Coordinate systems in Banach spaces and lattices

Abstract.

Using methods of descriptive set theory, in particular, the determinacy of infinite games of perfect information, we answer several questions from the literature, including [taylor, Question 2.5, Question 2.10], [gumenchuk, Problem 1.3, Problem 5.2] and [Rancourt, Question 2], regarding different notions of bases in Banach spaces and lattices.

For the case of Banach lattices, our results follow from a general theorem stating that (under the assumption of projective determinacy), every order basis $(e_{n})$ for a Banach lattice $X=[e_{n}]$ is also a $\sigma$-order basis for $X$, every $\sigma$-order basis for $X$ is a uniform basis for $X$, and every uniform basis is Schauder.

Regarding Banach spaces, we address two problems concerning filter Schauder bases for Banach spaces, i.e., in which the norm convergence of partial sums is replaced by norm convergence along some appropriate filter on $\mathbb{N}$. We first provide an example of a Banach space admitting such a filter Schauder basis, but no ordinary Schauder basis. Secondly, we show that every filter Schauder basis with respect to an analytic filter is also a filter Schauder basis with respect to a Borel filter.

1. Introduction

1.1. Order bases in Banach lattices

Suppose $X$ is a Banach lattice. Then the lattice structure on $X$ gives rise to three classical notions of sequential convergence, not available in a general Banach space. Namely,

• a sequence $(x_{n})_{n=1}^{\infty}$ converges uniformly to $x$, denoted $x_{n}\mathop{\overset{\sf u}{\longrightarrow}}x$, if there is some $z\in X_{+}$ so that

  $$\forall m\;\forall^{\infty}n\;|x_{n}-x|\leqslant\tfrac{z}{m},$$

• a sequence $(x_{n})_{n=1}^{\infty}$ $\sigma$-order converges to $x$, denoted $x_{n}\mathop{\overset{\sf\sigma\sf o}{\longrightarrow}}x$, if there is some sequence $z_{m}\downarrow 0$ so that

  $$\forall m\;\forall^{\infty}n\;|x_{n}-x|\leqslant z_{m},$$

• a sequence $(x_{n})_{n=1}^{\infty}$ order converges to $x$, denoted $x_{n}\mathop{\overset{\sf o}{\longrightarrow}}x$, if there is some net $z_{\mu}\downarrow 0$ so that

  $$\forall\mu\;\forall^{\infty}n\;|x_{n}-x|\leqslant z_{\mu}.$$

Here the notation $\forall^{\infty}n$ means for all but finitely many $n$, i.e., $\exists N\;\forall n\geqslant N$, while $z_{m}\downarrow 0$ and $z_{\mu}\downarrow 0$ mean that $(z_{m})$ and $(z_{\mu})$ are decreasing and have infimum $0$. It can be shown that, in all cases above, the limit is unique whenever it exists. Thus, if $C$ is one of the above notions of convergence and $\sum_{n=1}^{\infty}x_{n}$ is a series in $X$, we can unambiguously write

$$x=^{C}\sum_{n=1}^{\infty}x_{n}$$

to denote that the sequence of partial sums $(\sum_{n=1}^{m}x_{n})$ $C$-converges to $x$.

All three notions of convergence are evidently compatible with the algebraic structure of $X$, in the sense that if $(x_{n})$ and $(y_{n})$ converge to $x$ and $y$ respectively, then the sequence of sums $(x_{n}+y_{n})$ converges to $x+y$ and similarly for scalar products. Nevertheless, neither uniform nor order convergence arise in general from Hausdorff topologies on $X$ [Tay, Section 18].

As is evident from the definitions, uniform convergence implies norm convergence and $\sigma$-order convergence, whereas $\sigma$-order convergence implies order convergence. However, in absence of other hypotheses on $X$, no other implications hold, which is recorded in Figure 1.

Figure 1. Implications between convergence types.

Definition 1.1.

Let $X$ be a Banach lattice and $C$ one of the following convergence types: norm, uniform, order or $\sigma$-order. Then a sequence $(e_{n})_{n=1}^{\infty}$ in $X$ is said to be a $C$-basis for $X$ provided that, for every $x\in X$, there is a unique sequence of scalars $(a_{n})\in\mathbb{R}^{\mathbb{N}}$ so that

$$x=^{C}\sum_{n=1}^{\infty}a_{n}e_{n}.$$

For example, it is easy to see that the standard unit vector sequence $(e_{n})$ forms an order basis for each of the spaces $\ell_{p}$, $1\leqslant p\leqslant\infty$, and also for $c_{0}$. We are thus in the unfamiliar situation that the same sequence $(e_{n})$ is an order basis both for $\ell_{\infty}$ and its subspace (even sublattice) $c_{0}$. In particular, we see that the norm-closed linear span $[e_{n}]$ of $(e_{n})$ may be strictly smaller than the lattice $X$ for which it is an order basis.

When $(e_{n})$ is a $C$-basis for $X$, we may define functionals $X\mathop{\overset{e^{\sharp}_{k}}{\longrightarrow}}\mathbb{R}$ by letting

$${e^{\sharp}_{k}}(x)=a_{k},$$

where $(a_{n})$ is the uniquely defined sequence referenced above. Similarly, we let $X\mathop{\overset{P_{m}}{\longrightarrow}}[e_{1},\ldots,e_{m}]$ denote the corresponding sequence of basis projections,

$$P_{m}(x)=\sum_{n=1}^{m}e^{\sharp}_{n}(x)e_{n}.$$

Since the sequence constantly equal to $e_{n}$ will $C$-converge to $e_{n}$, we find that ${e^{\sharp}_{k}}(e_{n})=\delta_{k,n}$ for all $k,n$, that is, the functionals $e^{\sharp}_{k}$ are biorthogonal to the sequence $(e_{n})$. Observe however that a priori it is not clear that the functionals $e^{\sharp}_{k}$ or the operators $P_{m}$ are continuous (with respect to the norm topology on $X$; Banach lattices often admit no order continuous linear functionals).

Norm bases are of course more commonly known as Schauder bases and we shall employ that terminology here. Moreover, it is a classical result [Banach, p. 111] that the biorthogonal functionals $e^{\sharp}_{k}$ associated to a Schauder basis are always continuous.

Biorthogonal functionals associated with some sequence are typically denoted by $e^{*}_{k}$, but, since the very continuity of the functionals is at play here, we shall only use the notation $e^{*}_{k}$ if we already know that they are continuous.

In our first main theorem, which settles the relationships between the different types of bases, for some of the implications we resort to additional set theoretical axioms, namely the determinacy of certain infinite games on $\mathbb{N}$ [Kechris, Definition 26.3]. Nevertheless, this usage should not be too disturbing as projective determinacy is arguably part of the right set theoretical foundations of mathematics.

Theorem 1.2 (Projective determinacy).

Suppose $(e_{n})$ is a sequence of vectors in a Banach lattice $X=[e_{n}]$ and $(e^{\sharp}_{n})$ is a sequence of (possibly discontinuous) biorthogonal functionals for $(e_{n})$. Consider the following properties:

1. $(e_{n})$ is an order basis for $X$ with corresponding functionals $(e^{\sharp}_{n})$,

2. $(e_{n})$ is a $\sigma$-order basis for $X$ with corresponding functionals $(e^{\sharp}_{n})$,

3. $(e_{n})$ is a uniform basis for $X$ with corresponding functionals $(e^{\sharp}_{n})$,

4. $(e_{n})$ is a Schauder basis for $X$ with corresponding functionals $(e^{\sharp}_{n})$.

Then (1)$\Rightarrow$(2)$\Rightarrow$(3)$\Rightarrow$(4) and so the $e_{n}^{\sharp}$ are continuous in all cases above.

As hinted above, projective determinacy is only needed for parts of the theorem. Indeed, (3)$\Rightarrow$(4) holds without any additional set theoretical assumptions, whereas (2)$\Rightarrow$(3) is proved under the weaker assumption of analytic determinacy.

Coupled with [taylor, Theorem 2.1], we obtain the following characterisation of uniform bases for Banach lattices.

Corollary 1.3.

The following conditions are equivalent for a sequence $(e_{n})$ of non-zero vectors in a Banach lattice $X=[e_{n}]$:

1. $(e_{n})$ is a uniform basis for $X$,

2. $(e_{n})$ is a Schauder basis for $X$ so that, for every $x\in X$, the sequence of partial sums

   $$P_{m}x=\sum_{n=1}^{m}e^{*}_{n}(x)e_{n}$$

   is order bounded,

3. there is a constant $M$ so that, for all finite tuples of scalars $(a_{n})_{n=1}^{m}$ one has

   $$\begin{split}{\Bigg{\lVert}\bigvee_{k=1}^{m}\Big{|}\sum_{n=1}^{k}a_{n}e_{n}\Big{|}\,\Bigg{\rVert}\leqslant M\Bigg{\|}\sum_{n=1}^{m}a_{n}e_{n}\Bigg{\|}.}\end{split}$$

Remark 1.4.

It is a classical fact that, if $X$ is a Banach lattice of measurable functions, we have $f_{n}\mathop{\underset{n}{\overset{\sf o}{\longrightarrow}}}f$ if and only if $f_{n}\mathop{\underset{n}{\overset{a.e.}{\longrightarrow}}}f$ and, moreover, there exists a $g\in X_{+}$ satisfying $|f_{n}|\leqslant g$ for all $n$ (i.e., the sequence $(f_{n})$ is order bounded). Therefore, a uniform basis $(e_{n})$ of a Banach lattice $X$ can be thought of as a coordinate system which guarantees both norm and dominated almost everywhere convergence of the basis expansions.

Statement (3) of Corollary 1.3 is simply the standard inequality for Schauder basic sequences with the supremum $\bigvee_{k=1}^{m}$ pulled inside the norm. The equivalence between (1) and (3) in Corollary 1.3 therefore shows that bounding the maximal function of a basic sequence is equivalent to establishing strong convergence properties of the series, even in the general setting of Banach lattices.

We remark that ${\sf u}$-bases (and ${\sf u}$-basic sequences; see Remark 2.6) occur frequently in applications. For example, it follows from Doob’s inequality that martingale difference sequences in $L_{p}(\mu)$ ($1<p<\infty$ and $\mu$ a probability measure) are ${\sf u}$-basic. On the other hand, the combination of the Burkholder–Davis–Gundy and Khintchine inequalities yields that unconditional blocks of the Haar basis in $L_{1}[0,1]$ are ${\sf u}$-basic, and the Carleson–Hunt theorem [Hunt] establishes the inequality in Corollary 1.3 (3) for the trigonometric basis. For several more examples and non-examples of ${\sf u}$-basic sequences, the reader may consult [Tay, taylor].

Analogously to Corollary 1.3, we may characterise order bases as follows, and note that there is a similar characterisation of $\sigma$-order bases.

Corollary 1.5 (Projective determinacy).

The following conditions are equivalent for a sequence $(e_{n})$ of non-zero vectors in a Banach lattice $X=[e_{n}]$:

1. $(e_{n})$ is an order basis for $X$,

2. $(e_{n})$ is a uniform basis for $X$ such that $0=^{\sf o}\sum_{n=1}^{\infty}0e_{n}$ is the unique order expansion of $0$.

Theorem 1.2 immediately solves several problems listed in the literature regarding the relationships between these basis notions.

Problem 1.6.

[taylor, Question 2.10] If $(e_{n})$ is simultaneously a Schauder basis and a $\sigma$-order basis for a Banach lattice $X$, do the coefficients in the norm and in the order expansions of the same vector agree? That is, are the two associated sets of biorthogonal functionals equal?

By the implications (2)$\Rightarrow$(3)$\Rightarrow$(4), proved under the assumption of analytic determinacy, the answer to Problem 1.6 is therefore positive.

For our next applications, we need to recall some facts about order continuous Banach lattices from [MR0569521].

Lemma 1.7.

