Weaving Schauder frames

Peter G. Casazza Department of Mathematics
University of Missouri
Columbia, MO 65211
casazzap@mail.missouri.edu
Daniel Freeman Department of Mathematics and Computer Science
Saint Louis University
St Louis, MO 63103
dfreema7@slu.edu
 and  Richard G. Lynch Department of Mathematics
University of Missouri
Columbia, MO 65211
rglz82@mail.missouri.edu
Abstract.

We extend the concept of weaving Hilbert space frames to the Banach space setting. Similar to frames in a Hilbert space, we show that for any two approximate Schauder frames for a Banach space, every weaving is an approximate Schauder frame if and only if there is a uniform constant C1𝐶1C\geq 1 such that every weaving is a C𝐶C-approximate Schauder frame. We also study weaving Schauder bases, where it is necessary to introduce two notions of weaving. On one hand, we can ask if two Schauder bases are woven when considered as Schauder frames with their biorthogonal functionals, and alternatively, we can ask if each weaving of two Schauder bases remains a Schauder basis. We will prove that these two notions coincide when all weavings are unconditional, but otherwise they can be different. Lastly, we prove two perturbation theorems for approximate Schauder frames.

2000 Mathematics Subject Classification:
46B20, 42C15
The first and third authors were supported by NSF DMS 1307685; NSF ATD 1321779. The second author was supported by grant 353293 from the Simon’s foundation.

1. Introduction

The concept of weaving Hilbert space frames was first introduced in [BCGLL]. Two frames {xj0}jJsubscriptsuperscriptsubscript𝑥𝑗0𝑗𝐽\{x_{j}^{0}\}_{j\in J} and {xj1}jJsubscriptsuperscriptsubscript𝑥𝑗1𝑗𝐽\{x_{j}^{1}\}_{j\in J} for a Hilbert space H𝐻H are called woven if there exist constants 0<AB0𝐴𝐵0<A\leq B so that the weaving {xj0}jσ{xj1}jσcsubscriptsuperscriptsubscript𝑥𝑗0𝑗𝜎subscriptsuperscriptsubscript𝑥𝑗1𝑗superscript𝜎𝑐\{x_{j}^{0}\}_{j\in\sigma}\cup\{x_{j}^{1}\}_{j\in\sigma^{c}} is a frame with bounds A,B𝐴𝐵A,B for every choice of partition σJ𝜎𝐽\sigma\subset J.

Weaving frames is motivated by distributed signal processing. We can think of each jJ𝑗𝐽j\in J as a sensor or node, and for each one we measure a signal with either xj0superscriptsubscript𝑥𝑗0x_{j}^{0} or xj1superscriptsubscript𝑥𝑗1x_{j}^{1}. Can a signal be recovered robustly regardless of how the measurements are taken at each node? Another way to think of this is if (xj0)jJsubscriptsuperscriptsubscript𝑥𝑗0𝑗𝐽(x_{j}^{0})_{j\in J} and (xj1)jJsubscriptsubscriptsuperscript𝑥1𝑗𝑗𝐽(x^{1}_{j})_{j\in J} are two sensor networks then we are interested in whenever parts of one network can be used to replace parts of the other. In [BCGLL], [CL], and [DV] the sensors are modeled by inner product with a vector in a frame or Riesz basis for a Hilbert space. In this paper, we extend the concept of weaving frames to the Banach space setting. We will model sensors by evaluation by linear functionals associated with an approximate Schauder frame or Schauder basis. We will be specifically considering weaving two bases or frames together, but we note that all of our results can be easily extended to weaving any finite number of bases or frames.

We begin by recalling the definition of a frame for a Hilbert space.

Definition 1.1.

A set of vectors (xj)jJsubscriptsubscript𝑥𝑗𝑗𝐽(x_{j})_{j\in J} in a Hilbert space H𝐻H is called a frame for H𝐻H if there exists positive constants A𝐴A and B𝐵B (called the lower and upper frame bounds respectively) such that

Ax2jJ|x,xi|2Bx2for all xH.formulae-sequence𝐴superscriptnorm𝑥2subscript𝑗𝐽superscript𝑥subscript𝑥𝑖2𝐵superscriptnorm𝑥2for all 𝑥𝐻A\|x\|^{2}\leq\sum_{j\in J}|\langle x,x_{i}\rangle|^{2}\leq B\|x\|^{2}\quad\textrm{for all }x\in H. (1)

Given a set of vectors (xj)jJsubscriptsubscript𝑥𝑗𝑗𝐽(x_{j})_{j\in J} in a Hilbert space H𝐻H, we define an operator S:HH:𝑆𝐻𝐻S:H\rightarrow H (called the frame operator) by S(x):=jJx,xixiassign𝑆𝑥subscript𝑗𝐽𝑥subscript𝑥𝑖subscript𝑥𝑖S(x):=\sum_{j\in J}\langle x,x_{i}\rangle x_{i}, for all xH𝑥𝐻x\in H, assuming that the series converges. The frame operator can be used to give a simple characterization of frames. Indeed, (xj)jJsubscriptsubscript𝑥𝑗𝑗𝐽(x_{j})_{j\in J} is a frame if and only if the frame operator is well defined, bounded, and has bounded inverse. Furthermore, suppose that (xj)jJsubscriptsubscript𝑥𝑗𝑗𝐽(x_{j})_{j\in J} is a frame with frame operator S𝑆S, optimal lower frame bound A𝐴A, and optimal upper frame bound B𝐵B. Then S=Bnorm𝑆𝐵\|S\|=B and S1=A1normsuperscript𝑆1superscript𝐴1\|S^{-1}\|=A^{-1}.

The characterization of a frame in terms of the frame operator is a useful perspective when considering how to generalize frame theory to Banach spaces. Frames have been generalized to Banach spaces in multiple ways. One way is to generalize the frame inequality (1) as in [G, FG], but the following definition of a Schauder frame and approximate Schauder frame instead generalizes the definition of the frame operator. Schauder frames were first used in [CDOSZ] with the goal of creating a procedure to represent vectors using quantized coefficients and are a generalization of framings of Banach spaces which were introduced in [CHL]. Approximate Schauder frames were defined in [FOSZ] and were used in the construction of a Schauder frame for Lp(d)subscript𝐿𝑝superscript𝑑L_{p}(\mathbb{R}^{d}) with p>2𝑝2p>2 formed by translations of a single function.

Definition 1.2.

Let X𝑋X be a Banach space with dual space Xsuperscript𝑋X^{*}. A sequence (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} in X×X𝑋superscript𝑋X\times X^{*} is called an approximate Schauder frame for X𝑋X if the operator S:XX:𝑆𝑋𝑋S:X\rightarrow X (called the frame operator) defined by S(x):=i=1fi(x)xiassign𝑆𝑥superscriptsubscript𝑖1subscript𝑓𝑖𝑥subscript𝑥𝑖S(x):=\sum_{i=1}^{\infty}f_{i}(x)x_{i} is well defined and is bounded with bounded inverse. We say that (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is a C𝐶C-approximate Schauder frame if S,S1Cnorm𝑆normsuperscript𝑆1𝐶\|S\|,\|S^{-1}\|\leq C. We say that (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is a Schauder frame if the frame operator equals the identity operator.

An approximate Schauder frame is called unconditional if the series fπ(i)(x)xπ(i)subscript𝑓𝜋𝑖𝑥subscript𝑥𝜋𝑖\sum f_{\pi(i)}(x)x_{\pi(i)} converges to S(x)𝑆𝑥S(x) for every permutation π𝜋\pi of \mathbb{N}. For Cu1subscript𝐶𝑢1C_{u}\geq 1, an unconditional approximate Schauder frame is called Cusubscript𝐶𝑢C_{u}-unconditional if for all xX𝑥𝑋x\in X and all choices of εi{1,1}subscript𝜀𝑖11\varepsilon_{i}\in\{-1,1\}, iεifi(x)xiCuS(x)normsubscript𝑖subscript𝜀𝑖subscript𝑓𝑖𝑥subscript𝑥𝑖subscript𝐶𝑢norm𝑆𝑥\|\sum_{i\in\mathbb{N}}\varepsilon_{i}f_{i}(x)x_{i}\|\leq C_{u}\|S(x)\|. For Cs1subscript𝐶𝑠1C_{s}\geq 1, an unconditional approximate Schauder frame is Cssubscript𝐶𝑠C_{s}-suppression unconditional if for all xX𝑥𝑋x\in X and all ΓΓ\Gamma\subseteq\mathbb{N}, iΓfi(x)xiCsS(x)normsubscript𝑖Γsubscript𝑓𝑖𝑥subscript𝑥𝑖subscript𝐶𝑠norm𝑆𝑥\|\sum_{i\in\Gamma}f_{i}(x)x_{i}\|\leq C_{s}\|S(x)\|. As is the case for unconditional Schauder bases, we have that an approximate Schauder frame is unconditional if and only if it suppression unconditional and CsCu2Cssubscript𝐶𝑠subscript𝐶𝑢2subscript𝐶𝑠C_{s}\leq C_{u}\leq 2C_{s}.

Note that if (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is an approximate Schauder frame for a Banach space X𝑋X with frame operator S𝑆S then (xi,(S1)fi)i=1superscriptsubscriptsubscript𝑥𝑖superscriptsuperscript𝑆1subscript𝑓𝑖𝑖1(x_{i},(S^{-1})^{*}f_{i})_{i=1}^{\infty} is a Schauder frame for X𝑋X. This is a generalization of the often used result that if (xi)i=1superscriptsubscriptsubscript𝑥𝑖𝑖1(x_{i})_{i=1}^{\infty} is a frame for a Hilbert space H𝐻H with frame operator S𝑆S then (S1/2xi)i=1superscriptsubscriptsuperscript𝑆12subscript𝑥𝑖𝑖1(S^{-1/2}x_{i})_{i=1}^{\infty} is a Parseval frame for H𝐻H.

Our goal is to extend the notion of weaving frames in a Hilbert space to weaving frames and bases for Banach spaces and to prove the analogous results in this setting. This is a different generalization of weaving frames than given in [DV], where they instead consider weaving together infinitely many Hilbert space frames.

Since ordering plays a crucial role, we index the weavings with elements from {0,1}superscript01\{0,1\}^{\mathbb{N}}, the set of functions from \mathbb{N} to {0,1}01\{0,1\}. Note that {0,1}superscript01\{0,1\}^{\mathbb{N}} is a compact Hausdorff space in the product topology. We will repeatedly make use of this fact.

Definition 1.3.

[BCGLL, Definition 3] Let (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} be two frames for a Hilbert space H𝐻H. We say that (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are woven if there are positive constants A𝐴A and B𝐵B such that for every function σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} we have that (xjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j})_{j=1}^{\infty} is a frame for H𝐻H with lower frame bound A𝐴A and upper frame bound B𝐵B.

We give the following natural generalization of weaving frames for Hilbert spaces to weaving approximate Schauder frames for Banach spaces.

Definition 1.4.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be two approximate Schauder frames for a Banach space X𝑋X. A sequence (xjσ(j),fjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗subscriptsuperscript𝑓𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j},f^{\sigma(j)}_{j})_{j=1}^{\infty} with σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} is called a weaving of (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty}. Given C1𝐶1C\geq 1, we say that (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} are C𝐶C-woven if every weaving is a C𝐶C-approximate Schauder frame. We say that (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} are woven if they are C𝐶C-woven for some C1𝐶1C\geq 1.

Note that this is a true generalization of the definition of woven frames for Hilbert spaces as two frames (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} of a Hilbert space H𝐻H are woven if and only if the approximate Schauder frames (xj0,xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j},x^{0}_{j})_{j=1}^{\infty} and (xj1,xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j},x^{1}_{j})_{j=1}^{\infty} are woven (where we identify a vector yH𝑦𝐻y\in H with the linear functional xx,ymaps-to𝑥𝑥𝑦x\mapsto\langle x,y\rangle).

A Riesz basis for a Hilbert space H𝐻H is any sequence of vectors that is image of some orthonormal basis under an invertible operator (this is equivalent to being a semi-normalized unconditional basis for a Hilbert space). In [BCGLL], it is proven that if (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are two Riesz bases for a Hilbert space H𝐻H then (xjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j})_{j=1}^{\infty} is a Riesz basis for H𝐻H for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} if and only if (xjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j})_{j=1}^{\infty} is a frame for H𝐻H for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. Thus, there is no need to distinguish between two Riesz bases having all weavings be Riesz bases and two Riesz bases having all weavings be frames. However, we will show that these two notions are not always the same for Schauder bases and thus we will require a second definition of weaving in this context.

A sequence (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} in a Banach space X𝑋X is a Schauder basis for X𝑋X if for every xX𝑥𝑋x\in X there exists a unique sequence of scalars (aj)j=1superscriptsubscriptsubscript𝑎𝑗𝑗1(a_{j})_{j=1}^{\infty} such that x=j=1ajxj𝑥superscriptsubscript𝑗1subscript𝑎𝑗subscript𝑥𝑗x=\sum_{j=1}^{\infty}a_{j}x_{j}, and a basis is called unconditional if the series converges in every order for every xX𝑥𝑋x\in X. The biorthogonal functionals of (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} is the sequence (xj)j=1superscriptsubscriptsubscriptsuperscript𝑥𝑗𝑗1(x^{*}_{j})_{j=1}^{\infty} in Xsuperscript𝑋X^{*} such that xj(x)=ajsubscriptsuperscript𝑥𝑗𝑥subscript𝑎𝑗x^{*}_{j}(x)=a_{j} for all j𝑗j\in\mathbb{N}. A sequence (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} in a Banach space X𝑋X is called a basic sequence if it is a basis of its closed span [xj]j=1superscriptsubscriptdelimited-[]subscript𝑥𝑗𝑗1[x_{j}]_{j=1}^{\infty}. The basis constant of a basic sequence (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} is the least constant C1𝐶1C\geq 1 such that i=1naixiCi=1Naixinormsuperscriptsubscript𝑖1𝑛subscript𝑎𝑖subscript𝑥𝑖𝐶normsuperscriptsubscript𝑖1𝑁subscript𝑎𝑖subscript𝑥𝑖\|\sum_{i=1}^{n}a_{i}x_{i}\|\leq C\|\sum_{i=1}^{N}a_{i}x_{i}\| for all choices of scalars (ai)i=1Nsuperscriptsubscriptsubscript𝑎𝑖𝑖1𝑁(a_{i})_{i=1}^{N} and all nN𝑛𝑁n\leq N.

Definition 1.5.

Let (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} be two Schauder bases for a Banach space X𝑋X. A sequence (xjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j})_{j=1}^{\infty} with σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} is called a weaving of (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty}. We say that (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are woven if every weaving is a Schauder basis for X𝑋X.

Let (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} be two Schauder bases for a Banach space X𝑋X with biorthogonal functionals (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0*}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1*}_{j})_{j=1}^{\infty}. This implies that (xj0,xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j},x^{0*}_{j})_{j=1}^{\infty} and (xj1,xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j},x^{1*}_{j})_{j=1}^{\infty} are both Schauder frames for X𝑋X. We can now consider whether or not the two bases (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are woven and whether or not the two Schauder frames (xj0,xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j},x^{0*}_{j})_{j=1}^{\infty} and (xj1,xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j},x^{1*}_{j})_{j=1}^{\infty} are woven. In Example 3.2, we show that it is possible for two Schauder bases to be woven but for the corresponding Schauder frames to not be woven. The key idea for why this is possible is that when we weave approximate Schauder frames we are weaving both two sequences of vectors and two sequences of functionals, but if we weave two Schauder bases and obtain a basis the new corresponding biorthogonal functionals can be completely unrelated to the biorthogonal functionals of the two bases that we started with. This gives us more flexibility when weaving Schauder bases, but it is actually a problem when we consider our motivation as determining when different sensor networks can be combined. In that context, the given linear functionals are fundamental components of the sensors and thus we cannot change them based on what weaving we choose to use. For this reason, it is important for applications for two Schauder bases to be woven both as bases and as approximate Schauder frames. We prove that this is the case when all weavings are unconditional. In particular, in Theorem 4.5 we prove a number of equivalent properties, including that if (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are two unconditional Schauder bases then every weaving of (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} is an unconditional basis if and only if every weaving of (xj0,xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j},x^{0*}_{j})_{j=1}^{\infty} and (xj1,xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j},x^{1*}_{j})_{j=1}^{\infty} is an unconditional approximate Schauder frame. In Theorem 4.5 we prove as well that if (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are two unconditional Schauder bases of a Banach space X𝑋X then every weaving of (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} is an unconditional basis of X𝑋X if and only if every weaving is an unconditional basic sequence. Surprisingly, we show that this is not always the case if some weavings are conditional, and in Example 3.5 we give two unconditional Schauder bases for 1subscript1\ell_{1} such that every weaving is a basic sequence, but there exists a weaving which is not a Schauder basis for all of 1subscript1\ell_{1}.

Lastly, in Section 5 we prove two perturbation theorems for approximate Schauder frames and prove that these result in woven approximate Schauder frames. For a thorough approach to the basics of frame theory, see [CK, CL2, C]. For background on Banach space theory see [FHHMPZ] and [LT].

2. Uniform constants for woven approximate Schauder frames

One of the major results in [BCGLL] is a proof that if (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are two frames for a Hilbert space H𝐻H, then every weaving is a frame for H𝐻H if and only if there is a uniform constant C1𝐶1C\geq 1 such that (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are C𝐶C-woven. We shall prove in Theorem 2.4 that this same uniformity result holds more generally for approximate Schauder frames.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be two approximate Schauder frames of X𝑋X such that every weaving is an approximate Schauder frame. For I𝐼I\subseteq\mathbb{N} and σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, we define a linear operator Pσ,I:XX:subscript𝑃𝜎𝐼𝑋𝑋P_{\sigma,I}:X\rightarrow X by Pσ,I(x)=jIfjσ(j)(x)xjσ(j)subscript𝑃𝜎𝐼𝑥subscript𝑗𝐼subscriptsuperscript𝑓𝜎𝑗𝑗𝑥superscriptsubscript𝑥𝑗𝜎𝑗P_{\sigma,I}(x)=\sum_{j\in I}f^{\sigma(j)}_{j}(x)x_{j}^{\sigma(j)}. Note that this series will converge for every interval I𝐼I\subseteq\mathbb{N}, but that the series may not converge if the approximate Schauder frame is conditional and I𝐼I is not an interval. In this notation, we have that Pσ,subscript𝑃𝜎P_{\sigma,\mathbb{N}} is the frame operator of the weaving (xjσ(j),fjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗subscriptsuperscript𝑓𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j},f^{\sigma(j)}_{j})_{j=1}^{\infty}. Thus, to prove that two approximate Schauder frames (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} are C𝐶C-woven, we need to prove that Pσ,Cnormsubscript𝑃𝜎𝐶\|P_{\sigma,\mathbb{N}}\|\leq C and Pσ,1Cnormsuperscriptsubscript𝑃𝜎1𝐶\|P_{\sigma,\mathbb{N}}^{-1}\|\leq C for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}.

