Introduction
The study of special subsets of the real line is a classical topic at the intersection of topology and set theory. Such sets are defined by properties concerning descriptive complexity, Baire category and measure theoretic notions. For example, a λ \lambda λ -set is a set in which every countable subset is a relative G δ G_\delta G δ , a σ \sigma σ -set is one in which every G δ G_\delta G δ subset is a relative F σ F_\sigma F σ , and a Q Q Q -set is one in which every subset is a relative G δ G_\delta G δ . The possible cardinalities of these sets and the relationships between them are closely tied to cardinal characteristics of the continuum and to additional set-theoretic assumptions. We refer to [16 ] for an overview of this field of research.
These classes are also deeply connected to the study of separation properties of topological spaces, such as normality. This connection led to the introduction of weak λ \lambda λ -sets in [20 ] , a weakening of λ \lambda λ -sets, while studying separation properties of Isbell–Mrówka spaces.
Definition 1.1. A topological space X X X is a weak λ \lambda λ -space if, whenever A , B ⊆ X A, B \subseteq X A , B ⊆ X are disjoint and countable, there is a set H ⊆ X H \subseteq X H ⊆ X which is both F σ F_\sigma F σ and G δ G_\delta G δ in X X X such that A ⊆ H A \subseteq H A ⊆ H and B ⊆ X ∖ H B \subseteq X \setminus H B ⊆ X ∖ H . A separable metrizable weak λ \lambda λ -space is called a weak λ \lambda λ -set.
Every λ \lambda λ -set is a weak λ \lambda λ -set. Rodrigues, Ronchim and Szeptycki asked whether the converse can consistently fail [20 Question 4.6] .
An almost disjoint family is a collection A ⊆ [ ω ] ω \mathcal{A} \subseteq[\omega]^\omega A ⊆ [ ω ] ω whose distinct members have finite intersection. Given such a family, the Isbell–Mrówka space Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is obtained by adjoining to the discrete space ω \omega ω one point for each member of A \mathcal{A} A . A basic neighborhood of a ∈ A a \in\mathcal{A} a ∈ A is of the form { a } ∪ ( a ∖ F ) \{a\} \cup(a \setminus F) { a } ∪ ( a ∖ F ) for some finite F ⊆ a F \subseteq a F ⊆ a . The space Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is Hausdorff, separable, scattered, zero-dimensional, first countable, locally compact and regular. Its topological properties are often related to combinatorial properties of the underlying almost disjoint family A \mathcal{A} A : for instance, Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is pseudocompact if and only if A \mathcal{A} A is maximal, and Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is metrizable if and only if A \mathcal{A} A is countable. An infinite almost disjoint family A \mathcal{A} A is maximal (a MAD family) if every infinite subset of ω \omega ω meets some member of A \mathcal{A} A infinitely. For background on Isbell–Mrówka spaces, see [9 , 11 ] .
Since A \mathcal{A} A is closed and discrete, Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is not countably compact when A \mathcal{A} A is infinite [9 , 11 ] . A normal pseudocompact space is countably compact, so Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is not normal when A \mathcal{A} A is MAD. Also, Jones’ Lemma shows that Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is not normal whenever ∣ A ∣ = c |\mathcal{A}| = \mathfrak{c} ∣ A ∣ = c . Moreover, in ZFC, there exist almost disjoint families of cardinality ω 1 \omega_1 ω 1 which are not normal, e.g., Luzin families (see, e.g., [11 Sections 3.1–3.2] ). Nevertheless, it is consistent with ZFC that there are uncountable almost disjoint families A \mathcal{A} A for which Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is normal [22 ] .
Several weakenings of normality for Ψ \Psi Ψ -spaces have been studied, such as almost-normality, quasi-normality, partly-normality, and mild normality [6 ] . In this paper, we focus on almost-normality. A space is almost-normal if any two disjoint closed sets, one of which is regular closed, have disjoint open neighborhoods.
For X ⊆ 2 ω X \subseteq2^\omega X ⊆ 2 ω , let A X \mathcal{A}_X A X consist of the branches { x ↾ n : n ∈ ω } \{x \mathbin{\upharpoonright} n : n \in\omega\} { x ↾ n : n ∈ ω } , for x ∈ X x \in X x ∈ X . The space Ψ ( A X ) \Psi(\mathcal{A}_X) Ψ ( A X ) is normal if and only if X X X is a Q Q Q -set, almost-normal if and only if X X X is a σ \sigma σ -set, and pseudonormal if and only if X X X is a λ \lambda λ -set [20 Corollary 2.3, Theorem 2.9 and Proposition 3.3] .
Another separation property related to Isbell–Mrówka spaces is strong ℵ 0 \aleph_0 ℵ 0 -separation. We say that A \mathcal{A} A is strongly ℵ 0 \aleph_0 ℵ 0 -separated if any two disjoint countable closed subsets of Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) can be separated by clopen sets. For A X \mathcal{A}_X A X as above, this is equivalent to X X X being a weak λ \lambda λ -set [20 Proposition 4.3] .
