• Nie Znaleziono Wyników

Uniformly non-l

N/A
N/A
Protected

Academic year: 2021

Share "Uniformly non-l"

Copied!
10
0
0

Pełen tekst

(1)

Seria I: PRACE MATEMATYCZNE XLV (1) (2005), 23-32

Mohamed Morsli, Fatiha Boulahia

Uniformly non-l

1n

Besicovitch-Orlicz space of almost periodic functions

Abstract. We consider uniformly non-l1nproperty of the Besicovitch-Orlicz space of almost periodic functions. It is shown that this property is equivalent to the reflexiv- ity of this space.

2000 Mathematics Subject Classification: 46B20.

Key words and phrases: Uniformly non-ln1, B-convexity, reflexivity, Besicovitch- Orlicz, almost periodic functions.

1. Introduction. Recall at the beginning that a Banach space (X, k.k) is said to be uniformly non-l1n, ( n ∈ N , n ≥ 2) or ((n, δ)−convex) if there is a δ ∈ (0, 1) such that for each choice of x1, x2, ...xnfrom the unit ball of X we have :

kx1± x2± ... ± xnk ≤ n(1 − δ)

for some choice of the signs ±. The definition remains true if we replace kxik ≤ 1 by kxik = 1 for i = 1, 2, ..., n (see [5]).

This geometric property has been introduced by R. C. James in [6]. Banach spaces that are uniformly non - l21 are called uniformly non-square (see [6]). Any uniformly non -square Banach space is reflexive (see [6]). Banach spaces that are uniformly non-l1nfor some integer n ≥ 2 are called B−convex. This notion is useful in the probability theory (see [1], [2], [4]. B-convexity is a topological notion i.e., it is invariant under equivalent renorming (see [4]). Any B-convex Banach space which has an unconditional Schauder bases or which is a Banach lattice is reflexive.

In the class of Orlicz spaces, B-convexity coincides with reflexivity and also with super reflexivity (see [3]).

The purpose of this paper is to study these questions in the Besicovitch-Orlicz space of almost periodic functions.

Now, we shall give some notations and definitions.

(2)

2. Preliminaries.

2.1. Orlicz functions. A function φ : R → R+is said to be an Orlicz function if it is even, convex, φ (0) = 0, φ (u) > 0 iff u 6= 0 and lim

u→0 φ(u)

u = 0, lim

u→∞

φ(u) u = +∞.

An Orlicz function admits a derivative φ0 except perhaps on a countable set of points. It satisfies φ0(0) = 0, φ0(|u|) > 0 whenever u > 0 and lim

|u|→∞

φ0(|u|) = +∞, so that is increasing to infinity (see [3], [11])

The function ψ (y) = sup {x |y| − φ (x) , x ≥ 0} is called conjugate to φ. It is an Orlicz function when φ is. The pair (φ, ψ) satisfies the Young inequality

xy ≤ φ (x) + ψ (y) , x ∈ R, y ∈ R.

Let us note that equality holds in the Young’s inequality iff x = ψ0(y) or y = φ0(x) .

We say that an Orlicz function satisfies ∆2-condition if there exist K > 2 and u0≥ 0 for which

φ (2u) ≤ Kφ (u) , ∀u ≥ u0.

In this case we write φ ∈ 42. If φ ∈ 42and ψ ∈ 42we write φ ∈ 42∩ ∇2. When φ /∈ 42 then there exists a sequence (an)n≥1 of positive reals numbers increasing to infinity for which (see [3])

φ



1 + 1 n

 an



≥ 2nφ (an) , ∀ n ≥ 1.

2.2. The Besicovitch-Orlicz space of almost periodic functions. Let M (R, E) be the set of all real Lebesgue measurable functions with value in a Banach space (E, k.k).

The functional,

ρBφ: M (R, E) → [0, +∞] , ρBφ(f ) = lim

T →+∞

1 2T

+T

Z

−T

φ (kf (t)k) dt

is a pseudomodular (see [9], [10]).