Lemma 2.1.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be two approximate Schauder frames for X𝑋X such that every weaving is an approximate Schauder frame for X𝑋X. If xX𝑥𝑋x\in X and ε>0𝜀0\varepsilon>0, then there exists N𝑁N\in\mathbb{N} such that Pσ,[m,n]x<εnormsubscript𝑃𝜎𝑚𝑛𝑥𝜀\|P_{\sigma,[m,n]}x\|<\varepsilon for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and nmN𝑛𝑚𝑁n\geq m\geq N.

Proof.

Let xX𝑥𝑋x\in X and ε>0𝜀0\varepsilon>0. Assume the contrary that for all N𝑁N\in\mathbb{N}, we may choose σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and nmN𝑛𝑚𝑁n\geq m\geq N such that Pσ,[m,n]xεnormsubscript𝑃𝜎𝑚𝑛𝑥𝜀\|P_{\sigma,[m,n]}x\|\geq\varepsilon. We may then inductively choose (σj)j=1superscriptsubscriptsubscript𝜎𝑗𝑗1(\sigma_{j})_{j=1}^{\infty} and m1n1<m2n2<subscript𝑚1subscript𝑛1subscript𝑚2subscript𝑛2m_{1}\leq n_{1}<m_{2}\leq n_{2}<... such that Pσj,[mj,nj]xεnormsubscript𝑃subscript𝜎𝑗subscript𝑚𝑗subscript𝑛𝑗𝑥𝜀\|P_{\sigma_{j},[m_{j},n_{j}]}x\|\geq\varepsilon. As, ([mj,nj])j=1superscriptsubscriptsubscript𝑚𝑗subscript𝑛𝑗𝑗1([m_{j},n_{j}])_{j=1}^{\infty} is a sequence of disjoint intervals, we may choose σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} such that σ|[mj,nj]=σj|[mj,nj]evaluated-at𝜎subscript𝑚𝑗subscript𝑛𝑗evaluated-atsubscript𝜎𝑗subscript𝑚𝑗subscript𝑛𝑗\sigma|_{[m_{j},n_{j}]}=\sigma_{j}|_{[m_{j},n_{j}]} for all j𝑗j\in\mathbb{N}. Thus, we have that Pσ,[mj,nj]xεnormsubscript𝑃𝜎subscript𝑚𝑗subscript𝑛𝑗𝑥𝜀\|P_{\sigma,[m_{j},n_{j}]}x\|\geq\varepsilon for all j𝑗j\in\mathbb{N}. Since we assumed that (xjσ(j),fjσ(j))subscriptsuperscript𝑥𝜎𝑗𝑗subscriptsuperscript𝑓𝜎𝑗𝑗(x^{\sigma(j)}_{j},f^{\sigma(j)}_{j}) is an approximate Schauder frame, this contradicts the fact that Pσ,xsubscript𝑃𝜎𝑥P_{\sigma,\mathbb{N}}x converges. ∎

Lemma 2.2.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be two approximate Schauder frames for X𝑋X such that every weaving is an approximate Schauder frame for X𝑋X. There exists a constant D>0𝐷0D>0 such that Pσ,[m,n]Dnormsubscript𝑃𝜎𝑚𝑛𝐷\|P_{\sigma,[m,n]}\|\leq D for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and m,n𝑚𝑛m,n\in\mathbb{N}.

Proof.

Let xX𝑥𝑋x\in X. By Lemma 2.1 there exists N𝑁N\in\mathbb{N} such that Pσ,[n,)x<1normsubscript𝑃𝜎𝑛𝑥1\|P_{\sigma,[n,\infty)}x\|<1 for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and all nN𝑛𝑁n\geq N. We have that

supσ{0,1},nNPσ,[n,)xsupσ{0,1},nNPσ,[n,N)x+supσ{0,1}Pσ,[N,)x.subscriptsupremumformulae-sequence𝜎superscript01𝑛𝑁normsubscript𝑃𝜎𝑛𝑥subscriptsupremumformulae-sequence𝜎superscript01𝑛𝑁normsubscript𝑃𝜎𝑛𝑁𝑥subscriptsupremum𝜎superscript01normsubscript𝑃𝜎𝑁𝑥\sup_{\sigma\in\{0,1\}^{\mathbb{N}},n\leq N}\|P_{\sigma,[n,\infty)}x\|\leq\sup_{\sigma\in\{0,1\}^{\mathbb{N}},n\leq N}\|P_{\sigma,[n,N)}x\|+\sup_{\sigma\in\{0,1\}^{\mathbb{N}}}\|P_{\sigma,[N,\infty)}x\|.

Thus, we have

supσ{0,1},nNPσ,[n,)x<subscriptsupremumformulae-sequence𝜎superscript01𝑛𝑁normsubscript𝑃𝜎𝑛𝑥\sup_{\sigma\in\{0,1\}^{\mathbb{N}},n\leq N}\|P_{\sigma,[n,\infty)}x\|<\infty

as supσPσ,[N,)x1subscriptsupremum𝜎normsubscript𝑃𝜎𝑁𝑥1\sup_{\sigma}\|P_{\sigma,[N,\infty)}x\|\leq 1 and the set {Pσ,[n,N)x}σ{0,1},nNsubscriptsubscript𝑃𝜎𝑛𝑁𝑥formulae-sequence𝜎superscript01𝑛𝑁\{P_{\sigma,[n,N)}x\}_{\sigma\in\{0,1\}^{\mathbb{N}},n\leq N} is finite. Hence, there exists a constant K>0𝐾0K>0 such that Pσ,[n,)xKnormsubscript𝑃𝜎𝑛𝑥𝐾\|P_{\sigma,[n,\infty)}x\|\leq K for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and all n𝑛n\in\mathbb{N}. Therefore, for all m,n𝑚𝑛m,n\in\mathbb{N} with mn𝑚𝑛m\leq n we have that

Pσ,[m,n]x=Pσ,[m,)xPσ,[n+1,)x2K.normsubscript𝑃𝜎𝑚𝑛𝑥normsubscript𝑃𝜎𝑚𝑥subscript𝑃𝜎𝑛1𝑥2𝐾\|P_{\sigma,[m,n]}x\|=\|P_{\sigma,[m,\infty)}x-P_{\sigma,[n+1,\infty)}x\|\leq 2K.

Hence, {Pσ,[m,n]}σ{0,1},mnsubscriptsubscript𝑃𝜎𝑚𝑛formulae-sequence𝜎superscript01𝑚𝑛\{P_{\sigma,[m,n]}\}_{\sigma\in\{0,1\}^{\mathbb{N}},m\leq n} is a set of bounded operators such that for all xX𝑥𝑋x\in X, {Pσ,[m,n]x}subscript𝑃𝜎𝑚𝑛𝑥\{P_{\sigma,[m,n]}x\} is uniformly bounded in norm. By the uniform boundedness principle, the set of operators {Pσ,[m,n]}subscript𝑃𝜎𝑚𝑛\{P_{\sigma,[m,n]}\} is uniformly bounded in norm. ∎

Lemma 2.3.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be two approximate Schauder frames of X𝑋X such that every weaving is an approximate Schauder frame of X𝑋X. For all xX𝑥𝑋x\in X there exists δ>0𝛿0\delta>0 such that for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} we have that Pσ,xδxnormsubscript𝑃𝜎𝑥𝛿norm𝑥\|P_{\sigma,\mathbb{N}}x\|\geq\delta\|x\|

Proof.

Let xX𝑥𝑋x\in X such that x0𝑥0x\neq 0. By Lemma 2.1, for all n𝑛n\in\mathbb{N} there exists Mnsubscript𝑀𝑛M_{n}\in\mathbb{N} such that Pγ,[Mn,)x<1nnormsubscript𝑃𝛾subscript𝑀𝑛𝑥1𝑛\|P_{\gamma,[M_{n},\infty)}x\|<\frac{1}{n} for all γ{0,1}𝛾superscript01\gamma\in\{0,1\}^{\mathbb{N}}. Assume for the sake of contradiction that for all n𝑛n\in\mathbb{N} there exists σn{0,1}subscript𝜎𝑛superscript01\sigma_{n}\in\{0,1\}^{\mathbb{N}} such that Pσn,x<1/nnormsubscript𝑃subscript𝜎𝑛𝑥1𝑛\|P_{\sigma_{n},\mathbb{N}}x\|<1/n. After passing to a subsequence, we may assume that (σn)n=1superscriptsubscriptsubscript𝜎𝑛𝑛1(\sigma_{n})_{n=1}^{\infty} converges to some σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and σn|[1,Mn)=σ|[1,Mn)evaluated-atsubscript𝜎𝑛1subscript𝑀𝑛evaluated-at𝜎1subscript𝑀𝑛\sigma_{n}|_{[1,M_{n})}=\sigma|_{[1,M_{n})} for all n𝑛n\in\mathbb{N}. For each n𝑛n\in\mathbb{N}, we have the following upperbound on the norm of Pσ,xsubscript𝑃𝜎𝑥P_{\sigma,\mathbb{N}}x.

Pσ,xnormsubscript𝑃𝜎𝑥\displaystyle\|P_{\sigma,\mathbb{N}}x\| =Pσn,xPσn,[Mn,)x+Pσ,[Mn,)xabsentnormsubscript𝑃subscript𝜎𝑛𝑥subscript𝑃subscript𝜎𝑛subscript𝑀𝑛𝑥subscript𝑃𝜎subscript𝑀𝑛𝑥\displaystyle=\|P_{\sigma_{n},\mathbb{N}}x-P_{\sigma_{n},[M_{n},\infty)}x+P_{\sigma,[M_{n},\infty)}x\|
Pσn,x+Pσn,[Mn,)x+Pσ,[Mn,)xabsentnormsubscript𝑃subscript𝜎𝑛𝑥normsubscript𝑃subscript𝜎𝑛subscript𝑀𝑛𝑥normsubscript𝑃𝜎subscript𝑀𝑛𝑥\displaystyle\leq\|P_{\sigma_{n},\mathbb{N}}x\|+\|P_{\sigma_{n},[M_{n},\infty)}x\|+\|P_{\sigma,[M_{n},\infty)}x\|
<1/n+1/n+1/nabsent1𝑛1𝑛1𝑛\displaystyle<1/n+1/n+1/n

Thus, Pσ,x<3/nnormsubscript𝑃𝜎𝑥3𝑛\|P_{\sigma,\mathbb{N}}x\|<3/n for all n𝑛n\in\mathbb{N} and hence Pσ,x=0subscript𝑃𝜎𝑥0P_{\sigma,\mathbb{N}}x=0. This contradicts that (xjσ(j),fjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗subscriptsuperscript𝑓𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j},f^{\sigma(j)}_{j})_{j=1}^{\infty} is an approximate Schauder frame. ∎

Theorem 2.4.

Given two approximate Schauder frames for a Banach space X𝑋X, they are woven if and only if every weaving is an approximate Schauder frame. That is, if every weaving is an approximate Schauder frame then there is a uniform constant C1𝐶1C\geq 1 such that every weaving is a C𝐶C-approximate Schauder frame. Furthermore, if every weaving is an unconditional approximate Schauder frame then there exists a constant D1𝐷1D\geq 1 such that every weaving is a D𝐷D-unconditional approximate Schauder frame.

Proof.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be two approximate Schauder frames of X𝑋X such that every weaving is an approximate Schauder frame. Let yX𝑦𝑋y\in X. By Lemma 2.3, there exists δ>0𝛿0\delta>0 such that y=Pσ,(Pσ,1y)δPσ,1ynorm𝑦normsubscript𝑃𝜎superscriptsubscript𝑃𝜎1𝑦𝛿normsuperscriptsubscript𝑃𝜎1𝑦\|y\|=\|P_{\sigma,\mathbb{N}}(P_{\sigma,\mathbb{N}}^{-1}y)\|\geq\delta\|P_{\sigma,\mathbb{N}}^{-1}y\| for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. Thus, Pσ,1y(1/δ)ynormsuperscriptsubscript𝑃𝜎1𝑦1𝛿norm𝑦\|P_{\sigma,\mathbb{N}}^{-1}y\|\leq(1/\delta)\|y\| for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. We have that {Pσ,1}σ{0,1}subscriptsubscriptsuperscript𝑃1𝜎𝜎superscript01\{P^{-1}_{\sigma,\mathbb{N}}\}_{\sigma\in\{0,1\}^{\mathbb{N}}} is a collection of bounded operators such that supσ{0,1}Pσ,1x<subscriptsupremum𝜎superscript01normsubscriptsuperscript𝑃1𝜎𝑥\sup_{\sigma\in\{0,1\}^{\mathbb{N}}}\|P^{-1}_{\sigma,\mathbb{N}}x\|<\infty for all xX𝑥𝑋x\in X. Hence, supσ{0,1}Pσ,1<subscriptsupremum𝜎superscript01normsubscriptsuperscript𝑃1𝜎\sup_{\sigma\in\{0,1\}^{\mathbb{N}}}\|P^{-1}_{\sigma,\mathbb{N}}\|<\infty by the uniform boundedness principle. Also, supσ{0,1}Pσ,<subscriptsupremum𝜎superscript01normsubscript𝑃𝜎\sup_{\sigma\in\{0,1\}^{\mathbb{N}}}\|P_{\sigma,\mathbb{N}}\|<\infty by Lemma 2.2. Thus, there is a uniform constant C1𝐶1C\geq 1 such that Pσ,Cnormsubscript𝑃𝜎𝐶\|P_{\sigma,\mathbb{N}}\|\leq C and Pσ,1Cnormsuperscriptsubscript𝑃𝜎1𝐶\|P_{\sigma,\mathbb{N}}^{-1}\|\leq C for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and hence (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} are C𝐶C-woven approximate Schauder frames of X𝑋X.

We now assume that every weaving of (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} is an unconditional approximate Schauder frame. We have that there exists C1𝐶1C\geq 1 such that every weaving of (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} is a C𝐶C-approximate Schauder frame. For the sake of contradiction we assume that for all n𝑛n\in\mathbb{N}, there exists σn{0,1}subscript𝜎𝑛superscript01\sigma_{n}\in\{0,1\}^{\mathbb{N}} such that the approximate Schauder (xjσn(j),fjσn(j))j=1superscriptsubscriptsubscriptsuperscript𝑥subscript𝜎𝑛𝑗𝑗subscriptsuperscript𝑓subscript𝜎𝑛𝑗𝑗𝑗1(x^{\sigma_{n}(j)}_{j},f^{\sigma_{n}(j)}_{j})_{j=1}^{\infty} is not n𝑛n-suppression unconditional. After passing to a subsequence, we may assume that there exists σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} such that σnσsubscript𝜎𝑛𝜎\sigma_{n}\rightarrow\sigma in the product topology of {0,1}superscript01\{0,1\}^{\mathbb{N}} and that σn|[1,n]=σ|[1,n]evaluated-atsubscript𝜎𝑛1𝑛evaluated-at𝜎1𝑛\sigma_{n}|_{[1,n]}=\sigma|_{[1,n]} for all n𝑛n\in\mathbb{N}. We choose a subsequence (mn)n=1superscriptsubscriptsubscript𝑚𝑛𝑛1(m_{n})_{n=1}^{\infty} of \mathbb{N} such that there exists a finite set Γn[1,mn]subscriptΓ𝑛1subscript𝑚𝑛\Gamma_{n}\subseteq[1,m_{n}] and xnXsubscript𝑥𝑛𝑋x_{n}\in X with xn=1normsubscript𝑥𝑛1\|x_{n}\|=1 so that Pσn,ΓnxnnPσn,xnnC1normsubscript𝑃subscript𝜎𝑛subscriptΓ𝑛subscript𝑥𝑛𝑛normsubscript𝑃subscript𝜎𝑛subscript𝑥𝑛𝑛superscript𝐶1\|P_{\sigma_{n},\Gamma_{n}}x_{n}\|\geq n\|P_{\sigma_{n},\mathbb{N}}x_{n}\|\geq nC^{-1}.

We inductively define a sequence of natural numbers (Mn)n=1superscriptsubscriptsubscript𝑀𝑛𝑛1(M_{n})_{n=1}^{\infty} by M1=m1subscript𝑀1subscript𝑚1M_{1}=m_{1} and for n𝑛n\in\mathbb{N} we let Mn+1=mMnsubscript𝑀𝑛1subscript𝑚subscript𝑀𝑛M_{n+1}=m_{M_{n}}. Let γ{0,1}𝛾superscript01\gamma\in\{0,1\}^{\mathbb{N}} be such that γ|[1,m1]=σ1|[1,m1]evaluated-at𝛾1subscript𝑚1evaluated-atsubscript𝜎11subscript𝑚1\gamma|_{[1,m_{1}]}=\sigma_{1}|_{[1,m_{1}]} and for n𝑛n\in\mathbb{N}, γ|(Mn,Mn+1]=σMn|(Mn,Mn+1]evaluated-at𝛾subscript𝑀𝑛subscript𝑀𝑛1evaluated-atsubscript𝜎subscript𝑀𝑛subscript𝑀𝑛subscript𝑀𝑛1\gamma|_{(M_{n},M_{n+1}]}=\sigma_{M_{n}}|_{(M_{n},M_{n+1}]}. Let D1𝐷1D\geq 1 be the suppression unconditionality constant of the weaving (xjσ(j),fjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗subscriptsuperscript𝑓𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j},f^{\sigma(j)}_{j})_{j=1}^{\infty}. For n𝑛n\in\mathbb{N}, we have that

Pγ,ΓMn(Mn,Mn+1]normsubscript𝑃𝛾subscriptΓsubscript𝑀𝑛subscript𝑀𝑛subscript𝑀𝑛1\displaystyle\|P_{\gamma,\Gamma_{M_{n}}\cap(M_{n},M_{n+1}]}\| =PσMn,ΓMn(Mn,Mn+1]absentnormsubscript𝑃subscript𝜎subscript𝑀𝑛subscriptΓsubscript𝑀𝑛subscript𝑀𝑛subscript𝑀𝑛1\displaystyle=\|P_{\sigma_{M_{n}},\Gamma_{M_{n}}\cap(M_{n},M_{n+1}]}\|
PσMn,ΓMnPσMn,ΓMn[1,Mn]absentnormsubscript𝑃subscript𝜎subscript𝑀𝑛subscriptΓsubscript𝑀𝑛normsubscript𝑃subscript𝜎subscript𝑀𝑛subscriptΓsubscript𝑀𝑛1subscript𝑀𝑛\displaystyle\geq\|P_{\sigma_{M_{n}},\Gamma_{M_{n}}}\|-\|P_{\sigma_{M_{n}},\Gamma_{M_{n}}\cap[1,M_{n}]}\|
=PσMn,ΓMnPσ,ΓMn[1,Mn]MnC1DC.absentnormsubscript𝑃subscript𝜎subscript𝑀𝑛subscriptΓsubscript𝑀𝑛normsubscript𝑃𝜎subscriptΓsubscript𝑀𝑛1subscript𝑀𝑛subscript𝑀𝑛superscript𝐶1𝐷𝐶\displaystyle=\|P_{\sigma_{M_{n}},\Gamma_{M_{n}}}\|-\|P_{\sigma,\Gamma_{M_{n}}\cap[1,M_{n}]}\|\geq M_{n}C^{-1}-DC.