Suppose $X$ is an order continuous Banach lattice. Then, for all sequences $(x_{n})$ and vectors $x$,

$$x_{n}\mathop{\overset{\sf o}{\longrightarrow}}x\quad\Leftrightarrow\quad x_{n}\mathop{\overset{\sigma{\sf o}}{\longrightarrow}}x\quad\Leftrightarrow\quad x_{n}\mathop{\overset{\sf u}{\longrightarrow}}x.$$

Similarly, if $X$ is $\sigma$-order continuous, then

$$x_{n}\mathop{\overset{\sigma{\sf o}}{\longrightarrow}}x\quad\Leftrightarrow\quad x_{n}\mathop{\overset{\sf u}{\longrightarrow}}x.$$

Thus, in an order continuous Banach lattice $X$, the notions of order, $\sigma$-order and uniform bases coincide and they will have the same associated biorthogonal functionals. Similarly, if $X$ is $\sigma$-order continuous, the notions of $\sigma$-order and uniform bases coincide and have the same biorthogonal functionals. In particular, these types of bases will automatically also be Schauder bases with the same biorthogonal functionals.

Corollary 1.8.

Let $(e_{n})$ be an order basis for an order continuous Banach lattice. Then $(e_{n})$ is also a Schauder basis with the same biorthogonal functionals.

This in turn provides a positive answer to the following questions.

Problem 1.9.

[gumenchuk, Problem 1.3] Let $(x_{n})$ be an order basis of an order continuous Banach lattice E. Is then $(x_{n})$ a Schauder basis of $E$? What about $E=L_{p}$ with $1\leqslant p<\infty$?

In the same paper, the authors consider the specific example of $L_{1}$ and conjecture a negative answer to the following question.

Problem 1.10.

[gumenchuk, Problem 5.2] Does $L_{1}$ have a $\sigma$-order basis?

Regarding this, they show that $L_{1}$ does not admit a sequence $(e_{n})$ that is simultaneously a Schauder and an order basis for $L_{1}$ [gumenchuk, Theorem 5.1]. However, given that $L_{1}$ is order continuous, every order basis for $L_{1}$ is also a Schauder basis, which gives a negative answer to Problem 1.10.

Corollary 1.11.

The Banach lattice $L_{1}$ admits no $\sigma$-order basis.

A few cautionary remarks on the terminology are in order. Namely, our notion of $\sigma$-order convergence for sequences is simply called order convergence for sequences in [gumenchuk] and therefore our notion of $\sigma$-order basis is similarly designated order basis in [gumenchuk]. In the same paper, a sequence $(e_{n})$ in a Banach lattice $X$ which is simultaneously a Schauder basis and a $\sigma$-order basis for $X$ is denoted a bibasis. By Theorem 1.2, under analytic determinacy, bibases are thus simply $\sigma$-order bases. However, the authors exclusively work in $\sigma$-order continuous Banach lattices, where of course the notions of $\sigma$-order and uniform bases coincide. On the other hand, in [taylor], a sequence $(e_{n})$ in a Banach lattice $X$ is called a bibasis for $X$ provided that it is both a Schauder basis and a uniform basis. Because of Theorem 1.2, these two competing notions of bibases are superfluous as they just correspond to $\sigma$-order and uniform bases respectively. To avoid any confusion, we shall exclusively employ the terminology of Definition 1.1 and eschew the, in hindsight, unnecessary notion of bibasis.

1.2. Filter bases

The second topic of our study concerns a generalisation of Schauder bases in the context of general Banach spaces, not lattices. Assume that $F$ is a filter of subsets of $\mathbb{N}$, that is, $F\subseteq\mathcal{P}(\mathbb{N})$ is closed under taking intersections and supersets,

• $a,b\in F\;\Rightarrow\;a\cap b\in F$,

• $a\subseteq b\;\&\;a\in F\;\Rightarrow\;b\in F$.

For reasons that will become apparent later, we shall also assume that all filters are proper, i.e., $\emptyset\notin F$, and contain the Fréchet filter consisting of all cofinite subsets of $\mathbb{N}$. Recall that a sequence $(x_{n})$ is said to converge along $F$ to $x$, denoted $x_{n}\mathop{\underset{n\to F}{\overset{}{\longrightarrow}}}x$, if

$$\big{\{}m\in\mathbb{N}\;\big{|}\;\,\lVert x_{n}-x\rVert<\epsilon\big{\}}\in F$$

for all $\epsilon>0$. In complete analogy with Definition 1.1, we have the following definition due to M. Ganichev and V. Kadets [ganichev].

Definition 1.12.

A sequence $(e_{n})$ in a Banach space $X$ is said to be an $F$-basis for $X$ provided that, for all $x\in X$, there is a unique sequence $(a_{n})\in\mathbb{R}^{\mathbb{N}}$ so that $\sum_{n=1}^{m}a_{n}e_{n}\mathop{\underset{m\to F}{\overset{}{\longrightarrow}}}x$, which we denote by

$$x=^{F}\sum_{n=1}^{\infty}a_{n}e_{n}.$$

More generally, $(e_{n})$ is said to be a filter basis for $X$ if it is an $F$-basis for $X$ for some filter $F$, in which case $F$ is said to be compatible with $(e_{n})$.

Let us note that, if $F$ is just the Fréchet filter itself, then an $F$-basis $(e_{n})$ is nothing but a Schauder basis. Although [kadets, Example 1] provides a basis for $\ell_{2}$ with respect to the ideal of sets of density $1$, which however is not a Schauder basis, T. Kania asked whether there is an example of a Banach space without a Schauder basis that nevertheless has an $F$-basis for some appropriate filter $F$. We answer this by the following simple example.

Example 1.13 (A Banach space with a filter basis, but no Schauder basis).

Let $X$ be a Banach space with a finite-dimensional decomposition $(X_{n})_{n=1}^{\infty}$, but without a Schauder basis. That such spaces exist follows for example from [szarek, Theorem 1.1]. Choose now sequences $(e_{i})_{i\in\mathbb{N}}$ and $0=k_{0}<k_{1}<k_{2}<\ldots$ so that

$$\big{\{}e_{i}\;\big{|}\;k_{n-1}<i\leqslant{k_{n}}\big{\}}$$

is a basis for $X_{n}$ for all $n$. Let also

$$F=\big{\{}a\subseteq\mathbb{N}\;\big{|}\;k_{n}\in a\text{ for all but finitely many }n\big{\}}.$$

Since $(X_{n})$ is an F.D.D. for $X$, we have that, for every $x\in X$, there are unique vectors $x_{n}\in X_{n}$ so that $x=\sum_{n=1}^{\infty}x_{n}$. Writing $x_{n}=\sum_{i=k_{n-1}+1}^{k_{n}}a_{i}e_{i}$ for appropriate scalars $a_{i}\in\mathbb{R}$, we see that, for all $\epsilon>0$,

$$k_{n}\in\Big{\{}m\in\mathbb{N}\;\Big{|}\;\,\Big{\lVert}x-\sum_{i=1}^{m}a_{i}e_{i}\Big{\rVert}<\epsilon\Big{\}}$$

for all but finitely many $n$ and hence that the latter set belongs to $F$. Furthermore, by the uniqueness of the $x_{n}$, $(a_{i})$ is the only such sequence, which shows that $(e_{i})$ is an $F$-basis for $X$.

Note that, if $F$ is a filter and $(e_{n})$ is an $F$-basis for $X$, we may define the associated biorthogonal functionals $e^{\sharp}_{n}$ just as for order bases etc. However, it might be possible that $(e_{n})$ is simultaneously an $F^{\prime}$-basis for $X$ with respect to some other filter $F^{\prime}$, in which case it is unclear whether the biorthogonal functionals associated with $F^{\prime}$ are the same as those associated with $F$. Furthermore, even under additional set theoretical axioms, it is no longer clear whether any of these functionals are continuous. To discuss these issues, we must introduce a more refined concept.

Definition 1.14.

Let $(e_{n},e^{\sharp}_{n})$ be a biorthogonal system in a Banach space $X$, i.e., $(e_{n})$ is a sequence of vectors in $X$ and $e^{\sharp}_{n}\colon X\to\mathbb{R}$ are (possibly discontinuous) functionals biorthogonal to the $e_{n}$. We say that $(e_{n},e^{\sharp}_{n})$ is a filter basis system for $X$ provided that there is a filter $F$ so that $(e_{n})$ is an $F$-basis for $X$ with associated biorthogonal functionals $e^{\sharp}_{n}$. Such a filter $F$ is said to be compatible with $(e_{n},e^{\sharp}_{n})$.

Let us note that, to every filter basis system $(e_{n},e^{\sharp}_{n})$, there is a smallest compatible filter $F$, which we shall return to later on. Recall also that a sequence $(e_{n})$ in a Banach space is said to be minimal if, for all $k$, we have that $e_{k}\notin[e_{n}]_{n\neq k}$.

Lemma 1.15.

Let $(e_{n},e^{\sharp}_{n})$ be a filter basis system for $X$. Then the sequence $(e_{n})$ is minimal if and only if the functionals $e^{\sharp}_{n}$ are continuous.

Thus, the continuity of the associated biorthogonal functionals can be detected directly on the filter basis $(e_{n})$ itself without even involving the functionals.

Although Ganichev and Kadets [ganichev] operate with a slightly more general notion of filter basis, the following problem remains open even in our setting.

Problem 1.16.

[ganichev] Suppose $(e_{n})$ is a filter basis for a Banach space $X$. Is $(e_{n})$ necessarily a minimal sequence?

The papers [kania, Rancourt] address Problem 1.16 and show that a filter basis $(e_{n})$ is minimal if and only if it admits a compatible filter $F$ that is analytic when viewed as a subset of $\mathcal{P}(\mathbb{N})=\{0,1\}^{\mathbb{N}}$ [Rancourt, Theorem A, Theorem B]. In connection with this, [Rancourt, Question 2] asks whether it is possible to improve this so as to get the filter $F$ to be Borel. We resolve this even while keeping the associated biorthogonal functionals $e^{\sharp}_{n}$ fixed.

Theorem 1.17.

Let $(e_{n},e^{\sharp}_{n})$ be a filter basis system for $X$. Then the following are equivalent.

1. The functionals $e^{\sharp}_{n}$ are continuous,

2. the sequence $(e_{n})$ is minimal,

3. the smallest compatible filter is analytic,

4. there is a compatible analytic filter,

5. there is a compatible Borel filter.

1.3. A higher order Fatou property

It is of course natural to ask whether the use of projective and analytical determinacy in Theorem 1.2 is really necessary, i.e., if there is not some other more insightful proof bypassing these issues. We do not know the answer to this question, but note that the problem resides in the fact that the very notions of order and $\sigma$-order convergence are a priori of too high descriptive complexity. Indeed, in the proofs of Theorems 2.3 and 2.4, we are only able to show that, in a general separable Banach lattice $X$, the sets

$$\begin{split}{\Big{\{}\big{(}(x_{n}),x\big{)}\in X^{\mathbb{N}}\times X\;\Big{|}\;x_{n}\mathop{\underset{n}{\overset{\sf o}{\longrightarrow}}}x\Big{\}}\qquad\text{and}\qquad\Big{\{}\big{(}(x_{n}),x\big{)}\in X^{\mathbb{N}}\times X\;\Big{|}\;x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}x\Big{\}}}\end{split}$$

are respectively ${\bf\Pi}^{1}_{2}$ and ${\bf\Sigma}^{1}_{2}$. On the other hand, if we can show them to be ${\bf\Sigma}^{1}_{1}$ (i.e., analytic) instead, then our proofs no longer necessitate the additional set theoretical assumptions. This happens, for example, if the Banach lattice $X$ admits a countable $\pi$-basis, i.e., a countable set $P\subseteq X$ of positive elements so that, for every $x>0$, there is some $p\in P$ with $0<p\leqslant x$.

In Section 4 we introduce a hierarchy, indexed by countable ordinal numbers $\alpha$, of properties similar to the well-known Fatou property of separable Banach lattices (see Definition 4.8). Our main result in this context is the following.

Theorem 1.18.

The following conditions are equivalent for a separable Banach lattice $X$.

1. The set

   $$\Big{\{}(x_{n})_{n=1}^{\infty}\in X^{\mathbb{N}}\;\Big{|}\;x_{1}\geqslant x_{2}\geqslant\ldots\geqslant 0=\inf_{n}x_{n}\Big{\}}$$

   is Borel,

2. the set

   $$\Big{\{}\big{(}(x_{n})_{n=1}^{\infty},x\big{)}\in X^{\mathbb{N}}\times X\;\Big{|}\;x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}x\Big{\}}$$

   is analytic,

3. $X$ is $\alpha$-Fatou for some $\alpha<\omega_{1}$.

