1. Introduction
Let be an ideal on the nonnegative integers , that is,
a subset of closed under taking subsets and finite unions. Unless otherwise stated, it is assumed that is admissible, namely, and that contains the family of finite subsets of .
Intuitively, the ideal represents the family of “small” subsets of . An important example is the family of asymptotic density zero sets
|
|
|
Define .
Ideals are regarded as subsets of the Cantor space , hence we can speak about their topological complexity. For instance, is a -ideal, and is a -ideal which is not .
Pick also a sequence be taking values in a topological space .
Then, we denote by the set of -limit points of , that is, the set of all for which there exists a subsequence such that
|
|
|
It is well known that, even in the case where is a real bounded sequence, it is possible that is the empty set, see [13, Example 4].
The topological nature of the sets of -limits points and their relationship with the slightly weaker variant of -cluster points have been studied in [2], cf. also [16, 17].
In this work, we continue along this line of research. To this aim, we introduce our main definition:
Definition 1.1.
Let be a topological space and be an ideal on . We denote by the family of sets of -limit points of sequences taking values in together with the emptyset, that is,
|
|
|
If the topological space is understood, we write simply .
A remark is in order about the the addition of in the above definition: it has been proved by Meza-Alcántara in [28, Section 2.7] that, if , then there exists a -valued sequence such that if and only if there exists a function such that for every set with at most finitely many limit points; cf. also [3, Proposition 6.4] and, more generally, [10] for analogues in compact uncountable spaces. On the other hand, if is not compact, it is easy to see that, for every ideal , there is a sequence with no -limit points.
Thus, the addition of in the above definition avoids the repetition of known results in the literature and
to add further subcases based on the topological structure of the underlying space .
We summarize in Theorem 1.2 and Theorem 1.3 below the known results from [2, 16, 17] about the families .
For, given a topological space and an ordinal , we use the standard Borel pointclasses notations and , so that stands for the open sets of , for the closed sets, for the -sets, etc.;
we denote by the ambiguous classes;
also, if is a Polish space, stands for the analytic sets, for the coanalytic sets, etc., see e.g.
[25, Section 11.B] or [31, Section 3.6]. Again, we suppress the reference to the underlying space if it is clear from the context.
Recall that an ideal is a P-ideal if it is -directed modulo finite sets, that is, for every sequence with values in there exists such that is finite for all . Important examples include the P-ideals, which are known to be necessarily .
A topological space is discrete if it contains only isolated points.
Theorem 1.2.
Let be a nondiscrete first countable Hausdorff space. Then:
-
(i)
if and only if is a ideal, provided that is a P-ideal.
-
(ii)
, provided that is a P-ideal.
If, in addition, all closed subsets of are separable, then:
-
(iii)
, provided that is a ideal.
-
(iv)
, provided that is P-ideal which is not .
Proof.
It follows by [2, Theorems 2.2, 2.5, 2.7, and 3.4].
∎
To state the next result, we recall some further definitions.
An ideal has the hereditary Baire property if the restriction has the Baire property for every . Note that all analytic ideals have the hereditary Baire property: indeed, the proof goes verbatim as in
[24, Theorem 3.13], considering that analytic sets are closed under continuous preimages, and that they have the Baire property, see e.g. [31, Proposition 4.1.2 and Theorem 4.3.2]. In addition, there exist ideals with the Baire property but without the hereditary Baire property, see e.g. [32, Proposition 2.1].
Also, an ideal on is said to be a -ideal if, for every decreasing sequence with values in , there exists such that is finite for all . It is known that all -ideals are -ideals, see
[23] and
[22, Observation 2.2].
We remark that -ideal may have arbitrarily high Borel complexity, as it has been proved in [21, p. 2031] and [22, Example 2.6],
cf. also [14].
Moreover, an ideal is called a Farah ideal if there exists a sequence of hereditary compact subsets in such that if and only if for all there exists such that , see [8, 19, 20]. It is known that all analytic P-ideals are Farah and that all Farah ideals are . On the other hand, it is still unknown whether
the converse holds, namely, all ideals are Farah ideals,
see [7] and [19, p. 60].
Theorem 1.3.