Thus, we have that supnPγ,ΓMn(Mn,Mn+1]=subscriptsupremum𝑛normsubscript𝑃𝛾subscriptΓsubscript𝑀𝑛subscript𝑀𝑛subscript𝑀𝑛1\sup_{n\in\mathbb{N}}\|P_{\gamma,\Gamma_{M_{n}}\cap(M_{n},M_{n+1}]}\|=\infty and hence (xjγ(j),fjγ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝛾𝑗𝑗subscriptsuperscript𝑓𝛾𝑗𝑗𝑗1(x^{\gamma(j)}_{j},f^{\gamma(j)}_{j})_{j=1}^{\infty} is not unconditional.

3. Weaving bases

We now consider the case of weaving Schauder bases instead of approximate Schauder frames. Recall that if (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} is a Schauder basic sequence then we say that it is C𝐶C-basic for some C1𝐶1C\geq 1 if j=1NajxjCj=1ajxjnormsuperscriptsubscript𝑗1𝑁subscript𝑎𝑗subscript𝑥𝑗𝐶normsuperscriptsubscript𝑗1subscript𝑎𝑗subscript𝑥𝑗\|\sum_{j=1}^{N}a_{j}x_{j}\|\leq C\|\sum_{j=1}^{\infty}a_{j}x_{j}\| for all N𝑁N\in\mathbb{N} and all sequences of scalars (aj)j=1superscriptsubscriptsubscript𝑎𝑗𝑗1(a_{j})_{j=1}^{\infty} such that j=1aixisuperscriptsubscript𝑗1subscript𝑎𝑖subscript𝑥𝑖\sum_{j=1}^{\infty}a_{i}x_{i} converges. We say that an unconditional Schauder basic sequence is C𝐶C-suppression unconditional if jEajxjCj=1ajxjnormsubscript𝑗𝐸subscript𝑎𝑗subscript𝑥𝑗𝐶normsuperscriptsubscript𝑗1subscript𝑎𝑗subscript𝑥𝑗\|\sum_{j\in E}a_{j}x_{j}\|\leq C\|\sum_{j=1}^{\infty}a_{j}x_{j}\| for all E𝐸E\subseteq\mathbb{N} and all sequences of scalars (aj)j=1superscriptsubscriptsubscript𝑎𝑗𝑗1(a_{j})_{j=1}^{\infty} such that j=1ajxjsuperscriptsubscript𝑗1subscript𝑎𝑗subscript𝑥𝑗\sum_{j=1}^{\infty}a_{j}x_{j} converges.

In Section 2 we proved that if every weaving of two approximate Schauder frames is an approximate Schauder frame then there is a uniform constant C1𝐶1C\geq 1 such that the two approximate Schauder frames are C𝐶C-woven. We now prove that this same uniformity theorem holds for weaving Schauder bases. The proof will follow the same concatonation argument as in the furthermore part of the proof of Theorem 2.4.

Proposition 3.1.

Let (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} be two Schauder basic sequences in a Banach space X𝑋X. If (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is a basic sequence for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, then there is a uniform constant C1𝐶1C\geq 1 such that for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is C𝐶C-basic. Likewise, if (xjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j})_{j=1}^{\infty} is an unconditional basic sequence for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, then there is a uniform constant D1𝐷1D\geq 1 such that for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j})_{j=1}^{\infty} is D𝐷D-unconditional.

Proof.

Assume by way of contradiction that (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is a basic sequence for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, but that for all n𝑛n\in\mathbb{N}, there exists σn{0,1}subscript𝜎𝑛superscript01\sigma_{n}\in\{0,1\}^{\mathbb{N}} such that (xiσn(i))i=1superscriptsubscriptsubscriptsuperscript𝑥subscript𝜎𝑛𝑖𝑖𝑖1(x^{\sigma_{n}(i)}_{i})_{i=1}^{\infty} is not n𝑛n-basic. After passing to a subsequence, we may assume that there exists σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} such that σnσsubscript𝜎𝑛𝜎\sigma_{n}\rightarrow\sigma in the product topology of {0,1}superscript01\{0,1\}^{\mathbb{N}}. Let C1𝐶1C\geq 1 be such that (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is C𝐶C-basic. After passing to a further subsequence, we may assume that σn|[1,n]=σ|[1,n]evaluated-atsubscript𝜎𝑛1𝑛evaluated-at𝜎1𝑛\sigma_{n}|_{[1,n]}=\sigma|_{[1,n]} for all n𝑛n\in\mathbb{N}. Thus, for all n𝑛n\in\mathbb{N}, (xiσn(i))i=1nsuperscriptsubscriptsuperscriptsubscript𝑥𝑖subscript𝜎𝑛𝑖𝑖1𝑛(x_{i}^{\sigma_{n}(i)})_{i=1}^{n} is C𝐶C-basic, and hence (xiσn(i))i=n+1superscriptsubscriptsuperscriptsubscript𝑥𝑖subscript𝜎𝑛𝑖𝑖𝑛1(x_{i}^{\sigma_{n}(i)})_{i=n+1}^{\infty} is not (nC)𝑛𝐶(n-C)-basic. Let (mn)n=1superscriptsubscriptsubscript𝑚𝑛𝑛1(m_{n})_{n=1}^{\infty} be an increasing subsequence of natural numbers such that for all n𝑛n\in\mathbb{N}, (xiσn(i))i=n+1mnsuperscriptsubscriptsuperscriptsubscript𝑥𝑖subscript𝜎𝑛𝑖𝑖𝑛1subscript𝑚𝑛(x_{i}^{\sigma_{n}(i)})_{i=n+1}^{m_{n}} is not (nC)𝑛𝐶(n-C)-basic. We can now create γ{0,1}𝛾superscript01\gamma\in\{0,1\}^{\mathbb{N}} by γ|[1,m1]=σ1|[1,m1]evaluated-at𝛾1subscript𝑚1evaluated-atsubscript𝜎11subscript𝑚1\gamma|_{[1,m_{1}]}=\sigma_{1}|_{[1,m_{1}]}, γ|(m1,mm1]=σm1|(m1,mm1]evaluated-at𝛾subscript𝑚1subscript𝑚subscript𝑚1evaluated-atsubscript𝜎subscript𝑚1subscript𝑚1subscript𝑚subscript𝑚1\gamma|_{(m_{1},m_{m_{1}}]}=\sigma_{m_{1}}|_{(m_{1},m_{m_{1}}]}, γ|(mm1,mmm1]=σmm1|(mm1,mmm1]evaluated-at𝛾subscript𝑚subscript𝑚1subscript𝑚subscript𝑚subscript𝑚1evaluated-atsubscript𝜎subscript𝑚subscript𝑚1subscript𝑚subscript𝑚1subscript𝑚subscript𝑚subscript𝑚1\gamma|_{(m_{m_{1}},m_{m_{m_{1}}}]}=\sigma_{m_{m_{1}}}|_{(m_{m_{1}},m_{m_{m_{1}}}]}, and so on. We have that (xiγ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝛾𝑖𝑖𝑖1(x^{\gamma(i)}_{i})_{i=1}^{\infty} is not basic because it is formed by concatonating finite length basic sequences whose basis constants go to infinity, which is a contradiction.

A similar argument shows that if for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is an unconditional basic sequence then there is a universal constant D1𝐷1D\geq 1 such that for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is D𝐷D-unconditional. ∎

Suppose that (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are two Schauder bases for a Banach space X𝑋X with biorthogonal functionals (xi0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑗1(x^{0*}_{i})_{j=1}^{\infty} and (xi1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑗1(x^{1*}_{i})_{j=1}^{\infty}. We can consider the problem of whether or not (xi0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑗1(x^{0}_{i})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} are woven Schauder bases, and we can consider the problem of whether or not (xj0,xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j},x^{0*}_{j})_{j=1}^{\infty} and (xj1,xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j},x^{1*}_{j})_{j=1}^{\infty} are woven approximate Schauder frames. The first natural question to ask is if these two notions are equivalent. In other words, are (xi0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑗1(x^{0}_{i})_{j=1}^{\infty} and (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} woven Schauder bases if and only if (xj0,xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j},x^{0*}_{j})_{j=1}^{\infty} and (xj1,xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j},x^{1*}_{j})_{j=1}^{\infty} are woven approximate Schauder frames? Surprisingly, the answer is no.

Example 3.2.

We have the following examples in c0subscript𝑐0c_{0} and 1subscript1\ell_{1}.

  1. (1)

    Let (ej)j=1superscriptsubscriptsubscript𝑒𝑗𝑗1(e_{j})_{j=1}^{\infty} be the standard unit vector basis for c0subscript𝑐0c_{0} with biorthogonal functionals (ej)j=1superscriptsubscriptsubscriptsuperscript𝑒𝑗𝑗1(e^{*}_{j})_{j=1}^{\infty}, and let (sj)j=1superscriptsubscriptsubscript𝑠𝑗𝑗1(s_{j})_{j=1}^{\infty} be the summing basis for c0subscript𝑐0c_{0} with biorthogonal functionals (sj)j=1superscriptsubscriptsubscriptsuperscript𝑠𝑗𝑗1(s^{*}_{j})_{j=1}^{\infty}. That is, sj=i=1jeisubscript𝑠𝑗superscriptsubscript𝑖1𝑗subscript𝑒𝑖s_{j}=\sum_{i=1}^{j}e_{i} for all j𝑗j\in\mathbb{N}. Then, (sj)j=1superscriptsubscriptsubscript𝑠𝑗𝑗1(s_{j})_{j=1}^{\infty} is a conditional Schauder basis for c0subscript𝑐0c_{0} which is woven with the unconditional Schauder basis (ej)j=1superscriptsubscriptsubscript𝑒𝑗𝑗1(e_{j})_{j=1}^{\infty}, but (sj,sj)j=1superscriptsubscriptsubscript𝑠𝑗subscriptsuperscript𝑠𝑗𝑗1(s_{j},s^{*}_{j})_{j=1}^{\infty} and (ej,ej)j=1superscriptsubscriptsubscript𝑒𝑗subscriptsuperscript𝑒𝑗𝑗1(e_{j},e^{*}_{j})_{j=1}^{\infty} are not woven approximate Schauder frames.

  2. (2)

    Let (ej)j=1superscriptsubscriptsubscript𝑒𝑗𝑗1(e_{j})_{j=1}^{\infty} be the standard unit vector basis for 1subscript1\ell_{1} with biorthogonal functionals (ej)j=1superscriptsubscriptsubscriptsuperscript𝑒𝑗𝑗1(e^{*}_{j})_{j=1}^{\infty}. Let x1=e1subscript𝑥1subscript𝑒1x_{1}=e_{1} and xn=enen1subscript𝑥𝑛subscript𝑒𝑛subscript𝑒𝑛1x_{n}=e_{n}-e_{n-1} for n>1𝑛1n>1. Then, (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} is a conditional Schauder basis for 1subscript1\ell_{1} which is woven with the unconditional Schauder basis (ej)j=1superscriptsubscriptsubscript𝑒𝑗𝑗1(e_{j})_{j=1}^{\infty}, but (xj,xj)j=1superscriptsubscriptsubscript𝑥𝑗subscriptsuperscript𝑥𝑗𝑗1(x_{j},x^{*}_{j})_{j=1}^{\infty} and (ej,ej)j=1superscriptsubscriptsubscript𝑒𝑗subscriptsuperscript𝑒𝑗𝑗1(e_{j},e^{*}_{j})_{j=1}^{\infty} are not woven approximate Schauder frames, where (xj)j=1superscriptsubscriptsubscriptsuperscript𝑥𝑗𝑗1(x^{*}_{j})_{j=1}^{\infty} is the sequence of biorthogonal functionals to (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty}.

Example 3.2 shows that it is possible for a conditional Schauder basis to be woven with an unconditional Schauder basis. The following proposition shows that this is impossible for woven approximate Schauder frames.

Proposition 3.3.

Let (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} be C𝐶C-approximate Schauder frames for a Banach space X𝑋X. Suppose that (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} is K𝐾K-suppression unconditional and that (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} is not D𝐷D-suppression unconditional. Then (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} are not (DC1KC)𝐷superscript𝐶1𝐾𝐶(DC^{-1}-KC)-woven.

Proof.

Let S0subscript𝑆0S_{0} be the frame operator for (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and let S1subscript𝑆1S_{1} be the frame operator for (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty}. Let xX𝑥𝑋x\in X and A𝐴A\subset\mathbb{N} such that jAfj1(x)xj1>DS1xDC1xnormsubscript𝑗𝐴subscriptsuperscript𝑓1𝑗𝑥subscriptsuperscript𝑥1𝑗𝐷normsubscript𝑆1𝑥𝐷superscript𝐶1norm𝑥\|\sum_{j\in A}f^{1}_{j}(x)x^{1}_{j}\|>D\|S_{1}x\|\geq DC^{-1}\|x\|. Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} be the characteristic function of A𝐴A. Then

jfjσ(j)(x)xjσ(j)normsubscript𝑗superscriptsubscript𝑓𝑗𝜎𝑗𝑥superscriptsubscript𝑥𝑗𝜎𝑗\displaystyle\bigg{\|}\sum_{j\in\mathbb{N}}f_{j}^{\sigma(j)}(x)x_{j}^{\sigma(j)}\bigg{\|} jAfj1(x)xj1jAfj0(x)xj0absentnormsubscript𝑗𝐴subscriptsuperscript𝑓1𝑗𝑥subscriptsuperscript𝑥1𝑗normsubscript𝑗𝐴subscriptsuperscript𝑓0𝑗𝑥subscriptsuperscript𝑥0𝑗\displaystyle\geq\bigg{\|}\sum_{j\in A}f^{1}_{j}(x)x^{1}_{j}\bigg{\|}-\bigg{\|}\sum_{j\not\in A}f^{0}_{j}(x)x^{0}_{j}\bigg{\|}
>DS1xKS0xabsent𝐷normsubscript𝑆1𝑥𝐾normsubscript𝑆0𝑥\displaystyle>D\|S_{1}x\|-K\|S_{0}x\|
DC1xKCx.absent𝐷superscript𝐶1norm𝑥𝐾𝐶norm𝑥\displaystyle\geq DC^{-1}\|x\|-KC\|x\|.

Thus the frame operator of (xjσ(j),fjσ(j))j=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑗𝑗subscriptsuperscript𝑓𝜎𝑗𝑗𝑗1(x^{\sigma(j)}_{j},f^{\sigma(j)}_{j})_{j=1}^{\infty} has norm greater than (DC1KC)𝐷superscript𝐶1𝐾𝐶(DC^{-1}-KC) and hence (xj0,fj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑓0𝑗𝑗1(x^{0}_{j},f^{0}_{j})_{j=1}^{\infty} and (xj1,fj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑓1𝑗𝑗1(x^{1}_{j},f^{1}_{j})_{j=1}^{\infty} are not (DC1KC)𝐷superscript𝐶1𝐾𝐶(DC^{-1}-KC)-woven. ∎

We conclude the section with two examples that again demonstrate the significance of conditional compared to unconditional convergence. The next construction shows that it is possible for two woven unconditional Schauder bases to have some conditional weavings.

Example 3.4.

Let (ej)j=1superscriptsubscriptsubscript𝑒𝑗𝑗1(e_{j})_{j=1}^{\infty} be the standard unit vector basis for 1subscript1\ell_{1}. Define (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} by x2n10=ensuperscriptsubscript𝑥2𝑛10subscript𝑒𝑛x_{2n-1}^{0}=e_{n} and x2n0=e2ne2n1subscriptsuperscript𝑥02𝑛subscript𝑒2𝑛subscript𝑒2𝑛1x^{0}_{2n}=e_{2n}-e_{2n-1} for all n𝑛n\in\mathbb{N} and define (xi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1𝑖1(x_{i}^{1})_{i=1}^{\infty} by x11=e1superscriptsubscript𝑥11subscript𝑒1x_{1}^{1}=e_{1}, x21=e2superscriptsubscript𝑥21subscript𝑒2x_{2}^{1}=e_{2}, and x2n11=e2n1e2n2superscriptsubscript𝑥2𝑛11subscript𝑒2𝑛1subscript𝑒2𝑛2x_{2n-1}^{1}=e_{2n-1}-e_{2n-2} and x2n1=e2nsuperscriptsubscript𝑥2𝑛1subscript𝑒2𝑛x_{2n}^{1}=e_{2n} for all n2𝑛2n\geq 2. Then (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1𝑖1(x_{i}^{1})_{i=1}^{\infty} are woven Schauder bases, each of which are unconditional. However, the alternating weaving given by (x11,x20,x31,x40,)=(e1,e2e1,e3e2,e4e3,)subscriptsuperscript𝑥11subscriptsuperscript𝑥02subscriptsuperscript𝑥13subscriptsuperscript𝑥04subscript𝑒1subscript𝑒2subscript𝑒1subscript𝑒3subscript𝑒2subscript𝑒4subscript𝑒3(x^{1}_{1},x^{0}_{2},x^{1}_{3},x^{0}_{4},...)=(e_{1},e_{2}-e_{1},e_{3}-e_{2},e_{4}-e_{3},...) is the same conditional basis used in Example 3.2.

The following construction shows it is possible for all the weavings of two unconditional Schauder bases to be Schauder basic sequences, but there exists a weaving which does not have dense span. In [BCGLL], it is proven that this is impossible for Riesz bases for Hilbert spaces, and in Theorem 4.5 we prove that this is impossible if all weavings are unconditional.

Example 3.5.

