Result.Math. 54 (2009), 397–413 c 2009 Birkh¨auser Verlag Basel/Switzerland 1422-6383/030397-17, published online July 14, 2009 DOI 10.1007/s00025-009-0362-4 Results in Mathematics A Characterization of the K -Functional for the Algebraic Version of the Trigonometric Jackson Integrals G s,n and the K -Functionals for Cao– Gonska Operators G ∗ s,n and G + s,n Teodora Dimova Zapryanova Abstract. We construct moduli of functions which are computable character- istic equivalents of the error in approximation by algebraic versions of trigono- metric Jackson integrals and Cao–Gonska operators. Mathematics Subject Classification (2000). Primary 41A25; Secondary 41A36. Keywords. K-functional, moduli of functions, linear operators. 1. Introduction Let X be a normed space. For a given differential operator D we set X ∩D −1 (X)= {g ∈ X : Dg ∈ X} . Given a subspace Y of X ∩ D −1 (X), we define for every f ∈ X and t> 0 the Peetre K-functional by K(f,t; X,Y,D) := inf {f − gX + t DgX : g ∈ Y } . (1) Let I be the identity operator and Lf denote the linear function interpolating f at −1 and 1, i.e., L(f,x) := f (1) 1+ x 2 + f (−1) 1 − x 2 , −1 ≤ x ≤ 1 . Let X be the space C[−1, 1]. We denote the norm in C[−1, 1] by .. Exam- ples of the operator D are D 1 g := Hg, where H := (H 1 ) 2 , (H 1 g)(x) := (1 − x 2 ) 1/2 d dx g(x) , D 2 g := H(I − L)g and D 3 g := (I − L)Hg.
17
Embed
A characterization of the k functional for the algebraic version of the trigonometric jackson integr
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
DOI 10.1007/s00025-009-0362-4 Results in Mathematics
A Characterization of the K-Functional for theAlgebraic Version of the Trigonometric JacksonIntegrals Gs,n and the K-Functionals for Cao–Gonska Operators G∗
s,n and G+s,n
Teodora Dimova Zapryanova
Abstract. We construct moduli of functions which are computable character-istic equivalents of the error in approximation by algebraic versions of trigono-metric Jackson integrals and Cao–Gonska operators.
Keywords. K-functional, moduli of functions, linear operators.
1. Introduction
Let X be a normed space. For a given differential operator D we set X∩D−1(X) ={g ∈ X : Dg ∈ X} . Given a subspace Y of X∩D−1(X), we define for every f ∈ Xand t > 0 the Peetre K-functional by
K(f, t;X,Y,D) := inf {‖f − g‖X + t ‖Dg‖X : g ∈ Y } . (1)
Let I be the identity operator and Lf denote the linear function interpolatingf at −1 and 1, i.e.,
L(f, x) := f(1)1 + x
2+ f(−1)
1 − x
2, −1 ≤ x ≤ 1 .
Let X be the space C[−1, 1]. We denote the norm in C[−1, 1] by ‖.‖ . Exam-ples of the operator D are
D1g := Hg , where H := (H1)2 , (H1g)(x) := (1 − x2)1/2 d
dxg(x) ,
D2g := H(I − L)g and D3g := (I − L)Hg .
398 T.D. Zapryanova Result.Math.
In these cases the K-functionals defined in (1) are
K(f, t;C[−1, 1], C2,H
):= inf{‖f − g‖ + t‖Hg‖ : g ∈ C2} , (2)
K∗(f, t;C[−1, 1], C2,H(I − L))
:= inf{‖f − g‖ + t‖H(I − L)g‖ : g ∈ C2} , (3)
K+(f, t;C[−1, 1], C2, (I − L)H
):= inf{‖f − g‖ + t‖(I − L)Hg‖ : g ∈ C2} . (4)
The K-functional (2) is equivalent to the approximation error of Jacksontype operators Gs,n in uniform norm, while the K-functional (3) is equivalentto the approximation error of operators G∗
s,n (such equivalence was establishedin [5]). The K-functional (4) is equivalent to the approximation error of operatorsG+
s,n (such equivalence was established in [6]).We recall that the operators Gs,n :C[−1, 1] → Πsn−s, G
Notation Φ(f, t) ∼ Ψ(f, t) means that there is a positive constant γ, inde-pendent of f and t, such that γ−1Ψ(f, t) ≤ Φ(f, t) ≤ γΨ(f, t).