Furthermore, we show that, if a separable Banach lattice $X$ satisfies these three equivalent conditions, then every $\sigma$-order basis for $X$ is also a uniform basis for $X$ without any additional set theoretical assumptions. As noted above, this happens, for example, if $X$ has a countable $\pi$-basis.

In Section 5 we construct a sequence $(X_{\alpha})_{\alpha<\omega_{1}}$ of separable Banach lattices with countable $\pi$-bases so that $X_{\alpha}$ fails to be $\beta$-Fatou for all $\beta\leqslant\alpha$. However, the question of whether one can find a fixed separable Banach lattice failing to be $\alpha$-Fatou for all $\alpha<\omega_{1}$ remains open:

Problem 1.19.

Is every separable Banach lattice necessarily $\alpha$-Fatou for some $\alpha<\omega_{1}$?

An intriguing case is when $X$ is assumed to be $\sigma$-complete, that is, every countable set $\{z_{n}\}_{n\in\mathbb{N}}\subseteq X$ that is bounded above has a least upper bound, $\bigvee_{n\in\mathbb{N}}z_{n}\in X$. In this case, we have that

$$x_{n}\mathop{\underset{n}{\overset{\sf o}{\longrightarrow}}}x\quad\Leftrightarrow\quad x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}x$$

for all $(x_{n})$ and $x$, which means that both conditions are simultaneously ${\bf\Pi}^{1}_{2}$ and ${\bf\Sigma}^{1}_{2}$, that is, ${\bf\Delta}^{1}_{2}$. However, using the fact that separable $\sigma$-complete Banach lattices are order continuous [LT2, Proposition 1.a.7], it is easy to see that these conditions are in fact ${\bf\Sigma}^{1}_{1}$, which by Theorem 1.18 ensures that

$$z=\bigvee_{n\in\mathbb{N}}z_{n}$$

is Borel in the pair $\big{(}z,(z_{n})\big{)}$. Therefore, any counterexample to Problem 1.19 cannot be $\sigma$-complete.

Although Problem 1.19 remains open in general, our construction in Section 5 implies that there is no global Borel definition of when a sequence in a separable Banach lattice has infimum zero. In order to make such a statement precise, one needs to consider a Borel structure on the collection of all separable Banach lattices, as has been done by M. A. Tursi [TursiPhD] following the ideas initiated by Bossard in Banach space theory. Using this language, it can be derived from the results of this paper that the collection of pairs $(X,(x_{n}))$ such that $X$ is a separable Banach lattice and $(x_{n})_{n=1}^{\infty}\in X^{\mathbb{N}}$ satisfy $x_{1}\geqslant x_{2}\geqslant\ldots\geqslant 0=\inf_{n}x_{n}$ in $X$ constitutes a true coanalytic set.

2. Proofs for order bases

As is well-known from the case of Banach spaces, it is often useful to operate with basic sequences as opposed to bases. So let us introduce this notion in our context. Recall first that a sequence $(e_{n})$ in a Banach space $X$ is said to be Schauder basic in case it is a Schauder basis for its closed linear span $[e_{n}]$. When dealing with uniform, $\sigma$-order and order convergence, extra caution is required since the closed linear span $[e_{n}]$ of a sequence in $X$ need not be a sublattice. Furthermore, even the notions of $\sigma$-order and order convergence are not absolute, but depend on the ambient lattice. On the other hand, uniform convergence is absolute [taylor, Proposition 2.12]. In fact, as shown in Lemma 2.1 below, uniform convergence can be equivalently reformulated so as to avoid any reference to the ambient lattice.

Lemma 2.1.

For a sequence $(x_{n})$ and vector $x$ in a Banach lattice $X$, we have that $x_{n}\mathop{\underset{n}{\overset{\sf u}{\longrightarrow}}}x$ if and only if

$$\forall\epsilon>0\;\exists k\;\forall m\geqslant k\quad\Bigg{\lVert}\bigvee_{n=k}^{m}|x_{n}-x|\Bigg{\rVert}<\epsilon.$$

In particular, the set

$$\Big{\{}\big{(}(x_{n})_{n=1}^{\infty},x\big{)}\in X^{\mathbb{N}}\times X\;\Big{|}\;x_{n}\mathop{\underset{n}{\overset{\sf u}{\longrightarrow}}}x\Big{\}}$$

is Borel.

Proof.

Suppose first that $x_{n}\mathop{\underset{n}{\overset{\sf u}{\longrightarrow}}}x$ and find some $z>0$ so that

$$\forall l\;\forall^{\infty}n\;|x_{n}-x|\leqslant\tfrac{z}{l}.$$

Thus, if $\epsilon>0$ is given, choose $l$ large enough that $\lVert\tfrac{z}{l}\rVert<\epsilon$ and find $k$ so that $|x_{n}-x|\leqslant\tfrac{z}{l}$ for all $n\geqslant k$. We then see that

$$\Bigg{\lVert}\bigvee_{n=k}^{m}|x_{n}-x|\Bigg{\rVert}\leqslant\lVert\tfrac{z}{l}\rVert<\epsilon$$

for all $m\geqslant k$.

Conversely, suppose that

$$\forall\epsilon>0\;\exists k\;\forall m\geqslant k\quad\Bigg{\lVert}\bigvee_{n=k}^{m}|x_{n}-x|\Bigg{\rVert}<\epsilon.$$

We choose $0=k_{0}<k_{1}<k_{2}<\ldots$ so that

$$\Bigg{\lVert}\bigvee_{n=k_{l}}^{m}|x_{n}-x|\Bigg{\rVert}<\tfrac{1}{4^{l}}$$

whenever $l\geqslant 1$ and $m\geqslant k_{l}$. This implies that the series

$$\sum_{l=0}^{\infty}2^{l}\bigvee_{n=k_{l}}^{k_{l-1}-1}|x_{n}-x|$$

converges in norm to some element $z\in X_{+}$. Moreover, for any $l\geqslant 1$ and all $n\geqslant k_{l}$, we have that $2^{l}|x_{n}-x|\leqslant z$ and hence $|x_{n}-x|\leqslant\tfrac{z}{2^{l}}$. Therefore, $x_{n}\mathop{\underset{n}{\overset{\sf u}{\longrightarrow}}}x$. ∎

We may thus define a sequence $(e_{n})$ in a Banach lattice $X$ to be ${\sf u}$-basic in case, for every $x\in[e_{n}]$, there is a unique sequence $(a_{n})\in\mathbb{R}^{\mathbb{N}}$ so that $x=^{\sf u}\sum_{n=1}^{\infty}a_{n}e_{n}$. By Lemma 2.1, this notion is intrinsically defined and independent of the choice of the ambient Banach lattice, which we may therefore always assume to be the separable Banach lattice $[e_{n}]_{\wedge}\subseteq X$ generated by $(e_{n})$. Note however that $[e_{n}]$ itself will in general only be a Banach space, not a lattice. The corresponding biorthogonal functionals $e^{\sharp}_{k}\colon[e_{n}]\to\mathbb{R}$ are defined as before.

The following establishes the implication (3)$\Rightarrow$(4) of Theorem 1.2.

Theorem 2.2.

Suppose that $(e_{n})$ is a $\sf u$-basic sequence in a Banach lattice $X$. Then the biorthogonal functionals $e^{\sharp}_{n}$ are continuous and hence $(e_{n})$ is Schauder basic.

To simplify notation, it is slightly easier to work with the operator $[e_{n}]\mathop{\overset{E}{\longrightarrow}}\mathbb{R}^{\mathbb{N}}$ defined by $Ex=\big{(}e_{n}^{\sharp}(x)\big{)}_{n}$ in place of the biorthogonal functionals themselves. Observe that $E$ is continuous if and only if all the $e^{\sharp}_{n}$ are continuous.

Proof.

We recall that, since both $[e_{n}]$ and $\mathbb{R}^{\mathbb{N}}$ are separable Fréchet spaces and hence Polish spaces, the operator $E$ is continuous if and only if the graph $\mathcal{G}E\subseteq[e_{n}]\times\mathbb{R}^{\mathbb{N}}$ is Borel. Indeed, if $\mathcal{G}E$ is Borel or even analytic, then $E$ is Borel measurable [Kechris, Theorem 14.12] and therefore continuous [Kechris, Theorem 9.10]. Now,

$$\begin{split}{\big{(}x,(a_{n})\big{)}\in\mathcal{G}E&\;\Leftrightarrow\;\sum_{n=1}^{m}a_{n}e_{n}\mathop{\underset{m}{\overset{\sf u}{\longrightarrow}}}x,}\end{split}$$

which is Borel by Lemma 2.1. Thus, the biorthogonal functionals $e^{\sharp}_{n}$ are all continuous.

To see that $(e_{n})$ is Schauder basic, i.e., a Schauder basis for $[e_{n}]$, note that, for every $x\in[e_{n}]$, we have that $x=^{\sf u}\sum_{n=1}^{\infty}e^{\sharp}_{n}(x)e_{n}$ and hence also $x=^{\lVert\cdot\rVert}\sum_{n=1}^{\infty}e^{\sharp}_{n}(x)e_{n}$. Also, if $(a_{n})\in\mathbb{R}^{\mathbb{N}}$ is any sequence so that $x=^{\lVert\cdot\rVert}\sum_{n=1}^{\infty}a_{n}e_{n}$, we see that

$$a_{k}=\lim_{m}e^{\sharp}_{k}\Big{(}\sum_{n=1}^{m}a_{n}e_{n}\Big{)}=e^{\sharp}_{k}\Big{(}\lim_{m}\sum_{n=1}^{m}a_{n}e_{n}\Big{)}=e^{\sharp}_{k}(x),$$

so the norm-expansion $x=^{\lVert\cdot\rVert}\sum_{n=1}^{\infty}e^{\sharp}_{n}(x)e_{n}$ is unique. ∎

In order to obtain a similar result for either $\sigma$-order or order convergence, we are forced to rely on additional set theoretical assumptions, namely, the determinacy of increasingly complicated sets.

Theorem 2.3 (${\bf\Sigma}^{1}_{1}$-determinacy or $\text{MA}+\neg\text{CH}$).

Suppose that $(e_{n})$ is a $\sigma$-order basis for a separable Banach lattice $X$. Then the biorthogonal functionals $e^{\sharp}_{n}$ are continuous and hence $(e_{n})$ is Schauder basic.

Proof.

Observe that, for $x\in X$ and $(a_{n})\in\mathbb{R}^{\mathbb{N}}$, we have

$$\begin{split}{x=^{\sigma\sf o}\sum_{n=1}^{\infty}a_{n}e_{n}&\;\Leftrightarrow\;\exists(z_{m})\in X_{+}^{\mathbb{N}}\quad\Big{(}z_{m}\downarrow 0\quad\&\quad\forall m\;\forall^{\infty}n\;\;\;\Big{|}x-\sum_{k=1}^{n}a_{k}e_{k}\Big{|}\leqslant z_{m}\Big{)}.}\end{split}$$

Clearly, the condition $\forall m\;\forall^{\infty}n\;\Big{|}x-\sum_{k=1}^{n}a_{k}e_{k}\Big{|}\leqslant z_{m}$ is Borel in the tuple $\big{(}x,(a_{n}),(z_{m})\big{)}\in X\times\mathbb{R}^{\mathbb{N}}\times X_{+}^{\mathbb{N}}$, but unfortunately the condition $z_{m}\downarrow 0$ appears only to be ${\bf\Pi}^{1}_{1}$ in $(z_{m})\in X_{+}^{\mathbb{N}}$,

$$z_{m}\downarrow 0\;\;\Leftrightarrow\;\;\forall m\;z_{m}\geqslant z_{m+1}\quad\&\quad\forall y>0\;\exists m\;y\not<z_{m}.$$

Thus, a priori, the graph $\mathcal{G}E$ of $X\mathop{\overset{E}{\longrightarrow}}\mathbb{R}^{\mathbb{N}}$ is only ${\bf\Sigma}^{1}_{2}$, which means that the inverse image $E^{-1}(U)$ of an open set $U\subseteq\mathbb{R}^{\mathbb{N}}$ is ${\bf\Sigma}^{1}_{2}$ and therefore has the property of Baire if we assume either ${\bf\Sigma}^{1}_{1}$-determinacy [Kechris, Theorem 36.20] or $\text{MA}+\neg\text{CH}$ [Kechris, Exercise 38.8] and [Just, Theorem 19.23]. Therefore, $E$ is continuous by [Kechris, Theorem 14.12] and hence so are the associated partial sum projections $X\mathop{\overset{P_{m}}{\longrightarrow}}[e_{1},\ldots,e_{m}]$.

