arXiv:math/9804067v1 [math.FA] 14 Apr 1998 xml fawa ibr pc hc sntaHletsaei convexifie is space Hilbert a not is which space Hilbert weak a and of them spaces, example dual called subspaces, and to passing 2 under spaces stable cotype studied are weak spaces [P1] and Pisier 2 Later, type [MP]. weak Pisier of G. and Milman V. by studied otatMM-703/97. contract Abstract: faBnc pc hc osntcnana smrhccp of copy isomorphic an contain not does [T] which Tsirelson space B.S. Banach by introduced a was of space Tsirelson’s P1]. J2, J1, [CS, h upiigfc that fact surprising the aea motn nltcdsrpino h omdet iiladJ and Figiel to due norm introduced the Johnson of description analytic important an have oa,w eoeby denote we Today, hspit l h o-rva xmlso ekHletsae ie th f which (i.e. subspaces had spaces and Hilbert bases weak unconditional of had examples spaces) non-trivial Hilbert the all point, this digo E nrdcdacaso ekHletsae ihuncondit with spaces Hilbert weak ℓ of class a introduced [E] Edgington ibr pc u h pc tefi o smrhct ibr space pro Hilbert [KT]) a Tomczak-Jaegermann to and isomorphic Komorowski not generally is more itself (or space the but space Hilbert a 2 strtd hti,eeysbpc ftesaecnan fur a contains space the of subspace every is, That -saturated. h hr uhrwsprilyspotdb h ugra iityof Ministry Bulgarian the by supported partially was author 9706108. DMS third NSF The by supported was author second The 1991 h aahsaeproperties space Banach The ahmtc ujc Classification. Subject Mathematics EREADOLKS EE .CSZA N EK .KUTZAROV N. DENKA AND CASAZZA, G. PETER ANDROULAKIS, GEORGE egv e xmlso ekHletspaces. Hilbert weak of examples new give We oie srlo space Tsirelson modified OEMR EKHLETSPACES HILBERT WEAK MORE SOME T h ulsaeo h rgnleapeo srlo ic in since Tsirelson of example original the of space dual the T M sntrlyioopi oteoiia srlo space Tsirelson original the to isomorphic naturally is ektp 2 type weak 1. 46B. Introduction 1 and T M ae,CszaadOel[O proved [CO] Odell and Casazza Later, . ekHletspaces Hilbert weak ekctp 2 cotype weak utetsae.Tecanonical The spaces. quotient hrsbpc smrhcto isomorphic subspace ther dcto n cec under Science and Education e htteeaeweak are there that ved ie ocontain to ailed c 0 oa ae hc are which bases ional eeitoue and introduced were .Kmrwk [K] Komorowski R. . hsn[J.I [J1], In [FJ]. ohnson srlo space Tsirelson d or stefis example first the as s hc r not are which ose ℓ p ekHilbert Weak . hc r both are which 1 , ≤ A < p T ℓ 2 T At . A. . ∞ we T 2 . 2 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA

Hilbert spaces with no unconditional basis. In fact, they show that T 2 has such subspaces. In another surprise, Nielsen and Tomczak-Jaegermann [NTJ] showed that all weak Hilbert spaces with unconditional bases are very much like T (2). There are still many open questions concerning weak Hilbert spaces and T (2), due partly to the shortage of non-trivial examples in this area. For example, it is still a major open question in the field whether a Banach space for which every subspace has an unconditional basis (or just local unconditional structure - LUST) must be isomorphic to a Hilbert space. If there are such examples, they will probably come from the class of weak Hilbert spaces. It is an open question whether every weak Hilbert space has a basis, although Maurey and Pisier (see [M]) showed that separable weak Hilbert spaces have finite dimensional decompositions. Nielsen and Tomczak-Jaegermann have shown that weak Hilbert spaces that are Banach lattices have the property that every subspace of every quotient space has a basis. But it is unknown whether every weak Hilbert space can be embedded into a weak Hilbert space with a unconditional basis. In fact, it is unknown if a weak Hilbert space embeds into a Banach lattice of finite cotype. It turns out that this question is equivalent to the question of whether every subspace of a weak Hilbert space must have the GL-Property [CN] which is slightly weaker than having LUST. In this note we extend the list of non-trivial examples of weak Hilbert spaces by producing examples which are ℓ2-saturated but have no subspaces isomorphic to subspaces of the previously known examples.