Let (ej)j=1superscriptsubscriptsubscript𝑒𝑗𝑗1(e_{j})_{j=1}^{\infty} be the standard unit vector basis for 1subscript1\ell_{1}. Define (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} by x2n10=e2n1+e2nsuperscriptsubscript𝑥2𝑛10subscript𝑒2𝑛1subscript𝑒2𝑛x_{2n-1}^{0}=e_{2n-1}+e_{2n} and x2n0=e2n1e2nsuperscriptsubscript𝑥2𝑛0subscript𝑒2𝑛1subscript𝑒2𝑛x_{2n}^{0}=e_{2n-1}-e_{2n} for all n𝑛n\in\mathbb{N} and define (xi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1𝑖1(x_{i}^{1})_{i=1}^{\infty} by x11=e1superscriptsubscript𝑥11subscript𝑒1x_{1}^{1}=e_{1} and x2n1=e2n+e2n+1superscriptsubscript𝑥2𝑛1subscript𝑒2𝑛subscript𝑒2𝑛1x_{2n}^{1}=e_{2n}+e_{2n+1} and x2n+11=e2ne2n+1superscriptsubscript𝑥2𝑛11subscript𝑒2𝑛subscript𝑒2𝑛1x_{2n+1}^{1}=e_{2n}-e_{2n+1} for all n𝑛n\in\mathbb{N}. Then (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1𝑖1(x_{i}^{1})_{i=1}^{\infty} are both unconditional Schauder bases of 1subscript1\ell_{1} and every weaving is a Schauder basic sequence. However, the alternating weaving (x10,x21,x30,x41,)=(e1+e2,e2+e3,e3+e4,e4+e5,)subscriptsuperscript𝑥01subscriptsuperscript𝑥12subscriptsuperscript𝑥03subscriptsuperscript𝑥14subscript𝑒1subscript𝑒2subscript𝑒2subscript𝑒3subscript𝑒3subscript𝑒4subscript𝑒4subscript𝑒5(x^{0}_{1},x^{1}_{2},x^{0}_{3},x^{1}_{4},...)=(e_{1}+e_{2},e_{2}+e_{3},e_{3}+e_{4},e_{4}+e_{5},...) does not contain e1subscript𝑒1e_{1} in its closed span and thus is not a basis.

4. Weaving unconditional Schauder bases

In this section, we consider weaving unconditional bases. First, we show that two woven unconditional bases are necessarily equivalent. We conclude the section by proving that two unconditional bases that are woven and have all unconditional weavings is equivalent to five other characterizations, including in particular that all weavings are unconditional basic sequences and all weavings of the corresponding approximate Schauder frames are unconditional approximate Schauder frames. Recall that in Section 3 we gave counterexamples to these equivalences when it is not the case that all weavings are unconditional. We first cite the following theorem, which can be found in [LT] along with a thorough introduction to Banach lattices.

Theorem 4.1.

Let (xi)i=1superscriptsubscriptsubscript𝑥𝑖𝑖1(x_{i})_{i=1}^{\infty} be an unconditional basis for a Banach lattice X𝑋X. There exists a constant D1𝐷1D\geq 1 such that for every n𝑛n\in\mathbb{N} and sequence of scalars (ai)i=1nsuperscriptsubscriptsubscript𝑎𝑖𝑖1𝑛(a_{i})_{i=1}^{n}, we have that

D1(i=1n|aixi|2)1/2i=1naixiD(i=1n|aixi|2)1/2.superscript𝐷1normsuperscriptsuperscriptsubscript𝑖1𝑛superscriptsubscript𝑎𝑖subscript𝑥𝑖212normsuperscriptsubscript𝑖1𝑛subscript𝑎𝑖subscript𝑥𝑖𝐷normsuperscriptsuperscriptsubscript𝑖1𝑛superscriptsubscript𝑎𝑖subscript𝑥𝑖212D^{-1}\left\|\left(\sum_{i=1}^{n}|a_{i}x_{i}|^{2}\right)^{1/2}\right\|\leq\left\|\sum_{i=1}^{n}a_{i}x_{i}\right\|\leq D\left\|\left(\sum_{i=1}^{n}|a_{i}x_{i}|^{2}\right)^{1/2}\right\|.

When we apply Theorem 4.1, the Banach lattice structure for X𝑋X will be defined by a different 1-unconditional basis (ui)i=1superscriptsubscriptsubscript𝑢𝑖𝑖1(u_{i})_{i=1}^{\infty} with biorthogonal functionals (ui)i=1superscriptsubscriptsubscriptsuperscript𝑢𝑖𝑖1(u^{*}_{i})_{i=1}^{\infty}. In this case, if (xi)i=1superscriptsubscriptsubscript𝑥𝑖𝑖1(x_{i})_{i=1}^{\infty} is a sequence in X𝑋X, and (ai)i=1nsuperscriptsubscriptsubscript𝑎𝑖𝑖1𝑛(a_{i})_{i=1}^{n} is a finite sequence of scalars then the vector (i=1n|aixi|2)1/2superscriptsuperscriptsubscript𝑖1𝑛superscriptsubscript𝑎𝑖subscript𝑥𝑖212\left(\sum_{i=1}^{n}|a_{i}x_{i}|^{2}\right)^{1/2} is given by

(i=1n|aixi|2)1/2=j=1(i=1n|aiuj(xi)|2)1/2uj.superscriptsuperscriptsubscript𝑖1𝑛superscriptsubscript𝑎𝑖subscript𝑥𝑖212superscriptsubscript𝑗1superscriptsuperscriptsubscript𝑖1𝑛superscriptsubscript𝑎𝑖subscriptsuperscript𝑢𝑗subscript𝑥𝑖212subscript𝑢𝑗\left(\sum_{i=1}^{n}|a_{i}x_{i}|^{2}\right)^{1/2}=\sum_{j=1}^{\infty}\left(\sum_{i=1}^{n}|a_{i}u^{*}_{j}(x_{i})|^{2}\right)^{1/2}\!\!\!u_{j}. (2)

We have previously given an example of a conditional Schauder basis for 1subscript1\ell_{1} such that every weaving with the unit vector basis for 1subscript1\ell_{1} is a Schauder basis. In particular, this gives an example of two non-equivalent Schauder bases for a Banach space X𝑋X such that every weaving is a Schauder basis for X𝑋X. Our next result shows that this is impossible if both bases are unconditional. Recall that two Schauder basic sequences (xj)j=1superscriptsubscriptsubscript𝑥𝑗𝑗1(x_{j})_{j=1}^{\infty} and (yj)j=1superscriptsubscriptsubscript𝑦𝑗𝑗1(y_{j})_{j=1}^{\infty} are said to be equivalent if there are constants c,C>0𝑐𝐶0c,C>0 such that cajxjajyjCajxj𝑐normsubscript𝑎𝑗subscript𝑥𝑗normsubscript𝑎𝑗subscript𝑦𝑗𝐶normsubscript𝑎𝑗subscript𝑥𝑗c\|\sum a_{j}x_{j}\|\leq\|\sum a_{j}y_{j}\|\leq C\|\sum a_{j}x_{j}\| for all (aj)j=1c00superscriptsubscriptsubscript𝑎𝑗𝑗1subscript𝑐00(a_{j})_{j=1}^{\infty}\in c_{00}.

Theorem 4.2.

Let (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} be two semi-normalized unconditional bases for a Banach space X𝑋X such that for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i})_{i=1}^{\infty} is a basic sequence. Then, (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} are equivalent.

Proof.

After renorming, we may assume that (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} is 1-unconditional. Let X𝑋X have the Banach lattice structure defined by the 1-unconditional basis (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty}. Let D𝐷D be the constant given in Theorem 4.1. As every weaving of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} is a basic sequence, there is a constant C>0𝐶0C>0 such that |xi1(xi0)|Csubscriptsuperscript𝑥1𝑖subscriptsuperscript𝑥0𝑖𝐶|x^{1*}_{i}(x^{0}_{i})|\geq C for all i𝑖i\in\mathbb{N}. For any (ai)i=1c00superscriptsubscriptsubscript𝑎𝑖𝑖1subscript𝑐00(a_{i})_{i=1}^{\infty}\in c_{00}, we have the following estimate:

i=1aixi0normsuperscriptsubscript𝑖1subscript𝑎𝑖subscriptsuperscript𝑥0𝑖\displaystyle\left\|\sum_{i=1}^{\infty}a_{i}x^{0}_{i}\right\| D1(i=1|aixi0|2)1/2 by Theorem 4.1,absentsuperscript𝐷1normsuperscriptsuperscriptsubscript𝑖1superscriptsubscript𝑎𝑖subscriptsuperscript𝑥0𝑖212 by Theorem 4.1\displaystyle\geq D^{-1}\left\|\left(\sum_{i=1}^{\infty}|a_{i}x^{0}_{i}|^{2}\right)^{1/2}\right\|\quad\textrm{ by Theorem }\ref{square},
=D1j=1(i=1|aixj1(xi0)|2)1/2xj1 by (2),absentsuperscript𝐷1normsuperscriptsubscript𝑗1superscriptsuperscriptsubscript𝑖1superscriptsubscript𝑎𝑖subscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥0𝑖212subscriptsuperscript𝑥1𝑗 by italic-(2italic-)\displaystyle=D^{-1}\left\|\sum_{j=1}^{\infty}\left(\sum_{i=1}^{\infty}|a_{i}x^{1*}_{j}(x^{0}_{i})|^{2}\right)^{1/2}x^{1}_{j}\right\|\quad\textrm{ by }\eqref{E:lattice},
CD1j=1ajxj1 as (xj1) is 1-unconditional and |xj1(xj0)|C for all j.formulae-sequenceabsent𝐶superscript𝐷1normsuperscriptsubscript𝑗1subscript𝑎𝑗subscriptsuperscript𝑥1𝑗 as subscriptsuperscript𝑥1𝑗 is 1-unconditional and subscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥0𝑗𝐶 for all 𝑗\displaystyle\geq CD^{-1}\left\|\sum_{j=1}^{\infty}a_{j}x^{1}_{j}\right\|\quad\textrm{ as }(x^{1}_{j})\textrm{ is }1\textrm{-unconditional and }|x^{1*}_{j}(x^{0}_{j})|\geq C\textrm{ for all }j\in\mathbb{N}.

Therefore, we have that the basis (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} dominates the basis (x1)i=1superscriptsubscriptsuperscript𝑥1𝑖1(x^{1})_{i=1}^{\infty}, and by the same argument (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} dominates (x0)i=1superscriptsubscriptsuperscript𝑥0𝑖1(x^{0})_{i=1}^{\infty}. Hence, (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} and (x1)i=1superscriptsubscriptsuperscript𝑥1𝑖1(x^{1})_{i=1}^{\infty} are equivalent. ∎

Note that in proving Theorem 4.2 it was only required that there exists C>0𝐶0C>0 such that |xj0(xj1)|Csubscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥1𝑗𝐶|x^{0*}_{j}(x^{1}_{j})|\geq C and |xj1(xj0)|Csubscriptsuperscript𝑥1𝑗subscriptsuperscript𝑥0𝑗𝐶|x^{1*}_{j}(x^{0}_{j})|\geq C for j𝑗j\in\mathbb{N} which is a much weaker property than (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} are woven.

The following lemmas will be used in characterizing when every weaving of two unconditional Schauder bases is also an unconditional Schauder basis.

Lemma 4.3.

Suppose X𝑋X is a Banach space having subspaces Y𝑌Y and Z𝑍Z. Let P:XY:𝑃𝑋𝑌P:X\to Y be a projection from X𝑋X onto Y𝑌Y and assume P|Z:ZY:evaluated-at𝑃𝑍𝑍𝑌P|_{Z}:Z\to Y is invertible. Then Q=(P|Z)1P𝑄superscriptevaluated-at𝑃𝑍1𝑃Q=(P|_{Z})^{-1}P is a projection onto Z𝑍Z.

Proof.

The fact that Q𝑄Q is onto Z𝑍Z is clear. We check Q𝑄Q is a projection:

Q2=(P|Z)1(P(P|Z)1P)=(P|Z)1P=Qsuperscript𝑄2superscriptevaluated-at𝑃𝑍1𝑃superscriptevaluated-at𝑃𝑍1𝑃superscriptevaluated-at𝑃𝑍1𝑃𝑄\displaystyle Q^{2}=(P|_{Z})^{-1}\big{(}P(P|_{Z})^{-1}P\big{)}=(P|_{Z})^{-1}P=Q

since if xX𝑥𝑋x\in X, then PxY𝑃𝑥𝑌Px\in Y and so P(P|Z)1Px=Px𝑃superscriptevaluated-at𝑃𝑍1𝑃𝑥𝑃𝑥P(P|_{Z})^{-1}Px=Px. ∎

For a Banach space X𝑋X with subspaces X0,X1Xsubscript𝑋0subscript𝑋1𝑋X_{0},X_{1}\subseteq X, the distance between X0subscript𝑋0X_{0} and X1subscript𝑋1X_{1} measures how close a point in the sphere of one space can be to a point in the other space. That is,

d(X0,X1)=inf{xy:xSXj,yX1j and j{0,1}.}d(X_{0},X_{1})=\inf\{\|x-y\|\,:\,x\in S_{X_{j}},y\in X_{1-j}\textrm{ and }j\in\{0,1\}.\}
Lemma 4.4.

Suppose X𝑋X is a Banach space and X=X1X2=Y1Y2𝑋direct-sumsubscript𝑋1subscript𝑋2direct-sumsubscript𝑌1subscript𝑌2X=X_{1}\oplus X_{2}=Y_{1}\oplus Y_{2} for some subspaces X1subscript𝑋1X_{1}, X2subscript𝑋2X_{2}, Y1subscript𝑌1Y_{1}, Y2subscript𝑌2Y_{2} of X𝑋X. Assume P𝑃P is a projection onto X1subscript𝑋1X_{1} with (IP)𝐼𝑃(I-P) a projection onto X2subscript𝑋2X_{2}, and Q𝑄Q is a projection on Y1subscript𝑌1Y_{1} with (IQ)𝐼𝑄(I-Q) projecting onto Y2subscript𝑌2Y_{2}. Assume Q|X1:X1Y1:evaluated-at𝑄subscript𝑋1subscript𝑋1subscript𝑌1Q|_{X_{1}}:X_{1}\to Y_{1} and (IP)|Y2:Y2X2:evaluated-at𝐼𝑃subscript𝑌2subscript𝑌2subscript𝑋2(I-P)|_{Y_{2}}:Y_{2}\to X_{2} are invertible and d(X1,Y2)>0𝑑subscript𝑋1subscript𝑌20d(X_{1},Y_{2})>0. Then X=X1Y2𝑋direct-sumsubscript𝑋1subscript𝑌2X=X_{1}\oplus Y_{2}.

Proof.

By Lemma 4.3, we have that (Q|X1)1Qsuperscriptevaluated-at𝑄subscript𝑋11𝑄(Q|_{X_{1}})^{-1}Q is a projection onto X1subscript𝑋1X_{1} and ((IP)|Y2)1(IP)superscriptevaluated-at𝐼𝑃subscript𝑌21𝐼𝑃\big{(}(I-P)|_{Y_{2}}\big{)}^{-1}(I-P) is a projection onto Y2subscript𝑌2Y_{2}. Furthermore, both

(Q|X1)1Q((IP)|Y2)1(IP)=0and((IP)|Y2)1(IP)(Q|X1)1Q=0.formulae-sequencesuperscriptevaluated-at𝑄subscript𝑋11𝑄superscriptevaluated-at𝐼𝑃subscript𝑌21𝐼𝑃0andsuperscriptevaluated-at𝐼𝑃subscript𝑌21𝐼𝑃superscriptevaluated-at𝑄subscript𝑋11𝑄0(Q|_{X_{1}})^{-1}Q\big{(}(I-P)|_{Y_{2}}\big{)}^{-1}(I-P)=0\quad\mbox{and}\quad\big{(}(I-P)|_{Y_{2}}\big{)}^{-1}(I-P)(Q|_{X_{1}})^{-1}Q=0.

Thus, the operator R:XX:𝑅𝑋𝑋R:X\to X defined by

R:=(Q|X1)1Q+((IP)|Y2)1(IP)assign𝑅superscriptevaluated-at𝑄subscript𝑋11𝑄superscriptevaluated-at𝐼𝑃subscript𝑌21𝐼𝑃R:=(Q|_{X_{1}})^{-1}Q+\big{(}(I-P)|_{Y_{2}}\big{)}^{-1}(I-P)

is a projection from X𝑋X onto X1Y2direct-sumsubscript𝑋1subscript𝑌2X_{1}\oplus Y_{2}. We will show that the projection R𝑅R is one-to-one (and hence the identity map) to prove that X=X1Y2𝑋direct-sumsubscript𝑋1subscript𝑌2X=X_{1}\oplus Y_{2}. Let xX𝑥𝑋x\in X be so that Rx=0𝑅𝑥0Rx=0. Then

Rx=(Q|X1)1Qx+((IP)|Y2)1(IP)x=0\|Rx\|=\|(Q|_{X_{1}})^{-1}Qx+\big{(}(I-P)|_{Y_{2}}\big{)}^{-1}(I-P)x\|=0

and since d(X1,Y2)>0𝑑subscript𝑋1subscript𝑌20d(X_{1},Y_{2})>0, it follows that

(Q|X1)1Qx=0and((IP)|Y2)1(IP)x=0.formulae-sequencesuperscriptevaluated-at𝑄subscript𝑋11𝑄𝑥0andsuperscriptevaluated-at𝐼𝑃subscript𝑌21𝐼𝑃𝑥0(Q|_{X_{1}})^{-1}Qx=0\quad\mbox{and}\quad\big{(}(I-P)|_{Y_{2}}\big{)}^{-1}(I-P)x=0.

Thus, Qx=0=(IP)x𝑄𝑥0𝐼𝑃𝑥Qx=0=(I-P)x so that x=Px=(IQ)x𝑥𝑃𝑥𝐼𝑄𝑥x=Px=(I-Q)x. However, since (Q|X1)1Qsuperscriptevaluated-at𝑄subscript𝑋11𝑄(Q|_{X_{1}})^{-1}Q is a projection onto X1subscript𝑋1X_{1}, we have Px=(Q|X1)1QPx=(Q|X1)1Qx=0𝑃𝑥superscriptevaluated-at𝑄subscript𝑋11𝑄𝑃𝑥superscriptevaluated-at𝑄subscript𝑋11𝑄𝑥0Px=(Q|_{X_{1}})^{-1}QPx=(Q|_{X_{1}})^{-1}Qx=0 and therefore x=0𝑥0x=0. ∎

Given a sequence (xi)isubscriptsubscript𝑥𝑖𝑖(x_{i})_{i\in\mathbb{N}} in a Banach space X𝑋X and a set ΓΓ\Gamma\subseteq\mathbb{N}, we let [xi]iΓsubscriptdelimited-[]subscript𝑥𝑖𝑖Γ[x_{i}]_{i\in\Gamma} denote the subspace span¯{xi}iΓ¯spansubscriptsubscript𝑥𝑖𝑖Γ\overline{{\rm span}}\{x_{i}\}_{i\in\Gamma}.