By c we denote positive constants, independent of f and t, that may differat each occurrence.
For a natural number r number we putCr[a, b] = {f : f, f
′, . . . , f (r) ∈ C[a, b](continuous function in [a, b])}.
For r ∈ N and 1 ≤ p ≤ ∞ we denoteW r
p [a, b] = {f : f, f′, . . . , f (r−1) ∈ AC[a, b], f (r) ∈ Lp[a, b]}.
Here, in the case p = ∞ we mean only boundedness of the r-th derivative.Thus, W r
∞ = Cr.
2. Known results
The idea for the equivalence of the approximation errors of a given sequence ofoperators and the values of proper K-functionals was studied systematically in [1].Such equivalence was established for the algebraic version of trigonometric Jacksonintegrals Gs,n and K-functionals (2) and for the operators G∗
s,n and K-functionals(3) in uniform norm in [5] and later for the operators G+
s,n and K-functionals (4)in [6] (see Theorem A a), b), c) , respectively).
Vol. 54 (2009) A Characterization of K-functionals 399
Theorem A.a) For every f ∈ C[−1, 1] and s, n ∈ N, s ≥ 3 we have
‖f − Gs,nf‖ ∼ K
(f,
1n2
;C[−1, 1], C2,H
).
b) For every f ∈ C[−1, 1] and s, n ∈ N,n ≥ n0, s ≥ 3 we have∥∥f − G∗
s,nf∥∥ ∼ K∗
(f,
1n2
;C[−1, 1], C2,H(I − L))
.
c) For every f ∈ C[−1, 1] and s, n ∈ N, s ≥ 3 we have∥∥f − G+
s,nf∥∥ ∼ K+
(f,
1n2
;C[−1, 1], C2, (I − L)H)
.
On the other hand the idea for the equivalence of K- functionals and moduli ofsmoothness can be traced back to the 1960,s in the works of Peetre (see [2, Chap. 6,Theorem 2.4]).
Theorem B. For every 0 < t ≤ 1, f ∈ Lp[0, 1], 1 ≤ p ≤ ∞, r ∈ N and Dg = g(r)
we haveK
(f, tr;Lp[0, 1],W r
p ,D)∼ ωr(f, t)p .
The classical moduli of smoothness are defined by
ωr(f, t)p := sup0<h≤t
‖Δrhf(.)‖p ,
and the finite difference with a fixed step h is given by
Δrhf(x) =
{r∑k=0 (−1)r−k
(rk
)f(x + kh) , if x, x + rh ∈ [a, b] ,
0 , otherwise .
Similar characterizations of some weighted Peetre K-functionals in terms of weight-ed moduli were established in [3].
3. New results
The aim of this paper is to introduce moduli that are equivalent to the K-functionals (2), (3) and (4), respectively.
We define the modulus Ω2(f, t) for characterizing the K-functional K(f, t;C[−1, 1], C2,H) which describes the rate of approximation by operators Gs,n.
For a function f on [−1, 1] we consider
Δ2h
(f ◦ θ, θ−1(x)
):= (f ◦ θ)(y + h) + (f ◦ θ)(y − h) − 2(f ◦ θ)(y)
= f(cos(y + h)
)+ f
(cos(y − h)
)− 2f(cos y)
= f(cos(arccos x + h)
)+ f
(cos(arccos x − h)
)− 2f(x) ,
where θ(x) = cos x, y = θ−1(x).
400 T.D. Zapryanova Result.Math.
Definition 1. For given f ∈ C[−1, 1] and t > 0 we set
Ω2(f, t) := sup0<h≤t
sup−1≤x≤1
∣∣Δ2
h
(f ◦ θ, θ−1(x)
)∣∣ .
Obviously, Ω2(f, t) inherits the properties of the classical modulus ω2(f, t) as(f, g ∈ C[−1, 1]):
Lemma 1. Let Y be the space from Definition 4, g ∈ Y and g(σ) := g(cos σ).Theng ∈ C2(R) and g′′(σ) = Hg(cos σ) for σ ∈ R.
Proof. To prove the lemma we observe that
g(σ) = g(cos σ) = g(cos arccos cos σ) = g(arccos cos σ) .
We have
arccos cos σ ={
σ − 2kπ , if σ ∈(2kπ, (2k + 1)π
), k = 0 ,±1 ,±2 . . .