For a class of spaces P \mathcal{P} P , write non ( P ) \operatorname{non}(\mathcal{P}) non ( P ) for the least cardinality of a separable metrizable space outside P \mathcal{P} P . Let b \mathfrak{b} b and s \mathfrak{s} s denote the bounding and splitting numbers, respectively (for their definitions, see § 2 ). It is well-known that non ( λ ) = non ( σ ) = b \operatorname{non}(\lambda) = \operatorname{non}(\sigma) = \mathfrak{b} non ( λ ) = non ( σ ) = b [25 Theorem 9.1] . We also write q = non ( Q ) \mathfrak{q} = \operatorname{non}(Q) q = non ( Q ) for the least cardinality of a set of reals that is not a Q Q Q -set [3 p. 13] . The least cardinality of a set of reals that is not a weak λ \lambda λ -set, that is, non ( w λ ) \operatorname{non}(w\lambda) non ( w λ ) , was left open [20 ] . In this paper, we settle this question by proving that non ( w λ ) = max { b , s } \operatorname{non}(w\lambda) = \max\{\mathfrak{b}, \mathfrak{s}\} non ( w λ ) = max { b , s } .
Theorem A. non ( w λ ) = max { b , s } \operatorname{non}(w\lambda) = \max\{\mathfrak{b}, \mathfrak{s}\} non ( w λ ) = max { b , s } .
The following diagram summarizes these relationships for families of branches.
Figure 1. Special sets of reals and separation properties of their families of branches. The top row gives the least cardinality of a set of reals outside each class. Horizontal arrows denote implications and vertical arrows denote equivalences.
Rodrigues, Ronchim and Szeptycki asked whether there is, at least consistently, a weak λ \lambda λ -set which is not a λ \lambda λ -set [20 Question 4.6] . The preceding theorem, together with non ( λ ) = b \operatorname{non}(\lambda)=\mathfrak{b} non ( λ ) = b , gives a positive answer when b < s \mathfrak{b}<\mathfrak{s} b < s . Since each of the inequalities b < s \mathfrak{b}<\mathfrak{s} b < s and s < b \mathfrak{s}<\mathfrak{b} s < b is consistent relative to ZFC, the observation above and the following theorem show that the existence of a weak λ \lambda λ -set which is not a λ \lambda λ -set is independent of ZFC, relative to its consistency.
Theorem B. If s < b \mathfrak{s}<\mathfrak{b} s < b , then every weak λ \lambda λ -set is a λ \lambda λ -set.
The remaining question is whether there exists a weak λ \lambda λ -set which is not a λ \lambda λ -set if s = b \mathfrak{s}=\mathfrak{b} s = b . In the model obtained by adding ω 2 \omega_2 ω 2 Cohen reals over a model of CH, we have s = b = ℵ 1 \mathfrak{s}=\mathfrak{b}=\aleph_1 s = b = ℵ 1 [2 Section 11.3] . Let C ω 2 \mathbb{C}_{\omega_2} C ω 2 denote the forcing for adding ω 2 \omega_2 ω 2 Cohen reals. We prove the following.
Theorem C. Assume CH. Then C ω 2 ⊩ \mathbb{C}_{\omega_2}\Vdash C ω 2 ⊩ every weak λ \lambda λ -set is a λ \lambda λ -set.
On the other hand, we show that if p = c \mathfrak{p}=\mathfrak{c} p = c , then there is a weak λ \lambda λ -set which is not a λ \lambda λ -set, where p \mathfrak{p} p is the pseudointersection number (for its definition, see § 2 ). This hypothesis implies s = b = c \mathfrak{s}=\mathfrak{b}=\mathfrak{c} s = b = c . Under this assumption, we obtain a positive answer to the question above. This result was also obtained independently by Osipov and Lipin [18 Theorem 1.1] .
Theorem D. If p = c \mathfrak{p}=\mathfrak{c} p = c , then there is a weak λ \lambda λ -set which is not a λ \lambda λ -set.
Together, these results show that, relative to the consistency of ZFC, the existence of a weak λ \lambda λ -set which is not a λ \lambda λ -set is independent of ZFC + s = b +\mathfrak{s}=\mathfrak{b} + s = b .
We now turn to almost-normal MAD families. As noted above, a MAD family cannot be normal. García-Balan and Szeptycki asked whether there is an almost-normal MAD family [6 Question 4.3] . The second and third authors used forcing to construct an almost-normal almost disjoint family which is not normal [19 ] . The first author showed that CH implies the existence of an almost-normal almost disjoint family which is not normal, and that the Proper Forcing Axiom implies that no MAD family is almost-normal [4 ] . An earlier version of the first author’s work claimed the existence of an almost-normal MAD family under CH, but the argument contained a gap [5 Theorem 2.7] . In § 7 , we obtain the following result.
Theorem E. Assuming CH, there exists an almost-normal MAD family.
For the opposite direction, we use Todorčević’s Open Coloring Axiom O C A T \mathrm{OCA}_{\mathrm{T}} OCA T and the cardinal characteristic of the continuum a p \mathfrak{ap} ap [3 p. 14] . In § 8 , we recall the definition of a p \mathfrak{ap} ap and strengthen the first author’s result by proving the following.
Theorem F. Assume O C A T + a p > ω 1 \mathrm{OCA}_{T}+\mathfrak{a}_{p}>\omega_{1} OCA T + a p > ω 1 . Then almost-normal MAD families do not exist.
Together, the last two theorems stated above show that the existence of an almost-normal MAD family is independent of ZFC.
The paper is organized as follows.
Section 2: we introduce the notation and preliminary results used throughout the paper.
Section 3: we prove Theorem A, proving that non ( w λ ) = max { b , s } \operatorname{non}(\mathrm{w}\lambda)=\max\{\mathfrak{b},\mathfrak{s}\} non ( w λ ) = max { b , s } , and obtain as a corollary a consistent example of a weak λ \lambda λ -set which is not a λ \lambda λ -set under b < s \mathfrak{b}<\mathfrak{s} b < s .
