This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Shrinking Schauder Frames and their Associated Bases

Kevin Beanland Department of Mathematics, Washington & Lee University, 204 W. Washington St. Lexington, VA, 24450 beanlandk@wlu.edu kbeanland.academic.wlu.edu  and  Daniel Freeman Department of Mathematics and Statistics, St Louis University, St Louis, MO daniel.freeman@slu.edu mathstat.slu.edu/ freeman/
(Date: August 6, 2025)
Abstract.

For a Banach space XX with a shrinking Schauder frame (xi,fi)(x_{i},f_{i}) we provide an explicit method for constructing a shrinking associated basis. In the case that the minimal associated basis is not shrinking, we prove that every shrinking associated basis of (xi,fi)(x_{i},f_{i}) dominates an uncountable family of incomparable shrinking associated bases of (xi,fi)(x_{i},f_{i}). By adapting a construction of Pełczyński, we characterize spaces with shrinking Schauder frames as space having the ww^{*}-bounded approximation property.

Key words and phrases:
frames, Schauder frames, shrinking bases, bounded approximation property
2020 Mathematics Subject Classification:
46B15, 42C15, 46B10
The first author was supported by a Lenfest Summer Research Grant through Washington & Lee University. The second author was supported by grant 706481 from the Simons Foundation.

1. Introduction

A frame for a separable infinite dimensional Hilbert space HH is a sequence of vectors (xj)j=1(x_{j})_{j=1}^{\infty} in HH such that there exists constants 0<AB0<A\leq B so that Ax2|x,xj|2Bx2A\|x\|^{2}\leq\sum|\langle x,x_{j}\rangle|^{2}\leq B\|x\|^{2} for all xHx\in H. If (xj)j=1(x_{j})_{j=1}^{\infty} is a frame of HH then there exists a possibly different frame (fj)j=1(f_{j})_{j=1}^{\infty} of HH called a dual frame such that

(1) x=j=1fj,xxj for all xH.x=\sum_{j=1}^{\infty}\langle f_{j},x\rangle x_{j}\hskip 28.45274pt\textrm{ for all }x\in H.

That is, frames can be used like a basis to give a linear reconstruction formula for vectors in HH. The difference between frames and bases is that a frame allows for redundancy in that the coefficients given for reconstruction in (1) are not required to be unique.

Frames have been generalized to Banach spaces in various ways such as atomic decompositions [7, 9], framings [6], and Schauder frames [3, 5]. We will focus on Schauder frames in this paper which are a direct generalization of the reconstruction formula given in (1). Let XX be a separable infinite dimensional Banach space. A sequence of pairs (xj,fj)j=1(x_{j},f_{j})_{j=1}^{\infty} in X×XX\times X^{*} is called a Schauder frame for XX if

(2) x=j=1fj(x)xj for all xX.x=\sum_{j=1}^{\infty}f_{j}(x)x_{j}\hskip 28.45274pt\textrm{ for all }x\in X.

We make the convention that xj0x_{j}\not=0 for jj\in\mathbb{N}. Though Schauder frames were not explicitly defined until 2008, the first appearance of a Schauder frame, without the name Schauder frame, is in A. Pełczyński’s proof [16] that every space with the bounded approximation property (BAP) is isomorphic to complemented subspace of a space with a basis111In the same year (1971) Johnson, Rosenthal, and Zippin [10] proved the same result with a completely different method.. In 1987, S. Szarek showed that spaces with the BAP but without bases exist [17]. Indeed, Pełczyński showed that a separable Banach space XX has the BAP if and only if it has a Schauder frame (in the above sense) and that, furthermore, there is a space ZZ with a basis so that the identity on XX factors through the identity on ZZ in a natural way. Formally, if (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is a Schauder frame for XX and ZZ is a Banach space with a Schauder basis (zi)i=1(z_{i})_{i=1}^{\infty} then ZZ is an associated space for (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} (and (zi)i=1(z_{i})_{i=1}^{\infty} is an associated basis for the frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty}) if the maps T:XZT:X\to Z (analysis operator) and S:ZXS:Z\to X (synthesis operator) defined by T(i=1fi(x)xi)=i=1fi(x)ziT(\sum_{i=1}^{\infty}f_{i}(x)x_{i})=\sum_{i=1}^{\infty}f_{i}(x)z_{i} and S(i=1aizi)=i=1aixiS(\sum_{i=1}^{\infty}a_{i}z_{i})=\sum_{i=1}^{\infty}a_{i}x_{i} are bounded.

In the case that both (xi)i=1(x_{i})_{i=1}^{\infty} and (fi)i=1(f_{i})_{i=1}^{\infty} are frames for a Hilbert space HH, then the associated space ZZ can be chosen to be 2\ell_{2} and the associated basis (zi)i=1(z_{i})_{i=1}^{\infty} can be chosen to be the unit vector basis for 2\ell_{2}. This is of fundamental importance in frame theory as well as in applications such as signal processing. Indeed, given some vector xHx\in H, the analysis operator maps xx to T(x)=(fi,x)i=12T(x)=(\langle f_{i},x\rangle)_{i=1}^{\infty}\in\ell_{2}. One can then apply filters to the sequence of frame coefficients (fi,x)i=1(\langle f_{i},x\rangle)_{i=1}^{\infty} to obtain a sequence (bi)i=12(b_{i})_{i=1}^{\infty}\in\ell_{2}. Applying the synthesis operator then gives a vector S((bi)i=1)=i=1bixiS((b_{i})_{i=1}^{\infty})=\sum_{i=1}^{\infty}b_{i}x_{i} which is an approximation of xx but is improved in some way such as being compressed or having noise or artifacts removed.

If one wishes to use similar techniques for a Schauder frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} then it is advantageous to construct the associated basis (zi)i=1(z_{i})_{i=1}^{\infty} to be as nice as possible. That is, if (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} has some desirable property such as being unconditional, shrinking, or boundedly complete then one would like (zi)i=1(z_{i})_{i=1}^{\infty} to share the property as well. If (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is unconditional, then it is straightforward to construct an unconditional associated basis. In [1], the authors of the current paper and R. Liu prove that if (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is shrinking then it has a shrinking associated basis. However, the construction in [1] is relatively difficult and involves the method of bounds on branches of weakly null trees developed by E. Odell and Th. Schlumprecht [15] [8]. One of the main goals of this paper is to give a more direct and much simpler construction of a shrinking associated basis, which we state in the following theorem.

Theorem 1.1.

Let (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} be a shrinking Schauder frame for a Banach space XX. For each mnm\leqslant n, we denote P[m,n]:XXP_{[m,n]}:X\rightarrow X to be the operator P[m,n](x)=i=mnfi(x)xiP_{[m,n]}(x)=\sum_{i=m}^{n}f_{i}(x)x_{i}. Then there exists an increasing sequence of natural numbers (Nk)k=1(N_{k})_{k=1}^{\infty} such that

(3) supm0<n0kNkmnP[m0,n0]P[m,n]x2kx for all xX.\sup_{\begin{subarray}{c}m_{0}<n_{0}\leqslant k\\ N_{k}\leqslant m\leqslant n\end{subarray}}\|P_{[m_{0},n_{0}]}P_{[m,n]}x\|\leqslant 2^{-k}\|x\|\hskip 28.45274pt\textrm{ for all }x\in X.

Furthermore, if (Nk)k=1(N_{k})_{k=1}^{\infty} satisfies (3) then Z(Nk)Z_{(N_{k})} is an associated space of (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} and (zi)i=1(z_{i})_{i=1}^{\infty} is a shrinking associated basis where for aiziZ(Nk)\sum a_{i}z_{i}\in Z_{(N_{k})}, the norm is given by

(4) aizi(Nk)=supmnminaixisupm0n0kNkmn2kP[m0,n0]minaixi.\Big{\|}\sum a_{i}z_{i}\Big{\|}_{(N_{k})}=\sup_{m\leqslant n}\Big{\|}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}\vee\sup_{\begin{subarray}{c}m_{0}\leqslant n_{0}\leq k\\ N_{k}\leqslant m\leqslant n\end{subarray}}2^{k}\Big{\|}P_{[m_{0},n_{0}]}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}.

Given a Schuader frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} with xi0x_{i}\neq 0 for all ii\in\mathbb{N}, the most natural associated space is now referred to as the minimal associated space [5][12] and is defined as follows. Denote by (zi)i=1(z_{i})_{i=1}^{\infty} the unit vector basis for c00c_{00} and for (ai)c00(a_{i})\in c_{00} consider the norm

(5) aizimin=supm<nminaixi.\Big{\|}\sum a_{i}z_{i}\Big{\|}_{\min}=\sup_{m<n}\Big{\|}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}.

The minimal associated space ZminZ_{min} is defined to be the completion of c00c_{00} under the above norm and the basis (zi)i=1(z_{i})_{i=1}^{\infty} is called the minimal associated basis. A Schauder frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} may have many non-equivalent associated bases. However, the basis (zi)i=1(z_{i})_{i=1}^{\infty} defined in (5) is minimal in the sense that if (yi)i=1(y_{i})_{i=1}^{\infty} is any associated basis for (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} then (yi)i=1(y_{i})_{i=1}^{\infty} dominates (zi)i=1(z_{i})_{i=1}^{\infty}. That is, there exists a constant K>0K>0 so that iaiziKiaiyi\|\sum_{i}a_{i}z_{i}\|\leq K\|\sum_{i}a_{i}y_{i}\| for all (ai)c00(a_{i})\in c_{00} [12].

