IN VARIABLE EXPONENT LEBESGUE SPACES
DANIYAL M. ISRAFILOV and ELIFE YIRTICI
Communicated by Vasile Brˆınz˘anescu
In the variable exponent Lebesgue spaces a convolution is defined and its es-timations in the variable exponent Lebesgue spaces by the best approximation numbers are obtained.
AMS 2010 Subject Classification: 41A10, 42A10.
Key words: convolution, variable exponent Lebesgue space, best approximation.
1. INTRODUCTION AND MAIN RESULTS
Let p(·) : [0, 2π] → [1, ∞) be a Lebesgue measurable function. We define the modular functional
ρp(·)(f ) :=
Z 2π
0
|f (x)|p(x)dx
on the Lebesgue measurable functions f on [0, 2π]. By Lp(·)2π we denote the class of 2π periodic Lebesgue measurable functions f , such that for a constant λ = λ (f ) > 0
ρp(·)(f /λ) < ∞. Equipped with the norm
kf kp(·):= infλ > 0 : ρp(·)(f /λ) ≤ 1
the class Lp(·)2π creates a Banach space.
The variable exponent Lebesgue spaces are a generalization of the classical Lebesgue spaces, replacing the constant exponent p with a variable exponent function p(·). Interest in the variable exponent Lebesgue spaces has increased since 1990s, because of their use in the different applications to problems in mechanic, especially in fluid dynamic for the modeling of electrorheological fluids and also in the study of image processing and some problems in physics (see, for example the monographs [4, 5, 14] and the references cited therein). Nowadays there are sufficiently wide investigations relating to the fundamen-MATH. REPORTS 18(68), 4 (2016), 497–508
tal problems of these spaces, in view of potential theory, maximal and singular integral operator theory and others. The sufficiently wide presentation of the corresponding results can be found in the monographs [4, 5, 14].
In the variable exponent Lebesgue spaces some fundamental problems of approximation theory were investigated also. Some necessary and sufficient conditions in term of the variable exponent for the basicity of the well known classical systems of functions were obtained [15, 16], the different modulus of smoothness were defined [7, 8, 17] and the direct and inverse theorems of ap-proximation theory in these spaces defined on the intervals of the real line and on the domains of the complex plane were proved [1–3, 7–9, 17]. The detailed information on these results and also on the general aspects of approximation theory in the variable exponent Lebesgue spaces can be found in the mono-graph [18].
Note that in the variable exponent Lebesgue space theory some convolu-tion operators were commonly used. This type of operators have some applica-tions also in the approximation theory, in particular for the construction of the approximation polynomials. Therefore, the estimation problem of convolution operators by using the best approximation numbers is an actual problem of ap-proximation theory. In this work, we investigate this problem in the variable exponent Lebesgue spaces. For the formulation of the main results obtained in this work we give some notations.
By β2π we denote the class of the exponents p (·), for which 1 < p− ≤
p+ < ∞, where p−:= essinf x∈[0,2π] p (x) , and p+:= essup x∈[0,2π] p (x) .
Definition 1. Let p(·) : [0, 2π] → [1, ∞) be a 2π periodic, measurable function. We say that p(·) is a log-Holder continuous function on [0, 2π] if (1.1) |p (x) − p (y)| ≤ c0
− log (|x − y|), x, y ∈ [0, 2π] with |x − y| ≤ 1/2 with some constant c0 > 0.
If p(·) ∈ β2πand satisfies the condition (1.1), then we say that p(·) ∈ bβ2π.
In the space Lp(·)2π with p(·) ∈ bβ2π we define a mean value operator σh
(σhf ) (x, u) := 1 2h Z h −h f (x + tu) dt, 0 < h < π, x ∈ [0, π] , − ∞ < u < ∞, which is linear and bounded by [6], moreover k(σhf )kp(·)≤ c1kf kp(·) for some
constant c1 > 0.