The associated modular space,

Bφ(R, E) = n

f ∈ M (R, E) , lim

α→0ρBφ(αf ) = 0o

= {f ∈ M (R, E) , ρBφ(λf ) < +∞, for some λ > 0}

is called the Besicovitch-Orlicz space.

This space is endowed with the Luxemberg pseudonorm kf kBφ= inf



k > 0, ρBφ f k



≤ 1



, f ∈ Bφ(R, E) called the Luxemburg norm (see [9], [10]).

(3)

Let now A be the linear set of generalized trigonometric polynomials, i.e

A =

 P (t) =

n

X

j=1

αjexp (iλjt) , λj∈ R , αj∈ E, n ∈ N

 .

The Besicovitch-Orlicz space of almost periodic functions denoted by Bφa.p. (R, E) ( resp. eBφa.p. (R, E) ) is the closure of A in Bφ(R, E) with respect to the pseudonorm k.kBφ ( resp. to the modular convergence ), more exactly :

Bφa.p. (R, E) =

f ∈ Bφ, ∃ (Pn)n≥1⊂ A, s.t. lim

n→+∞kf − PnkBφ= 0 ff

=

f ∈ Bφ, ∃ (Pn)n≥1⊂ A, s.t. ∀k > 0, lim

n→+∞ρBφ(k (f − Pn)) = 0 ff

Beφa.p. (R, E) =

f ∈ Bφ, ∃ (Pn)n≥1⊂ A, s.t. ∃k > 0, lim

n→+∞ρBφ(k (f − Pn)) = 0 ff

. Clearly, Bφa.p. (R, E) ⊂ eBφa.p. (R, E) and equality holds when φ ∈ ∆2.

Some structural and topological properties of these spaces are considered in [7],[9]

and [10].

From [9], we know that when f ∈ Bφa.p. (R, E) and φ ∈ ∆2the limits exists in the expression of ρBφ, i.e :

ρBφ(f ) = lim

T →+∞

1 2T

+T

Z

−T

φ (kf (t)k) dt, f ∈ Bφa.p. (R, E) .

This fact is very useful in our computations.

We now state some fundamental results that will be used below.

From now on, we always denote by φ, ψ a pair of Orlicz functions complementary to each other.

Lemma 2.1 (see [9]). Let f ∈ Bφa.p. (R, E), and suppose the Orlicz function φ satisfies the ∆2-condition. Then :

1. kf kBφ≤ 1 if and only if ρBφ(f ) ≤ 1.

2. kf kBφ= 1 if and only if ρBφ(f ) = 1.

3. ∀ε > 0, ∃η > 0 such that kf kBφ > ε ⇒ ρBφ(f ) > η and the converse is also true.

4. ∀ε > 0, ∃δ (ε) > 0 such that ρBφ(f ) ≤ 1 − ε ⇒ kf kBφ ≤ 1 − δ (ε) .

Lemma 2.2 (see [8]). Let E be a uniformly non-l1nnormed linear space, if ψ satisfies the 42−condition, there exists 0 < r < 1 such that for all x1, x2, ..., xn in E

X

±

φ



x1± x2± ... ± xn

n



≤ 2n−1r n

n

X

i=1

φ (kxik) ,

where the first sum is taken over the 2n−1 choices of signs ±.

(4)

Lemma 2.3 (see [9]). Let {an}n≥1, an > 0, be a sequence of real numbers. With every n ≥ 1, we associate a measurable set An⊂ [0, 1] such that :

a) Ai∩ Aj= ∅ if i 6= j and S

n≥1

An⊂ [0, α] , where 0 < α < 1.

b) P

n≥1

φ (an) µ (An) < +∞.

Consider the function f = P

n≥1

anχAn on [0, 1] and let ef be the periodic extension of f to the whole R, with period τ = 1. Then ef ∈ eBφa.p. (R, R).