We now consider the problem of determining if a shrinking Schauder frame has a minimal shrinking associated basis. That is, if (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is a shrinking Schauder frame, then when does there exist a shrinking associated basis (wi)i=1(w_{i})_{i=1}^{\infty} of (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} such that if (yi)i=1(y_{i})_{i=1}^{\infty} is any shrinking associated basis of (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} then (yi)i=1(y_{i})_{i=1}^{\infty} dominates (wi)i=1(w_{i})_{i=1}^{\infty}? In Section 4 we prove that a Schauder frame has a minimal shrinking associated basis if and only if the minimal associated basis defined in (5) is shrinking. Our construction of a shrinking associated basis in Theorem 1.1 is defined solely in terms of the shrinking Schauder frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} and some sequence of natural numbers (Nk)k=1[]ω(N_{k})_{k=1}^{\infty}\in[\mathbb{N}]^{\omega}. In Section 4 we prove that if (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is a shrinking Schauder frame and (yi)i=1(y_{i})_{i=1}^{\infty} is any shrinking associated basis then there exists (Nk)k=1[]ω(N_{k})_{k=1}^{\infty}\in[\mathbb{N}]^{\omega} such that the resulting shrinking associated basis (zi)i=1(z_{i})_{i=1}^{\infty} from our construction is dominated by (yi)i=1(y_{i})_{i=1}^{\infty}. In other words, Theorem 1.1 produces a set of shrinking associated bases such that every shrinking associated basis of (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} dominates some basis in that set. Furthermore, we prove that if the minimal associated basis is not shrinking then for every shrinking associated basis (yi)i=1(y_{i})_{i=1}^{\infty} there exists uncountably many mutually incomparable shrinking associated bases which are all dominated by (yi)i=1(y_{i})_{i=1}^{\infty}. Hence, except for the trivial case where the minimal associated basis is shrinking, we have that the collection of shrinking associated bases will have a very rich lattice structure under the domination partial order.

In the final section we make some observations about how this work relates to now classical results about the BAP and give an alternative proof of the theorem of Johnson, Rosenthal, and Zippin that for a Banach space XX with separable dual, XX^{*} has the BAP if and only if XX is isomorphic to a complemented subspace of a Banach space with a shrinking basis.

2. Shrinking Schauder bases and shrinking Schauder frames

A sequence of vectors (xi)i=1(x_{i})_{i=1}^{\infty} in a separable Banach space XX is called a Schauder basis if for all xXx\in X there exists a unique sequence of scalars (ai)i=1(a_{i})_{i=1}^{\infty} such that x=i=1aixix=\sum_{i=1}^{\infty}a_{i}x_{i}. If XX is a Banach space with dual XX^{*} then a Schauder basis (xi)i=1(x_{i})_{i=1}^{\infty} is called shrinking if the biorthogonal functionals (xi)i=1(x_{i}^{*})_{i=1}^{\infty} form a Schauder basis for XX^{*}. In particular, a Banach space with a shrinking basis necessarily has a separable dual with a basis. There are, however, Banach spaces with bases whose duals are separable but fail the approximation property [11, Theorem 1.e.7.(b)]. Naturally, a Schauder frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} for XX is shrinking if and only if (fi,xi)i=1(f_{i},x_{i})_{i=1}^{\infty} is a Schauder frame for XX^{*}. Using the terminology atomic decomposition instead of Schauder frame, Carando and Lassalle [2] give the following useful characterization of shrinking Schauder frames which is analogous to James’ well-known characterization for Schauder bases.

Theorem 2.1 ([2], Theorem 1.4).

Let (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} be a Schauder frame for XX. For each interval II\subseteq\mathbb{N}, let PI:XXP_{I}:X\rightarrow X be the operator PI(x)=iIfi(x)xiP_{I}(x)=\sum_{i\in I}f_{i}(x)x_{i}. Then (fi,xi)i=1(f_{i},x_{i})_{i=1}^{\infty} is a Schauder frame for XX^{*} if and only if for each fXf\in X^{*} we have that

(6) limnfP[n,)=0.\lim_{n\to\infty}\|f\circ P_{[n,\infty)}\|=0.

We sketch a short proof for completeness.

Proof.

Consider the reverse direction and assume that (6) holds. Let fXf\in X^{*}. It suffices to show that f(xi)fi\sum f(x_{i})f_{i} is a Cauchy sequence. This follows readily from (6) as

lim supm,ni=mnf(xi)fi\displaystyle\limsup_{m,n\to\infty}\Big{\|}\sum_{i=m}^{n}f(x_{i})f_{i}\Big{\|} =lim supm,nsupxSXi=mnf(xi)fi(x)\displaystyle=\limsup_{m,n\to\infty}\sup_{x\in S_{X}}\sum_{i=m}^{n}f(x_{i})f_{i}(x)
=lim supm,nsupxSXf(i=mnfi(x)xi)\displaystyle=\limsup_{m,n\to\infty}\sup_{x\in S_{X}}f\Big{(}\sum_{i=m}^{n}f_{i}(x)x_{i}\Big{)}
=lim supm,nfP[m,n]=0.\displaystyle=\limsup_{m,n\to\infty}\|f\circ P_{[m,n]}\|=0.

Therefore (fi,xi)i=1(f_{i},x_{i})_{i=1}^{\infty} is a Schauder frame for XX^{*}. A similar proof shows that the converse holds. ∎

Let XX be a Banach space with a Schauder frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} and let ZZ be a Banach space with a Schauder basis (zi)i=1(z_{i})_{i=1}^{\infty}. Recall that ZZ is said to be an associated space of the Schauder frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} and (zi)i=1(z_{i})_{i=1}^{\infty} is said to an associated basis if the maps T:XZT:X\rightarrow Z and S:ZXS:Z\rightarrow X are bounded where T(x)=i=1fi(x)xiT(x)=\sum_{i=1}^{\infty}f_{i}(x)x_{i} for all xXx\in X and S(i=1aizi)=i=1aixiS(\sum_{i=1}^{\infty}a_{i}z_{i})=\sum_{i=1}^{\infty}a_{i}x_{i} for all i=1aiziZ\sum_{i=1}^{\infty}a_{i}z_{i}\in Z. It follows immediately that if a Schauder frame has a shrinking associated basis then the Schauder frame must be shrinking. In [1], the authors of the current paper and R. Liu prove a more general and technical theorem which implies that every shrinking Schauder frame has a shrinking associated basis. Unfortunately, the argument does not provide an explicit construction of the associated basis and the proofs are relatively difficult. In Section 3 we give an explicit method which will give a shrinking associated basis for any shrinking Schauder frame. Before proceeding we show that the minimal associated basis for a shrinking Schauder frame need not be shrinking.

Example 1.

Let (ei)(e_{i}) be the unit vector basis for 2\ell_{2}. Let x1=e1x_{1}=e_{1} and f1=e1f_{1}=e_{1}^{*}. For all ii\in\mathbb{N} we let x2i+1=e1x_{2i+1}=e_{1}, x2i=ei+1x_{2i}=e_{i+1}, f2i+1=0f_{2i+1}=0, and f2i=ei+1f_{2i}=e^{*}_{i+1}. Then (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is a shrinking Schauder frame for 2\ell_{2} but the minimal associated space for (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} has the norm

(7) aizimin2=sup{|iIi odd ai|2+iIi even |ai|2:I, is an interval}.\big{\|}\sum a_{i}z_{i}\big{\|}_{\min}^{2}=\sup\bigg{\{}\big{|}\sum_{\begin{subarray}{c}i\in I\\ i\text{ odd }\end{subarray}}a_{i}\big{|}^{2}+\sum_{\begin{subarray}{c}i\in I\\ i\text{ even }\end{subarray}}|a_{i}|^{2}:I\subset\mathbb{N},\text{ is an interval}\bigg{\}}.

The basis (zi)(z_{i}) is not shrinking since the sequence of odd vectors (z2i1)i(z_{2i-1})_{i\in\mathbb{N}} is equivalent to the summing basis of c0c_{0}.

The above example, although simple, is rather instructive in that it reveals that redundancy in a frame can make the minimal associated basis not shrinking.

A sequence (yi)(y_{i}) in a Banach space is called α\alpha-1+\ell_{1}^{+} for some α>0\alpha>0 if aiyiαai\|\sum a_{i}y_{i}\|\geqslant\alpha\sum a_{i} whenever (ai)(a_{i}) is a summable sequence of non-negative scalars. We conclude this section by recalling the following useful and well known characterization of shrinking bases.

Lemma 2.2.

Let XX be a Banach spaces with a Schauder basis (xi)(x_{i}). The following are equivalent:

  1. 1.

    (xi)(x_{i}) is not shrinking.

  2. 2.

    There is a normalized block sequence (yi)(y_{i}) of (xi)(x_{i}) that is not weakly null.

  3. 3.

    There is a normalized block sequence (yi)(y_{i}) of (xi)(x_{i}) that is α\alpha-1+\ell_{1}^{+} for some α>0\alpha>0.

3. Constructing a shrinking associated basis

We start by setting some notation. Let (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} be a shrinking Schauder frame for a Banach space XX. Let ZZ be an associated space and (zi)i=1(z_{i})_{i=1}^{\infty} be an associated basis. The analysis operator is the map T:XZT:X\rightarrow Z given by T(x)=i=1fi(x)ziT(x)=\sum_{i=1}^{\infty}f_{i}(x)z_{i} for all xXx\in X. The synthesis operator is the map S:ZXS:Z\rightarrow X given by S(i=1aizi)=i=1aixiS(\sum_{i=1}^{\infty}a_{i}z_{i})=\sum_{i=1}^{\infty}a_{i}x_{i} for all i=1aiziZ\sum_{i=1}^{\infty}a_{i}z_{i}\in Z. For m<nm<n, we use the following notation when we wish to use partial sums.

  1. 1.

    S[m,n](iaizi)=i=mnaixiS_{[m,n]}(\sum_{i}a_{i}z_{i})=\sum_{i=m}^{n}a_{i}x_{i} for all aiziZ\sum a_{i}z_{i}\in Z,

  2. 2.

    R[m,n](iaizi)=i=mnaiziR_{[m,n]}(\sum_{i}a_{i}z_{i})=\sum_{i=m}^{n}a_{i}z_{i} for all aiziZ\sum a_{i}z_{i}\in Z,

  3. 3.

