ISSN: 1311-1728 (printed version); ISSN: 1314-8060 (on-line version)
doi:http://dx.doi.org/10.12732/ijam.v33i1.2
ON LINEAR OPERATORS GIVING HIGHER ORDER APPROXIMATION OF FUNCTIONS
IN Lpσ(R+)
Ali M. Musayev
Azerbaijan State Oil and Industry University Azadlig av. 20, AZ 1601, Baku, AZERBAIJAN
Abstract: Numerical investigations of different authors were devoted to con-vergence of singular integrals and approximation of functions by linear oper-ators. Asymptotic values of approximation of a function by linear operators were obtained here.
AMS Subject Classification: 41A35, 42A38, 42B20
Key Words: singular integrals, approximation, linear operators, Fourier transformation, Fejer operator
1. Introduction
In this field, the works of Lebesgue, I.P. Natanson, I.P. Korovkin, F.I. Kharshi-ladze, A.Z. Turetskiy, R.G. Mamedov, A.D. Hajiyev and others should be men-tioned.
One of the important problems in approximation theory is finding saturation classes of linear operators in various spaces. The saturation problem was first stated by Favour in 1937.
Aleksich, Zamanskiy, Alyanchich, Kharshiladze, Turetskiy, Butzer, Berents, Nessel, Sunouchi, Mamedov and others have obtained many important results in solving the problem on definition of saturation classes. Berents and Butzer [1] determined the class and order of saturation of approximation of the function f(t) (e−c t
f(t)∈Lp(0,∞), C >0, p ≥1) by the linear operator
R(f;x) =λ
x
Z
0
f(x−t)Kλ(t)dt
with a positive kernelKλ(t) =λK(λt)>0 in the metric of the spaceLp(0;∞).
They showed that the saturation order is 0(λ−γ
) (0< γ≤1).
Using the Fourier transformation, Butzer, Nessel [3], Sunouchi [4], Mamedov [6] and others determined the order and class of saturation of different singular integrals and linear operators in the space Lp(−∞;∞).
The basic results obtained for the last years by different authors solution of the saturation problem were stated in detail in the monographs of Mamedov [6], and Butzer-Berents [2].
In this paper, we suggest a method for constructing a new linear operator on the base of the given operators that gives a high approximation order to multiply differentiable functions.
Let Lpσ(R+) be a space of measurable in R+(0; +∞) functions f(x), for
which kfkLp
σ(R+)<∞(σ >0), where
kf(x)kLp σ(R+)=
∞ R 0
|e−σx
f(x)|pdx 1/p
for 1≤p <∞
supvrai |f(x)e−σx
| for p=∞ .
Let the function f(x)∈Lpσ(R+), then the Laplace transformation of the
func-tion f(x) is determined by
f∧ (s) =
∞ Z
0 e−sx
f(x)dx (s=σ+iτ, Res=σ >0),
where the integral converges absolutely forRes >0. Now, letϕ(x) be a function with a bounded variation on the segment [0, r] for any r > 0, i.e. ϕ(x) ∈
BV(0, r) and ∞ R 0
e−σi
Then the Laplace-Stieltjes transformation of the functionϕ(x) is determined in the following way ([2])
ϕ∧ (s) =
∞ Z
0 e−sx
d(ϕ(x)) (s=σ+iτ),
where the integral converges absolutely for Res=σ >0. Consider approxima-tion of funcapproxima-tions f(x) by linear operators of the form
Q[λm],l(f;x) =Rλ,1(Rλ,2...(Rλ,e(f;x))...))− m−1
X
v=1 αv(l)
λv f
(v)(x), (1)
wherel∈N, αv(l) (v= 0, N −1) are real numbers and
Rλ,n(g;x) =λ x
Z
0
g(x−t)Kn(λt)dt (n= 1, e), (2)
Kλ,n(t) =λKn(λt) are the functions determined onR+ called kernels with the
properties:
Kλ,n(t)∈L(R+),
∞ Z
0
Kλ,n(t)dt= 1
and
x
Z
0
Kλ,n(t)dt→1 (λ→ ∞), (n= 1, e).
Obviously, if f(v)(x) ∈Lp
σ(R+) (1≤p < ∞, v = 0, m−1),then the integral
operator (1) exists almost everywhere onR+ and Q[λm],l(f;x) ∈Lpσ(R+).
