arXiv:1910.02335v2 [math.FA] 2 Mar 2020 o n 1 any for endb zeki S.Temi ups fti ae st rvd na an provide to is paper this of purpose main case The remaining the [S]. for in Szlenk by defined atta the that fact unique a n[LS eto .]ta hoe . olne od fw replace we if holds longer no 1.1 Theorem that 9.2] Section [BLMS, in space eebe htof that resembles tutr mle htasaei Asymptotic is space a that implies structure 1 hoe 1.1 Theorem ra fsqecs( sequences of array ℓ oti nAsymptotic an contain lc eune fafie ai,if basis, fixed a of sequences block oe ihrsetto respect with model rbe 1 Problem fnraie ekynl eune and sequences null weakly normalized of eso rm[M]o h oino asymptotic of notion the of [MMT] from version rmtefloigPolmo abie n dl rm[O n subse a and [HO] from Odell space Banach and a mo Given Halbeisen its [FOSZ]. of from draws result Problem remarkable question following The the [MT]. from from Asymptotic Tomczak-Jaegermann be and Milman must space a the in that models spreading implies joint of space uniqueness uniform the equivalently, or ubr S M-600 n S DMS-1912897. NSF and DMS-1600600 NSF Numbers 1 ≤ twsosre yBuir ace,Klo,tetidato,an author, third the Kalton, Lancien, Baudier, by observed was It naypoi oe santo hc ecie h smttcbeha asymptotic the describes which notion a is model asymptotic An nti ae esuyteqeto hte h nqeeso asy of uniqueness the whether question the study we paper this In h hr uhrsrsac a upre yteNtoa Sc National the by supported was research author’s third The Classification: Subject 2010 n disaunique a admits and < p PRSA RYO,AEADO ERIU N ALSMOTAKIS PAVLOS AND GEORGIOU, ALEXANDROS ARGYROS, A. SPIROS X xssarflxv aahsaewt nucniinlbssa basis n presented, unconditional is an type with [MR] space Rosenthal Banach P. reflexive H. a - exists Maurey B. of result ereplace we nqeaypoi oe hra n usqec ftebas the of subsequence any whereas non-Asymptotic model asymptotic unique a htalaypoi oesgnrtdb ekynl sequenc of null basis weakly vector unit by the generated to models asymptotic all that htweee eaal aahsaentcontaining not space Banach separable a whenever that Abstract. sAsymptotic is ℓ ∞ NON-ASYMPTOTIC 1 < p < rs.Asymptotic (resp. , smttcmdlwihmasavr togpeec fasympto of presence strong very a means which model asymptotic ([HO]) ℓ 1 ([FOSZ]) nr stelretoeadhne suigta h pc admits space the that assuming hence, and one largest the is -norm ℓ c ∞ p 0 eetrsl fFemn dl,Sr,adZeg[OZ sta [FOSZ] Zheng and Sari, Odell, Freeman, of result recent A . (resp. with x h oneeape r pcsvr iia otespace the to similar very spaces are counterexamples The . Let j i ℓ F ) 1 p c ℓ j . p 0 , 0 ℓ subspace. c 1 . 1 , .Nt httemi btuto nti aei the is case this in obstruction main the that Note 1. = ( Let X 0 i X SMTTCMODEL ASYMPTOTIC c hnti euti olne re oevr stronger a Moreover, true. longer no is result this then 0 ∈ smttcmdlwt epc to respect with model asymptotic ≤ ,o ihrsetto respect with or ), .Rmral,i oecssuiu smttcarray asymptotic unique cases some in Remarkably, ). eaBnc pc htamt nqeasymptotic unique a admits that space Banach a be X c N < p 0 1. c ntecnrr pc sAsymptotic is space a contrary the On . easprbeBnc pc htde o contain not does that space Banach separable a be hntesaei Asymptotic is space the then 0 fteaypoi eairo h hl space whole the of behavior asymptotic the if ) X Introduction ∞ ℓ 1 a one. has ra Asymptotic an or , rmr 60,4B6 62,46B45. 46B25, 46B06, 46B03, Primary PCSWT UNIQUE WITH SPACES 1 F b ( c X 0 . eoetecaso l normalized all of class the denote ) ℓ p F pcswt cadrbssby basis Schauder a with spaces b ℓ ( X p X hsi oriaefree coordinate a is This . let , if ) ec onainudrGrant under Foundation ience c ℓ c 1 0 0 X eso htif that show We . F a h property the has subspace? saeequivalent are es 0 dmitting a ai.Does basis. a has sgnrtsa generates is ( X ml,there amely, F eoeteclass the denote ) 0 poi models, mptotic ( Schlumprecht d X ie Banach given .Te the Then ). ℓ 1 ℓ 1 tes c as iro an of vior 0 tivation with ℓ nswer quent p tic for , X ℓ ℓ p 1 2 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS structure, it is not obvious how to preserve a tree structure in the space which has norm smaller than ℓ1. Theorem 1.2. There exists a reflexive X with an unconditional basis that admits a unique ℓ1 asymptotic model with respect to F0(X), whereas it is not an Asymptotic ℓ1 space. ξ In fact, for every countable ordinal ξ, there is a space Tinc, as in Theorem 1.2, ξ ξ that contains a weakly null ℓ2-tree of height ω . An easy modification of Tinc can yield a space containing a weakly null ℓp-tree, for any 1

ℓp. These are spaces with an unconditional (et)t∈Tξ indexed over ξ a well-founded and infinite branching countable tree Tξ of height ω . The norm of JT ξ is defined as follows: if x = a e and S is a segment of T define r,p t∈Tξ t t ξ r r kS(x)kr = t∈S |at| and P P n 1/p p n (1.8) kxk ξ = sup kSi(x)k : (Si) disjoint segments of Tξ . JT r,p  r i=1   Xi=1  ξ The space Tinc from Theorem 1.2 is defined on the same tree. We say that two segments S1, S2 of Tξ are incomparable if any node of S1 is incomparable to any node of S2. We relabel the basis of the Tsirelson space T as (et)t∈Tξ so that the ξ order is compatible with the initial one and define the norm of Tinc as follows : for x = a e define kS(x)k2 = |a |2 and t∈Tξ t t 2 t∈S t P P n n kxkT ξ = sup kSi(x)k2emin Si : (Si)i=1 incomparable segments of Tξ . inc  T  Xi=1 However, we will not use the above description of the norms. Instead we revert to the notion of norming sets and norming functionals. This makes some parts of the proof easier and it can also be potentially useful to show similar results on more complicated spaces based on these norms. Finally, we should mention that Problem 1 is only one of several concerning the separation of different asymptotic structures in Banach space theory. For example, in [AM3] the first and third author showed that there exist spaces with a uniformly unique spreading model, which can be chosen to be any ℓp or c0, that have no Asymptotic ℓp or c0 subspace. This answers a question by Odell in [O1] and Junge, Kutzarova, and Odell in [JKO]. Moreover, in [KM] Kutzarova and the third author showed that certain spaces by Beanland, the first author, and the third author from [ABM] are asymptotically symmetric and have no Asymptotic ℓp or c0 subspaces, answering a question from [JKO]. NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTICMODEL 3

Notation. By N = {1, 2,...} we denote the set of all positive integers. We will use capital letters as L,M,N,... (resp. lower case letters as s, t, u, . . .) to denote infinite subsets (resp. finite subsets) of N. For every infinite subset L of N, the notation [L]∞ (resp. [L]<∞) stands for the set of all infinite (resp. finite) subsets of L. For every s ∈ [N]<∞, by |s| we denote the cardinality of s. For L ∈ [N]∞ and k ∈ N,[L]k (resp. [L]≤k) is the set of all s ∈ [L]<∞ with |s| = k (resp. |s|≤ k). For every s,t ∈ [N]<∞, we write s

2. Asymptotic structures Let us recall the definitions of the asymptotic notions that appear in the results of this paper and were mentioned in the introduction. Namely, asymptotic models, joint spreading models and the notions of Asymptotic ℓp and Asymptotic c0 spaces. For a more thorough discussion, including several open problems and known results, we refer the reader to [AM3, Section 3]. i N Definition 2.1 ([HO]). An infinite array of sequences (xj )j , i ∈ , in a Banach space X, is said to generate a sequence (ei)i, in a seminormed space E, as an asymptotic model if for every ε> 0 and n ∈ N, there is a k0 ∈ N such that for any natural numbers k0 ≤ k1 < ··· < kn and any choice of scalars a1,...,an in [−1, 1] we have that n n a xi − a e < ε. i ki i i

Xi=1 Xi=1

A Banach space X is said to admit a unique asymptotic model with respect to a family F of normalized sequences in X if whenever two infinite arrays, consisting of sequences from F , generate asymptotic models then those must be equivalent. Typical families under consideration are those of normalized weakly null sequences, denoted F0(X), normalized Schauder basic sequences, denoted F (X), or the family all normalized block sequences of a fixed basis of X, if it has one, denoted Fb(X). The notion of plegma families was first introduced by Kanellopoulos, Tyros, and the first author in [AKT]. We will use the slightly modified definition of from [AGLM]. Definition 2.2 ([AGLM]). Let M ∈ [N]∞ and k ∈ N. A plegma (resp. strict k l k plegma) family in [M] is a finite sequence (si)i=1 in [M] satisfying the following.