For f ∈ Lp(·)2π we define also the best approximation number En(f )p(·):= inf
Tn
by trigonometric polynomials Tn(x) = n X k=0 ckeikx
of degree at most n and by Tn∗the best approximation trigonometric polynomial of degree ≤ n, such that
kf − Tn∗kp(·)= En(f )p(·).
Let f ∈ Lp(·)2π . We define a convolution type operator Z ∞
−∞
(σhf ) (·, u) dσ (u)
with a bounded variation function σ (u) on the real line R and denote D (f, σ, h, p (·)) := Z ∞ −∞ (σhf ) (·, u) dσ (u) p(·) .
In this work, we estimate the quantity D (f, σ, h, p (·)) using the best approximation number En(f )p(·).
Our new results are following:
Theorem 1. If f ∈ Lp(·)2π with p(·) ∈ˆß2π, then
D (f, σ, h, p (·)) ≤ c
m
X
k=0
E2k−1(f )p(·)δ2k,h+ cp(·)E2m+1(f )p(·)
for every m ∈ N, where δ2k,h : = 2k+1−1 X l=2k |ˆσ (lh) − ˆσ ((l + 1) h)| + σˆ 2kh , ˆ σ (x) : = Z ∞ −∞ sin (ux) ux dσ (u) , 0 < h < π.
Theorem 2. Let f ∈ Lp(·)2π with p(·) ∈ˆß2πp(·) and let F (x) be a function
with bounded variation, i.e. kF (x)k ≤ c1,
2µ+1−1
X
θ=2µ
|F (θh) − F ((θ + 1) h)| ≤ c2, h ≤ 2−m−1.
If σ1 and σ2 are two functions satisfying the condition
ˆ σ1(x) = ˆσ2(x) F (x) , |x| < 1, then D (f, σ1, h, p(·)) = c h D (f, σ2, h, p(·)) + E2m+1(f )p(·) i .
Using the usual shift f (x + t) in the definition of the convolution operator Theorems 1 and 2 in the Orlicz spaces were proved in [13]. In the weighted Orlicz spaces in term of the mean value operator (σhf ) (x, u), Theorems 1 and
2 were obtained in [10]. We also use this operator in the variable exponent spaces Lp(·)2π , because Lp(·)2π in general is non invariant with respect to the usual shift f (x + t).
Throughout this paper, the constant c denotes a generic constant, i.e. a constant whose values can change even between different occurrences in a chain of inequalities.
2. AUXILIARY RESULTS
As was proved in [15], if p (·) ∈ ˆß2π, then the systemeikx
k∈Z of
expo-nents greats a base in the space Lp(·)2π , which is equivalent to the inequality kSn(f )kp(·)≤ cp(·)kf kp(·) (n = 0, 1, 2, ...) , where Sn(f ) (x) := n X k=−n ˆ fk(x) eikx
is the nth partial sum of the Fourier series
∞ X k=1 ˆ fkeikx ; fˆk(x) = 1 2π Z 2π 0 f (y) e−ikydy of f . This inequality implies the following
Lemma 1. If f ∈ Lp(·)2π with p(·) ∈ˆß2π, then
kf − Snf kp(·)≤ cEn(f )p(·). Proof. kf − Snf kp(·) ≤ kf − Tnkp(·)+ kSnf − Tnkp(·) ≤ En(f )p(·)+ kSn(f − Tn)kp(·) ≤ En(f )p(·)+ c k(f − Tn)kp(·) ≤ c1En(f )p(·)
with some constant c1=c1(p)¿ 0.
The proofs of the following two Theorems A and B can be found in [4, p. 27, Theorem 2.26] and [18, p. 39, Theorem 1.6.5], respectively.
Theorem A. Let p : [0, 2π] → [1, ∞) be a 2π periodic, measurable func-tion. If f ∈ Lp(·)2π and g ∈ Lp2π0(·), where 1/p (·) + 1/p0(·) = 1, then f g ∈ L1
2π and Z .2π 0 |f (x) g (x)| dx ≤ Kp(·)kf kp(·)kgkp0(·) with Kp(·)= 1/p−− 1/p++ 1.
