x)j−2 1
b[j/2+1]n
+ 1
bjn
+x 1
bjn
≤K1(q)
x2(1 +x)q−1 1
b[(q+3)/2]n
+x2 1+x
bqn + 1+x
bq−1n + x(1+x)q−3
b[(q+1)/2]+1 n
+x(1+x)
bq+1n
≤K2(q)
x2(1 +x)q−1 1
b[(q+3)/2]n
+x(1+x)
bq+1n
,
and inequality (10) follows now immediately.
Identity (3), estimate (10) and the known Schwarz inequality lead to Lemma 3. Let q ∈N0, x∈R0 . Then, forn∈N
∞
X
k=0
k an −x
k bn −x
q+1
pn,k(x;an)≤ (12)
≤K(q)
x(1+x)q+1/2 bq/2+1n
+
√
x(1+x) bq+3/2n
,
where (an)∞1 ,(bn)∞1 , are fixed.
3. MAIN RESULT
In this Section, we consider only the points x∈[0;∞) at which wx(δ;f)<
∞ for all δ >0.
Theorem4. Letf :R0→Ris integrable in the Lebesgue or Denjoy-Perron sesnse on every compact interval contained in R0 and let n ∈ N, x ∈ R0. Given any number q∈N, we have
|Mn∗f(x)−f(x)| ≤K(q)
(1 +x)q+1/2+
1+x xbq+1n
1/2 (13)
∞
X
v=0
wx
√v+1 bn
1 (v+1)q.
Proof. For the sake of brevity we will write f(x+t)−f(x) = ϕx(t) and wx(δ;f) =wx(δ). In view of (3) we have
Mn∗f(x)−f(x) =bn
∞
X
k=0
pn,k(x;an)
Z (k+1)/bn
k/bn
(f(t)−f(x))dt
=bn
∞
X
k=0
pn,k(x;an)
Z (k+1)/bn−x k/bn−x
ϕx(t)dt
=bn
∞
X
k=1
(pn,k−1(x;an)−pn,k(x;an))
Z k/bn−x 0
ϕx(t)dt
−bnexp(−anx) Z −x
0
ϕx(t)dt.
It is easy to see that
x(pn,k−1(x;an)−pn,k(x;an)) =pn,k(x;an) k
an −x
.
Consequently
x(Mn∗f(x)−f(x)) =bn
∞
X
k=0
pn,k(x;an)
k
an −xZ k/bn−x 0
ϕx(t)dt.
Applying the obviously inequality
Rh
0 ϕx(t)dt
≤ |h|wx(|h|), we obtain x|Mn∗f(x)−f(x)| ≤bn
∞
X
k=0
pn,k(x;an)
k an −x
k bn −x
wx
k bn −x
≤
∞
X
v=0
Tv(n)(λ;x)wx((v+ 1)λ), whereλis an arbitrary positive number and
Tv(n)(λ;x) = X
vλ<|k/bn−x|≤(v+1)λ
bn
k an −x
k bn −x
pn,k(x;an).
We shall estimate the factors Tv(n)(λ;x) for v = 0 and v ≥ 1, respectively.
Clearly, in view of (12)
T0(n)(λ;x) ≤bn
∞
X
k=0
k an −x
k bn −x
pn,k(x;an)
≤K
x(1 +x)1/2+
x(1+x) bn
1/2 .
Next, if v≥1, we have, by Lemma 3, Tv(n)(λ;x) ≤ vbqnλq
∞
X
k=0
k an −x
k bn −x
q+1
pn,k(x;an)
≤ K(q)vqλq
x(1+x)q+1/2
bq/2n
+x1/2(1+x)1/2
bq+1/2n
.
Now, takingλ= 1/b1/2n , we easily get x|Mn∗f(x)−f(x)| ≤
x(1 +x)q+1/2+x1/2(1+x)1/2
b(q+1)/2n
· Kwx
√1 bn
+K(q)
∞
X
v=1
wx
v+1√ bn
1 (v+1)q
! .
This last relation is equivalent to (13).
Remark 5. Let Dloc∗ (R) be the class of all functions integrable in the Denjoy-Perron sense on every compact interval contained in R. Clearly, if f ∈Dloc∗ (R), then the function
F(x) = Z x
0
f(t)dt
is ACG∗ on every [a;b]⊂R0 and F0(x) =f(x) almost everywhere [5].