We claim that the sequence of operators $(P_{m})_{m=1}^{\infty}$ is uniformly bounded, which by Grunblum’s criterion [albiac, Proposition 1.1.9] implies that $(e_{n})$ is Schauder basic. To see this, it suffices by the principle of uniform boundedness to show that $(P_{m}x)_{m=1}^{\infty}$ is bounded in norm for each $x\in X$. However, given $x$, observe that, as $P_{m}x\underset{m\to\infty}{\overset{\sigma\sf o}{\longrightarrow}}x$, there is $z\geqslant 0$ for which $|x-P_{m}x|\leqslant z$ for all but finitely many $m$, which shows that the sequence $\big{(}\lVert P_{m}x\rVert\big{)}_{m=1}^{\infty}$ is bounded. ∎

Theorem 2.4 (Projective determinacy).

Suppose that $(e_{n})$ is an order basis for a separable Banach lattice $X$. Then the biorthogonal functionals $e^{\sharp}_{n}$ are continuous and hence $(e_{n})$ is Schauder basic.

Proof.

We claim that, for any sequence $(x_{n})$ and vector $x$ in $X$, we have

$$\begin{split}{x_{n}\mathop{\underset{}{\overset{\sf o}{\longrightarrow}}}x\quad\Leftrightarrow\quad\forall y>0\;\exists z\;\big{(}y\not\leqslant z\;\;\&\;\;\forall^{\infty}n\;|x_{n}-x|\leqslant z\big{)}.}\end{split}$$

As the right-hand side is clearly a ${\bf\Pi}^{1}_{2}$-condition on $\big{(}(x_{n}),x\big{)}\in X^{\mathbb{N}}\times X$, this will imply that the biorthogonal operator $X\mathop{\overset{E}{\longrightarrow}}\mathbb{R}^{\mathbb{N}}$ has ${\bf\Pi}^{1}_{2}$-graph and hence is continuous under the assumption of projective determinacy [Kechris, Theorem 38.17]. As in the proof of Theorem 2.3, this in turn ensures that $(e_{n})$ is Schauder basic.

To establish our claim, suppose first that $x_{n}\mathop{\underset{}{\overset{\sf o}{\longrightarrow}}}x$. This means that there is a decreasing net $(z_{\mu})$ with $\inf_{\mu}z_{\mu}=0$ so that $\forall\mu\;\forall^{\infty}n\;|x_{n}-x|\leqslant z_{\mu}$. In particular, if $y>0$, then $y\not\leqslant z_{\mu}$ for some $\mu$ and hence the sequence $(x_{n})$ satisfies

(1) $$\forall y>0\;\exists z\;\big{(}y\not\leqslant z\;\;\&\;\;\forall^{\infty}n\;|x_{n}-x|\leqslant z\big{)}.$$

Conversely, suppose that $(x_{n})$ satisfies (1). Then the set

$$A=\{z\in X\;\big{|}\;\forall^{\infty}n\;|x_{n}-x|\leqslant z\}$$

becomes directed under the ordering $z\prec z^{\prime}\Leftrightarrow z^{\prime}\leqslant z$. For if $z,z^{\prime}\in A$, then also $z,z^{\prime}\prec z\wedge z^{\prime}\in A$. If follows that $(A,\prec)$ can be viewed as a decreasing net with infimum $0$ witnessing that $x_{n}\mathop{\underset{}{\overset{\sf o}{\longrightarrow}}}x$. ∎

The proofs of Theorem 1.2 and Corollaries 1.3 and 1.5 heavily rely on the relationship between the different types of convergence of partial sums established in [taylor, Theorem 3.1] (see also [gumenchuk, Theorem 2.3] for a related earlier result).

Theorem 2.5.

[taylor, Theorem 3.1] The following statements are equivalent for a Schauder basic sequence $(e_{n})$ in a Banach lattice $X$ with associated basis projections $[e_{n}]\mathop{\overset{P_{m}}{\longrightarrow}}[e_{n}]$.

1. For all $x\in[e_{n}]$, $P_{m}x\xrightarrow{\sf u}x$,

2. For all $x\in[e_{n}]$, $P_{m}x\xrightarrow{\sigma\sf o}x$,

3. For all $x\in[e_{n}]$, $P_{m}x\xrightarrow{\sf o}x$,

4. For all $x\in[e_{n}]$, $(P_{m}x)$ is order bounded in $X$,

5. For all $x\in[e_{n}]$, $(\bigvee_{n=1}^{m}\left|P_{n}x\right|)$ is norm bounded,

6. There is $M\geq 1$ so that, for all $n\in\mathbb{N}$ and scalars $a_{1},\ldots,a_{n}$, one has

   $$\begin{split}{\Bigg{\lVert}\bigvee_{m=1}^{n}\Big{|}\sum_{k=1}^{m}a_{k}e_{k}\Big{|}\,\Bigg{\rVert}\leqslant M\Bigg{\|}\sum_{k=1}^{n}a_{k}e_{k}\Bigg{\|}.}\end{split}$$

Proof of Theorem 1.2.

(1)$\Rightarrow$(2): Assume that $(e_{n})$ is an order basis for $X$ with corresponding functionals $(e^{\sharp}_{n})$. Then, by Theorem 2.4, the biorthogonal functionals $(e^{\sharp}_{n})$ are continuous and $(e_{n})$ is a Schauder basis for $X=[e_{n}]$. Furthermore, for all $x\in X$, we have $P_{m}x\mathop{\underset{}{\overset{\sf o}{\longrightarrow}}}x$, which means that $(e_{n})$ satisfies condition (iii) of [taylor, Theorem 3.1] and hence must also satisfy condition (ii) of the same theorem, namely that, for every $x\in X$, $P_{m}x\mathop{\underset{}{\overset{\sigma\sf o}{\longrightarrow}}}x$, i.e., $x=^{\sigma\sf o}\sum_{n=1}^{\infty}e^{\sharp}_{n}(x)e_{n}$. On the other hand, if $(a_{n})$ is any sequence so that $x=^{\sigma\sf o}\sum_{n=1}^{\infty}a_{n}e_{n}$, then also $x=^{\sf o}\sum_{n=1}^{\infty}a_{n}e_{n}$, whereby $a_{n}=e^{\sharp}_{n}(x)$ as $(e_{n})$ is an order basis. This shows uniqueness of the $\sigma$-order expansion and hence implies that $(e_{n})$ is a $\sigma$-order basis for $X$.

The implication (2)$\Rightarrow$(3) is entirely analogous using conditions (ii) and (i) of [taylor, Theorem 2.1] in place of conditions (iii) and (ii), together with Theorem 2.3. Finally, the implication (3)$\Rightarrow$(4) follows directly from Theorem 2.2. ∎

Proof of Corollary 1.3.

Fix a sequence $(e_{n})$ in a Banach lattice $X$ so that $X=[e_{n}]$. Assume first that condition (3) holds, i.e., that, for some constant $M$ and all finite tuples of scalars $(a_{n})_{n=1}^{m}$ one has

$$\begin{split}{\Bigg{\lVert}\bigvee_{k=1}^{m}\Big{|}\sum_{n=1}^{k}a_{n}e_{n}\Big{|}\,\Bigg{\rVert}\leqslant M\Bigg{\|}\sum_{n=1}^{m}a_{n}e_{n}\Bigg{\|},}\end{split}$$

whereby also

$$\begin{split}{\Bigg{\lVert}\sum_{n=1}^{k}a_{n}e_{n}\Bigg{\rVert}\leqslant M\Bigg{\|}\sum_{n=1}^{m}a_{n}e_{n}\Bigg{\|},}\end{split}$$

for all $k\leqslant m$. Thus, by Grunblum’s criterion [albiac, Proposition 1.1.9], we see that $(e_{n})$ is a Schauder basis for $X$. We let $P_{m}$ denote the corresponding basis projections and $e^{*}_{n}$ the biorthogonal functionals. By the implication (vi)$\Rightarrow$(i) of [taylor, Theorem 3.1], we find that, for all $x\in X$, $P_{m}x\mathop{\underset{}{\overset{\sf u}{\longrightarrow}}}x$, i.e., that $x=^{\sf u}\sum_{n=1}^{\infty}e^{*}_{n}(x)e_{n}$. On the other hand, to see that this expansion is unique, note that, if $x=^{\sf u}\sum_{n=1}^{\infty}a_{n}e_{n}$ for some sequence $(a_{n})$, then also $x=^{\lVert\cdot\rVert}\sum_{n=1}^{\infty}a_{n}e_{n}$, which in turn implies that $a_{n}=e^{*}_{n}(x)$ for all $n$. This shows that $(e_{n})$ is also a uniform basis for $X$ and hence verifies the implication (3)$\Rightarrow$(1).

Now, assume instead that $(e_{n})$ is a uniform basis for $X$. Then, by Theorem 2.2, $(e_{n})$ is also a Schauder basis for $X$. Let again $P_{m}$ denote the corresponding basis projections. Then, for all $x\in X$, $P_{m}x\mathop{\underset{}{\overset{\sf u}{\longrightarrow}}}x$, which implies that the sequence $(P_{m}x)$ is order bounded. Thus (1)$\Rightarrow$(2).

Finally, the implication (2)$\Rightarrow$(3) is a direct consequence of [taylor, Theorem 3.1]. ∎

Proof of Corollary 1.5.

By Theorem 1.2, if $(e_{n})$ is an order basis for $X$, it is also a uniform basis for $X$. Note also, that because $(e_{n})$ is an order basis, $0=^{\sf o}\sum_{n=1}^{\infty}0e_{n}$ must be the unique order expansion of $0$. This shows that (1)$\Rightarrow$(2).

Conversely, if (2) holds, then, by Theorem 1.2, $(e_{n})$ is also a Schauder basis and must satisfy condition (3) of Corollary 1.3. So, if $e^{*}_{n}$ denote the biorthogonal functionals associated to the Schauder basis $(e_{n})$, then by [taylor, Theorem 2.1] we have that $x=^{\sf o}\sum_{n=1}^{\infty}e^{*}_{n}(x)e_{n}$ for all $x\in X$. To see that this order expansion of $x$ is unique, note that, if $x=^{\sf o}\sum_{n=1}^{\infty}a_{n}e_{n}$ for some sequence $(a_{n})$, then $0=^{\sf o}\sum_{n=1}^{\infty}\big{(}e^{*}_{n}(x)-a_{n}\big{)}e_{n}$ and so $a_{n}=e^{*}_{n}(x)$ by the uniqueness of the order expansion for $0$. Thus, $(e_{n})$ is an order basis for $X$. ∎

Remark 2.6.

As a consequence of the above discussion, it follows that a sequence $(e_{n})$ of non-zero vectors in a Banach lattice $X$ is ${\sf u}$-basic if and only if the inequality in Theorem 2.5 (vi) holds. This shows that the assumption that $X=[e_{n}]$ is not required for the equivalence of the statements in Corollary 1.3. Moreover, it yields a significant generalization of Grunblum’s criterion [albiac, Proposition 1.1.9] for Schauder basic sequences. Indeed, if $(e_{n})$ is a sequence of non-zero vectors in a Banach space $E$, then we may always view $E$ as contained in the Banach lattice $X=C(B_{E^{*}})$. In $C(K)$-spaces, it is clear that uniform convergence agrees with norm convergence – hence the notions of ${\sf u}$-basic and Schauder basic coincide – and the supremum in Theorem 2.5 (vi) commutes with the norm. Therefore, we recover the standard Grunblum criterion [albiac, Proposition 1.1.9] in the particular case $X=C(B_{E^{*}})$.