(i) si1 (j1)

(ii) si1 (j) ≤ si2 (j) resp. si1 (j) 0 and n ∈ N, N l N n there is a k0 ∈ such that for any (si)i=1 ∈ S-Plm([ ] ) with k0 ≤ s1(1) and any l × n matrix A = (aij ) with entries in [−1, 1] we have that l n l n a xi − a ei < ε. ij si(j) ij j

Xi=1 Xj=1 Xi=1 Xj=1

A Banach space X is said to admit a uniformly unique joint spreading model with respect to a family of normalized sequences F in X, if there exists a constant 4 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

i i C such that whenever two arrays (xj )j and (yj )j , 1 ≤ i ≤ l, of sequences from F generate joint spreading models then those must be C-equivalent. Moreover, a Banach space admits a uniformly unique joint spreading model with respect to a family F if and only if it admits a unique asymptotic model with respect to F (see, e.g., [AGLM, Remark 4.21] or [AM3, Proposition 3.12]). It was proved in [AGLM] that whenever a Banach space admits a uniformly unique joint spreading model with respect to some family satisfying certain stabil- ity conditions, then it satisfies a property concerning its bounded linear operators, called the Uniform Approximation on Large Subspaces property (see [AGLM, The- orem 5.17] and [AGLM, Theorem 5.23]).

Definition 2.4 ([MT] and [MMT]). A Banach space X is called Asymptotic ℓp, 1 ≤ p < ∞, (resp. Asymptotic c0) if there exists a constant C such that in a two-player n-turn game G(n,p,C), where in each turn k =1,...,n player (S) picks a finite codimensional subspace Yk of X and then player (V) picks a normalized vector xk ∈ Yk, player (S) has a winning strategy to force player (V) to pick a n n ∞ sequence (xk)k=1 that is C-equivalent to the unit vector basis of ℓp (resp. ℓn ).

The typical example of a non-classical Asymptotic ℓp space is the Tsirelson space as defined by Figiel and Johnson in [FJ]. This is a reflexive Asymptotic ℓ1 space and it is the dual of Tsirelson’s original space from [T] that is Asymptotic c0. Finally, whenever a Banach space is Asymptotic ℓp or Asymptotic c0, then it admits a uniformly unique joint spreading model with respect to F0(X) (see, e.g., [AGLM, Corollary 4.12]).

3. A family of non-asymptotic ℓ1 spaces admitting uniformly unique ℓ1 joint spreading models ξ In this section we define the spaces Tinc, for each countable ordinal ξ, and we prove that they admit a uniformly unique ℓ1 joint spreading model with respect to F ξ b(Tinc) and are not Asymptotic ℓ1. The spaces are defined in terms of norming sets and norming functionals as this is more convenient to prove the desired result. 3.1. Measures on Well-Founded Countable Compact Trees. We start with ξ a key result that will be used later to prove that Tinc admits a uniformly unique joint spreading model equivalent to the unit vector basis of ℓ1. Let  be a partial order on some infinite subset M of the naturals, which is compatible with the standard order, i.e. n  m implies n ≤ m, for all n,m ∈ M. Assume that, for each n ∈ M, the set Sn = {m ∈ M : m  n} is finite and totally ordered with respect to , that is, T = (M, ) is a tree. Let us also assume that the tree T is well-founded, i.e., it contains no infinite totally ordered sets, and infinite branching, i.e., every non-maximal node has infinitely many immediate successors. Observe that T = ({St : t ∈T}, ⊂) is also a tree and is in fact isomorphic to T via the mapping t 7→ S . Given t ∈ T , we will denote S by t˜. Moreover, two nodes e t t t˜1, t˜2 are incomparable in T if and only if the nodes t1,t2 are incomparable in T , i.e. not comparable in the respective order. For t˜ ∈ T , we denote by V the set e t˜ consisting of t˜ and all of its successors. Note that T is a countable when equipped with the pointwise convergence topology and hence M(T ), the set of all regular measures on T , is e isometric to ℓ (T ). In particular, each µ ∈ M(T ) is of the form µ = e a δ , 1 e t˜∈T et˜ t˜ where δ is the Dirac measure centered on t˜, and kµk = e |a |. Finally,P the t˜ e e t˜∈T t˜ ˜ support of µ is defined as suppµ = {t ∈ T : at˜ 6= 0}. We willP prove the following proposition, starting with Lemma 3.2 e Proposition 3.1. Let (µj )j be a sequence of positive regular measures on T with N disjoint finite supports and let c > 0 be such that µj (T ) < c for all j ∈ .e Then, e NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTICMODEL 5

∞ for every ε > 0, there is an L ∈ [N] and a subset Gj of suppµj for each j ∈ L, satisfying the following.

(i) µj (T\ Gj ) ≤ ε for every j ∈ L. (ii) The sets G , j ∈ N, are pairwise incomparable. e j

Lemma 3.2. Let (µj )j be a sequence of positive regular measures on T with disjoint N finite supports and let c > 0 be such that µj (T ) < c for all j ∈ .e Assume that w∗- lim µ = µ = e a δ . Then, for every t˜ ∈ suppµ and ε > 0, there is an j j t˜∈T t˜ t˜ e N ∞ t˜ L ∈ [ ] and a subsetP Gj of suppµj for each j ∈ L, satisfying the following.

t˜ (i) Gj ⊂ Vt˜ for every j ∈ L. t˜ (ii) |µj (Gj ) − at˜| <ε for every j ∈ L. t˜ (iii) The sets Gj , j ∈ L, are pairwise incomparable.

Proof. Recall that the nodes of T are in fact naturals numbers. Hence identifying {t : t˜∈ suppµj }, j ∈ N, as subsets of the naturals and passing to a subsequence, we may assume that they are successive. Let (t˜j )j be an enumeration of the immediate successors of t˜ and for each j ∈ N t˜ j t˜ define Wj = Vt˜ \ ∪i=1Vt˜i . Observe that (Wj )j is a decreasing sequence of clopen t˜ ˜ t˜ t˜ t˜ subsets of T with ∩j Wj = {t} and hence limj µ(Wj )= at˜ and limj µj (Wi )= µ(Wi ) ∞ for all i ∈ N. We can thus find N ∈ [N] and pass to a subsequence of (µj )j , e t˜ t˜ which we relabel for convenience, so that limj∈N |µj (Wj ) − µ(Wj )| = 0 and define t˜ t˜ t˜ Gj = suppµj ∩ Wj for each j ∈ N. Note then that limj∈N µj (Gj ) = at˜ and j µj | t˜(∪ V˜ ) = 0 for all j ∈ N. Gj i=1 ti ˜ t˜ There is at most one j ∈ N such that t ∈ Gj and hence, passing to a subsequence, ˜ t˜ t˜ we may assume that t∈ / Gj for all j ∈ N. Moreover, since limj∈N µj (Gj )= at˜, we t˜ may even pass to a further subsequence such that |µj (Gj )−at˜| <ε for all j ∈ N. For the remaining part of the proof we will choose, by induction, an L ∈ [N]∞ such that t˜ Gj , j ∈ L, are pairwise incomparable. Set l1 = min N and suppose that we have N t˜ t˜ chosen l1 <...

Proof of Proposition 3.1. Passing to a subsequence, since T is compact with respect to the pointwise convergence topology and (µ ) are uniformly bounded, we may j j e ∗ e assume that (µj )j w -converges to some measure µ = t˜∈T at˜δt˜ in M(T ). N Let δ > 0 be such (1 − δ)(µ(T ) − δ) > µ(T ) − Pε/2 and pick n0 e∈ such n0 ˜ that i=1 at˜i ≥ µ(T ) − δ. Applying the previous lemma successively for each ti, ∞e e i =1,...,nP 0, we obtain an L ∈ [N] and, for each j ∈ L and i =1,...,n0, a subset i e ˜ Gj of suppµj satisfying items (i) - (iii) of Lemma 3.2 for ti and δat˜i . Note that if ˜ ˜ i1 ti1 , ti2 are incomparable for some 1 ≤ i1,i2 ≤ n0, then by item (i), the sets Gj1 and i2 ˜ ˜ Gj2 are also incomparable for any j1, j2 ∈ L. If the nodes ti1 , ti2 are comparable, ˜ ˜ ˜ i1 say ti1 ⊂ ti2 , then there exists at most one j ∈ L such that ti2 ∈ Gj . Hence by a i finite induction argument, we may pass to a subsequence such that the sets Gj , for n0 i i = 1,...,n0 and j ∈ L, are pairwise incomparable. Define Gj = ∪i=1Gj , j ∈ L, 6 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS and conclude that

n0 n0 i ε µ (G )= µ (G ) ≥ a˜ − δa˜ ≥ (µ(T ) − δ)(1 − δ) >µ(T ) − . j j j j ti ti 2 Xi=1 Xi=1 e e Finally, passing to a further subsequence if necessary, we may also assume that |µj (T ) − µ(T )| < ε/2 and hence µj (T\ Gj ) <ε for every j ∈ L.  e e e ξ 3.2. Tsirelson Extension of a Ground Set. In order to define Tinc, we first introduce some necessary concepts used in the construction of Tsirelson type spaces.

Definition 3.3. A subset W of c00(N) is called a norming set if it satisfies the following conditions. ∗ N (i) W is symmetric and ei ∈ W for every i ∈ . (ii) kfk∞ ≤ 1 for every f ∈ W . (iii) W is closed under the restriction of its elements to intervals of N.

A norming set W induces a norm k·kW on c00(N) defined as

kxkW = sup{f(x): f ∈ W }.