Theorem B. Let p (·) : [0, 2π] → [1, ∞) and let f : [0, 2π] × [0, 2π] → R be a measurable function and f (·, y) ∈ Lp(·)2π for every y ∈ [0, 2π]. Then
2π Z 0 f (·, y) dy p(·) ≤ c 2π Z 0 kf (·, y)kp(·)dy
with some positive constant c = cp(·).
Theorem C ([6]). If f ∈ Lp(·)2π with p(·) ∈ˆß2π, then the maximal operator
M (f ) (x) := sup I x 1 |I| Z I |f (t)| dt is bounded in Lp(·)2π and kM f kp(·)≤ c kf kp(·) with some constant c = cp(·)> 0.
Theorem CF ([4, p. 212]). Let = be a family of pairs (f, g) of non-negative, measurable function f and g defined on (0, 2π) such that for all ω ∈ Ap0(0, 2π) with some p0≥ 1 2π Z 0 fp0(x) ω (x) dx ≤ c 2π Z 0 gp0(x) ω (x) dx, (f, g) ∈ =,
where a constant c independent of (f, g). If p(·) ∈ˆß2π, then for every r,
1 < r < ∞, and sequence {(fi, gi)} ⊂ = X i fir !1/r p(·) ≤ cp(·) X i gir !1/r p(·) .
Lemma 2 ([11]). Let f ∈ Lp(·)2π with p (·) ∈ˆß2π. If Al(x) = eilxfˆl and A2−1(x) = 0, l = 0, 1, 2, ..., then c1kf kp(·)≤ ∞ X j=0 2j−1 X l=2j−1 Al(x) 2 1/2 p(·) ≤ c2kf kp(·).
Lemma 3. Let {fn}∞1 be a sequence of functions fn∈Lp(·)2π with p (·) ∈ˆß2π
and let Sn,kn be the kth partial sum of fn with k = kn. Then
(2.1) ∞ X n=1 |Sn,kn(x)| 2 !1/2 p(·) ≤ c ∞ X n=1 |fn(x)|2 !1/2 p(·)
with some constant c > 0 independent of fn.p
Proof. The inequality (2.1) is a consequence of the extrapolation Theo-rem CF (for the case of r = 2, gi := fi and fi := |Si,ki(x)|, i = 1, 2, ..., )
and of the norm inequality, proved by Kurtz in [12] in the weighted Lebesgue spaces Lpω.
Let L∞comp be the set of all bounded functions with compact support on [0, 2π]and let {fn}∞1 be a sequence in L∞comp. If
f (x) := ∞ X n=1 |fn(x)|2 !1/2 and T f (x) := ∞ X n=1 |Sn,kn(x)| 2 !1/2 , then by [12] for a number p0 with 1 < p0 < ∞ there is a constant c such that
for all weight ω ∈ Ap0
kT f kp
0,ω ≤ c kf kp0,ω.
Hence the conditions of the above cited extrapolation Theorem CF are fulfilled which implies the inequality (2.1) for the sequence in {fn}∞1 from L∞comp. Since
the set L∞comp is dense in Lp(·)2π the inequality (2.1) is also valid for all sequences {fn}∞1 from Lp(·)2π .
Theorem D. Let λ0, λ1, ... be a sequence of the numbers such that
(2.2) |λl| ≤ M ,
2l+1−1
X
υ=2l
|λυ− λυ+1| ≤ M (l = 0, 1, 2, ...) .
If p (·) ∈ˆß2π and aυ, bυ be the Fourier coefficients of f ∈ L p(·)
2π , then the series
a0λ0/2 + ∞
X
υ=1
is the Fourier series of function F ∈ Lp(·)2π and kF kp(·)≤ c kf kp(·) with some constant c = cp(·)> 0 independent of f .
Proof. Let for s > 2µ−1, µ = 1, 2, ..., ∆µ,s : = s X υ=2µ−1 Aυ(x) , Aυ(x) := aυcos υx + bυsin υx ∆µ : = 2µ−1 X υ=2µ−1 Aυ(x) and ∆ 0 µ:= 2µ−1 X υ=2µ−1 λυAυ(x) .