Section 4: we prove Theorem D, constructing a weak λ \lambda λ -set which is not a λ \lambda λ -set under p = c \mathfrak{p}=\mathfrak{c} p = c . This result was also obtained independently by Osipov and Lipin [18 Theorem 1.1] using similar techniques before the announcement of the present work.
Section 5: we prove a strengthening of Theorem B, showing that s < b \mathfrak{s}<\mathfrak{b} s < b implies that every first countable weak λ \lambda λ -space is a λ \lambda λ -space. This shows that the existence of a weak λ \lambda λ -set which is not a λ \lambda λ -set is independent of ZFC.
Section 6: we prove Theorem C, showing that every weak λ \lambda λ -set is a λ \lambda λ -set in the Cohen model. Together with Theorem D, this yields that the existence of a weak λ \lambda λ -set which is not a λ \lambda λ -set is independent of ZFC + s = b +\mathfrak{s}=\mathfrak{b} + s = b .
Section 7: we prove Theorem E, constructing an almost-normal MAD family under CH.
Section 8: we prove Theorem F showing that no almost-normal MAD family exists under O C A T + a p > ω 1 \mathrm{OCA}_{T}+\mathfrak{a}_{p}>\omega_{1} OCA T + a p > ω 1 as a consequence of known results from the literature. Together with Theorem E, this implies that the existence of an almost-normal MAD family is independent of ZFC.
Preliminaries
Our terminology is mainly standard: given two sets a , b a,b a , b , we write a ⊆ ∗ b a\subseteq^{*}b a ⊆ ∗ b when a ∖ b a\setminus b a ∖ b is finite and a = ∗ b a=^{*}b a = ∗ b when a △ b a\mathbin{\triangle}b a △ b is finite, where a △ b a\mathbin{\triangle}b a △ b denotes the symmetric difference of a a a and $b. We denote by P ( X ) \mathcal{P}(X) P ( X ) the power set of X X X , by [ X ] κ [X]^{\kappa} [ X ] κ the set of all subsets of X X X with cardinality κ \kappa κ , and [ X ] < κ = ⋃ η < κ [ X ] η [X]^{<\kappa}=\bigcup_{\eta<\kappa}[X]^{\eta} [ X ] < κ = ⋃ η < κ [ X ] η .
For an almost disjoint family over a countably infinite set N N N , we use Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) for the space defined in the introduction, with N N N in place of ω \omega ω . When forming N ∪ A N\cup\mathcal{A} N ∪ A , we assume N ∩ [ N ] ω = ∅ N\cap[N]^{\omega}=\varnothing N ∩ [ N ] ω = ∅ .
We will use the following shorthand.
Definition 2.1. We say that A \mathcal{A} A satisfies the topological property P P P if the space Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) satisfies P P P .
Given L , R ⊆ P ( ω ) \mathcal{L},\mathcal{R}\subseteq\mathcal{P}(\omega) L , R ⊆ P ( ω ) , we write L ⊥ R \mathcal{L}\perp\mathcal{R} L ⊥ R if ∣ L ∩ R ∣ < ω |L\cap R|<\omega ∣ L ∩ R ∣ < ω for every L ∈ L L\in\mathcal{L} L ∈ L and R ∈ R R\in\mathcal{R} R ∈ R . In this case, we say that L \mathcal{L} L and R \mathcal{R} R are orthogonal .
Definition 2.2. Let L , R ⊆ P ( ω ) \mathcal{L},\mathcal{R}\subseteq\mathcal{P}(\omega) L , R ⊆ P ( ω ) . A separator for L \mathcal{L} L and R \mathcal{R} R is a set D ⊆ ω D\subseteq\omega D ⊆ ω such that
L ⊆ ∗ D ( ∀ L ∈ L ) and ∣ R ∩ D ∣ < ω ( ∀ R ∈ R ) . L\subseteq^{*}D\quad(\forall L\in\mathcal{L})\qquad\text{and}\qquad|R\cap D|<\omega\quad(\forall R\in\mathcal{R}). L ⊆ ∗ D ( ∀ L ∈ L ) and ∣ R ∩ D ∣ < ω ( ∀ R ∈ R ) . If D D D is a separator for L \mathcal{L} L and R \mathcal{R} R , we say that D D D separates L \mathcal{L} L from R \mathcal{R} R .
Notice that if D D D separates L \mathcal{L} L from R \mathcal{R} R , then L ⊥ R \mathcal{L} \perp\mathcal{R} L ⊥ R . Note that this concept is not symmetric. However, if D D D separates L \mathcal{L} L from R \mathcal{R} R , then its complement separates R \mathcal{R} R from L \mathcal{L} L . It is well known that countable orthogonal families can be separated. We include its proof for the convenience of the reader.
Lemma 2.3. Let L , R ⊆ P ( ω ) \mathcal{L},\mathcal{R} \subseteq\mathcal{P}(\omega) L , R ⊆ P ( ω ) be countable collections such that L ⊥ R \mathcal{L} \perp\mathcal{R} L ⊥ R , and let C ∈ L C \in\mathcal{L} C ∈ L . Then there exists a separator D ⊇ C D \supseteq C D ⊇ C for L \mathcal{L} L and R \mathcal{R} R .
