Seria I: PRACE MATEMATYCZNE XLV (1) (2005), 71-86
Satit Saejung∗
Some Convexity properties in Musielak-Orlicz sequence spaces endowed with the Luxemburg
Norm
Abstract. Criteria for k-strict convexity, uniform convexity in every direction, prop- erty K, property H, and property G in Musielak-Orlicz sequence spaces and their subspaces endowed with the Luxemburg norm are presented. In particular, we obtain a characterization of such properties of Nakano sequence spaces.
2000 Mathematics Subject Classification: 46B20, 46E30.
Key words and phrases: k-strict convexity, uniform convexity in every direction, property K, property H, and property G, Musielak-Orlicz sequence space.
1. Introduction. Convexity properties of Banach spaces play essential role in the theory of approximation and optimization. For example, the property of k-strict convexity, introduced by I. Singer, ensures that the dimension of the set PM(x), the Chebyshev map or the best approximation operator, is not greater than k and vice versa (see [16]).
Now we introduce the basic notions and definitions. A convex function ϕ : R → R+= [0, ∞) is called an Orlicz function if it vanishes at zero and is convex even on the whole line R and is not identically equal to zero. Denote by l the space of all real sequences x = (x(i)). For a given Musielak-Orlicz function Φ = (Φi), i.e. a sequence Φ = (Φi) of Orlicz functions, we define a convex semimodular IΦ: l → [0, ∞] by the formula
IΦ(x) =
∞
X
i=1
Φi(x(i)).
The Musielak-Orlicz sequence space lΦis the space
lΦ:= {x ∈ l : IΦ(cx) < ∞ for some c > 0}.
∗Supported by Thailand Research Fund under grant BRG/01/2544.
We consider lΦ equipped with the Luxemburg norm kxk = inf{k > 0 : IΦ(x/k) ≤ 1}.
To simplify notation, we put lΦ:= (lΦ, k · k). It is known that lΦis a Banach space (see [13]).
The subspace hΦ, called the space of finite (or order continuous) elements, is defined by
hΦ:= {x ∈ lΦ: IΦ(λx) < ∞ for all λ > 0}.
We say a Musielak-Orlicz function Φ = (Φi) satisfies the δ2-condition (Φ ∈ δ2) if there exist constants K ≥ 2, u0> 0 and a sequence (ci) of positive numbers such thatP∞
i=1ci< ∞ and the inequality
Φi(2u) ≤ KΦi(u) + ci holds for every i ∈ N and u ∈ R satisfying Φi(u) ≤ u0.
It is well known that hΦ= lΦif and only if Φ ∈ δ2(see [8]).
We also say a Musielak-Orlicz function Φ = (Φi) satisfies the condition (∗) if for any ε ∈ (0, 1) there exists a δ > 0 such that, for all i ∈ N and u ∈ R, Φi((1 + δ)u) ≤ 1 whenever Φi(u) ≤ 1 − ε (see [11]).
The paper is organized as follows: In Section 2, we give a criteria for a point in the unit sphere of Musielak-Orlicz sequence space lΦ and of the subspace hΦ to be a k- extreme point and then k-strict convexity of lΦand of hΦare characterized. Uniform convexity in every direction of lΦ and hφare equivalent and presented in Section 3.
Also, property K, property H, and property G of lΦand of hΦ are characterized in Section 4. Finally, in Section 5, we conclude all corresponding results in the Nakano sequence spaces.
2. k-Strict Convexity. Let X be a Banach space. Denoted by S(X) and B(X) the unit sphere and the unit ball of X, respectively. A point x ∈ S(X) is called an extreme point if for any two elements x1 and x2 in B(X) satisfying x = x1+x2 2 we have that x1= x2.
A point x ∈ S(X) is called a k-extreme point, for k ∈ N, if for any k + 1 elements x1, x2, . . . , xk+1 in B(X) satisfying x =x1+x2+···+xk+1 k+1 implies that the set {x1, x2, . . . , xk+1} is linearly dependent. It is clear from the definition that if the dimension of a Banach space X is less than or equal to k, then every point in S(X) is always a k-extreme point.
Remark 2.1 With the above notation we have
1. A point x ∈ S(X) is a 1-extreme point if and only if it is an extreme point.
2. If a point x in S(X) is a k-extreme point, then it is also a (k + 1)-extreme point.
An interval [a, b] is called a structural affine interval (SAI) of an Orlicz function ϕ if ϕ is affine on [a, b], i.e.,
ϕ(λa + (1 − λ)b) = λϕ(a) + (1 − λ)ϕ(b)
for all λ ∈ [0, 1], but not affine either on [a − ε, b] or [a, b + ε] for any ε > 0. Let {[an, bn]}nbe the set of all SAIs of ϕ. Put
SCϕ= R \ ∪n(an, bn).
Let aϕ= sup{u ∈ R : ϕ(u) = 0}.