2. Basic Constructions

If F is a finite dimensional Banach space then let d(F ) denote the Banach-Mazur distance dimF between F and ℓ2 . The fundamental notion of this note is the one of the weak Hilbert space. Recall the following definition as one of the many equivalent ones (cf [P1] Theorem 2.1).

Definition 2.1. A Banach space X is said to be a weak Hilbert space if there exist δ > 0 and C ≥ 1 such that for every finite dimensional subspace E of X there exists a subspace F ⊆ E and a projection P : X → F such that dimF ≥ δdimE, d(F ) ≤ C and kP k ≤ C. WEAK HILBERT 3

We need to recall the definition of the Schreier sets Sn, n ∈ N [AA]. For F,G ⊂ N, we write F

S0 = {{n} : n ∈ N}∪{∅}.

If n ∈ N ∪{0} and Sn has been defined,

n Sn+1 = {∪1 Fi : n ∈ N, n ≤ F1 < F2 < ··· < Fn and Fi ∈ Sn for 1 ≤ i ≤ n}.

For n ∈ N a family of finite non-empty subsets (Ei) of N is said to be Sn-admissible if

E1 < E2 < ··· and (min(Ei)) ∈ Sn. Also, (Ei) is said to be Sn-allowable if Ei ∩ Ej = ∅ for i =6 j and (min(Ei)) ∈ Sk.

Every Banach space with a basis can be viewed as the completion of c00 (the linear space of finitely supported real valued sequences) under a certain norm. (ei) will denote the unit vector basis for c00 and whenever a Banach space (X, k·k) with a basis is regarded as the completion of (c00, k·k), (ei) will denote this (normalized) basis. If x ∈ c00 and E ⊆ N,

Ex ∈ c00 is the restriction of x to E; (Ex)j = xj if j ∈ E and 0 otherwise. Also the support of x, supp (x), (w.r.t. (ei)) is the set {j ∈ N : xj =6 0}. If f : R → R is a function with f(0) = 0 then for x ∈ c00 f(x) will denote the vector f(x)=(f(xi)) in c00.

Let (X, k.k) be a Banach space with an unconditional basis. The norm of X is 2-convex provided that

k(x2 + y2)1/2k ≤ (kxk2 + kyk2)1/2

(2) for all vectors x, y ∈ X. The 2-convexification of (X, k.k) is the Banach space (X , k.k(2)) with an unconditional basis, where x ∈ X(2) if and only if x2 ∈ X and

2 1/2 kxk(2) = kx k .

Of course k.k(2) is 2-convex. For C > 0, 1 ≤ p < ∞, n ∈ N and x1, x2,...,xn ∈ X we say n n that (xi)i=1 is C-equivalent to the unit vector basis of ℓp if there exist constants A,B > 0 4 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA with AB ≤ C such that 1 ( |a |p)1/p ≤k a x k ≤ B( |a |p)1/p A i i i i X n X X for every sequence of scalars (ai)i=1. For C > 0, we say that X is an asymptotic ℓp space

(resp. asymptotic ℓp space for vectors with disjoint supports) with constant C if for every n n n and for every sequence of vectors (xi)i=1 such that (supp (xi))i=1 is S1-admissible (resp. n n S1-allowable), we have that (xi)i=1 are C-equivalent to the unit vector basis of ℓp .

If (k.k)n is a sequence of norms in c00 then Σ(k.kn) will denote the completion of c00 under the norm ∞

kxkΣ(k.kn) = kxkn. n=1 X

αn Fix a sequence α =(α ) N of elements of (0, 1) and ℓ,u ∈ (0, 1) with 0 <ℓ ≤ +1 ≤ u< 1 n n∈ αn for all n and n αn = 1 (the existence of numbers ℓ,u ∈ (0, 1) such that the last relationships are valid willP always be assumed whenever a sequence (αn) will be considered in these notes).

