Seria I: PRACE MATEMATYCZNE XLV (1) (2005), 57-70
Agnieszka Tyliba, Eugeniusz Wachnicki
On some class of exponential type operators
Abstract. Starting from a differential equation ∂t∂W (λ, t, u) = λ(u−t)p(t) W (λ, t, u) − βW (λ, t, u) for the kernel of an operator Sλ(f, t) = RB
AW (λ, t, u)f (u)du with the normalization condition RB
AW (λ, t, u)du = 1 we prove some properties which are similar to properties proved by Ismail and May for the exponential operators. In particular, we show that all these operators are approximation operators. Moreover, a method of determining Sλfor a given function p is introduced.
2000 Mathematics Subject Classification: 41A36, 41A25, 35K15.
Key words and phrases: exponential operators, Voronovskaya type theorem, rate of convergence, limit value problems.
1. Introduction. C. R. May in [3], M. E. Ismail and C. R. May in [2] studied family of exponential operators. They considered the integral S
λdefined by (1) S
λ(f, t) =
Z
B AW (λ, t, u)f (u) du, −∞ ≤ A < B ≤ +∞
under the following assumptions: a kernel W is a positive function, satisfying the following homogenous partial differential equation
(2) ∂
∂t W (λ, t, u) = λ
p(t) W (λ, t, u)(u − t), λ ∈ R, u, t ∈ (A, B), p is analytic and positive for t ∈ (A, B) and
(3) S
λ(e
0, t) = 1, t ∈ (A, B),
where e
0(u) = 1 for u ∈ (A, B). They presented some well-known operators satis-
fying (2). For example, the Bernstein polynominals and operators of Szasz, Post-
Widder, Gauss-Weierstrass and Baskakov. The above operators satisfy the condition
S
λ(e
1, t) = e
1(t), where e
1(u) = u for u ∈ (A, B).
Our purpose is to extend the results of May and Ismail to a family of operators, in which there are operators S
λsuch that S
λ(e
1, t) 6= e
1(t). We investigate a similar family of operators. Instead of the equation (2), however, we consider the following
(4) ∂
∂t W (λ, t, u) = λ
p(t) W (λ, t, u)(u − t) − βW (λ, t, u),
where β is a non-negative real number, λ ∈ R, u, t ∈ (A, B), p is analitic and positive for t ∈ (A, B). For these operators we obtain similar results as May and Ismail for exponential operators. We state some estimates of the rate of convergence of S
λ. We also prove the Voronovskaya type theorem for these operators. In section 4 we give examples of operators satisfying (4) and on the strength of one of these we define the function U which is a solution of some limit problem.
Let C(A, B) be a set of all continuous, real-valued functions on (A, B). We define the space (C
N, || ||
CN) as follows
C
N= n
f ∈ C(A, B) : ∃ M > 0 ∀ t ∈ (A, B) |f (t)| ≤ M e
N |t|o
and ||f ||
CN= sup n
e
−N |t||f (t)| : t ∈ (A, B) o .
2. Preliminaries. In this section we give some properties of the above opera- tors, which we use in the proofs of the main theorems. The proofs of these lemmas follow by the same method as the proofs of the appropriate properties of exponential operators in [2].
Lemma 2.1 For each λ ∈ R and t ∈ (A, B) we have S
λ(e
1, t) = e
1(t) + βp(t)
λ and
S
λ(e
2, t) = e
2(t) + p(t) + 2βtp(t)
λ + βp(t)p
0(t) + β
2p
2(t)
λ
2,
where e
2(u) = u
2for u ∈ (A, B).
Lemma 2.2 For every n ∈ N
∗we get S
λ(e
n, t) = e
n(t) +
n
X
k=1
ϕ
k,n(t)
λ
k, t ∈ (A, B), where e
n(u) = u
nfor u ∈ (A, B), the function ϕ
k,nis defined by
ϕ
k,n(t) = tϕ
k,n−1(t) + p(t)
βϕ
k−1,n−1(t) + ∂
∂t ϕ
k−1,n−1(t)
for 1 ≤ k ≤ n and we put
ϕ
0,n= e
n, ϕ
n,n−1≡ 0.