3. Auxiliary results.

Lemma 3.1 Let φ satisfies the ∆2-condition, then the mapping I : (Lφ([0, 1]) , k.kφ) → Bφa.p. (R, R)

f → fe

(where ef is the periodic extension of f ) is an isometry for the respective modulars and also an isometry for the respective norms. Lφ([0, 1]) is the classical Orlicz space on [0, 1] and k.kφdenotes it’s usual Luxemburg norm.

The results of this lemma remains true if we take an interval [a, b] , a, b ∈ R.

Proof We have only to prove that ef ∈ Bφa.p. (R, R) . Indeed, if f ∈ Lφ([0, 1]) , then for every ε > 0, there exists a step function fε=

n

P

i=1

aiχAi d´efined on [0, 1] such that kf − fεkφε4 (see [3], [11]). Since f is absolutely continuous (cf.[3]), we may choose δ > 0 such that µ(A) ≤ δ ⇒ kf χAk ≤ε4. We take α > 0 with 1 − α ≤ δ and put Aαi = Ai∩ [0, α], i = 1, 2, · · · , n. Let fεα=

n

P

i=1

aiχAα

i, then fεα∈ Lφ([0, 1]).

If ef and efεα are the respective periodic extensions of f and fεα to the whole R (with the same period τ = 1), we will have

f − ee fεα

Bφ

≤ kf − fεαkφ

(f − fεα[0,α]

φ+

(f − fεα[α,1]

φ

≤ kf − fεkφ+ f χ[α,1]

φ

≤ ε

4+ε 4= ε

2

From Lemma 2.3 since φ ∈ ∆2we know that efεα∈ Bφa.p. (R, R) , and then there exists a trigonometric polynomial Pεsuch that

feεα− Pε Bφ

ε2, hence,

f − Pe ε

Bφ

≤ f − ee fεα

Bφ

+

feεα− Pε Bφ

≤ ε 2+ε

2 = ε which means that ef ∈ Bφa.p. (R, R) .

The mapping is clearly a modular isometry and also an isometry for the norms. The following Lemma is a direct consequence of Theorem 1.13 in [3]:

(5)

Lemma 3.2 Suppose that there exist two constants η, ξ ∈ (0, 1) and δ ≥ 0 such that φ (ηu) ≤ ηξφ (u) ∀u ≥ δ.

Then ψ satisfies the 42−-condition.

Proposition 3.3 We suppose that φ doesn’t satisfies the ∆2-condition. Then Beφa.p. (R, E) contains an isometric copy of C0.

(C0is the classical Banach space of sequences converging to 0).

Proof Let Sn, n ≥ 1, be a family of disjoints subsets of [0, 1[ , such that µ (Sn) = 21n

and S

n≥1

Sn⊂ [0, 1[ .

Let an,1, n ≥ 1, be a sequence of real numbers with an,1> 2n. Since φ does’nt satisfies the 42−condition, for every n ≥ 1, we may find a sequence (an,k)k≥1 increasing to infinity and such that

φ



1 +1 k

 an,k



> 2n+kφ (an,k) , k ≥ 1.

Since the measure µ is nonatomic, by the Laponov’s theorem {µ (A) , A is Lebesgue measurable} = [0. + ∞[ .

Consequently we can find disjoint sets {Sn} of [0, 1] with µ (Sn) = 21n and for each n a sequence (Sn,k)k≥1a disjoint subsets of Snwith µ (Sn,k) = 2n+kφ(a1 n,k).

We define fn= P

k≥1

an,kχSn,k, n = 1, 2, ... on [0, 1[ and fn= efn, n = 1, 2, ... with e ∈ E, kek = 1.