    P[m,n](x)=i=mnfi(x)xiP_{[m,n]}(x)=\sum_{i=m}^{n}f_{i}(x)x_{i}.

It follows from the uniform boundedness principle that supmnS[m,n]\sup_{m\leq n}\|S_{[m,n]}\|, supmnR[m,n]\sup_{m\leq n}\|R_{[m,n]}\|, and supmnP[m,n]\sup_{m\leq n}\|P_{[m,n]}\| are all finite. The value supmnR[m,n]\sup_{m\leq n}\|R_{[m,n]}\| is the basis constant of (zi)i=1(z_{i})_{i=1}^{\infty} and the value supmnP[m,n]\sup_{m\leq n}\|P_{[m,n]}\| is called the frame constant of (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty}.

The following proposition is contained in [1] and we include the short proof for completeness.

Proposition 3.1.

Let (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} be a shrinking Schauder frame for a Banach space XX. Then there is an increasing sequence (Nk)k=1(N_{k})_{k=1}^{\infty} of natural numbers so that

(8) supm0<n0kNk<m<nP[m0,n0]P[m,n]x2kx for all xX.\sup_{\begin{subarray}{c}m_{0}<n_{0}\leqslant k\\ N_{k}<m<n\end{subarray}}\|P_{[m_{0},n_{0}]}P_{[m,n]}x\|\leqslant 2^{-k}\|x\|\hskip 28.45274pt\textrm{ for all }x\in X.
Proof.

Let kk\in\mathbb{N} and ε>0\varepsilon>0. It suffices to show that there is an Nk>kN_{k}>k satisfying

supm0n0kNkmnP[m0,n0]P[m,n]x<ε for all xX with x=1.\sup_{\begin{subarray}{c}m_{0}\leqslant n_{0}\leqslant k\\ N_{k}\leqslant m\leqslant n\end{subarray}}\|P_{[m_{0},n_{0}]}P_{[m,n]}x\|<\varepsilon\hskip 28.45274pt\textrm{ for all $x\in X$ with $\|x\|=1$.}

As (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} is shrinking we have that (fi,xi)i=1(f_{i},x_{i})_{i=1}^{\infty} is a Schauder frame for XX^{*}. Thus, for sufficiently large NkN_{k} we have that

supNkmni=mnfi(xj)fi<εkxjfor all 1jk.\sup_{N_{k}\leqslant m\leqslant n}\|\sum_{i=m}^{n}f_{i}(x_{j})f_{i}\|<\frac{\varepsilon}{k\|x_{j}\|}\hskip 28.45274pt\textrm{for all $1\leqslant j\leqslant k$.}

This NkN_{k} suffices as for fixed m0n0km_{0}\leqslant n_{0}\leqslant k, NkmnN_{k}\leqslant m\leqslant n, and xXx\in X with x=1\|x\|=1 we have that

P[m0,n0]P[m,n]x=j=m0n0fj(i=mnfi(x)xi)xjksup1jkfj(i=mnfi(x)xi)xjksup1jki=mnfi(xj)fixxj<ε\begin{split}\|P_{[m_{0},n_{0}]}P_{[m,n]}x\|&=\Big{\|}\sum_{j=m_{0}}^{n_{0}}f_{j}(\sum_{i=m}^{n}f_{i}(x)x_{i})x_{j}\Big{\|}\\ &\leqslant k\sup_{1\leqslant j\leqslant k}\Big{\|}f_{j}(\sum_{i=m}^{n}f_{i}(x)x_{i})x_{j}\Big{\|}\\ &\leqslant k\sup_{1\leqslant j\leqslant k}\Big{\|}\sum_{i=m}^{n}f_{i}(x_{j})f_{i}\Big{\|}\|x\|\|x_{j}\|<\varepsilon\end{split}

The claim follows. ∎

Remark 3.2.

Note that Proposition 3.1 still holds if we replace (2k)k=1(2^{-k})_{k=1}^{\infty} with any positive sequence which converges to 0. Moreover, if (Nk)k=1(N_{k})_{k=1}^{\infty} satisfies (8) and (Mk)k=1(M_{k})_{k=1}^{\infty} is any increasing sequence of natural numbers with NkMkN_{k}\leqslant M_{k} for all kk\in\mathbb{N} then (Mk)k=1(M_{k})_{k=1}^{\infty} also satisfies (8).

We now define a norm on c00c_{00} which we will later prove gives a shrinking associated basis.

Definition 3.3.

Let (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} be a shrinking Schauder frame and let (Nk)k=1(N_{k})_{k=1}^{\infty} satisfy the hypothesis of Proposition 3.1. We let (zi)i=1(z_{i})_{i=1}^{\infty} denote the unit vector basis of c00c_{00} and consider the following norm for aizic00\sum a_{i}z_{i}\in c_{00}.

(9) aizi(Nk)=supmnminaixisupm0n0kNkmn2kP[m0,n0]minaixi.\Big{\|}\sum a_{i}z_{i}\Big{\|}_{(N_{k})}=\sup_{m\leqslant n}\Big{\|}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}\vee\sup_{\begin{subarray}{c}m_{0}\leqslant n_{0}\leq k\\ N_{k}\leqslant m\leqslant n\end{subarray}}2^{k}\Big{\|}P_{[m_{0},n_{0}]}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}.

Let Z(Nk)Z_{(N_{k})} denote the completion of c00c_{00} under this norm. For zZ(Nk)z\in Z_{(N_{k})} and kk^{\prime}\in\mathbb{N} it will be convenient to denote the second part of (9) as

(10) zk:=supm0n0kNkmn2kP[m0,n0]S[m,n]z.\|z\|_{k^{\prime}}:=\sup_{\begin{subarray}{c}m_{0}\leqslant n_{0}\leq k^{\prime}\\ N_{k^{\prime}}\leqslant m\leqslant n\end{subarray}}2^{k^{\prime}}\|P_{[m_{0},n_{0}]}S_{[m,n]}z\|.

Proposition 3.1 gives a condition satisfied by each shrinking Schauder frame. The idea behind the definition of the norm above is to force the associated space to satisfy some version of this condition. The goal then is to show that satisfying this condition is sufficient to establish that the associated basis is shrinking.

Remark 3.4.

A slight weakening of the norm (Nk)\|\cdot\|_{(N_{k})} was introduced in [1] where the authors prove that the basis (zi)(z_{i}) is strongly shrinking relative to (xi,fi)(x_{i},f_{i}), which is a weaker condition than shrinking.

Theorem 3.5.

Let (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} be a shrinking frame for a Banach space XX and let (Nk)k=1(N_{k})_{k=1}^{\infty} satisfy Proposition 3.1. Then Z(Nk)Z_{(N_{k})} is an associated space for (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} and (zi)i=1(z_{i})_{i=1}^{\infty} is a shrinking basis for the space Z(Nk)Z_{(N_{k})}.

Proof.

Assuming (Nk)k=1(N_{k})_{k=1}^{\infty} satisfies Proposition 3.1, we will first show that Z(Nk)Z_{(N_{k})} is as an associated space to the frame (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty}. This is the only place in the proof we use Proposition 3.1. Let us see that the analysis operator T:XZ(Nk)T:X\to Z_{(N_{k})} satisfies TC\|T\|\leq C where C:=supmnP[m,n]C:=\sup_{m\leqslant n}\|P_{[m,n]}\| is the frame constant of (xi,fi)(x_{i},f_{i}). Let xXx\in X. Then, the first part of Tx(Nk)=fi(x)zi(Nk)\|Tx\|_{(N_{k})}=\|\sum f_{i}(x)z_{i}\|_{(N_{k})} in (9) satisfies

supmni=mnfi(x)xi=supmnP[m,n]xCx.\sup_{m\leq n}\Big{\|}\sum_{i=m}^{n}f_{i}(x)x_{i}\Big{\|}=\sup_{m\leq n}\|P_{[m,n]}x\|\leq C\|x\|.

We now fix kk\in\mathbb{N}, m0n0km_{0}\leqslant n_{0}\leqslant k, and NkmnN_{k}\leqslant m\leqslant n. Then by Proposition 3.1 we have that the second part of (9) satisfies

2kP[m0,n0]S[m,n]fi(x)zi=2kP[m0,n0]P[m,n]fi(x)xix.2^{k}\Big{\|}P_{[m_{0},n_{0}]}S_{[m,n]}\sum f_{i}(x)z_{i}\Big{\|}=2^{k}\Big{\|}P_{[m_{0},n_{0}]}P_{[m,n]}\sum f_{i}(x)x_{i}\Big{\|}\leqslant\|x\|.

Thus, we have that Tx(Nk)Cx\|Tx\|_{(N_{k})}\leq C\|x\| and hence TC\|T\|\leq C. The synthesis operator S:Z(Nk)XS:Z_{(N_{k})}\to X is bounded, since it is bounded on S:ZminXS:Z_{\min}\to X and zminz(Nk)\|z\|_{\min}\leqslant\|z\|_{(N_{k})}. Thus, Z(Nk)Z_{(N_{k})} is an associated space to (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty}.

Let (yi)(y_{i}) be a normalized block sequence of (zi)(z_{i}) in Z(Nk)Z_{(N_{k})}. In order to show that (zi)(z_{i}) is shrinking it suffices to show by Lemma 2.2 that there is a subsequence of (yi)(y_{i}) which is weakly null. We claim that we may pass to a subsequence of (yi)(y_{i}) and find an increasing sequence (ki)(k_{i}) in \mathbb{N} so that for all ii\in\mathbb{N} we have that

  • (i)

    yispanNkijki+1(zj)y_{i}\in\operatorname{span}_{N_{k_{i}}\leqslant j\leqslant k_{i+1}}(z_{j}),

  • (ii)

    P[m,n]Syi2ki\displaystyle\|P_{[m,n]}Sy_{i}\|\leqslant 2^{-k_{i}} for all m,nm,n\in\mathbb{N} with ki+1mnk_{i+1}\leqslant m\leqslant n.