−σ + 2kπ , if σ ∈((2k − 1)π, 2kπ
), k = 0 ,±1 ,±2 . . .
.
Hence
(arccos cos σ)′=
{1 , if σ ∈
(2kπ, (2k + 1)π
), k = 0 ,±1 ,±2 . . .
−1 , if σ ∈((2k − 1)π, 2kπ
), k = 0 ,±1 ,±2 . . .
,
dg(σ)dσ
= g′(arccos cos σ)(arccos cos σ)′ .
From g ∈ Y and H1g(cos σ) = −dg(σ)dσ we get limσ→kπ±0
dg(σ)dσ = 0 for k =
0,±1,±2 . . .. Hence the left and the right derivative of the function g(σ) at pointsσ = kπ are equal to zero, dg(σ)
dσ = 0 at σ = kπ , k = 0,±1,±2 . . . and thereforeg(σ) ∈ C1(−∞,∞).
Vol. 54 (2009) A Characterization of K-functionals 403
For σ ∈ (2kπ, (2k + 1)π), k = 0,±1,±2 . . . and g ∈ Y we have
dg(σ)dσ
= g′(arccos cos σ)(arccos cos σ)′ = g′(σ − 2kπ) = g′(σ) .
d2g(σ)dσ2
= g′′(σ) = g′′(σ − 2kπ) .
From Hg ∈ C[−1, 1] (g ∈ Y ) and d2g(σ)dσ2 = Hg(cos σ) we get
limσ→2kπ+0
d2g(σ)dσ2
= limσ→0+0
g′′(σ) = limx→1−0
Hg(x) = Hg(1) .
For σ ∈ ((2k − 1)π, 2kπ), k = 0,±1,±2 . . . and g ∈ Y we have
dg(σ)dσ
= g′(arccos cos σ)(arccos cos σ)′ = g′(−σ + 2kπ)(−1) = −g′(−σ) .
d2g(σ)dσ2
= g′′(−σ) = g′′(−σ + 2kπ) .
From Hg ∈ C[−1, 1] (g ∈ Y ) and d2g(σ)dσ2 = Hg(x) we get
limσ→2kπ−0
d2g(σ)dσ2
= limσ→0−0
g′′(−σ) = limσ→0+0
g′′(σ) = limx→1−0
Hg(x) = Hg(1) .
Therefore the left and the right second derivative of the function g(σ) at pointsσ = 2kπ are equal. Similarly, using 2π periodisity of g′′(σ) we get that the left andthe right second derivative of the function g(σ) at points σ = (2k + 1)π are equal.This proves the lemma. �
Lemma 2. Let Y be the space from Definition 4. Then for every f ∈ C[−1, 1] andt > 0, we have
K(f, t;C[−1, 1], Y,H
)= K
(f, t;C[−1, 1], C2,H
),
K+(f, t;C[−1, 1], Y, (I − L)H
)= K+
(f, t;C[−1, 1], C2, (I − L)H
).
Proof. From C2 ⊂ Y we see that K(f, t;C[−1, 1], Y,H) ≤ K(f, t;C[−1, 1], C2,H).In order to prove K(f, t;C[−1, 1], C2,H) ≤ K(f, t;C[−1, 1], Y,H) it is sufficientto show (see Lemma 2 in [4, p. 116]) that for every g ∈ Y and ε > 0 there existsG ∈ C2 such that ‖G − g‖ ≤ ε and ‖HG‖ ≤ ‖Hg‖ + ε. Let g(x) ∈ Y. We putx = cos σ and consider g(σ) := g(cos σ). Since g(x) ∈ Y, g(σ) ∈ C2 (see Lemma 1).We use the Jackson integrals of the following type
Jn(g, σ) :=
π∫
−π
g(σ + v)K1s,n(v)dv =
π∫
−π
g(v)K1s,n(σ − v)dv, (5)
where K1s,n(v) := λs,n( sin mv/2sin v/2 )2s,
π∫−π
K1s,n(v)dv = 1, n, s > 0,m := [n/s] + 1.