The condition Dm(α(e), B(e)): If for real numbers αv(e) (v = 0, m−1,
α0(e) = 1) andB(e)6= 0 it holds
lim
λ→∞ s
λ
−m" e Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
=B(e)6= 0,
for some fixed s(Res >0), it is said that the kernel Kn(t) operator (2) satisfies
the condition Dm(α(e), B(e)), where
Denote by ap(sν;f) a class of functionsf(x), having derivatives f(ν)(x) ∈
Lpσ(R+) andf(ν)(x)∈ACloc(R+) (ν = 1, N−1).
Introduce the class of functions
bp(sv;f)
=
f(x)∈a1(sv;f)/smf∧(s) =h∨1(s), ifh1(t)∈BVσ(R+)
R∞ 0 e
−σt
|dh1(t)|<+∞, if p= 1
f(x)∈ap(sv;f)/smf∧(s) =h∨2(s), if h2(t)∈Lpσ(R+), p >1
.
2. On linear operators giving higher order approximation of functions in Lpσ(R+)
Theorem 1. Let the kernel Kλ,n(t) = λKn(λt) (n = 1, e) of operator
(2) satisfy the condition Dm(α(e), B(e)) and f(t) ∈ap(sv, f).Then, if for the
functiong(t) (g(t)∈Lpσ(R+), p ≥1) the condition
λ
−m
[Rλ,1(Rλ,2(...(Rλ,e(f :x))...))
−
m−1 X
v=0 αv(e)
λv f
(v)(x)−g(x)
Lp σ(R+)
= 0(1) (3)
is fulfilled as λ→ ∞, the function f(x) has a m order derivative and g(x) =
B(e)f(m)(x) almost everywhere.
Proof. Letp= 1,sincef(x)∈a1(sv;f),then λm
" e Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
f∧
(s)−g∧ (s)
= ∞ Z
0 e−xs
{λm[Rλ,1(Rλ,2(...(Rλ,e(f;x))...) )
−
m−1 X
v=0 αv(e)
λv f
(v)(x) #
−g(x) )
dx. (4)
By the conditions of the theorem, from the last equality we find
λm " e
Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
f∧
(s)−g∧ (s)
asλ→ ∞.
Since Kn(t) (n= 1, e) satisfies the condition Dm(α(e), B(e)), then
B(e)smf∧
(s) =g∧
(s), (5)
for each s (Res >0), or
g(x) =B(e)f(m)(x)
almost everywhere.
Let 1< p <∞. For 0< σ′
< σby means of the H¨older inequality, from (4) we get
λm " e
Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
f∨
(s)−g∧ (s)
≤
e −σ′t
{[Rλ,1(Rλ,2(...(Rλ,e(f;x))...))
−
m−1 X
v=0 αv(e)
λv f
(v)(x) #
λm−g(x)
Lp(R+)
·
1 (σ−σ′)·q
1q
, (6)
wherepq=p+q.
The further reasoning is conducted similarly to the proof of the casep= 1.
Theorem 2. Let the kernel Kλ,n(t) =λKn(λt) (n= 1, e) of operator (2)
satisfy the condition Dm(α(e), B(e)). If for the function f(t) ∈ Lpσ(R+) the
condition
Rλ,1(Rλ,2(...(Rλ,e(f;x)...))− m−1
X
v=0 αv(e)
λv f
(v)(x)
Lp σ(R+)
= 0(λ−m
) (7)
is fulfilled asλ→ ∞, thenf(x)∈bp(sv;f) (v= 0, m, p≥1).
Proof. Letp= 1. Consider the partial integral Sτ,λ(x)
= 1 2πi
σ+iτ
Z
σ−iτ esx
" e Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
f∧
(s)ds, (8)
Feier’s mean valuesσr,λ(x) of partial integralsSτ,λ(x) equal
σr′,λ(x) =
1 r′
r′
Z
0
Sτ,λ(x)dτ =
1 2πi
σ+ir′
Z
σ−ir′
(1−|τ| r′ )e
sx· e Y n=1 K∧ n s λ −
m−1 X
v=0 αv(e)
s λ
v
ds= 1 2πi
r′
Z
−r′
(1−|τ| r′ )e
(σ+iτ)x
∞ Z 0
e−(σ+iτ)u
×Rλ,1(Rλ,2(...(Rλ,e(f;u)...))− m−1
X
v=0 αv(e)
λv f
(v)(u) #
du )
dτ.
Since by Lemma A (see [1]), the internal integral converges uniformly on −r′
≤τ < r′ , then
σr′,λ(x) =
2 πr′
∞ Z
0
eσ(x−u) ·sin
2r′ ·x−u
2
(x−u)2 {Rλ,1(Rλ,2(...(Rλ,e(f;u))...)
