148 (1995)
Classification of function spaces with the pointwise topology determined
by a countable dense set
by
Tadeusz D o b r o w o l s k i (Norman, Okla.) and Witold M a r c i s z e w s k i (Warszawa)
Abstract. We are concerned with C
A(X), the space of continuous real valued func- tions on X considered with the topology of pointwise convergence on A, where A is a countable dense subset of X. We focus on the Borel and the topological classifications of the spaces C
A(X). For example, we prove that for countable nondiscrete X, C
A(X) is homeomorphic to σ
ω, the countable product of σ = {(x
i) ∈ R
ω| x
i= 0 a.e.}, provided C
A(X) ∈ F
σδ.
1. Introduction. All spaces under consideration are completely regu- lar. For a space X, C
p(X) denotes the space of all continuous real valued functions on X with the pointwise convergence topology. If A is a dense subset of X, then by C
A(X) we denote the space of continuous real valued functions on X with the topology of pointwise convergence on A. Hence, we have C
A(X) = {f |A | f is continuous on X} ⊆ R
Aand C
X(X) = C
p(X).
Throughout this paper we will assume that A is countable; consequently, C
A(X) is a dense linear subspace of R
A, a countable product of lines.
Recently, a lot of work has been done on the topological classification of Borel and projective function spaces C
p(X) for countable spaces X (for references see [DMM] and [CDM]). It has been shown that C
p(X), while Borel, is always of an exact multiplicative class [CDM] and it is conjectured that the topological and the Borel classifications coincide [DMM]. For the countable spaces X the spaces C
A(X) seem to be a natural generalization
1991 Mathematics Subject Classification: Primary 46E10, 57N20, 54C35.
Key words and phrases: function space, topology determined by a countable set, bor- elian filters.
Research of the second author partially supported by KBN grant 2 1113 91 01.
The results of this paper were presented by the second author in January 1993 at the 21st Winter School on Abstract Analysis, Podˇebrady, Czech Republic.
[35]
of the spaces C
p(X). This paper is to initiate an investigation on the Borel complexity and the topological classification of the spaces C
A(X). The Borel structure of the spaces C
A(X) seems to be much more complicated than in the case of C
p(X). However, we are able to indicate some similarities between the topological classifications of the spaces C
A(X) and C
p(X).
The spaces C
A(X) also allow us to use some methods of descriptive set theory and infinite-dimensional topology (applicable for separable metriz- able spaces) for the investigation of the function spaces on uncountable separable spaces. In this context the spaces C
A(X) have appeared in the literature in several natural situations. Before we describe some of them we have to recall a few notions.
A map f : M → N between separable metrizable spaces is of the first Baire class if f
−1(U ) is an F
σ-subset of M for every open U ⊂ N (if N is additionally a linear space then this means that f is the pointwise limit of a sequence of continuous maps M → N ). Let P be the space of irrationals.
It turns out that compact spaces that can be embedded in B
1(P ), the space of real valued first Baire class functions on P with the topology of pointwise convergence, are of great importance in topology and Banach space theory (see [BFT] and [Ne, Section 1]); they are called Rosenthal compacta. The following result of Godefroy [Go, Theorem 4] shows how C
A(X) spaces are involved in dealing with Rosenthal compacta.
1.1. A separable compact space K is a Rosenthal compactum if and only if for every countable dense subset A of K the space C
A(K) is analytic (i.e., a continuous image of P ).
The problem of Borel classification of the spaces C
A(K) for separable Rosenthal compact spaces K has been disscussed in [Ma1].
Another important fact involving C
A(X) spaces is the following factor- ization result [Ma1, Lemma 3.4]:
1.2. Let X and Y be separable spaces and ϕ : C
p(X) → C
p(Y ) be a homeomorphism. For any countable sets C ⊆ X and D ⊆ Y there exist countable dense sets A ⊆ X and B ⊆ Y such that C ⊆ A, D ⊆ B and the map π
Bϕπ
A−1: C
A(X) → C
B(Y ) is a homeomorphism (π
A: C
p(X) → C
A(X) and π
B: C
p(Y ) → C
B(X) are the standard projections).
This result shows that the problem of the topological classification of the function spaces C
p(X) for separable spaces X is related to the problem of the classification of the spaces C
A(X).
Let us observe that the topology of C
A(X) is precisely the weak topology
on C
p(X) induced by the family of evaluation functionals at a, a ∈ A; this
family is countable, consists of continuous linear functionals, and separates
points of C
p(X). Any continuous linear functional on C
p(X) is a linear com-
bination of finitely many evaluation functionals (at x ∈ X). The following example shows that the weak topology on C
p(X) induced by a countable family of continuous linear functionals which separate points of C
p(X) is not necessarily that of C
A(X) for some countable dense A ⊆ X.
1.3. Example. Let {q
n}
∞n=1be an enumeration of the rationals in the line R. The sequence {ϕ
n}
∞n=1of continuous linear functionals on C
p(R) given by ϕ
n(f ) = f (q
n) + f (q
n+ π/n), f ∈ C
p(R), n = 1, 2, . . . , separates points of C
p(R). However, every evaluation functional on C
p(R) is discon- tinuous in the weak topology induced by {ϕ
n}
∞n=1on C
p(R).