404 T.D. Zapryanova Result.Math.
Since 12m ( sin mv/2
sin v/2 )2 = 12+
m−1∑k=0 (1 − k/m) cos kv it follows that K1s,n is an
even non-negative trigonometric polynomial of degree ≤ n. Jn(g, σ) is a trigono-metric polynomial of degree ≤ n, which is even as g is even. From Jackson’stheorem (see [3, Chap. 7, Theorem 2.2]) we get
‖g − Jn(g)‖ ≤ cω2
(g,
1n
)≤ O
(1n2
). (6)
By the substitution σ = arccos x in Jn(∼g, σ) we obtain an algebraic polynomial,
which is the desired function from C2. We set
G(x) = Jn(g, arccos x) . (7)
From ‖g − G‖C[−1,1] = ‖g − Jn(g)‖C[0,π] and (6) we get ‖g − G‖ ≤ O( 1n2 ). From
(5) we get
d2
dσ2Jn(g, σ) =
π∫
−π
g′′(σ + v)K1s,n(v)dv =
π∫
−π
g′′(v)K1s,n(σ − v)dv = Jn(g′′, σ) .
Using the Jackson theorem we get
‖g′′ − Jn(g′′)‖C[0,π] ≤ cω2
(g′′,
1n
). (8)
Since (Hg)(x) = d2
dσ2 g(σ) and (HG)(x) = d2
dσ2 Jn(g, σ) (8) implies
‖Hg − HG‖C[−1,1] ≤ cω2
(g′′,
1n
). (9)
For given ε > 0 we choose n such that cω2(g′′, 1n ) < ε and obtain ‖HG‖ ≤ ‖Hg‖+ε.
To prove the second statement in Lemma 2 it is sufficient to show thatfor every g ∈ Y and ε > 0 there exists G ∈ C2 such that ‖G − g‖ ≤ ε and‖(I − L)HG‖ ≤ ‖(I − L)Hg‖ + ε. Let G(x) be defined by (7). From (9) we get‖(I − L)H(g − G)‖ ≤ 2 ‖Hg − HG‖ ≤ cω2(g′′, 1
n ). For given ε > 0 we choose n
such that cω2(g′′, 1n ) < ε and obtain‖(I − L)HG‖ ≤ ‖(I − L)Hg‖ + ε. This com-
pletes the proof of the lemma. �
Proof of Theorem 1. The theorem will be proved if we show that
K
(f,
1n2
, C[−1, 1], Y,H
)∼ Ω2
(f,
1n
)
(see Lemma 2). First we show that Ω2(f, 1n ) ≤ cK(f, 1
n2 , C[−1, 1], Y,H). For g ∈ Y
we have Ω2(f, 1n ) ≤ Ω2(f − g, 1
n ) + Ω2(g, 1n ). But
Ω2
(f − g,
1n
)≤ 4 ‖f − g‖ ,
Vol. 54 (2009) A Characterization of K-functionals 405
and
Ω2
(g,
1n
)= sup
0<h≤ 1n
−1≤x≤1
∣∣g
(cos(arccos x + h)
)+ g
(cos(arccos x − h)
)− 2g(x)
∣∣
= sup0<h≤ 1
n
0≤z≤π
∣∣g
(cos(z + h)
)+ g
(cos(z − h)
)− 2g(cos z)
∣∣ (z = arccos x)
= sup0<h≤ 1
n
0≤z≤π
|g(z + h) + g(z − h) − 2g(z)|(g(z) := g(cos z)
)
= sup0<h≤ 1
n
0≤z≤π
∣∣∣∣∣∣∣
h2∫
−h2
h2∫
−h2
g′′(z + u1 + u2)du1du2
∣∣∣∣∣∣∣
≤ 1n2
‖g′′‖C[0,π] =1n2
‖Hg‖C[−1,1] .
We used that for g ∈ Y we have (see Lemma 1) g ∈ C2(−∞,∞). Therefore
Ω2
(f,
1n
)≤ 4 ‖f − g‖ +
1n2
‖Hg‖ ≤ cK
(f,
1n2
, C[−1, 1], Y,H
).
In order to prove K(f, 1n2 , C[−1, 1], Y,H) ≤ cΩ2(f, 1
n ) it is enough to showthat for every function f ∈ C[−1, 1] there exists function g ∈ Y such that:
a) ‖f − g‖ ≤ cΩ2(f, 1n );
b) 1n2 ‖Hg‖ ≤ cΩ2(f, 1
n ).