Theorem 2.2 A point x = (x(i)) ∈ S(lΦ) is a k-extreme point if and only if the following conditions are satisfied:
(i) IΦ(x) = 1,
(ii) #{i ∈ N : |x(i)| ∈ [0, aΦi)} ≤ k − 1 and (iii) #{i ∈ N : x(i) 6∈ SCΦi} ≤ k.
In particular, a point x = (x(i)) ∈ S(lΦ) is an extreme point if and only if IΦ(x) = 1,
#{i ∈ N : |x(i)| ∈ [0, aΦi)} = 0, and #{i ∈ N : x(i) 6∈ SCΦi} ≤ 1. Here #A denotes the cardinality of the set A.
Proof Necessity. We note that if x = (x(i)) ∈ S(lΦ) is a k-extreme point, then (|x(i)|) is also a k-extreme point. So we may assume that x(i) ≥ 0 for all i ∈ N.
Suppose that (i) does not hold, i.e., IΦ(x) < 1. By the continuity of each Φi, there exists ε > 0 such that
Φi(x(i) ± ε) ≤ Φi(x(i)) +1 − IΦ(x)
2 ,
for all i = 1, . . . , k + 1. Without loss of generality, we assume in addition that x(1) > 0. We define
x1= (x(1) + ε)e1+ (x(k + 1) − ε)ek+1+ X
i6=1,k+1
x(i)ei
and, for each j ∈ {2, . . . , k + 1}, we also define
xj= (x(j − 1) − ε)ej−1+ (x(j) + ε)ej+ X
i6=j−1,j
x(i)ei.
Here eiis the sequence which has the i-th term 1 and all other terms 0. It is easy to see that x =x1+x2k+1+···+xk+1 and {x1, . . . , xk+1} ⊂ B(lΦ). To prove that x1, . . . , xk+1 are linearly independent, let a1, . . . , ak+1∈ R be such that a1x1+· · ·+ak+1xk+1= 0.
In particular, we have
k+1
X
i=1
ai
!
x(1) + (a1− a2)ε = 0,
k+1
X
i=1
ai
!
x(2) + (a2− a3)ε = 0, ...
k+1
X
i=1
ai
!
x(k + 1) + (ak+1− a1)ε = 0.
Clearly, (Pk+1
i=1 ai)(Pk+1
i=1x(i)) = 0. Knowing that Pk+1
i=1 x(i) ≥ x(1) > 0, thus Pk+1
i=1ai= 0 and this gives (a1− a2)ε = (a2− a3)ε = · · · = (ak+1− a1)ε = 0. Hence a1= · · · = ak+1 = 0 and x1, . . . , xk+1 are proved to be linearly independent. This implies that x is not a k-extreme point, a contradiction.
Suppose (iii) does not hold. Without loss of generality, we assume x(i) 6∈ SCΦi for all i ∈ N where N = {1, . . . , k + 1}. Hence x(i) ∈ (bi, ci) where [bi, ci] is a structural affine interval of Φi. Let Φi(u) = αiu + βi for u ∈ (bi, ci) for some constants αi≥ 0 and βi∈ R for each i ∈ N. We note that βi= 0 whenever αi= 0.
Let A = {i ∈ N : αi> 0}. Consider the following cases:
Case 1: #(A) ≥ 2. For simplicity, we assume th7at A = {1, . . . , m} where 2 ≤ m ≤ k + 1. Choose ε1, . . . , εk+1> 0 such that
α1ε1= · · · = αmεm and x(i) ± εi∈ (ai, bi) for all i ∈ N.
We define
x1 = (x(1) + ε1)e1+ (x(m) − εm)em+ X
i6=1,m
x(i)ei, xj = (x(j − 1) − εj−1)ej−1+ (x(j) + εj)ej+ X
i6=j−1,j
x(i)ei for j = 2, . . . , m − 1,
xm = (x(m − 1) − εm−1)em−1+ (x(m) + εm)em +(x(m + 1) + εm+1)em+1+ X
i6=m−1,m,m+1
x(i)ei xj = (x(j + 1) + εj+1)ej+1+ X
i6=j+1
x(i)ei for j = m + 1, . . . , k, and finally, xk+1 =
k+1
X
i=m+1
(x(i) − εi)ei+ X
i6=m+1,...,k+1
x(i)ei.
It is easy to see that x = x1+···+xk+1k+1. Moreover, IΦ(x1) = · · · = IΦ(xk+1) = 1.
Indeed, by the fact that α1ε1= αmεm, we have
IΦ(x1) = Φ1(x(1) + ε1) + Φm(x(m) − εm) + X
i6=1,m
Φi(x(i))
= α1x(1) + α1ε1+ β1+ αmx(m) − αmεm+ βm+ X
i6=1,m
Φi(x(i))
= Φ1(x(1)) + Φm(x(m)) + X
i6=1,m
Φi(x(i)) = IΦ(x) = 1.