Definition 3.4. Let G be a norming set on c00(N). The Tsirelson extension of G, denoted by WG, is the minimal subset of c00(N) that contains G and is closed under the (S, 1/2)-operation, i.e., if f1,...,fn are in WG and n ≤ suppf1 <...< suppfn, n then 1/2 i=1 fi is also in WG. We call G the ground set of WG. P N Note that WG is a norming set on c00( ). Moreover, the induced norm k·kWG satisfies the following implicit equation 1 n kxk = max kxk , sup kE xk WG G 2 i WG n Xi=1 o where the supremum is taken over all finite collections E1,...,En of successive intervals of N with n ≤ E1.

Definition 3.5. Let f ∈ WG. For a finite tree A, a family (fα)α∈A is said to be a tree analysis of f if the following are satisfied.

(i) A has a unique root denoted by 0 and f0 = f. (ii) For every maximal node α ∈ A we have that fα ∈ G. (iii) Let α be a non-maximal node of A and denote by S(α) set of immediate successors of α. Then fα ∈ WG and the ranges of fs, s ∈ S(α), are disjoint and fα =1/2 s∈S(α) fs. P It follows, by minimality, that every f ∈ WG admits a tree analysis.

Proposition 3.6. Let f ∈ WG with a tree analysis (fα)α∈A and denote by M the set of all maximal nodes of A. Then the following hold. N kα (i) For every α ∈M, there is a kα ∈ ∪{0} such that f = α∈M fα/2 . kα (ii) If N ⊂M, then g = α∈N fα/2 is in WG and g = f|∪{Psuppfα:α∈N }. P For an extensive review on Tsirelson’s space we refer the reader to [CS].

ξ ξ 3.3. Definition of the space Tinc. We define the space Tinc as the completion of c00(N) with respect to the norm induced by a norming set Wξ. This norming set ξ is a subset of the Tsirelson extension of a ground set G2, the functionals of which ξ satisfy a certain property. Both this property and G2 are defined via an infinite branching well-founded tree Tξ on the natural numbers. N ∞ We start by fixing a partition of the naturals = ∪j=0Nj into infinite sets and N <∞ N an injection φ :[ ] → . Recall the definition of the Schreir families (Sξ)ξ<ω1 . NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTICMODEL 7

Definition 3.7. Let ξ be a countable ordinal. We define, by transfinite induction, <∞ the Schreier family Sξ ⊂ [N] as follows.

(i) If ξ = 0, then S0 = {{n} : n ∈ N}∪{∅}. (ii) If ξ = α + 1, then n N Sξ = {∪j=1Ej : n ∈ , E1 <...

(iii) If ξ is a limit ordinal we choose a fixed sequence (α(ξ, j))j ⊂ [1, ξ) which increases to ξ and set

Sξ = {E ⊂ N : there exists j ∈ N such that E ∈ Sα(ξ,j) and j ≤ E}.

We now define the tree Tξ, by defining a partial order ξ on N.

Definition 3.8. Fix a countable ordinal ξ and define the partial order ξ on N as follows: n ξ m if there exists {n0,...,nk} ∈ Sξ such that

(i) n0 ∈ N0 and ni ∈ Nφ(n0,...,ni−1) with ni−1

Remark 3.9. Note that Tξ = (N, ξ) is an infinite branching tree and it is also well-founded since Sξ is a compact family, i.e., {χE : E ∈ Sξ} is a compact subset N of {0, 1} . Moreover, the partial order ξ is compatible with the standard order on ξ the naturals and finally, standard inductive arguments yield that Tξ is of height ω .

Definition 3.10. Define the following norming set on c00(N)

ξ ∗ 2 G2 = aiei : S is a segment of Tξ and ai ≤ 1 n Xi∈S Xi∈S o and denote by Wξ the subset of W ξ containing all f with tree analysis (fα)α∈A G2 such that there exist pairwise incomparable segments Sα of Tξ with suppfα ⊂ Sα for ξ N every maximal node α ∈ A. Denote by Tinc the completion of c00( ) with respect to the norm k·kWξ induced by the norming set Wξ.

Remark 3.11. The unit vector basis (ej )j of c00(N) forms a 1-unconditional ξ ξ Schauder basis for the space Tinc. Moreover it is boundedly complete, since Tinc admits ℓ1 as a uniformly unique spreading model as shown in Proposition 3.12. ξ First, we show that Tinc admits a uniformly unique joint spreading model with F ξ respect to b(Tinc), that is equivalent to the unit vector basis of ℓ1. ξ Proposition 3.12. The space Tinc admits a uniformly unique joint spreading model F ξ with respect to b(Tinc), which is equivalent to the unit vector basis of ℓ1.

i ξ Proof. Let (xj )j , 1 ≤ i ≤ l, be an array of normalized block sequences in Tinc and i1 i2 ε > 0. Passing to a subsequence, we assume that suppxj < suppxj+1 for every i1,i2 = 1,...,l and j ∈ N. For every i = 1,...,l and j ∈ N, pick a functional i i i kj,α i i i i f = i f /2 in Wξ such that f (x ) ≥ 1 − ε and f (x ) > 0 for every j α∈Mj j,α j j j,α j Pi i α ∈ Mj , where Mj denotes the set of all maximal nodes of a fixed tree analysis i i N i i i i kj,α of fj . For every i = 1,...,l, j ∈ and α ∈ Mj , define λj,α = fj,α(xj )/2 and i i N tj,α = min suppfj,α. Moreover, for each j ∈ , define the probability measure

l 1 1 i µj = λj,αδt˜i . l f i(xi ) j,α i=1 j j i X αX∈Mj ∞ Then, Proposition 3.1 yields an L ∈ [N] and a sequence (Gj )j∈L of pairwise incomparable subsets of Tξ such that µj (Gj ) ≥ 1 − δ for every j ∈ L and for some e 8 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

δ sufficiently small such that for any i =1,...,l and j ∈ L

1 i 2 (2.8.1) λ δ˜i (Gj ) ≥ (1 − ε) . i i j,α tj,α fj (xj ) i αX∈Mj N l k Let k ∈ and (si)i=1 ∈ S-Plml([L] ) with kl ≤ xs1(1). Then, for i = 1,...,l and j ∈ L, if N i = {α ∈ Mi : t˜i ∈ G }, item (ii) of Proposition 3.6 yields j si(j) j,α si(j) that i 1 i gj = i fsi(j),α ∈ Wξ. ks (j),α i 2 i αX∈Nj Moreover, (2.8.1) implies gi (xi ) ≥ (1 − ε)2 for all i = 1,...,l and j ∈ L, and j si (j) l k i since Gj , j ∈ L, are pairwise incomparable, we have that g = 1/2 i=1 j=1 gj l,k is in Wξ. Then for any choice of scalars (aij )i=1,j=1, due to unconditionality,P P we conclude that l k l k 2 l k i i (1 − ε) aij xs (j) ≥ |aij |xs (j) ≥ |aij |. i i 2 Xi=1 Xj=1 Xi=1 Xj=1 Xi=1 Xj=1 

ξ Proposition 3.13. The space Tinc is reflexive. ξ Proof. Since Tinc admits a boundedly complete unconditional Schauder basis, it does not contain c0 (see [LT, Theorem 1.c.10]) and hence it suffices to show that it does not contain ℓ1 as follows from [J1, Theorem 2]. N ξ ∗ Fix n ∈ . Let (xj )j be a normalized block sequence in Tinc and f = i∈S biei ξ in G2. For each j =1,...,n, define Ij = {i ∈ S : i ∈ suppxj } and note thatP 2 2 bjxj (i) ≤ bi .  iX∈Ij  iX∈Ij

Then, for any choice of scalars a1,...,an, we have that

n n n 1 n 2 1 2 2 2 f aj xj = aj bixj (i) ≤ aj bixj (i)  Xj=1  Xj=1 Xi∈S  Xj=1   Xj=1  iX∈Ij  

n 1 n 1 n 1 2 2 2 2 2 2 ≤ aj bi ≤ aj  Xj=1   Xj=1 iX∈Ij   Xj=1  and hence n n 1 2 2 aj xj ≤ aj . Gξ 2   Xj=1 Xj=1 ξ That is, for any normalized block sequence (xj )j in Tinc, there exists a block sub- sequence (yj )j with kyjk ξ → 0. G2 ξ We show that Tinc does not contain ℓ1 in a similar manner as in the proof of ξ the reflexivity for the classical Tsirelson space [FJ]. Suppose that Tinc contains ℓ1. Then James’ ℓ1 distortion theorem [J2] implies that, for ε < 1/4, there exists a ξ normalized block sequence (xj )j in Tinc such that n n aj xj ≥ (1 − ε) |aj |

Xj=1 Xj=1 for any n ∈ N and any choice of scalars a1,...,an. Applying the result of the previous paragraph, we may also assume that kxj k ξ < 1/2 for every j ∈ N and G2 NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTICMODEL 9 hence, for any n ≥ 2, we have that n+1 n+1 1 1 (2.9.1) x1 + xi > x1 + xi ξ . n n G2 Xi=2 Xi=2 Moreover, for any n ∈ N, we have that n+1 1 x1 + xi ≥ 2(1 − ε). n Xi=2 k ξ Observe that (2.9.1) implies that there exists f =1/2 j=1 fj ∈ Wξ \ G2 such that n+1 n+1 P 1 1 5 f x1 + xi > x1 + xi − ε ≥  n  n 4 Xi=2 Xi=2 and that min suppf1 ≤ max suppx1, since otherwise 1 n+1 1 n+1 f x + x = f(x ) ≤ 1. 1 n i n i  Xi=2  Xi=2

Therefore, k ≤ max suppx1. Note that there are at most ki’s such that the support of xi intersects the supports of at least two fj ’s and hence n+1 1 k n − k n + max suppx1 3 f x1 + xi ≤ 1+ + ≤ 1+ −−−−→ . n n 2n 2n n→∞ 2  Xi=2  This yields a contradiction for sufficiently large n since 3/2 < 2(1 − ε). 