Indeed, (i) is easily obtained as (yi)(y_{i}) is a block sequence of (zi)(z_{i}). We may obtain (ii) by choosing ki+1k_{i+1} sufficiently large as jfj(Syi)xj\sum_{j\in\mathbb{N}}f_{j}(Sy_{i})x_{j} is convergent for all ii\in\mathbb{N}. The following additional properties are implied by (i) and (ii).

  1. (iii)

    P[m0,n0]S[m,n]yi2ki\|P_{[m_{0},n_{0}]}S_{[m,n]}y_{i}\|\leqslant 2^{-k_{i}} for m0n0kim_{0}\leqslant n_{0}\leqslant k_{i} and mnm\leqslant n,

  2. (iv)

    For each ii\in\mathbb{N}, P[ki,ki+1)SyiSyi2ki+1\|P_{[k_{i},k_{i+1})}Sy_{i}\|\geqslant\|Sy_{i}\|-2^{-k_{i}+1},

  3. (v)

    For iji\not=j in \mathbb{N}, P[kj,kj+1)Syi2ki\|P_{[k_{j},k_{j+1})}Sy_{i}\|\leqslant 2^{-k_{i}}.

Item (iii) follows from (i) and the fact that yikiyi(Nk)=1\|y_{i}\|_{k_{i}}\leqslant\|y_{i}\|_{(N_{k})}=1. Item (v) follows from (iii) if j<ij<i and follows from (ii) if j>ij>i. Item (iv) is a consequence of (ii) and (iii) as

P[ki,ki+1)SyiSyiP[1,ki)SyiP[ki+1,)SyiSyi2ki+1.\|P_{[k_{i},k_{i+1})}Sy_{i}\|\geqslant\|Sy_{i}\|-\|P_{[1,k_{i})}Sy_{i}\|-\|P_{[k_{i+1},\infty)}Sy_{i}\|\geqslant\|Sy_{i}\|-2^{-k_{i}+1}.

Before dividing the proof into two cases, we fix (ai)c00(a_{i})\in c_{00} and kk\in\mathbb{N} and will show that

(11) aiyik=supm0<n0kNkm<n2kP[m0,n0]S[m,n]aiyi2sup|ai|.\big{\|}\sum a_{i}y_{i}\big{\|}_{k}=\sup_{\begin{subarray}{c}m_{0}<n_{0}\leq k\\ N_{k}\leqslant m<n\end{subarray}}2^{k}\big{\|}P_{[m_{0},n_{0}]}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\leqslant 2\sup|a_{i}|.

Let m0n0km_{0}\leqslant n_{0}\leqslant k and NkmnN_{k}\leqslant m\leqslant n. Let i0i_{0} be the least integer such that mki0+1m\leqslant k_{i_{0}+1}. By (i), we have that S[m,n]yi=0S_{[m,n]}y_{i}=0 for all i<i0i<i_{0}. Since yi0=1\|y_{i_{0}}\|=1, (9) implies that 2kP[m0,n0]S[m,n]ai0yi0|ai0|2^{k}\|P_{[m_{0},n_{0}]}S_{[m,n]}a_{i_{0}}y_{i_{0}}\|\leqslant|a_{i_{0}}|. By (iii),

(12) i=i0+12kP[m0,n0]S[m,n]aiyi2ki=i0+112ki|ai|sup|ai|.\sum_{i=i_{0}+1}^{\infty}2^{k}\|P_{[m_{0},n_{0}]}S_{[m,n]}a_{i}y_{i}\|\leqslant 2^{k}\sum_{i=i_{0}+1}\frac{1}{2^{k_{i}}}|a_{i}|\leqslant\sup|a_{i}|.

Thus, we have that

2kP[m0,n0]S[m,n]aiyi\displaystyle 2^{k}\|P_{[m_{0},n_{0}]}S_{[m,n]}\sum a_{i}y_{i}\| 2kP[m0,n0]S[m,n]ai0yi0+i=i0+12kP[m0,n0]S[m,n]aiyi\displaystyle\leqslant 2^{k}\|P_{[m_{0},n_{0}]}S_{[m,n]}a_{i_{0}}y_{i_{0}}\|+\sum_{i=i_{0}+1}^{\infty}2^{k}\|P_{[m_{0},n_{0}]}S_{[m,n]}a_{i}y_{i}\|
2sup|ai|.\displaystyle\leqslant 2\sup|a_{i}|.

This proves (11). We now pass to a further subsequence of (yi)(y_{i}) such that exactly one of the following holds.

  • (vi a.)

    Syi2i\|Sy_{i}\|\leqslant 2^{-i} for each ii\in\mathbb{N}.

  • (vi b.)

    For some c>2k1+3c>2^{-k_{1}+3}, we have cSyi1c\leqslant\|Sy_{i}\|\leqslant 1.

Assume (vi a.) holds. Fix (ai)c00(a_{i})\in c_{00}. Let mnm\leqslant n and let i0i_{0} be the least integer such that mmaxsuppyi0m\leqslant\max\operatorname{supp}\,y_{i_{0}} and let i1i_{1} be the greatest integer such that nminsuppyi1n\geqslant\min\operatorname{supp}y_{i_{1}}. We have the following.

S[m,n]aiyi\displaystyle\|S_{[m,n]}\sum a_{i}y_{i}\| =S[m,n]i=i0i1aiyi\displaystyle=\|S_{[m,n]}\sum_{i=i_{0}}^{i_{1}}a_{i}y_{i}\|
S[m,n]ai0yi0+i=i0+1i11Saiyi+S[m,n]ai1yi1\displaystyle\leq\|S_{[m,n]}a_{i_{0}}y_{i_{0}}\|+\sum_{i=i_{0}+1}^{i_{1}-1}\|Sa_{i}y_{i}\|+\|S_{[m,n]}a_{i_{1}}y_{i_{1}}\|
|ai0|+i=i0+1i112i|ai|+|ai1| by (vi),\displaystyle\leq|a_{i_{0}}|+\sum_{i=i_{0}+1}^{i_{1}-1}2^{-i}|a_{i}|+|a_{i_{1}}|\hskip 28.45274pt\textrm{ by (vi),}
3supi|ai|\displaystyle\leq 3\sup_{i\in\mathbb{N}}|a_{i}|

Combining this with (11) gives that (yi)(y_{i}) is 33-dominated by the unit vector basis of c0c_{0} and therefore (yi)(y_{i}) is weakly null, which completes the proof for this case.

We now assume item (vi b.) holds. We will prove that in this case that (Syi)(Sy_{i}) and (yi)(y_{i}) are equivalent basic sequences and that (Syi)(Sy_{i}) is weakly null. Let CC be the frame constant of (xi,fi)(x_{i},f_{i}). We first prove that (Syi)(Sy_{i}) is 2C2C-basic.

Let (ai)c00(a_{i})\in c_{00} and let jj\in\mathbb{N} be such that |aj|=sup|ai||a_{j}|=\sup|a_{i}|. We have that

CaiSyi\displaystyle C\big{\|}\sum a_{i}Sy_{i}\big{\|} aiP[kj,kj+1)Syi\displaystyle\geqslant\|\sum a_{i}P_{[k_{j},k_{j+1})}Sy_{i}\|
ajP[kj,kj+1)SyjijaiP[kj,kj+1)Syi\displaystyle\geqslant\big{\|}a_{j}P_{[k_{j},k_{j+1})}Sy_{j}\big{\|}-\sum_{i\not=j}\big{\|}a_{i}P_{[k_{j},k_{j+1})}Sy_{i}\big{\|}
(|aj|Syi|aj|2kj+1)ij|ai|2ki by (iv), and (v)\displaystyle\geqslant(|a_{j}|\|Sy_{i}\|-|a_{j}|2^{-k_{j}+1})-\sum_{i\not=j}|a_{i}|2^{-k_{i}}\hskip 14.22636pt\textrm{ by (iv), and (v)}
|aj|c|aj|2k1+2 by (vi b.)\displaystyle\geqslant|a_{j}|c-|a_{j}|2^{-k_{1}+2}\hskip 28.45274pt\textrm{ by (vi b.) }
c21|aj| as c>2k1+3 by (vi b.).\displaystyle\geqslant c2^{-1}|a_{j}|\hskip 56.9055pt\textrm{ as $c>2^{-k_{1}+3}$ by (vi b.). }

Thus, we have that

(13) CaiSyic21sup|ai|.C\|\sum a_{i}Sy_{i}\|\geqslant c2^{-1}\sup|a_{i}|.

We now fix MM\in\mathbb{N} and consider the following partial sum.

i=1M\displaystyle\Big{\|}\sum_{i=1}^{M} aiSyiP[1,kM+1)i=1MaiSyi+P[kM+1,)i=1MaiSyi\displaystyle a_{i}Sy_{i}\Big{\|}\leqslant\Big{\|}P_{[1,k_{M+1})}\sum_{i=1}^{M}a_{i}Sy_{i}\Big{\|}+\Big{\|}P_{[k_{M+1},\infty)}\sum_{i=1}^{M}a_{i}Sy_{i}\Big{\|}
P[1,kM+1)aiSyi+P[1,kM+1)i=M+1aiSyi+P[kM+1,)i=1MaiSyi\displaystyle\leqslant\big{\|}P_{[1,k_{M+1})}\sum a_{i}Sy_{i}\big{\|}+\Big{\|}P_{[1,k_{M+1})}\sum_{i=M+1}^{\infty}a_{i}Sy_{i}\Big{\|}+\Big{\|}P_{[k_{M+1},\infty)}\sum_{i=1}^{M}a_{i}Sy_{i}\Big{\|}
CaiSyi+i=M+1|ai|2ki+i=1M|ai|2ki by (iii) and (ii)\displaystyle\leqslant C\big{\|}\sum a_{i}Sy_{i}\big{\|}+\sum_{i=M+1}^{\infty}|a_{i}|2^{-k_{i}}+\sum_{i=1}^{M}|a_{i}|2^{-k_{i}}\hskip 14.22636pt\textrm{ by (iii) and (ii)}
CaiSyi+sup|ai|2k1+1\displaystyle\leqslant C\big{\|}\sum a_{i}Sy_{i}\big{\|}+\sup|a_{i}|2^{-k_{1}+1}
CaiSyi+sup|ai|c22 as c>2k1+3 by (vi b.)\displaystyle\leqslant C\big{\|}\sum a_{i}Sy_{i}\big{\|}+\sup|a_{i}|c2^{-2}\hskip 28.45274pt\textrm{ as $c>2^{-k_{1}+3}$ by (vi b.)}
(C+21C)aiSyi2CaiSyi by (13).\displaystyle\leqslant(C+2^{-1}C)\big{\|}\sum a_{i}Sy_{i}\big{\|}\leqslant 2C\big{\|}\sum a_{i}Sy_{i}\big{\|}\hskip 28.45274pt\textrm{ by \eqref{more lower}}.