Similarly, we also have IΦ(xj) = 1 for all j = 2, . . . , k + 1. Now we prove that x1, . . . , xk+1 are linearly independent and as a consequence we obtain a contradic- tion. Let a1, . . . , ak+1∈ R be such that a1x1+ · · · + ak+1xk+1= 0. Hence
a1x1(i) + · · · + ak+1xk+1(i) = 0
for all i ∈ N. In particular, we have
k+1
X
i=1
ai
!
x(1) + (a1− a2)ε1 = 0,
k+1
X
i=1
ai
!
x(2) + (a2− a3)ε2 = 0, ...
k+1
X
i=1
ai
!
x(m) + (am− a1)εm = 0.
Combining all these we have
k+1
X
i=1
ai
! x(1)
ε1 + · · · +x(m) εm
= 0.
Since x(1)ε
1 + · · · +x(m)ε
m 6= 0, Pk+1
i=1ai= 0. Therefore a1= · · · = am. Furthermore, for all j = m, . . . , k + 1, we have
0 =
k+1
X
i=1
ai
!
x(j) + (aj− ak+1)εj= (aj− ak+1)εj.
Again we obtain am= · · · = ak+1 and so a1= · · · = ak+1= 0.
Case 2: #(A) = 1. We assume that A = {1}. Choose ε2, . . . , εk+1> 0 such that x(i) ± εi∈ (ai, bi) for i = 2, . . . , k + 1.
We define, for j = 1, . . . , k,
xj = (x(j + 1) + εj+1)ej+1+ X
i6=j+1
x(i)ei and
xk+1 =
k+1
X
j=2
(x(j) − εj)ej+ X
i6=2,...,k+1
x(i)ei.
Clearly, x =x1+···+xk+1k+1, and IΦ(x1) = · · · = IΦ(xk+1) = 1. To prove the linear inde- pendence of x1, . . . , xk+1, let a1, . . . , ak+1∈ R be such that a1x1+· · ·+ak+1xk+1= 0.
Hence a1x(1) + · · · + ak+1x(1) = 0. Note that x(1) > 0. Otherwise, x(1) ∈ SCΦ1 which contradicts to our assumption. Thus,Pk+1
i=1ai= 0 and so 0 =
k+1
X
i=1
ai
!
x(j + 1) + (aj− ak+1)εj+1= (aj− ak+1)εj+1
for all j = 1, . . . , k. Therefore a1= · · · = ak+1= 0.
Case 3: #(A) = 0. Since IΦ(x) = 1, there exists i06∈ N such that Φi0(x(i0)) > 0.
Let us consider the following subcases.
Subcase 3.1: x(i0) 6∈ SCΦ
i0. If we put A0= (A \ {1}) ∪ {i0} and repeat the proof of Case 2, then we obtain a contradiction.
Subcase 3.2: x(i0) ∈ SCΦ
i0. Choose ε1, . . . , εk+1> 0 such that x(i) ± εi∈ (bi, ci) for i = 1, . . . , k + 1.
Define
x1= (x(1) + ε1)e1+ (x(k + 1) − εk+1)ek+1+ X
i6=1,k+1
x(i)ei
and, for each j ∈ {2, . . . , k + 1}, we also define
xj= (x(j − 1) − εj−1)ej−1+ (x(j) + εj)ej+ X
i6=j−1,j
x(i)ei.
Again, we have x1, . . . , xk+1∈ S(lΦ) and x =x1+···+xk+1k+1. We now prove the linear independence of x1, . . . , xk+1. If a1x1+ · · · + ak+1xk+1= 0 where a1, a2, . . . , ak+1∈ R, then a1x1(i) + · · · + ak+1xk+1(i) = 0 for all i ∈ N. Since x(i0) ∈ SCΦ
i0 and IΦ(x1+···+xk+1k+1) = IΦ(x) = 1, we have x1(i0) = · · · = xk+1(i0) = x(i0) > 0. This implies that a1+ a2+ · · · + ak+1= 0. Now, for all j = 1, . . . , k,
aj− aj+1= (aj− aj+1)εj+ (a1+ a2+ · · · + ak+1)x(j) = 0.
Hence, we get a1= · · · = ak+1= 0.
In all cases we encounter with contradictions since x is a k-extreme point and thus the necessity of (iii) is established.
To prove (ii). Suppose that x(i) ∈ [0, aΦi) for all i = 1, . . . , k. Choose ε > 0 so that x(i) ± ε ∈ (−aΦi, aΦi) for all i = 1, . . . , k. For j = 1, . . . , k, we define
xj= (x(j) − ε)ej+X
i6=j
x(i)ei
and
xk+1=
k
X
i=1
(x(i) + ε)ei+
∞
X
i=k+1
x(i)ei.
Obviously x =x1+x2k+1+···+xk+1 and {x1, . . . , xk+1} ⊂ S(lΦ). Now we prove the linear independence of these elements. If a1x1+ · · · + ak+1xk+1 = 0, then a1x1(i) + · · · + ak+1xk+1(i) = 0 for all i ∈ N. Since IΦ(x) = 1, there exists an index i0≥ k such that Φi0(x(i0)) > 0. This implies that x(i0) 6= 0. It follows from (iii) that x(i0) ∈ SCΦi0, and then that x1(i0) = · · · = xk+1(i0) = x(i0) > 0. Hence a1+ · · · + ak+1 = 0.