Proof. If R = ∅ \mathcal{R} = \varnothing R = ∅ , take D = ω D = \omega D = ω . Otherwise, enumerate both collections with repetitions if needed as L = { L n : n < ω } \mathcal{L} = \{L_n : n < \omega\} L = { L n : n < ω } and R = { R n : n < ω } \mathcal{R} = \{R_n : n < \omega\} R = { R n : n < ω } . Let
D = C ∪ ⋃ n < ω ( L n ∖ ⋃ j ≤ n R j ) . D = C \cup\bigcup_{n<\omega}\left(L_n \setminus\bigcup_{j\leq n} R_j\right). D = C ∪ n < ω ⋃ ( L n ∖ j ≤ n ⋃ R j ) . For every n ∈ ω n \in\omega n ∈ ω , L n ∖ D ⊆ ⋃ j ≤ n ( L n ∩ R j ) L_n \setminus D \subseteq\bigcup_{j\leq n}(L_n \cap R_j) L n ∖ D ⊆ ⋃ j ≤ n ( L n ∩ R j ) , so L n ⊆ ∗ D L_n \subseteq^* D L n ⊆ ∗ D . On the other hand, for each m ∈ ω m \in\omega m ∈ ω ,
R m ∩ D ⊆ R m ∩ C ∪ ⋃ n < m ( R m ∩ L n ) . R_m \cap D \subseteq R_m \cap C \cup\bigcup_{n<m}(R_m \cap L_n). R m ∩ D ⊆ R m ∩ C ∪ n < m ⋃ ( R m ∩ L n ) . The right-hand side is finite, so ∣ R m ∩ D ∣ < ω |R_m \cap D| < \omega ∣ R m ∩ D ∣ < ω . Thus D D D is a separator for L \mathcal{L} L and R \mathcal{R} R . □ \square □
A separator that separates an almost disjoint family A \mathcal{A} A into two orthogonal subfamilies is called a partitioner for A \mathcal{A} A .
Definition 2.4. Let A \mathcal{A} A be an almost disjoint family.
(1) A set D ⊆ ω D \subseteq\omega D ⊆ ω is a partitioner for A \mathcal{A} A if it separates some subfamily B ⊆ A \mathcal{B} \subseteq\mathcal{A} B ⊆ A from A ∖ B \mathcal{A} \setminus\mathcal{B} A ∖ B .
(2) For C ⊆ ω C \subseteq\omega C ⊆ ω , let A C = { a ∈ A : ∣ a ∩ C ∣ = ω } \mathcal{A}_C = \{a \in\mathcal{A} : |a \cap C| = \omega\} A C = { a ∈ A : ∣ a ∩ C ∣ = ω } .
The following proposition reformulates [19 Theorem 3.6] and [4 Proposition 2.3] . Recall from [6 ] that the regular closed subsets of Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) are precisely the sets cl Ψ ( A ) C = C ∪ A C \operatorname{cl}_{\Psi(\mathcal{A})} C = C \cup\mathcal{A}_C cl Ψ ( A ) C = C ∪ A C , with C ⊆ ω C \subseteq\omega C ⊆ ω .
Proposition 2.5. For an almost disjoint family A \mathcal{A} A , the following are equivalent.
(1) Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is almost-normal.
(2) For every C ⊆ ω C \subseteq\omega C ⊆ ω , there is a partitioner for A \mathcal{A} A which separates A C \mathcal{A}_C A C from A ∖ A C \mathcal{A} \setminus\mathcal{A}_C A ∖ A C .
We shall use the following cardinal characteristics of the continuum. An unbounded family is a set B ⊆ ω ω \mathcal{B} \subseteq\omega^\omega B ⊆ ω ω such that for every f ∈ ω ω f \in\omega^\omega f ∈ ω ω , there exists g ∈ B g \in\mathcal{B} g ∈ B with f ≤ ∞ g f \leq^\infty g f ≤ ∞ g , i.e., f ( n ) ≤ g ( n ) f(n) \leq g(n) f ( n ) ≤ g ( n ) for infinitely many n ∈ ω n \in\omega n ∈ ω . The bounding number is
b = min { ∣ B ∣ : B ⊆ ω ω is unbounded } . \mathfrak{b} = \min\{|\mathcal{B}| : \mathcal{B} \subseteq\omega^\omega\text{ is unbounded}\}. b = min { ∣ B ∣ : B ⊆ ω ω is unbounded } .
A set S ⊆ ω S \subseteq\omega S ⊆ ω splits A ∈ [ ω ] ω A \in[\omega]^\omega A ∈ [ ω ] ω if both A ∩ S A \cap S A ∩ S and A ∖ S A \setminus S A ∖ S are infinite. A splitting family is a set S ⊆ P ( ω ) \mathcal{S} \subseteq\mathcal{P}(\omega) S ⊆ P ( ω ) such that every A ∈ [ ω ] ω A \in[\omega]^\omega A ∈ [ ω ] ω is split by some S ∈ S S \in\mathcal{S} S ∈ S . The splitting number is
s = min { ∣ S ∣ : S ⊆ P ( ω ) is a splitting family } . \mathfrak{s} = \min\{|\mathcal{S}| : \mathcal{S} \subseteq\mathcal{P}(\omega) \text{ is a splitting family}\}. s = min { ∣ S ∣ : S ⊆ P ( ω ) is a splitting family } .