3. Proofs for filter bases

In the following, we shall identify the powerset $\mathcal{P}(\mathbb{N})$ with the Cantor space $\{0,1\}^{\mathbb{N}}$. If $(e_{n})$ is any sequence in a Banach space $X$, we may define a Borel measurable function

$$X\times\mathbb{R}^{\mathbb{N}}\times\mathbb{R}_{+}\mathop{\overset{\theta}{\longrightarrow}}\mathcal{P}(\mathbb{N})$$

by letting, for all $x\in X$, $(a_{n})\in\mathbb{R}^{\mathbb{N}}$ and $\epsilon>0$,

(2) $$\theta\big{(}x,(a_{n}),\epsilon\big{)}=\Big{\{}m\in\mathbb{N}\;\Big{|}\;\,\Big{\lVert}x-\sum_{n=1}^{m}a_{n}e_{n}\Big{\rVert}<\epsilon\Big{\}}.$$

Assume now that $(e_{n},e^{\sharp}_{n})$ is a fixed biorthogonal system in $X$. Then a filter $F$ on $\mathbb{N}$ (always assumed to be proper and containing the Fréchet filter of all cofinite sets) is compatible with $(e_{n},e^{\sharp}_{n})$ if, for all $x\in X$,

$$\sum_{n=1}^{m}e^{\sharp}_{n}(x)e_{n}\mathop{\underset{m\to F}{\overset{}{\longrightarrow}}}x$$

and, for all sequences $(a_{n})\in\mathbb{R}^{\mathbb{N}}$ other than $\vec{0}=(0,0,\ldots)$, we have

$$\sum_{n=1}^{m}a_{n}e_{n}\mathop{\underset{m\to F}{\overset{}{\not}\!\!\longrightarrow}}0.$$

Indeed, these two conditions taken together ensure that $\sum_{n=1}^{\infty}e^{\sharp}_{n}(x)e_{n}$ is the unique $F$-expansion of an element $x\in X$. Rewriting these conditions in terms of $\theta$, we find that the filter $F$ is compatible with $(e_{n},e^{\sharp}_{n})$ if and only if

(3) $$\theta\big{(}x,(e^{\sharp}_{n}(x)),\epsilon\big{)}\in F$$

for all $x\in X$ and $\epsilon>0$ and, moreover, $F$ satisfies the property $\Phi$ defined by

(4) $$\Phi(F)\;\;\Leftrightarrow\;\;\forall(a_{n})\in\mathbb{R}^{\mathbb{N}}\setminus\{{\vec{0}}\}\;\;\exists k\in\mathbb{N}\quad\theta\big{(}0,(a_{n}),\tfrac{1}{k}\big{)}\notin F.$$

To simplify notation, if $(e_{n},e^{\sharp}_{n})$ is a biorthogonal system, we let $X\mathop{\overset{E}{\longrightarrow}}\mathbb{R}^{\mathbb{N}}$ denote the biorthogonal operator $Ex=\big{(}e^{\sharp}_{n}(x)\big{)}$. In particular, $E$ is continuous if and only if all the $e^{\sharp}_{n}$ are continuous.

Lemma 3.1.

Every filter basis system $(e_{n},e^{\sharp}_{n})$ for a Banach space $X$ has a smallest compatible filter.

Proof.

Observe that

(5) $$A=\Big{\{}a\in\mathcal{P}(\mathbb{N})\;\Big{|}\;\bigcap_{i=1}^{m}\theta\big{(}x_{i},Ex_{i},\epsilon\big{)}\subseteq a\text{ for some }x_{i}\in X\text{ and }\epsilon>0\Big{\}}$$

is the smallest filter on $\mathbb{N}$ containing all images $\theta\big{(}x,Ex,\epsilon\big{)}$ for $x\in X$ and $\epsilon>0$. In particular, $A$ is contained in every compatible filter. On the other hand, if $F$ is a compatible filter and $(a_{n})\neq\vec{0}$, then there is some $k$ so that

$$\theta(0,(a_{n}),\tfrac{1}{k})\notin F\supseteq A,$$

which shows that $\Phi(A)$ holds and hence that $A$ is a compatible filter for $(e_{n},e^{\sharp}_{n})$. ∎

Note that, if $E$ is continuous, then the smallest compatible filter $A$ (see Equation (5)) is analytic when viewed as a subset of $\mathcal{P}(\mathbb{N})$. This is [Rancourt, Theorem B]. Observe also that, if $F$ is a compatible analytic filter, then $E$ has analytic graph,

$$\big{(}x,(a_{n})\big{)}\in\mathcal{G}E\;\Leftrightarrow\;\forall k\in\mathbb{N}\;\;\;\theta\big{(}x,(a_{n}),\tfrac{1}{k}\big{)}\in F,$$

and thus is Borel measurable [Kechris, Theorem 14.12] and therefore continuous [Kechris, Theorem 9.10]. This is [Rancourt, Theorem A].

Proof of Theorem 1.17.

Let $(e_{n},e^{\sharp}_{n})$ be a fixed filter basis system for $X=[e_{n}]$. We remark that the implications (4)$\Rightarrow$(1)$\Rightarrow$(3) have been noted above. Also, the implications (5)$\Rightarrow$(4) and (1)$\Rightarrow$(2) are trivial, so it suffices to show (2)$\Rightarrow$(1) and (3)$\Rightarrow$(5).

(2)$\Rightarrow$(1): Assume that $(e_{n})$ is minimal, i.e., that $e_{k}\notin[e_{n}]_{n\neq k}$ for all $k$. This means that there is a set $(e^{*}_{k})$ of continuous functionals $e^{*}_{k}$ on $X=[e_{n}]$ biorthogonal to the vectors $e_{n}$. Now, for each $x\in X$, the series $\sum_{n=1}^{\infty}e^{\sharp}_{n}(x)e_{n}$ converges in norm along a filter to $x$ and so, for some increasing sequence $(m_{i})$, we have that

$$x=\lim_{i\to\infty}\sum_{n=1}^{m_{i}}e^{\sharp}_{n}(x)e_{n},$$

whereby

$$\begin{split}{e^{*}_{k}(x)&=e^{*}_{k}\Big{(}\lim_{i\to\infty}\sum_{n=1}^{m_{i}}e^{\sharp}_{n}(x)e_{n}\Big{)}\\ &=\lim_{i\to\infty}e^{*}_{k}\Big{(}\sum_{n=1}^{m_{i}}e^{\sharp}_{n}(x)e_{n}\Big{)}\\ &=\lim_{i\to\infty}\sum_{n=1}^{m_{i}}e^{\sharp}_{n}(x)e^{*}_{k}(e_{n})\\ &=e^{\sharp}_{k}(x)}\end{split}$$

for all $k$. So, $e^{\sharp}_{k}=e^{*}_{k}$ is continuous for each $k$.

(3)$\Rightarrow$(5): Assume that the minimal compatible filter $A$ is analytic. We define a binary predicate $\Psi$ on subsets of $\mathcal{P}(\mathbb{N})$ by letting for $B,C\subseteq\mathcal{P}(\mathbb{N})$

$$\begin{split}{\Psi(B,C)\;\;\Leftrightarrow\;\;&\forall x\subseteq y\subseteq\mathbb{N}\;(x\in B\to y\notin C)\;\;\&\\ &\forall x,y\subseteq\mathbb{N}\;(x,y\in B\to x\cap y\notin C)\;\;\&\\ &\forall x\subseteq\mathbb{N}\;(x\text{ is cofinite }\to x\notin C)\;\;\&\\ &\emptyset\notin B.}\end{split}$$

Observe that, if $\sim\!F$ denotes the complement of a set $F\subseteq\mathcal{P}(\mathbb{N})$, then $\Psi(F,\sim\!F)$ holds if and only if $F$ is a proper filter on $\mathbb{N}$ containing all cofinite sets.

Consider now the conjunction

$$\Gamma(B,C)\;\;\Leftrightarrow\;\;\Phi(B)\;\;\&\;\;\Psi(B,C),$$

where $\Phi$ is defined as in (4), and observe that $\Gamma$ is a hereditary predicate in both variables, i.e., passes to subsets, and is continuous upwards in the second variable, i.e., if $C_{1}\subseteq C_{2}\subseteq\ldots$ and $\Gamma(B,C_{n})$ hold for all $n$, then also $\Gamma(B,\bigcup_{n}C_{n})$. Furthermore, $\Gamma$ is ${\bf\Pi}^{1}_{1}$ on ${\bf\Sigma}^{1}_{1}$. That is, if $Y$ is a Polish space and $B,C\subseteq Y\times\mathcal{P}(\mathbb{N})$ are ${\bf\Sigma}^{1}_{1}$, then

$$\big{\{}y\in Y\;\big{|}\;\Gamma(B_{y},C_{y})\text{ holds}\big{\}}$$

is ${\bf\Pi}^{1}_{1}$.

By the discussion above, we see that, if $A\subseteq F\subseteq\mathcal{P}(\mathbb{N})$, then $F$ is a compatible filter if and only if $\Gamma(F,\sim\!F)$. In particular, $\Gamma(A,\sim\!A)$ and hence, by the Second Reflection Theorem [Kechris, Theorem 35.16], there is some Borel set $F\subseteq\mathcal{P}(\mathbb{N})$ so that $A\subseteq F$ and $\Gamma(F,\sim\!F)$. Thus, $F$ is a compatible Borel filter. ∎

Remark 3.2.

Observe that, if the biorthogonal operator $X\mathop{\overset{E}{\longrightarrow}}\mathbb{R}^{\mathbb{N}}$ associated with some $F$-filter basis $(e_{n})$ for $X$ is continuous, then the operator range $E[X]$ is the continuous injective image of a separable Banach space and is therefore a Borel linear subspace of $\mathbb{R}^{\mathbb{N}}$ [Kechris, Theorem 15.1]. However, if $F$ is actually Borel, we have explicit bounds on the Borel complexity of $E[X]$ in terms of the Borel complexity of $F$. Indeed,

$$(a_{n})\in E[X]\;\Leftrightarrow\;\forall l\;\exists k\;\Big{\{}m\in\mathbb{N}\;\Big{|}\;\,\Big{\lVert}\sum_{n=1}^{m}a_{n}e_{n}-\sum_{n=1}^{k}a_{n}e_{n}\Big{\rVert}<\tfrac{1}{l}\Big{\}}\in F.$$

To see this, note that the implication from left to right is immediate. For the implication from right to left, note that, if $k_{l}$ are such that

$$\Big{\{}m\in\mathbb{N}\;\Big{|}\;\,\Big{\lVert}\sum_{n=1}^{m}a_{n}e_{n}-\sum_{n=1}^{k_{l}}a_{n}e_{n}\Big{\rVert}<\tfrac{1}{l}\Big{\}}\in F$$

for all $l$, then $\big{(}\sum_{i=1}^{k_{l}}a_{n}e_{n}\big{)}_{l}$ is Cauchy and converges to some $x$ so that $Ex=(a_{n})$.

4. Higher order Fatou properties

We now return to the setting of Banach lattices and investigate conditions under which the assumptions of projective or analytic determinacy may be eliminated from Theorem 1.2. Recall that a $\pi$-basis for a Banach lattice $X$ is a subset $B\subseteq X$ for which $b>0$ for all $b\in B$ and so that, for all $x>0$, there is $b\in B$ with $b<x$. Observe that, for example, $C([0,1])$ and the sequence spaces $c_{0}$ and $\ell_{p}$, $1\leqslant p\leqslant\infty$, all have countable $\pi$-bases, while $L_{1}[0,1]$ fails to have a countable $\pi$-basis.

Lemma 4.1.

Suppose $X$ is a separable Banach lattice with a countable $\pi$-basis $B$. Then, for all sequences $(x_{n})$ and vectors $x$, we have

$$\begin{split}{x_{n}\mathop{\underset{n}{\overset{\sf o}{\longrightarrow}}}x&\quad\Leftrightarrow\quad x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}x\\ &\quad\Leftrightarrow\quad\forall b\in B\;\exists z\;\big{(}\forall^{\infty}n\;|x_{n}-x|\leqslant z\;\&\;b\not<z\big{)}.}\end{split}$$

Thus, order and $\sigma$-order convergence of sequences coincide and define an analytic relation on $(x_{n})$ and $x$.