This proves that (Syi)(Sy_{i}) is 2C2C-basic.

Since SS is a bounded linear operator, (yi)(y_{i}) dominates (Syi)(Sy_{i}). We now prove that (Syi)(Sy_{i}) is equivalent to (yi)(y_{i}) by proving that (Syi)(Sy_{i}) dominates (yi)(y_{i}). Fix (ai)c00(a_{i})\in c_{00}. Let j0j_{0}\in\mathbb{N} with |aj0|=maxi|ai||a_{j_{0}}|=\max_{i}|a_{i}| and let Ij0I_{j_{0}}\subset\mathbb{N} be the smallest interval containing suppyj0\operatorname{supp}y_{j_{0}}. Thus by (vi b.),

(14) supmnS[m,n]aiyiSIj0aiyi=|aj|Syjcsup|ai|.\sup_{m\leq n}\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\geqslant\big{\|}S_{I_{j_{0}}}\sum a_{i}y_{i}\big{\|}=|a_{j}|\|Sy_{j}\|\geqslant c\sup|a_{i}|.

We now have that

aiyi\displaystyle\big{\|}\sum a_{i}y_{i}\big{\|} =supmnS[m,n]aiyisupkaiyik\displaystyle=\sup_{m\leq n}\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\vee\sup_{k}\big{\|}\sum a_{i}y_{i}\big{\|}_{k}
supmnS[m,n]aiyi2sup|ai| by (11),\displaystyle\leqslant\sup_{m\leq n}\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\vee 2\sup|a_{i}|\hskip 28.45274pt\textrm{ by \eqref{upperc0}},
supmnS[m,n]aiyi2csupmnS[m,n]aiyi by (14),\displaystyle\leqslant\sup_{m\leq n}\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\vee\frac{2}{c}\sup_{m\leq n}\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\hskip 28.45274pt\textrm{ by \eqref{at least c0}},
=2csupmnS[m,n]aiyi\displaystyle=\frac{2}{c}\sup_{m\leq n}\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}

Therefore, to prove that (Syi)i=1(Sy_{i})_{i=1}^{\infty} dominates (yi)i=1(y_{i})_{i=1}^{\infty} it will suffice to prove that for fixed m<nm<n we have that

(15) S[m,n]aiyi2C(1+2c1)aiSyi.\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|}\leqslant 2C(1+2c^{-1})\big{\|}\sum a_{i}Sy_{i}\big{\|}.

As, (yi)(y_{i}) is a block sequence of (zi)(z_{i}), there exists j1j2j_{1}\leqslant j_{2} so that

S[m,n]iaiyi=i=j1j2ajSyjS[1,m)aj1yj1S[n,)aj2yj2S_{[m,n]}\sum_{i}a_{i}y_{i}=\sum_{i=j_{1}}^{j_{2}}a_{j}Sy_{j}-S_{[1,m)}a_{j_{1}}y_{j_{1}}-S_{[n,\infty)}a_{j_{2}}y_{j_{2}}

By taking the norm of both sides we now have that

S[m,n]aiyi\displaystyle\big{\|}S_{[m,n]}\sum a_{i}y_{i}\big{\|} i=j1j2ajSyj+S[1,m)aj1yj1+S[n,)aj2yj2\displaystyle\leqslant\Big{\|}\sum_{i=j_{1}}^{j_{2}}a_{j}Sy_{j}\Big{\|}+\|S_{[1,m)}a_{j_{1}}y_{j_{1}}\|+\|S_{[n,\infty)}a_{j_{2}}y_{j_{2}}\|
i=j1j2ajSyj+|aj1|+|aj2|\displaystyle\leqslant\Big{\|}\sum_{i=j_{1}}^{j_{2}}a_{j}Sy_{j}\Big{\|}+|a_{j_{1}}|+|a_{j_{2}}|
i=j1j2ajSyj+c1aj1Syj1+c1aj2Syj2\displaystyle\leqslant\Big{\|}\sum_{i=j_{1}}^{j_{2}}a_{j}Sy_{j}\Big{\|}+c^{-1}\|a_{j_{1}}Sy_{j_{1}}\|+c^{-1}\|a_{j_{2}}Sy_{j_{2}}\|
2C(1+2c1)aiSyias (Syi) is 2C-basic.\displaystyle\leqslant 2C(1+2c^{-1})\big{\|}\sum a_{i}Sy_{i}\big{\|}\hskip 28.45274pt\textrm{as $(Sy_{i})$ is $2C$-basic.}

This proves (15) and hence (Syi)(Sy_{i}) and (yi)(y_{i}) are equivalent basic sequences. All that remains is to prove that (Syj)(Sy_{j}) is weakly null. Let fXf\in X^{*} be some functional.

limj|f(Syj)|\displaystyle\lim_{j\rightarrow\infty}|f(Sy_{j})| =limnlimj|f(P[n,)Syj)|by (iii),\displaystyle=\lim_{n\rightarrow\infty}\lim_{j\rightarrow\infty}|f(P_{[n,\infty)}Sy_{j})|\hskip 28.45274pt\textrm{by (iii),}
limnlimjfP[n,)Syj\displaystyle\leq\lim_{n\rightarrow\infty}\lim_{j\rightarrow\infty}\|f\circ P_{[n,\infty)}\|\|Sy_{j}\|
=0 as the Schauder frame (xi,fi) is shrinking.\displaystyle=0\hskip 56.9055pt\textrm{ as the Schauder frame $(x_{i},f_{i})$ is shrinking.}

Thus, (Syj)(Sy_{j}) is weakly null which implies that (yj)(y_{j}) is weakly null as they are equivalent basic sequences. Hence, (zi)(z_{i}) is a shrinking basis as every normalized block sequence is weakly null. ∎

The proof of the above theorem admits the following corollary.

Corollary 3.6.

Let (xi,fi)(x_{i},f_{i}) be a shrinking frame for a Banach space XX and let (Nk)(N_{k}) satisfy Proposition 3.1. Let (yi)(y_{i}) be a normalized block sequence in Z(Nk)Z_{(N_{k})}.

  1. 1.

    If there is a subsequence (yi)(y^{\prime}_{i}) of (yi)(y_{i}) so that Syi0Sy^{\prime}_{i}\to 0, then there is a further subsequence of (yi)(y^{\prime}_{i}) that is equivalent the the unit vector basis of c0c_{0}.

  2. 2.

    If there is no subsequence (yi)(y^{\prime}_{i}) of (yi)(y_{i}) so that Syi0Sy^{\prime}_{i}\to 0, then there is a subsequence (yi)(y^{\prime}_{i}) of (yi)(y_{i}) so that (yi)(y^{\prime}_{i}) is equivalent to (Syi)(Sy^{\prime}_{i}).

Note that the example given in Section 2 shows that Corollary 3.6 is false for the minimal associated ZminZ_{min}. In that example, (z2j1)j=1(z_{2j-1})_{j=1}^{\infty} is a normalized block sequence in ZminZ_{min} with Sz2j1=1\|Sz_{2j-1}\|=1 for all jj\in\mathbb{N}, but (z2j1)j=1(z_{2j-1})_{j=1}^{\infty} has no subsequence which is equivalent to a sequence in XX.

Remark 3.7.

Let XX have a Schauder frame (xi,fi)(x_{i},f_{i}) with associated space ZZ, analysis operator T:XZT:X\to Z and synthesis operator S:ZXS:Z\to X. The following are fundamental properties of Schauder frames.

  1. 1.

    XX is isomorphic to TXTX, which is a complemented subspace of ZZ.

  2. 2.

    TS:ZZTS:Z\rightarrow Z is a projection of ZZ onto TXTX.

  3. 3.

    Z/TXZ/TX is isomorphic to the range of IZTSI_{Z}-TS, where IZI_{Z} is the identity operator on ZZ.

Proposition 3.8.

Let (xi,fi)(x_{i},f_{i}) be a shrinking frame for a Banach space XX and let (Nk)(N_{k}) satisfy Proposition 3.1. Then Z(Nk)/TXZ_{(N_{k})}/TX is c0c_{0} saturated. That is, every infinite dimensional subspace of Z(Nk)/TXZ_{(N_{k})}/TX contains a further subspace which is isomorphic to c0c_{0}.

Proof.

Using Remark 3.7, Z(Nk)/TXZ_{(N_{k})}/TX is isomorphic to the range of IZTSI_{Z}-TS in Z(Nk)Z_{(N_{k})}. Let YY be an infinite dimensional subspace of the range of IZTSI_{Z}-TS. There exists a normalized block sequence (yi)(y_{i}) in Z(Nk)Z_{(N_{k})} and a sequence (wi)(w_{i}) in YY so that yiwi0\|y_{i}-w_{i}\|\to 0. After passing to a subsequence, we may assume that (yi)(y_{i}) and (wi)(w_{i}) are equivalent basic sequences. As YY is contained in the range of IZTSI_{Z}-TS and TSTS is a projection operator, we have that TSwi=0TSw_{i}=0 for all ii\in\mathbb{N}. Therefore TSyi0TSy_{i}\to 0. Since TT is a an isomorphic embedding, we have that Syi0Sy_{i}\to 0. Therefore we are in the first alternative of Corollary 3.6 and so (yi)(y_{i}) has a subsequence equivalent to the unit vector basis of c0c_{0}. ∎

The above can be compared to the result of Liu-Zheng [13] in which the authors prove that if ZminZ_{min} is the minimal associated space for a Schauder frame of XX, then Zmin/XZ_{min}/X contains an isomorphic copy of c0c_{0} if and only if Zmin/XZ_{min}/X is infinite dimensional.