Definition 2.6. An interval partition of ω \omega ω is a sequence ( I n ) n ∈ ω (I_n)_{n\in\omega} ( I n ) n ∈ ω of the form I n = [ t n , t n + 1 ) I_n = [t_n,t_{n+1}) I n = [ t n , t n + 1 ) , where 0 = t 0 < t 1 < ⋯ 0 = t_0 < t_1 < \cdots 0 = t 0 < t 1 < ⋯ are natural numbers. A set B ⊆ ω B \subseteq\omega B ⊆ ω block-splits this partition if
∣ { n ∈ ω : I n ⊆ B } ∣ = ∣ { n ∈ ω : I n ∩ B = ∅ } ∣ = ω . |\{n \in\omega: I_n \subseteq B\}| = |\{n \in\omega: I_n \cap B = \varnothing\}| = \omega. ∣ { n ∈ ω : I n ⊆ B } ∣ = ∣ { n ∈ ω : I n ∩ B = ∅ } ∣ = ω . A block-splitting family is a set B ⊆ P ( ω ) \mathcal{B} \subseteq\mathcal{P}(\omega) B ⊆ P ( ω ) such that every interval partition is block-split by some B ∈ B B \in\mathcal{B} B ∈ B . The block-splitting number is
s b l o c k = min { ∣ B ∣ : B is a block-splitting family } . \mathfrak{s}_{\mathrm{block}} = \min\{|\mathcal{B}| : \mathcal{B} \text{ is a block-splitting family}\}. s block = min { ∣ B ∣ : B is a block-splitting family } .
The following result is due to Kamburelis and Węglorz.
Proposition 2.7 ([13 Proposition 2.1] ). s b l o c k = max { b , s } \mathfrak{s}_{\mathrm{block}}=\max\{\mathfrak{b},\mathfrak{s}\} s block = max { b , s } .
A collection C ⊆ [ ω ] ω \mathcal{C}\subseteq[\omega]^\omega C ⊆ [ ω ] ω is centered if every finite nonempty subcollection of C \mathcal{C} C has an infinite intersection. A pseudointersection of a centered collection C \mathcal{C} C is an infinite set P ⊆ ω P\subseteq\omega P ⊆ ω such that P ⊆ ∗ C P\subseteq^* C P ⊆ ∗ C for every C ∈ C C\in\mathcal{C} C ∈ C . The pseudointersection number p \mathfrak{p} p is the least size of a centered collection F ⊆ [ ω ] ω \mathcal{F}\subseteq[\omega]^\omega F ⊆ [ ω ] ω with no pseudointersection.
Bell’s theorem [1 ] states that p \mathfrak{p} p is the least cardinal κ \kappa κ such that Martin’s axiom for κ \kappa κ many dense subsets of σ \sigma σ -centered posets fails; that is, such that there exists a σ \sigma σ -centered poset P \mathbb{P} P and a collection D \mathcal{D} D of κ \kappa κ many dense subsets of P \mathbb{P} P with no filter G ⊆ P G\subseteq\mathbb{P} G ⊆ P meeting every D ∈ D D\in\mathcal{D} D ∈ D . For a proof in this language, see [14 Theorem III.3.61] . It is also well-known that p ≤ b \mathfrak{p}\leq\mathfrak{b} p ≤ b (see [2 ] ).
For disjoint subfamilies B , C \mathcal{B},\mathcal{C} B , C of an almost disjoint family, we say that B \mathcal{B} B is weakly separated from C \mathcal{C} C if some X ⊆ ω X\subseteq\omega X ⊆ ω satisfies
∣ b ∩ X ∣ = ω ( ∀ b ∈ B ) and ∣ c ∩ X ∣ < ω ( ∀ c ∈ C ) . \lvert b\cap X\rvert=\omega\quad(\forall b\in\mathcal{B})\qquad\text{and}\qquad\lvert c\cap X\rvert<\omega\quad(\forall c\in\mathcal{C}). ∣ b ∩ X ∣ = ω ( ∀ b ∈ B ) and ∣ c ∩ X ∣ < ω ( ∀ c ∈ C ) .
Following Brendle [3 p. 14] , let a p \mathfrak{ap} ap be the least cardinality of an almost disjoint family A \mathcal{A} A for which some B ⊆ A \mathcal{B}\subseteq\mathcal{A} B ⊆ A is not weakly separated from A ∖ B \mathcal{A}\setminus\mathcal{B} A ∖ B .
For t ∈ 2 < ω t\in2^{<\omega} t ∈ 2 < ω , the basic clopen set generated by t t t is denoted by [ t ] = { x ∈ 2 ω : t ⊆ x } [t]=\{x\in2^\omega:t\subseteq x\} [ t ] = { x ∈ 2 ω : t ⊆ x } .
We also use the following generalization of λ \lambda λ -sets to arbitrary topological spaces.
Definition 2.8. A topological space X X X is a λ \lambda λ -space if every countable subset of X X X is G δ G_\delta G δ in X X X . A separable metrizable λ \lambda λ -space is called a λ \lambda λ -set.
Notice that weak λ \lambda λ -spaces are T 1 T_1 T 1 . For distinct points x , y x,y x , y of a weak λ \lambda λ -space X X X , choose a G δ G_\delta G δ set H H H with x ∈ H x\in H x ∈ H and y ∉ H y\notin H y ∈ / H , and write H = ⋂ n ∈ ω U n H=\bigcap_{n\in\omega}U_n H = ⋂ n ∈ ω U n with each U n U_n U n open in X X X . Some U n U_n U n contains x x x and omits y y y .
We allow λ \lambda λ -sets and weak λ \lambda λ -sets to be countable. This convention differs from the one in [20 Definition 1.1] , which requires λ \lambda λ -sets to be uncountable. This does not affect the question of whether there exists a weak λ \lambda λ -set which is not a λ \lambda λ -set: since every countable set is trivially a λ \lambda λ -set, any counterexample must be uncountable.
We will use the following notation.