ξ Proposition 3.14. The space Tinc is not Asymptotic ℓ1. ξ N 2 Proof. Suppose that Tinc is C-Asymptotic ℓ1 and let n ∈ be such that n > C . ξ Since Tinc is reflexive, we may assume that player (S) chooses tail subspaces (see [AGLM, Lemma 5.18]) throughout any winning strategy in the game G(n, 1, C). Let us assume the role of player (V) and let Y1 be the tail subspace with which player (S) initiates the game. Then, as player (V ), we choose an element of the basis ej1 ∈ Y1, such that |S| ≥ n for every maximal segment S of Tξ with min S = j1. Suppose that in the k + 1 turn of the game, for k

Gξ = a e∗ : S is a segment of T and |a |q ≤ 1 p  i i ξ i  Xi∈S Xi∈S where p−1 + q−1 = 1, to obtain a reflexive Banach space admitting a uniformly ξ unique ℓ1 joint spreading model, that contains a weakly null ℓp-tree of height ω or ξ ∗ N a weakly null c0-tree if we replace G2 with G = {±ei : i ∈ }. 10 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

4. A stronger separation of the two properties ξ The spaces Tinc constructed in the previous section, yield a separation between the properties of being an Asymptotic ℓ1 space and admitting a unique ℓ1 asymp- totic model. It is easy however to see that these spaces contain subsequences of their bases that generate Asymptotic ℓ1 subspaces. For example, consider any subspace ξ generated by a subsequence (ej)j∈M of the basis of some Tinc, such that the ele- ments of M are pairwise incomparable in Tξ. In this section we show that, for any ξ countable ordinal ξ, there is a reflexive Banach space Tess-inc that admits a unique F ξ ℓ1 asymptotic model with respect to b(Tess-inc) and any subsequence of its basis generates a non-Asymptotic ℓ1 subspace. To some extent, this family of spaces is the Maurey - Rosenthal [MR] analogue of the two aforementioned properties. Start by fixing a countable ordinal ξ and let (mj )j≥0, (nj )j≥0 be increasing sequences of natural numbers such that : 2 (i) m0 = 2, m1 = 4 and mj ≥ mj−1 for every j ≥ 2 and 2 (ii) n0 = 1, n1 = 6 and nj > log2 mj + nj−1 for every j ≥ 2.

Let Q denote the collection of all finite sequences ((g1,mj1 ),..., (gk,mjk )), where N N gi : → {−1, 0, 1} has finite support and ji ∈ for 1 ≤ i ≤ k, and mj1 < N ··· < mjk . Let σ : Q→{mj : j ∈ } be an injection so that each sequence

((g1,mj1 ),..., (gk,mjk )) is mapped to some mj with mjk

Definition 4.1. Let Tξ be the set of all finite sequences ((g1,mj1 ),..., (gk,mjk )) satisfying the following conditions. e (i) gi : N → {−1, 0, 1} for i =1,...,k with suppg1 <...< suppgk. (ii) suppg ∈ S for i =1,...,k, where n = n and n <...

(iii) mj1 = m1 and mji = σ((g1,mj1 ),..., (gi−1,mji−1 )) for every i =2,...,k. (iv) {min suppgi : i =1,...,k} ∈ Sξ.

Note that item (iii) of the above definition implies that Tξ, equipped with the partial order ≤ e where t˜1 ≤ e t˜2 if t˜1 is an initial segment of t˜2, is a tree. Moreover, Tξ Tξ e it is easy to see that it is infinite-branching, and as follows from item (iv) and standard inductive arguments, it is also well founded and of height ωξ. In particular, the above remain true if for an infinite subset of the naturals M we additionally require that suppgi ⊂ M for every i =1,...,k, in Definition 4.1. We may also identify Tξ as a closed subset, with respect to the pointwise conver- −1 N gence topology, of {±mej }j∈N ∪{0} via the mapping  −1 −1 ((g1,mj1 ),..., (gk,mjk )) 7→ mj1 g1 + ··· + mjk gk. −1 −1 N The fact that limj mj = 0 implies that {±mj }j∈N ∪{0} is compact with respect to the pointwise convergence topology of [−1, 1]N. ˜ Observe that, as a consequence of item (iii), any t = ((g1,mj1 ),..., (gk,mjk )) in

Tξ is uniquely determined by the pair (gk,mjk ), which we will denote by (gt,mjt ) or just by t (i.e., t = (g ,m ). Taking advantage of this we may define T = {(g ,m ): e t jt ξ t jt ˜ ˜ t ∈ Tξ}, which is in bijection with Tξ via the mapping t = (gt,mjt ) 7→ t. Note that ≤ e induces a natural order, denoted by ≤T , on Tξ, where (gt ,mj ) ≤T (gt ,mj ) Tξ e e ξ 1 t1 ξ 2 t2 if t˜ ≤ e t˜ . Clearly, the tree (T , ≤ ) is isomorphic to (T , ≤ e ) via the mapping 1 T 2 ξ Tξ ξ Tξ ˜ t = (gt,mjt ) 7→ t. e

Definition 4.2. Let Wξ be the set of all finite sequences (mj1 ,mj2 ,...,mjk ) for which there exist g ,...,g : N → {−1, 0, 1} with ((g ,m ),..., (g ,m )) ∈ T . 1 f k 1 j1 k jk ξ

The initial segment order ≤ f is a partial order on Wξ and is in fact natu-e Wξ rally induced by the order ≤ e . Moreover, it is easy to verify that (Wξ, ≤ f ) is a Tξ f Wξ f NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTIC MODEL 11 well founded infinite-branching tree of height ωξ. It is also isomorphic to the tree (W , ≤ ), where W = {m : t˜ ∈ T } and m ≤ m if t˜ ≤ e t˜ . This ξ Wξ ξ jt ξ jt1 Wξ jt2 1 Tξ 2 correspondence between Wξ and Wξ ise identical to that of Tξ and Tξ. Remark 4.3. (i) If m , m are incomparable nodes in W , then for every fjt1 jt2 e ξ g ,g : N → {−1, 0, 1} such that (g ,m ) and (g ,m ) are in T , these 1 2 1 jt1 2 jt2 ξ are also incomparable. (ii) Note that there exist nodes t˜1 and t˜2 which are incomparable in Tξ, whereas mj and mj are comparable in Wξ. To see this, consider any node t1 t2 e ˜ t = ((g1,mj1 ),..., (gk,mjk )) in Tξ with k> 1 and, for each i =1,...,k − 1, let h : N → {−1, 0, 1} be such that h 6= g and t = (h ,m ) is in T . i e i i i i ji ξ Then, item (iii) of Definition 4.1 implies that the nodes t˜i and t˜ are incom-

parable whereas mjti and mjt are comparable for every i = 1,...,k − 1, since t˜∈ Tξ.

Definition 4.4. Wee say that a subset X of Tξ is essentially incomparable if when- ever (g ,m ), (g ,m ) are in X with m < m and g : N → {−1, 0, 1} is t1 jt1 t2 jt2 jt1 Wξ jt2 the unique sequence such that (g,m ) ≤ (g ,m ), then suppg< suppg . jt1 Tξ t2 jt2 t1

Remark 4.5. Let X be an essentially incomparable subset of Tξ and ht : N →

{−1, 0, 1} with suppht ⊂ suppgt for every t ∈ X. Then {(ht,mjt ): t ∈ X} is also an essentially incomparable subset of Tξ. The following lemma is an extension of Proposition 3.1 and is the main ingredient ξ of the proof that the space Tess-inc admits a uniformly unique joint spreading model.

Lemma 4.6. Let (µi)i be a sequence of positive regular measures on Tξ with finite N supports and let C > 0 be such that µi(Tξ) < C for all i ∈ . Assumee that the sets ∪{suppg : t˜ ∈ suppµ }, i ∈ N, are disjoint. Then, for every ε > 0, there exists an t i e N ∞ 1 2 M ∈ [ ] and Gi , Gi subsets of Tξ for each i ∈ M, such that 1 2 (i) Gi , Gi are disjoint subsetse of suppµi for every i ∈ M, 1 2 (ii) µi(Tξ \ Gi ∪ Gi ) <ε for every i ∈ M, (iii) {t ∈ T : t˜∈ ∪ G1} is essentially incomparable and e ξ i∈M i (iv) for every i 6= i in M, every t˜ ∈ G2 and t˜ ∈ G2 , the nodes m and 1 2 1 i1 2 i2 jt1 m are incomparable in W . jt2 ξ Before we are able to prove this Lemma it is necessary to introduce the notion of successor limits of measures. We find this limit notion to be of independent interest and therefore we use broader terminology to define it and prove its properties. Notation 4.7. Let T be a countably branching well founded tree. For each t ∈ T we denote succT (t) the set of immediate successors of t. In particular, if t is maximal then succT (t) is empty. For t ∈ T we denote Vt = {s ∈ T : t ≤ s}. We view T as topological space with the topology generated by the sets Vt and T\Vt, t ∈ T . This is a compact metric topology for which the sets of the form Vt \ (∪s∈F Vs), t ∈ T and F ⊂ succT (t) finite, form a base of clopen sets. We denote by M+(T ) the cone of all bounded positive measures µ : P(T ) → [0, +∞). For µ ∈ M+(T ) we define the support of µ to be the set supp(µ)= {t ∈ T : µ({t}) > 0}. A set A in M+(T ) is called bounded if supµ∈A µ(T ) < ∞. ∗ Recall that a sequence (µi) in M+(T ) converges in the w -topology to a µ ∈ M+(T ) if and only if for all clopen sets V ⊂ T we have limi µi(V ) = µ(V ) if and only if for all t ∈ T we have limi µi(Vt)= µ(Vt).