4. Comparing Associated Spaces with Shrinking Bases

Our next result illustrates that associated spaces of the form Z(Nk)Z_{(N_{k})} are a minimal collection, with respect to domination, among associated spaces with shrinking bases.

Theorem 4.1.

Suppose that (xj,fj)(x_{j},f_{j}) is a shrinking Schauder frame for XX and that WW is an associated space of (xj,fj)(x_{j},f_{j}) with a shrinking associated basis (wj)(w_{j}). Then there exists (Nk)(N_{k}) so that, the basis (zj)(z_{j}) of Z(Nk)Z_{(N_{k})} is dominated by (wj)(w_{j}).

We isolate the following remark.

Remark 4.2.

Let 𝒜\mathcal{A} be a finite collection of finite rank operators on a Banach space WW with a shrinking basis (wj)(w_{j}). Then,

limnsupA𝒜AR[n,)=0 where R[n,)(aiwi)=inaiwi.\lim_{n\to\infty}\sup_{A\in\mathcal{A}}\|A\circ R_{[n,\infty)}\|=0\text{ where }R_{[n,\infty)}(\sum a_{i}w_{i})=\sum_{i\geqslant n}a_{i}w_{i}.
Proof of Theorem 4.1.

Fix (xi,fi)(x_{i},f_{i}), (wj)(w_{j}) and WW as in the statement of the theorem and R[n,)R_{[n,\infty)} as in Remark 4.2. Since (wi)(w_{i}) dominates the minimal associated basis of (xi,fi)(x_{i},f_{i}) there exists K1K\geq 1 so that

(16) supmni=mnaixiKaiwi for all (ai)c00.\sup_{m\leqslant n}\|\sum_{i=m}^{n}a_{i}x_{i}\|\leqslant K\|\sum a_{i}w_{i}\|\hskip 28.45274pt\textrm{ for all }(a_{i})\in c_{00}.

Let SW:WXS_{W}:W\to X be the synthesis operator. For each kk\in\mathbb{N}, the set {P[m0,n0]SW:m0<n0k}\{P_{[m_{0},n_{0}]}\circ S_{W}:m_{0}<n_{0}\leqslant k\} is a finite collection of finite rank operators on WW. By the previous remark, for each kk\in\mathbb{N} there exists NkN_{k}\in\mathbb{N} so that (P[m0,n0]SW)R[m,n]<2k\|(P_{[m_{0},n_{0}]}\circ S_{W})\circ R_{[m,n]}\|<2^{-k} for all m0<n0km_{0}<n_{0}\leqslant k and NkmnN_{k}\leqslant m\leqslant n. Thus we have for all (ai)c00(a_{i})\in c_{00} that

Kaiwi\displaystyle K\Big{\|}\sum a_{i}w_{i}\Big{\|} supmnminaixisupm0n0kNkmn2kP[m0,n0]minaixi\displaystyle\geqslant\sup_{m\leqslant n}\Big{\|}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}\vee\sup_{\begin{subarray}{c}m_{0}\leqslant n_{0}\leq k\\ N_{k}\leqslant m\leqslant n\end{subarray}}2^{k}\Big{\|}P_{[m_{0},n_{0}]}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\Big{\|}
=aizi(Nk)\displaystyle=\Big{\|}\sum a_{i}z_{i}\Big{\|}_{(N_{k})}

Hence, we have that (wi)(w_{i}) KK-dominates the basis (zi)(z_{i}) of Z(Nk)Z_{(N_{k})}. ∎

Let (xj,fj)(x_{j},f_{j}) be a shrinking Schauder frame and let (Nj)(N_{j}) be an increasing sequence of natural numbers which satisfies Proposition 3.1. If (kj)(k_{j}) is an increasing subsequence of natural numbers then we denote Z(kj),(Nj)Z_{(k_{j}),(N_{j})} to be the Banach space with basis (zi(kj))i=1(z^{(k_{j})}_{i})_{i=1}^{\infty} which is the completion of c00c_{00} under the norm:

(17) aizi(kj)(kj),(Nj)=supmnminaixisupjaizikj.\big{\|}\sum a_{i}z^{(k_{j})}_{i}\big{\|}_{(k_{j}),(N_{j})}=\sup_{m\leq n}\big{\|}\sum_{m\leqslant i\leqslant n}a_{i}x_{i}\big{\|}\vee\sup_{j}\big{\|}\sum a_{i}z_{i}\big{\|}_{k_{j}}.

Recall that for zZ(Nj)z\in Z_{(N_{j})} and jj\in\mathbb{N},

(18) zkj:=supm0n0kjNkjmn2kjP[m0,n0]S[m,n]z.\|z\|_{k_{j}}:=\sup_{\begin{subarray}{c}m_{0}\leqslant n_{0}\leq k_{j}\\ N_{k_{j}}\leqslant m\leqslant n\end{subarray}}2^{k_{j}}\big{\|}P_{[m_{0},n_{0}]}S_{[m,n]}z\big{\|}.

It follows from Remark 3.2 and Theorem 3.5 that Z(kj),(Nj)Z_{(k_{j}),(N_{j})} is an associated space of (xi,fi)i=1(x_{i},f_{i})_{i=1}^{\infty} and that (zi(kj))i=1(z^{(k_{j})}_{i})_{i=1}^{\infty} is a shrinking basis for Z(kj),(Nj)Z_{(k_{j}),(N_{j})}.

Theorem 4.3.

Suppose that (xj,fj)(x_{j},f_{j}) is a shrinking Schauder frame for a Banach space XX so that the minimal associated basis is not shrinking. Then for any sequence (Nk)(N_{k}) satisfying Proposition 3.1, there exists an increasing sequence (ki)i=1(k_{i})_{i=1}^{\infty} so that for all infinite subsets L,ML,M\subset\mathbb{N}, the following are equivalent.

  1. 1.

    (zj(ki)iM)(z^{(k_{i})_{i\in M}}_{j}) dominates (zj(ki)iL)(z^{(k_{i})_{i\in L}}_{j}).

  2. 2.

    LML\setminus M is finite.

Before proving Theorem 4.3 we state and prove the following corollary.

Corollary 4.4.

Let (xj,fj)(x_{j},f_{j}) be a shrinking Schauder frame so that the minimal associated basis is not shrinking. Then for each shrinking associated basis (wj)(w_{j}) there are increasing sequences of natural numbers (ki)(k_{i}) and (Ni)(N_{i}) and a set of increasing sequences of natural numbers (Mα)αΔ(M_{\alpha})_{\alpha\in\Delta} with Δ\Delta having cardinality the continuum so that

  1. 1.

    For each αΔ\alpha\in\Delta, the basis (zj(ki)iMα)(z^{(k_{i})_{i\in M_{\alpha}}}_{j}) of Z(ki)iMα,(Ni)Z_{(k_{i})_{i\in M_{\alpha}},(N_{i})} is a shrinking associated basis of (xj,fj)(x_{j},f_{j}) which is dominated by (wj)(w_{j}).

  2. 2.

    For αβ\alpha\not=\beta in Δ\Delta, the bases (zj(ki)iMα)(z^{(k_{i})_{i\in M_{\alpha}}}_{j}) and (zj(ki)iMβ)(z^{(k_{i})_{i\in M_{\beta}}}_{j}) are incomparable.

Proof.

Fix a shrinking associated basis (wj)(w_{j}). By Theorem 4.1, there is a sequence (Ni)(N_{i}) so that the basis (zj)(z_{j}) of Z(Ni)Z_{(N_{i})} is dominated by (wj)(w_{j}). Let (ki)(k_{i}) be a sequence which satisfies the conclusion of Theorem 4.3. For every infinite MM\subseteq\mathbb{N}, the basis (zj(ki)iM)(z^{(k_{i})_{i\in M}}_{j}) of Z(ki)iM,(Ni)Z_{(k_{i})_{i\in M},(N_{i})} is a shrinking associated basis of (xi,fi)(x_{i},f_{i}) which is dominated by (zj)(z_{j}). Hence, (zj(ki)iM)(z^{(k_{i})_{i\in M}}_{j}) is also dominated by (wj)(w_{j}). Let (Mα)αΔ(M_{\alpha})_{\alpha\in\Delta} be a collection of infinite subsets of \mathbb{N} with cardinality the continuum so that MαMβM_{\alpha}\cap M_{\beta} is finite for all αβ\alpha\not=\beta. This is called a collection of almost disjoint sets and is known to exist. In particular, MαMβM_{\alpha}\setminus M_{\beta} is infinite for all αβ\alpha\neq\beta. By Theorem 4.3, (zj(ki)iMα)(z^{(k_{i})_{i\in M_{\alpha}}}_{j}) and (zj(ki)iMβ)(z^{(k_{i})_{i\in M_{\beta}}}_{j}) are incomparable basic sequences for all αβ\alpha\neq\beta. ∎

Proof of Theorem 4.3.

Let (xi,fi)(x_{i},f_{i}) be a shrinking Schauder frame for XX so that the minimum associated basis is not shrinking. Let (Nk)(N_{k}) be an increasing sequence of natural numbers which satisfies Proposition 3.1. By Theorem 3.5 the basis (zi)(z_{i}) of Z(Nk)Z_{(N_{k})} is a shrinking associated basis of (xi,fi)(x_{i},f_{i}) and, moreover for each sequence (kj)(k_{j}), the basis (zi(kj))(z_{i}^{(k_{j})}) of Z(kj),(Nkj)Z_{(k_{j}),(N_{k_{j}})} is a shrinking associated space of (xi,fi)(x_{i},f_{i}).