Let us denote
(5) A
m(λ, t) = λ
mS
λ((u − t)
m, t).
Lemma 2.3
A
0(λ, t) = 1, A
1(λ, t) = βp(t) and
(6) A
m+1(λ, t) = p(t) d
dt A
m(λ, t) + βp(t)A
m(λ, t) + mλp(t)A
m−1(λ, t) for m ≥ 1.
Lemma 2.4 For all m ∈ N the operators A
2mand A
2m+1are polynomials in λ, whose the degree is m. Moreover, the coefficient of the term λ
mis
c
1mp
m(t) for A
2mand
c
2mp
m+1(t) + c
3mp
m(t)p
0(t) for A
2m+1, where c
1m, c
2mand c
3mare constants.
Lemma 2.5 If
q(t) = Z
tc
dv
p(v) , c ∈ (A, B) and
g(q(t)) = q(g(t)) = t, then the following equality holds
∞
X
m=0
S
λ((u − t)
m, t) x
mm! = exp
(
−xt + λ
Z
g(
q(t)+xλ)
t
vdv p(v) + β
g q(t) + x
λ
− t )
.
3. Main results.
Theorem 3.1 Let N > 0 and a, b be such that A < a < b < B, then for a sufficiently large λ we have
||S
λ(e
N |u|, t)||
C[a,b]< ∞,
where C[a, b] is the set of all continuous, real-valued functions on [a, b] with the
suppremum norm.
Proof By the definition of S
λwe have S
λ(e
N u, t) = e
N tZ
B AW (λ, t, u)e
N (u−t)du
= e
N t∞
X
m=0
N
mm! S
λ((u − t)
m, t).
From Lemma 2.5 we obtain S
λ(e
N u, t) = exp
( λ
Z
g(
q(t)+Nλ)
t
θ p(θ) dθ
) exp
β
g
q(t) + N λ
− t
.
Letting λ −→ ∞ and the application of de L’Hospital’s formula implies that
(7) lim
λ→+∞
S
λ(e
N u, t) = e
N tuniformly in C[a, b].
Hence for a sufficiently large λ
||S
λ(e
N u, t)||
C[a,b]< ∞.
Similarly,
||S
λ(e
−N u, t)||
C[a,b]< ∞.
S
λis a positive and linear operator, so we have
||S
λ(e
N |u|, t)||
C[a,b]≤ ||S
λ(e
N u, t)||
C[a,b]+ ||S
λ(e
−N u, t)||
C[a,b]< ∞, because e
N |u|< e
N u+ e
−N u. This completes the proof.
Theorem 3.2 Suppose that m, η are positive numbers, N is a non-negative number and [a, b] ⊂ (A, B). Then
(8) Z
|u−t|≥η
W (λ, t, u)e
N udu = O(λ
−m) uniformly on [a, b] as λ → +∞.
Proof By Cauchy-Schwarz’s inequality, (7) and Lemma 2.4 we have Z
|u−t|≥η
W (λ, t, u)e
N udu
≤ Z
|u−t|≥η
W (λ, t, u) du Z
|u−t|≥η
W (λ, t, u)e
2N udu
!
12≤ η
−4mλ
−4mA
4m(λ, t)
12S
λ(e
2N u, t)
12= O(λ
−m) for λ → +∞,
and (8) is proved.
Theorem 3.3 Let f ∈ C
N. Then
(9) lim
λ→∞
S
λ(f, t) = f (t), uniformly on every [a, b] ⊂ (A, B).