We define the following functions
g1,t(x) :=1t
t2∫
− t2
f(cos(arccos x + u)
)du =
1t
t2∫
− t2
f(z + u)du ≡ g1,t(z) .
g2,t(x) :=1t2
t2∫
− t2
t2∫
− t2
f(cos(arccos x + u1 + u2)
)du1du2
=1t2
t2∫
− t2
t2∫
− t2
f(z + u1 + u2)du1du2
=1t
t2∫
− t2
g1,t(z + u)du ≡ g2,t(z) ,
406 T.D. Zapryanova Result.Math.
where we have put z = arccos x, f(z) := f(cos z). We have
g2,kt(x) =1
k2t2
kt2∫
− kt2
kt2∫
− kt2
f(z + u1 + u2)du1du2
(ui = kvi,−
t
2≤ vi ≤
t
2
)
=1t2
t2∫
− t2
t2∫
− t2
f(z + k(v1 + v2)
)dv1dv2 .
We define
G2,t(x) :=12g2, t
2(x) +
12g2,− t
2(x) . (10)
We show later that G2,t ∈ Y. This function will serve for g in the above inequalitiesa) and b). We have
(G2,t−f)(x) =1
2t2
t2∫
−t2
t2∫
−t2
(f
(z+
12(u1+u2)
)+ f
(z − 1
2(u1+u2)
))du1du2−f(x)
=1
2t2
t2∫
− t2
t2∫
− t2
(f
(z +
12(u1 + u2)
)+ f
(z − 1
2(u1 + u2)
)
− 2f(z))
du1du2 .
Then
|(f − G2,t)(x)| ≤ 12t2
t2 sup0<u≤t
sup−1≤x≤1
∣∣∣f
(cos
(arccos x +
u
2
))
+f(cos
(arccos x − u
2
))− 2f(x)
∣∣∣ ,
and hence
‖f − G2,t‖ ≤ 12Ω2(f, t) .
We take t = 1n and obtain a).
Now we will prove b) with g = G2,t.We have
d
dzg2,t(z) =
1t
(g1,t
(z +
t
2
)− g1,t
(z − t
2
)), (11)
d2
dz2g2,t(z) =
1t
[1t
(f(z + t) − f(z)
)− 1
t
(f(z) − f(z − t)
)]
=1t2
(f(z + t) + f(z − t) − 2f(z)
),
Vol. 54 (2009) A Characterization of K-functionals 407
and henced2
dz2g2,kt(z) =
1k2t2
(f(z + kt) + f(z − kt) − 2f(z)
). (12)
If we set G2,t(z) = 12 g2, t
2(z) + 1
2 g2,− t2(z) and use that d2
dz2 G2,t(z) = HG2,t(x) weget
HG2,t(x) =12
d2
dz2g2, t
2(z) +
12
d2
dz2g2,− t
2(z)
=12
1t2
4
(f
(cos
(arccos x +
t
2
))+ f
(cos
(arccos x − t
2
))− 2f(x)
)
+12
1t2
4
(f
(cos
(arccos x− t
2
))+f
(cos
(arccos x+
t
2
))−2f(x)
).
Hence t2 ‖HG2,t‖ ≤ 4Ω2(f, t). This completes the proof of b) .Finally, we observe that G2,t(x) ∈ Y. (10) implies that G2,t(x) ∈ C[−1, 1]. (11)
implies that ddz G2,t(z) ∈ C[0, π] and d
dz G2,t(z) |z=0,π= 0. From ddz G2,t(z) =
−H1G2,t(x) we get H1G2,t(x) ∈ C[−1, 1] and H1G2,t(±1) = 0. (12) implies thatd2
dz2 G2,t(z) ∈ C[0, π]. From d2
dz2 G2,t(z) = HG2,t(x) we get HG2,t(x) ∈ C[−1, 1].This completes the proof of the equivalence
K
(f,
1n2
, C[−1, 1], Y,H
)∼ Ω2
(f,
1n
).
Applying the first statement of Lemma 2 we prove the theorem. �
Proof of Theorem 2. According to Theorem A b) we have K∗(f, 1n2 ;C[−1, 1], C2,
Lemma 3. Let Y be the space from Definition 4. Then for f ∈ C[−1, 1] and naturalnumber n we have
K+
(f,
1n2
;C[−1, 1], Y, (I − L)H)
= infg∈Y
{‖ f − g‖ +
1n2
‖(I − L)Hg‖}
∼ infg∈Y
{‖(I−L)( f−g)‖ +
1n2
‖(I−L)Hg‖}
.
Proof. Obviously for every function g ∈ Y we have
‖(I − L)(f − g)‖ +1n2
‖(I − L)Hg‖ ≤ 2 ‖f − g‖ +1n2
‖(I − L)Hg‖ .