Definition 2.9. Let
non ( w λ ) = min { ∣ X ∣ : X is separable, metrizable, and not a weak λ -set } \operatorname{non}(w\lambda)=\min\{\lvert X\rvert:X\text{ is separable, metrizable, and not a weak }\lambda\text{-set}\} non ( w λ ) = min {∣ X ∣ : X is separable, metrizable, and not a weak λ -set } .
non ( λ ) = min { ∣ X ∣ : X is separable, metrizable, and not a λ -set } \operatorname{non}(\lambda)=\min\{\lvert X\rvert:X\text{ is separable, metrizable, and not a }\lambda\text{-set}\} non ( λ ) = min {∣ X ∣ : X is separable, metrizable, and not a λ -set } .
The following result is due to Rothberger.
Proposition 2.10 ([21 Lemma 3 and Theorem 1] ). non ( λ ) = b \operatorname{non}(\lambda)=\mathfrak{b} non ( λ ) = b .
For a proof of the preceding result, the reader may also refer to [25 Theorem 9.1] , where non ( λ ) \operatorname{non}(\lambda) non ( λ ) is denoted by b λ \mathfrak{b}_\lambda b λ .
We identify P ( ω ) \mathcal{P}(\omega) P ( ω ) with 2 ω 2^\omega 2 ω by characteristic functions. For x ∈ [ ω ] ω x\in[\omega]^\omega x ∈ [ ω ] ω , let e x ∈ ω ω e_x\in\omega^\omega e x ∈ ω ω be its increasing enumeration, so x = { e x ( n ) : n ∈ ω } x=\{e_x(n):n\in\omega\} x = { e x ( n ) : n ∈ ω } . The map x ↦ e x x\mapsto e_x x ↦ e x is continuous, since for every n , m ∈ ω n,m\in\omega n , m ∈ ω , the condition e x ( n ) = m e_x(n)=m e x ( n ) = m is determined by x ∩ ( m + 1 ) x\cap(m+1) x ∩ ( m + 1 ) .
The following criterion is known and is a variation of, e.g., [25 Lemma 9.3] and [17 Theorem 2.1] . As we did not find a reference for this exact formulation and the proof is short, we include it here for the convenience of the reader.
Lemma 2.11. Let [ ω ] < ω ⊆ X ⊆ 2 ω [\omega]^{<\omega} \subseteq X \subseteq2^\omega [ ω ] < ω ⊆ X ⊆ 2 ω . If { e x : x ∈ X ∖ [ ω ] < ω } \{e_x : x \in X \setminus[\omega]^{<\omega}\} { e x : x ∈ X ∖ [ ω ] < ω } is unbounded in ( ω ω , ≤ ∗ ) (\omega^\omega,\leq^*) ( ω ω , ≤ ∗ ) , then [ ω ] < ω [\omega]^{<\omega} [ ω ] < ω is not G δ G_\delta G δ in X X X . In particular, X X X is not a λ \lambda λ -set.
Proof. Suppose that [ ω ] < ω = X ∩ G [\omega]^{<\omega}=X\cap G [ ω ] < ω = X ∩ G for some G δ G_\delta G δ set G ⊆ 2 ω G\subseteq2^\omega G ⊆ 2 ω . Then C = 2 ω ∖ G C=2^\omega\setminus G C = 2 ω ∖ G is F σ F_\sigma F σ in the compact space 2 ω 2^\omega 2 ω , hence σ \sigma σ -compact, and X ∖ [ ω ] < ω ⊆ C ⊆ [ ω ] ω X\setminus[\omega]^{<\omega}\subseteq C\subseteq[\omega]^\omega X ∖ [ ω ] < ω ⊆ C ⊆ [ ω ] ω .
Write C = ⋃ j ∈ ω K j C=\bigcup_{j\in\omega}K_j C = ⋃ j ∈ ω K j , where each K j K_j K j is compact. For every j , n ∈ ω j,n\in\omega j , n ∈ ω , the set { e x ( n ) : x ∈ K j } \{e_x(n):x\in K_j\} { e x ( n ) : x ∈ K j } is finite, as the continuous image of a compact space in the discrete space ω \omega ω . Define g j , g ∈ ω ω g_j,g\in\omega^\omega g j , g ∈ ω ω by:
g j ( n ) = max ( { e x ( n ) : x ∈ K j } ∪ { 0 } ) and g ( n ) = max j ≤ n g j ( n ) . g_j(n)=\max\left(\{e_x(n):x\in K_j\}\cup\{0\}\right)
\qquad\text{and}\qquad
g(n)=\max_{j\leq n}g_j(n). g j ( n ) = max ( { e x ( n ) : x ∈ K j } ∪ { 0 } ) and g ( n ) = j ≤ n max g j ( n ) .
If x ∈ K j x\in K_j x ∈ K j , then e x ( n ) ≤ g j ( n ) ≤ g ( n ) e_x(n)\leq g_j(n)\leq g(n) e x ( n ) ≤ g j ( n ) ≤ g ( n ) for every n ≥ j n\geq j n ≥ j , so e x ≤ ∗ g e_x\leq^*g e x ≤ ∗ g . Thus { e x : x ∈ X ∖ [ ω ] < ω } \{e_x:x\in X\setminus[\omega]^{<\omega}\} { e x : x ∈ X ∖ [ ω ] < ω } is bounded, contradicting the hypothesis. Since [ ω ] < ω [\omega]^{<\omega} [ ω ] < ω is countable, X X X is not a λ \lambda λ -set.