Theorem 4.5.

Let (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1𝑖1(x_{i}^{1})_{i=1}^{\infty} be two C𝐶C-unconditional bases for a Banach space X𝑋X with biorthogonal functionals (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0*})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1𝑖1(x_{i}^{1*})_{i=1}^{\infty}, respectively. The following are equivalent:

  1. (i)

    For every σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)})_{i=1}^{\infty} is an unconditional basis for X𝑋X.

  2. (ii)

    There is a K>0𝐾0K>0 so that for any σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)})_{i=1}^{\infty} is a K𝐾K-unconditional basis for X𝑋X.

  3. (iii)

    For every σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i),xiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖superscriptsubscript𝑥𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)},x_{i}^{\sigma(i)*})_{i=1}^{\infty} is an unconditional approximate Schauder frame.

  4. (iv)

    For every σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, (xiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)})_{i=1}^{\infty} is an unconditional basic sequence.

  5. (v)

    There exists a D>0𝐷0D>0 so that for every σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}},

    d([xi0]iσ1(0),[xi1]iσ1(1))D1.𝑑subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎11superscript𝐷1d\left([x_{i}^{0}]_{i\in\sigma^{-1}(0)},[x_{i}^{1}]_{i\in\sigma^{-1}(1)}\right)\geq D^{-1}.
  6. (vi)

    There exists E>0𝐸0E>0 such that for any σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, if P𝑃P is the basis projection of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} onto [xi0]iσ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10[x_{i}^{0}]_{i\in\sigma^{-1}(0)} and Q𝑄Q is the basis projection of (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} onto [xi1]iσ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10[x_{i}^{1}]_{i\in\sigma^{-1}(0)} then

    P|[xi1]iσ1(0):[xi1]iσ1(0)[xi0]iσ1(0)andQ|[xi0]iσ1(0):[xi0]iσ1(0)[xi1]iσ1(0):evaluated-at𝑃subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10andevaluated-at𝑄subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10:subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10P|_{[x_{i}^{1}]_{i\in\sigma^{-1}(0)}}:{[x_{i}^{1}]_{i\in\sigma^{-1}(0)}}\!\!\rightarrow{[x_{i}^{0}]_{i\in\sigma^{-1}(0)}}\quad\textrm{and}\quad Q|_{[x_{i}^{0}]_{i\in\sigma^{-1}(0)}}:{[x_{i}^{0}]_{i\in\sigma^{-1}(0)}}\!\!\rightarrow{[x_{i}^{1}]_{i\in\sigma^{-1}(0)}}

    are invertible with inverses having norm at most E𝐸E.

Proof.

Note that (i)\Rightarrow(ii) follows from Proposition 3.1, and that (ii)\Rightarrow(iv) is immediate. We will prove (iv)\Rightarrow(v)\Rightarrow(vi)\Rightarrow(i) and (iii)\Rightarrow(vi)\Rightarrow(iii).

The case (iv)\Rightarrow(v) is well known, but we include a short proof here for completeness. By Proposition 3.1 there exists K>0𝐾0K>0 such that (xiσ(i))isubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖𝑖(x^{\sigma(i)}_{i})_{i\in\mathbb{N}} is K𝐾K-suppression unconditional for all σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}, x[xi0]iσ1(0)𝑥subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎10x\in[x^{0}_{i}]_{i\in\sigma^{-1}(0)}, and y[xi1]iσ1(1)𝑦subscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖superscript𝜎11y\in[x^{1}_{i}]_{i\in\sigma^{-1}(1)}. We have that xy1Kxnorm𝑥𝑦1𝐾norm𝑥\|x-y\|\geq\frac{1}{K}\|x\|, and hence d([xi0]iσ1(0),[xi1]iσ1(1))1K𝑑subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎111𝐾d\big{(}[x_{i}^{0}]_{i\in\sigma^{-1}(0)},[x_{i}^{1}]_{i\in\sigma^{-1}(1)}\big{)}\geq\frac{1}{K}.

For (v)\Rightarrow(vi), we assume that (v) holds for some D>0𝐷0D>0. Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. For ΓΓ\Gamma\subseteq\mathbb{N} we let PΓsubscript𝑃ΓP_{\Gamma} be the basis projection of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} onto [xi0]iΓsubscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖Γ[x_{i}^{0}]_{i\in\Gamma}. Note that this implies that IPΓ=PΓc𝐼subscript𝑃Γsubscript𝑃superscriptΓ𝑐I-P_{\Gamma}=P_{\Gamma^{c}} is the basis projection of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} onto [xi0]iΓcsubscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscriptΓ𝑐[x_{i}^{0}]_{i\in\Gamma^{c}}. Let x[xi1]iσ1(0)𝑥subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10x\in[x_{i}^{1}]_{i\in\sigma^{-1}(0)} with x=1norm𝑥1\|x\|=1. Then

Pσ1(0)(x)=x(xPσ1(0)(x))d([xi1]iσ1(0),[xi0]iσ1(1))D1.normsubscript𝑃superscript𝜎10𝑥norm𝑥𝑥subscript𝑃superscript𝜎10𝑥𝑑subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎11superscript𝐷1\left\|P_{\sigma^{-1}(0)}(x)\right\|=\left\|x-(x-P_{\sigma^{-1}(0)}(x))\right\|\geq d\left([x_{i}^{1}]_{i\in\sigma^{-1}(0)},[x_{i}^{0}]_{i\in\sigma^{-1}(1)}\right)\geq D^{-1}. (3)

Thus, we have that Pσ1(0)subscript𝑃superscript𝜎10P_{\sigma^{-1}(0)} is an isomorphism of [xi1]iσ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10[x_{i}^{1}]_{i\in\sigma^{-1}(0)} onto a closed subspace of [xi0]σ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖0superscript𝜎10[x_{i}^{0}]_{\sigma^{-1}(0)} with inverse having norm at most E:=Dassign𝐸𝐷E:=D. It remains to prove that Pσ1(0)([xi1]σ1(0))subscript𝑃superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1superscript𝜎10P_{\sigma^{-1}(0)}\left([x_{i}^{1}]_{\sigma^{-1}(0)}\right) is dense in [xi0]iσ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10[x_{i}^{0}]_{i\in\sigma^{-1}(0)}.

Let yspan(xi0)iσ1(0)𝑦spansubscriptsuperscriptsubscript𝑥𝑖0𝑖superscript𝜎10y\in{\rm span}(x_{i}^{0})_{i\in\sigma^{-1}(0)} with y=1norm𝑦1\|y\|=1. As (xi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0𝑖1(x_{i}^{0})_{i=1}^{\infty} and (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} are unconditional bases there exists N𝑁N\in\mathbb{N} such that yspan1iNxi0𝑦subscriptspan1𝑖𝑁subscriptsuperscript𝑥0𝑖y\in{\rm span}_{1\leq i\leq N}x^{0}_{i} and for all Γ(N,)Γ𝑁\Gamma\subset(N,\infty) we have that iΓxi1(y)xi1<εnormsubscript𝑖Γsubscriptsuperscript𝑥1𝑖𝑦subscriptsuperscript𝑥1𝑖𝜀\big{\|}\sum_{i\in\Gamma}x^{1*}_{i}(y)x^{1}_{i}\big{\|}<\varepsilon. For nN𝑛𝑁n\geq N, we let σn{0,1}subscript𝜎𝑛superscript01\sigma_{n}\in\{0,1\}^{\mathbb{N}} be defined by σn|[1,n)=σ|[1,n)evaluated-atsubscript𝜎𝑛1𝑛evaluated-at𝜎1𝑛\sigma_{n}|_{[1,n)}=\sigma|_{[1,n)} and σn(i)=1subscript𝜎𝑛𝑖1\sigma_{n}(i)=1 for in𝑖𝑛i\geq n. A simple induction argument gives that (xiσn(i))i=1superscriptsubscriptsubscriptsuperscript𝑥subscript𝜎𝑛𝑖𝑖𝑖1(x^{\sigma_{n}(i)}_{i})_{i=1}^{\infty} is an unconditional basis for X𝑋X as we have changed only finitely many of the coordinates of (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} and d([xi0]iσn1(0),[xi1]iσn1(1))>0𝑑subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscriptsubscript𝜎𝑛10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscriptsubscript𝜎𝑛110d\big{(}[x_{i}^{0}]_{i\in\sigma_{n}^{-1}(0)},[x_{i}^{1}]_{i\in\sigma_{n}^{-1}(1)}\big{)}>0. Furthermore, Pσn1(0)subscript𝑃superscriptsubscript𝜎𝑛10P_{\sigma_{n}^{-1}(0)} is an isomorphism of [xi1]σn1(0)subscriptdelimited-[]subscriptsuperscript𝑥1𝑖superscriptsubscript𝜎𝑛10[x^{1}_{i}]_{\sigma_{n}^{-1}(0)} onto [xi0]σn1(0)subscriptdelimited-[]subscriptsuperscript𝑥0𝑖superscriptsubscript𝜎𝑛10[x^{0}_{i}]_{\sigma_{n}^{-1}(0)} because these spaces are finite dimensional and have the same dimension. Hence, for each nN𝑛𝑁n\geq N there exists unique znspan(xi1)iσn1(0)subscript𝑧𝑛spansubscriptsubscriptsuperscript𝑥1𝑖𝑖superscriptsubscript𝜎𝑛10z_{n}\in{\rm span}(x^{1}_{i})_{i\in\sigma_{n}^{-1}(0)} such that y=Pσn1(0)(zn)=iσn1(0)xi0(zn)xi0𝑦subscript𝑃superscriptsubscript𝜎𝑛10subscript𝑧𝑛subscript𝑖superscriptsubscript𝜎𝑛10subscriptsuperscript𝑥0𝑖subscript𝑧𝑛subscriptsuperscript𝑥0𝑖y=P_{\sigma_{n}^{-1}(0)}(z_{n})=\sum_{i\in\sigma_{n}^{-1}(0)}x^{0*}_{i}(z_{n})x^{0}_{i} and znDnormsubscript𝑧𝑛𝐷\|z_{n}\|\leq D by (3). Therefore,

Pσ1(0)zny=iσ1(0)xi0(zn)xi0iσn1(0)xi0(zn)xi0=iσ1(0)(n,)xi0(zn)xi0.subscript𝑃superscript𝜎10subscript𝑧𝑛𝑦subscript𝑖superscript𝜎10subscriptsuperscript𝑥0𝑖subscript𝑧𝑛subscriptsuperscript𝑥0𝑖subscript𝑖superscriptsubscript𝜎𝑛10subscriptsuperscript𝑥0𝑖subscript𝑧𝑛subscriptsuperscript𝑥0𝑖subscript𝑖superscript𝜎10𝑛subscriptsuperscript𝑥0𝑖subscript𝑧𝑛subscriptsuperscript𝑥0𝑖P_{\sigma^{-1}(0)}z_{n}-y=\sum_{i\in\sigma^{-1}(0)}x^{0*}_{i}(z_{n})x^{0}_{i}-\sum_{i\in\sigma_{n}^{-1}(0)}x^{0*}_{i}(z_{n})x^{0}_{i}=\sum_{i\in\sigma^{-1}(0)\cap(n,\infty)}x^{0*}_{i}(z_{n})x^{0}_{i}. (4)

If (Pσ1(0)zn)n=1superscriptsubscriptsubscript𝑃superscript𝜎10subscript𝑧𝑛𝑛1(P_{\sigma^{-1}(0)}z_{n})_{n=1}^{\infty} has a sequence of convex combinations which converges to y𝑦y then we can conclude that yPσ1(0)([xi1]iσ1(0))𝑦subscript𝑃superscript𝜎10subscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖superscript𝜎10y\in P_{\sigma^{-1}(0)}([x^{1}_{i}]_{i\in\sigma^{-1}(0)}) as the set is closed and convex. We assume by contradiction that this is not the case. Since any convex combination of y𝑦y is itself, there exists F>0𝐹0F>0 such that for all nN𝑛𝑁n\geq N,

iσ1(0)(n,)j=nxi0(ajzj)xi0>2F for all (aj)j=n[0,1] with j=naj=1.normsubscript𝑖superscript𝜎10𝑛superscriptsubscript𝑗𝑛subscriptsuperscript𝑥0𝑖subscript𝑎𝑗subscript𝑧𝑗subscriptsuperscript𝑥0𝑖2𝐹 for all superscriptsubscriptsubscript𝑎𝑗𝑗𝑛superscript01 with superscriptsubscript𝑗𝑛subscript𝑎𝑗1\left\|\sum_{i\in\sigma^{-1}(0)\cap(n,\infty)}\sum_{j=n}^{\infty}x^{0*}_{i}(a_{j}z_{j})x^{0}_{i}\right\|>2F\textrm{ for all }(a_{j})_{j=n}^{\infty}\in[0,1]^{\mathbb{N}}\textrm{ with }\sum_{j=n}^{\infty}a_{j}=1. (5)

Let εn0subscript𝜀𝑛0\varepsilon_{n}\searrow 0 such that n=1nεn<1superscriptsubscript𝑛1𝑛subscript𝜀𝑛1\sum_{n=1}^{\infty}n\varepsilon_{n}<1 and n=1εn<Fsuperscriptsubscript𝑛1subscript𝜀𝑛𝐹\sum_{n=1}^{\infty}\varepsilon_{n}<F. We inductively choose a subsequence (Mn)n=1superscriptsubscriptsubscript𝑀𝑛𝑛1(M_{n})_{n=-1}^{\infty} of \mathbb{N} with M1=1subscript𝑀11M_{-1}=1 and M0=N+1subscript𝑀0𝑁1M_{0}=N+1 and sequences (yn)n=1,(wn)n=0Xsuperscriptsubscriptsubscript𝑦𝑛𝑛1superscriptsubscriptsubscript𝑤𝑛𝑛0𝑋(y_{n})_{n=1}^{\infty},(w_{n})_{n=0}^{\infty}\subseteq X such that

  1. (a)

    iσ1(0)[Mn+1,)xi0(zMj)xi0<εn+1normsubscript𝑖superscript𝜎10subscript𝑀𝑛1subscriptsuperscript𝑥0𝑖subscript𝑧subscript𝑀𝑗subscriptsuperscript𝑥0𝑖subscript𝜀𝑛1\big{\|}\sum_{i\in\sigma^{-1}(0)\cap[M_{n+1},\infty)}x^{0*}_{i}(z_{M_{j}})x^{0}_{i}\big{\|}<\varepsilon_{n+1} for all 1jn1𝑗𝑛1\leq j\leq n and n𝑛n\in\mathbb{N}.

  2. (b)

    yn:=iσ1(0)[Mn,Mn+1)xi0(zMn)xi0assignsubscript𝑦𝑛subscript𝑖superscript𝜎10subscript𝑀𝑛subscript𝑀𝑛1subscriptsuperscript𝑥0𝑖subscript𝑧subscript𝑀𝑛subscriptsuperscript𝑥0𝑖y_{n}:=\sum_{i\in\sigma^{-1}(0)\cap[M_{n},M_{n+1})}x^{0*}_{i}(z_{M_{n}})x^{0}_{i} for all n𝑛n\in\mathbb{N} and jajyj>Fnormsubscript𝑗subscript𝑎𝑗subscript𝑦𝑗𝐹\big{\|}\sum_{j\in\mathbb{N}}a_{j}y_{j}\big{\|}>F for all (aj)j[0,1]subscriptsubscript𝑎𝑗𝑗superscript01(a_{j})_{j\in\mathbb{N}}\in[0,1]^{\mathbb{N}} with jaj=1subscript𝑗subscript𝑎𝑗1\sum_{j\in\mathbb{N}}a_{j}=1.

  3. (c)

    wn:=limjiσ1(0)[Mn1,Mn)xi1(zMj)xi1assignsubscript𝑤𝑛subscript𝑗subscript𝑖superscript𝜎10subscript𝑀𝑛1subscript𝑀𝑛subscriptsuperscript𝑥1𝑖subscript𝑧subscript𝑀𝑗subscriptsuperscript𝑥1𝑖w_{n}:=\lim\limits_{j\rightarrow\infty}\sum_{i\in\sigma^{-1}(0)\cap[M_{n-1},M_{n})}x^{1*}_{i}(z_{M_{j}})x^{1}_{i} for all n0𝑛0n\geq 0.

  4. (d)

    wniσ1(0)[Mn1,Mn)xi1(zMj)xi1<εj/jnormsubscript𝑤𝑛subscript𝑖superscript𝜎10subscript𝑀𝑛1subscript𝑀𝑛subscriptsuperscript𝑥1𝑖subscript𝑧subscript𝑀𝑗subscriptsuperscript𝑥1𝑖subscript𝜀𝑗𝑗\big{\|}w_{n}-\sum_{i\in\sigma^{-1}(0)\cap[M_{n-1},M_{n})}x^{1*}_{i}(z_{M_{j}})x^{1}_{i}\big{\|}<\varepsilon_{j}/j for all j>n0𝑗𝑛0j>n\geq 0.

Note that (a) can be achieved because the series converges and we are only working with finitely many zjsubscript𝑧𝑗z_{j} at each step. We have that (b) follows from (5), (a), and n=1εn<Fsuperscriptsubscript𝑛1subscript𝜀𝑛𝐹\sum_{n=1}^{\infty}\varepsilon_{n}<F. The definition of wnsubscript𝑤𝑛w_{n} in (c) can be made because the sum is over the finite interval [Mn1,Mn)subscript𝑀𝑛1subscript𝑀𝑛[M_{n-1},M_{n}) and the sequence (zMj)j=n+1superscriptsubscriptsubscript𝑧subscript𝑀𝑗𝑗𝑛1(z_{M_{j}})_{j=n+1}^{\infty} is bounded, so we may pass to a subsequence of (Mj)jsubscriptsubscript𝑀𝑗𝑗(M_{j})_{j\in\mathbb{N}} in which the limit exists. Finally, since the limit exists by (c), we can pass to yet another subsequence of (Mj)jsubscriptsubscript𝑀𝑗𝑗(M_{j})_{j\in\mathbb{N}} to obtain (d).