−
m−1 X
v=0 αv(e)
λv f
(v)(u) )
du. (9)
Further, taking into account (7), we get
kσr;λ(x)kLσ(−∞;+∞)≤ kRλ,1(Rλ,2(...(Rλ,e(f;u))...))
−
m−1 X
v=0 αv(e)
λv f
(v)(x)
Lpσ(R+)
= 0(λ−m
) (10)
asλ→ ∞. Consequently, 1 2πi
σ+ir′
Z
σ−ir′
1−|τ|
r′ esx " e Y n=1 K∧ n s λ −
m−1 X
v=0 αv(e)
λv s v # f∧ (s)ds
Lσ(−∞;+∞)
=O(λ−m
Furthermore, by the conditionDm(α(e), B(e)),we have λm " e Y n=1 K∧ n s λ −
m−1 X
v=0 αv(e)
s λ v #
≤B(e)|sm|
and
e−σx
(1−|τ| r′ )e
sx " e Y n=1 K∧ n s λ −
m−1 X
v=0 αv(e)
s λ
v #
f∧
(s)·λm
≤2B(e)· |sm| ·f ∧
(s) for each λ > λ0 |τ|< r′,wheres=σ+iτ, σ >0.
Consequently,
2B(e)
r′
Z
−r′
|σ+iτ|m f
∧
(σ+iτ)
dτ ≤2B(e)
r′
Z
−r′
(σ2+τ2)m2
× ∞ Z 0 e−σt
|f(t)|dt
dτ ≤M1 <∞.
Since e−σx
σr′,λ(x)
λm>0 and estimation (9) is valid, then ∞
Z
−∞ e−σx
σr′,λ(x)
λmdx <∞.
Thus, the requirements of the Fatou lemma are fulfilled and taking into account (9), we find:
1 2πi
σ+ir′
Z
σ−ir′
(1−|τ| r′ )e
sxB(e)smf∧ (s)ds
Lσ(−∞;+∞)
= 0(1).
To complete the proof of f(x) ∈bp(sv;f), (v= 0, m) for p= 1 we should
apply Lemma A in [1].
Let 1 < p < ∞. By (4) and the theorem on weak compactness in the
space Lp(R+), there exists a sequence {λ
j}
lim
j→∞λj =∞
qe[m](x)∈Lp(R+) such that
lim
λ→∞ ∞ Z
0 e−εx
g(x)λjmhQλj[m],e(f;x)−f(x)i dx
= ∞ Z
0
g(x)qe[m](x)dx, (12)
for each g(t)∈Lp(R+), pq=p+q (ε >0 is any number), whereQ[λm],e
j (f;t) is
determined from formula (12). Then, it is obvious that h(t) =eεtq[em](t).
Now, if in equality (12) instead ofg(x) we take the function of the form
g(x) =e−[(σ−ε)+iυ]x
(σ >0, −∞< τ <∞),
we have
lim
λj→∞λ
m j
∞ Z
0 e−sx
[Q[λjm],e(f;x)−f(x)]dx= ∞ Z
0 e−sx
g(x)dx (Res >0).
Hence, by the condition Dm(α(e), B(e)), we find B(e)smf∧(s) = g∧(s), i.e.
f(x) ∈ bp(sv;f). This is the required relation for 1 < p < ∞. In the case
p=∞, the required one is proved in the same way.
Theorem 3. Let the kernel Kλ,n(t) = λKn(λt) (n = 1, e) satisfy the
condition Dm(α(e), B(e))and be such that the function
θm,e
s λ
= 1 B(e)
s λ
−m " e
Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
,
is the Laplace-Stielties transformation of the normalized function with bounded
variation onR+forp= 1or the Laplace transformation of functions in the space
L(R+) forp >1. Then fromf(x)∈b
p(sv;f) (v= 0, m), the relation
kRλ,1(Rλ,2(...(Rλ,e(f;x))...))− m−1
X
v=0 αv(e)
λv f
(v)(x)
Lpσ(R+)
=O(λ−m )
Proof. Let p = 1. Since f(x) ∈ b1(sv;f) (v = 0, m) and θm,e λs
is a Laplace-Stielties transformation of the normalized function with bounded
varia-tion on R+ i.e. θ
m,e sλ = Nm,ev λs
, (Nm,e(x)∈BV(0, r)).