Moreover, we have
0 = a1x(1) − a1ε + a2x(1) + · · · + aix(1) + ak+1x(1) − ak+1ε
= a1x(1) + a2x(1) + · · · + akx(1) + ak+1x(1) + ak+1ε − a1ε
= ak+1ε − a1ε.
This gives a1 = ak+1. Similarly, we have aj = ak+1 for j = 2, . . . , k. Hence a1= · · · = ak+1= 0. Therefore x cannot be a k-extreme point.
Sufficiency. Let x ∈ S(lΦ) be such that the conditions (i)-(iii) hold. Take elements x1, . . . , xk+1 in the unit sphere of lΦ with
x = x1+ x2+ · · · + xk+1
k + 1 .
By the condition (i) and the convexity of the modular, we obtain IΦ(x1) = · · · = IΦ(xk+1) = 1. Furthermore, for each i ∈ N, {xj(i) : j = 1, . . . , k + 1} is either a singleton or a set contained in the same SAI of Φi. To prove that {x1, . . . , xk+1} is linearly dependent, we shall find a1, . . . , ak+1∈ R such that a1x1+· · ·+ak+1xk+1= 0 where ai’s are not all zero. It follows by the condition (iii) that, for all but k coordinates, {xj(i) : j = 1, . . . , k + 1} is a singleton. For the sake of convenience we assume that {xj(i) : j = 1, . . . , k + 1} is a singleton for all i ≥ k + 1. Then
(?) IΦ
k
X
i=1
x1(i)ei
!
= · · · = IΦ
k
X
i=1
xk+1(i)ei
! .
We also assume in the worst case that {i ∈ N : x(i) 6∈ SCΦi} = {1, . . . , k}. Let {i ∈ N : |x(i)| ∈ [0, aΦi)} = {1, . . . , m} where m ≤ k − 1 and let K = {1, . . . , k} \ {1, . . . , m}. If x(i) = 0 for all i ≥ k + 1, the following system of equations
a1x1(1) + a2x2(1) + · · · + ak+1xk+1(1) = 0, a1x1(2) + a2x2(2) + · · · + ak+1xk+1(2) = 0,
... a1x1(k) + a2x2(k) + · · · + ak+1xk+1(k) = 0.
always has a nontrivial solution. On the other hand, if there exists a coordinate i ≥ k + 1 such that x(i) 6= 0, then
a1+ a2+ · · · + ak+1= 0.
Consider the matrix
x1(1) x2(1) · · · xk+1(1) x1(2) x2(2) · · · xk+1(2)
... ... . .. ... x1(k) x2(k) · · · xk+1(k)
1 1 · · · 1
.
For k ∈ K, let Φk(u) = αku + βk when u ∈ [bk, ck], where [bk, ck] is a structural affine interval of Φk containing x(k), αk> 0 and βk∈ R. By (?), we have
X
k∈K
(αkx1(k) + βk) = X
k∈K
(αkx2(k) + βk) = · · · = X
k∈K
(αkxk+1(k) + βk).
This implies that the above matrix is equivalent to this following matrix
x1(1) x2(1) · · · xk+1(1)
x1(2) x2(2) · · · xk+1(2)
... ... . .. ...
x1(m) x2(m) · · · xk+1(m)
Φm+1(x1(m + 1)) Φm+1(x2(m + 1)) · · · Φm+1(xk+1(m + 1))
... ... . .. ...
Φk(x1(k)) Φk(x2(k)) · · · Φk(xk+1(k))
0 0 · · · 0
.
Then there exists a nontrivial solution {ai: i = 1, . . . , k + 1} for the above system.
This implies the linear dependence of {x1, . . . , xk+1}. A Banach space X is said to be k-strictly convex if each point in its unit sphere is a k-extreme point (see [16]). Also, a Banach space X is said to be strictly convex if each point in its unit sphere is an extreme point.
Let σi = sup{u ≥ 0 : Φiis strictly convex on [0, u] and Φi(u) ≤ 1}. By the previous theorem, we obtain the following characterizations.
Corollary 2.3 The Musielak-Orlicz sequence space lΦ is k-strictly convex if and only if the following conditions are satisfied
1. Φ = (Φi) satisfies the δ2-condition,
2. each Φi vanishes only at zero for all but k − 1 indices i’s and
3. Φi1(σi1) + Φi2(σi2) + · · · + Φik+1(σik+1) ≥ 1 for all k + 1 distinct indices i1, i2, . . . , ik+1.
In particular, hΦ is k-strictly convex if and only if the conditions (2) and (3) are satisfied.