For all n𝑛n\in\mathbb{N}, we set vn:=iσ1(0)[Mn1,Mn)xi1(zMn)xi1assignsubscript𝑣𝑛subscript𝑖superscript𝜎10subscript𝑀𝑛1subscript𝑀𝑛subscriptsuperscript𝑥1𝑖subscript𝑧subscript𝑀𝑛subscriptsuperscript𝑥1𝑖v_{n}:=\sum_{i\in\sigma^{-1}(0)\cap[M_{n-1},M_{n})}x^{1*}_{i}(z_{M_{n}})x^{1}_{i} and obtain the following:

  1. (e)

    zMn(vn+j=0n1wj)<εnnormsubscript𝑧subscript𝑀𝑛subscript𝑣𝑛superscriptsubscript𝑗0𝑛1subscript𝑤𝑗subscript𝜀𝑛\big{\|}z_{M_{n}}-(v_{n}+\sum_{j=0}^{n-1}w_{j})\big{\|}<\varepsilon_{n} for all n𝑛n\in\mathbb{N}.

  2. (f)

    Pσ1(0)zMn(y+yn)<εn+1normsubscript𝑃superscript𝜎10subscript𝑧subscript𝑀𝑛𝑦subscript𝑦𝑛subscript𝜀𝑛1\big{\|}P_{\sigma^{-1}(0)}z_{M_{n}}-(y+y_{n})\big{\|}<\varepsilon_{n+1} for all n𝑛n\in\mathbb{N}.

We have that (e) follows from (d) and the fact that zMn[xj1]j=1Mn1subscript𝑧subscript𝑀𝑛superscriptsubscriptdelimited-[]superscriptsubscript𝑥𝑗1𝑗1subscript𝑀𝑛1z_{M_{n}}\in[x_{j}^{1}]_{j=1}^{M_{n}-1}. We have that (f) follows from (a), (b), and (3).

Let M𝑀M\in\mathbb{N} and Ω=σ1(0)n=12M[M2n1,M2n)\Omega=\sigma^{-1}(0)\cap\cup_{n=1}^{2M}[M_{2n-1},M_{2n}). Then

12Mn=12M(1)nv2n[xi1]iΩand12Mn=12M(1)n(Pσ1(1)zM2n+y2n)[xi0]iΩ.formulae-sequence12𝑀superscriptsubscript𝑛12𝑀superscript1𝑛subscript𝑣2𝑛subscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖Ωand12𝑀superscriptsubscript𝑛12𝑀superscript1𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛subscript𝑦2𝑛subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖Ω\frac{1}{2M}\sum_{n=1}^{2M}(-1)^{n}v_{2n}\in[x^{1}_{i}]_{i\in\Omega}\quad\mbox{and}\quad\frac{1}{2M}\sum_{n=1}^{2M}(-1)^{n}(P_{\sigma^{-1}(1)}z_{M_{2n}}+y_{2n})\in[x^{0}_{i}]_{i\in\mathbb{N}\setminus\Omega}. (6)

Since (yn)n=Nsuperscriptsubscriptsubscript𝑦𝑛𝑛𝑁(y_{n})_{n=N}^{\infty} is a block sequence of (xi0)iσ1(0)subscriptsubscriptsuperscript𝑥0𝑖𝑖superscript𝜎10(x^{0}_{i})_{i\in\sigma^{-1}(0)}, it is also C𝐶C-unconditional and therefore by (b) we have that

12Mn=12M(1)n(Pσ1(1)zM2n+y2n)C112Mn=12My2n>C1F.norm12𝑀superscriptsubscript𝑛12𝑀superscript1𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛subscript𝑦2𝑛superscript𝐶1norm12𝑀superscriptsubscript𝑛12𝑀subscript𝑦2𝑛superscript𝐶1𝐹\left\|\dfrac{1}{2M}\sum_{n=1}^{2M}(-1)^{n}\left(P_{\sigma^{-1}(1)}z_{M_{2n}}+y_{2n}\right)\right\|\geq C^{-1}\left\|\dfrac{1}{2M}\sum_{n=1}^{2M}y_{2n}\right\|>C^{-1}F. (7)

As (wi)i=0superscriptsubscriptsubscript𝑤𝑖𝑖0(w_{i})_{i=0}^{\infty} is a block sequence of (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty}, we have for all n𝑛n\in\mathbb{N} and Γ{0,1,,n}Γ01𝑛\Gamma\subseteq\{0,1,...,n\} that

iΓwiCi=1nwi=Climjiσ1(0)[1,Mn)xi1(zMj)xi1lim supjC2zMjC2D.normsubscript𝑖Γsubscript𝑤𝑖𝐶normsuperscriptsubscript𝑖1𝑛subscript𝑤𝑖𝐶normsubscript𝑗subscript𝑖superscript𝜎101subscript𝑀𝑛subscriptsuperscript𝑥1𝑖subscript𝑧subscript𝑀𝑗subscriptsuperscript𝑥1𝑖subscriptlimit-supremum𝑗superscript𝐶2normsubscript𝑧subscript𝑀𝑗superscript𝐶2𝐷\left\|\sum_{i\in\Gamma}w_{i}\right\|\leq C\left\|\sum_{i=1}^{n}w_{i}\right\|=C\left\|\lim_{j\rightarrow\infty}\sum_{i\in\sigma^{-1}(0)\cap[1,M_{n})}x^{1*}_{i}(z_{M_{j}})x^{1}_{i}\right\|\leq\limsup_{j\rightarrow\infty}C^{2}\|z_{M_{j}}\|\leq C^{2}D. (8)

Thus, we have the following estimate:

12Mn=12M(1)n(Pσ1(1)zM2n+y2n)12Mn=12M(1)nv2nnorm12𝑀superscriptsubscript𝑛12𝑀superscript1𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛subscript𝑦2𝑛12𝑀superscriptsubscript𝑛12𝑀superscript1𝑛subscript𝑣2𝑛\displaystyle\left\|\frac{1}{2M}\sum_{n=1}^{2M}\right.(-1)^{n}\left.(P_{\sigma^{-1}(1)}z_{M_{2n}}+y_{2n})-\frac{1}{2M}\sum_{n=1}^{2M}(-1)^{n}v_{2n}\right\|
=12Mn=12M(1)n(Pσ1(1)zM2n+y2nv2n)absent12𝑀normsuperscriptsubscript𝑛12𝑀superscript1𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛subscript𝑦2𝑛subscript𝑣2𝑛\displaystyle=\frac{1}{2M}\left\|\sum_{n=1}^{2M}(-1)^{n}(P_{\sigma^{-1}(1)}z_{M_{2n}}+y_{2n}-v_{2n})\right\|
12Mn=12M(1)n(y+y2n+Pσ1(1)zM2n(v2n+j=02n1wj))+12Mj=1Mw4j2+w4j1absent12𝑀normsuperscriptsubscript𝑛12𝑀superscript1𝑛𝑦subscript𝑦2𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛subscript𝑣2𝑛superscriptsubscript𝑗02𝑛1subscript𝑤𝑗12𝑀normsuperscriptsubscript𝑗1𝑀subscript𝑤4𝑗2subscript𝑤4𝑗1\displaystyle\leq\frac{1}{2M}\left\|\sum_{n=1}^{2M}(-1)^{n}(y+y_{2n}+P_{\sigma^{-1}(1)}z_{M_{2n}}-(v_{2n}+\sum_{j=0}^{2n-1}w_{j}))\right\|+\frac{1}{2M}\left\|\sum_{j=1}^{M}w_{4j-2}+w_{4j-1}\right\|
12Mn=12My+y2n+Pσ1(1)zM2n(v2n+j=02n1wj)+C2D2M by (8)absent12𝑀superscriptsubscript𝑛12𝑀norm𝑦subscript𝑦2𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛subscript𝑣2𝑛superscriptsubscript𝑗02𝑛1subscript𝑤𝑗superscript𝐶2𝐷2𝑀 by italic-(8italic-)\displaystyle\leq\frac{1}{2M}\sum_{n=1}^{2M}\left\|y+y_{2n}+P_{\sigma^{-1}(1)}z_{M_{2n}}-(v_{2n}+\sum_{j=0}^{2n-1}w_{j})\right\|+\frac{C^{2}D}{2M}\quad\textrm{ by }\eqref{E:sumwi}
<12Mn=12MPσ1(0)zM2n+Pσ1(1)zM2nzM2n+12Mn=12M(ε2n+ε2n+1)+C2D2M by (e) and (f)bra12𝑀superscriptsubscript𝑛12𝑀subscript𝑃superscript𝜎10subscript𝑧subscript𝑀2𝑛subscript𝑃superscript𝜎11subscript𝑧subscript𝑀2𝑛conditionalsubscript𝑧subscript𝑀2𝑛12𝑀superscriptsubscript𝑛12𝑀subscript𝜀2𝑛subscript𝜀2𝑛1superscript𝐶2𝐷2𝑀 by (e) and (f)\displaystyle<\frac{1}{2M}\sum_{n=1}^{2M}\left\|P_{\sigma^{-1}(0)}z_{M_{2n}}\!+\!P_{\sigma^{-1}(1)}z_{M_{2n}}\!-\!z_{M_{2n}}\right\|+\frac{1}{2M}\sum_{n=1}^{2M}(\varepsilon_{2n}+\varepsilon_{2n+1})\!+\!\frac{C^{2}D}{2M}\textrm{ by (e) and (f)}
<12M+C2D2M.absent12𝑀superscript𝐶2𝐷2𝑀\displaystyle<\frac{1}{2M}+\frac{C^{2}D}{2M}.

The above estimate together with (6) and (7) gives that d([xi0]iΩ,[xi1]iΩ)<CF1(1+C2D)(2M)1𝑑subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖Ωsubscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖Ω𝐶superscript𝐹11superscript𝐶2𝐷superscript2𝑀1d([x^{0}_{i}]_{i\in\mathbb{N}\setminus\Omega},[x^{1}_{i}]_{i\in\Omega})<CF^{-1}(1+C^{2}D)(2M)^{-1}. This contradicts (v) by choosing M𝑀M large enough. An identical proof shows the claim for Q𝑄Q.

For (vi)\Rightarrow(i), we assume that (vi) holds for some E>0𝐸0E>0. Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. For jσ1(0)𝑗superscript𝜎10j\in\sigma^{-1}(0) we let fj:=(Q|[xi0]iσ1(0)1Q)xj0assignsubscript𝑓𝑗superscriptevaluated-at𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄subscriptsuperscript𝑥0𝑗f_{j}:=\big{(}Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}Q\big{)}^{*}x^{0*}_{j} and for jσ1(1)𝑗superscript𝜎11j\in\sigma^{-1}(1) we let fj:=((IP)|[xi1]iσ1(1)1(IP))xj1assignsubscript𝑓𝑗superscriptevaluated-at𝐼𝑃subscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖superscript𝜎111𝐼𝑃subscriptsuperscript𝑥1𝑗f_{j}:=\big{(}(I-P)|_{[x^{1}_{i}]_{i\in\sigma^{-1}(1)}}^{-1}(I-P)\big{)}^{*}x^{1*}_{j}. We first prove that (xiσ(i),fi)i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscript𝑓𝑖𝑖1(x^{\sigma(i)}_{i},f_{i})_{i=1}^{\infty} is a biorthogonal system. Let j,kσ1(0)𝑗𝑘superscript𝜎10j,k\in\sigma^{-1}(0), then

fj(xk0)=((Q|[xi0]iσ1(0)1Q)xj0)(xk0)=xj0((Q|[xi0]iσ1(0)1Q)xk0)=xj0(xk0)=δj,ksubscript𝑓𝑗subscriptsuperscript𝑥0𝑘superscriptevaluated-at𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄subscriptsuperscript𝑥0𝑗superscriptsubscript𝑥𝑘0subscriptsuperscript𝑥0𝑗evaluated-at𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄subscriptsuperscript𝑥0𝑘subscriptsuperscript𝑥0𝑗subscriptsuperscript𝑥0𝑘subscript𝛿𝑗𝑘f_{j}(x^{0}_{k})=\left(\big{(}Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}Q\big{)}^{*}x^{0*}_{j}\right)(x_{k}^{0})=x^{0*}_{j}\left(\big{(}Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}Q\big{)}x^{0}_{k}\right)=x^{0*}_{j}(x^{0}_{k})=\delta_{j,k}

For jσ1(0)𝑗superscript𝜎10j\in\sigma^{-1}(0) and kσ1(1)𝑘superscript𝜎11k\in\sigma^{-1}(1),

fj(xk1)=((Q|[xi0]iσ1(0)1Q)xj0)(xk1)=xj0((Q|[xi0]iσ1(0)1Q)xk1)=xj0(0)=0subscript𝑓𝑗subscriptsuperscript𝑥1𝑘superscriptevaluated-at𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄subscriptsuperscript𝑥0𝑗superscriptsubscript𝑥𝑘1subscriptsuperscript𝑥0𝑗evaluated-at𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄subscriptsuperscript𝑥1𝑘subscriptsuperscript𝑥0𝑗00f_{j}(x^{1}_{k})=\left(\big{(}Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}Q\big{)}^{*}x^{0*}_{j}\right)(x_{k}^{1})=x^{0*}_{j}\left(\big{(}Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}Q\big{)}x^{1}_{k}\right)=x^{0*}_{j}(0)=0

Thus, if jσ1(0)𝑗superscript𝜎10j\in\sigma^{-1}(0) and k𝑘k\in\mathbb{N}, then fj(xkσ(k))=δj,ksubscript𝑓𝑗subscriptsuperscript𝑥𝜎𝑘𝑘subscript𝛿𝑗𝑘f_{j}(x^{\sigma(k)}_{k})=\delta_{j,k}. Likewise, if jσ1(1)𝑗superscript𝜎11j\in\sigma^{-1}(1) and k𝑘k\in\mathbb{N}, then fj(xkσ(k))=δj,ksubscript𝑓𝑗subscriptsuperscript𝑥𝜎𝑘𝑘subscript𝛿𝑗𝑘f_{j}(x^{\sigma(k)}_{k})=\delta_{j,k}. Thus, (fi)i=1superscriptsubscriptsubscript𝑓𝑖𝑖1(f_{i})_{i=1}^{\infty} are biorthogonal functionals to (xiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)})_{i=1}^{\infty}. Let ΓΓ\Gamma\subseteq\mathbb{N} and xX𝑥𝑋x\in X.

iΓfi(x)xiσ(i)normsubscript𝑖Γsubscript𝑓𝑖𝑥subscriptsuperscript𝑥𝜎𝑖𝑖\displaystyle\left\|\sum_{i\in\Gamma}f_{i}(x)x^{\sigma(i)}_{i}\right\| iΓσ1(0)(Q|[xi0]iσ1(0)1Q)xi0(x)xi0+iΓσ1(1)((IP)|[xi0]iσ1(0)1(IP))xi1(x)xi1\displaystyle\leq\left\|\sum_{i\in\Gamma\cap\sigma^{-1}(0)}\!\!\!\!\!\!(\!Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!\!\!\!\!Q)^{*}x^{0*}_{i}(x)x^{0}_{i}\right\|\!+\!\left\|\sum_{i\in\Gamma\cap\sigma^{-1}(1)}\!\!\!\!\!\!(\!(I-P)|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!\!\!\!\!\!\!(I-P))^{*}x^{1*}_{i}(x)x^{1}_{i}\right\|
Cixi0(Q|[xi0]iσ1(0)1Qx)xi0+Cixi1((IP)|[xi0]iσ1(0)1(IP)x)xi1\displaystyle\leq C\left\|\sum_{i\in\mathbb{N}}x^{0*}_{i}(Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!\!\!Qx)x^{0}_{i}\right\|+C\left\|\sum_{i\in\mathbb{N}}x^{1*}_{i}((I-P)|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!\!\!\!\!(I-P)x)x^{1}_{i}\right\|
=CQ|[xi0]iσ1(0)1Qx+C(IP)|[xi0]iσ1(0)1(IP)xabsent𝐶superscriptsubscriptdelimited-‖|𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄𝑥norm𝐶superscriptsubscriptdelimited-‖|𝐼𝑃subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝐼𝑃𝑥\displaystyle=C\left\|Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!Qx\right\|+C\left\|(I-P)|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!\!\!\!\!(I-P)x\right\|
CQ|[xi0]iσ1(0)1Qx+C(IP)|[xi0]iσ1(0)1(IP)xabsent𝐶superscriptsubscriptdelimited-‖|𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝑄normnorm𝑥𝐶superscriptsubscriptdelimited-‖|𝐼𝑃subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎101𝐼𝑃norm𝑥\displaystyle\leq C\left\|Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!Q\right\|\|x\|+C\left\|(I-P)|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\!\!\!\!\!\!(I-P)\right\|\|x\|
2CExabsent2𝐶𝐸norm𝑥\displaystyle\leq 2CE\|x\|

Thus, (xσ(i))i=1superscriptsubscriptsuperscript𝑥𝜎𝑖𝑖1(x^{\sigma(i)})_{i=1}^{\infty} is an unconditional basic sequence. By Lemma 4.4, we have that [xiσ(i)]i=Xsubscriptdelimited-[]subscriptsuperscript𝑥𝜎𝑖𝑖𝑖𝑋[x^{\sigma(i)}_{i}]_{i\in\mathbb{N}}=X and hence (xσ(i))i=1superscriptsubscriptsuperscript𝑥𝜎𝑖𝑖1(x^{\sigma(i)})_{i=1}^{\infty} is an unconditional basis of X𝑋X.

For (vi)\Rightarrow(iii), we assume that (vi) holds for some E>0𝐸0E>0. Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. We let P𝑃P be the basis projection of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} onto X1:=[xi0]iσ1(0)assignsubscript𝑋1subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10X_{1}:=[x_{i}^{0}]_{i\in\sigma^{-1}(0)} and Q𝑄Q be the basis projection of (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} onto Y1:=[xi1]iσ1(0)assignsubscript𝑌1subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10Y_{1}:=[x_{i}^{1}]_{i\in\sigma^{-1}(0)}. Note that (IP)𝐼𝑃(I-P) is the basis projection of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} onto X2:=[xi0]iσ1(1)assignsubscript𝑋2subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎11X_{2}:=[x_{i}^{0}]_{i\in\sigma^{-1}(1)} and (IQ)𝐼𝑄(I-Q) is the basis projection of (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} onto Y2:=[xi1]iσ1(1)assignsubscript𝑌2subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎11Y_{2}:=[x_{i}^{1}]_{i\in\sigma^{-1}(1)}. We have that P|Y1evaluated-at𝑃subscript𝑌1P|_{Y_{1}}, Q|X1evaluated-at𝑄subscript𝑋1Q|_{X_{1}}, (IP)|Y2evaluated-at𝐼𝑃subscript𝑌2(I-P)|_{Y_{2}}, and (IQ)|Y1evaluated-at𝐼𝑄subscript𝑌1(I-Q)|_{Y_{1}} are all invertible with bounded inverses. We let S𝑆S be the frame operator of (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty}. Then for all xX𝑥𝑋x\in X, we have that

S(x)=i=1xiσ(i)(x)xiσ(i)=iσ1(0)xi0(x)xi0+iσ1(1)xi1(x)xi1=P(x)+(IQ)(x).𝑆𝑥superscriptsubscript𝑖1subscriptsuperscript𝑥𝜎𝑖𝑖𝑥subscriptsuperscript𝑥𝜎𝑖𝑖subscript𝑖superscript𝜎10subscriptsuperscript𝑥0𝑖𝑥subscriptsuperscript𝑥0𝑖subscript𝑖superscript𝜎11subscriptsuperscript𝑥1𝑖𝑥subscriptsuperscript𝑥1𝑖𝑃𝑥𝐼𝑄𝑥S(x)=\sum_{i=1}^{\infty}x^{\sigma(i)*}_{i}(x)x^{\sigma(i)}_{i}=\sum_{i\in\sigma^{-1}(0)}x^{0}_{i}(x)x^{0}_{i}+\sum_{i\in\sigma^{-1}(1)}x^{1}_{i}(x)x^{1}_{i}=P(x)+(I-Q)(x).