Recall that if M is a separable metrizable space and α is a countable ordi- nal then A
α(M ) (resp., M
α(M )) denotes the family of subsets of M that are Borel of additive (resp., multiplicative) class α. By A
α(resp., M
α) we de- note the class of spaces that are absolute Borel of additive (resp., multiplica- tive) class α. If A ∈ A
α\M
α(resp., M
α\A
αor A
α∩M
α\ S
β<α
(A
β∪M
β)), then we say that A is of the exact additive (resp., multiplicative or ambi- guous) class α. By P
n, n ≥ 0, we denote the nth projective class. A map f : M → N between separable metrizable spaces is Borel of class α if f
−1(U ) ∈ A
α(M ) for every open U ⊂ N (this means that f is Borel of class α = 1 precisely when f is of the first Baire class). The above terminology is that of [Kur].
In Section 2 we address some questions concerning the Borel (projective) structure of spaces C
A(X). For instance, fixing X, we wonder how the Borel (projective) class of C
A(X) will change when varying A. The following in- teresting question arises. What is the relationship between the exact Borel (projective) classes of C
p(X), C
A(X) and C
p(A) for countable X? For each countable ordinal α, we provide an example of a countable space X
αand a dense subset A
αso that C
p(X
α) ∈ M
2and C
Aα(X
α) 6∈ M
α(see 2.6).
In Section 3 we deal with spaces X with exactly one nonisolated point.
Such spaces can be identified with N
F= N ∪ {∞} topologized by isolating the points of N = {1, 2, . . .} and using the family {A ∪ {∞} | A ∈ F } as a neighborhood base at ∞, where F is a filter on N. In our considerations, if F is a filter on a countable set T , then we usually identify T and N and most often we let T = ω = {0, 1, . . .}. We will always assume that F contains the Fr´echet filter F
0consisting of all cofinite sets in ω. We will write
c
F= {f ∈ R
ω| ∀(ε > 0) (f
−1((−ε, ε)) ∈ F )}
and
C
F= {f ∈ R
ω| ∃(x ∈ R)∀(ε > 0) (f
−1((x − ε, x + ε)) ∈ F )}.
It is known [Ma2, Lemma 2.1] that c
Fis (linearly) homeomorphic to C
p(N
F).
The space C
Fcan be identified with C
ω(N
F). Moreover, we have
C
F= {f ∈ R
ω| ∀(ε > 0)∃(A ∈ F )∀(a
1, a
2∈ A) (|f (a
1) − f (a
2)| < ε)}.
In these definitions the space ω can be replaced by any countable infinite set T . A filter F on T will be treated as a subset of 2
T, a copy of the Cantor set. We address the following question. What is the relation between the Borel (projective) exact class of F and that of C
F? Our answer is contained in 3.4 and is less satisfactory compared to the case of c
F(see [DMM]).
Section 5 is devoted to the topological identification of F
σδ-spaces C
A(X).
The main result states that for countable nondiscrete X and dense A ⊆ X, the space C
A(X) is homeomorphic to σ
ωprovided C
A(X) ∈ F
σδ; here σ = {(x
n) ∈ R
ω| x
n= 0 a.e.}. (This fact for A = X is the main result of [DMM].) The same result holds for (not necessarily countable) Fr´echet spaces X. Let us recall that X is a Fr´echet space if given a subset Y ⊂ X and a point x ∈ Y there exists a sequence of elements of Y that converges to x. The main tool to obtain these identification results is the method of absorbing sets (belonging to infinite-dimensional topology, see [BM]).
An application of this technique to function spaces was initiated by van Mill in [vM]. We do not explicitly refer to this method; instead, we use some factorization facts and rely on results of [CDM]. However, we were not able to avoid another notion of infinite-dimensional topology: so-called Z
σ-spaces. We devote Section 4 to Z
σ-spaces C
A(X). Many of our results concern also spaces C
A∗(X), subsets of C
A(X) consisting of all bounded functions.
In Section 6 we settle the case of metrizable X. It turns out that for a metrizable X, C
A(X) is analytic only if X is σ-compact (and then, it is homeomorphic to σ
ω).
In the last section we provide examples of spaces C
Fwith arbitrarily high Borel complexity. This is achieved by employing filters F previously used in [CDM]. Here, as in [CDM], not only do we show that for every α ≥ 2 there exists a filter F such that C
F∈ M
α\ A
a, but actually it follows that C
Fis an absorbing set for the class of M
α. Hence, C
Fis homeomorphic to the standard, M
α-absorbing model Ω
α(see [BM]).
2. C
A(X) versus C
B(X). How much can they differ? For a space X and countable dense subsets A and B of X, we will be interested in how much the Borel (projective) classes of C
A(X) and C
B(X) can differ. The following estimate was established in [Ma1, Theorem 2.2].
2.1. Proposition. Let X be a Fr´echet space and let A, B ⊆ X be count- able dense sets. For every countable ordinal α and every n ∈ ω, we have:
(a) if C
A(X) ∈ M
α, then C
B(X) ∈ M
1+α,
(b) if C
A(X) ∈ A
α, then C
B(X) ∈ A
1+α,
(c) if C
A(X) ∈ P
n, then C
B(X) ∈ P
n.
In particular , for infinite ordinals α and n ∈ ω, the exact Borel (projective) classes of C
A(X) and C
B(X) coincide.
The following example shows that for non-Fr´echet spaces X the gap between Borel classes can be as big as we wish (for a space Z, βZ denotes the ˇ Cech–Stone compactification of Z).