As the minimal associated basis of ZminZ_{\min} is not shrinking, there exists a normalized block sequence (yn)(y_{n}) in ZminZ_{\min} which is α\alpha-1+\ell_{1}^{+} for some α>0\alpha>0. The Z(Nk)Z_{(N_{k})} norm 11-dominates the ZminZ_{\min} norm. Hence, for all non-negative scalars (an)(a_{n}) we have that

αananynZminanyn(Ni).\alpha\sum a_{n}\leq\big{\|}\sum a_{n}y_{n}\big{\|}_{Z_{\min}}\leq\big{\|}\sum a_{n}y_{n}\big{\|}_{(N_{i})}.

As (zi)(z_{i}) is a shrinking basis for Z(Nk)Z_{(N_{k})}, every bounded block sequence converges weakly to 0 and is hence not 1+\ell_{1}^{+}. Thus, (yn)(y_{n}) cannot be norm bounded as a sequence in Z(Ni)Z_{(N_{i})}. After passing to a subsequence, we assume that yn(Ni)2n\|y_{n}\|_{(N_{i})}\geq 2^{n} for all nn\in\mathbb{N}.

Let C>0C>0 be the frame constant of (xi,fi)(x_{i},f_{i}). Thus, for all kk\in\mathbb{N}, and zZminz\in Z_{min} we have that zkC2kzZmin\|z\|_{k}\leq C2^{k}\|z\|_{Z_{min}}. As ynZmin=1\|y_{n}\|_{Z_{min}}=1 for all nn\in\mathbb{N} and (yn)(y_{n}) is unbounded in Z(Ni)Z_{(N_{i})}, after passing to a subsequence of (yn)(y_{n}) we may assume that there exists a sequence (ki)(k_{i}) so that for all jj\in\mathbb{N}, yj+1kj+122kj\|y_{j+1}\|_{k_{j+1}}\geq 2^{2k_{j}} and supp(ytj+1)[Nkj,Nkj+2)\operatorname{supp}(y_{t_{j+1}})\subseteq[N_{k_{j}},N_{k_{j+2}}). We now assume that LML\setminus M is infinite and will prove that the basis (zj(ki)iM)(z_{j}^{(k_{i})_{i\in M}}) does not dominate the basis (zj(ki)iL)(z_{j}^{(k_{i})_{i\in L}}). Let dLMd\in L\setminus M with d>1d>1. We have that the following holds

yd(ki)iM,(Ni)\displaystyle\|y_{d}\|_{(k_{i})_{i\in M},(N_{i})} =ydZminsupiMydki\displaystyle=\|y_{d}\|_{Z_{min}}\vee\sup_{i\in M}\|y_{d}\|_{k_{i}}
ydZminydkd1 as supp(yd)[Nkd1,Nkd+1)\displaystyle\leqslant\|y_{d}\|_{Z_{min}}\vee\|y_{d}\|_{k_{d-1}}\hskip 28.45274pt\textrm{ as }\operatorname{supp}(y_{d})\subseteq[N_{k_{d-1}},N_{k_{d+1}})
ydZminC2kd1ydZmin\displaystyle\leqslant\|y_{d}\|_{Z_{min}}\vee C2^{k_{d-1}}\|y_{d}\|_{Z_{min}}
=C2kd1\displaystyle=C2^{k_{d-1}}
C2kd1yd(ki)iL,(Ni) as ydkd22kd1\displaystyle\leqslant C2^{-k_{d-1}}\|y_{d}\|_{(k_{i})_{i\in L},(N_{i})}\hskip 28.45274pt\textrm{ as }\|y_{d}\|_{k_{d}}\geq 2^{2k_{d-1}}

Thus, the basis (zj(ki)iM)(z_{j}^{(k_{i})_{i\in M}}) does not dominate the basis (zj(ki)iL)(z_{j}^{(k_{i})_{i\in L}}) as LML\setminus M is infinite. We now assume that LML\setminus M is finite. Let mm^{\prime} be the least element of MM so that L[m,)ML\cap[m^{\prime},\infty)\subseteq M. Let zspaniNkmziz\in\textrm{span}_{i\geq N_{k_{m^{\prime}}}}z_{i}. Thus, zkizkm\|z\|_{k_{i}}\leq\|z\|_{k_{m^{\prime}}} for all imi\leq m^{\prime}. We have that

z(ki)iL,(Ni)\displaystyle\|z\|_{(k_{i})_{i\in L},(N_{i})} =zZminsupiLzki\displaystyle=\|z\|_{Z_{min}}\vee\sup_{i\in L}\|z\|_{k_{i}}
zZminsupiM,imzki\displaystyle\leq\|z\|_{Z_{min}}\vee\sup_{i\in M,i\geq m^{\prime}}\|z\|_{k_{i}}
=z(ki)iM,(Ni)\displaystyle=\|z\|_{(k_{i})_{i\in M},(N_{i})}

Thus, (zj(ki)iM)j=Nkm(z_{j}^{(k_{i})_{i\in M}})_{j=N_{k_{m^{\prime}}}}^{\infty} 11-dominates the basis (zj(ki)iL)j=Nkm(z_{j}^{(k_{i})_{i\in L}})_{j=N_{k_{m^{\prime}}}}^{\infty}. This implies that (zj(ki)iM)j=1(z_{j}^{(k_{i})_{i\in M}})_{j=1}^{\infty} KK-dominates (zj(ki)iL)j=1(z_{j}^{(k_{i})_{i\in L}})_{j=1}^{\infty} for some K1K\geq 1. ∎

5. Shrinking Bounded Approximation Property

A separable Banach space XX has the Bounded Approximation Property BAP if there is a sequence (Bn)(B_{n}) of finite rank operators on XX so that limnxBnx=0\lim_{n}\|x-B_{n}x\|=0 for all xXx\in X. The uniform boundedness principle implies that whenever this condition holds there is a λ>0\lambda>0 with Bnλ\|B_{n}\|\leqslant\lambda for each nn. A space with this property for λ\lambda is said to have the λ\lambda-AP. Moreover, setting A1=B1A_{1}=B_{1} and An=BnBn1A_{n}=B_{n}-B_{n-1} for n>1n>1 we can replace BnB_{n} with i=1nAi\sum_{i=1}^{n}A_{i}. We note that the definition of the λ\lambda-AP is that there for each ε>0\varepsilon>0 and compact set KK in XX there is finite rank operator TT with Tλ\|T\|\leqslant\lambda so that xTxε\|x-Tx\|\leqslant\varepsilon for all xKx\in K.

To mirror the definition of shrinking basis, one may wish to define a space XX to have the shrinking-BAP if there are finite rank operators (Bn)(B_{n}) on XX so that limnxBnx=0\lim_{n}\|x-B_{n}x\|=0 for all xXx\in X and limnfBnf=0\lim_{n}\|f-B_{n}^{*}f\|=0 for all fXf\in X^{*}. That is, the operators in the space approximating the identity also have the property that their dual operator approximate the identity. This is analogous to: A basis (xn)(x_{n}) is shrinking if and only if the coordinate functionals (xn)(x_{n}^{*}) form a basis for XX^{*}.

The above definition of shrinking-BAP is formally stronger than simply XX^{*} having the BAP and has been isolated before under the name duality-BAP [4, page 288]. The surprising fact that these notions are equivalent is the content of the following proposition [4, Proposition 3.5].

Theorem 5.1.

A space XX has the shrinking BAP if and only if XX^{*} has the BAP.

In fact a lot more is known: A dual space XX^{*} has the AP if and only if XX and XX^{*} have the 11-AP (i.e. the metric approximation property). Another result related to the current work is the following [4, Theorem 4.9].

Theorem 5.2.

Let XX be a Banach space with separable dual. Then XX^{*} has the BAP if and only if XX embeds complementably in a Banach space with a shrinking basis.

Theorem 5.2 is a complemented version of Zippin’s theorem [18] stating that every Banach space with a separable dual embeds into a space with a shrinking basis. It is also a refinement of the aforementioned theorem of Pełczynski and Johnson-Rosethal-Zippin stating that every space with the BAP embeds complementably into a space with a basis. The proof of Theorem 5.2 (as stated [4]) follows the results in [10]. Here the authors show that if XX^{*} has the BAP then XCpX\oplus C_{p} has a shrinking basis where CpC_{p} is the p\ell_{p} sum of finite dimensional spaces (En)(E_{n}) which are dense (with the Banach Mazur distance) in the space of all finite dimensional spaces.

We present an alternative proof Theorem 5.2 using the language of frames and modifying the proof of Pełczyński[16]. The technique we employ for this proof is also used by Mujica and Vieira in [14] to give a quantitative improvement of Pełczyński’s theorem.

Theorem 5.3.

Let XX be a Banach space. Then XX^{*} has the BAP only if and only if XX has a shrinking Schauder frame.

Note that Theorem 5.2 follows immediately from combining Theorem 5.3 with Theorem 3.5.

Proof of Theorem 5.3.