Let ffn be the periodic extension of fn to the whole R, with period τ = 1. From Lemma 2.3 we know that ffn∈ eBφa.p. (R, E) . Moreover,

ρBφ

 ffn

=

1

Z

0

φ (|fn(t)|) dt

= X

k≥1

φ (an,k) µ (Sn,k) =X

k≥1

1

2n+k < 1, ∀n = 1, 2, ... . Now if λ > 1 there exists k0∈ N such that 1 +1k < λ, ∀k ≥ k0and thus

ρBφ

 λffn

=

1

Z

0

φ (λ |fn(t)|) dt =X

k≥1

φ (λan,k) µ (Sn,k)

≥ X

k≥k0

φ



1 +1 k

 an,k

 µ (Sn,k)

≥ X

k≥k0

2n+kφ (an,k) µ (Sn,k) ≥ X

k≥k0

1 = +∞, ∀n ≥ 1.

(6)

So that ffn

Bφ

= 1, ∀n ≥ 1.

Consider now the mapping,

P : C0 −→ X

c = (cn) 7−→ fec= P

n≥1

cnffn,

where X = (

P

n≥1

cnffn, (cn)n≥1∈ C0 )

.

First, we prove that efc∈ eBφa.p. (R, E) . Indeed, if efN =

N

P

n=1

cnffnwe have clearly feN ∈ eBφa.p. (R, E) and then it is sufficient to show that the sequencen

feNo

N ≥1

is norm convergent to efc or that for each λ > 0,

lim

N →+∞ρBφ

 λ

 fec− efN



= 0.

We have

ρBφ

 λ

 fec− efN



= ρBφ

 X

n≥N

λcnffn

=

1

Z

0

φ

 X

n≥N

λcn|fn(t)|

dt = X

n≥N 1

Z

0

φ (λcn|fn(t)|) dt

Taking N such that λcn≤ 1; ∀ n ≥ N, we will have,

ρBφ

 λ

fec− efN

= X

n≥N

 X

k≥1

φ (λcnan,k) µ (Sn,k)

≤ X

n≥N

X

k≥1

φ (an,k) µ (Sn,k)

≤ X

n≥N

 X

k≥1

1 2n+k

≤ 1

2N −1

which tends to zero when N → +∞. Hence the space being complete (see [10]), we get efc∈ eBφa.p. (R, E) .

Now, we show that P is an isometry.

Let λ < kck, then there exists n0such that cn0λ = α0> 1.

Consequently :

ρBφ

fec λ

!

=

1

Z

0

φ

X

n≥1

cn λfn(t)

dt ≥

1

Z

0

φ (α0|fn0(t)|) dt = +∞

(7)

If λ ≥ kck, we will have,

ρBφ

fec λ

!

=

1

Z

0

φ

X

n≥1

cn λfn(t)

dt

≤ X

n≥1 1

Z

0

φ (|fn(t)|) dt ≤X

n≥1

 X

k≥1

1 2n+k

≤ 1.

Finally, fec

Bφ= kck and P is an isometry. 

4. Uniform non-l1n of eBφa.p. (R, E).

Theorem 4.1 The space eBφa.p. (R, E) is uniformly non-l1nif and only if E is uni- formly non-l1nand both φ and ψ satisfy the 42−condition.

Proof Sufficiency. Suppose E is uniformly non-l1n and both φ and ψ satisfy the 42−condition. Let n be the natural number from the definition of uniformly non-ln1 of E and let f1, f2, ..., fnbe norm one elements in eBφa.p. (R, E) . Since φ ∈ 42we have also ρBφ(fi) = 1 for all i = 1, 2, · · · , n. By Lemma 2.2, there exists 0 < r < 1 such that

(1) X

±

φ



f1(t) ± f2(t) ± ... ± fn(t) n



≤2n−1r n

n

X

i=1

φ (kfi(t)k) . Integrating inequality (1) over [−T, T ] yields

X

± T

Z

−T

φ



f1(t) ± f2(t) ± ... ± fn(t) n



dt ≤ 2n−1r n

n

X

i=1 T

Z

−T

φ (kfi(t)k) dt

and since the limits exist, letting T → +∞ we get:

X

±

lim

T →+∞

1 2T

T

Z

−T

φ



f1(t) ± f2(t) ± ... ± fn(t) n

 dt

≤ 2n−1r n

n

X

i=1

lim

T →+∞

1 2T

T

Z

−T

φ (kfi(t)k) dt

≤ 2n−1r n

n

X

i=1

ρBφ(fi) = 2n−1r Consequently,

ρBφ

 f1± f2± ... ± fn

n



= lim

T →+∞

1 2T

T

Z

−T

φ



f1(t) ± f2(t) ± ... ± fn(t) n

 dt

≤ r < 1

(8)

for some choice of signs ±. Hence there exists δ ∈ (0, 1) (δ depends on r) such that

f1± f2± ... ± fn n

Bφ

≤ 1 − δ for some choice of signs ±.

Therefore Bφa.p. (R, E) is uniformly non-ln1.

Necessity. Suppose φ does not satisfy the 42−condition, then by Proposition 3.3, eBφa.p. (R, E) will contain an isometric copy of C0and so eBφa.p. (R, E) can’t be uniformly non- ln1 (it is well known (cf.[4]) that a B-convex Banach space contains no subspace isomorphic to C0).

Now, assume that φ satisfies the 42-condition and ψ does not satisfy the 42- condition, then by Lemma 3.2 we have :

∀η, ξ ∈ (0, 1) and δ ≥ 0, there exists u0≥ δ such that φ (ηu0) > ηξφ (u0) Putting η = 1n. we get

φ 1 nu0



> 1 nξφ (u0) In particular we can write :

∀ξ ∈ (0, 1) , ∃uξ such that φ (uξ) > n and φuξ n



> ξ nφ (uξ) .

By Lapunov’s theorem, there are pairwise disjoint sets A1, A2, ..., An such that µ (Ai) = φ(u1

ξ) i = 1, 2, ..., n;

n

S

i=1

Ai⊂ [0, 1[ , this implies from φ (uξ) > n, µ (Ai) <

1 n.

Define fi = uξ χAi on [0, 1[ , i = 1, 2, ..., n. Let efi be the periodic extension of fi to the whole R, with period τ = 1. From Lemma 2.3, we know that efi ∈ Beφa.p. (R, R) , ∀i = 1, 2, ..., n.

Now let fi = e efiwhere e ∈ E with kek = 1 for i = 1, 2, ..., n. We have : ρBφ

 f1± f2± ... ± fn



≥ 1

ξρBφ

 f1± f2± ... ± fn

n



≥ 1

ξ lim

T →+∞

1 2T

T

Z

−T

φ

f1(t) ± f2(t) ± ... ± fn(t) n

! dt

≥ 1

ξ

1

Z

0

φ |f1(t) ± f2(t) ± ... ± fn(t)|

n

 dt

≥ 1

ξ

n

X

i=1

φuξ n

 µ (Ai)

≥ n

ξφuξ n

 1

φ (uξ) ≥ n ξ ξ n = 1.

Thus,

f1± f2± ... ± fn

Bφ ≥ nξ, for any choice of signs ±. ξ ∈ (0, 1) being arbitrary, this means that eBφa.p. (R, E) is not uniformly non-l1n.

(9)

The necessity of the uniform non-ln1of E is a direct consequence of the following injective isometry

i : E −→ Beφa.p. (R, E)

x 7−→ xf

where f ∈ eBφa.p. (R, E) satisfies kf kBφ = 1 i.e. E is embedded isometrically into

Beφa.p. (R, E) . 

Remark 4.2 The necessity of the 42−condition on ψ may be proved as follows:

Since φ satisfies the 42−condition, the mapping I : (Lφ([0, 1]) , k.kφ) → Bφa.p. (R, R) is a modular isometry for the respective norms (see Lemma 3.1). Then, from the uniformly non-l1nof Lφ([0, 1]), it follows that ψ must be of 42−type (see [3]).

Remark 4.3 If E is a Hilbert space the space eBφa.p. (R, E) is reflexive iff φ ∈

2∩ ∇2.

We have only to show the necessity.