Thus, S=P+(IQ)𝑆𝑃𝐼𝑄S=P+(I-Q) is the frame operator of (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty}, and hence is well defined and bounded. We consider the operator T=P|Y11Q|X11Q+(IQ)|X21(IP)|Y21(IP)𝑇evaluated-atevaluated-at𝑃subscript𝑌11𝑄subscript𝑋11𝑄evaluated-atevaluated-at𝐼𝑄subscript𝑋21𝐼𝑃subscript𝑌21𝐼𝑃T=P|^{-1}_{Y_{1}}Q|^{-1}_{X_{1}}Q+(I-Q)|^{-1}_{X_{2}}(I-P)|^{-1}_{Y_{2}}(I-P) and will prove that T𝑇T is the inverse of S𝑆S and hence (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty} would be an approximate Schauder frame. We have that

ST𝑆𝑇\displaystyle ST =(P+(IQ))P|Y11Q|X11Q+(P+(IQ))(IQ)|X21(IP)|Y21(IP)absentevaluated-atevaluated-at𝑃𝐼𝑄𝑃subscript𝑌11𝑄𝑋11𝑄evaluated-atevaluated-at𝑃𝐼𝑄𝐼𝑄subscript𝑋21𝐼𝑃subscript𝑌21𝐼𝑃\displaystyle=(P+(I-Q))P|^{-1}_{Y_{1}}Q|^{-1}_{X1}Q+(P+(I-Q))(I-Q)|^{-1}_{X_{2}}(I-P)|^{-1}_{Y_{2}}(I-P)
=PP|Y11Q|X11Q+(IQ)(IQ)|X21(IP)|Y21(IP)absentevaluated-atevaluated-at𝑃𝑃subscript𝑌11𝑄𝑋11𝑄evaluated-atevaluated-at𝐼𝑄𝐼𝑄subscript𝑋21𝐼𝑃subscript𝑌21𝐼𝑃\displaystyle=PP|^{-1}_{Y_{1}}Q|^{-1}_{X1}Q+(I-Q)(I-Q)|^{-1}_{X_{2}}(I-P)|^{-1}_{Y_{2}}(I-P)
=Q|X11Q+(IP)|Y21(IP)=I.absentevaluated-at𝑄subscript𝑋11𝑄evaluated-at𝐼𝑃subscript𝑌21𝐼𝑃𝐼\displaystyle=Q|^{-1}_{X_{1}}Q+(I-P)|^{-1}_{Y_{2}}(I-P)=I.

where the second equation follows from (IQ)P|Y11=0=P(IQ)|X21evaluated-at𝐼𝑄𝑃subscript𝑌110evaluated-at𝑃𝐼𝑄subscript𝑋21(I-Q)P|^{-1}_{Y_{1}}=0=P(I-Q)|^{-1}_{X_{2}} and the last equation by Lemma 4.4. Thus, ST=I𝑆𝑇𝐼ST=I. To show TS=I𝑇𝑆𝐼TS=I, we have

TS𝑇𝑆\displaystyle TS =P|Y11QX11Q(P+(IQ))+(IQ)|X21(IP)|Y21(IP)(P+(IQ))absentevaluated-at𝑃subscript𝑌11subscriptsuperscript𝑄1𝑋1𝑄𝑃𝐼𝑄evaluated-atevaluated-at𝐼𝑄subscript𝑋21𝐼𝑃subscript𝑌21𝐼𝑃𝑃𝐼𝑄\displaystyle=P|^{-1}_{Y_{1}}Q^{-1}_{X1}Q(P+(I-Q))+(I-Q)|^{-1}_{X_{2}}(I-P)|^{-1}_{Y_{2}}(I-P)(P+(I-Q))
=P|Y11QX11QP+(IQ)|X21(IP)|Y21(IP)(IQ)absentevaluated-at𝑃subscript𝑌11subscriptsuperscript𝑄1𝑋1𝑄𝑃evaluated-atevaluated-at𝐼𝑄subscript𝑋21𝐼𝑃subscript𝑌21𝐼𝑃𝐼𝑄\displaystyle=P|^{-1}_{Y_{1}}Q^{-1}_{X1}QP+(I-Q)|^{-1}_{X_{2}}(I-P)|^{-1}_{Y_{2}}(I-P)(I-Q)
=P1|Y1P+(IQ)|X21(IQ)=Iabsentevaluated-atsuperscript𝑃1subscript𝑌1𝑃evaluated-at𝐼𝑄subscript𝑋21𝐼𝑄𝐼\displaystyle=P^{-1}|_{Y_{1}}P+(I-Q)|^{-1}_{X_{2}}(I-Q)=I

where the last equation follows from Lemma 4.4. Hence, we have that T𝑇T is the inverse of S𝑆S and (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty} is an approximate Schauder frame. We now check that (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty} is an unconditonal approximate Schauder frame. Let xX𝑥𝑋x\in X and ΓΓ\Gamma\subseteq\mathbb{N}. We have that

iΓxiσ(i)(x)xiσ(i)normsubscript𝑖Γsubscriptsuperscript𝑥𝜎𝑖𝑖𝑥subscriptsuperscript𝑥𝜎𝑖𝑖\displaystyle\left\|\sum_{i\in\Gamma}x^{\sigma(i)*}_{i}(x)x^{\sigma(i)}_{i}\right\| iΓσ1(0)xi0(x)xi0+iΓσ1(1)xi1(x)xi1absentnormsubscript𝑖Γsuperscript𝜎10subscriptsuperscript𝑥0𝑖𝑥subscriptsuperscript𝑥0𝑖normsubscript𝑖Γsuperscript𝜎11subscriptsuperscript𝑥1𝑖𝑥subscriptsuperscript𝑥1𝑖\displaystyle\leq\left\|\sum_{i\in\Gamma\cap\sigma^{-1}(0)}x^{0*}_{i}(x)x^{0}_{i}\right\|+\left\|\sum_{i\in\Gamma\cap\sigma^{-1}(1)}x^{1*}_{i}(x)x^{1}_{i}\right\|
Cixi0(x)xi0+Cixi1(x)xi1absent𝐶normsubscript𝑖subscriptsuperscript𝑥0𝑖𝑥subscriptsuperscript𝑥0𝑖𝐶normsubscript𝑖subscriptsuperscript𝑥1𝑖𝑥subscriptsuperscript𝑥1𝑖\displaystyle\leq C\left\|\sum_{i\in\mathbb{N}}x^{0*}_{i}(x)x^{0}_{i}\right\|+C\left\|\sum_{i\in\mathbb{N}}x^{1*}_{i}(x)x^{1}_{i}\right\|
=2Cx.absent2𝐶norm𝑥\displaystyle=2C\|x\|.

Thus, (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty} is a 2C2𝐶2C-unconditional approximate Schauder frame.

For (iii)\Rightarrow(vi), we assume that for every ϕ{0,1}italic-ϕsuperscript01\phi\in\{0,1\}^{\mathbb{N}}, we have that (xiϕ(i),xiϕ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥italic-ϕ𝑖𝑖subscriptsuperscript𝑥italic-ϕ𝑖𝑖𝑖1(x^{\phi(i)}_{i},x^{\phi(i)*}_{i})_{i=1}^{\infty} is an unconditional approximate Schauder frame. By Theorem 2.4 there exists E1𝐸1E\geq 1 such that for every ϕ{0,1}italic-ϕsuperscript01\phi\in\{0,1\}^{\mathbb{N}}, we have that (xiϕ(i),xiϕ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥italic-ϕ𝑖𝑖subscriptsuperscript𝑥italic-ϕ𝑖𝑖𝑖1(x^{\phi(i)}_{i},x^{\phi(i)*}_{i})_{i=1}^{\infty} is a E𝐸E-unconditional approximate Schauder frame whose frame operator S𝑆S satisfies S,S1Enorm𝑆normsuperscript𝑆1𝐸\|S\|,\|S^{-1}\|\leq E. Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and y[xi0]iσ1(0)𝑦subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎10y\in[x^{0}_{i}]_{i\in\sigma^{-1}(0)}. We let P𝑃P be the basis projection of (xi0)i=1superscriptsubscriptsubscriptsuperscript𝑥0𝑖𝑖1(x^{0}_{i})_{i=1}^{\infty} onto [xi0]iσ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10[x_{i}^{0}]_{i\in\sigma^{-1}(0)} and Q𝑄Q be the basis projection of (xi1)i=1superscriptsubscriptsubscriptsuperscript𝑥1𝑖𝑖1(x^{1}_{i})_{i=1}^{\infty} onto [xi1]iσ1(0)subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10[x_{i}^{1}]_{i\in\sigma^{-1}(0)}. Thus, the frame operator of (xiσ(i),xiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖subscriptsuperscript𝑥𝜎𝑖𝑖𝑖1(x^{\sigma(i)}_{i},x^{\sigma(i)*}_{i})_{i=1}^{\infty} is given by S=P+(IQ)𝑆𝑃𝐼𝑄S=P+(I-Q). There exists unique xX𝑥𝑋x\in X with xEynorm𝑥𝐸norm𝑦\|x\|\leq E\|y\| such that S(x)=y𝑆𝑥𝑦S(x)=y. We have that

y=S(x)=(P+(IQ))(Q+(IQ))x=PQx+P(IQ)x+(IQ)x𝑦𝑆𝑥𝑃𝐼𝑄𝑄𝐼𝑄𝑥𝑃𝑄𝑥𝑃𝐼𝑄𝑥𝐼𝑄𝑥y=S(x)=(P+(I-Q))(Q+(I-Q))x=PQx+P(I-Q)x+(I-Q)x

For the sake of contradiction, we assume that (IQ)x0𝐼𝑄𝑥0(I-Q)x\neq 0. As, y[xi0]iσ1(0)𝑦subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10y\in[x_{i}^{0}]_{i\in\sigma^{-1}(0)}, we have that (IP)y=0𝐼𝑃𝑦0(I-P)y=0. Hence, 0=(IP)y=(IP)(IQ)x0𝐼𝑃𝑦𝐼𝑃𝐼𝑄𝑥0=(I-P)y=(I-P)(I-Q)x. Thus, 0=(Q+(IP))(IQ)x0𝑄𝐼𝑃𝐼𝑄𝑥0=(Q+(I-P))(I-Q)x and hence the operator (Q+(IP))𝑄𝐼𝑃(Q+(I-P)) is not invertible. This is a contradiction because (Q+(IP))𝑄𝐼𝑃(Q+(I-P)) is the frame operator of (xi1σ(i),xi(1σ(i)))i=1superscriptsubscriptsubscriptsuperscript𝑥1𝜎𝑖𝑖subscriptsuperscript𝑥1𝜎𝑖𝑖𝑖1(x^{1-\sigma(i)}_{i},x^{(1-\sigma(i))*}_{i})_{i=1}^{\infty}. Thus, (IQ)x=0𝐼𝑄𝑥0(I-Q)x=0 and x=Q(x)[xi1]iσ1(0)𝑥𝑄𝑥subscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖superscript𝜎10x=Q(x)\in[x^{1}_{i}]_{i\in\sigma^{-1}(0)}. This implies P|[xi1]iσ1(0):[xi1]iσ1(0)[xi0]iσ1(0):evaluated-at𝑃subscriptdelimited-[]subscriptsuperscript𝑥1𝑖𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10P|_{[x^{1}_{i}]_{i\in\sigma^{-1}(0)}}\!\!:{[x_{i}^{1}]_{i\in\sigma^{-1}(0)}}\!\!\rightarrow{[x_{i}^{0}]_{i\in\sigma^{-1}(0)}} is invertible and P|[xi1]iσ1(0)1E\|P|_{[x^{1}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\|\leq E. Likewise, Q|[xi0]iσ1(0):[xi0]iσ1(0)[xi1]iσ1(0):evaluated-at𝑄subscriptdelimited-[]subscriptsuperscript𝑥0𝑖𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖0𝑖superscript𝜎10subscriptdelimited-[]superscriptsubscript𝑥𝑖1𝑖superscript𝜎10Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}\!\!:{[x_{i}^{0}]_{i\in\sigma^{-1}(0)}}\!\!\rightarrow{[x_{i}^{1}]_{i\in\sigma^{-1}(0)}} is invertible and Q|[xi0]iσ1(0)1E\|Q|_{[x^{0}_{i}]_{i\in\sigma^{-1}(0)}}^{-1}\|\leq E. ∎

5. Perturbations

Perturbation theorems are powerful tools in both creating new coordinate systems and in showing that coordinate systems are resilient to error. We first recall the following classical perturbation theorem for Schauder bases which is often referred to as the small perturbation lemma (see Theorem 6.18 in [FHHMPZ]).

Theorem 5.1.

Let (xj0)j=1superscriptsubscriptsuperscriptsubscript𝑥𝑗0𝑗1(x_{j}^{0})_{j=1}^{\infty} be a Schauder basis for a Banach space X𝑋X with biorthogonal functionals (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0*}_{j})_{j=1}^{\infty}. If (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} is a sequence in X𝑋X such that j=1xj0xj1xj0<1superscriptsubscript𝑗1normsuperscriptsubscript𝑥𝑗0subscriptsuperscript𝑥1𝑗normsubscriptsuperscript𝑥0𝑗1\sum_{j=1}^{\infty}\|x_{j}^{0}-x^{1}_{j}\|\|x^{0*}_{j}\|<1 then (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} is a Schauder basis for X𝑋X and is equivalent to (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty}.

Note that if (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} is a perturbation of (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} in the sense of Theorem 5.1, then every weaving is also a perturbation of (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} and hence (xj1)j=1superscriptsubscriptsubscriptsuperscript𝑥1𝑗𝑗1(x^{1}_{j})_{j=1}^{\infty} and (xj0)j=1superscriptsubscriptsubscriptsuperscript𝑥0𝑗𝑗1(x^{0}_{j})_{j=1}^{\infty} are woven. In other words, a perturbation of a Schauder basis is a Schauder basis which is woven with it. Perturbation theorems have been considered for Hilbert space frames and Schauder frames as well [BCGLL, CLZ]. However, we note that perturbation theorems are particularly natural in the context of approximate Schauder frames because when a Schauder frame is perturbed, one might expect the frame operator to be perturbed as well. Hence, the perturbation of a Schauder frame would be expected to be an approximate Schauder frame, and indeed the name “approximate Schauder frame” was chosen for this very reason. We prove two perturbation theorems for approximate Schauder frames and prove they result in woven approximate Schauder frames.

Theorem 5.2.

Suppose (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is a C𝐶C-suppression unconditional approximate Schauder frame for X𝑋X with frame operator S𝑆S. If T𝑇T is a bounded operator satisfying

IdT<C1,norm𝐼𝑑𝑇superscript𝐶1\|Id-T\|<C^{-1},

then (Txi,fi)i=1superscriptsubscript𝑇subscript𝑥𝑖subscript𝑓𝑖𝑖1(Tx_{i},f_{i})_{i=1}^{\infty} is an unconditional approximate Schauder frame for X𝑋X and is woven with (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty}.

Proof.

Set (xi0,fi0)i=1=(xi,fi)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖0𝑖1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i}^{0},f_{i}^{0})_{i=1}^{\infty}=(x_{i},f_{i})_{i=1}^{\infty} and (xi1,fi1)i=1=(Txi,fi)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1superscriptsubscript𝑓𝑖1𝑖1superscriptsubscript𝑇subscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i}^{1},f_{i}^{1})_{i=1}^{\infty}=(Tx_{i},f_{i})_{i=1}^{\infty} for ease of notation. Let xX𝑥𝑋x\in X and σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}}. We will first prove that the series i=1fiσ(i)(x)xiσ(i)superscriptsubscript𝑖1superscriptsubscript𝑓𝑖𝜎𝑖𝑥superscriptsubscript𝑥𝑖𝜎𝑖\sum_{i=1}^{\infty}f_{i}^{\sigma(i)}(x)x_{i}^{\sigma(i)} is unconditionally Cauchy and hence the frame operator of (xiσ(i),fiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖superscriptsubscript𝑓𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)},f_{i}^{\sigma(i)})_{i=1}^{\infty} is well defined. Let ε>0𝜀0\varepsilon>0 and choose N𝑁N\in\mathbb{N} so that for all J𝐽J\subseteq\mathbb{N} with min(J)N𝐽𝑁\min(J)\geq N, we have that (1+T)iJfi(x)xi<ε1norm𝑇normsubscript𝑖𝐽subscript𝑓𝑖𝑥subscript𝑥𝑖𝜀(1+\|T\|)\|\sum_{i\in J}f_{i}(x)x_{i}\|<\varepsilon. For a fixed J𝐽J\subseteq\mathbb{N} with min(J)N𝐽𝑁\min(J)\geq N, we have that

iJfiσ(i)(x)xiσ(i)normsubscript𝑖𝐽subscriptsuperscript𝑓𝜎𝑖𝑖𝑥subscriptsuperscript𝑥𝜎𝑖𝑖\displaystyle\bigg{\|}\sum_{i\in J}f^{\sigma(i)}_{i}(x)x^{\sigma(i)}_{i}\bigg{\|} =iJσ1(0)fi(x)xi+iJσ1(1)fi(x)T(xi)absentnormsubscript𝑖𝐽superscript𝜎10subscript𝑓𝑖𝑥subscript𝑥𝑖subscript𝑖𝐽superscript𝜎11subscript𝑓𝑖𝑥𝑇subscript𝑥𝑖\displaystyle=\bigg{\|}\sum_{i\in J\cap\sigma^{-1}(0)}f_{i}(x)x_{i}+\sum_{i\in J\cap\sigma^{-1}(1)}f_{i}(x)T(x_{i})\bigg{\|}
iJσ1(0)fi(x)xi+iJσ1(1)fi(x)xiTabsentnormsubscript𝑖𝐽superscript𝜎10subscript𝑓𝑖𝑥subscript𝑥𝑖normsubscript𝑖𝐽superscript𝜎11subscript𝑓𝑖𝑥subscript𝑥𝑖norm𝑇\displaystyle\leq\bigg{\|}\sum_{i\in J\cap\sigma^{-1}(0)}f_{i}(x)x_{i}\bigg{\|}+\bigg{\|}\sum_{i\in J\cap\sigma^{-1}(1)}f_{i}(x)x_{i}\bigg{\|}\|T\|
<ε.absent𝜀\displaystyle<\varepsilon.