Definition 4.8. Let T be a countably branching well founded tree, (µi) be a disjointly supported sequence in M+(T ) and ν ∈ M+(T ). We say that ν is the successor-determined limit of (µi) if for all t ∈ T we have µ({t}) = limi µi(succT (t)). In this case we write ν = succ-limi µi. 12 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

Remark 4.9. It is possible for a disjointly supported and bounded sequence (µi) ∈ ∗ ≤2 M+(T ) to satisfy w - limi µi 6= succ-limi µi. Take for example T = [N] (all subsets of N with at most two elements with the partial order of initial segments). ∗ Define µi = δ{i,i}. Then, w - limi µi = δ∅ whereas succ-limi µi = 0. Although these limits are not the necessarily the same, there is an explicit formula ∗ relating succ-limi µi to w - limi µi.

Lemma 4.10. Let T be a countable well founded tree, (µi) be a bounded and ∗ disjointly supported sequence in M+(T ) so that w - limi µi = µ exists and for all t ∈ T the limit ν({t}) = limi µi(succT (t)) exists as well. Then, for every t ∈ T and enumeration (tj ) of succT (t) we have

(1) µ({t})= ν({t}) + lim lim µi ∪k≥j (Vtk \{tk}) . j i   In particular, µ({t})= ν({t}) if and only if the double limit in (1) is zero. N Proof. For j ∈ we have {t} ∪ (∪k≥j Vtk )= Vt \ (∪k

(2) lim µi {t} ∪ (∪k≥j Vt ) = µ {t} ∪ (∪k≥j Vt ) . i k k   Because (µi) is disjointly supported we observe that for all j ∈ N

(3) lim µi({tk : k ≥ j}) = lim µi(succT (t)) = ν({t}). i i We calculate (2) µ({t}) = lim µ {t} ∪ (∪k≥j Vtj ) = lim lim µi {t} ∪ (∪k≥j Vtj ) j→∞   j i  

= lim lim µi(∪k≥j Vtj ) = lim lim µi({tk : k ≥ j} ∪ (∪k≥j (Vt \{tk}))) j i j i k

= lim lim µi({tk : k ≥ j}) + lim lim µi ∪k≥j (Vtk \{tk}) . j i j i   Thus, (3) yields the conclusion. 

Corollary 4.11. Let T be a countable well founded tree and (µi) be a bounded and disjointly supported sequence in M+(T ). Then, there exist a subsequence (µin ) of

(µi) and ν ∈M+(T ) with ν = succ-limn µin .

∗ Proof. By passing to a subsequence, µ = w - limi µi exists and for all t ∈ T the limit ν({t}) = limi µi(succT (t)) exists as well. By (1) for all t ∈ T we have ν({t}) ≤  µ({t}). Thus t∈T ν({t}) ≤ µ(T ), i.e., ν defines a bounded positive measure P Lemma 4.12. Let T be a countable well founded tree and (µi) be a bounded and disjointly supported sequence in M+(T ) so that succ-limi µi = ν exists. Then, there exist an infinite L ⊂ N and partitions Ai, Bi of supp(µi), i ∈ L, so that the following are satisfied. 1 1 (i) If for all i ∈ L we define the measure µi given by µi (C)= µi(C ∩ Ai), then ∗ 1 1 ν = w - limi∈L µi = succ-limi∈L µi . 2 2 (ii) If for all i ∈ L we define the measure µi given by µi (C)= µi(C ∩ Bi) then 2 for all t ∈ T the sequence (µi (succT (t)))i is eventually zero. In particular, 2 succ-limi∈L µi = 0.

Proof. Enumerate T = {sn : n ∈ N} and assume, passing if necessary to a subse- quence, that for all n ∈ N and i>n we have 1 (4) |µ (succ (s )) − ν(s )| < . i T n n 2n NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTIC MODEL 13

Let us point out that for m 6= n the sets succT (sm) and succT (sn) are disjoint and ∪nsuccT (sn)= T\{t0}, where t0 denotes the root of the tree T . We may, and will, assume that for all i ∈ N, t0 6∈ supp(µi). Define for each i ∈ N the sets

i ∞ Ai = supp(µi) ∩ ∪n=1 succT (sn) and Bi = supp(µi) ∩ ∪n=i+1 succT (sn) .     We point out that for all i ∈ N, Ai, Bi forms a partition of supp(µi) and we will show that it has the desired properties. Statement (ii) follows directly from the fact that for every t ∈ T the sequence of sets (Bi ∩ succT (t))i is eventually empty. To show that (i) holds we fix t ∈ T and ∞ N let (tj ) be an enumeration of succT (t). Define Lj = ∪k=j {n ∈ : tk ≤ sn}, for each j ∈ N, and observe that ∩j Lj = ∅. Also observe that for all j ∈ N we have

∪k≥j (Vtk \{tk})= ∪n∈Lj succT (sn). Therefore we have

i µi Ai ∩ ∪k≥j (Vtk \{tk}) = µi ∪n=1 succT (sn) ∩ ∪n∈Lj succT (sn)     

= µi ∪n∈Lj ∩[1,i] succT (sn) = µi succT (sn)   n∈LXj ∩[1,i]  (4) 1 ≤ ν(s )+ ≤ ν({s : n ∈ L })+2− min(Lj )+1 n 2n n j n∈LXj ∩[1,i] n∈LXj ∩[1,i]

− min(Lj )+1 = ν(∪k≥j Vtk )+2 .

Therefore, limj supi µi Ai ∩ ∪k≥j (Vtk \{tk}) = 0 and by Lemma 4.10, (i) is satisfied.    Proof of Lemma 4.6. Apply Lemma 4.12 so that, by passing to a subsequence of (µi), there are, for each i ∈ N, partitions Ai, Bi of supp(µi) so that the conclusion N 1 2 of that Lemma it satisfied. Define, for each i ∈ , the measures µi , µi given by 1 2 ∗ 1 1 µi (C) = µi(Ai ∩ C) and µi (C) = µi(Bi ∩ C). Let ν = w - limi µi = succ-limi µi . ∗ 1 Pick a finite subset F of Tξ so that ν(Tξ \ F ) < ε/2. Then, because ν = w - limi µi 1 1 we have limi µi (Tξ)= ν(Teξ) and becausee ν = succ-limi µi 1 1 ˜ 1 ˜ ε lim µi (Tξ) − µei (∪t˜∈F succ(e t)) = ν(Tξ) − lim µi (succ(t)) = ν(Tξ \ F ) < . i i 2 tX˜∈F e e e We can find i0 ∈ N so that for all i ≥ i0 we have ˜ 1 1 ˜ ε (5) µi(Ai) − µi Ai ∩ ∪t˜∈F succ(t) = µi (Tξ) − µi (∪t˜∈F succ(t)) < .   2  We may, using the fact that the sets ∪{supp gt : t˜e∈ suppµi} for i ∈ N are disjoint, find j0 ≥ i0 ∈ N such that ˜ 1 (6) ∪s˜∈F suppgs < suppgt for every t ∈ ∪i≥j0 supp(µi ). 1 ˜ We define Gi = Ai ∩ (∪t˜∈F succ(t)), i ≥ j0. By (5) we have that for all i ≥ 1 ˜ i j0, |µi(Ai) − µi(Gi )| < ε/2. Additionally, {t ∈ Tξ : t ∈ ∪i≥j0 G1} is essentially incomparable. Indeed, lets ˜ , s˜ ∈ ∪ G1 with m < m and (h,m ) ∈ T 1 2 i≥j0 i js1 W js2 js1 ξ be such that and (h,m ) ≤ s . Then (6) implies that supph< suppg . js1 Tξ 2 s1 N 2 For the remaining part of the proof, since for all i ∈ the set Bi = suppµi is finite (as a subset of the finite support of µi) and for each t˜ ∈ Tξ the se- 2 ˜ quence (µi (succ(t)))i is eventually zero, we may pass to a subsequence so that ˜ 2 ˜ 2 e for all i < j we have {mjt : t ∈ suppµi }∩{mjt : t ∈ suppµj } = ∅. We can therefore define the bounded sequence of disointly supported measures (νi) on Wξ with ν ({(w ,...,w )}) = µ2({t˜ ∈ T : m = w }). Hence, applying Proposi- i 1 k i ξ jt k f tion 3.1 and passing to a subsequence, we obtain a subset E of suppν such that e i i νi(Wξ \ Ei) < ε/2 and the sets Ei, i ∈ N, are pairwise incomparable. It is easy f 14 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

2 ˜ N to verify that Gi = {t ∈ Bi : mjt ∈ 2i}, i ∈ , are pairwise incomparable and 2 2 2 N  |µi(Bi) − µi(Gi )| = µi (Tξ \ Gi ) < ε/2 for every i ∈ .

e ξ ξ We now define the space Tess-inc in a similar way to Tinc, that is, using the notion of the Tsirelson extension WG of a ground set G.