3. Uniform Convexity in Every Direction. A Banach space X is said to be uniformly convex in every direction (UCED) if for each ε ∈ (0, 2] and any z ∈ S(X) there is a δ = δ(ε, z) > 0 such that for any x, y ∈ S(X) with x − y = αz for some scalar α, the condition kx − yk ≥ ε implies that kx + yk ≤ 2(1 − δ). Equivalently, if xn, z ∈ X, kxnk, kxn+zk → 1 and k2xn+zk → 2 imply z = 0 (see [17, Proposition 1, page 7]). This property was introduced by A. L. Garkavi in 1962 (see [6]). Moreover, he showed that every UCED space has weak normal structure and hence enjoys the fixed point property (see also [1, 7]).
It is easy to see that every UCED space is strictly convex.
Lemma 3.1 ([9]) Let vi∈ R, i = 1, . . . , 4 and v1< v2 < v3 < v4. If ϕ is strictly convex on [v2, v3], then there exists p ∈ (0, 1) such that
ϕ u + v 2
≤ 1 − p
2 (ϕ(u) + ϕ(v)) for all u ∈ [v1, v2] and v ∈ [v3, v4].
Lemma 3.2 ([10]) Let ϕ be strictly convex on [−a, a]. Then for each ε > 0, d1, d2∈ (0, a], d1< d2, there exists p ∈ (0, 1) such that
ϕ u + v 2
≤ 1 − p
2 (ϕ(u) + ϕ(v)) if all |u − v| ≥ ε max(|u|, |v|) and max(|u|, |v|) ∈ [d1, d2].
Lemma 3.3 ([11]) If the Musielak-Orlicz function Φ satisfies the δ2-condition and the condition (∗), then for each ε > 0, there exists δ > 0 such that kxk ≤ 1 − δ whenever IΦ(x) ≤ 1 − ε.
Theorem 3.4 The following statements are equivalent:
1. lΦ is UCED;
2. hΦ is UCED;
3. the following conditions are satisfied:
(a) Φ satisfies the δ2-condition and the condition (∗), (b) each Φi vanishes only at zero,
(c) Φi(σi) + Φj(σj) ≥ 1 for all i 6= j.
Proof (1)⇒(2) is trivial. To prove (2)⇒(3), it suffices to prove only the necessity of (a). Because (2) implies that hΦis strictly convex. Suppose first that Φ 6∈ δ2, then there exists x = (x(i)) ∈ S(lΦ) such that IΦ(x) ≤ ε0< 1 and IΦ(λx) = ∞ for all λ > 1. We can find a strictly increasing sequence 0 = i1< i2< · · · < in< in+1< · · · of nonnegative integers so that
IΦ n + 1 n
in+1
X
i=in+1
x(i)ei
!
> 1
for all n ∈ N. Then, for all n ∈ N,
1 ≥
in+1
X
i=in+1
x(i)ei
> n n + 1. Define xn = Pin+1
i=in+1x(i)ei. Then kxnk → 1. We may assume that x(1) 6= 0.
Put z = x(1)e16= 0. Then we have kxn+ zk → 1 and k2xn+ zk → 2. This is a contradiction.
We next prove that Φ satisfies the condition (∗). For an arbitrary ε ∈ (0, 1) and i 6= 1, let u ∈ R be such that Φi(u) ≤ 1 − ε. Put z = 2ae1where Φ1(a) = ε. Since hΦ is UCED, there exists δ0 > 0 such that kx +z2k ≤ 1 − δ0 for any x ∈ hΦ with kxk, kx + zk ≤ 1. If we put x = uen− ae1, then kxk ≤ 1, kx + zk = 1. Hence kuenk = kx + z2k ≤ 1 − δ0. This implies that Φi 1−δu0 ≤ 1 for all i 6= 1. By the
continuity of Φ1, if Φ1(u) ≤ 1 − , there exists δ00 > 0 such that Φ1((1 + δ00)u) ≤ 1.
Put δ = min δ0
1−δ0, δ00 . Then the necessity of the condition (∗) is proved.
(3)⇒(1) Let z = (z(i)) ∈ lΦ be a nonzero element. Consider the set A = {x = (x(i)) : IΦ(x) = 1 and IΦ(x + z) ≤ 1}.
We first consider these two following cases:
I : There exists an index k such that σk> 0, z(k) 6= 0, |x(k)| ≤ σk and |x(k) + z(k)| ≤ σk
II : There exist an index k and numbers t1, t2 ∈ (−Φ−1k (1), Φ−1k (1)), t1 < t2, σk> 0 such that
(i) Φk is strictly convex on [t1, t2],
(ii) x(k) ≤ t1and x(k) + z(k) ≥ t2or x(k) + z(k) ≤ t1and x(k) ≥ t2, and (iii) Φk(x(k)) ≥ Φk(σk) or Φk(x(k) + z(k)) ≥ Φk(σk).
We will estimate the value of IΦ x +z2.
I : Let n ∈ N be such that Φk(z(k)) ≤ n. Then
|z(k)| ≥Φk(z(k))
n max(|x(k)|, |x(k) + z(k)|).