Edgington defined a sequence of norms (k.kE,n) on c00 by

2 2 kxkE,0 = kxk∞, kxkE,n+1 = sup{ kEixkE,n :(Ei)i is S1-admissible}. i X

Then Edgington defined the norm k.kEα by 1/2 2 kxkEα = αnkxkE,n . n ! X Let Eα denote the completion of c00 with respect to k.kE. It is shown in [E] that Eα is a weak Hilbert space which is not isomorphic to ℓ2, yet it is ℓ2-saturated. It is easy to see that the spaces constructed by Edgington are asymptotic ℓ2 spaces for vectors with disjoint supports. The main theorem that we prove in these notes (Theorem 3.1) shows that such spaces are weak Hilbert spaces.

Let (|.|n)n∈N denote the sequence of the Schreier norms on c00:

|x|n = sup |xj| S∈S n j∈S X WEAK HILBERT 5

(if x = j xjej). Then the weak Hilbert space Eα that was constructed by Edgington

[E] is theP 2-convexification of Σ(αn|.|n). One can see that Σ(αn|.|n) is an asymptotic ℓ1 space for vectors with disjoint supports which is ℓ1-saturated, yet not isomorphic to ℓ1. In these notes we give examples of sequences of norms that can replace (|.|n) in Σ(αn|.|n) to obtain asymptotic ℓ1 spaces for vectors with disjoint supports which are ℓ1-saturated yet not isomorphic to ℓ1. The 2-convexification of each of these spaces will give ℓ2 saturated weak

Hilbert spaces which are not isomorphic to ℓ2.

′ ′ Definition of the spaces V , W , V and W : Let (θn)n∈N be a sequence of real numbers in

(0, 1) with limn θn = 0 (this assumption will always be valid whenever a sequence (θn) will be considered in these notes) and let s ∈ N. The asymptotic ℓ1 spaces V = TM (θn,Sn)n and

W = TM(s)(θn,Sn)n were introduced in [AD] and [ADKM] as the completion of c00 under the norms:

kxkV = kxk∞ ∨ sup sup{θn kEixkV :(Ei) is Sn allowable}, n i X kxkW = kxk∞ ∨ sup sup{θn kEixkW :(Ei) is Sn allowable} n≤s i X ∨ sup sup{θn kEixkW :(Ei) is Sn admissible}, n≥s+1 i X respectively. These norms can also be defined as limits of appropriate sequences. For x ∈ c00 let

kxkV,0 = kxkW,0 = kxk∞ and for m ∈ N define:

kxkV,m+1 = kxk∞ ∨ sup sup{θn kEixkV,m :(Ei) is Sn allowable}, n i X kxkW,m+1 = kxk∞ ∨ sup sup{θn kEixkW,m :(Ei) is Sn allowable} n≤s i X ∨ sup sup{θn kEixkW,m :(Ei) is Sn admissible}, n≥s+1 i X Then

kxkV = lim kxkV,m, kxkW = lim kxkW,m. m m 6 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA

′ ′ Then one can construct the spaces V = Σ(αnk.kV,n), and W = Σ(αnk.kW,n). We show that ′ ′ V and W are ℓ1-saturated asymptotic ℓ1 spaces for vectors with disjoint supports.

n It is known [CO] that if θn = δ for some δ ∈ (0, 1) then one can replace the “allowable” by

“admissible” in the definition of k.kV to obtain an equivalent norm for V . For this choice of (θn) the variant of the norm k.kV,m+1 by replacing “allowable” by “admissible” can be minorized and majorized up to a uniform multiplicative constant by the norms k.kV,m and k.kV,m+1 respectively. This is enough to conclude that the new norms lead to an equivalent norm for V ′ (see [B]).

3. The main Theorems

The following results is the main tool of our paper for constructing weak Hilbert spaces.