Consequently,
(14) limδ→0+wx(δ;f) = 0 a.e. on R.
Suppose that f ∈D∗loc(R) and that
||f|| ≡sup−∞<v<∞
sup0≤u≤1
Z v+u v
f(t)dt
<∞.
The operators Mn∗f(x) are well-defined for all n ∈ N. This follows at once from the inequality
Z (k+1)/n k/n
f(t)dt
≤ ||f||.
In view of (14), given any > 0 there is a δ0 > 0 such that wx(δ;f) <
whenever 0< δ≤δ0. In case δ≥δ0 we have wx(δ;f)< +|f(x)|+ supδ0≤|h|≤δ
1 h
Z h 0
f(x+t)dt
≤+|f(x)|+δ1
0||f||
provided that δ ≤1. If δ > 1, then putting µ = [δ] + 1 and xj = x+jh/µ (j = 0,1, ..., µ), we get
Z h 0
f(x+t)dt
≤
µ−1
X
j=0
Z xj+1
xj
f(t)dt
≤µ||f|| ≤2δ||f||.
Hence
wx(δ;f)< +|f(x)|+δ1
0(1 + 2δ)||f|| for all δ >0.
This inequality and the condition (14) ensure that the right-hand side of the estimate (13) (with arbitraryq ≥3) converges to zero asn→ ∞.
Letm ∈N0. Denote byCm(R0) the class of all measurable functions f on R0, such that
||f||m = supx∈R1+x|f(x)|2m <∞.
It is easy to see, that operators Mn∗f are well-defined for every function f ∈ Cm(R0).
Further, for continuousf ∈Cm(R0), let us introduce the weighted modulus of continuity
ω(δ;f)m = sup|h|≤δ||f(·+h)−f(·)|| (δ >0).
Theorem6. Letm∈N0,x∈R0,f ∈Cm(R0)and(an)∞1 , (bn)∞1 are fixed.
Then
(1 +x2m)−1|Mn∗f(x)−f(x)| ≤
≤K(m)
(1 +x)2m+3+ 1+xx 1/2 1 bm+5/4n
ω(b−1/2n ;f)m, for all n∈N.
Proof. It is easy to see, that for anyx∈R0,r∈N
|f(x+rt)−f(x)| ≤
r−1
X
v=0
|f(x+vt+t)−f(x+vt)|
1+(x+vt)2m (1 + (x+vt)2m)
≤ω(|t|;f)m
r−1
X
v=0
(1 + (|x|+v|t|)2m)
≤(1 + (|x|+ (r−1)|t|)2mrω(|t|;f)m. Therefore, for any x∈R,δ >0,r ∈N
wx(rδ;f)≤(1 + (2x)2m+ (2(r−1)δ)2m)rω(δ;f)m.
This inequality and Theorem 4, with q= 2m+ 5/2, lead to Theorem 6.
Acknowledgement. We are grateful to the referee’s valuable suggestions.
REFERENCES
[1] Inspir, N.and Atakut, C., Approximation by modified Szasz-Mirakyan operators on weighted spaces, Proc. Indian Acad. Sci.,112 (4), pp. 571–578, 2002.
[2] Herman, T.,On the Szasz-Mirakyan operator, Acta Math. Acad. Sci. Hung.,32 (1-2), pp. 163–173, 1978.
[3] Totik, V., Approximation by Szasz-Mirakyan-Kantorovich operators in Lp (p > 1), Analysis Math.,9, pp. 147–167, 1983.
[4] Totik, V.,Uniform approximation by Szasz-Mirakyan type operators, Acta Math. Hung., 41, pp. 241–307, 1983.
[5] Saks, S.,Theory of the Integral, New York, 1937.
[6] Walczak, Z.,On certain modified Szasz-Mirakyan operators for functions of two vari- ables, Demonstratio Math.,33 (1), pp. 91–100, 2000.
[7] Walczak, Z.,On modified Szasz-Mirakyan operators, Novi Sad J. Math.,33 (1), pp. 93–
107, 2003.
Received by the editors: January 20, 2009.