Proof.

Assume first that $x_{n}\mathop{\underset{n}{\overset{\sf o}{\longrightarrow}}}x$. Then there is a decreasing net $(z_{\mu})$ with infimum $0$ so that every $z_{\mu}$ bounds all but finitely many of the expressions $|x_{n}-x|$. In particular, if $b\in B$, then we have that $b\not<z_{\mu}$ for some $\mu$, which shows that

$$\forall b\in B\;\exists z\;\big{(}\forall^{\infty}n\;|x_{n}-x|\leqslant z\;\&\;b\not<z\big{)}.$$

Assume now, in turn, that $\forall b\in B\;\exists z\;\big{(}\forall^{\infty}n\;|x_{n}-x|\leqslant z\;\&\;b\not<z\big{)}$. Enumerate $B$ as $B=\{b_{1},b_{2},\ldots\}$ and, for each $k$, choose some $z_{k}$ so that $\forall^{\infty}n\;|x_{n}-x|\leqslant z_{k}$, whereas $b_{k}\not<z_{k}$. Let also $y_{m}=\bigwedge_{k=1}^{m}z_{k}$. Then $y_{m}\downarrow 0$ and, for every $m$, we have $|x_{n}-x|\leqslant y_{m}$ for all but finitely many $n$, i.e., $x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}x$. As $\sigma$-order convergence implies order convergence, this finishes the proof. ∎

Corollary 4.2.

Suppose $X$ is a separable Banach lattice with a countable $\pi$-basis. Then Theorem 1.2 holds without the additional assumption of projective determinacy.

We now turn to other weaker conditions than the existence of a countable $\pi$-basis. So, in the following, let $X$ denote a fixed separable Banach lattice. Let us begin by recalling the following simple calculation.

Lemma 4.3.

For all $x,y\in X$, we have

$$(y-x)^{+}=y-(x\wedge y).$$

Proof.

Recall that $+$ distributes over the lattice operation $\vee$, so

$$\begin{split}{y-(x\wedge y)&=y+\big{(}(-x)\vee(-y)\big{)}=(y-x)\vee 0=(y-x)^{+}}\end{split}$$

for all $x,y\in X$. ∎

If $\Sigma$ is any set, we let $\Sigma^{<\mathbb{N}}$ denote the set of all finite strings of elements of $\Sigma$. Recall that a subset $T\subseteq\Sigma^{<\mathbb{N}}$ is a tree provided that $T$ contains the empty string $\emptyset$ and is closed under taking initial segments. Recall also that a tree $T$ is said to be ill-founded provided that it has an infinite branch, i.e., if there is an infinite sequence $(y_{n})_{n=1}^{\infty}$ in $\Sigma$ so that

$$(y_{1},\ldots,y_{n})\in T$$

for all $n$. Otherwise, $T$ is well-founded. For a well-founded tree $T$, we may define an ordinal valued rank function $\rho_{T}\colon T\to{\sf Ord}$ by letting

$$\rho_{T}(s)=0\;\Leftrightarrow\;s\text{ has no proper extensions in }T$$

and otherwise

$$\rho_{T}(s)=\sup\big{\{}\rho_{T}(t)+1\;\big{|}\;t\in T\;\&\;s\subsetneq t\}.$$

We also define the rank of $T$ itself by

$$\rho(T)=\sup\big{\{}\rho_{T}(t)+1\;\big{|}\;t\in T\}=\rho_{T}(\emptyset)+1.$$

A binary relation $\prec$ on a set $\Omega$ is said to be well-founded if there is no infinite sequence of elements $p_{n}\in\Omega$ so that

$$\ldots\prec p_{3}\prec p_{2}\prec p_{1}.$$

In this case, we may similarly define a rank $\rho_{\prec}\colon\Omega\to{\sf Ord}$ by

$$\rho_{\prec}(p)=0\;\Leftrightarrow\;p\text{ is minimal, i.e., }q\not\prec p\text{ for all }q\in\Omega$$

and

$$\rho_{\prec}(p)=\sup\big{\{}\rho_{\prec}(q)+1\;\big{|}\;q\prec p\}.$$

As for trees, we set $\rho(\prec)=\sup\big{\{}\rho_{\prec}(p)+1\;\big{|}\;p\in\Omega\}$. For example, if $T$ is a well-founded tree, we may let $\prec$ be the relation $\supsetneq$ on $T$, that is, for $s,t\in T$, we have $s\prec t$ if $t$ is a proper initial segment of $s$. Then $\rho_{T}=\rho_{\prec}$.

We let $Tr_{\Sigma}$ denote the set of all trees $T\subseteq\Sigma^{<\mathbb{N}}$ and $WF_{\Sigma}$ denote the subset of all well-founded trees. If $\Sigma$ is a countable set, then $Tr_{\Sigma}$ is a closed subset of the Polish space $\{0,1\}^{\Sigma^{<\mathbb{N}}}$, whereas $WF_{\Sigma}$ is a coanalytic subset of $Tr_{\Sigma}$. Moreover, by [Kechris, Exercise 34.6], $\rho$ is a coanalytic rank on $WF_{\Sigma}$. In particular, this means that, for all $\lambda<\omega_{1}$, the set

$$WF_{\Sigma}^{\lambda}=\{T\in Tr_{\Sigma}\;\big{|}\;\rho(T)\leqslant\lambda\}$$

is Borel.

Set

$$X_{\downarrow}=\big{\{}(z_{n})_{n=1}^{\infty}\in X^{\mathbb{N}}\;\big{|}\;z_{1}\geqslant z_{2}\geqslant\ldots\geqslant 0\big{\}}$$

and define, for every sequence $(z_{n})\in X_{\downarrow}$, a tree $\Psi\big{(}(z_{n})\big{)}\subseteq\big{(}X_{+}\!\setminus\{0\}\big{)}^{<\mathbb{N}}$ by

$$\begin{split}{(y_{1},\ldots,y_{k})\in\Psi&\big{(}(z_{n})\big{)}\Leftrightarrow\\ &\forall i\;\forall n\quad\lVert(y_{i}-z_{n})^{+}\rVert<\tfrac{\lVert y_{i}\rVert}{2^{i}}\;\;\&\;\;\lVert y_{i+1}-y_{i}\rVert<\tfrac{\lVert y_{i}\rVert}{2^{i}}.}\end{split}$$

Lemma 4.4.

Let $P\subseteq X_{+}\setminus\{0\}$ be a countable norm-dense subset. Then, for all $(z_{n})_{n=1}^{\infty}\in X_{\downarrow}$, the following conditions are equivalent.

1. $\inf_{n}z_{n}=0$,

2. $\Psi\big{(}(z_{n})\big{)}$ is well-founded,

3. $\Psi\big{(}(z_{n})\big{)}\cap P^{<\mathbb{N}}$ is well-founded.

Proof.

Note first that the implication (2)$\Rightarrow$(3) is trivial.

(3)$\Rightarrow$(1): Assume that $0$ is not the infimum of the sequence $(z_{n})$, that is, that there is some $y\in X_{+}$ satisfying

$$z_{1}\geqslant z_{2}\geqslant\ldots\geqslant y>0.$$

Pick then elements $y_{i}\in B\big{(}y,\tfrac{\lVert y\rVert}{4^{i}}\big{)}\cap P$, which implies that

$$\lVert y_{i+1}-y_{i}\rVert\leqslant\lVert y_{i+1}-y\rVert+\lVert y-y_{i}\rVert<\frac{\lVert y\rVert}{4^{i+1}}+\frac{\lVert y\rVert}{4^{i}}\leqslant\frac{\lVert y_{i}\rVert}{2^{i}}$$

for all $i$. Furthermore,

$$\lVert(y_{i}-z_{n})^{+}\rVert\leqslant\lVert(y-z_{n})^{+}\rVert+\lVert y-y_{i}\rVert<\frac{\lVert y_{i}\rVert}{2^{i}},$$

showing that $(y_{1},y_{2},\ldots)$ is an infinite branch of $\Phi\big{(}(z_{n})\big{)}\cap P^{<\mathbb{N}}$.

(1)$\Rightarrow$(2): Suppose that $\Psi\big{(}(z_{n})\big{)}$ is ill-founded, i.e., that $\Psi\big{(}(z_{n})\big{)}$ has an infinite branch $(y_{1},y_{2},\ldots)$. Then the conditions $\lVert y_{i+1}-y_{i}\rVert<\frac{\lVert y_{i}\rVert}{2^{i}}$ ensure that $(y_{i})$ is a Cauchy sequence converging to some $y>0$. Furthermore, as $\lVert(y_{i}-z_{n})^{+}\rVert<\frac{\lVert y_{i}\rVert}{2^{i}}$ for all $n$ and $i$, we find that also

$$\big{\lVert}y-(z_{n}\wedge y)\big{\rVert}=\big{\lVert}(y-z_{n})^{+}\big{\rVert}=0,$$

i.e., that $y\leqslant z_{n}$ for all $n$. So, $y$ is a strictly positive lower bound for $(z_{n})$. ∎

Lemma 4.5.

The following conditions are equivalent.

1. The set

   $$X_{\downarrow 0}=\big{\{}(z_{n})_{n=1}^{\infty}\in X_{\downarrow}\;\big{|}\;\inf_{n}z_{n}=0\big{\}}$$

   is Borel,

2. $$\sup\Big{\{}\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\;\Big{|}\;{(z_{n})_{n=1}^{\infty}\in X_{\downarrow 0}}\Big{\}}<\omega_{1}.$$

Proof.

(1)$\Rightarrow$(2): Note that, if $X_{\downarrow 0}$ is Borel, so is the set

$$\Omega=\big{\{}\big{(}(z_{n})_{n=1}^{\infty},(y_{1},\ldots,y_{m})\big{)}\in X_{\downarrow 0}\times\big{(}X_{+}\!\setminus\{0\}\big{)}^{<\mathbb{N}}\;\big{|}\;(y_{1},\ldots,y_{m})\in\Psi\big{(}(z_{n})\big{)}\big{\}}.$$

We may then define a Borel quasiordering $\prec$ on $\Omega$ by setting

$$\begin{split}{\big{(}(z_{n})_{n=1}^{\infty},&(y_{1},\ldots,y_{m})\big{)}\prec\big{(}(u_{n})_{n=1}^{\infty},(v_{1},\ldots,v_{k})\big{)}\\ &\Leftrightarrow\;(z_{n})_{n=1}^{\infty}=(u_{n})_{n=1}^{\infty}\quad\&\quad k<m\quad\&\quad(v_{1},\ldots,v_{k})=(y_{1},\ldots,y_{k})}\end{split}$$

and observe that $\prec$ is well-founded by Lemma 4.4. It then follows from the boundedness theorem for analytic well-founded relations [Kechris, Theorem 31.1] that

$$\begin{split}{\sup\Big{\{}\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\;\Big{|}\;{(z_{n})_{n=1}^{\infty}\in X_{\downarrow 0}}\Big{\}}&=\sup\Big{\{}\rho_{\Psi\big{(}(z_{n})\big{)}}(\emptyset)+1\;\Big{|}\;{(z_{n})_{n=1}^{\infty}\in X_{\downarrow 0}}\Big{\}}\\ &=\sup\Big{\{}\rho_{\prec}\big{(}(z_{n}),\emptyset\big{)}+1\;\Big{|}\;{(z_{n})_{n=1}^{\infty}\in X_{\downarrow 0}}\Big{\}}\\ &=\rho(\prec)\\ &<\omega_{1}.}\end{split}$$

(2)$\Rightarrow$(1): Observe that, if $P\subseteq X_{+}\setminus\{0\}$ is a fixed countable norm-dense subset and

$$\sup\Big{\{}\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\;\Big{|}\;{(z_{n})_{n=1}^{\infty}\in X_{\downarrow 0}}\Big{\}}<\omega_{1},$$

then also

$$\lambda=\sup\Big{\{}\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\cap P^{<\mathbb{N}}\Big{)}\;\Big{|}\;{(z_{n})_{n=1}^{\infty}\in X_{\downarrow 0}}\Big{\}}<\omega_{1}.$$