Proof We have
|S
λ(f, t) − f (t)| ≤ Z
|u−t|<δ
|f (u) − f (t)|W (λ, t, u) du +
Z
|u−t|≥δ
|f (u) − f (t)|W (λ, t, u) du
= I
1+ I
2for every δ > 0. Let > 0. By the continuity of f in [a, b] there exists δ > 0 such that |f (u) − f (t)| < for |u − t| < δ. Hence
I
1≤ 2 Z
|u−t|<δ
W (λ, t, u) du ≤ 2
Z
B AW (λ, t, u) du = 2 . On the other hand, from (8) for a sufficiently large λ
I
2≤ 2||f ||
CNZ
|u−t|≥δ
W (λ, t, u)e
N udu ≤ M ||f ||
CN1 λ <
2 ,
which proves the theorem.
Theorem 3.4 Let ξ ∈ (A, B). If f ∈ C
Nand f
00(ξ) exists, then
(10) lim
λ→∞
λ(S
λ(f, ξ) − f (ξ)) = βp(ξ)f
0(ξ) + 1
2 p(ξ)f
00(ξ).
Proof By Taylor’s formula, (3) and Lemma 2.1
S
λ(f, ξ) = f (ξ) + f
0(ξ)S
λ(e
1, ξ) − ξf
0(ξ) + 1
2 f
00(ξ)S
λ(e
2, ξ)
−ξf
00(ξ)S
λ(e
1, ξ) + 1 2 ξ
2f
00(ξ) +
Z
B AW (λ, ξ, u)(u, ξ)(u − ξ)
2du.
Lemma 2.1 now implies
S
λ(f, ξ) = f (ξ) + βp(ξ)f
0(ξ)
λ + 1
2
p(ξ)f
00(ξ)
λ + 1
2
βp(ξ)p
0(ξ)f
00(ξ) λ
2+ 1
2
β
2p
2(ξ)f
00(ξ) λ
2+ I.
Hence
λ (S
λ(f, ξ) − f (ξ)) = βp(ξ)f
0(ξ) + 1
2 p(ξ)f
00(ξ) + 1 2
βp(ξ)p
0(ξ)f
00(ξ) λ + 1
2
β
2p
2(ξ)f
00(ξ)
λ + Iλ.
On the other hand,
I ≤ Z
BA
W (λ, t, u)
2(u, ξ) du
!
12Z
BA
W (λ, t, u)(u − ξ)
4du
!
12. Applying Theorem 3.3 we see that
lim
λ→∞
Z
B AW (λ, t, u)
2(u, ξ) du = 0.
Therefore
Z
B AW (λ, t, u)(u − ξ)
4du
!
12= A
4(λ, ξ) λ
4 12= O(λ
−1)
by Lemma 2.4. Thus I = o(λ
−1), and (10) is proved.
Before the next theorem we recall some notations. Let < a, b > denote the interval [a, b] or (a, b) or (a, b] etc. If f :< a, b >−→ R, then
ω(f ; h, a, b) = sup {|f (t) − f (t + δ)|; t, t + δ ∈< a, b >, |δ| ≤ h}
and
ω
2(f ; h, a, b)
= sup {|f (t) − 2f (t + δ) + f (t + 2δ)|; t, t + 2δ ∈< a, b >, |δ| ≤ h}
for h > 0.
Theorem 3.5 Suppose that f ∈ C
Nand A < a < a
1< b
1< b < B. Then for every m > 0 there exists a constant K
msuch that
||S
λ(f, ·) − f ||
C[a1,b1]≤ K
mh
λ
−12ω(f ; λ
−12, a, b) + ω
2(f ; λ
−12, a, b) + λ
−m||f ||
CNi . Proof Let δ > 0 and δ ≤
12min{a
1− a, b − b
1}. Define
g
δ(x) = 1 2δ
2Z
δ2−δ2
Z
δ2−δ2
[f (x + u + v) + f (x − u − v)] du dv.
Since S
λis linear we get
kS
λ(f, ·) − f k
C[a1,b1]≤ kS
λ(f − g
δ, ·)k
C[a1,b1]+ kS
λ(g
δ, ·) − g
δk
C[a1,b1]+ kg
δ− f k
C[a1,b1]= I
1+ I
2+ I
3.