Otherwise, since Φk vanishes only at zero, Φk(z(k)) < Φk(z(k))
n max(Φk(x(k)), Φk(x(k) + z(k)))
≤ Φk(z(k)) n Φk(σk)
≤ Φk(z(k)) n
which is impossible. Moreover we also have
|z(k)|
2 ≤ max(|x(k)|, |x(k) + z(k)|) ≤ σk.
Now we apply Lemma 3.2 with Φk(z(k))n , |z(k)|2 , σk in place of ε, d1, d2, respectively.
Then there exists pk∈ (0, 1) such that Φk
x(k) +z(k) 2
≤ 1 − pk
2 (Φk(x(k)) + Φk(x(k) + z(k))).
This implies IΦ
x +z 2
≤ 1 −pk
2(Φk(x(k)) + Φk(x(k) + z(k)))
≤ 1 −pk
2 max(Φk(x(k)), Φk(x(k) + z(k)))
≤ 1 −pk
2Φk z(k) 2
.
II : Applying Lemma 3.1 with −Φ−1k (1), t1, t2, Φ−1k (1) in place of vi, respectively, we obtain pk∈ (0, 1) such that
Φk
x(k) +z(k) 2
≤ 1 − pk
2 (Φk(x(k)) + Φk(x(k) + z(k))).
This implies IΦ
x +z
2
≤ 1 −pk
2(Φk(x(k)) + Φk(x(k) + z(k)))
≤ 1 −pk 2Φk(σk).
Without loss of generality, we assume that
Φ1(z(1)) = max{Φi(z(i)) : i ∈ N}, Φ2(z(2)) = max{Φi(z(i)) : i ∈ N, i 6= 1}, and define the following sets
A1 = {x ∈ A : |x(1)| ≤ σ1and |x(1) + z(1)| ≤ σ1}, A2 = {x ∈ A : |x(1)| > σ1and |x(1) + z(1)| > σ1}, A3 = {x ∈ A : |x(1)| ≤ σ1and |x(1) + z(1)| > σ1}, and A4 = {x ∈ A : |x(1)| > σ1and |x(1) + z(1)| ≤ σ1}.
It is evident that A = ∪4i=1Ai.
We note that if z(i) = 0 for all i ≥ 2, then IΦ
x +z 2
= Φ1
x(1) +z(1) 2
+
∞
X
i=2
Φi(x(i))
= Φ1
x(1) +z(1) 2
+ 1 − Φ1(x(1))
≤ 1 − Φ1 z(1) 2
.
From now on, we may assume that z(2) 6= 0.
First, if σ1= 0, then by (c) we have Φi(σi) = 1 for all i ≥ 2. We apply I with k = 2.
Secondly, if Φ1(σ1) = 1, then we shall apply the case I with k = 1.
We now assume that σi > 0 for all i ∈ N. In the virtue of I it is enough to consider only the sets A2, A3and A4.
Suppose that the numbers x(1) and x(1)+z(1) are of the different sign. Then, for such x from A2∪A3∪A4, it falls in the case II when k = 1 by putting t1, t2∈ {±σ1, 0}.
Now assume that x(1) and x(1) + z(1) are of the same sign. If x ∈ A2 then Φ2(x(2)) ≤ Φ2(σ2) and Φ2(x(2) + z(2)) ≤ Φ2(σ2) since Φ1(σ1) + Φ2(σ2) ≥ 1. There- fore, case I is applicable for k = 2. Now let x ∈ A3. Note that the signs of x(1) and z(1) must be the same. Let m ∈ N such that σ1−|z(1)|m > 0 and let
B3=
x ∈ A3: |x(1)| ≤ σ1−|z(1)|
m
.
Putting t1, t2as ±(σ1−|z(1)|m ), ±σ1, so elements in B3satisfy the assumption of II.
Denoted by fB3the complement of B3in A3i.e.
Bf3=
x ∈ A3: |x(1)| > σ1−|z(1)|
m
.
Then |x(1) + z(1)| = |x(1)| + |z(1)| > σ1−|z(1)|m + |z(1)| = σ1+(m−1)|z(1)|
m . There- fore, Φ1(x(1) + z(1)) > Φ1
σ1+(m−1)|z(1)|
m
, which implies Φ2(x(2) + z(2)) ≤ 1 − Φ1
σ1+(m−1)|z(1)|
m
< Φ2(σ2). If |x(2)| ≤ σ2then we are in case I with k = 2.
If |x(2)| > σ2then we are in case II for k = 2 with t1, t2are chosen respectively from
±
Φ−12
1 − Φ1
σ1+(m − 1)|z(1)|
m
, ±σ2.
For x ∈ A4we also make analogous considerations. Note that x(1) and z(1) must have the different signs. However x(1) and x(1) + z(1) have the same sign. Thus
|x(1)| > |z(1)| and |x(1) + z(1)| = |x(1)| − |z(1)|. Let B4=
x ∈ A4: |x(1) + z(1)| ≤ σ1−|z(1)|
m
. If x ∈ B4then the conditions of Case II are satisfied. Put
Bf4=
x ∈ A4: |x(1) + z(1)| > σ1−|z(1)|
m
.