2.2. Example. We have
(a) C
ω(βω) ∈ A
1\M
1(actually, C
ω(βω) can be identified with the space Σ = {(x
n) ∈ R
ω| (x
n) is bounded}),
(b) if p is an ultrafilter on ω (i.e., p ∈ βω \ω), then C
ω∪{p}(βω) 6∈ P
1∪P
2(actually, C
ω∪{p}(βω) contains a closed copy of the ultrafilter p).
The statement of 2.2(b) can be reversed in the following way.
2.3. Proposition. Let X be a compact space. Assume there are count- able dense sets A, B ⊆ X such that C
A(X) is analytic and C
B(X) is non- analytic. Then X contains a copy of βω.
P r o o f. Write S = C
A(X) and consider i : X → R
Sdefined by i(x)(f ) = arctan(f (x)), f ∈ S, x ∈ X.
It follows that i is an embedding and i(K) is norm-bounded (here, we con- sider the sup norm on the space of bounded functions on S). Moreover, for a ∈ A, i(a) is continuous on S. Using the Godefroy characterization of Rosenthal compacta and our assumption, we get i(K) * B
1(S). Now, our result is a consequence of the following statement from [P, p. 34] (see also [BFT]): “For an arbitrary norm-bounded sequence {f
j| j ∈ ω} of continu- ous functions on P (or, more generally, on an analytic space X) one and only one possibility occurs: either all pointwise accumulation points of {f
j} are of first Baire class, or there exists a subsequence {f
j| j ∈ ω} which behaves on some Cantor set T in P like the sequence of projections p
j: ω
ω→ {0, 1}
(in particular, in the second case, all accumulation points of {f
j| j ∈ ω} are non-Borel and the pointwise closure of the set {f
j| j ∈ ω} is homeomorphic to the ˇ Cech–Stone compactification of the natural numbers”.
To get our assertion, enumerate A = {a
j| j ∈ ω} and apply the above statement to the sequence {i(a
j) | j ∈ ω}.
The following result is a direct consequence of 2.3 and 1.1.
2.4. Corollary. Let X be a compact space and let A be a countable dense subset of X. If C
A(X) is analytic then either X is a Rosenthal com- pactum or else X contains a copy of βω.
It has been shown [CDM, Theorem 5.1] that for every countable space
X, the space C
p(X), when Borel, is of an exact multiplicative class. In
connection with this and 2.2(a) we ask:
2.5. Question. Is there an example of a (countable) space X such that C
A(X) ∈ A
α\ M
α, α > 1, for some countable dense set A ⊆ X?
Next we show that the gap between Borel classes of C
p(X) = C
X(X) and C
A(X) can be large even for countable X.
2.6. Proposition. For every countable ordinal α, there exists a count- able space X
αand a dense set A ⊆ X
αsuch that C
p(X
α) ∈ M
2and C
A(X
α) 6∈ M
α.
P r o o f. We will use induction on α. Set X
0= A = ω + 1. It is obvious that C
p(X
0) ∈ M
2; by [DGM, Corollary 1.2], C
p(X
0) is homeomorphic to σ
ωand hence it belongs to M
2\ A
2. Let us distinguish x
0= ω ∈ X
0and B
0= ω; note that B
0is the set of all isolated points of X
0and A = B
0∪{x
0}.
Suppose the spaces X
βhave been constructed (for all β < α), sets B
βof all isolated points of X
βand points x
β∈ X
β\ B
βhave been determined so that A
β= B
β∪ {x
β} is a countable dense subset of X
βso that C
Aβ(X
β) 6∈
M
αand C
p(X
β) ∈ M
2. For every n ∈ ω, let α
n= β provided α = β +1. If α is a limit ordinal, fix a sequence of ordinals {α
n}
n∈ωwith α
n< α, n ∈ ω, and sup
nα
n= α. Form a direct sum Y
α= ( S
n∈ω
{n} × X
αn) ⊕ (ω + 1). Consider an equivalence relation R on Y
αwhich identifies (n, x
αn) with n ∈ ω + 1.
We set X
α= Y
α/R to be the quotient space. Write q : Y
α→ X
αfor the quotient map. We let
B
α= {x ∈ X
α| x is isolated in X
α}
and x
α= q(ω), where ω ∈ ω + 1. By the construction B
α= S
n∈ω
q({n} × B
αn). Finally, we let A
α= B
α∪ {x
α} and observe that A
αis countable and dense in X
α. (Note that X
1and A
1are the Arens spaces described in Examples 1.6.19 and 1.6.20 of [En], respectively).
Observe that f is continuous on X
αif and only if f |q({n} × X
αn) and f |q(ω + 1), n ∈ ω, are continuous. It then inductively follows that C
p(X
a) ∈ M
2. We now show that C
Aα(X
α) 6∈ M
α. Consider the following subspaces E
αand F
αof C
Bα(X
α):
E
α= {f ∈ C
Bα(X
α) ∩ {0, 1}
Bα| f (x
α) = 0}, F
α= {f ∈ C
Bα(X
α) ∩ {0, 1}
Bα| f (x
α) = 1}.
Claim. Write G
α= E
α∪ F
α. The pairs (G
α, E
α) and (G
α, F
α) are Wadge (2
ω, A
α∪ M
α(2
ω))-complete (i.e., for every C ∈ A
α∪ M
α(2
ω) there exists a map ϕ : 2
ω→ G
αso that ϕ
−1(E
α) = C or ϕ
−1(F
α) = C, respec- tively).