Let η =
12min{a
1− a, b − b
1}, then for x ∈ [a
1− η, b
1+ η] we get
|f (x) − g
δ(x)| =
f (x) − 1 2δ
2Z
δ2−δ2
Z
δ2−δ2
[f (x + u + v) + f (x − u − v)] du dv
=
1 2δ
2Z
δ2−δ2
Z
δ2−δ2
[f (x + u + v) − 2f (x) + f (x − u − v)] du dv
.
As |u + v| ≤ δ, the above equality shows that
(11) |f (x) − g
δ(x)| ≤ 1
2 ω
2(f ; δ, a, b).
Thus
(12) I
3≤ 1
2 ω
2(f ; δ, a, b).
The estimation of I
1follows from (11) and (8). Indeed, for t ∈ [a
1, b
1] we have
|S
λ(f − g
δ, t)| ≤ Z
|u−t|<η
W (λ, t, u)|f (u) − g
δ(u)| du +
Z
|u−t|≥η
W (λ, t, u)|f (u) − g
δ(u)| du
= L
1+ L
2.
Since |u − t| < η and η =
12min{a
1− a, b − b
1}, (11) yields
|f (u) − g
δ(u)| ≤ 1
2 ω
2(f ; δ, a, b).
Hence
L
1≤ 1
2 ω
2(f ; δ, a, b) Z
|u−t|<η
W (λ, t, u) du
≤ 1
2 ω
2(f ; δ, a, b).
Moreover,
L
2= Z
|u−t|≥η
W (λ, t, u)e
N u|(f − g
δ)(u)|e
−N udu .
As f − g
δ∈ C
Nwe have |(f − g
δ)(u)|e
−N u≤ 4kf k
CN. We conclude from (8) that L
2≤ 4M λ
−mkf k
CN,
and finally that
(13) I
1≤ M
1(ω
2(f ; δ, a, b) + λ
−mkf k
CN).
To estimate I
2we first compute g
0δ(x) = 1
2δ
2Z
δ2−δ2
[f (x − u − δ
2 ) − f (x − u + δ
2 )] − [f (x + u − δ
2 ) − f (x + u + δ 2 )] du.
Hence for x ∈ [a
1− η, b
1+ η] we have
(14) kg
δ0k
C[a1−η,b1+η]≤ δ
−1ω(f ; δ, a, b).
Moreover,
g
00δ(x) = δ
−2[f (x − δ) − 2f (x) + f (x + δ)]
and
(15) kg
δ00k
C[a1−η,b1+η]≤ δ
−2ω
2(f ; δ, a, b).
Condition (3) now shows that
|S
λ(g
δ, t) − g
δ(t)| =
Z
B AW (λ, t, u)(g
δ(u) − g
δ(t)) du
, t ∈ [a, b].
By the above and by Taylor’s formula we have
|S
λ(g
δ, t) − g
δ(t)| ≤
g
0δ(t) Z
BA
W (λ, t, u)(u − t) du +
Z
|u−t|<η
W (λ, t, u)g
00δ(ξ)(u − t)
2du +
Z
|u−t|≥η
W (λ, t, u)g
00δ(ξ)(u − t)
2du
= J
1+ J
2+ J
3.
From Lemma 2.4, (5) and (14) for t ∈ [a
1− η, b
1+ η] we obtain J
1= |g
δ0(t)λ
−1A
1(λ, t)| ≤ M λ
−1δ
−1ω(f ; δ, a, b).
On the other hand, (5), Lemma 2.4 and (15) show that
J
2≤ δ
−2ω
2(f ; δ, a, b)|λ
−2A
2(λ, t)| ≤ Kλ
−1δ
−2ω
2(f ; δ, a, b), t ∈ [a
1, b
1].
It is clear that g
00δ∈ C
Nand kg
δ00k
CN≤ 4kf k
CN. From this, Cauchy-Schwarz’s inequality and Lemma 2.4 we deduce that
J
3≤ 4kf k
CNZ
|u−t|≥η
W (λ, t, u)e
2Ndu
!