408 T.D. Zapryanova Result.Math.
Therefore
infg∈Y
{‖(I − L)(f − g)‖ +
1n2
‖(I − L)Hg‖}
≤ 2K+
(f,
1n2
;C[−1, 1], Y, (I − L)H)
.
We will prove the inequality in the opposite direction. Let
infg∈Y
{‖(I − L)(f − g)‖ +
1n2
‖(I − L)Hg‖}
≥ ‖(I − L)(f − g0)‖ +1n2
‖(I − L)Hg0‖ − ε ,
i.e., the infimum is almost achieved for g = g0 (ε is a small positive arbitrarynumber). Set g1 = g0 + L(f − g0). We have g1 ∈ Y and
‖f − g1‖ +1n2
‖(I − L)Hg1‖
= ‖f − g0 − L(f − g0)‖ +1n2
∥∥(I − L)H
(g0 + L(f − g0)
)∥∥
= ‖(I − L)(f − g0)‖ +1n2
‖(I − L)Hg0‖ , as (I − L)HL(f − g0) ≡ 0 .
Hence
K+
(f,
1n2
;C[−1, 1], Y, (I − L)H)
≤ ‖f − g1‖ +1n2
‖(I − L)Hg1‖
≤ ‖(I − L)(f − g0)‖ +1n2
‖(I − L)Hg0‖
≤ inf{‖(I−L)(f−g)‖+
1n2
‖(I−L)Hg‖}
+ε .
The last inequality shows that
K+
(f,
1n2
;C[−1, 1], Y, (I−L)H)
≤ infg∈Y
{‖(I − L)(f−g)‖+
1n2
‖(I−L)Hg‖}
.
This proves the lemma. �
Proof of Theorem 3. The theorem will be proved if we show that
K+
(f,
1n2
, C[−1, 1], Y, (I − L)H)
∼ Ω+2
(f,
1n
)
Vol. 54 (2009) A Characterization of K-functionals 409
(see Lemma 2). First we show that Ω+2 (f, 1
n ) ≤ cK+(f, 1n2 ;C[−1, 1], Y, (I −L)H).
For g ∈ Y we have Ω+2 (f, 1
n ) = Ω+2 (f − g + g, 1
n ) ≤ Ω+2 (f − g, 1
n ) + Ω+2 (g, 1
n ). But
Ω+2
(f − g,
1n
)≤ 8 ‖f − g‖ , as ‖I − L‖ = 2 and
Ω+2
(g,
1n
)= sup
0<h≤ 1n
−1≤x≤1
∣∣∣(I − L)
[g(cos(arccos x + h)
)
+ g(cos(arccos x − h)
)− 2g(x)
]∣∣∣
= sup0<h≤ 1
n
0≤z≤π
∣∣∣(I − L)
[g(cos(z + h)
)+ g
(cos(z − h)
)− 2g(cos z)
]∣∣∣
= sup0<h≤ 1
n
0≤z≤π
∣∣(I − L)
[g(z + h) + g(z − h) − 2g(z)
]∣∣
= sup0<h≤ 1
n
0≤z≤π
∣∣∣∣∣∣∣(I − L)
h2∫
−h2
h2∫
−h2
g′′(z + u1 + u2)du1du2
∣∣∣∣∣∣∣
= sup0<h≤ 1
n
−1≤x≤1
∣∣∣∣∣∣∣(I − L)
h2∫
−h2
h2∫
−h2
Hg(cos(arccos x + u1 + u2)
)du1du2
∣∣∣∣∣∣∣,
where we have put z = arccos x , g(z) := g(cos z) and
L(g, z) := g(0)1 + cos z
2+ g(π)
1 − cos z
2, 0 ≤ z ≤ π .
We used that for g ∈ Y we have (see Lemma 1) g ∈ C2(−∞,∞).Set cos(arccos x + u1 + u2) = y, u = u1 + u2. Next we have
Hg(cos(arccos x + u)
)= g′′
(cos(arccos x + u)
)sin2(arccos x + u)
− g′(cos(arccos x + u)
)cos(arccos x + u)
and
Hg(y) = (1 − y2)g′′(y) − yg′(y) .
Thus
Hg(y) = Hg(cos(arccos x + u1 + u2)
).
Further we have
LHg(y) =(Hg)(1) − (Hg)(−1)
2y +
(Hg)(1) + (Hg)(−1)2
.