Hence, the frame operator of (xiσ(i),fiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖superscriptsubscript𝑓𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)},f_{i}^{\sigma(i)})_{i=1}^{\infty} is well defined and converges unconditionally.

Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and let Sσsubscript𝑆𝜎S_{\sigma} be the frame operator of (xiσ(i),fiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖superscriptsubscript𝑓𝑖𝜎𝑖𝑖1(x^{\sigma(i)}_{i},f_{i}^{\sigma(i)})_{i=1}^{\infty}. We will prove that IdSσS1<1norm𝐼𝑑subscript𝑆𝜎superscript𝑆11\|Id-S_{\sigma}S^{-1}\|<1, hence showing that Sσsubscript𝑆𝜎S_{\sigma} is bounded with bounded inverse. If xX𝑥𝑋x\in X and y=S1x𝑦superscript𝑆1𝑥y=S^{-1}x, then

(IdSσS1)xnorm𝐼𝑑subscript𝑆𝜎superscript𝑆1𝑥\displaystyle\left\|(Id-S_{\sigma}S^{-1})x\right\| =SySσyabsentnorm𝑆𝑦subscript𝑆𝜎𝑦\displaystyle=\|Sy-S_{\sigma}y\|
=i=1fi0(y)xi0i=1fiσ(i)(y)xiσ(i)absentnormsuperscriptsubscript𝑖1subscriptsuperscript𝑓0𝑖𝑦subscriptsuperscript𝑥0𝑖superscriptsubscript𝑖1subscriptsuperscript𝑓𝜎𝑖𝑖𝑦subscriptsuperscript𝑥𝜎𝑖𝑖\displaystyle=\left\|\sum_{i=1}^{\infty}f^{0}_{i}(y)x^{0}_{i}-\sum_{i=1}^{\infty}f^{\sigma(i)}_{i}(y)x^{\sigma(i)}_{i}\right\|
=iσ1(1)[fi(y)xifi(y)T(xi)]absentnormsubscript𝑖superscript𝜎11delimited-[]subscript𝑓𝑖𝑦subscript𝑥𝑖subscript𝑓𝑖𝑦𝑇subscript𝑥𝑖\displaystyle=\left\|\sum_{i\in\sigma^{-1}(1)}\big{[}f_{i}(y)x_{i}-f_{i}(y)T(x_{i})\big{]}\right\|
iσ1(1)fi(y)xiIdTabsentnormsubscript𝑖superscript𝜎11subscript𝑓𝑖𝑦subscript𝑥𝑖norm𝐼𝑑𝑇\displaystyle\leq\left\|\sum_{i\in\sigma^{-1}(1)}f_{i}(y)x_{i}\right\|\|Id-T\|
CSyIdTabsent𝐶norm𝑆𝑦norm𝐼𝑑𝑇\displaystyle\leq C\|Sy\|\|Id-T\|
=CxIdT.absent𝐶norm𝑥norm𝐼𝑑𝑇\displaystyle=C\|x\|\|Id-T\|.

Thus, IdSσS1CIdT<1norm𝐼𝑑subscript𝑆𝜎superscript𝑆1𝐶norm𝐼𝑑𝑇1\|Id-S_{\sigma}S^{-1}\|\leq C\|Id-T\|<1, and hence Sσsubscript𝑆𝜎S_{\sigma} is bounded with bounded inverse. ∎

Note that Theorem 5.2 is only stated for the case that (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is an unconditional approximate Schauder frame. It is natural to ask if the same conclusion holds for conditional approximate Schauder frames as well. This is not possible because if (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is a conditional approximate Schauder frame and α𝛼\alpha is any scalar other than 111 then (xi,fi)i=1superscriptsubscriptsubscript𝑥𝑖subscript𝑓𝑖𝑖1(x_{i},f_{i})_{i=1}^{\infty} is not woven with (αxi,fi)i=1superscriptsubscript𝛼subscript𝑥𝑖subscript𝑓𝑖𝑖1(\alpha x_{i},f_{i})_{i=1}^{\infty}. However, the following perturbation result holds for both unconditional and conditional approximate Schauder frames.

Theorem 5.3.

If (xi0,fi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖0𝑖1(x_{i}^{0},f_{i}^{0})_{i=1}^{\infty} is an approximate Schauder frame for X𝑋X with frame operator S𝑆S and (xi1,fi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1superscriptsubscript𝑓𝑖1𝑖1(x_{i}^{1},f_{i}^{1})_{i=1}^{\infty} is a sequence in X×X𝑋superscript𝑋X\times X^{*} satisfying

i=1(fi0fi1xi0+xi0xi1fi1)<S11superscriptsubscript𝑖1normsuperscriptsubscript𝑓𝑖0superscriptsubscript𝑓𝑖1normsuperscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑥𝑖1normsuperscriptsubscript𝑓𝑖1superscriptnormsuperscript𝑆11\sum_{i=1}^{\infty}\left(\|f_{i}^{0}-f_{i}^{1}\|\|x_{i}^{0}\|+\|x_{i}^{0}-x_{i}^{1}\|\|f_{i}^{1}\|\right)<\|S^{-1}\|^{-1}

then (xi1,fi1)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖1superscriptsubscript𝑓𝑖1𝑖1(x_{i}^{1},f_{i}^{1})_{i=1}^{\infty} is an approximate Schauder frame for X𝑋X that is woven with (xi0,fi0)i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖0𝑖1(x_{i}^{0},f_{i}^{0})_{i=1}^{\infty}.

Proof.

Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and xX𝑥𝑋x\in X. We first show that the series i=1fiσ(i)(x)xiσ(i)superscriptsubscript𝑖1superscriptsubscript𝑓𝑖𝜎𝑖𝑥superscriptsubscript𝑥𝑖𝜎𝑖\sum_{i=1}^{\infty}f_{i}^{\sigma(i)}(x)x_{i}^{\sigma(i)} is Cauchy and hence the frame operator of (xiσ(i),fiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖superscriptsubscript𝑓𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)},f_{i}^{\sigma(i)})_{i=1}^{\infty} is well defined. Let ε>0𝜀0\varepsilon>0 and choose N𝑁N\in\mathbb{N} so that for all m>nN𝑚𝑛𝑁m>n\geq N,

i=nmfi0(x)xi0+i=nm(fi0fi1xi0+xi0xi1fi1)x<ϵ.normsuperscriptsubscript𝑖𝑛𝑚superscriptsubscript𝑓𝑖0𝑥superscriptsubscript𝑥𝑖0superscriptsubscript𝑖𝑛𝑚normsuperscriptsubscript𝑓𝑖0superscriptsubscript𝑓𝑖1normsuperscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑥𝑖1normsuperscriptsubscript𝑓𝑖1norm𝑥italic-ϵ\left\|\sum_{i=n}^{m}f_{i}^{0}(x)x_{i}^{0}\right\|+\sum_{i=n}^{m}(\|f_{i}^{0}-f_{i}^{1}\|\|x_{i}^{0}\|+\|x_{i}^{0}-x_{i}^{1}\|\|f_{i}^{1}\|)\|x\|<\epsilon.

Thus,

i=nmfiσ(i)(x)xiσ(i)normsuperscriptsubscript𝑖𝑛𝑚superscriptsubscript𝑓𝑖𝜎𝑖𝑥superscriptsubscript𝑥𝑖𝜎𝑖\displaystyle\left\|\sum_{i=n}^{m}f_{i}^{\sigma(i)}(x)x_{i}^{\sigma(i)}\right\| i=nmfi0(x)xi0+i[n,m]σ1(1)[fi0(x)xi0fi1(x)xi1]absentnormsuperscriptsubscript𝑖𝑛𝑚superscriptsubscript𝑓𝑖0𝑥superscriptsubscript𝑥𝑖0normsubscript𝑖𝑛𝑚superscript𝜎11delimited-[]superscriptsubscript𝑓𝑖0𝑥superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑥superscriptsubscript𝑥𝑖1\displaystyle\leq\left\|\sum_{i=n}^{m}f_{i}^{0}(x)x_{i}^{0}\right\|+\left\|\sum_{i\in[n,m]\cap\sigma^{-1}(1)}\big{[}f_{i}^{0}(x)x_{i}^{0}-f_{i}^{1}(x)x_{i}^{1}\big{]}\right\|
i=nmfi0(x)xi0+i=nm[fi0(y)xi0fi1(y)xi0+fi1(x)xi0fi1(x)xi1]absentnormsuperscriptsubscript𝑖𝑛𝑚superscriptsubscript𝑓𝑖0𝑥superscriptsubscript𝑥𝑖0superscriptsubscript𝑖𝑛𝑚delimited-[]normsuperscriptsubscript𝑓𝑖0𝑦superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑦superscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑓𝑖1𝑥superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑥superscriptsubscript𝑥𝑖1\displaystyle\leq\left\|\sum_{i=n}^{m}f_{i}^{0}(x)x_{i}^{0}\right\|+\sum_{i=n}^{m}\big{[}\left\|f_{i}^{0}(y)x_{i}^{0}-f_{i}^{1}(y)x_{i}^{0}\right\|+\left\|f_{i}^{1}(x)x_{i}^{0}-f_{i}^{1}(x)x_{i}^{1}\right\|\big{]}
i=nmfi0(x)xi0+i=nm(fi0fi1xi0+xi0xi1fi1)xabsentnormsuperscriptsubscript𝑖𝑛𝑚superscriptsubscript𝑓𝑖0𝑥superscriptsubscript𝑥𝑖0superscriptsubscript𝑖𝑛𝑚normsuperscriptsubscript𝑓𝑖0superscriptsubscript𝑓𝑖1normsuperscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑥𝑖1normsuperscriptsubscript𝑓𝑖1norm𝑥\displaystyle\leq\left\|\sum_{i=n}^{m}f_{i}^{0}(x)x_{i}^{0}\right\|+\sum_{i=n}^{m}(\|f_{i}^{0}-f_{i}^{1}\|\|x_{i}^{0}\|+\|x_{i}^{0}-x_{i}^{1}\|\|f_{i}^{1}\|)\|x\|
<ϵ.absentitalic-ϵ\displaystyle<\epsilon.

Hence, the frame operator of (xiσ(i),fiσ(i))i=1superscriptsubscriptsuperscriptsubscript𝑥𝑖𝜎𝑖superscriptsubscript𝑓𝑖𝜎𝑖𝑖1(x_{i}^{\sigma(i)},f_{i}^{\sigma(i)})_{i=1}^{\infty} is well defined.

Let σ{0,1}𝜎superscript01\sigma\in\{0,1\}^{\mathbb{N}} and let T𝑇T be the frame operator of (xiσ(i),fiσ(i))i=1superscriptsubscriptsubscriptsuperscript𝑥𝜎𝑖𝑖superscriptsubscript𝑓𝑖𝜎𝑖𝑖1(x^{\sigma(i)}_{i},f_{i}^{\sigma(i)})_{i=1}^{\infty}. We will prove that IdTS1<1norm𝐼𝑑𝑇superscript𝑆11\|Id-TS^{-1}\|<1, hence showing that T𝑇T is bounded with bounded inverse. If xX𝑥𝑋x\in X and y=S1x𝑦superscript𝑆1𝑥y=S^{-1}x, then

(IdTS1)xnorm𝐼𝑑𝑇superscript𝑆1𝑥\displaystyle\left\|(Id-TS^{-1})x\right\| =SyTyabsentnorm𝑆𝑦𝑇𝑦\displaystyle=\|Sy-Ty\|
=i=1fi0(y)xi0i=1fiσ(i)(y)xiσ(i)absentnormsuperscriptsubscript𝑖1subscriptsuperscript𝑓0𝑖𝑦subscriptsuperscript𝑥0𝑖superscriptsubscript𝑖1subscriptsuperscript𝑓𝜎𝑖𝑖𝑦subscriptsuperscript𝑥𝜎𝑖𝑖\displaystyle=\left\|\sum_{i=1}^{\infty}f^{0}_{i}(y)x^{0}_{i}-\sum_{i=1}^{\infty}f^{\sigma(i)}_{i}(y)x^{\sigma(i)}_{i}\right\|
=iσ1(1)[fi0(y)xi0fi1(y)xi1]absentnormsubscript𝑖superscript𝜎11delimited-[]superscriptsubscript𝑓𝑖0𝑦superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑦superscriptsubscript𝑥𝑖1\displaystyle=\left\|\sum_{i\in\sigma^{-1}(1)}\big{[}f_{i}^{0}(y)x_{i}^{0}-f_{i}^{1}(y)x_{i}^{1}\big{]}\right\|
i=1fi0(y)xi0fi1(y)xi1absentsuperscriptsubscript𝑖1normsuperscriptsubscript𝑓𝑖0𝑦superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑦superscriptsubscript𝑥𝑖1\displaystyle\leq\sum_{i=1}^{\infty}\|f_{i}^{0}(y)x_{i}^{0}-f_{i}^{1}(y)x_{i}^{1}\|
i=1fi0(y)xi0fi1(y)xi0+fi1(y)xi0fi1(y)xi1absentsuperscriptsubscript𝑖1normsuperscriptsubscript𝑓𝑖0𝑦superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑦superscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑓𝑖1𝑦superscriptsubscript𝑥𝑖0superscriptsubscript𝑓𝑖1𝑦superscriptsubscript𝑥𝑖1\displaystyle\leq\sum_{i=1}^{\infty}\|f_{i}^{0}(y)x_{i}^{0}-f_{i}^{1}(y)x_{i}^{0}\|+\|f_{i}^{1}(y)x_{i}^{0}-f_{i}^{1}(y)x_{i}^{1}\|
(i=1fi0fi1xi0+xi0xi1fi1)yabsentsuperscriptsubscript𝑖1normsuperscriptsubscript𝑓𝑖0subscriptsuperscript𝑓1𝑖normsuperscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑥𝑖1normsuperscriptsubscript𝑓𝑖1norm𝑦\displaystyle\leq\left(\sum_{i=1}^{\infty}\|f_{i}^{0}-f^{1}_{i}\|\|x_{i}^{0}\|+\|x_{i}^{0}-x_{i}^{1}\|\|f_{i}^{1}\|\right)\|y\|
(i=1fi0fi1xi0+xi0xi1fi1)S1x.absentsuperscriptsubscript𝑖1normsuperscriptsubscript𝑓𝑖0subscriptsuperscript𝑓1𝑖normsuperscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑥𝑖1normsuperscriptsubscript𝑓𝑖1normsuperscript𝑆1norm𝑥\displaystyle\leq\left(\sum_{i=1}^{\infty}\|f_{i}^{0}-f^{1}_{i}\|\|x_{i}^{0}\|+\|x_{i}^{0}-x_{i}^{1}\|\|f_{i}^{1}\|\right)\|S^{-1}\|\|x\|.

Thus, IdTS1(i=1fi0fi1xi0+xi0xi1fi1)S1<1norm𝐼𝑑𝑇superscript𝑆1superscriptsubscript𝑖1normsuperscriptsubscript𝑓𝑖0subscriptsuperscript𝑓1𝑖normsuperscriptsubscript𝑥𝑖0normsuperscriptsubscript𝑥𝑖0superscriptsubscript𝑥𝑖1normsuperscriptsubscript𝑓𝑖1normsuperscript𝑆11\|Id-TS^{-1}\|\leq\left(\sum_{i=1}^{\infty}\|f_{i}^{0}-f^{1}_{i}\|\|x_{i}^{0}\|+\|x_{i}^{0}-x_{i}^{1}\|\|f_{i}^{1}\|\right)\|S^{-1}\|<1 proving that T𝑇T is bounded with bounded inverse. ∎

References

  • [BCGLL] T. Bemrose, P. G. Casazza, K. Gröchenig, M. C. Lammers, and R. G. Lynch, Weaving frames, submitted, see arXiv:1503.03947.
  • [CDOSZ] P. G. Casazza, S. J. Dilworth, E. Odell, Th. Schlumprecht, and A. Zsak, Coefficient Quantization for Frames in Banach Spaces, J. Math. Anal. Appl. 348 (2008), 66–86.
  • [CHL] P. G. Casazza, D. Han, and D. Larson, Frames for Banach spaces, Cont. Math., 247 (1999), 149-182.
  • [CK] P.G. Casazza and G. Kutyniok, Finite frames: Theory and applications, Springer (2013).
  • [CL] P. G. Casazza and R. G. Lynch, Weaving properties of Hilbert space frames, Proceedings of SampTA (2015).
  • [CL2] P. G. Casazza and R. G. Lynch, A brief introduction to Hilbert space frame theory and its applications, submitted.
  • [CLZ] D.Y. Chen, L. Li, and B.T. Zheng, Perturbations of frames, Acta Math. Sinica, English Series, 30 (2014), no. 7, 1089–1108.
  • [C] O. Christensen, An Introduction to Frames and Riesz Bases, Birkhäuser, Boston (2003).
  • [DV] Deepshikha and L.K. Vashisht, On weaving frames, preprint.
  • [FHHMPZ] M. Fabian, P. Habala, P. Hájek, V. Montesinos Santalucía, J. Pelant, and V. Zizler, Functional Analysis and Infinite-Dimensional Geometry, CMS Books Math./Ouvrages Math. SMC, vol. 8 Springer, New York (2001).
  • [FOSZ] D. Freeman, E. Odell, Th. Schlumprecht, and A. Zsak,Unconditional structures of translates for Lp(d)subscript𝐿𝑝superscript𝑑L_{p}(\mathbb{R}^{d}). Israel J. Math., 203 (2014), no. 1, 189–209.
  • [FG] H. G. Feichtinger and K. Gröchenig, A unified approach to atomic decompositions via integrable group representations, In: Proc. Conf. Function Spaces and Applications, Lecture Notes Math. 1302. BerlinHeidelberg New York: Springer (1988), 52-73.
  • [G] K. Gröchenig, Describing functions: Atomic decompositions versus frames, Monatshefte fur Math. 112 (1991), 1–41.
  • [LT] J. Lindenstrauss and L. Tzafriri, Classical Banach spaces I and II, Springer, New York (1977).