Therefore |x(1)| − |z(1)| = |x(1) + z(1)| > σ1−|z(1)|m which implies |x(1)| > σ1+
(m−1)|z(1)|
m . Hence Φ2(x(2)) ≤ 1 − Φ1
σ1+(m−1)|z(1)|
m
< Φ2(σ2). If |x(2) + z(2)| ≤ σ2then we are in case I with k = 2. If |x(2) + z(2)| > σ2, then we are in case II for k = 2 with t1, t2are chosen respectively from
±
Φ−12
1 − Φ1
σ1+(m − 1)|z(1)|
m
, ±σ2.
Thus, for all x ∈ A, we have that IΦ x +z2 ≤1−p2 for some p > 0. The number p depends only on z. Indeed, p depends on the numbers
±σ1, ±σ2, 0, ±Φ−12
1 − Φ1
σ1+(m − 1)|z(1)|
m
, and
±
σ1−|z(1)|
m
.
Hence by the condition (∗) and the δ2-condition there exists δ > 0 such that x +z2
≤ 1 − δ for all x ∈ S(lΦ) with kx + zk ≤ 1. The proof is now complete.
4. Property K, Property H and Property G. A point x ∈ S(X) is called an H-point if xn→ x whenever (xn) ⊂ X such that kxnk → 1 and xn→ x. A pointw x ∈ S(X) is called a PC-point if the identity map id : B(X) → B(X) is weak-to- norm continuous at x. Equivalently, for any ε > 0 there exist δ > 0 and finitely many linear functionals x∗1, x∗2, . . . , x∗n∈ X∗such that
ky − xk < ε
whenever kyk ≤ 1 and |x∗i(y − x)| < δ for all i = 1, 2, . . . , n.
It is easy to see that every PC-point is an H-point. Moreover, if X is reflexive, both notions are the same ([1]).
Lemma 4.1 ([11]) If the Musielak-Orlicz function Φ = (Φi) satisfies the δ2-condition and the condition (∗) and each Φi vanishes only at zero, then for each ε > 0, there exist δ > 0 such that |IΦ(x) − IΦ(y)| < ε whenever IΦ(x) ≤ 1, IΦ(y) ≤ 1 and IΦ(x − y) ≤ δ .
Lemma 4.2 ([15]) If Φ does not satisfy the δ2-condition, then S(lΦ) contains no H-points.
Theorem 4.3 Suppose that a Musielak-Orlicz function Φ satisfies the condition (∗) and each Φi vanishes only at zero. Then the following statements are equivalent:
1. x ∈ S(lΦ) is a PC-point;
2. x is an H-point;
3. Φ ∈ δ2.
Proof (1)⇒(2) is obvious. See [15] for a proof of the implication (2)⇒(3).
(3)⇒(1) Suppose Φ ∈ δ2. Given ε > 0. There exists δ ∈ (0, ε) such that kyk < ε
2 whenever IΦ(y) ≤ 2δ and there exists δ0∈ (0, δ) such that
|IΦ(y) − IΦ(z)| < δ whenever IΦ(y − z) ≤ δ0, IΦ(y) ≤ 1, IΦ(z) ≤ 1.
Choose i0∈ N so thatP∞
i=i0+1Φi(x(i)) < δ. Note that α := min
i=1,...,i0
Φ−1i (δ0 i0) > 0.
Put Aδ= {y ∈ B(lΦ) : |hy − x, eii| = |y(i) − x(i)| < α for all i = 1, . . . , i0}. For any y ∈ Aδ, we have
i0
X
i=1
Φi(y(i) − x(i)) <
i0
X
i=1
Φi(α) ≤ δ0.
Moreover, for y ∈ Aδ, we also have
∞
X
i=i0+1
Φi(y(i)) ≤ 1 −
i0
X
i=1
Φi(y(i))
=
i0
X
i=1
Φi(x(i)) −
i0
X
i=1
Φi(y(i)) +
∞
X
i=i0+1
Φi(x(i)) ≤ 2δ.
These yield
IΦ y − x 2
≤
i0
X
i=1
Φi y(i) − x(i) 2
+1
2
∞
X
i=i0+1
Φi(y(i)) + Φi(x(i))
!
≤ 2δ.
Hence ky − xk < ε, i.e. Aδ⊂ x + εB(lΦ). Therefore x is a PC-point.
A Banach space X is said to have property H (property K, resp.) if each point in its unit sphere is an H-point (PC-point, resp.). Sometimes, the property H is also called the Randon-Riesz property or the Kadets-Klee property. See [12] for references concerning the history and related results.
Corollary 4.4 Suppose that a Musielak-Orlicz function Φ satisfies the condition (∗) and each Φi vanishes only at zero. Then the following statements are equivalent:
1. lΦ has property K;
2. hΦ has property K;
3. lΦ has property H;
4. hΦ has property H;
5. Φ ∈ δ2.