Let be a first countable Hausdorff space. Then:
-
(i)
, provided that is -ideal.
-
(ii)
if and only if is a -ideal, provided that has the hereditary Baire property and is nondiscrete metrizable.
-
(iii)
, provided that is a Farah ideal.
-
(iv)
if and only if is maximal, provided that
If, in addition, is second countable, then:
-
(v)
, provided that has the hereditary Baire property.
-
(vi)
, provided that has the hereditary Baire property and is not a -ideal.
Proof.
It follows by [16, Proposition 1.4, Corollary 3.9, and Theorem 4.5] and [17, Corollary 2.5 and Theorems 2.4, 2.8, and 2.10]. (We added the only if part of item (iv), which is straighforward.)
In what follows, we divide our main results into four sections. First, we show in Section 2 that can be equal to families of arbitrarily high Borel complexity.
Also, we prove
that we cannot have the equality : more precisely, if and only if .
Then, we show that, if is a first countable space and is a simply coanalytic ideal (see Section 3 for details), then every is simply analytic.
Lastly, we study structural and topological properties of “smallest” ideals such that contains a given subsets of .
2. Large and small Borel complexities
Our first main result
computes explicitly some families , proving that they may attain arbitrarily large Borel complexity (Theorem 2.5 below). This is somehow related to [16, Question 3.11], which asks about the existence of a Borel ideal such that contains sets with large Borel complexities.
For,
given (possibly nonadmissible) ideals and on two countably infinite sets and , respectively, we define their Fubini product by
|
|
|
which is an ideal on the countably infinite set , see e.g. [7, Chapter 1]. Hence,
recursively,
for all integers is an ideal on .
Theorem 2.1.
Let be (possibly nonadmissible) ideals on such that is an admissible ideal on .
Fix also a bijection and let be a sequence with values in a first countable space .
For each , define the sequence by
for all .
Then
|
|
|
(1) |
where stands for the family .
Proof.
Let and be the left and right hand side of (1), respectively.
Inclusion . The inclusion is clear if . Otherwise fix a point .
Hence there exists a set such that and the subsequence is convergent to .
By the definition of Fubini product ,
|
|
|
At this point, for each , the subsequence is convergent to and . Since , we obtain that .
Inclusion .
The inclusion is clear if . Otherwise fix a point and let be a decreasing local base of neighborhoods at . Hence there exists such that is a -limit point of for each , let us say for some . Upon removing finitely many elements, we can suppose without loss of generality that
|
|
|
Now, set and note that .
It follows that the subsequence is convergent to : indeed there are only finitely many elements of the subsequence outside each . Therefore , which concludes the proof.
∎
Remark 2.2.
The above result holds for every topological space if . Indeed, in the second part of the proof it is enough to let be a singleton.
It is worth noting that if is an ideal on a countably infinite set , is a bijection, and is a -indexed sequence with values in a topological space , then is an ideal on and
where is the sequence defined by for all . Hence
we may use interchangeably or .
In particular, Theorem 2.1 allows to compute explicitly families of the type .
For, we state two consequence of Theorem 2.1:
Corollary 2.3.
Let be a topological space and be an ideal on . Then
|
|
|
Proof.
It follows by Theorem 2.1 and Remark 2.2 that
|
|
|
completing the proof.
∎
Corollary 2.4.
Let be a first countable space and be an ideal on . Then
|
|
|
Proof.
It follows by Theorem 2.1 that
|
|
|
completing the proof.
∎
At this point, recall that is a -ideal, and that is an analytic -ideal which is not , see [7, Example 1.2.3] and [25, Exercise 23.1]. Thus, it follows by Theorem 1.2 that
|
|
|
(2) |
With the above premises, we are able to extend
(2)
to certain ideals with large Borel complexity families .
Theorem 2.5.
Let be a complete metrizable space. Then, for each positive integer , we have:
-
(i)
is a -ideal and .
-
(ii)
is a -ideal and .
Proof.
(i)
The complexity of is obtained applying recursively [28, Proposition 1.6.16], while the computation of is obtained putting together the base case (2), Corollary 2.4, and [25, Exercise 23.5(i)].