We will provide an inductive proof. If α = 0, then E
0and F
0are copies of
the rationals and both E
0and F
0are dense in G
0. Let C ∈ M
0(2
ω) (i.e., C
is a closed subset of 2
ω). Pick a sequence {x
n}
∞n=1⊂ E
0such that lim x
n=
x
0∈ F
0. Find a map ϕ : 2
ω→ {x
n| n ∈ ω} such that ϕ
−1({x
0}) = C. If
C ∈ A
0(2
ω) (i.e., C is open in 2
ω), apply the same argument interchanging the roles of E
0and F
0. Let C ∈ A
αand pick C
n∈ M
αn, C
n⊆ C
n+1, n ∈ ω, so that C = S
n∈ω
C
n. We have E
α= FP(G
αn, E
αn) and F
α= FP(G
αn, F
αn), where for a sequence {(X
n, A
n)}
n∈ωof pairs of spaces
FP(X
n, A
n) = n
(x
n) ∈ Y
n∈ω
X
nx
n∈ A
na.e.
o
(notation of [CDM, Section 8]). Using the inductive assumption, we can find ϕ
n: 2
ω→ G
αn⊂ R
Bαnsuch that ϕ
−1n(E
αn) = C
n. Take ϕ = ∆ϕ
n: 2
ω→ R∪
Bαngiven by ϕ(p) = (ϕ
n(p)) and observe that ϕ( S
n∈ω
C
n) ⊂ FP(G
αn, E
αn) and ϕ(2
ω\ S
n∈ω
C
n) ⊂ Q
n∈ω
F
αn⊂ F
α(cf. [CDM, Proof of 8.1]). This settles the case where C ∈ A
α. By symmetry the same argu- ment works for C ∈ M
α.
An easy application of the Claim yields E
α6∈ M
α∪ A
α. Since E
αcan be identified with the set
{f ∈ C
Aα(X
α) ∩ {0, 1}
Aα| f (x
α) = 0}, a closed subset of C
Aα(X
α), our assertion follows.
2.7. R e m a r k. Let us note that even for a compact X the Borel class of C
A(X) can be as high as we wish. Such spaces X can be taken as Rosenthal compacta and were provided in [Ma1]. Let us indicate another way of finding such X. By [LvMP, Theorem 4.1], there exists a countable regular space A with exactly one nonisolated point such that the Borel class of C
p(A) (and hence, of C
p∗(A)) is as high as we wish. Let X = βA and observe that C
A(X) = C
p∗(A).
2.8. R e m a r k. Consider X = N
Ffor a filter F on ω. Applying 3.3, we see that C
ω(N
F) = C
Fcontains a closed copy of F ; and hence can have as complicated Borel structure as we wish. At the same time C
p(ω) = R
ω∈ M
1.
The statements 2.6 and 2.8 show that, within the Borel hierarchy, the Borel class of C
A(X) cannot be estimated by either the class of C
p(X) or C
p(A).
2.9. Question. Let X be a countable space and let A be a dense subset of X. Is the Borel (projective) class of C
p(X) (resp., C
p(A)) determined by that of C
A(X)? Is C
p(X) Borel (analytic) if C
A(X) is? Can the exact Borel (projective) class of C
p(X) be greater than that of C
A(X)?
2.10. R e m a r k. Let π
A: C
p(X) → C
A(X) be the projection map. Since
C
A(X) = π
A(C
p(X)) and π
Ais injective, C
A(X) is Borel provided C
p(X)
is (see [Kur]). We claim that also C
p(A) is analytic provided C
A(X) is. The
latter can be shown as follows. If C
A(X) is analytic, then for every accumu-
lation point a ∈ A the filter F
a= {Y ∩A | Y ∈ 2
X\{a}and a ∈ Int(Y ∪{a})}
is an analytic filter on A \ {a} (cf. the proof of [DMM, Corollary 3.6]). If now by A
Fawe denote the space A topologized by isolating the points of A \ {a} and by using the family {A ∪ {a} | A ∈ F
a} as neighborhood base at a, we have C
p(A) = T
{C
p(A
Fa) | a is an accumulation point of A}.
Since each C
p(A
Fa) is analytic, so is C
p(A) (this argument was taken from [DMM, Lemma 4.3]).
Let us notice that the last two questions of 2.9 have negative answers for bounded function spaces.
2.11. Example. For any ultrafilter p ∈ βω \ ω we have, C
ω∗(ω ∪ {p}) = Σ ∈ A
1\ M
1and C
p∗(ω ∪ {p}) 6∈ P
1∪ P
2.
Observe that, for every p from Example 2.11, we also have C
ω(ω ∪{p}) 6∈
P
1∪ P
2. Hence, we see that the gap between Borel classes of C
A∗(X) and C
A(X) can be as big as we wish.
3. Borel type of spaces C
F. For the Fr´echet filter F
0on ω we use the classical functional analysis symbols c
0and c to denote the spaces c
F0and C
F0. The spaces c
0and c considered as Banach spaces (with the sup norm) are linearly isomorphic. The following fact shows a dramatic difference if one considers c
0and c as subspaces of R
ω.
3.1. Proposition. The spaces c
0, c ⊂ R
ωare not linearly isomorphic.