12λ
−4A
4(λ, t)
12≤ Lλ
−mkf k
CN, and finally that
(16) I
2≤ M
2[δ
−1λ
−1ω(f ; δ, a, b) + δ
−2λ
−1ω
2(f ; δ, a, b) + λ
−mkf k
CN].
Put δ = λ
−12into (12), (13) and (16). This ends the proof.
Let 0 < α ≤ 1 and A ≤ a < b ≤ B. The Lipschitz class Lip(α, a, b) is defined by
Lip(α; a, b) = {f : ω(f ; h, a, b) ≤ M h
α}.
From Theorem 3.5 we have
Corollary 3.6 If A < a < a
1< b
1< b < B and f ∈ Lip(α; a, b), then kS
λ(f, ·) − f k
C[a1,b1]≤ M λ
−α2.
Theorem 3.7 Let f ∈ C
0, 0 < α ≤ 1. If kS
λ(f, ·) − f k
C(A,B)≤ M λ
−α, then
f ∈ Lip(α, A, B).
We divide the proof into a sequence of the following lemmas.
Lemma 3.8 Let f ∈ C
0, 0 < α ≤ 1. If kS
λ(f, ·) − f k
C(A,B)≤ M λ
−α, then there exists M
0> 0 such that
ω(f ; h, A, B) ≤ M
0[λ
−α+ hλω(f ; λ
−1, A, B)]
for 0 < h ≤ 1 and λ > 1.
Proof Let x, y ∈ (A, B) be such that |y − x| ≤ h. Then
|f (y) − f (x)| ≤ |f (y) − S
λ(f, y)| + |S
λ(f, y) − S
λ(f, x)| + |S
λ(f, x) − f (x)|
≤ 2M λ
−α+ P.
It remains to estimate the term P . Let us first examine |S
λ0(f, t)|
|S
λ0(f, t)| =
Z
B A∂
∂t W (λ, t, u)f (u) du
=
λ p(t)
Z
B AW (λ, t, u)(u − t)f (u) du − β Z
BA
W (λ, t, u)f (u) du . Since
λ p(t)
Z
B AW (λ, t, u)(u − t)f (t) du = f (t)
p(t) A
1(λ, t) = βf (t), we have
|S
0λ(f, t)| ≤
λ p(t)
Z
B AW (λ, t, u)(u − t)[f (u) − f (t)] du
+ |β[f (t) − S
λ(f, t)]|
≤ λ
p(t) Z
BA
W (λ, t, u)|u − t|ω(f ; λ|u − t|λ
−1, A, B) du + βM λ
−α≤ λM
1Z
BA
W (λ, t, u)|u − t|[λ|u − t| + 1]ω(f ; λ
−1, A, B) du + βM λ
−α≤ λM
1ω(f ; λ
−1, A, B)
"
A
2(λ, t)
λ +
Z
B AW (λ, t, u)|u − t| du
#
+ βM λ
−α.
From Cauchy-Schwarz’s inequality and Lemma 2.4 we conclude that
|S
λ0(f, t)| ≤ M
2λ
−α+ λω(f ; λ
−1, A, B) . We are now in a position to estimate P . We have
P ≤
Z
y x|S
λ0(f, t)|dt
≤ M
0λ
−α+ hλω(f ; λ
−1, A, B) ,
and the lemma follows.
Lemma 3.9 Let f ∈ C
0and 0 < α ≤ 1. If
ω(f ; h, A, B) ≤ M
0[λ
−α+ hλω(f ; λ
−1, A, B)]
for 0 < h ≤ 1 and λ > 1, then f ∈ Lip(α, A, B).
Proof It is sufficient to show that ω(f ; h, A, B) ≤ M
00h
αfor 0 < h < 1, where M
00is a positive constant. Let K > 1 be such that 2M
0< K
1−α. Choose M
1= max{ω(f, 1, A, B); 2M
0K
α}. Define h
n= K
−n, n = 1, 2, . . ..
By induction it is easy to check that for every positive integer n (17) ω(f ; h
n, A, B) ≤ M
1h
αn.