∞ R 0
e−σx
|dNm,e(x)|<∞, we apply Lemma A in [1] and find
∞ Z
0 e−sx
"
Rλ,1(Rλ,2(...(Rλ,e(f;x)...))− m−1
X
v=0 αv(e)
λv f
(v)(x) #
λmdx
=λmf∧ (s)
" e Y
n=1 K∧
n
s λ
−
m−1 X
v=0 αv(e)
s λ
v #
= ∨ [
λ,m,e
(s)≡ ∞ Z
0 e−sx
d [
λ,m,e
(x), (13)
where
Uλ,m,e(x) = x
Z
0
h1(x−t)dNm,e(λt), Uλ,m,e(x)∈BV(0, r)
for each r >0 and the relation
∞ Z
0 e−σx
|dUλ,m,e(x)| ≤M2<∞ (14)
is satisfied for allλ >0.
According to the uniqueness of the Laplace-Stielties transformation ([2], p. 62), from (13) we have:
Uλ,m,e(x) =λm x
Z
0 "
Rλ,1(Rλ,2(...(Rλ,e(f;t))...))− m−1
X
v=0 αv(e)
λv f
(v)(t) #
dt.
Consequently, by (15) we find
λm Z x
0 e−σx
|Rλ,1(Rλ,2(...(Rλ,e(f;x))...)) − m−1
X
v=0 αv(e)
λv f
asλ→ ∞. This is relation (4) forp= 1. Let 1< p <∞. Sincef(x)∈bp(sv;f)
andθm,e λs
is a Laplace transformation of the function from the spaceL(R+), then ([4], p. 92), we have
"
Rλ,1(Rλ,2(...(Rλ,e(f;x))...))− m−1
X
v=0 αv(e)
λv f
(v)(x)λm
#∧ (s)
= λB(e)
x
Z
0
f(m)(x−u)h2,m,e(λu)du
∧
(s),
forRes >0, whereQm,e λs=h∧r,m,e λs
, (hr,m,e(t)∈L(R+)).
By the theorem on uniqueness of the Laplace transformation ([6], p. 63), from the last relation we find
"
Rλ,1(Rλ,2(...(Rλ,e(f;x))...))− m−1
X
v=0 αv(e)
λv f
(v)(x) #
λm
=λB(e)
x
Z
0
f(m)(x−t)hr,m,e(λt)dt. (15)
Furthermore, by the H¨older inequality, we have
λB(e)
x
Z
0
f(m)(x−t)h2,m,e(t)dt
p
Lpσ(R+)
≤B(e)· khr,m,e(t)k p q+1 L(R+)·
f
(m)(x)
p
Lp σ(R+)
=O(1)
asλ→ ∞.
Then, the validity of relation (4) follows from equality (15). Notice that for p=∞ the reasonings are conducted similarly.
It follows from Theorem 2 that the family of operators (1) is saturated in the spaceLpσ(R+) (p≥1) with orderO(λ−m) andbp(sv;f) is a saturation class.
Apply the obtained results to one concrete linear operator:
p[λm],e(f;x) =Rλ(Rλ(...(Rλ(f;x))...))− m−1
X
v=0 αv(e)
λv f
(v)(x), (16)
Rλ(f;x) =λ x
Z
0
f(x−t)e−λt dt
and
αv(e) =
(−1)ve(e+ 1)...(e+v−1)
v! .
It is easily verified that the operator (16) is saturated in the spaceLpσ(R+)
(p≥1) with orderO(λ−ν
) and bp(sv;f) (v= 0, m), is its saturation order.
References
[1] H. Berens and P. Butzer, On the best approximation for approximation for singular integrals by Laplace-transform methods, Bull. of the Amer. Math. Soc.,70(1964), 180-184.
[2] H. Berens and P. Butzer, ¨Uber die Darstelling holomorpher Funktionen-durch Laplace-und Laplace Stieltjes Integrale,Mat. Z.,81(1963), 124-134. [3] P. Butzer, R. Nessel, Fourier Analysis an Approximation, Vol. 1.,
Birkhauser Verlag, Basel-Stuttgart, 1971.
[4] G. Cynoymu (G. Sunouchi), Direct theorems in the theory of approxima-tion,Acta Math.,20, No 3-4 (1969), 409-420.
[5] A.D. Hajiyev, Collected Papers, Elm, Baku, 2003.
[6] R.G. Mamedov, Mellin Transformation and Approximation Theory, Elm, Baku, 1991.
[7] A.M. Musayev, To the question of approximation of functions by the Mellin type operators in the spaceXσ1,σ2(E
+).Proc. of IMM of NAS Azerbaijan, 28(2008), 69-73.
[8] A.M. Musayev, Multiparameter approximation of function of general vari-ables by singular integrals, Azerb. Techn. Univ. Baku, No 2 (2014), 212-218.