Theorem 3.1. If X is an asymptotic ℓ2 space for vectors with disjoint supports then X is a weak Hilbert space.

Proof Since the ideas needed for the proof of Theorem 3.1 exist in the literature, we will just outline the proof. Recall that the fast growing hierarchy is a sequence of functions on the natural numbers (gn) which is defined inductively by: g0(n) = n +1, and for i ≥ 0, n n 0 gi+1(n)= gi (n) where g is the n-fold iteration of g and g = I.

Step I

If X is asymptotic-ℓ2 with constant C for vectors with disjoint supports, then for

every i ≥ 0 any gi(n) normalized disjointly supported vectors with supports after i n are C -equivalent to the unit vector basis of ℓ2. We proceed by induction on i with the case i = 0 being trivial. So, assume Step I holds for some i ≥ 0 and let {xk : 1 ≤ k ≤ gi+1(n)} be a sequence of disjointly supported vectors in X with supports after n. For 1 ≤ j ≤ n let

j−1 j Ej = {k : gi+1 (n) ≤ k ≤ gi+1 − 1}. WEAK HILBERT 7

Then, n 1/2 gi+1(n) n C k x k ≈ k x k2 k  k  j=1 Xk=1 X kX∈Ej Applying the induction hypotheses to each sum on the right we continue this equivalence as

n 1/2 gi+1(n) Ci+1 ≈ kx k2 .  k  Xk=1  

Step II

If X is asymptotic-ℓ2 with constant C for vectors with disjoint supports then every n-dimensional subspace of X supported after n is 8C3-isomorphic to a Hilbert space and 8C3-complemented in X.

n If E is a 5(5 )-dimensional subspace of X supported after n, then by a result of Johnson (See

Proposition V.6 of [CS]) there is a subspace G of X spanned by ≤ g3(n) disjointly supported 1 vectors supported after n and an operator V : E → G with kV x − xk ≤ 2 kxk, for all x ∈ E. Now, by Step I, we have that E is 2C3-isomorphic to a Hilbert space. It follows [J2] that every 5n-dimensional space of X∗ supported after n is 4C3-isomorphic to a Hilbert space and 4C3-complemented in X∗. Therefore, every n-dimensional subspace of X supported after n is 8C3-isomorphic to a Hilbert space and 8C3-complemented in X.

Step III

Every asymptotic-ℓ2 space for vectors with disjoint supports is a weak Hilbert space.

If E is a 2n-dimensional subspace of X, let F =: E ∩ (span k≥nek). Then F is supported after n and dim F ≥ n implies F is K-isomorphic to a Hilbert space and K-complemented in X by Step II, where K = 8C3. It follows from Definition 2.1 that X is a weak Hilbert space. ✷ The 2-convexification of certain Tsirelson spaces for obtaining weak Hilbert spaces was first used in [ADKM]. More generally we have the following: 8 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA

(2) Corollary 3.2. If X is an asymptotic ℓ1 space for vectors with disjoint supports then X is a weak Hilbert space.

Proof: Let (X, k.k) be an asymptotic ℓ1 space for vectors with disjoint supports. Then there exists C > 0 such that for every sequence of vectors (xi) with (supp xi) being S1-allowable, (2) we have that C kxik≤k xik. It suffices to prove that X is an asymptotic ℓ2 space (2) for vectors withP disjoint support.P Let (yi) be a sequence of vectors in X with (supp yi) being S1-allowable. Then

1/2 2 1/2 1/2 2 1/2 2 1/2 2 1/2 C ( kyik(2)) = C ( kyi k) ≤k yi k = k( yi ) k. X X X X Also,

2 1/2 2 1/2 2 1/2 2 1/2 k( yi ) k(2) = k yi k ≤ ( kyi k) =( kyik(2)) . X X X X ✷

′ ′ The spaces V , W , V and W are asymptotic ℓ1 spaces for vectors with disjoint supports. ′ n Indeed, this is obvious for V and W . To see this for V let n ∈ N and vectors (xi)i=1 with disjoint supports with n ≤ xi for all i. Then:

n ∞ n ∞ n

k xik = αmk xikV,m ≥ αmθ1 kxikV,m−1 i=1 m=1 i=1 m=1 i=1 X X n ∞X X Xn

≥ θ1u αm−1kxikV,m−1 ≥ θ1u kxikV ′ i=1 m=1 i=1 X X X The proof for W ′ is similar. Thus V (2), W (2), V ′(2) and W ′(2) are weak Hilbert spaces.