Definition 4.13. Define the following norming sets on c00(N). ∗ N G0 = ± en : n ∈  1 ∗ N N G1 = g(n)en : j ∈ and g : → {−1, 0, 1} with suppg ∈ Snj . mj n nX∈N o −1 ∗ For each f = mj n∈N g(n)en in G1, set tf = (g,mj ). Moreover, if G = G1 ∪ G0 and f is in WG withP a tree analysis (fα)α∈A, define 1 Mf = {α : α is a maximal node of A and fa ∈ G1}. 1 Let W be the subset of WG containing all functionals f such that {tfα : α ∈Mf } ξ is an essentially incomparable subset of Tξ. Denote by Tess-inc the completion of c00(N) with respect to the norm k·kW induced by W .

Remark 4.14. (i) The standard basis (ej )j of c00(N) forms a 1-unconditional ξ ξ basis for the space Tess-inc and it is also boundedly complete since Tess-inc admits a uniformly unique ℓ1 spreading model as shown in Proposition 4.15. 1 (ii) If f ∈ WG with a tree analysis (fα)α∈A and mjt , for α ∈M , are pairwise fα f incomparable nodes in Wξ , then f ∈ W . ξ (iii) The norming set of Tess-inc contains the norming set of Tsirelson’s original space, i.e., the Tsirelson extension of G0. ξ Proposition 4.15. The space Tess-inc admits ℓ1 a uniformly unique joint spreading F ξ model with respect to b(Tess-inc).

i ξ Proof. Let (xj )j , 1 ≤ i ≤ l, be an array of normalized block sequences in Tess-inc i1 i2 and fix ε> 0. Passing to a subsequence, we may assume that suppxj < suppxj+1 for all i1,i2 =1,...,l and j ∈ N. For each i =1,...,l and j ∈ N, pick a functional i i i kj,α i i i i f = i f /2 in W with f (x ) ≥ 1 − ε and f (x ) > 0 for every j α∈Mj j,α j j j,α j Pi i α ∈Mj , where Mj denotes the set of all maximal nodes of a fixed tree analysis of i 1 i i i fj . Moreover, for each α ∈ M i = {α ∈ Mj : fj,α ∈ G1}, define tj,α = tf i and, fj j,α for each j ∈ N, the measure µj as follows: i i fj,α(xj ) µj = i δt˜i . kj,α j,α Xi α∈MX1 2 fi j

Passing to a subsequence assume that limj µj (Tξ) = c. If c = 0, then we may assume that f i ∈ G for every i =1,...,l, j ∈ N and α ∈Mi , in which case the j,α 0 e j desired result is immediate. Hence, if c> 0, applying Lemma 4.6 and passing to a 1 2 1 2 subsequence, we obtain (Gj )j , (Gj )j satisfying items (i) - (iv) with µj (Tξ \Gj ∪Gj ) < 1/8. Then, for each pair (i, j), set e 1 i i 1 2 j 1 M = {α ∈M i : t ∈ G } and M = M \M i,j fj j,α j i,j i i,j and i k i kj,α fi,j = fj,α/2 , k =1, 2. k α∈MXi,j NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTIC MODEL 15

1 2 Note in particular that, for every pair (i, j), the fact that µj (Tξ \ Gj ∪ Gj ) < 1/8 implies that |f i(xi )−(f 1 (xi )+f 2 (xi ))| < 1/8 and hence that there exists k =1, 2 j j i,j j i,j j e k i such that fi,j (xj ) ≥ (7 − ε)/16. Set k i Ak = {(i, j): fi,j (xj ) ≥ (7 − 8ε)/16}, k =1, 2. N l,n l N k Let n ∈ , {λij }i=1,j=1 ⊂ [−1, 1] with i,j |λij | = 1 and s = (si)i=1 ∈ S-Plml([ ] ) with ln ≤ min suppx1 . Then let kP= 1, 2 be such that |λ | ≥ 1/2 s1(1) (i,si(j))∈Ak ij and observe that f =1/2 f k is in W . Hence,P we calculate (i,si(j))∈Ak i,si(j) P l n l n 1 7 − 8ε |λ |xi ≥ f |λ |xi = |λ |f k xi ≥ ij si (j)  ij si(j) 2 ij i,si(j) si(j) 32 Xi=1 Xj=1 Xi=1 Xj=1 (i,siX(j))∈Ak 

and due to unconditionality this yields that l n i 7 − 8ε λij x ≥ . si(j) 32 Xi=1 Xj=1

 N ∞ ξ It remains to show that for every M ∈ [ ] , the space Tess-inc contains a c0-tree ξ of height ω supported by (ej )j∈M . To this end, let us recall the following definition. Definition 4.16. Let n ∈ N and ε > 0. We say that a convex combination N x = i∈∆ λiei in c00( ) is an (n,ε)-special convex combination if P(i) ∆ ∈ Sn and ′ (ii) i∈∆′ λi <ε for every ∆ ∈ Sm with m 0, there is a k ∈ N such that, for every maximal subset F of Sn with k < F , there exists an (n,ε)-special convex combination x in c00(N) with suppx = F .

For a functional f in W with tree analysis (fα)α∈A, we define the height of f, denoted by h(f), as the maximum of |a| over all maximal nodes α ∈ A. Moreover, −1 ∗ if f = mj n∈N g(n)en is in W , we say that f is a weighted functional and define the weight ofPf as w(f)= mj .

Lemma 4.18. Let j ∈ N and f be a functional in W with a tree analysis (fα)α∈A 1 such that w(fα) mj } and ∆2 = ∆ \ ∆1. 1 1 Consider the tree analysis (fα)α∈A of f1 = f|∆1 and note that w(fα) < mj for every α ∈M1 . Indeed, if w(f 1)= m ′ ≥ m for some α, then for any i ∈ suppf 1 f1 α j j α −1 we have that |f(ei)| ≤ mj′ and this is a contradiction. Moreover, the fact that 16 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

−1 |f1(ei)| >mj for every i ∈ ∆1 = suppf1 implies that h(f1) < log2 mj−1 and hence the previous proposition yields that suppf1 ∈ Sl, where l ≤ log2 mj + nj−1 < nj . −2 Therefore, since x = i∈∆ λiei is an (nj ,mj )-special convex combination, we have that P 1 |f|∆1 ( λiei)|≤ λi < 2 . mj iX∈∆ iX∈∆1 We also calculate 1 1 |f|∆2 ( λiei)|≤ λi ≤ mj mj iX∈∆ iX∈∆2 −1 −2 and conclude that kxkW ≤ mj + mj . For the remaining part notice that the −1 ∗  functional f = mj i∈∆ ei is in W . P N −2 Proposition 4.20. Let j ∈ and x = i∈∆ λiei be an (nj ,mj )-special convex −2 combination. Then |f(x)| < 2mj , for everyP f ∈ W with a tree analysis (fα)α∈A 1 such that w(fα) 6= mj for all α ∈Mf .

−2 1 Proof. Define ∆1 = {i ∈ ∆ : |f(ei)| >mj } and ∆2 = ∆ \ ∆1 and let (fα)α∈A1 be the tree analysis of f1 = f|∆1 . Similar arguments as in the previous proof yield that 1 2 1 1 w(fα) < mj < mj+1 and hence w(fα) < mj for all α ∈ Mf1 , since w(fα) 6= mj 1 2 for all α ∈ Mf . Moreover, since |f1(ei)| > mj for all i ∈ suppf1, we have that 2 h(f1) < log2 mj and therefore Proposition 4.18 yields that ∆1 = suppf1 ∈ Sl with 2 −2 l ≤ log2 mj + nj−1 < nj . The fact that x = i∈∆ λiei is an (nj ,mj )-special convex combination implies that P 1 |f1( λiei)|≤ λi < 2 . mj iX∈∆ iX∈∆1 We also calculate 1 1 |f|∆2 ( λiei)|≤ 2 λi ≤ 2 mj mj iX∈∆ iX∈∆2 and this completes the proof. 

Let M be an infinite subset of the naturals and consider the collection Tξ(M) of all finite sequences (x1,...,xk) of vectors in c00(N) such that (i) x = m λ e , where λ e is an (n ,m−2)-special convex com- l jl i∈∆l i i i∈∆l i i jl jl binationP for every l =1,...,kP, (ii) ∆l is a subset of M for every l =1,...,k and χ χ (iii) (( ∆1 ,mj1 ),..., ( ∆k ,mjk )) ∈ Tξ. Note that Tξ(M), equipped with the initiale segment order, is a well-founded infinite branching tree of height ωξ.

Proposition 4.21. Let M be an infinite subset of the naturals and (x1,...,xk) be any node of Tξ(M), then kx1 + ... + xkkW ≤ 3.