As in [19, p. 347] we have the estimation ∆ 0 µ 2 ≤ 2M 2µ−1 X s=2µ−1 |∆µ,s|2|λs− λs+1| + |∆µ|2|λ2µ| ! .
Since p (x) ≤ p+ by Lemma 3 and (2.2)
ρp(·) ∞ X µ=1 ∆ 0 µ 2 1/2 = Z 2π 0 ∞ X µ=1 ∆ 0 µ 2 p(x)/2 dx ≤ Z 2π 0 ∞ X µ=1 2M 2µ−1 X s=2µ−1 |∆µ,s|2|λs− λs+1| + |∆µ|2|λ2µ| ! p(x)/2 dx = Z 2π 0 (2M )p(x)/2 ∞ X µ=1 2µ−1 X s=2µ−1 |∆µ,s|2|λs− λs+1| + |∆µ|2|λ2µ| ! p(x)/2 dx ≤ c Z 2π 0 (2M )p(x)/2 ∞ X µ=1 |∆µ|2 2µ−1 X s=2µ−1 |λs− λs+1| + |λ2µ| ! p(x)/2 dx ≤ c Z 2π 0 (2M )p+ ∞ X µ=1 |∆µ|2 p(x)/2 dx ≤ c3 Z 2π 0 ∞ X µ=1 |∆µ|2 p(x)/2 dx = c3ρp(·) ∞ X µ=1 |∆µ|2 1/2 ,
which by Lemma 2 implies that kF kp(·) ≤ ∞ X µ=1 ∆ 0 µ 2 1/2 p(·) ≤ c3 ∞ X µ=1 |∆µ|2 1/2 p(·) ≤ c4kf kp(·).
3. PROOFS OF MAIN RESULTS
Proof of Theorem 1. Let f ∈ Lp(·)2π , m ∈ N and let S2m+1 be the 2m+1th
partial sum of its Fourier series. Let also h ≤ 2−m−1. Then D (f, σ, h, p(·)) ≤ Z ∞ −∞ [(σhf ) (x, u) − (σhS2m+1f ) (x, u)]dσ (u) p(·) + Z ∞ −∞ (σhS2m+1f ) (x, u)]dσ (u) p(·) . (3.1)
By applying Theorems B and C and Lemma 1 in the first term, we have Z ∞ −∞ [(σhf ) (x, u) − (σhS2m+1f ) (x, u)]dσ (u) p(·) ≤ Kp(·) Z ∞ −∞ k(σhf ) (·, u) − (σhS2m+1f ) (·, u)kp(·)dσ (u) = Kp(·) Z ∞ −∞ k(σh(f − S2m+1f )) (·, u) kp(·)dσ (u) ≤ c5Kp(·) Z ∞ −∞ kf − S2m+1f kp(·)dσ (u) ≤ c5Kp(·)E2m+1(f )p(·) Z ∞ −∞ dσ (u) ≤ c6E2m+1(f )p(·).
Without loss of generality, we suppose that the Fourier series of f is
∞ X k=1 ˆ Ak(x) := ∞ X k=1 ˆ fkeikx. Then Z ∞ −∞ (σhS2m+1f ) (x, u) dσ (u)
= Z ∞ −∞ [ 1 2h Z h −h S2m+1f (x + tu) dt]dσ (u) = Z ∞ −∞ [ 1 2h Z h −h 2m+1−1 X k=1 ˆ fkeik(x+tu)dt]dσ (u) = Z ∞ −∞ [ 1 2h 2m+1−1 X k=1 ˆ fkeikx Z h −h eiktudt]dσ (u) = Z ∞ −∞ [ 1 2h 2m+1−1 X k=1 ˆ Ak(x) Z h −h eiktudt]dσ (u) = 2m+1−1 X k=1 ˆ Ak(x) Z ∞ −∞ eikhu− e−ikhu 2ikhu dσ (u) = 2m+1−1 X k=1 ˆ Ak(x) ˆσ (kh) , (3.2)
and hence by (3.2) and (3.1)
(3.3) D (f, σ, h, p(·)) ≤ 2m+1−1 X k=1 ˆ Ak(x) ˆσ (kh) p(·) + c6E2m+1(f )p(·).