P r o o f. Assume T : c
0→ c establishes a linear topological isomorphism of c
0onto c. Let k · k be the sup-norm on c (and c
0). First we note that the graph of T ,
Γ = {(x, T x) | x ∈ c
0} ⊂ c
0× c,
is closed in the norm topology. This follows from the obvious fact that Γ is closed with respect to the coordinatewise topology on c
0×c and the fact that the latter topology is weaker than the norm topology on c
0×c. Consequently, by the Closed Graph Theorem, T establishes a linear isomorphism of (c
0, k·k) onto (c, k · k).
Consider the continuous linear functional ` : (c, k · k) → R given by
`((x
n)) = lim x
n. It follows that `◦T : (c
0, k·k) → R is continuous and `◦T 6≡
0. This obviously implies that ` ◦ T is continuous with respect to the weak
topology ω on c
0. Consequently, for every convex closed neighborhood U of
0 ∈ (c
0, k·k), ker(`◦T )∩U is closed in the ω-topology, and ker(`◦T )∩U 6= U .
Let B = {y ∈ c | kyk ≤ 1} be the unit ball in c. We will show that for
U = T
−1(B), ker(` ◦ T ) ∩ U is dense in U in the ω-topology, contradicting
the above fact. To this end, first notice that given y ∈ B, there exists a
sequence {y
n}
n∈ω⊂ B ∩ c
0so that {y
n}
n∈ωconverges to y in R
ω. We see
that {T
−1y
n}
n∈ωconverges to T
−1y in R
ωand T
−1y
n∈ ker(` ◦ T ). Since
the ω-topology coincides with the R
ω-topology on bounded sets, it follows that {T
−1y
n} converges to T
−1y in the ω-topology.
Before we enter a discussion on Borel types of C
F, let us formulate the following general fact which will be employed later on; its proof can be obtained easily by using the argument of the proof of [Ma2, Lemma 2.1].
3.2. Lemma. Let F be a filter on ω which is not the Fr´echet filter. Then C
Fis homeomorphic to the product C
F× R
ω.
We identify a filter F on a countable set T with a subset of a Cantor set 2
T. As shown in [DMM, Lemma 4.2] the exact Borel class of c
Fis entirely determined by that of F in 2
T. Here we try to recover this result for the space C
F. The first result shows that Borel (projective) classes of C
Fand C
F∗are not lower than that of F .
3.3. Proposition. Let F be a filter on a countable set T . For every countable ordinal α and n ∈ ω, we have:
(a) if C
F∈ M
α, then F ∈ M
α, (b) if C
F∈ A
α, then F ∈ A
α, (c) if C
F∈ P
n, then F ∈ P
n.
In (a)–(c), the space C
Fcan be replaced by C
F∗provided F is not an ultra- filter.
P r o o f. We may assume T = ω. Consider
Z = {f ∈ C
F| ∀(n ∈ ω) (f (n) = 0 or f (n) = n + 1)},
a closed subset of C
F. The map (x
n) → ((n + 1)(1 − x
n)) embeds 2
ωin R
ωand sends F onto Z. Thus, the C
F-part of our assertion follows.
Assume that F is not an ultrafilter on ω (see 2.11). There exists a subset M ⊂ ω such that neither M nor N = ω \ M belongs to F . Consider the filters
G = {A ∩ M | A ∈ F } and H = {A ∩ H | A ∈ F } induced by F on M and N , respectively. The sets
Y
1= {f ∈ C
F∗∩ {0, 1}
ω| f |N = 0}, Y
2= {f ∈ C
F∗∩ {0, 1}
ω| f |M = 0}
are closed subsets of C
F∗. It is clear that G and H are homeomorphic to Y
1and Y
2, respectively. Finally, since F can be identified with the product G × H, our assertion follows.
Here is our counterpart of [DMM, Lemma 4.2] for spaces C
F.
3.4. Proposition. Let F be a filter on ω. For every countable ordinal
α ≥ 1 and n ∈ ω, we have:
(a) if F ∈ A
α, then C
F∈ M
α+1,
(b) if F ∈ M
α, then C
F∈ M
α+1∩ A
α+1(more exactly, C
Fis a differ- ence of two sets belonging to M
α),
(c) if F ∈ P
n, then C
F∈ P
n.
P r o o f. First we provide an argument which shows (a) and (c). Then we adapt this argument to check that whenever F ∈ M
αthen C
F∩ [0, 1]
ω∈ M
α. Finally, since C
Fis homeomorphic to
{f ∈ C
F| ∀(n ∈ ω) (f (n) ∈ (0, 1) and lim
F
f (n) ∈ (0, 1))}
= (C
F∩ [0, 1]
ω) ∩ (0, 1)
ω\ (c
F∪ (1 + c
F)), the space C
Fis a difference of two sets belonging to M
α(by [DMM, Lem- ma 4.2], c
Fand hence 1 + c
F= {(x
n+ 1) ∈ R
ω| (x
n) ∈ c
F} belong to M
α).
Assume F ∈ A
α(resp., F ∈ P
n). For n ∈ ω and k ∈ Z, write T
n,k=
2k
2n + 2 , 2k + 1 2n + 2
. Set T
n= S
k∈Z
T
n,k. For m ∈ N and l ∈ Z, define S
m,l,n= [
k
T
n,kdist
l m , T
n,k≤ 1 m
.
Let %
m,l: Q
n∈ω
T
n→ 2
ωbe given by
%
m,l(f )(n) =
1 if f (n) ∈ S
m,l,n, 0 otherwise, for f ∈ Q
n∈ω
T
n. One can check that %
m,lis continuous. We claim that
(1) C
F∩ Y
n∈ω
T
n=
\
∞ m=1[
l∈Z
%
−1m,l(F ).