On the other hand, for every 0 < h ≤ 1 exists an integer n > 0 such that h
n< h ≤ h
n−1. From this
ω(f ; h, A, B) ≤ ω(f ; h
n−1, A, B) ≤ M
1h
αn−1= M
1K
αh
αn≤ M
00h
α,
and the lemma is proved.
Theorem 3.10 A kernel W can be obtained by the partial differential equation (4) and the condition (3).
Proof Let W satisfy (4) and (3). We define ξ as follows ξ(λ, t, u) = exp
−λ Z
tc
u − θ p(θ) dθ + βt
W (λ, t, u).
Then ∂ξ(λ, t, u)
∂t = 0,
hence ξ(λ, t, u) is depending only on λ and u. On the other hand,
(18) W (λ, t, u) = exp
λ
Z
t cu − θ p(θ) dθ − βt
C(λ, u).
From (3) we have exp
λ
Z
t cθ
p(θ) dθ + βt
= Z
BA
exp {λuq(t)} C(λ, u)du and
(19) exp
( λ
Z
g(t) cθ
p(θ) dθ + βg(t) )
= Z
BA
exp {λut} C(λ, u)du.
This and (18) give W .
4. Examples.
Example 4.1 Let p(t) ≡ 1, c = 0, A = −∞, B = +∞. Then q(t) = g(t) = t. From (19) we have
e
λt22 +βt= Z
∞−∞
e
λutC(λ, u)du.
Hence
e
λ2(
t+β2)
2= e
β22λZ
∞−∞
e
λutC(λ, u)du.
Put z = t +
β2, then we obtain e
λz22= e
β22λZ
∞−∞
e
λuze
−uβC(λ, u)du.
From this and [1] we conclude that
e
β22λ−uβC(λ, u) = r λ
2π e
−λu22. Therefore
C(λ, u) = r λ
2π e
−λu22 +uβ−β22λ. Applying (18) we get
W (λ, t, u) = r λ
2π e
−λ(u−t)22 +β(u−t)−β22λ. Combining this with (1) we obtain
S
λ(f, t) = r λ
2π Z
∞−∞
e
−λ(u−t)22 +β(u−t)−β22λf (u)du
= r λ
2π Z
∞−∞
e
−λ
(
u−t−βλ)
22
f (u)du.
Now consider the function
(20) U (x, t) :=
Z
∞−∞
K(u, t, x)f (u)du, where
K(u, t, x) = 1 2 √
πt e
−(u−x−2βt)24t.
Theorem 4.2 If f ∈ C
N, then the function U given by (20) belongs to C
∞[Ω], where Ω = {(x, t) : x ∈ R, t > 0}. Moreover, U is a solution of the equation
∂
2U
∂x
2+ 2β ∂U
∂x = ∂U
∂t
in Ω and lim
t→0U (x, t) = f (x).
Proof From Theorem 3.3 we conclude that lim
t→0U (x, t) = f (x).
Let t
0, T
0, x
0, X
0∈ R be such that 0 < t
0< T
0and x
0< X
0. Consider the set Ω(t
0, T
0, x
0, X
0) = {(x, t) : x
0≤ x ≤ X
0, t
0≤ t ≤ T
0}.
By induction we deduce that the integral (21)
Z
∞−∞
∂
n+m∂t
n∂x
mK(u, t, x)f (u) du is a linear combination of integrals of the form:
(22)
Z
∞−∞
β
q(u − x)
rt
s+12e
−(u−x−2βt)24tf (u) du, where q, r, s ∈ N.
Now we prove that the integral (22) is uniformly convergent in Ω(t
0, T
0, x
0, X
0).
Note that for any p, k, l ∈ N from the definition of the norm in C
Nwe have Z
∞−∞
β
p(u − x)
kt
l+12e
−(u−x−2βt)24tf (u)
du
≤ β
pkf k
CNZ
∞−∞
(u − x)
kt
l+12e
−(u−x−2βt)24t +N udu.
