C O L L O Q U I U M M A T H E M A T I C U M
VOL. 79 1999 NO. 2
A REMARK ON A MODIFIED SZ ´ ASZ–MIRAKJAN OPERATOR
BY
GUANZHEN Z H O U (NINGBO
ANDHANGZHOU)
ANDSONGPING Z H O U (NANGCHONG
ANDHANGZHOU)
Abstract. We prove that, for a sequence of positive numbers δ(n), if n
1/2δ(n) 6→ ∞ as n → ∞, to guarantee that the modified Sz´ asz–Mirakjan operators S
n,δ(f, x) converge to f (x) at every point, f must be identically zero.
1. Introduction. Let C
αbe the set of all continuous functions on [0, ∞) satisfying |f (t)| ≤ M t
αtfor some real numbers M > 0 and α > 0. For f ∈ C
αand x ∈ [0, ∞), the well-known Sz´ asz–Mirakjan operator is defined by S
n(f, x) =
∞
X
k=0
f k n
e
−nx(nx)
kk! =:
∞
X
k=0
f k n
p
k(nx).
From Hermann [2], we know that for f ∈ C
α, S
n(f, x) converges to f (x) uniformly on any closed subset of [0, ∞), hence in particular at every point x in [0, ∞). At the same time, Hermann also pointed out that C
α, for all α > 0, are the largest sets, in the usual sense, which guarantee S
n(f, x) to exist.
For computational reasons, Gr´ of [1] and Lehnhoff [3] suggested using a partial sum of S
n(f, x) (which only has a finite number of terms depending upon n and x) to approximate f (x). Let δ = δ(n) be a sequence of positive numbers. Lehnhoff examined the operator
S
n,δ(f, x) =
[n(x+δ)]
X
k=0
f k n
p
k(nx),
and he proved that, for all f in C
αsatisfying |f (t)| ≤ M
1+ M
2t
2mfor some positive numbers M
1, M
2and some natural number m, S
n,δ(f, x) converges to f (x) at every point on [0, ∞) if
(1) lim
n→∞
n
1/2δ(n) = ∞.
1991 Mathematics Subject Classification: Primary 41A36.
Key words and phrases: modified Sz´ asz–Mirakjan operator.
Research supported in part by National and Provincial Natural Science Foundations, and by State Key Laboratory of Southwest Institute of Petroleum.
[157]
158
G. Z. Z H O U AND S. P. Z H O URecently, Sun [4] showed that the condition (1) is a sharp necessary and sufficient condition for S
n,δ(f, x) to converge pointwise to f (x) for f in C
α. More precisely, he showed that for f ∈ C
αthe condition (1) is sufficient for S
n,δ(f, x) to converge to f (x) uniformly on any closed subset of [0, ∞), and he also proved that if (1) does not hold, then for the function f
0(x) = x
αx∈ C
α, S
n,δ(f
0, x) does not converge to f
0(x) at some point x.
A natural question is whether (1) can be weakened if we consider a subset of C
α(for example, the subset that Lehnhoff studied). Our result exhibits a surprising phenomenon that if (1) does not hold, then to guarantee that S
n,δ(f, x) converges to f (x) at every point, f must be identically zero.
2. Result and proof. In what follows, we always use C to indicate a positive constant, whose value may be different in different situations.
Theorem. Let δ = δ(n), n = 1, 2, . . . , be a sequence of positive numbers such that n
1/2δ(n) 6→ ∞ as n → ∞, and assume that f ∈ C
α. If f (x
0) 6= 0 for some x
0∈ [0, ∞), then S
δ(f, x
0) 6→ f (x
0) as n → ∞.
P r o o f. Suppose n
1/2δ(n) 6→ ∞ as n → ∞. Without loss of generality, there exists a constant A > 0 and a sequence {n
j} of positive integers such that n
1/2jδ(n
j) ≤ A, and f (x
0) > 0 for x
0∈ (0, ∞), say. There are M
0> 0 and ε
0> 0 such that f (x) > M
0for all x ∈ (x
0− ε
0, x
0+ ε
0) ⊂ (0, ∞).
Write
f
+(x) =
12(f (x) + |f (x)|), f
−(x) =
12(f (x) − |f (x)|).
Then
R
n(f, x) := S
n(f, x) − S
n,δ(f, x) = R
n(f
+, x) + R
n(f
−, x)
since R
nis a linear operator. At the same time, noting that R
nis also a positive operator, we calculate that
R
nj(f
+, x
0) =
∞
X
k=[nj(x0+δ)]+1
f (k/n
j)p
k(n
jx
0)
≥ X
njx0+An1/2j +1≤k≤njx0+2An1/2j +2
f (k/n
j)p
k(n
jx
0).
For n
jx
0+ An
1/2j+ 1 ≤ k ≤ n
jx
0+ 2An
1/2j+ 2 and sufficiently large j, k/n
j∈ (x
0− ε
0, x
0+ ε
0), so that
R
nj(f
+, x
0) ≥ M
0X
njx0+An1/2j +1≤k≤njx0+2An1/2j +2
p
k(n
jx
0)
≥ AM
0n
1/2jp
[njx0+2An1/2j ]
(n
jx
0)
MODIFIED SZ ´ASZ–MIRAKJAN OPERATOR
159
by the monotonicity of {p
k(nx
0)} for k ≥ nx
0. It is easy to obtain p
[njx0+2An1/2j ]
(n
jx
0) ≥ C(n
jx
0)
−1/2e
−4A2/x0from Stirling’s formula, hence
R
nj(f
+, x
0) ≥ CAM
0x
−1/20e
−4A2/x0> 0, that is,
(2) R
nj(f
+, x
0) 6→ 0 as j → ∞.
For R
n(f
−, x
0), we see that
R
n(f
−, x
0) = X
k/n−x0≥ε0
f (k/n)p
k(nx
0)
in view of f
−(x) = 0 for x ∈ (x
0− ε
0, x
0+ ε
0). Thus for any given ε > 0, there is an N > 0 such that
∞
X
k=N +1
f (k/n)p
k(nx
0) < ε.
Similarly to the standard proof of the Korovkin theorem, we have
N
X
k=[nx0+nε0]+1
f (k/n)p
k(nx
0)
≤ max
0≤t≤N/n
|f (t)| X
k/n−x0≥ε0