The proof of (ii) goes similarly, replacing Corollary 2.4 with Corollary 2.3.
∎
In particular, if is a complete metrizable space, . This provides a generalization of [2, Example 4.2], where it is proved constructively that there exists a real sequence such that is equal to (note that the latter is not a -set, hence is nonempty).
Our second main result deals with ideals with small topological complexity. Suppose that is a first countable space, and recall that
|
|
|
(3) |
provided is a -ideal, see [2, Theorem 2.3]; note that the Hausdorffness hypothesis is not needed here. In the next result, we are going to show that the same conclusion holds if is a -ideal.
Theorem 2.6.
Let be a first countable space, and suppose that is a -ideal. Then inclusion (3) holds.
It is worth noting that Theorem 2.6 is not a consequence of the former result as it really includes new cases: indeed, as remarked also by Solecki in [30, p. 345], there exists a -ideal on (hence, both and ) which is neither nor , see [34]; cf. also [4, 5] and [28, Proposition 1.2.1].
In addition, Theorem 2.6 allows us to prove a generalization of the folklore result that every -ideal is a -ideal, see [23]. For a different proof of the second part, see also [9, Proposition 10.1].
Corollary 2.7.
Let be a -ideal on . Then is a -ideal, and there exists a -ideal such that .
Proof.
The first part follows putting together Theorem 2.6 and Theorem 1.3.(ii) (note that is Borel, hence with the hereditary Baire property). The second part follows by the known fact that a Borel ideal on is contained in a -ideal if and only if it is contained in a -ideal, see [23].
∎
At this point, we divide the proof of Theorem 2.6 into two intermediate steps.
To this aim, we recall that properties of ideals can be often expressed by finding critical ideals with respect to some preorder, cf. e.g. the survey [19].
To this aim, let and be two ideals on two countably infinite sets and , respectively. Then we say that is below in the Rudin–Blass ordering, shortened as
|
|
|
if there is a finite-to-one map such that if and only if for all subsets .
The restriction of these orderings to maximal ideals , and the Borel cardinality of the quotients have been extensively studied, see e.g. [6, 18, 26] and references therein.
Lemma 2.8.
Let be a first countable Hausdorff space and be an ideal on with the hereditary Baire property. Suppose that inclusion (3) does not hold. Then .
Proof.
Since inclusion (3) fails and is first countable, there exists a sequence such that is not closed, hence not sequentially closed. Therefore there exists a sequence taking values in which is convergent to some limit . Since is Hausdorff, we may suppose without loss of generality that for all distinct . Now, for each , there exists such that . Define and, recursively, for all . Hence is a partition of into -positive sets, for each , the restriction is an ideal on with the Baire property.
It follows by Talagrand’s characterization of meager ideals that for each . More explicitly, for each there exists a finite-to-one map such that
|
|
|
(4) |
see [33, Theorem 2.1]; cf. also [1] for further characterizations of meager ideals based of -limit points of sequences.
Define the map
by
|
|
|
We claim that is a witnessing function for .
For, suppose that belongs to . Then
|
|
|
(5) |
Since each is finite-to-one, has finite intersection with each . Then either is finite or the subsequence is convergent to , while is not an -limit point of . Hence, in both cases, . Conversely, suppose that does not belong to , so that there exists such that . It follows by (4) and (5) that contains
|
|
|
Therefore if and only if for each .
∎
Corollary 2.9.
Under the same hypotheses of Lemma 2.8, .
Proof.
Thanks to Lemma 2.8 and [16, Proposition 3.8], we have . The claim follows by
Equation (2).
∎
For the next intermediate result, given topological spaces and subsets and , we say that is Wadge reducible to , shortened as
|
|
|
if there exists a continuous map such that (or, equivalently, if and only if for all ), see e.g. [25, Definition 21.13]. If, in addition, and are Polish spaces with zero-dimensional, then is said to be -hard if for some . Lastly, if is a set which is also -hard, then it is called -complete. (Analogous definitions can be given for other classes of sets in Polish spaces, see [25, Definition 22.9].)
Lemma 2.10.
Let be an ideal on such that . Then is -hard.
Proof.