′ Proposition 3.3. The spaces V and W do not contain an isomorph of ℓ1. The spaces V ′ and W are ℓ1-saturated without being isomorphic to ℓ1.

The following Lemma will be used in the proof of Proposition 3.3.

Lemma 3.4. For every m ∈ N the completion of (c00, k.kV,m) is a c0-saturated space.

In order to prove this Lemma we need a result of Fonf along with the notion of the boundary. WEAK HILBERT 9

Definition 3.5. A subset B of the unit sphere of the dual of a Banach space X is called a boundary for X if for every x ∈ X there exists f ∈ B such that f(x)= kxk.

Theorem 3.6. ([F1], see also [F2], [H] [DGZ]) Every Banach space with a countable bound- ary is c0-saturated.

Proof of Lemma 3.4 Define inductively on i ≤ m the sets Ki of the unit ball of the dual 0 i of (c00, k.kV,m). Let K = {±en : n ∈ N}. For i < m if K has been defined then let

i+1 i m r K = K ∪{θk(f1 +···+fr): Fj ∈ K , for all j, (supp fj)j=1 is Sk allowable k =1, 2,... }.

m Then K is a norming set for (c00, k.kV,m):

m kxkV,m = sup{|f(x)| : f ∈ K }.

m It is easy to see that K ∪{0} is a pointwise closed set since each Sk is pointwise closed and limk θk = 0. The previous Theorem of Fonf finishes the proof of the Lemma. ✷

Proof of Proposition 3.3 The statement for V and W is obvious since they are reflexive [AD], [ADKM]. ′ ′ V is ℓ1 saturated: Let (xi) be an arbitrary block basis of V . It is enough to construct a ′ normalized (in V ) block basis (vi) of (xi) and an increasing sequence of positive integers

1= p1

p −1 i+1 1 α kv k ≥ m i V,m 2 m=p Xi for all i. Once this is done then for (λi) ∈ c00

n ∞ n ∞ pj+1−1 n

k λivikV ′ = αmk λivikV,m = αmk λivikV,m i=1 m=1 i=1 j=1 m=p i=1 X X X X Xj X p −1 ∞ j+1 1 ∞ ≥ α kλ v k ≥ |λ | m j j V,m 2 j j=1 m=p j=1 X Xj X 10 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA which shows that (vi) is equivalent to the unit vector basis of ℓ1. In order to choose such

(vi) and (pi) we use that for every m ∈ N the norms k.kV,m and k.kV,m+1 are not equivalent.

Thus for every m,M,K ∈ N there is u in the span of (xi) with

1 M ≤ u, kuk < , and kuk ≥ K. V,m 4 V,m+1

Then m 1 ∞ α kuk < and α kuk ≥ α K. i V,i 4 i V,i m+1 i=1 i=1 X X Let v = u . By taking K large enough we can assume that kukV ′

m 1 α kvk < . i V,i 4 i=1 X Also choose m′ > m with ∞ 1 αikvkV,i < . ′ 4 i=Xm +1 Thus ′ m 1 α kvk ≥ . i V,i 2 i=m+1 X It only remains to show that for every m ∈ N the norms k.kV,m and k.kV,m+1 are not ′ equivalent on the span of (xi). By the previous Lemma there is a block sequence (yi) of(xi) such that kyikV,m = 1 for all i and ((yi), k.kV,m) is 2-equivalent to the unit vector basis of c0.