Now, by inequality (a + b)p < ap+ bp, which holds for every positive numbers a and b in the case of 0 < p < 1 and by Lemma 2
2m+1−1 X k=1 ˆ Ak(x) ˆσ (kh) p(·) ≤ c m X k=0 2k+1−1 X l=2k ˆ Al(x) ˆσ (lh) 2 1/2 p(·) : = c m X k=0 ∆2k,σ !1/2 p(·) < c m X k=0 ∆k,σ p(·) ≤ c m X k=0 k∆k,σkp(·).
On the other hand, applying the Abel transformation to the sum ∆k,σ= 2k+1−1 X l=2k ˆ Al(x) ˆσ (lh) we have ∆k,σ = 2k+1−1 X l=2k Sl(f, x) − S2k+1−1(f, x) [ˆσ (lh) − ˆσ ((l + 1) h)] +S2k+1−1(f, x) − S2k−1(f, x) ˆσ 2kh and then k∆k,σkp(·)≤ 2k+1−1 X l=2k kSl(f ) − S2k+1(f )kp(·)|ˆσ (lh) − ˆσ ((l + 1) h)| + S2k+1−1(f ) − S2k−1(f ) p(·) σˆ 2kh = kS2k(f ) − S2k+1(f )kp(·) σˆ 2kh− ˆσ2k+ 1h +... + S2k+1(f ) − S2k+1(f ) p(·) σˆ 2k+ 1 h − ˆσ 2k+ 2 h + S2k+1−1(f ) − S2k−1(f ) p(·) σˆ 2kh ≤hkS2k(f ) − f kp(·)+ kS2k+1(f ) − f kp(·) i σˆ 2kh − ˆσ 2k+ 1 h + ... + h S2k+1(f ) − f p(·)+ kS2k+1(f ) − f kp(·) i σˆ 2k+1− 1h − ˆσ 2kh +h S2k+1−1(f ) − f p(·)+ S2k−1(f ) − f p(·) i ˆσ 2kh ≤ cE2k−1(f )p(·)δ2k,h. Hence 2m+1−1 X k=1 ˆ Ak(x) ˆσ (kh) p(·) ≤ c m X r=0 E2k−1(f )p(·)δ2k,h
and by (3.3) we obtain the required inequality.
Proof of Theorem 2. Let f ∈ Lp(·)2π . Repeating the techniques used for the estimation of the quantity D (f, σ, h, p(·)) from Theorem 1, we have
(3.4) D (f, σ1, h, p(·)) ≤ Z ∞ −∞ (σhS2m+1f ) (·, u) dσ1(u) p(·) + cp(·)E2m+1(f )p(·).
On the other by (3.2) and Lemma 1 Z ∞ −∞ (σhS2m+1f ) (x, u) dσ1(u) p(·) = 2m+1−1 X k=1 ˆ Ak(x) ˆσ1(kh) p(·) = 2m+1−1 X k=1 ˆ Ak(x) ˆσ2(kh) F (kh) p(·) = 2m+1−1 X k=1 ˆ fkeikxσˆ2(kh) F (kh) p(·) ≤ c 2m+1−1 X k=1 ˆ fkeikxˆσ2(kh) p(·) = c Z ∞ −∞ (σhS2m+1f ) (x, u) dσ2(u) p(·) = c Z ∞ −∞ S2m+1(σhf ) (x, u) dσ2(u) p(·) = c S2m+1 Z ∞ −∞ (σhf ) (x, u) dσ2(u) p(·) ≤ Z ∞ −∞ σhf (x, u) dσ2(u) p(·) . The last inequality together with (3.4) implies the required relation. Acknowledgments. The authors sincerely thank the anonymous reviewers for their careful reading and constructive comments. This work was supported by TUBITAK grant 114F422: Approximation Problems in the Variable Exponent Lebesgue Spaces.