If XX has a shrinking Schauder frame (xj,fj)(x_{j},f_{j}) then (fj,xj)(f_{j},x_{j}) is a Schauder frame of XX^{*} and hence XX^{*} has the BAP. Before proving the reverse direction we prove the following finite dimensional result. Let XX be a Banach space and let A:XXA:X\rightarrow X be a finite rank operator. Let dd be the rank of AA and let mm\in\mathbb{N}. Then there exists (xj,fj)j=1mdX×X(x_{j},f_{j})_{j=1}^{md}\subseteq X\times X^{*} such that the sequence of rank one operators (fjxj)j=1md(f_{j}\otimes x_{j})_{j=1}^{md} satisfies

(19) j=1qdfjxj=qmA for all 0qm,j=qd+1qd+rfjxjrmA for all 0q<m and 0rd.\begin{split}&\sum_{j=1}^{qd}f_{j}\otimes x_{j}\!=\frac{q}{m}\!A\hskip 14.22636pt\textrm{ for all $0\leqslant q\leqslant m$,}\\ \Big{\|}&\sum_{j=qd+1}^{qd+r}f_{j}\otimes x_{j}\Big{\|}\!\leqslant\!\frac{r}{m}\|A\|\hskip 14.22636pt\textrm{ for all $0\leqslant q<m$ and $0\leqslant r\leqslant d$.}\end{split}

Indeed, let (ei)i=1d(e_{i})_{i=1}^{d} be an Auerbach basis of A(X)A(X) with bi-orthogonal functionals (ei)i=1d(e_{i}^{*})_{i=1}^{d}. In particular, eiei=1\|e_{i}^{*}\otimes e_{i}\|=1 for all 1id1\leq i\leq d and eiei\sum e^{*}_{i}\otimes e_{i} is the identity on A(X)A(X). For each 1jdm1\leq j\leq dm, we let xj=erx_{j}=e_{r} and fj=m1Aerf_{j}=m^{-1}A^{*}e_{r}^{*} where j=qd+rj=qd+r for some 0q<d0\leqslant q<d and 1rm1\leqslant r\leqslant m. Let xXx\in X and 0qm0\leqslant q\leqslant m. Then,

j=1qdfjxj=qr=1d1m(Aer)er=qm(r=1derer)A=qmA.\sum_{j=1}^{qd}f_{j}\otimes x_{j}=q\sum_{r=1}^{d}\frac{1}{m}(A^{*}e^{*}_{r})\otimes e_{r}=\frac{q}{m}\big{(}\sum_{r=1}^{d}e^{*}_{r}\otimes e_{r}\big{)}A=\frac{q}{m}A.

Thus the first part of (LABEL:E:Pel) holds. To prove the second part we let 0q<m0\leqslant q<m, and 1rm1\leqslant r\leqslant m. We have that,

j=qm+1qm+rfjxj=j=1r1m(Aej)ejj=1r1mejejA=rmA\Big{\|}\sum_{j=qm+1}^{qm+r}f_{j}\otimes x_{j}\Big{\|}=\Big{\|}\sum_{j=1}^{r}\frac{1}{m}(A^{*}e^{*}_{j})\otimes e_{j}\Big{\|}\leqslant\sum_{j=1}^{r}\frac{1}{m}\|e^{*}_{j}\otimes e_{j}\|\|A\|=\frac{r}{m}\|A\|

Thus we have proven (LABEL:E:Pel). We now assume that XX^{*} has the BAP and hence by Theorem 5.1 we have that XX has the shrinking-BAP. Let (Ak)(A_{k}) be a sequence of finite rank operators so that x=k=1Akxx=\sum_{k=1}^{\infty}A_{k}x for all xXx\in X and that for each fXf\in X we have that limnf(Ik=1nAk)=0\lim_{n}\|f\circ(I-\sum_{k=1}^{n}A_{k})\|=0 where II is the identity operator on XX. Let dkd_{k} be the dimension of Ak(X)A_{k}(X) and let (mk)k=1(m_{k})_{k=1}^{\infty} be a sequence of natural numbers with dk/mk0d_{k}/m_{k}\to 0.

As described above, we construct for all kk\in\mathbb{N} a finite sequence (xik,fik)i=1dkmk(x^{k}_{i},f^{k}_{i})_{i=1}^{d_{k}m_{k}} which satisfies (LABEL:E:Pel) for the finite rank operator AkA_{k}. We claim that the infinite sequence (xik,fik)k,1idkmk(x_{i}^{k},f_{i}^{k})_{k\in\mathbb{N},1\leq i\leq d_{k}m_{k}} is a shrinking Schauder frame of XX when enumerated in the natural way. We will prove that (xik,fik)k,1idkmk(x_{i}^{k},f_{i}^{k})_{k\in\mathbb{N},1\leq i\leq d_{k}m_{k}} is shrinking, and we note that the proof that (xik,fik)k,1idkmk(x_{i}^{k},f_{i}^{k})_{k\in\mathbb{N},1\leq i\leq d_{k}m_{k}} is a Schauder frame follows the same argument. Let fXf\in X^{*} and ε>0\varepsilon>0. Choose KK\in\mathbb{N} large enough so that for all kKk\geqslant K we have dkmkAk<ε\frac{d_{k}}{m_{k}}\|A_{k}\|<\varepsilon and fi=1kAif<ε\|f-\sum_{i=1}^{k}A_{i}^{*}f\|<\varepsilon. Thus we also have that Akf<2ε\|A^{*}_{k}f\|<2\varepsilon. Let lj=1Kdjmjl\geq\sum_{j=1}^{K}d_{j}m_{j}. Then l=j=1k1djmj+qdk+rl=\sum_{j=1}^{k-1}d_{j}m_{j}+qd_{k}+r for some 0q<mk0\leqslant q<m_{k}, 1rdk1\leqslant r\leqslant d_{k}, and k>Kk>K. The approximation of ff by the partial sum consisting of the first ll terms of f(xji)fji\sum f(x_{j}^{i})f_{j}^{i} satisfies

f(i=1k1\displaystyle\Big{\|}f-\Big{(}\sum_{i=1}^{k-1} j=1dimif(xji)fji+j=1qdkf(xjk)fjk+j=1rf(xjk)fjk)\displaystyle\sum_{j=1}^{d_{i}m_{i}}f(x_{j}^{i})f_{j}^{i}+\sum_{j=1}^{qd_{k}}f(x_{j}^{k})f_{j}^{k}+\sum_{j=1}^{r}f(x_{j}^{k})f_{j}^{k}\Big{)}\Big{\|}
=(fi=1k1Aif)qmkAkfj=1rfjk(f)xjk by (LABEL:E:Pel),\displaystyle=\Big{\|}\Big{(}f-\sum_{i=1}^{k-1}A^{*}_{i}f\Big{)}-\frac{q}{m_{k}}A^{*}_{k}f-\sum_{j=1}^{r}f_{j}^{k}(f)x_{j}^{k}\Big{\|}\hskip 14.22636pt\textrm{ by \eqref{E:Pel},}
fi=1k1Aif+qmkAkf+rmkAkf<4ε.\displaystyle\leq\Big{\|}f-\sum_{i=1}^{k-1}A^{*}_{i}f\Big{\|}+\frac{q}{m_{k}}\|A_{k}^{*}f\|+\frac{r}{m_{k}}\|A^{*}_{k}\|\|f\|<4\varepsilon.

Thus we have that (fik,xik)k,1idkmk(f_{i}^{k},x_{i}^{k})_{k\in\mathbb{N},1\leq i\leq d_{k}m_{k}} is a Schauder frame of XX^{*} and hence (xik,fik)k,1idkmk(x_{i}^{k},f_{i}^{k})_{k\in\mathbb{N},1\leq i\leq d_{k}m_{k}} is shrinking. ∎

References

  • [1] K. Beanland, D. Freeman, and R. Liu. Upper and lower estimates for Schauder frames and atomic decompositions. Fund. Math., 231(2):161–188, 2015.
  • [2] D. Carando and S. Lassalle. Duality, reflexivity and atomic decompositions in Banach spaces. Studia Math., 191(1):67–80, 2009.
  • [3] P. Casazza, O. Christensen, and D. T. Stoeva. Frame expansions in separable Banach spaces. J. Math. Anal. Appl., 307(2):710–723, 2005.
  • [4] P. G. Casazza. Approximation properties. In Handbook of the geometry of Banach spaces, Vol. I, pages 271–316. North-Holland, Amsterdam, 2001.
  • [5] P. G. Casazza, S. J. Dilworth, E. Odell, T. Schlumprecht, and A. Zsák. Coefficient quantization for frames in Banach spaces. J. Math. Anal. Appl., 348(1):66–86, 2008.
  • [6] P. G. Casazza, D. Han, and D. Larson. Frames for banach spaces. Cont. Math., 247:149–182, 1999.
  • [7] H. G. Feichtinger and K. Gröchenig. A unified approach to atomic decompositions via integrable group representations. Proc. Conf. Function Spaces and Applications, Lecture Notes Math., 1302:52–73, 1988.
  • [8] D. Freeman, E. Odell, T. Schlumprecht, and A. Zsák. Banach spaces of bounded Szlenk index. II. Fund. Math., 205(2):161–177, 2009.
  • [9] K. Gröchenig. Describing functions: Atomic decompositions versus frames. Monatshefte fur Math., 112:1–41, 1991.
  • [10] W. B. Johnson, H. P. Rosenthal, and M. Zippin. On bases, finite dimensional decompositions and weaker structures in Banach spaces. Israel J. Math., 9:488–506, 1971.
  • [11] J. Lindenstrauss and L. Tzafriri. Classical Banach spaces. I. Springer-Verlag, Berlin, 1977. Sequence spaces, Ergebnisse der Mathematik und ihrer Grenzgebiete, Vol. 92.
  • [12] R. Liu. On shrinking and boundedly complete Schauder frames of Banach spaces. J. Math. Anal. Appl., 365(1):385–398, 2010.
  • [13] R. Liu and B. Zheng. A characterization of Schauder frames which are near-Schauder bases. J. Fourier Anal. Appl., 16(5):791–803, 2010.
  • [14] J. Mujica and D. M. Vieira. Schauder bases and the bounded approximation property in separable Banach spaces. Studia Math., 196(1):1–12, 2010.
  • [15] E. Odell and T. Schlumprecht. Trees and branches in banach spaces. Trans. Amer. Math. Soc., 354:4085––4108, 2002.
  • [16] A. Pełczyński. Any separable Banach space with the bounded approximation property is a complemented subspace of a Banach space with a basis. Studia Math., 40:239–243, 1971.
  • [17] S. J. Szarek. A Banach space without a basis which has the bounded approximation property. Acta Math., 159(1-2):81–98, 1987.
  • [18] M. Zippin. Banach spaces with separable duals. Trans. Amer. Math. Soc., 310(1):371–379, 1988.