Suppose eBφa.p. (R, E) is reflexive, then φ ∈ ∆2since in the contrary case eBφa.p. (R, E) will contain an isometric copy of C0.

To show that we have also ψ ∈ ∆2, remark that since considering the canonical isometry of Lemma 3.1, we deduce that Lφ[0, 1] is also reflexive and so that ψ ∈ ∆2.

Remark 4.4 When E = R, uniform non-l1nproperties are all equivalent in eBφa.p.(R).

In fact, they are equivalent to the reflexivity of the space.

References

[1] A. Beck, On the strong law of large numbers, Ergodic Theory, Academic Press, New York 1963.

[2] A. Beck, A convexity condition in Banach space and the strong law of large numbers Proc.

Amer. Math. Soc. 13 (1962), 329-334.

[3] S. Chen, Geometry of Orlicz spaces, Dissertationes Math. 356, 1996.

[4] D. P. Giesy, On convexity condition in normed lineaire spaces, Trans. Amer. Soc. 125 (1966), 114-146.

[5] R. Grz¸slewicz, H. Hudzik and W. Orlicz, Uniform non-l1n property in some normed space, Bull. Polish Acad. Sci. Math. 34 (1986), 161-171.

[6] R. C. James, Uniformly non-square Banach spaces, Ann. of Maths 80 (1964), 542-550.

[7] T. R. Hillmann, Besicovitch-Orlicz spaces of almost periodic functions, Real and Stochastic Analysis, Wiley 1986, 119-167.

[8] A. Kami´nska and B. Turett, Uniformly non l1n Orlicz-Bochner spaces, Bull Acad. Polon. Sci.

Math. 35(3-4) (1987), 211-218.

[9] M. Morsli, On some convexity properties of the Besicovitch-Orlicz space of almost periodic functions, Comment. Math. 34 (1994), 137-152.

(10)

[10] M. Morsli, On modular approximation property in the Besicovitch-Orlicz space of almost periodic functions, Comment. Math. Univ. Carolinae 38(3) (1997), 485-496.

[11] M. M. Rao and Z. D. Ren, Theory of Orlicz spaces, Marcel Dekker, Inc. New-York 1991.

Mohamed Morsli University of Tizi-Ouzou

Department of Mathematics. Faculty of Sciences.University of Tizi-Ouzou Oued-Aissi, Algeria E-mail: morsli@ifrance.com

Fatiha Boulahia University of Tizi-Ouzou

Department of Mathematics. Faculty of Sciences.University of Tizi-Ouzou Oued-Aissi, Algeria E-mail: Boulahia@yahoo.fr

(Received: 15.06.2003; revised: 22.03.2004)

Cytaty

Powiązane dokumenty

Meskine, Existence of solutions for elliptic equations having natural growth terms in orlicz spaces, Abstr.. Meskine, Strongly nonlinear parabolic equations with natural growth terms

Complex extreme point, complex strict convexity and complex uniform convexity of complex Banach space are natural generalizations of extreme point, strict

ANNALES SOCIETAT1S MATHEMATICAE POLONAE Series I: COMMENTATIONES MATHEMATICAE XXVII (1987) ROCZNIKI POLSKIEGO TOWARZYSTWA MATEMATYCZNEGO.. Séria I: PRACE MATEMATYCZNE

A strongly exposing property (SE) is introduced as an appropriate assumption on the space R). An appropriate assumption on this space appears to be a strongly

ROCZNIKI PQLSKIEGO TOWARZYSTWA MATEMATYCZNEGO Séria I: PRACE MATEMATYCZNE XXIV (1984)M. Hence in this case p is

The norm generated by the function 99 is defined in the third chapter in which also one concerns the mutual relations of various kinds of convergence of the

The purpose of this note is to investigate a constructive characterization of generalized almost periodic functions;.. the set of these functions will be denoted

In Section 4, we develop a capacity theory based on this space; we study basic properties of capacity, including monotonicity and countable subadditivity, as well as several