(If f = (f (n)) ∈ Q
n∈ω
T
nand there exists x ∈ R such that for all ε > 0 we have f
−1((x − ε, x + ε)) ∈ F , then for every m ≥ 1 one can find l ∈ Z so that |x − l/m| ≤ 1/(2m) and check that {n | f (n) ∈ S
m,l,n} ∈ F ; this shows %
m,l(f ) ∈ F . Conversely, if for every m ≥ 1 one can find l ∈ Z so that {n | f (n) ∈ S
m,l,n} ∈ F , then there exists A ∈ F so that whenever i, j ∈ A then |f (i) − f (j)| < 4/m; this implies f ∈ C
F.)
Pick a map ψ
n: T
n→ R that transforms T
n,kaffinely onto
2k2n+2
,
2k+22n+2, k ∈ Z. Letting ψ = Q
n∈ω
ψ
n: Q
n∈ω
T
n→ R
ωwe see that ψ is a perfect map and that
(2) C
F∩ Y
n∈ω
T
n= ψ
−1(C
F).
Applying (1), C
F∩ Q
n∈ω
T
n∈ M
α+1(resp., C
F∩ Q
n∈ω
T
n∈ P
n). By (2)
and a result of Saint Raymond [SR1], C
F∈ M
α+1(resp., C
F∈ P
n).
Now, let F ∈ M
α. We must show that C
F∩ [0, 1]
ω∈ M
α. Using the sets T
n,kwe define
T
n=
n−1
[
k=0
T
n,k.
With these T
n, as previously, define for m ∈ N and l ∈ {0, . . . , m} the sets S
m,l,nand maps %
m,l. One can check
(1)
0C
F∩ Y
n∈ω
T
n=
\
∞ m=1[
m l=0%
−1m,l(F ).
Let ψ
n: T
n→ [0, 1] be a map such that T
n,kis affinely transformed onto
2k2n+2
,
2n+22k+2. Letting ψ = Q
n∈ω
ψ
n: Q
n∈ω
T
n→ [0, 1]
ω, we have
(2)
0C
F∩ Y
n∈ω
T
n= ψ
−1(C
F∩ [0, 1]
ω).
Using (1)
0, C
F∩ Q
n∈ω
T
n∈ M
α. As above, by (2)
0and a result of Saint Raymond [SR1], C
F∩ [0, 1]
ω∈ M
α.
3.5. Question. Let F ∈ M
αbe a filter on ω. Is C
F∈ M
α?
Let F be a filter on ω. Identify the space C
Fwith C
ω(N
F). Using this identification, let ` : C
F→ R be given by `(f ) = f (∞). We see that ` is well defined and can be viewed as `(f ) = lim
Ff (n).
3.6. Proposition. For a filter F ∈ A
α∪ M
α, and a countable ordinal α, α ≥ 1, ` : C
F→ R is Borel of class α.
P r o o f. Let U ⊆ R be an open set. We must show that `
−1(U ) ∈ A
α(C
F).
We shall use some of the notation employed in the proof of 3.4. For n ∈ ω and m ∈ N, let
P
n,m= [
{T
n,k| ∀(t ∈ T
n,k) (dist(t, R \ U ) > 1/m)}.
Recall that T
n= S
k∈Z
T
n,kand define ζ
m: Q
n∈ω
T
n→ {0, 1}
ωby letting for f ∈ Q
n∈ω
T
n,
ζ
m(f )(n) =
n 1 if f (n) ∈ P
n,m, 0 otherwise.
Clearly, ζ
mis continuous, m ≥ 1. Let ψ : Q
n∈ω
T
n→ R
ωbe that of the proof of 3.4.
Suppose F ∈ A
α. Consider S = S
∞m=1
ζ
m−1(F ). We see that S ∈ A
α(S ∈ A
1( Q
n∈ω
T
n) for α = 1). We claim that ψ
−1(ψ(S)) = S. Let f, g ∈ Q
n∈ω
T
nbe such that g ∈ S
∞m=1
ζ
m−1(F ) and ψ(f ) = ψ(g). Since ψ(f ) = ψ(g), we have |g(n) − f (n)| < 1/n. Take m
0such that g ∈ ζ
m−10(F ). If n > 2m
0and g(n) ∈ P
n,m0, then f (n) ∈ P
n,2m0. It follows that ζ
2m0(f ) ⊃ ζ
m0(g) \ {0, 1, . . . , 2m
0}; and hence ζ
2m0(f ) ∈ F . This shows that f ∈ ζ
2m−10(F ) ⊂ S.
Suppose F ∈ M
α. Let J = {A ⊂ ω | ω \ A ∈ F } be the dual ideal of F ; J is homeomorphic to F . Consider T = ( Q
n∈ω
T
n) \ T
∞m=1
ζ
n−1(J ). We see that T ∈ A
α. We claim that ψ
−1(ψ(T )) = T . Let f, g ∈ Q
n∈ω
T
nbe such that f 6∈ T
∞m=1
ζ
m−1(J ) and ψ(f ) = ψ(g). Take m
0such that f 6∈ ζ
m−10(J ).
If n > 2m
0and f (n) ∈ P
n,m0, then g(n) ∈ P
n,2m0. As previously, it follows that ζ
2m0(g) 6∈ J ; hence g ∈ T .
We now claim that
(1) C
F∩ S = C
F∩ T = Y
n∈ω
T
n∩ `
−1(U ).