410 T.D. Zapryanova Result.Math.
Thenh2∫
−h2
h2∫
−h2
LHg(y)du1du2
=
h2∫
−h2
h2∫
−h2
((Hg)(1) − H(g)(−1)
2y +
(Hg)(1) + H(g)(−1)2
)du1du2
=
h2∫
−h2
h2∫
−h2
(c1 cos(arccos x + u1 + u2) + c2
)du1du2 .
We calculateh2∫
−h2
h2∫
−h2
cos(arccos x + u1 + u2)du1du2
=
h2∫
−h2
h2∫
−h2
[x cos(u1 + u2) −
√1 − x2 sin(u1 + u2)
]du1du2
= x
h2∫
−h2
h2∫
−h2
cos(u1 + u2)du1du2 −√
1 − x2
h2∫
−h2
h2∫
−h2
sin(u1 + u2)du1du2 = c(h)x ,
because the second term is zero. Hence
(I − L)
h2∫
−h2
h2∫
−h2
LHg(y)du1du2 = (I − L)(c1(h)x + c2(h)
)≡ 0 .
Further, using that Hg(cos(arccos x + u)) = Hg(y) and the above relation we get
(I − L)
h2∫
−h2
h2∫
−h2
Hg(cos(arccos x + u1 + u2)
)du1du2
= (I − L)
h2∫
−h2
h2∫
−h2
(I − L)Hg(cos(arccos x + u1 + u2)
)du1du2 .
Vol. 54 (2009) A Characterization of K-functionals 411
Hence
Ω+2
(g,
1n
)= sup
0<h≤ 1n
−1≤x≤1
∣∣∣∣∣∣∣(I − L)
h2∫
−h2
h2∫
−h2
Hg(cos(arccos x + u1 + u2)
)du1du2
∣∣∣∣∣∣∣
= sup0<h≤ 1
n
−1≤x≤1
∣∣∣∣∣∣∣(I − L)
h2∫
−h2
h2∫
−h2
(I − L)Hg(cos(arccos x + u1 + u2)
)du1du2
∣∣∣∣∣∣∣
≤ 2 sup0<h≤ 1
n
−1≤x≤1
∣∣∣∣∣∣∣
h2∫
−h2
h2∫
−h2
(I − L)Hg(y)du1du2
∣∣∣∣∣∣∣
≤ 21n2
sup−1≤y≤1
|(I − L)Hg(y)| =2n2
‖(I − L)Hg‖ .
Therefore
Ω+2
(f,
1n
)≤ 8 ‖f − g‖+
2n2
‖(I − L)Hg‖ ≤ 8K+
(f,
1n2
;C[−1, 1], Y, (I − L)H)
.
Using Lemma 3 to prove K+(f, 1n2 ;C[−1, 1], Y, (I − L)H) ≤ cΩ+
2 (f, 1n ), it is
enough to show that for every f ∈ C[−1, 1] there exists g ∈ Y such that:
a) ‖(I − L)(f − g)‖ ≤ cΩ+2 (f, 1
n );b) 1
n2 ‖(I − L)Hg‖ ≤ cΩ+2 (f, 1
n ).
We define G2,t by (10). This function will serve for g in the above inequali-ties a) and b). In the proof of Theorem 1 we observed that G2,t ∈ Y . We have
(I − L)(G2,t − f)(x)
=1
2t2(I − L)
t2∫
− t2
t2∫
− t2
(I − L)(
f
(cos
(arccos x +
12(u1 + u2)
))
+ f
(cos
(arccos x − 1
2(u1 + u2)
))− 2f(x)
)du1du2
+1
2t2(I − L)
t2∫
− t2
t2∫
− t2
L
(f
(cos
(arccos x +
12(u1 + u2)
))
+ f
(cos
(arccos x − 1
2(u1 + u2)
))− 2f(x)
)du1du2 .
412 T.D. Zapryanova Result.Math.
As the last term is zero we get
|(I − L)(G2,t − f)(x)| ≤ 22t2
t2 sup−t≤u≤t
−1≤x≤1
∣∣∣(I − L)
[f
(cos
(arccos x +
u
2
))
+f(cos
(arccos x − u
2
))− 2f(x)
]∣∣∣ .
Hence‖(I − L)(f − G2,t)‖ ≤ Ω+(f, t) .