REFERENCES
[1] R. Akgun, Trigonometric approximation of functions in generalized Lebesgue spaces with variable exponent. Ukrainian Math. J. 63 (2011), 1, 3–23.
[2] R. Akgun, Polynomial approximation of functions in weighted Lebesgue and Smirnov spaces with nonstandard growth. Georgian Math. J. 18 (2011), 203–235.
[3] R. Akgun and V.M. Kokilashvili, The refined direct and converse inequalities of trigono-metric approximation in weighted variable exponent Lebesgue spaces. Georgian Math. J. 18 (2011), 399–423.
[4] D.V. Cruz-Uribe and A. Fiorenza, Variable Lebesgue Spaces Foundation and Harmonic Analysis. Birkh¨asuser, 2013.
[5] L. Diening, P. Harjulehto, P. H¨ast¨o and M. Ruzicka, Lebesgue and Sobolev Spaces with Variable Exponents. Springer, Heidelberg, Dordrecht, London, New York, 2011. [6] L. Diening, Maximal function on generalized Lebesgue spaces Lp(·). Math. Inequal.
Appl. 7 (2004), 245–253.
[7] A. Guven and D.M. Israfilov, Trigonometric approximation in generalized Lebesgue spaces Lp(x). J. Math. Inequal. 4 (2010), 2, 285–299.
[8] D.M. Israfilov, V.M. Kokilashvili and S. Samko, Approximation in Weighted Lebesgue and Smirnov Spaces with Variable Exponents. Proc. A. Razmadze Math. Inst. 143 (2007), 25–35.
[9] D.M. Israfilov and A. Testici, Approximation in Smirnov classes with variable exponent. Complex Var. Elliptic Equ. 60 (2015), 9, 1243–1253.
[10] D.M. Israfilov and Y.E. Yildirir, The properties of convolution type transforms in weighted Orlicz spaces, Gla. Mat. 45 (2010), 2, 461–474.
[11] T. Kopaliani, Interpolation theorems in the variable exponent Lebesgue spaces. J. Funct. Anal. 257 (2009), 3541–3551.
[12] D.S. Kurtz, Littlewood-Paley and multiplier theorems in weighted Lp spaces. Trans. Amer. Math. Soc. 259 (1980), 235–254.
[13] V.G. Ponomarenko and M.F. Timan, The properties of convolution type transforms in the Orlicz spaces. In: Theory of Approximation of Functions, Proc. Inst. Math. Mech. 3, Donetsk, 1998.
[14] M. R˚uˇziˇcka, Elektrorheological Fluids: Modeling and Mathematical Theory. Springer, 2000.
[15] I.I. Sharapudinov, The basis property of the Haar system in the space Lp(t)([0, 1]) and
the principle of localization in the mean. Mat. Sb. 130 (1986), 172, 275–283.
[16] I.I. Sharapudinov, The basis property of the Legendre polynomials in the variable expo-nent Lebesgue spaces system in the space Lp(x)(0, 1), and the principle of localization in
the mean. Sb. Math. 200 (2009), 1, 130–156.
[17] I.I. Sharapudinov, Approximation of functions in Lp(x)2π by trigonometric polynomials.
Izvestiya RAN: Ser. Math. 77 (2013), 2, 197-224. English transl., Izvestiya: Mathe-matics 77 (2013), 2, 407–434.
[18] I.I. Sharapudinov, Some questions of approximation theory in the Lebesgue spaces with variable exponent. Vladikavkaz. Mat. Zh., 2012.
[19] A. Zygmund, Trigonometric Series. Cambridge Univ. Press, Cambridge, 1968.
Received 6 March 2015 Balikesir University,
Faculty of Art and Science, Department of Mathematics,
10145, Balikesir, Turkey ANAS Institute of Mathematics
and Mechanics, Baku, Azerbaijan [email protected]
Balikesir University, Faculty of Art and Science, Department of Mathematics,
10145, Balikesir, Turkey [email protected]