A point x ∈ S(X) is called a denting point if for any ε > 0, x 6∈ co{B(X)\(x + εB(X))}. Recall that co(A) denotes the closed convex hull of A. If each point in S(X) is a denting point, we say that X has property G. The reader who is interested in a discussion of the relevance of denting points in connection with the Randon- Nikodym property is referred to the monographs [2] and [4].
Recently, B.-L. Lin, et al. ([14]) proved that x ∈ S(X) is a denting point if and only if it is a PC-point and an extreme point (see [14]). This gives the following characterizations:
Theorem 4.5 Suppose that a Musielak-Orlicz function Φ satisfies the condition (∗) and each Φi vanishes only at zero. Then x = (x(i)) ∈ S(lΦ) is a denting point if and only if Φ ∈ δ2and #{i ∈ N : x(i) 6∈ SCΦi} ≤ 1.
In particular, the following statements are equivalent:
1. lΦ has property G;
2. hΦ has property G;
3. lΦ is strictly convex.
5. Convexity Properties in Nakano Sequence Spaces. In this section, we give the characterizations of properties in the previous sections for Nakano sequence spaces. Recall that a Nakano sequence space l{pi} is a Musielak-Orlicz sequence space with
Φi(u) = |u|pi where 1 ≤ pi< ∞.
Theorem 5.1 For the Nakano sequence space l{pi}, we have
1. ([5, Theorem 3]) l{pi} is k-strictly convex if and only if #{i ∈ N : pi= 1} ≤ k and lim sup
i→∞
pi< ∞,
2. ([5, Theorem 22 and Final remark]) l{pi} is UCED if and only if l{pi} has property G; if and only if #{i ∈ N : pi= 1} ≤ 1 and lim sup
i→∞
pi< ∞, and 3. ([5, Theorem 6 and Final remark]) l{pi}has property K if and only if l{pi}has
property H; if and only if lim sup
i→∞
pi< ∞.
Acknowledgement. The author would like to thank Professor Sompong Dhom- pongsa for his helpful discussion and comments.
References
[1] A. G. Aksoy and M. A. Khamsi, Nonstandard methods in fixed point theory, Springer-Verlag, New York 1990.
[2] R. D. Bourgin, Geometric aspects of convex sets with the Radon-Nikod´ym property, Lecture Notes in Math. No. 993, Springer-Verlag, Berlin 1983.
[3] S. Chen, Geometry of Orlicz spaces, Dissertationes Math.(Rozprawy Matematyczne) 356, 1996.
[4] J. Diestel and J. J. Uhl, Vector measures, Mathematical Surveys, No. 15, American Mathe- matical Society, Providence, R.I. 1977.
[5] S. Dhompongsa, Convexity properties of Nakano spaces, Science Asia 26 (2000), 21-31.
[6] A. L. Garkavi, On the optimal net and best cross-section of a set in a normed space. (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 26 (1962) 87–106.
[7] K. Goebel and W. A. Kirk, Topics in metric fixed point theory, Cambridge Studies in Advanced Mathematics, 28. Cambridge University Press, Cambridge 1990.
[8] A. Kami´nska, Rotundity of sequence Musielak-Orlicz spaces, Bull. Polish Acad. Sci. Math. 29 (1981), 137-144.
[9] A. Kami´nska, On uniform convexity of Orlicz spaces, Nederl. Akad. Wetensch. Indag. Math.
44 (1982), 27-36.
[10] A. Kami´nska, The criteria for local uniform rotundity of Orlicz spaces, Studia Math., 79 (1984), 201-215.
[11] A. Kami´nska, Uniform rotundity of Musielak-Orlicz sequene spaces, J. Approx. Theory, 47 (1986), 302-322.
[12] R. E. Megginson, An intorduction to Banach space theory, Graduate Texts in Mathematics, 183, Springer-Verlag, New York 1998.
[13] J. Musielak, Orlicz spaces and modular spaces, Lecture Notes in Math. No. textbf1034, Springer-Verlag, Berlin 1983.
[14] B.-L. Lin, P.-K. Lin and S. L. Troyanski, Characterizations of denting points, Proc. Amer.
Math. Soc., 102 (1988), 526-528.
[15] S. Saejung and S. Dhompongsa, Extreme points in Musielak-Orlicz sequence spaces, Acta Math. Vietnamica, 7:2 (2002), 219-229.
[16] I. Singer, On the set of the best approximations of an element in a normed linear space, Rev.
Math. Pures Appl. 5 (1960), 383-402.
[17] V. Zizler, On some rotundity and smoothness properties of Banach spaces, Dissertationes Math.(Rozprawy Matematyczne) 87 (1971).
Satit Saejung
Department of Mathematics, Faculty of Science, Khon Kaen University Khon Kaen, 40002, Thailand
E-mail: satitz@yahoo.com
(Received: 7.01.2001; revised: 16.02.2005)