We will show that t2 ‖(I − L)Hg‖ ≤ cΩ+(f, t) for g = G2,t. As in the proofof Theorem 1 we get
(I − L)HG2,t(x)
= (I − L)
[12
1t2
4
(f
(cos
(arccos x +
t
2
))+ f
(cos
(arccos x − t
2
))− 2f(x)
)]
+ (I−L)
[12
1t2
4
(f
(cos
(arccos x− t
2
))+f
(cos
(arccos x+
t
2
))−2f(x)
)]
.
Hence t2 ‖(I − L)HG2,t‖ ≤ 4Ω+2 (f, t). This completes the proof of b) .
Thus
K+
(f,
1n2
;C[−1, 1], Y, (I − L)H)
∼ Ω+2
(f,
1n
).
Applying the second statement of Lemma 2 we prove the theorem. �
5. Examples
In this section we give the order of magnitude of Ditzian–Totik modulus of smooth-ness ω2√
1−x2(f, t)∞ and moduli Ω2(f, t),Ω∗2(f, t),Ω+
2 (f, t) for several functions. Werecall
ω2√1−x2(f, t)∞ := sup
−1≤x≤1,0<h≤t
∣∣∣Δ
2
h√
1−x2f(x)∣∣∣ ,
where the second difference is given by
Δ2
t f(x) ={
f(x + t) + f(x − t) − 2f(x) , if x, x ± t ∈ [−1, 1] ,0 , otherwise .
Example 5.1. Let f(x) = (1 − x)α |log(1 − x)|β , x ∈ (−1, 1), α ∈ (0, 2], β ∈ R orα = 0, β < 0. Then
ω2√1−x2(f, t)∞∼
⎧⎪⎪⎨
⎪⎪⎩
t2α |log t|β for α ∈ (0, 1) , β ∈ R or α = 0 , β < 0 ;t2 |log t|β−1 for α = 1 , β > 1 ;t2 for α∈(1, 2] , β∈R or α=1 , β∈(−∞, 0) ∪ (0, 1] ;0 for α = 1 , β = 0 .
Vol. 54 (2009) A Characterization of K-functionals 413
Ω2(f, t) ∼ Ω∗2(f, t) ∼ Ω+
2 (f, t) ∼ t2α |log t|β for α ∈ (0, 1), β ∈ R or α = 1, β > 0 orα = 0, β < 0.
Ω2(f, t) ∼ t2 , Ω∗2(f, t) = Ω+
2 (f, t) = 0 for α = 1 , β = 0 .
Ω2(f, t) ∼ Ω∗2(f, t) ∼ Ω+
2 (f, t) ∼ t2
for α ∈ (1, 2] , β ∈ R or α = 1 , β < 0 .
Comparing the moduli Ω2(f, t),Ω∗2(f, t) and Ω+
2 (f, t) with the modulusω2√
1−x2(f, t)∞ for the function f(x) = (1 − x)α |log(1 − x)|β we see the differ-ence of the order of magnitude for α = 1, β ≥ 0. Therefore we cannot use themodulus ω2√
1−x2(f, t)∞ for a characterization of K-functionals (2), (3) and (4).
References
[1] Z. Ditzian and K.Ivanov, Strong converse inequalities, J. Anal. Math. 61 (1993),61–111.
[2] Z. Ditzian and V. Totik, Moduli of Smoothness. Berlin: Springer Verlag 1987.
[3] R. A. DeVore, G. G. Lorentz, Constructive Approximation. Berlin: Springer Verlag1993.
[4] K. Ivanov, A Characterization Theorem for the K-functional for Kantorovich andDurrmeyer Operators, Approximation Theory: A volume dedicated to Borislav Bo-janov, Marin Drinov Academic Publishing House, Sofia, (2004), 112–125.
[5] T. Zapryanova, Approximation by Operators of Cao–Gonska type Gs,n and G∗s,n.
Direct and Converse Theorem, Thirty Third Spring Conference of the Union ofBulgarian Mathematicians, Borovets (2004), pp. 189–194.
[6] T. Zapryanova, Approximation by operators of Cao–Gonska type G+s,n. Direct and
converse theorem. Thirty Seventh Spring Conference of the Union of Bulgarian Math-ematicians, Borovets (2008), pp. 169–175.
Teodora Dimova ZapryanovaVarna University of EconomicsBlvd. Kniaz Boris I, 77BG-9002 VarnaBulgariae-mail: [email protected]