For n ∈ N let k ∈ N with n ≤ yk+1

k+n

k yikV,m ≤ 2 i=Xk+1 yet k+n

k yikV,m+1 ≥ θ1n. i=Xk+1 This proves the result. ′ W is ℓ1-saturated: Similar. WEAK HILBERT 11

′ ′ V is not isomorphic to ℓ1: If the statement were false then the basis of V would be isomor- phic to the unit vector basis of ℓ1 (since every normalized unconditional basic sequence in

ℓ1 is equivalent to the usual unit basis of ℓ1, [LP]). Observe that for x ∈ c00,

kxk = αmkxkV,m ≤ sup kxkV,m = kxkV . m m X

By [ADKM] the norm of V can become arbitrarily smaller that the ℓ1 norm on certain vectors. ′ W is not isomorphic to ℓ1: Similar. ✷

Remark 3.7. Note that a space X with an unconditional basis contains ℓ1 if and only if (2) (2) X contains ℓ2. Since V and W do not contain an isomorph of ℓ1, we obtain that V and (2) ′ W are weak Hilbert spaces which do not contain an isomorph of ℓ2. Since the spaces V ′ ′(2) and W are ℓ1 saturated without being isomorphic to ℓ1, we obtain that the spaces V , and ′(2) W are ℓ2-saturated weak Hilbert spaces which are not isomorphic to ℓ2.

′(2) Thus the essential properties of the space Eα constructed by Edgington are shared by V and W ′(2).

1/n Theorem 3.8. Let (θn) ⊂ (0, 1) with limn θn = 1, s ∈ N, and β = (βn) ⊂ (0, 1) with β =1 and 0 < inf βn+1 ≤ sup βn+1 < 1. Then V ′(2) and W ′(2) are not isomorphic to E . n n βn βn β P ′(2) ′(2) Proof: Let T : X → Eβ be an isomorphism where X is either V or W . Since T is an isomorphism there exists C > 0 such that

1 kT xk ≤kxk ≤ CkT xk C Eβ X Eβ for all x ∈ c00. Also, by [E] (proof of Theorem 7) there exists δ > 0 such that

(2) kT xkEβ ≤ CkT xkT (δ,S1).

Thus for x ∈ c00

2 (2) kxkX ≤ C kT xkT (δ,S1). (1) 12 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA

Since the unit vector basis (ei) of X is weakly null, we can select a subsequence (eki )of(ei), (2) a block sequence (ui) in T (δ, S1) with non-negative coefficients and a number K > 0 such that: ε 1 kT (e ) − u k (2) < and ≤ku k (2) ≤ K for all i, ki i T (δ,S1) 2i K i T (δ,S1) where ε > 0 will be chosen later. Let n ∈ N to be selected later. Let (xi)i∈I ⊂ (0, 1) for some I ∈ Sn so that u2 x2 =1, and k x2 i k ≤ δn + ε i i ku2k T (δ,S1) i∈I i∈I i T (δ,S1) X X ([OTW] Theorem 5.2 (a)). Then

(2) (2) (2) kT ( xieki )kT (δ,S1) ≤ k xiuikT (δ,S1) + xikT eki − uikT (δ,S1) i∈I i∈I i∈I X X X ≤ k x2u2k1/2 + ε i i T (δ,S1) Xi∈I u2 ≤ Kk x2 i k1/2 + ε i 2 T (δ,S1) kui kT (δ,S1) Xi∈I ≤ K(δn + ε)1/2 + ε.

On the other hand if Y = V ′ when X = V ′(2) or Y = W ′ when X = W ′(2) then ∞ 2 1/2 2 1/2 k xieki kX = k xi eki kY ≥ ( βmθn xi ) = θn i∈I i∈I m=1 i∈I X X X X p Therefore (1) gives

2 n 1/2 θn ≤ C (K(δ + ε) + ε). 1/n p But since limn θn = 1, n and ε can be chosen so that this inequality fails. ✷

References

[AA] D.E. Alspach and S.A. Argyros, Complexity of weakly null sequences, Dissertationes Mathematicae 321, 1992.