On the other hand, for (x, t) ∈ Ω(t
0, T
0, x
0, X
0) we have β
pkf k
CNZ
∞−∞
(u − x)
kt
l+12e
−(u−x−2βt)24t +N udu
≤ M β
pkf k
CNZ
∞−∞
(u − x)
ke
−(u−x−2βt)24t +N udu.
Put z = u − x, then Z
∞−∞
|u − x|
ke
−(u−x−2βt)24t +N udu
= e
N xZ
∞−∞
|z|
ke
−(z−2βt)24t +N zdz
= e
N xZ
∞−∞
|z|
ke
−4t1(z−2t(β+N ))2+t(βN +N2−β2)dz
= e
N x+t(βN +N2−β2)Z
∞−∞
|z|
ke
−4t1(z−2t(β+N ))2dz.
Hence for v = z − 2t(β + N ) we get Z
∞−∞
|u − x|
ke
−(u−x−2βt)24t +N udu
= e
N x+t(βN +N2−β2)Z
∞−∞
|v + 2t(β + N )|
ke
−4t1v2dv
≤ M
1Z
∞−∞
|v + α|
ke
−γv2dv,
where M
1, α, γ are positive constants depending only on the set Ω(x
0, X
0, t
0, T
0), β and N . This implies that
Z
∞−∞
β
p(u − x)
kt
l+12e
−(u−x−2βt)24tf (u) du
≤ M
2Z
∞−∞
|v + α|
ke
−γv2dv, where M
2is a positive constant. Observe that the integral
Z
∞−∞
|v + α|
ke
−γv2dv
is convergent, hence the integral (22) is uniformly convergent on Ω(t
0, T
0, x
0, X
0).
Thus
(23) ∂
n+m∂t
n∂x
mZ
∞−∞
K(u, t, x)f (u) du = Z
∞−∞
∂
n+m∂t
n∂x
mK(u, t, x)f (u) du.
Consequently, the function U is of the class C
∞in Ω. It is easy to check that
∂
2K(u, t, x)
∂x
2+ 2β ∂K(u, t, x)
∂x = ∂K(u, t, x)
∂t .
This completes the proof.
Example 4.3 For p(t) = t, c = 1, A = −∞, B = +∞ we have q(t) = ln t, g(t) = e
t. From (19) we get
e
λex−λ+βex= Z
∞−∞
e
λutC(λ, u) du.
Hence
e
(λ+β)ex= e
λZ
∞−∞
e
λutC(λ, u) du.
From this and [1] we obtain e
λC(λ, u) =
∞
X
k=0
(λ + β)
kk! δ(k − λu), where δ is Dirac distribution. Therefore
C(λ, u) = e
−λ∞
X
k=0
(λ + β)
kk! δ(k − λu).
Applying (18) we have
W (λ, t, u) = t
λue
−(λ+β)t∞
X
k=0
(λ + β)
kk! δ(k − λu).
Combining this with (1) we get S
λ(f, t) =
Z
∞−∞
t
λue
−(λ+β)t∞
X
k=0
(λ + β)
kk! δ(k − λu)f (u) du
= e
−(λ+β)t∞
X
k=0
(λ + β)
kt
kk! f k
λ
.
Acknowledgement. The authors is thankful to the referee for giving useful comments.
References
[1] J. Antoniewicz, Tables of functions for engineers (in Polish), PWN Warszawa 1980.
[2] M. E. Ismail i C. R. May, On a Family of Approximation Operators, J. Math. Anal. 63 (1978), 446-462.
[3] C. R. May, Saturation and Inverse theorems for combinations of a class of exponential type operators, Canad. J. Math. 28 (1976), 1224-1250.
Agnieszka Tyliba, Eugeniusz Wachnicki
Institute of Mathematics, Pedagogical University ul. Podchora¸ ˙zych 2, 30-084 Krak´ow, Poland E-mail: tyliba@ap.krakow.pl, euwachni@ap.krakow.pl
(Received: 22.12.2003; revised: 17.10.2004)