First, recall that is a -complete subset of , see e.g. [25, Exercise 23.1]. Hence, to complete the proof, it is sufficient to show that
|
|
|
By hypothesis, there exists a finite-to-one function such that if and only if .
Now, define the map
by , so that if and only if .
Hence, we only need to show that is continuous. For, fix and define .
It follows that, for all with ,
|
|
|
which proves the continuity of .
∎
Proof of Theorem 2.6.
Suppose that is a first countable space and is an ideal on such that inclusion (3) fails.
We claim that is not a -ideal.
If is not a Borel ideal, then the claim is trivial. Hence, let us suppose hereafter that is Borel. In particular, has the hereditary Baire property.
At this point, it follows by Lemma 2.8 and Lemma 2.10 that is a -hard. To sum up, is a Borel subset of a zero-dimensional Polish space and it is -hard. We conclude by [25, Therem 22.10] that the is not a -ideal.
∎
3. Simply analyticity
In [2, Proposition 4.1], the first and last-named authors proved the following:
Proposition 3.1.
Let be a sequence taking values in a Hausdorff regular first countable space . Let also be a coanalytic ideal on . Then is analytic.
However, the classical definition of analytic sets as projections of Borel subsets of is usually considered in Polish spaces , cf. e.g. [31, Chapter 4] or [25, Chapter 14].
In addition, the above result has been re-proved in [17, Theorem 4.1] for zero-dimensional Polish spaces , so that
|
|
|
whenever is a -ideal (notice that the above notation is meaningful).
In this Section, our aim is to reformulate and clarify the statement and the proof of Proposition 3.1, extending in turn both the latter and the special case treated in [17, Theorem 4.1].
Note that there are several papers in which the notion of analytic set is adapted to more general topological spaces, see e.g. [15, 29]. The theory of the latter extension, called -analytic sets, is important and nontrivial, and it does not proceed verbatim as in the classical one of analytic sets.
For our purposes, we generalize in a straight way one of possible definitions of analytic sets to the case of arbitrary topological spaces. Hereafter, stands for the usual projection on .
Definition 3.2.
Let be a topological space. A subset is said to be simply analytic, shortened as s-analytic, if there exists an uncountable Polish space and a Borel subset such that .
Note that Definition 3.2 is coherent with the classical notion of analytic set.
Indeed, it is well known that a subset of a Polish space is analytic if and only if there exists an uncountable Polish space and a Borel such that , cf. [31, Proposition 4.1.1]. At this point, let be two uncountable Polish spaces. Then there exists a Borel isomorphism , as it follows by [31, Theorem 3.3.13]. Let be a s-analytic subset for a Borel set . Then is a Borel subset of by Lemma 3.4 below and, clearly, . Therefore, in Definition 3.2 one may assume without loss of generality that .
Lastly, a subset is said to be s-coanalytic if is s-analytic.
Observe that every Borel set in is both s-analytic and s-coanalytic.
These observations enable us to consider Proposition 3.1 valid as it was stated in the original version in [2]. They allow, in addition, to remove in its statement the hypothesis of regularity (and also the property of being Hausdorff; however, the latter one has been used in [2] only because a first countable space is Hausdorff if and only if every sequence has at most one limit):
Theorem 3.3.
Let be a sequence taking values in a
first countable space . Let also be a s-coanalytic ideal on . Then is s-analytic.
For the proof of Theorem 3.3, we will need some intermediate lemmas.
Lemma 3.4.
Let and be Borel functions, where are topological spaces.
Then the map defined by
|
|
|
is Borel.
Proof.
Let and be arbitrary open sets. It is enough to show that is a Borel subset of . The latter set is equal to , which belongs to . The claim follows by the inclusion , see e.g. [11, Proposition 251E]. (For the converse inclusion of the latter, which does not hold for every and , see [12].)
∎
Lemma 3.5.
Let be a Borel function, where is a topological space and is an uncountable Polish space. If is an analytic set, then is a s-analytic subset of .
Proof.
Since is analytic, there exists a Borel subset such that is the projection on the first coordinate of .
At this point, define
.
Then is Borel set by Lemma 3.4 and . Therefore is a s-analytic set, which concludes the proof.