Note now that the map $X_{\downarrow}\mathop{\overset{\Theta}{\longrightarrow}}Tr_{P}$ defined by

$$\Theta\big{(}(z_{n})\big{)}=\Psi\big{(}(z_{n})\big{)}\cap P^{<\mathbb{N}}$$

is Borel measurable and satisfies

$$(z_{n})\in X_{\downarrow 0}\quad\Leftrightarrow\quad\Theta\big{(}(z_{n})\big{)}\in WF^{\lambda}_{P}.$$

Because $WF^{\lambda}_{P}$ is Borel, this shows that also $X_{\downarrow 0}$ is Borel. ∎

For every ordinal $\alpha$ and $(z_{n})_{n=1}^{\infty}\in X_{\downarrow}$, we define a game $G_{\alpha}\big{[}(z_{n})\big{]}$ between two players I and II as follows. Players I and II alternate in playing ordinals $\beta_{i}$ and vectors $y_{i}\in X_{+}\!\setminus\{0\}$,

$${\begin{array}[]{ccccccccccccccc}{\bf I}&&{\beta_{1}}&&{\beta_{2}}&&&\ldots&&{\beta_{k-1}}&&{\beta_{k}}&&\\ {\bf II}&&&{y_{1}}&&{y_{2}}&&\ldots&&&{y_{k-1}}&&{y_{k}}&\end{array}}$$

and where the positions played are subject to the conditions

$$\alpha>\beta_{1}>\beta_{2}>\ldots>\beta_{k-1}>\beta_{k}\geqslant 0$$

and

$$\lVert y_{i+1}-y_{i}\rVert<\frac{\lVert y_{i}\rVert}{2^{i}}.$$

The game ends when I plays $\beta_{k}=0$ and II plays its response $y_{k}$. This will eventually happen as the ordinals are well-ordered. Player II is then said to win a run of the game provided that

$$\lVert(y_{i}-z_{n})^{+}\rVert<\frac{\lVert y_{i}\rVert}{2^{i}}$$

for all $n\geqslant 1$ and $1\leqslant i\leqslant k$. Otherwise player I wins.

Example 4.6 (Banach lattices with the Fatou property).

Suppose $X$ is a Banach lattice with the Fatou property, that is, whenever we have elements $0\leqslant x_{1}\leqslant x_{2}\leqslant\ldots\leqslant x$ with $x=\sup_{n}x_{n}$, then $\lVert x\rVert=\sup_{n}\lVert x_{n}\rVert$. Assume also that $(z_{n})\in X_{\downarrow 0}$. Then it is easy to see that I has a winning strategy in the game $G_{1}\big{[}(z_{n})\big{]}$. Indeed, I simply plays $\beta_{1}=0$, to which II responds with some vector $y_{1}$. If II wins this run of the game, we must have

$$\lVert y_{1}-(z_{n}\wedge y_{1})\rVert=\lVert(y_{1}-z_{n})^{+}\rVert<\frac{\lVert y_{1}\rVert}{2}$$

for all $n$ and

$$0\leqslant y_{1}-(z_{1}\wedge y_{1})\leqslant y_{1}-(z_{2}\wedge y_{1})\leqslant y_{1}-(z_{3}\wedge y_{1})\leqslant\ldots\leqslant y_{1}=\sup_{n}\big{(}y_{1}-(z_{n}\wedge y_{1})\big{)},$$

which contradicts the Fatou property. This means that a winning strategy for I in $G_{1}\big{[}(z_{n})\big{]}$ is simply to play $\beta_{1}=0$.

Lemma 4.7.

For every $(z_{n})_{n=1}^{\infty}\in X_{\downarrow}$ and every ordinal $\alpha$,

$$\begin{split}{\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\leqslant{\alpha}\quad\Leftrightarrow\quad\text{ I has a winning strategy in the game }G_{\alpha}\big{[}(z_{n})\big{]}.}\end{split}$$

Proof.

A straightforward inspection shows that

$$\begin{split}{\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}>{\alpha}\quad\Leftrightarrow\quad\text{ II has a winning strategy in the game }G_{\alpha}\big{[}(z_{n})\big{]}.}\end{split}$$

Note also that, because the rules and winning conditions in both games are Borel, the games are determined, that is, either player I or II has a winning strategy [Kechris, Theorem 20.6]. Therefore,

$$\begin{split}{\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\leqslant{\alpha}&\quad\Leftrightarrow\quad\text{ II has no winning strategy in the game }G_{\alpha}\big{[}(z_{n})\big{]}\\ &\quad\Leftrightarrow\quad\text{ I has a winning strategy in the game }G_{\alpha}\big{[}(z_{n})\big{]}}\end{split}$$

as claimed. ∎

Definition 4.8.

The separable Banach lattice $X$ is said to be $\alpha$-Fatou provided that, for all $(z_{n})\in X_{\downarrow}$,

$$\begin{split}{\inf_{n}z_{n}=0&\quad\Leftrightarrow\quad\text{ I has a winning strategy in the game }G_{\alpha}\big{[}(z_{n})\big{]}\\ &\quad\Leftrightarrow\quad\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\leqslant{\alpha}.\ }\end{split}$$

By Example 4.6, Banach lattices with the Fatou property are $1$-Fatou.

Proof of Theorem 1.18.

(1)$\Rightarrow$(2): Suppose that (1) holds, i.e., that $X_{\downarrow 0}$ is Borel. Then, for all sequences $(x_{n})$ and vectors $x$, we have

$$\begin{split}{x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}x&\quad\Leftrightarrow\quad\exists(z_{n})\in X_{\downarrow 0}\;\;\forall m\;\;\forall^{\infty}n\;\;x_{n}\leqslant z_{m},}\end{split}$$

which is clearly ${\bf\Sigma}^{1}_{1}$.

(2)$\Rightarrow$(1): Observe that, for a sequence $x_{1}\geqslant x_{2}\geqslant\ldots\geqslant 0$, we have

$$\forall y>0\;\exists n\;\;y\not\leqslant x_{n}\quad\Leftrightarrow\quad\inf_{n}x_{n}=0\quad\Leftrightarrow\quad x_{n}\mathop{\underset{n}{\overset{\sigma\sf o}{\longrightarrow}}}0.$$

The first expression is clearly ${\bf\Pi}^{1}_{1}$, so, if the last is ${\bf\Sigma}^{1}_{1}$, then these equivalent expressions are all Borel and hence $X_{\downarrow 0}$ is a Borel set.

(1)$\Leftrightarrow$(3): Just observe that, by Lemma 4.5, the set $X_{\downarrow 0}$ is Borel if and only if there is some $\alpha<\omega_{1}$ so that

$$\rho\Big{(}\Psi\big{(}(z_{n})\big{)}\Big{)}\leqslant{\alpha}$$

for all $(z_{n})\in X_{\downarrow 0}$. ∎

Let us also note [gumenchuk, Theorem 2.6]. The authors there operate with a slightly stronger Fatou property, namely, the conjunction of $\sigma$-monotonic completeness and what we have termed the Fatou property here. Thus, translated in to our terminology, [gumenchuk, Theorem 2.6] states that a sequence $(e_{n})$ in a $\sigma$-monotonically complete, $\sigma$-order continuous Banach lattice $X=[e_{n}]$ with the Fatou property is simultaneously a Schauder and a $\sigma$-order basis for $X$ if and only if it satisfies condition (3) of Corollary 1.3.

We note that, in fact, a stronger result holds. Indeed, if $(e_{n})$ is a sequence in a $\sigma$-order continuous Banach lattice $X=[e_{n}]$, then $\sigma$-order bases are just uniform bases and the latter are automatically Schauder bases. So, $(e_{n})$ is simultaneously a Schauder and a $\sigma$-order basis for $X$ if and only if it satisfies condition (3) of Corollary 1.3.

5. Examples of spaces with higher order Fatou properties

Our next task is to show that the hierarchy of $\alpha$-Fatou properties does not collapse. That is, we will construct spaces that are $\alpha$-Fatou, but only for larger and larger $\alpha<\omega_{1}$.

Theorem 5.1.

For every countable ordinal $\alpha$, there is a separable Banach lattice with a countable $\pi$-basis that fails to be $\alpha$-Fatou.

Proof.

Our proof goes by induction on $1\leqslant\alpha<\omega_{1}$. For each $\alpha$, we will construct a separable Banach lattice $(X_{\alpha},\lVert\cdot\rVert_{\alpha})$ with a countable $\pi$-basis $B_{\alpha}\subseteq X_{\alpha}$ so that $X_{\alpha}$ fails the $\alpha$-Fatou property, together with a Banach lattice homomorphism

$$X_{\alpha}\mathop{\overset{\phi_{\alpha}}{\longrightarrow}}\mathbb{R}$$

of norm $1$. Furthermore, setting

$$S_{\alpha}=\big{\{}x\in X^{+}_{\alpha}\;\big{|}\;\lVert x\rVert_{\alpha}=\phi_{\alpha}(x)=1\big{\}},$$

we will construct a sequence in $S_{\alpha}$,

$$z^{\alpha}_{1}\geqslant z^{\alpha}_{2}\geqslant\ldots>0=\inf_{n}z^{\alpha}_{n}$$

and a tree $T_{\alpha}\subseteq\Psi\big{(}(z^{\alpha}_{n})\big{)}\cap S_{\alpha}^{<\mathbb{N}}$ so that

$$\rho(T_{\alpha})>\alpha,$$

which implies that

$$\rho\Big{(}\Psi\big{(}(z^{\alpha}_{n})\big{)}\Big{)}>{\alpha}$$

and thus that $X_{\alpha}$ fails the $\alpha$-Fatou property.

Base case, $\alpha=1$. We let $X_{1}$ be the space $c$ of convergent sequences endowed with the equivalent renorming

$$\big{\lVert}(t_{1},t_{2},\ldots)\big{\rVert}_{1}=\max\Big{\{}\tfrac{1}{3}\big{\lVert}(t_{1},t_{2},\ldots)\big{\rVert}_{\ell_{\infty}},\big{|}\lim_{n}t_{n}\big{|}\Big{\}}$$

and define $\phi_{1}$ by

$$\phi_{1}\big{(}(t_{1},t_{2},\ldots)\big{)}=\lim_{n}t_{n}.$$

Also, for each $n$, set

$$z^{1}_{n}=(\underbrace{0,0,\ldots,0}_{n\text{ times}},1,1,1,\ldots)\in S_{1}.$$

Finally, note that because $y=(1,1,1,\ldots)\in\Psi\big{(}(z^{\alpha}_{n})\big{)}$ and $\lVert y\rVert_{1}=\phi_{1}(y)=1$, we may simply let

$$T_{1}=\{\emptyset,y\},$$

whereby $\rho(T_{1})=2>1$.