Namely, we have f ∈ S (resp., f ∈ T ) if and only if there exists m ≥ 1 such that f ∈ ζ
m−1(F ) (resp., f 6∈ ζ
m−1(J )). Suppose f ∈ ( Q
n∈ω
T
n) ∩ `
−1(U ) and f (∞) = x ∈ U . Let ε = dist(x, R \ U ). For m, n > 4/ε we have (x − ε/2, x + ε/2) ∩ T
n⊆ P
n,m. It follows that f ∈ ζ
m−1(F ) and hence f 6∈ ζ
m−1(J ). Suppose now that f ∈ C
Fand f ∈ ζ
m−1(F ). Since f 6∈ ζ
m−1(J ) the set R = {n | f (n) ∈ P
n,m} is not in J . Take x 6∈ U . Then for every n ≥ 1, (x − 1/(2m), x + 1/(2m)) ∩ P
n,m= ∅. Therefore for Q = {n | f (n) ∈ (x − 1/(2m), x + 1/(2m))} we have R ∩ Q = ∅. Since R 6∈ J , we have Q 6∈ F . It follows that f (∞) 6= x, so f (∞) ∈ U .
Since lim
n(f (n)−ψ(f )(n)) = 0, one sees that lim
Ff (n) = lim
Fψ(f )(n);
and consequently
`
−1(U ) = ψ Y
n∈ω
T
n∩ `
−1(U )
.
This together with (1) and the fact that ψ
−1(C
F) = C
F∩ Q
n∈ω
T
nyields
`
−1(U ) = ψ(C
F∩ S) = C
F∩ ψ(S),
`
−1(U ) = ψ(C
F∩ T ) = C
F∩ ψ(T ).
The facts that ψ
−1(ψ(S)) = S and ψ
−1(ψ(T )) = T together with a result of Saint Raymond [SR1] show that, under our assumption, C
F∩ ϕ(S) and C
F∩ ϕ(T ) belong to A
α(C
F).
The following is a partial answer to Question 3.5.
3.7. Corollary. For n ∈ ω, let F
nbe a filter on ω with F
n∈ S
β<α
A
β, α ≥ 1. Then for the filter F = Q
n∈ω
F
non ω × ω, we have C
F∈ M
α. P r o o f. We obviously have
F = {A ⊂ ω × ω | ∀(n ∈ ω) ({m | (m, n) ∈ A} ∈ F
n)}.
Moreover, for f = (f (m, n)) ∈ R
ω×ω, f ∈ C
Fif and only if f (·, n) ∈ C
Fn, n ∈ ω, and lim
Fnf (·, n) = lim
Fkf (·, k) for n, k ∈ ω. Now our assertion follows by application of 3.4 and 3.6.
The assertion of Proposition 3.9 below shows that the result of [DDM, Lemma 4.2] is fully recovered for the spaces C
F∗. The proof of 3.9 requires a particular case of the following general fact.
3.8. Proposition. Let X be a separable metrizable space such that X = S
∞n=1
X
n, X
n∈ M
α, α ≥ 2 and X
n∈ A
β(X), where β < α. Then X ∈ M
α. P r o o f. Let Y be a metrizable compactification of X and let Y
n∈ A
βbe such that Y
n∩ X = X
n. Consider Z = S
∞n=1
Y
n∈ A
β⊆ M
α. We have Y
n\X
n∈ A
α; hence, S
∞n=1
(Y
n\X
n) ∈ A
αand X = Z\ S
∞n=1
(Y
n\X
n) ∈ M
α.
3.9. Proposition. Let F be a filter on ω, α be a countable ordinal, α ≥ 1, and let n ∈ ω. We have:
(a) if F ∈ A
α\ M
α, then C
F∗∈ M
α+1\ A
α+1, (b) if F ∈ M
α\ A
α, then C
F∗∈ M
α\ A
α, (c) if F ∈ M
α∩ A
α\ S
β<α
(A
β∪ M
β), then C
F∗∈ M
α\ A
α, (d) if F ∈ P
n, then C
F∗∈ P
n.
P r o o f. The proof of the second part of 3.4 yields that whenever F ∈ M
α, C
F∩ [−n, n]
ω∈ M
αfor n ≥ 1. Since C
F∗= S
∞n=1
(C
F∩ [−n, n]
ω) and each C
F∩ [−n, n]
ωis closed in C
F∗, it follows from 3.8 that C
F∗∈ M
α(observe that here α ≥ 2 since there are no filters of class M
1). If additionally F 6∈ A
αthen, by 3.3, C
F∗∈ M
α\A
α; hence (b) follows. Assume F ∈ M
α∩A
α\ S
β<α
(A
β∪M
β). Consider the decomposition F = G×H of F into filters G and H described in the proof of 3.3 (obviously, being Borel, F is not an ultrafilter). We may assume that G ∈ M
α∩A
α\ S
β<α
M
β(if not, H has this property). Recall that Σ = {(x
n) ∈ R
ω| (x
n) is bounded}. Applying [DMM, Lemma 4.2(3)], we find that c
G∈ M
α\ A
α. Since c
∗G= c
G∩ Σ and c
Gis homeomorphic to c
∗G∩ (−1, 1)
ω, we conclude that c
∗G∈ M
α\ A
α. Finally, since C
F∗contains c
∗Gas a closed set, it follows that C
F∗∈ M
α\ A
α; this shows (c).