∎
Lemma 3.6.
Let be a s-analytic set, where is a topological space and is an uncountable Polish space. Then is a s-analytic set.
Proof.
Let be an uncountable Polish space and be a Borel set such that .
Now, observe that and obviously is an uncountable Polish space.
∎
Lemma 3.7.
Let be a sequence of s-analytic subsets of a topological space .
Then both and are s-analytic.
Proof.
The proof proceeds verbatim as in [31, Proposition 4.1.2].
∎
We are ready for the proof of Theorem 3.3.
Proof of Theorem 3.3.
Note that
is s-analytic and let be the set of strictly increasing sequences of nonnegative integers.
Since can be regarded as a closed subset of , it is a Polish space by Alexandrov’s theorem, see e.g. [31, Theorem 2.2.1].
For each ,
fix a decreasing local base of open neighbourhoods of .
Then is an -limit point of if and only if there exists a sequence such that and
for all .
At this point, define the continuous function
|
|
|
where stands for the characteristic function of a set .
In addition, for each , define the function by
|
|
|
Identifying each with its charateristic function ,
it follows that
|
|
|
(6) |
Claim 1.
is a s-analytic subset of .
Proof.
Thanks to Lemma 3.5 and the fact that is continuous, is an (ordinary) analytic subset of . Hence there exists an uncountable Polish space and a Borel set such that . The claim follows from the fact that
|
|
|
see e.g. [11, Proposition 251E], and that .
∎
Claim 2.
For each and , the section is continuous.
Proof.
Fix and .
It is enough to show that the the section is sequentially continuous.
For, pick a sequence of elements of which is convergent to some .
Then, for each , there exists such that for all .
Hence for all . This proves that .
∎
Claim 3.
For each and , the section is Borel measurable.
Proof.
Fix and and define for convenience by for all .
It is enough to show that preimage is Borel in for every basic open set .
This is clear if .
Otherwise there exist and such that
.
Set and . It follows that
|
|
|
Since each is open, we conclude that is a Borel set.
∎
Claim 4.
For each , the set is Borel.
Proof.
Thanks to Claim 2, Claim 3, and [31, Theorem 3.1.30], each map is Borel measurable. The conclusion follows since is a -set.
∎
To conclude the proof of Proposition 3.3, we have by Claim 1 and Claim 4 that both and are s-analytic subsets of , hence also their intersection by Lemma 3.7.
By the identity (6), is the projection on of a s-analytic subset of , which is s-analytic by Lemma 3.6.
∎
4. Minimal ideals and their complexities
In this last Section, following the line of research initiated in [16, 17], we recall the definition of certain ideals and we study their structural and topological complexity. For, given a sequence taking values in a first countable Hausdorff space and a subset , define the ideal
|
|
|
where ,
see [16, Section 2]. More explicitly, belongs to if and only if is finite or, in the opposite, is infinite and there are no infinite subsets such that is convergent to some element of . Note that may be not admissible: for instance, if then .
The main reason of its introduction is the following:
Theorem 4.1.
Let be a sequence taking values in a first countable Hausdorff space and fix a subset . Then
|
|
|
(7) |
In particular,
if is separable, then
.
Proof.
The first part follows by [16, Theorem 2.2(i)]. This implies, if , then , so that . Hence, the second part is obtained by choosing a sequence with dense image.
∎
Note that that Theorem 4.1 proves that, even if , the family may contain sets which are not Borel, which answered an open question in [2].
It is also worth to remark that ideals are not the only ones for which (7) holds, see [16, Theorem 2.7]; cf. also [17, Theorems 3.4 and 3.5] for further refinements of Theorem 4.1 in regular and Polish spaces. However, we show that they are the smallest ideals with such property. For, we say that contains an isomorphic copy of if there exists a bijection on such that , where stands for the family .
Proposition 4.2.
Let be a sequence taking values in a
compact metric space
and fix a dense subset .
Let also be an ideal on such that . Suppose also that has dense image, and that
|
|
|
(8) |
Then contains an isomorphic copy of .
Proof.
By hypothesis, there exists a sequence taking values in such that .