The least cardinality of a non-weak λ \lambda λ -set
The question of whether every weak λ \lambda λ -set is a λ \lambda λ -set was raised in [20 ] . In this section we prove that non ( w λ ) = max { b , s } \operatorname{non}(\mathrm{w}\lambda)=\max\{\mathfrak{b},\mathfrak{s}\} non ( w λ ) = max { b , s } . Together with non ( λ ) = b \operatorname{non}(\lambda)=\mathfrak{b} non ( λ ) = b , this gives a weak λ \lambda λ -set which is not a λ \lambda λ -set whenever b < s \mathfrak{b}<\mathfrak{s} b < s .
We first prove the lower bound using the bounding and splitting numbers.
The following lemma is a well-known characterization of the splitting number s \mathfrak{s} s . For a proof, see also [2 Theorem 3.2] .
Lemma 3.1. The cardinal s \mathfrak{s} s is the minimum cardinality of a family S ⊆ 2 ω S\subseteq2^\omega S ⊆ 2 ω such that there is no infinite A ⊆ ω A\subseteq\omega A ⊆ ω for which r ↾ A r\mathbin{\upharpoonright}A r ↾ A converges for every r ∈ S r\in S r ∈ S .
We start proving our characterization.
Lemma 3.2. Every zero-dimensional T 1 T_1 T 1 space X X X of cardinality less than s \mathfrak{s} s is a weak λ \lambda λ -space.
Proof. Let Y , Z ⊆ X Y,Z\subseteq X Y , Z ⊆ X be disjoint and countable. If Y ∪ Z Y\cup Z Y ∪ Z is finite, then Y Y Y is separated from Z Z Z by a clopen set.
So assume that Y ∪ Z Y\cup Z Y ∪ Z is infinite and enumerate it without repetitions as { d n : n ∈ ω } \{d_n:n\in\omega\} { d n : n ∈ ω } . Since X X X is zero-dimensional and T 1 T_1 T 1 , for each n ∈ ω n\in\omega n ∈ ω choose ( U n , i : i ≤ n ) (U_{n,i}:i\leq n) ( U n , i : i ≤ n ) such that:
U n , i U_{n,i} U n , i is a clopen neighborhood of d i d_i d i for every i ≤ n i\leq n i ≤ n ;
For every two distinct i , j ≤ n i,j\leq n i , j ≤ n , the sets U n , i U_{n,i} U n , i and U n , j U_{n,j} U n , j are disjoint.
This is possible as T 1 T_1 T 1 zero-dimensional spaces are Hausdorff.
For each n ∈ ω n\in\omega n ∈ ω , define C n = ⋃ { U n , i : i ≤ n and d i ∈ Y } C_n=\bigcup\{U_{n,i}:i\leq n\text{ and }d_i\in Y\} C n = ⋃ { U n , i : i ≤ n and d i ∈ Y } . Each C n C_n C n is clopen. For each x ∈ X x\in X x ∈ X , let f x : ω → 2 f_x:\omega\to2 f x : ω → 2 be defined by
f x ( n ) = χ C n ( x ) = { 1 , if x ∈ C n , 0 , if x ∉ C n . f_x(n)=\chi_{C_n}(x)=
\begin{cases}
1, & \text{if }x\in C_n,\\
0, & \text{if }x\notin C_n.
\end{cases} f x ( n ) = χ C n ( x ) = { 1 , 0 , if x ∈ C n , if x ∈ / C n .
In particular, whenever i ≤ n < ω i\leq n<\omega i ≤ n < ω :
f d i ( n ) = { 1 , if d i ∈ Y , 0 , if d i ∈ Z . (1) f_{d_i}(n)=
\begin{cases}
1, & \text{if }d_i\in Y,\\
0, & \text{if }d_i\in Z.
\end{cases} \tag*{(1)} f d i ( n ) = { 1 , 0 , if d i ∈ Y , if d i ∈ Z . (1)
Apply Lemma 3.1 to S = { f x : x ∈ X } S=\{f_x:x\in X\} S = { f x : x ∈ X } , which is possible as ∣ X ∣ < s |X|<\mathfrak{s} ∣ X ∣ < s , to obtain A ⊆ ω A\subseteq\omega A ⊆ ω infinite such that f x ↾ A f_x\mathbin{\upharpoonright}A f x ↾ A converges for every x ∈ X x\in X x ∈ X . Thus, f x ↾ A f_x\mathbin{\upharpoonright}A f x ↾ A is eventually constant for every x ∈ X x \in X x ∈ X . Now let H H H be the set of all x ∈ X x \in X x ∈ X such that f x ↾ A f_x\mathbin{\upharpoonright}A f x ↾ A is eventually equal to 1 1 1 , that is,
H = ⋃ m ∈ ω ⋂ n ∈ A ∖ m { x ∈ X : f x ( n ) = 1 } = ⋃ m ∈ ω ⋂ n ∈ A ∖ m C n . H=\bigcup_{m\in\omega}\bigcap_{n\in A\setminus m}\{x\in X:f_x(n)=1\}=\bigcup_{m\in\omega}\bigcap_{n\in A\setminus m}C_n. H = m ∈ ω ⋃ n ∈ A ∖ m ⋂ { x ∈ X : f x ( n ) = 1 } = m ∈ ω ⋃ n ∈ A ∖ m ⋂ C n .