To get (a), suppose F ∈ A
α\ M
α. Since C
F∗= C
F∩ Σ, 3.4(a) yields C
F∗∈ M
α+1. Considering, as previously, F = G × H, we may assume G ∈ A
α\ M
α. By [DDM, Lemma 4.2(4)], c
G∈ M
α+1\ A
α+1. As before, this implies that c
∗G∈ M
α+1\ A
α+1and consequently C
F∗∈ M
α+1\ A
α+1.
The assertion (d) follows from 3.4(c) and the fact that C
F∗= C
F∩ Σ.
We now provide a counterpart of [DMM, Lemma 4.3]. Let X be a count-
able space. For every accumulation point a of X we consider the filter F
aon X \ {a} consisting of all neighborhoods of a; hence F
a= {Y ∈ 2
X\{a}| a ∈ Int(Y ∪ {a})}.
3.10. Lemma. Let X be a countable space and A ⊆ X be a dense set.
If there exists a countable ordinal α, α ≥ 1 (resp., an integer n ∈ ω) such that for every accumulation point a ∈ X the filter F
ais in A
α∪ M
α(resp., F
a∈ P
n), then C
A(X) ∈ M
α+1(resp., C
A(X) ∈ P
n).
P r o o f (cf. [DMM, Lemma 4.3]). Let a be an accumulation point of X.
Define a filter on A by letting
G
a= {Y ∩ A | Y ∈ F
a}.
The filter G
ais homeomorphic to
{Y ∈ F
a| X \ (A ∪ {a}) ⊆ Y },
a closed subset of F
a. Therefore G
a∈ A
a∪ M
α(resp., G
a∈ P
n). Let X
abe the space A ∪ {a} topologized by isolating the points of A \ {a} and using the family {Y ∪ {a} | Y ∈ G
a} as a neighborhood base at a (it may happen that a ∈ A). Then either C
A(X
a) is isomorphic to c
Ga(in case a ∈ A), or else C
A(X
α) is isomorphic to C
Gα(in case a 6∈ A). From Proposition 3.4 and [DMM, Lemma 4.2] (see also [CDM, Corollary 5.3(d)]) it follows that in both cases C
A(X
a) ∈ M
α+1(resp., C
A(X
a) ∈ P
n). Now, 3.10 is a consequence of the following observation:
C
A(X) = \
{C
A(X
a) | a is an accumulation point of X}.
3.11. R e m a r k. Let X
αbe the space from Proposition 2.6. Then C
p(X
α)
∈ F
σδ. Moreover, according to Lemma 3.10, there exists a ∈ X
αsuch that F
a6∈ S
β<α
A
β∪ M
β; in fact, one can show that a = x
αhas this prop- erty. (As shown in [DMM, Corollary 3.6] the filters F
aare analytic provided C
p(X) is analytic.) Observe that the filter F
ahas a base which is of the F
σδ-type. Namely, the filter base {Y ⊂ X \ {a} | Y ∪ {a} is a clopen subset of X containing a} is homeomorphic to
{f ∈ C
p(X
α) ∩ {0, 1}
Xα| f (a) = 1}, a closed subset of C
p(X
α).
4. Z
σ-property of C
A(X). We recall that a closed subset A of an absolute neighborhood retract M is a Z-set if every map f : K → M of a compactum K into M can be approximated by maps f : K → M \ A.
A space which is a countable union of its own Z-sets is called a Z
σ-space.
Obviously, every Z
σ-space is of the first category. It is a well-known fact
that a dense convex subset of a convex Z
σ-space is itself a Z
σ-space. Since
C
A(X) is dense in R
Aand since Σ = {(x
n) ∈ R
ω| (x
n) is bounded} is a
Z
σ-space, we have:
4.1. Lemma. Let X be a space and let A be a countable dense subset of X.
(a) The space C
A∗(X) is a Z
σ-space.
(b) If X is compact, then C
A(X) is a Z
σ-space.
4.2. Proposition. Let X be a countable nondiscrete space and let A be a dense subset of X. If C
A(X) is analytic, then C
A(X) and C
p(X) are Z
σ-spaces.
Let us recall that a filter F on a countable set T is of the first category if F belongs to the σ-field generated by the open sets and the first category sets of 2
T. We need the following fact.
4.3. Lemma. Let F be a first category filter on a countable set T . Then for every 0 < r ≤ ∞ the space C
F∩ [−r, r]
ωis a Z
σ-space.
P r o o f. By a result of Talagrand [Ta, Theorem 2.1] there exists a se- quence of pairwise disjoint finite sets A
n⊂ T such that every A ∈ F meets all but finitely many A
n, n ∈ N.
For finite r, let
X
n= {f ∈ C
F∩ [−r, r]
ω| ∀(k > n)∃(m ∈ A
k) (|f (m)| ≤ 2r/3)}
and
Y
n= {f ∈ C
F∩ [−r, r]
ω| ∀(k > n)∃(m ∈ A
k) (|f (m)| ≥ r/3)}.
An argument of the proof of [DMM, Proposition 3.3] shows that each X
nand Y
n, n ≥ 1, is a Z-set. Moreover, C
F∩ [−r, r]
ω= S
∞n=1
X
n∪ Y
n. For r = ∞, let
X
n,l= {f ∈ C
F| ∀(k > n)∃(m ∈ A
k) (|f (m)| ≤ l)}.
As previously, each X
n,lis a Z-set in C
Fand C
F= S
∞n,l=1