Successor case. Suppose that $(X_{\alpha},\lVert\cdot\rVert_{\alpha})$, $B_{\alpha}$, $\phi_{\alpha}$, $(z_{n}^{\alpha})$ and $T_{\alpha}$ have been defined as above. We then let $X_{\alpha+1}=X_{\alpha}$ with the new norm

$$\lVert x\rVert_{\alpha+1}=\max\Big{\{}\tfrac{1}{5}{\lVert x\rVert_{\alpha}},\big{|}\phi_{\alpha}(x)\big{|}\Big{\}}$$

and set $\phi_{\alpha+1}=\phi_{\alpha}$, $B_{\alpha+1}=B_{\alpha}$ and $z^{\alpha+1}_{n}=z^{\alpha}_{n}$ for all $n$. Note that $\lVert\cdot\rVert_{\alpha+1}\leqslant\lVert\cdot\rVert_{\alpha}$, $\big{\lVert}\phi_{\alpha+1}\big{\rVert}=1$ and that

$$\lVert x\rVert_{\alpha+1}=\lVert x\rVert_{\alpha}\;\Leftrightarrow\;\lVert x\rVert_{\alpha}=|\phi_{\alpha}(x)|,$$

from which it follows that $S_{\alpha}\subseteq S_{\alpha+1}$. Finally, let

$$T_{\alpha+1}=\big{\{}(z_{1}^{\alpha},y_{1},\ldots,y_{n})\;\big{|}\;(y_{1},\ldots,y_{n})\in T_{\alpha}\big{\}}\cup\{\emptyset\}\;\;\subseteq\;\;S_{\alpha}^{<\mathbb{N}}\;\;\subseteq\;\;S_{\alpha+1}^{<\mathbb{N}}$$

and note that $\rho(T_{\alpha+1})>\rho(T_{\alpha})>\alpha$. To see that $T_{\alpha+1}\subseteq\Psi\big{(}(z^{\alpha+1}_{n})\big{)}$, note first that, for all $x,y\in S_{\alpha+1}$, we have

$$\phi_{\alpha}\big{(}(x-y)^{+}\big{)}=\big{(}\phi_{\alpha}(x-y)\big{)}^{+}=0^{+}=0=\phi_{\alpha}(x-y),$$

whereby $\big{\lVert}(x-y)^{+}\big{\rVert}_{\alpha+1}=\tfrac{1}{5}\big{\lVert}(x-y)^{+}\big{\rVert}_{\alpha}$ and $\lVert x-y\rVert_{\alpha+1}=\tfrac{1}{5}\lVert x-y\rVert_{\alpha}$. Thus, for all $n$,

$$\begin{split}{\big{\lVert}(z_{1}^{\alpha}-z^{\alpha+1}_{n})^{+}\big{\rVert}_{\alpha+1}&=\tfrac{1}{5}\big{\lVert}(z_{1}^{\alpha}-z^{\alpha+1}_{n})^{+}\big{\rVert}_{\alpha}\\ &=\tfrac{1}{5}\big{\lVert}(z_{1}^{\alpha}-z^{\alpha}_{n})^{+}\big{\rVert}_{\alpha}\\ &\leqslant\tfrac{1}{5}\lVert z^{\alpha}_{1}\rVert_{\alpha}\\ &=\tfrac{1}{5}\lVert z^{\alpha}_{1}\rVert_{\alpha+1}\\ &<\frac{\lVert z_{1}^{\alpha}\rVert_{\alpha+1}}{2^{1}}}\end{split}$$

and

$$\begin{split}{\big{\lVert}(y_{i}-z^{\alpha+1}_{n})^{+}\big{\rVert}_{\alpha+1}&=\tfrac{1}{5}\big{\lVert}(y_{i}-z^{\alpha+1}_{n})^{+}\big{\rVert}_{\alpha}\\ &=\tfrac{1}{5}\big{\lVert}(y_{i}-z^{\alpha}_{n})^{+}\big{\rVert}_{\alpha}\\ &<\tfrac{1}{5}\frac{\|{y}_{i}\|_{\alpha}}{2^{i}}\\ &<\frac{\lVert y_{i}\rVert_{\alpha+1}}{2^{i+1}}.}\end{split}$$

Similarly,

$$\begin{split}{\lVert y_{1}-z_{1}^{\alpha}\rVert_{\alpha+1}&=\tfrac{1}{5}\big{\lVert}y_{1}-z^{\alpha}_{1}\big{\rVert}_{\alpha}\leqslant\frac{2}{5}<\frac{\lVert z_{1}^{\alpha}\rVert_{\alpha+1}}{2^{1}}}\end{split}$$

and

$$\begin{split}{\lVert{y}_{i+1}-{y}_{i}\rVert_{\alpha+1}=\tfrac{1}{5}\big{\lVert}y_{i+1}-y_{i}\big{\rVert}_{\alpha}<\tfrac{1}{5}\frac{\lVert y_{i}\rVert_{\alpha}}{2^{i}}<\frac{\lVert y_{i}\rVert_{\alpha+1}}{2^{i+1}}.}\end{split}$$

Limit case. Suppose that $\alpha_{1}<\alpha_{2}<\ldots<\alpha=\lim_{n}\alpha_{n}$. Assume also that the construction has been done for all ordinals smaller than $\alpha$. We then let

$$X_{\alpha}=\Big{\{}x=(x_{1},x_{2},\ldots)\in\prod_{n}X_{\alpha_{n}}\;\Big{|}\;\lim_{n}\phi_{\alpha_{n}}(2^{n}x_{n})\text{ exists }\;\&\;\lim_{n}\|x_{n}\|_{\alpha_{n}}=0\Big{\}}$$

equipped with the norm

$$\|x\|_{\alpha}=\max\Big{\{}\sup_{n}\lVert x_{n}\rVert_{\alpha_{n}},\big{|}\lim_{n}\phi_{\alpha_{n}}(2^{n}x_{n})\big{|}\Big{\}}$$

and define the homomorphism $\phi_{\alpha}$ by

$$\phi_{\alpha}(x)=\lim_{n}\phi_{\alpha_{n}}(2^{n}x_{n}).$$

Observe that the set $B_{\alpha}$ of elements of the form $(0,\ldots,0,b,0,0,\ldots)$ with $b\in B_{\alpha_{n}}$ in the $n$th position forms a countable $\pi$-basis for $X_{\alpha}$. Furthermore, for every $k$, we define

$$z^{\alpha}_{k}=\left(\frac{z^{\alpha_{1}}_{k}}{2^{1}},\frac{z^{\alpha_{2}}_{k}}{2^{2}},\frac{z^{\alpha_{3}}_{k}}{2^{3}},\ldots\right)\in S_{\alpha}$$

and remark that $z^{\alpha}_{1}\geqslant z^{\alpha}_{2}\geqslant\ldots>0=\inf_{k}z^{\alpha}_{k}$.

Now, suppose that $k\in\mathbb{N}$ is given and that, for every $m\geqslant k$, we have some vector $y_{m}\in S_{\alpha_{m}}$. Then also

$$\big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y_{k}}{2^{k}},\tfrac{y_{k+1}}{2^{k+1}},\tfrac{y_{k+2}}{2^{k+2}},\ldots\big{)}\;\in\;S_{\alpha}.$$

We therefore let $T_{\alpha}$ be the tree consisting of all finite strings of the form

$$\Big{(}\big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{1}_{k}}{2^{k}},\tfrac{y^{1}_{k+1}}{2^{k+1}},\tfrac{y^{1}_{k+2}}{2^{k+2}},\ldots\big{)}\,,\,\ldots\,,\,\big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{n}_{k}}{2^{k}},\tfrac{y^{n}_{k+1}}{2^{k+1}},\tfrac{y^{n}_{k+2}}{2^{k+2}},\ldots\big{)}\Big{)},$$

where $k,n\in\mathbb{N}$ and

$$(y^{1}_{m},y^{2}_{m},\ldots,y^{n}_{m})\in T_{\alpha_{m}}$$

for all $m\geqslant k$. To see that $T_{\alpha}\subseteq\Psi\big{(}(z^{\alpha}_{m})\big{)}$, observe first that, for all $1\leqslant i<n$,

$$\begin{split}{\Bigg{\lVert}\Big{(}\underbrace{0,0,\ldots,0}_{k-1}&,\tfrac{y^{i+1}_{k}}{2^{k}},\tfrac{y^{i+1}_{k+1}}{2^{k+1}},\tfrac{y^{i+1}_{k+2}}{2^{k+2}},\ldots\Big{)}-\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{i}_{k}}{2^{k}},\tfrac{y^{i}_{k+1}}{2^{k+1}},\tfrac{y^{i}_{k+2}}{2^{k+2}},\ldots\Big{)}\Bigg{\rVert}_{\alpha}\\ &=\Bigg{\lVert}\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{i+1}_{k}-y^{i}_{k}}{2^{k}},\tfrac{y^{i+1}_{k+1}-y^{i}_{k+1}}{2^{k+1}},\tfrac{y^{i+1}_{k+2}-y^{i}_{k+2}}{2^{k+2}},\ldots\Big{)}\Bigg{\rVert}_{\alpha}\\ &=\sup_{m\geqslant k}\Big{\lVert}\frac{y^{i+1}_{m}-y^{i}_{m}}{2^{m}}\Big{\rVert}_{\alpha_{m}}\\ &<\sup_{m\geqslant k}\tfrac{1}{2^{i}}\Big{\lVert}\frac{y^{i}_{m}}{2^{m}}\Big{\rVert}_{\alpha_{m}}\\ &\leqslant\frac{1}{2^{i}}\Bigg{\lVert}\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{i}_{k}}{2^{k}},\tfrac{y^{i}_{k+1}}{2^{k+1}},\tfrac{y^{i}_{k+2}}{2^{k+2}},\ldots\Big{)}\Bigg{\rVert}_{\alpha}.}\end{split}$$

Similarly, for all $k$ and $m$,

$$\begin{split}{\Bigg{\lVert}\Bigg{(}&\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{i}_{k}}{2^{k}},\tfrac{y^{i}_{k+1}}{2^{k+1}},\tfrac{y^{i}_{k+2}}{2^{k+2}},\ldots\Big{)}-z^{\alpha}_{m}\Bigg{)}^{+}\Bigg{\rVert}_{\alpha}\\ &=\Bigg{\lVert}\Bigg{(}\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{i}_{k}}{2^{k}},\tfrac{y^{i}_{k+1}}{2^{k+1}},\tfrac{y^{i}_{k+2}}{2^{k+2}},\ldots\Big{)}-\Big{(}\tfrac{z^{\alpha_{1}}_{m}}{2^{1}},\tfrac{z^{\alpha_{2}}_{m}}{2^{2}},\tfrac{z^{\alpha_{3}}_{m}}{2^{3}},\ldots\Big{)}\Bigg{)}^{+}\Bigg{\rVert}_{\alpha}\\ &=\Bigg{\lVert}\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{\big{(}{y^{i}_{k}}-{z^{\alpha_{k}}_{m}}\big{)}^{+}}{2^{k}},\tfrac{\big{(}{y^{i}_{k+1}}-{z^{\alpha_{k+1}}_{m}}\big{)}^{+}}{2^{k+1}},\tfrac{\big{(}{y^{i}_{k+2}}-{z^{\alpha_{k+2}}_{m}}\big{)}^{+}}{2^{k+2}},\ldots\Big{)}\Bigg{\rVert}_{\alpha}\\ &=\sup_{r\geqslant k}\Big{\lVert}\tfrac{\big{(}y^{i}_{r}-z^{\alpha_{r}}_{m}\big{)}^{+}}{2^{r}}\Big{\rVert}_{\alpha_{r}}\\ &<\sup_{r\geqslant k}\tfrac{1}{2^{i}}\Big{\lVert}\frac{y^{i}_{r}}{2^{r}}\Big{\rVert}_{\alpha_{r}}\\ &\leqslant\frac{1}{2^{i}}\Bigg{\lVert}\Big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{i}_{k}}{2^{k}},\tfrac{y^{i}_{k+1}}{2^{k+1}},\tfrac{y^{i}_{k+2}}{2^{k+2}},\ldots\Big{)}\Bigg{\rVert}_{\alpha}.}\end{split}$$

Finally, we verify that $\rho(T_{\alpha})>\alpha$, i.e., that $\rho_{T_{\alpha}}(\emptyset)\geqslant\alpha$. But, if $\beta<\alpha$, find some $k$ so that $\beta<\alpha_{k}$ and hence $\rho_{T_{\alpha_{m}}}(\emptyset)>\beta$ for all $m\geqslant k$. As $T_{\alpha}$ contains all strings

$$\Big{(}\big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{1}_{k}}{2^{k}},\tfrac{y^{1}_{k+1}}{2^{k+1}},\tfrac{y^{1}_{k+2}}{2^{k+2}},\ldots\big{)}\,,\,\ldots\,,\,\big{(}\underbrace{0,0,\ldots,0}_{k-1},\tfrac{y^{n}_{k}}{2^{k}},\tfrac{y^{n}_{k+1}}{2^{k+1}},\tfrac{y^{n}_{k+2}}{2^{k+2}},\ldots\big{)}\Big{)},$$

where $(y^{1}_{m},y^{2}_{m},\ldots,y^{n}_{m})\in T_{\alpha_{m}}$, this shows that also $\rho_{T_{\alpha}}(\emptyset)>\beta$. As $\beta<\alpha$ was arbitrary, we have that $\rho_{T_{\alpha}}(\emptyset)\geqslant\alpha$. ∎