Proof. Let f ∈ W with a tree analysis (fα)α∈A. Observe that there exists at most one 1 ≤ l ≤ k such that there is an α ∈M1 with w(f )= m and suppf ∩ ∆ 0 f α jl0 a l0 1 is non-empty. Indeed, suppose that there exist 1 ≤ l1 < l2 ≤ k and α1, α2 ∈ Mf with w(f ) = m , w(f ) = m , suppf ∩ ∆ 6= ∅ and suppf ∩ ∆ 6= ∅. α1 jl1 α2 jl2 α1 l1 α2 l2 Then since {t : α ∈M1 } is essentially incomparable and m < m we have fα f jl1 Wξ jl2 that ∆l1 < ∆tf = suppfα1 which is a contradiction. α1 1 Therefore, for any l 6= l0, we have that w(fα) 6= mjl for every α ∈ Mf and the −1 previous proposition yields that |f(xl)| < 2mjl . Moreover, Proposition 4.19 yields NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTIC MODEL 17

−1 that |f(xl0 )|≤ 1+ m and hence we conclude that jl0 k 1 |f(x1 + ··· + xk)|≤ 1+2 ≤ 3. mj Xl=1 l 

ξ The previous proposition and the fact that the tree Tξ(M) is of height ω yield the following result. N ∞ ξ Theorem 4.22. For every M ∈ [ ] , the space Tess-inc contains a c0-tree of ξ height ω , supported by (ej )j∈M . In particular, the space generated by (ej )j∈M is not Asymptotic ℓ1. Remark 4.23. There exist modifications of the ground set G that yield, for any 1

Proof. Note that since Tξ is a countable compact space with respect to the pointwise convergence topology, the completion of c00(N) with respect to k·kG is embedded in C[Tξ], i.e., the space of all continuous real functions on Tξ, and hence is c0-saturated. ξ Furthermore, Tess-inc admits a boundedly complete basis and therefore does not contain c0. The above imply that the identity operator Id : (c00(N), k·kW ) → (c00(N), k·kG) is strictly singular and hence for any normalized block sequence ξ (xj )j in Tess-inc there exists a subsequence (xj )j∈M such that limj∈M kxj kG = 0. The remainder of the proof is identical to the last paragraph of Proposition 3.13. 

5. More non-asymptotic ℓp spaces with uniformly unique ℓp joint spreading models In this final section we show that, for every 1

m m ξ N q ξ G1,p = bifi : m ∈ , |bi| ≤ 1,fi ∈ G1 for i =1,...,m and n Xi=1 Xi=1

suppf1,..., suppfm are pairwise disjoint . o ξ N Denote by JT1,p the completion of c00( ) with respect to the norm induced by the ξ norming set G1,p. We start with some necessary remarks on the above norming sets and a Ramsey type result. Remark 5.2. ξ j ∗ N Let (fj )j be a sequence in G1 with fj = i∈Sj ǫi ei , j ∈ , and N i for each i, j ∈ , define ǫj(i) = ǫj if i ∈ Sj and ǫj(i) =P 0 otherwise. Passing to a subsequence, we may assume that (Sj )j converges pointwise to a segment S, N since Tξ is well-founded, and that (ǫj )j also converges to some ǫ in {−1, 1} . Then, ∗ ξ clearly, (fj)j converges pointwise to f = i∈S ǫ(i)ei and f is in G1. P 18 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

ξ Remark 5.3. Let x be a normalized vector in JT1,p with finite support. ξ (i) If for some ε > 0 there is a family {fi}i∈I in G1 whose members have −p pairwise disjoint supports and |fi(x)|≥ ε for all i ∈ I, then #I ≤ ε . ξ (ii) Let f1,...,fm ∈ G1 have pairwise disjoint supports and suppfi ⊂ range(x) for i =1,...,m. Then, for any choice of scalars b1,...,bm, we have that

m q m q bifi(x) ≤ |bi| .

Xi=1 Xi=1 ξ Definition 5.4. We call a family (Fj )j of finite subsets of JT1,p a normalized block family if for any choice of xj ∈ Fj , j ∈ N, the sequence (xj )j is block and kxk =1 for any x ∈ Fj and j ∈ N. Moreover, for such a family, define M(Fj ) = max{suppx : x ∈ Fj } and r(Fj ) = #(M(Fj−1),M(Fj )], where M(F0) = 0. ξ Lemma 5.5. Let (Fj )j be a normalized block family in JT1,p with supj #Fj < ∞. ∞ Then, for every ε> 0 and n0 ∈ N, there is an L ∈ [N] such that, for any segment ξ S of Tξ with min S ≤ n0 and any f ∈ G1 with suppf = S, there is at most one j ∈ L with the property that |f(x)|≥ ε for some x ∈ Fj . ξ Proof. For a segment S of Tξ, let GS denote the set of all f ∈ G1 with suppf = S. If the conclusion is false for some ε > 0 and n0 ∈ N, then using Ramsey Theorem from [Ra], there exists an L ∈ [N]∞ such that, for any i < j in L, there is a segment Sij with min Sij ≤ n0, a functional fij ∈ GSij and xij ∈ Fi, yij ∈ Fj such that |fij (xij )| ≥ ε and |fij (yij )| ≥ ε. Assume for convenience that L = N. Since # supj Fj < ∞, using the pigeon hole principle and a diagonal argument we may assume that there exist sequences (xj )j , (yj )j such that xj ,yj ∈ Fj and, for every N i < j ∈ , a segment Sij of Tξ with min Sij ≤ n0 and an fij ∈ GSij such that |fij (xi)|≥ ε and |fij (yj )|≥ ε. For each i

ξ Lemma 5.6. Let ε > 0 and (Fj )j be a normalized block family in JT1,p with supj #Fj < ∞. Then there exists a strictly increasing sequence (nj )j of naturals and a decreasing sequence (εj )j of positive reals such that N ξ (i) for every j ∈ , every segment S of Tξ with min S ≤ M(Fnj ) and f ∈ G1 ′ with suppf = S, there exists at most one j > j such that |f(x)| ≥ εj for

some x ∈ Fnj′ and ∞ ∞ (ii) j=1 r(Fnj ) i=j (i + 1)εi <ε. P P ∞ Proof. Let (δj )j be a sequence of positive reals such that j=1 δj < ε. We will construct (nj )j and (εj )j by induction, along with a decreasingP sequence (Lj )j of infinite subsets of N. Set n1 = 1 and L1 = N and choose ε1 > 0 such that 2r(F1)ε1 <δ1. Suppose that n1,...,nj , ε1,...,εj and L1,...,Lj have been chosen ∞ for some j in N. Then, the previous lemma yields an Lj+1 ∈ [Lj] such that, for ξ every segment S of Tξ with min S ≤ M(Fnj ) and every f ∈ G1 with suppf = S, NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTIC MODEL 19

′ there is at most one j > j such that |f(x)| ≥ εj for some x ∈ Fnj′ . Choose nj+1 ∈ Lj+1 with nj+1 >nj and εj+1 <εj such that

(a) r(Fnj+1 )(j + 2)εj+1 <δj+1 and j+1 (b) r(Fnk ) i=k (i + 1)εi <δk for all k ≤ j. It follows quiteP easily that (nj )j and (εj )j are as desired. 

ξ Proposition 5.7. Let ε> 0 and (Fj )j be a normalized block family in JT1,p with supj #Fj < ∞ satisfying the following. N ξ (i) For every j ∈ , every segment S of Tξ with min S ≤ M(Fn) and f ∈ G1 ′ with suppf = S, there exists at most one j > j such that |f(x)| ≥ εj for some x ∈ Fj′ and ∞ ∞ (ii) j=1 r(Fj ) i=j (i + 1)εi <ε. Then, forP every n ∈ NP, every choice of x1,...,xn with xj ∈ Fj and scalars a1,...,an, we have that

n 1 n n 1 p p 1 p p |aj | ≤ aj xj ≤ (2 q + ε) |aj| .     Xj=1 Xj=1 Xj=1

ξ Proof. The lower inequality follows easily from the definition of G1,p. Let us first observe that if (xj )j is a sequence with each xj ∈ Fj , then, for any j ∈ N and any ξ segment S of Tξ with M(xj−1) < min S ≤ M(xj ) and f ∈ G1 with suppf = S, the following hold due to (i).

(a) #{i > j : |f(xi)|≥ εj }≤ 1. (b) #{i > j : εk−1 > |f(xi)|≥ εk}≤ k for all k > j. m ξ Let f = i=1 bifi be in G1,p with suppfi = Si, for i =1,...,m. For each i, we will denote byPji,1 the unique 1 ≤ j ≤ n such that M(xji,1−1) < min Si ≤ M(xji,1 ) and by ji,2 the unique, if there exists, ji,1 < j ≤ n such that |fi(xji,2 )| ≥ εji,1 . Denote by fi,1 the restriction of fi to range(xji,1 ) ∩ range(xji,2 ) and set fi,2 = fi − fi,1 for i = 1,...,m, and Ij = {i : j = ji,1 or j = ji,2} for j = 1,...,n. Note that, n q due to (a), each i appears in Ij for at most two j and hence j=1 i∈Ij |bi| ≤ 2. We thus calculate applying item (ii) of Remark 5.3 P P

m n n n 1 n 1 p p q q bifi,1 aj xj = aj bifi,1(xj ) ≤ |aj | bifi,1(xj ) Xi=1  Xj=1  Xj=1 iX∈Ij  Xj=1   Xj=1 iX∈Ij 

n 1 n 1 n 1 p p q q 1 p p ≤ |aj | |bi| ≤ 2 q |aj | .