Let be a compatible metric on and note that the denseness of implies that .
Since is compact, there exists a strictly increasing sequence of integers such that and, for each , the family of open balls is an open cover of .
At this point, define recursively the map as it follows:
-
(i)
is the smallest integer such that ;
-
(ii)
for each integer , is the smallest integer such that and ;
-
(iii)
for each with and , is the smallest integer such that and .
It follows by construction that is a bijection on which satisfies
|
|
|
(9) |
Fix and note that the set of -limit points of is contained both in and in . Since the latter is equal to , we obtain by (8) and (9) that
|
|
|
By the definition of , we get . Therefore .
∎
As remarked in the Introduction, the technical condition (8) has been already studied in the literature. For instance, if is compact, all -ideals satisfy (8), taking into account [2, Theorem 2.3] and [27, Lemma 3.1(vi)].
The following result, due to He et al. [16], deals with the topological complexity of ideals . Hereafter, we assume for simplicity that is the Cantor space and is an enumeration of the rationals (i.e., finitely supported sequences) of , with .
Theorem 4.3.
Let be a subset of the Cantor space . Then:
-
(i)
if is closed then is a -ideal;
-
(ii)
if is then is a -ideal;
-
(iii)
if is open then is an analytic -ideal.
Proof.
It follows by [16, Theorem 2.2(ii)]
by choosing a sequence with dense image (the case holds as well).
∎
Taking into account also the results obtained in Section 2, we prove some characterizations for the [non]closedness of the subset :
Theorem 4.4.
Let be a nonempty subset of the Cantor space such that has the hereditary Baire property. Then the following are equivalent:
-
(i)
is not closed;
-
(ii)
;
-
(iii)
is -hard;
-
(iv)
.
Proof.
(i) (ii).
Thanks to Theorem 4.1, , hence the inclusion fails. The claim follows by Lemma 2.8.
(iii) (i).
If is closed, then would be a -ideal by Theorem 4.3.
(iv) (i).
As in the previous implication, if is closed, then would be . Hence by Theorem 1.2. However, since is separable, there exists a countable dense subset, which is and not closed.
∎
On the same lines, we are able to characterize the openness of the set and the -property of the ideal , improving Theorem 4.3(iii):
Theorem 4.5.
Let be a subset of the Cantor space . Then the following are equivalent:
-
(i)
-
(ii)
is an analytic -ideal;
-
(iii)
is a -ideal.
Proof.
(iii) (i). Let us suppose that is not open, i.e., is not closed. Since is metrizable, is not sequentially closed. Hence it is possible to pick a sequence taking values in which is convergent to some . Let us denote by a compatible metric on , so that .
Since is dense, for each there exists an infinite set such that . Considering that , it follows that , hence .
At this point, pick a set such that for all . Set for convenience and define recursively a sequence of integers such that is the smallest element of for which and (note that this is well defined). It follows by the triangle inequality that
|
|
|
Therefore is an infinite subset of such that . Since , this proves that . Hence, by the definition of , we conclude that . To sum up, is a witnessing sequence of sets in which fails the -property for the ideal .
∎
As a consequence, we obtain that there are only three possibilities for the complexity of ideals :
Corollary 4.6.
Let be a subset of the Cantor space . Then exactly one of the following cases occurs:
-
(i)
is a -ideal;
-
(ii)
is a Borel ideal which is not ;
-
(iii)
is not a Borel ideal.
Proof.
First, if , then , which satisfies only (i). Hence suppose hereafter that is nonempty. Suppose also that is Borel, so that (iii) fails and has the hereditary Baire property. If is closed then is by Theorem 4.3, hence (i) holds. If is not closed then is -hard by Theorem 4.4, which is equivalent to be not by [25, Theorem 22.10], hence (ii) holds.
∎
In light of Theorem 4.3, one may ask whether the third case in Corollary 4.6 really occurs. We answer in the affirmative with our last main result:
Theorem 4.7.
Let be a Borel subset of the Cantor space which is not . Then is
not analytic, hence not Borel.
Putting together the above results, we have the following consequence:
Corollary 4.8.