[AD] S.A. Argyros and I. Deliyanni, Examples of asymptotic ℓ1 spaces, Trans. Amer. Math. Soc. 349 (1997), 973-995. [ADKM] S.A. Argyros, I. Deliyanni, D.N. Kutzarova and A. Manoussakis, Modified mixed Tsirelson spaces, preprint. WEAK HILBERT 13

[B] S.F. Bellenot, Tsirelson subspaces and ℓp, J. Funct. Analysis 69 (1986), 207-228. [CN] P.G. Casazza and N.J. Nielsen, A Gaussian Average Property of Banach spaces, Illinois J. Math. 41 No 4 (1997), 559-576. [CO] P.G. Casazza and E. Odell, Tsirelson’s space and minimal subspaces, Longhorn notes, University of Texas (1982-83), 61-72. [CS] P.G. Casazza and T. Shura, Tsirelson’s space, Lecture Notes in Math. 1363, Springer 1989. [DGZ] R. Deville, G. Godefroy and V. Zizler, Smoothness and renormings in Banach spaces, Pitman Monographs Surveys Pure Appl. Math. Vol. 64, Longman Sci. Tech., 1993. [E] A. Edgington, Some more weak Hilbert spaces Studia Math. 100 (1991), 1-11.

[FJ] T. Fiegel and W.B. Johnson, A uniformly convex Banach space which contains no ℓp, Compositio Mathematica, Vol. 29, Fasc 2 (1974), 191-196. [F1] V.P. Fonf, Weakly extremely properties of Banach spaces, Mat. Zametki 45(6) (1989) 83-92 (Rus- sian). [F2] V.P. Fonf, On exposed and smooth points of convex bodies in Banach spaces, Bull. London Math. Soc. 28 (1996) 51-58. [H] P. Hajek, Smooth norms that depend locally on finitely many coordinates, Proc. Amer. Math. Soc. 123 (1995), 3817-3821. [J1] W.B. Johnson, A reflexive Banach space which is not sufficiently Euclidean, Studia Math. 55 (1976), 201-205. [J2] W.B. Johnson, Banach spaces all of whose subspaces have the approximation property, Seminaire d’ Analyse Fonct. Expose 16 (1979-80). Ecole Polytechnique, Paris. [K] R. Komorowski, Weak Hilbert spaces without unconditional bases, Proc. Amer. Math. Soc. 120 (1994), 101-107. [KT] R. Komorowski and N. Tomczak-Jaegermann, Banach spaces without local unconditional structure, Israel J. Math 89 (1995), 205-226.

[LP] J. Lindenstrauss and A. Pelczynski, Absolutely summing operators in Lp-spaces and their applica- tions, Studia Math. 29 (1968) 275-326. [M] V. Mascioni, On Banach spaces isomorphic to their duals, Houston J. Math, 19 (1993), 27-38. [MP] V.D. Milman and G. Pisier, Banach spaces with a weak cotype 2 property, Israel J. Math. 54 (1986), 139-158. [NTJ] N.J. Nielsen and N. Tomczak-Jaegermann, On subspaces of Banach spaces with Property (H) and weak Hilbert spaces,

[OTW] E. Odell, N. Tomczak-Jaegermann and R. Wagner, Proximity to ℓ1 and Distortion in Asymptotic

ℓ1 spaces, preprint. 14 GEORGE ANDROULAKIS, PETER G. CASAZZA, AND DENKA N. KUTZAROVA

[P1] G. Pisier, Weak Hilbert spaces, Proc. London Math. Soc. 56 (1988), 547-579.

[T] B.S. Tsirelson, Not every Banach space contains ℓp or c0, Funct. Anal. Appl. 8 (1974), 138-141.

Department of Mathematics, Math. Sci. Bldg., University of Missouri-Columbia, Columbia MO 65211 E-mail address: [email protected]

Department of Mathematics, Math. Sci. Bldg., University of Missouri-Columbia, Columbia MO 65211 E-mail address: [email protected]

Current: Dept. Math. & Stat., Miami University, Oxford, Ohio, Permanent: Institute of Mathematics, Bulgarian Academy of Sciences, 1113 Sofia, Bulgaria E-mail address: [email protected]