 Xj=1   Xj=1 iX∈Ij   Xj=1  N Finally, for each j ∈ , set Gj = {i : M(xji,1 −1) < min Si ≤ M(xji,1 )}. Note that, n ∞ as follows from (b), #Gj ≤ r(Fj ) and |fi,2( k=1 xk)| < k=i(k + 1)εk for any m n i ∈ Gj . Hence (ii) yields that i=1 |fi,2( k=1Pxk)| <ε andP we conclude that

m n P n Pm n 1 p p bifi,2 aj xj = aj bifi,2(xj ) <ε |aj |     Xi=1 Xj=1 Xj=1 Xi=1 Xj=1 which along with the above calculation yield the desired result. 

ξ Proposition 5.8. The space JT1,p admits a uniformly unique joint spreading model F ξ with respect to b(JT1,p), equivalent to the unit vector basis of ℓp.

1 l ξ Proof. Let (xj )j ,..., (xj )j be normalized block sequences in JT1,p and let ε > 0. i Applying Lemma 5.6 and passing to a subsequence, we may assume that Fj = {xj : ξ i = 1,...,l} is a normalized block family in JT1,p satisfying items (i) and (ii) of 20 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

N l k Proposition 5.7. Then, for every k ∈ , every s = (si)i=1 in S-Plml([L] ) and any l,k choice of scalars (aij )i=1,j=1, we calculate

l k 1 l k l k 1 p 1 p p i q p |aij | ≤ aij xsi(j) ≤ (2 + ε) |aij | .     Xi=1 Xj=1 Xi=1 Xj=1 Xi=1 Xj=1

N ∞ i l A diagonal argument then yields that there exists L ∈ [ ] such that ((xj )j∈L)i=1 1 generates a joint spreading model 2 q -equivalent to the unit vector basis of ℓp. 

ξ Proposition 5.9. The space JT1,p is reflexive.

Proof. Note that the unit vector basis of c00(N) forms a boundedly complete un- ξ conditional Schauder basis for JT1,p, that is, it does not contain c0. Moreover, Proposition 5.7 yields that it does not contain ℓ1 and hence Theorem 2 from [J1] yields the desired result. 

ξ Proposition 5.10. The space JT1,p is not Asymptotic ℓp.

ξ 1 N q Proof. Suppose that JT1,p is C-Asymptotic ℓp and let n ∈ be such that C ≤ n . Then, following the same arguments as in Proposition 3.14, in the final outcome of G(n,p,C) we, as player (V), have chosen elements of the basis ej1 ,...,ejn such n that {j1,...,jn} is a segment of Tξ and hence {ej1 ,...,ejn } is isometric to ℓ1 . We then calculate n n 1 1 1 − p − p q n eji = n eji ξ = n G1 Xi=1 Xi=1 ξ whereas, since JT1,p is C-Asymptotic ℓp, we have that n 1 − p n eji ≤ C

Xi=1 and this is a contradiction. 

ξ ξ Remark 5.11. We may also define a conditional version of JT1,p, denoted as JTp , ξ by replacing the norming set G1 with

ξ ∗ Gsum = ei : S is a segment of Tξ . n Xi∈S o ξ Note that the above results hold for JTp . For the reflexivity part, notice that it ξ ∗ ξ ∗ suffices to show that (ej)j is shrinking for JTp . If not, then there is an x ∈ (JTp ) \ ∗ ∞ ∗∗ ξ ∗∗ ∗∗ ∗ N ∗∗ ∗ span{ej }j=1 and an x ∈ (JTp ) with x (ej ) = 0 for all j ∈ and x (x ) = 1. ∗∗ ∗ N Then, from Odell-Rosenthal Theorem [OR] and the fact that x (ej ) = 0, j ∈ , ξ ∗ ∗∗ we may find a seminormalized block sequence (xj )j in JTp with w -limj xj = x and, passing to a subsequence, we may assume that it also satisfies items (i) and ∗∗ ∗ (ii) of Proposition 5.7 for some ε> 0. Since x (x ) = 1, there exists n0 ∈ N such ∗ 1 − 1 that x (xn) ≥ 1/2 for all n ≥ n0. Then, for k ∈ N such that (2 q + ε)k q < 1/2, Proposition 5.7 yields that

1 1 xn0+1 + ... + xn0+k − x∗ ≤ (2 q + ε)k q  k  which is a contradiction. ξ Remark 5.12. Note that by replacing the norming set G1 with ′ ξ ∗ r Gr = biei : S is a segment of Tξ and |bi| ≤ 1 n Xi∈S Xi∈S o NON-ASYMPTOTIC ℓ1 SPACES WITH UNIQUE ℓ1 ASYMPTOTIC MODEL 21

−1 ′−1 ξ where r + r = 1 and 1

Acknowledgement The authors would like to thank the anonymous referees whose useful remarks helped us identify and correct a mistake in the original proof of Lemma 4.6.

References

[ABM] S.A. Argyros, K. Beanland, and P. Motakis, Strictly singular operators in Tsirelson like spaces, Illinois J. Math. 57 (2013), no. 4, 1173-1217. [AGLM] S. A. Argyros, A. Georgiou, A.-R. Lagos, and P. Motakis, Joint spreading models and uniform approximation of bounded operators arXiv:1712.07638 (2017). [AKT] S. A. Argyros, V. Kanellopoulos and K. Tyros, Finite Order Spreading Models, Adv. Math. 234 (2013), 574-617. [AMT] S. A. Argyros, S. Mercourakis and A. Tsarpalias, Convex unconditionality and summability of weakly null sequences, Israel J. Math. 107, no. 1 (1998), 157-193. [AM3] S. A. Argyros and P. Motakis, On the complete separation of asymptotic structures in Banach spaces, Adv. Math. (to appear). [AT] S. A. Argyros and A. Tolias, Methods in the theory of hereditarily indecomposable Banach spaces, Memoirs of the American Mathematical Society 170 (2004), vi+114. [BLMS] F. Baudier, G. Lancien, P. Motakis, and Th. Schlumprecht, A new coarsely rigid class of Banach spaces, arXiv:1806.00702v2 (2018). [CS] P. G. Casazza and T. J. Shura, Tsirelson’s space, Lecture Notes in Mathematics, 1363. Springer-Verlag, Berlin, 1989. [FJ] T. Figiel and W. B. Johnson, A uniformly convex Banach space which contains no ℓp, Com- positio Math. 29 (1974), 179-190. [FOSZ] D. Freeman, E. Odell, B. Sari and B. Zheng, On spreading sequences and asymptotic structures, Trans. Amer. Math. Soc. 370 (2018), no. 10, 6933-6953. [HO] L. Halbeisen and E. Odell, On asymptotic models in Banach spaces, Israel J. Math. 139 (2004), 253-291. [J1] R. C. James, Bases and reflexivity of Banach spaces, Ann. of Math. (2) 52 (1950), 518-527. [J2] R. C. James, Uniformly non-square Banach spaces, Annals of Mathematics (1964), Vol. 80, No. 3, 542–550. [JKO] M. Junge, D. Kutzarova, and E. Odell, On asymptotically symmetric Banach spaces, Studia Math. 173 (2006), no. 3, 203-231. [KM] D. Kutzarova and P. Motakis, Asymptotically symmetric spaces with hereditarily non-unique spreading models, arXiv:1902.10098 (2019). [LT] J. Lindenstrauss and L. Tzafriri, Classical Banach spaces I: sequence spaces, Springer Verlag (1977). [MMT] B. Maurey, V. D. Milman, and N. Tomczak-Jaegermann, Asymptotic infinite-dimensional theory of Banach spaces, Geometric aspects of (Israel, 1992-1994), 149-175, Oper. Theory Adv. Appl., 77, Birkh¨auser, Basel, 1995. [MR] B. Maurey and H. P. Rosenthal, Normalized weakly null sequence with no unconditional subsequence, Studia Math. 61 (1977), 77–98. [MT] V. D. Milman and N. Tomczak-Jaegermann, Asymptotic ℓp spaces and bounded distortion, Banach spaces (M´erida, 1992) Contemp. Math., vol. 144, Amer. Math. Soc., Providence, RI, 1993, pp. 173-195. [O1] E. Odell, Stability in Banach spaces, Extracta Math. 17 (2002), no. 3, 385-425. [O2] E. Odell, On the structure of separable infinite dimensional Banach spaces, Chern institute of mathematics, Nankai university, Tianjin, China, July 2007. [OR] E. Odell and H. P. Rosenthal, A double-dual characterization of separable Banach spaces containing ℓ1, Israel Journal of Mathematics. Vol. 20, 1975 [OS] E. Odell and Th. Schlumprecht, Trees and branches in Banach spaces, Trans. Amer. Math. Soc. 354 (2002), 4085-4108. 22 S.A.ARGYROS,A.GEORGIOU,ANDP.MOTAKIS

[Ra] F. P. Ramsey, On a problem of formal logic, Proc. London Math. Soc. 30, (1929), 264–286. [S] W. Szlenk, The non existence of a separable reflexive Banach space universal for all separable reflexive Banach spaces, Studia Math., 30 (1968), 53-61. [T] B. S. Tsirelson, Not every Banach space contains ℓp or c0, Functional Anal. Appl. 8 (1974), 138-141.

National Technical University of Athens, Faculty of Applied Sciences, Department of Mathematics, Zografou Campus, 157 80, Athens, Greece E-mail address: [email protected]

National Technical University of Athens, Faculty of Applied Sciences, Department of Mathematics, Zografou Campus, 157 80, Athens, Greece E-mail address: [email protected]

Department of Mathematics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, U.S.A. E-mail address: [email protected]