Let be a Borel subset of the Cantor space . Then the following are equivalent:
-
(i)
is ;
-
(ii)
is a -ideal;
-
(iii)
is a Borel ideal;
-
(iv)
is an analytic ideal.
We recall now some definition about trees, which will be needed in the proof of Theorem 4.7.
Denote by the set of finite -sequences. We say that is an extension of , shortened as , if and for all . Given an infinite sequence , we define for each . A tree on is a subset with the property that, for all such that . The body of a tree on , denoted by , is the set of all its infinite branches, that is, the set of all sequences such that for all .
A tree on is said to be pruned if every has a proper extension, that is, for all there exists such that and , see [25, Definition 2.1]. Accordingly, we define
|
|
|
Identifying a pruned tree with its characteristic function, the set can be regarded as a closed subset of the Polish space , see [25, Exercise 4.32]. Thanks to Alexandrov’s theorem [25, Theorem 3.11], is a Polish space.
With the above premises, let be the set of pruned trees on for which every infinite branch contains finitely many ones, that is,
|
|
|
see [25, Section 33.A]. It is known that is a -complete subset of , hence it is conanalytic but not analytic, see [25, Exercise 27.3].
More generally, given a nonempty subset of the Cantor space , let be the set of pruned trees on which contains an infinite branch in , that is,
|
|
|
It is immediate to check that coincides with , where is the set of irrationals .
With the above premises, we are ready for the proof of Theorem 4.7.
Proof of Theorem 4.7.
We divide the proof in some intermediate steps. Hereafter, as in the statement of the result, is a given Borel subset of the Cantor space which is not .
Claim 5.
Fix nonempty subsets such that . Then
|
|
|
(10) |
Proof.
Since is Wadge reducible to , there exists a continuous map such that , i.e., if and only if . Let be the family of compact subsets of , endowed with the Vietoris topology. It follows by [25, Exercise 4.32] that the map defined by
|
|
|
is a well-defined homeomorphism. In addition, its inverse map is given by for all . Thanks to [25, Exercise 4.29(vi)], the map defined by is continuous.
At this point, define the function by
|
|
|
so that for each pruned tree .
Now, it is sufficient to show that is a witnessing map for the claimed Wadge reduction (10). For, note that is a composition of three continuous maps, hence it is continuous. Moreover, the following chain of equivalences holds for each pruned tree on :
|
|
|
Therefore , which completes the proof.
∎
Claim 6.
.
Proof.
Thanks to [25, Theorem 22.10], the Borel set is -hard, that is, for all . In particular, . Hence by Claim 5, we conclude that .
∎
Claim 7.
.
Proof.
We need to show that there exists a continuous map such that
|
|
|
for all pruned trees on .
First, define the map as follows: for each , let be the unique nonnegative integer such that
|
|
|
Of course, is injective. Recalling that is the unique sequence with empty support, it is easy to see the image of is the set of positive integers.
At this point, define the map by
for all . Since is an infinite and is injective, then is infinite as well.
It is also not difficult to show that is continuous: indeed, a basic clopen set containing is of the type
|
|
|
for some finite sets such that every restriction of each does not belong to . Then, the set is an open set such that .
Next, we claim that
|
|
|
(11) |
On the one hand, fix . Then there exists a strictly increasing sequence of positive integers such that . For each , set and define so that . Hence
|
|
|
Fix . Then there exists such that for all . Since is injective, there exists such that . It follows that
|
|
|
Since is a tree and , we obtain . By the arbitrariness of , it follows that is an infinite branch of . Therefore .
On the other hand, fix , so that for all . It follows that and for all . It is clear that . Therefore the opposite inclusion holds.
Recall that if and only if . Thanks to (11), this is equivalent to , that is, .
∎
To conclude the proof, thanks to Claim 6, Claim 7, and the transitivity of , we get . Hence there exists a continuous map such that
|
|
|
The conclusion follows by the facts that is -complete (hence, not analytic) and that analytic sets are closed under continuous preimages, see e.g. [31, Proposition 4.1.2].
∎
It is possible to show that is coanalytic, hence -complete. We leave as open question for the interested reader to check whether is always coanalytic whenever is Borel and not .