The set H H H is F σ F_\sigma F σ in X X X . Its complement is
X ∖ H = ⋃ m ∈ ω ⋂ n ∈ A ∖ m { x ∈ X : f x ( n ) = 0 } = ⋃ m ∈ ω ⋂ n ∈ A ∖ m X ∖ C n . X\setminus H=\bigcup_{m\in\omega}\bigcap_{n\in A\setminus m}\{x\in X:f_x(n)=0\}=\bigcup_{m\in\omega}\bigcap_{n\in A\setminus m}X\setminus C_n. X ∖ H = m ∈ ω ⋃ n ∈ A ∖ m ⋂ { x ∈ X : f x ( n ) = 0 } = m ∈ ω ⋃ n ∈ A ∖ m ⋂ X ∖ C n .
This is also an F σ F_\sigma F σ set in X X X . Hence H H H is also G δ G_\delta G δ in X X X . (1) shows that Y ⊆ H Y\subseteq H Y ⊆ H and Z ⊆ X ∖ H Z\subseteq X\setminus H Z ⊆ X ∖ H .
Every metric space of cardinality < c <\mathfrak c < c is zero-dimensional. Indeed, if ( X , d ) (X,d) ( X , d ) is a metric space with ∣ X ∣ < c |X|<\mathfrak c ∣ X ∣ < c , then for each x ∈ X x\in X x ∈ X , R x = { d ( x , y ) : y ∈ X } R_x=\{d(x,y):y\in X\} R x = { d ( x , y ) : y ∈ X } has cardinality < c <\mathfrak c < c , so { B ( x , r ) : r ∈ ( 0 , ∞ ) ∖ R x } \{B(x,r):r\in(0,\infty)\setminus R_x\} { B ( x , r ) : r ∈ ( 0 , ∞ ) ∖ R x } is a local base of clopen neighborhoods of x x x . Thus, we have the following corollary.
Corollary 3.3. Every metrizable space of cardinality less than s \mathfrak s s is a weak λ \lambda λ -space.
Moreover, we have proved the desired lower bound for non ( w λ ) \operatorname{non}(\mathrm{w}\lambda) non ( w λ ) .
Corollary 3.4. max { b , s } ≤ non ( w λ ) \max\{\mathfrak b,\mathfrak s\}\leq\operatorname{non}(\mathrm{w}\lambda) max { b , s } ≤ non ( w λ ) .
Proof. As every λ \lambda λ -set is a weak λ \lambda λ -set, by Proposition 2.10 we have b = non ( λ ) ≤ non ( w λ ) \mathfrak b=\operatorname{non}(\lambda)\leq\operatorname{non}(\mathrm{w}\lambda) b = non ( λ ) ≤ non ( w λ ) . Combining this with Corollary 3.3 gives the desired inequality.
We now construct a non-weak λ \lambda λ -set of size max { b , s } \max\{\mathfrak b,\mathfrak s\} max { b , s } using Proposition 2.7.
Proof of Theorem A. By Corollary 3.4, it suffices to prove non ( w λ ) ≤ max { b , s } \operatorname{non}(\mathrm{w}\lambda)\leq\max\{\mathfrak b,\mathfrak s\} non ( w λ ) ≤ max { b , s } . For ε ∈ { 0 , 1 } \varepsilon\in\{0,1\} ε ∈ { 0 , 1 } , let:
Q ε = { x ∈ 2 ω : x ( n ) = ε for all sufficiently large n } = ⋃ N ∈ ω ⋂ n ≥ N { x ∈ 2 ω : x ( n ) = ε } . Q_\varepsilon=\{x\in2^\omega:x(n)=\varepsilon\text{ for all sufficiently large }n\}=\bigcup_{N\in\omega}\bigcap_{n\geq N}\{x\in2^\omega:x(n)=\varepsilon\}. Q ε = { x ∈ 2 ω : x ( n ) = ε for all sufficiently large n } = N ∈ ω ⋃ n ≥ N ⋂ { x ∈ 2 ω : x ( n ) = ε } .
The sets Q 0 Q_0 Q 0 and Q 1 Q_1 Q 1 are disjoint and countable. By Proposition 2.7, choose a block-splitting family B \mathcal B B of size max { b , s } \max\{\mathfrak b,\mathfrak s\} max { b , s } , and put X = Q 0 ∪ Q 1 ∪ { χ B : B ∈ B } ⊆ 2 ω X=Q_0\cup Q_1\cup\{\chi_B:B\in\mathcal B\}\subseteq2^\omega X = Q 0 ∪ Q 1 ∪ { χ B : B ∈ B } ⊆ 2 ω . Clearly ∣ X ∣ = max { b , s } |X|=\max\{\mathfrak b,\mathfrak s\} ∣ X ∣ = max { b , s } . We show that Q 0 , Q 1 Q_0,Q_1 Q 0 , Q 1 admit no relative G δ G_\delta G δ –F σ F_\sigma F σ separator in X X X .
Suppose, towards a contradiction, that a relative G δ G_\delta G δ –F σ F_\sigma F σ set H ⊆ X H\subseteq X H ⊆ X satisfies Q 0 ⊆ H Q_0\subseteq H Q 0 ⊆ H and Q 1 ⊆ X ∖ H Q_1\subseteq X\setminus H Q 1 ⊆ X ∖ H . Since both H H H and X ∖ H X\setminus H X ∖ H are F σ F_\sigma F σ in X X X , write:
H = X ∩ ⋃ i ∈ ω F i , X ∖ H = X ∩ ⋃ i ∈ ω E i , H=X\cap\bigcup_{i\in\omega}F_i,\qquad X\setminus H=X\cap\bigcup_{i\in\omega}E_i, H = X ∩ i ∈ ω ⋃ F i , X ∖ H = X ∩ i ∈ ω ⋃ E i ,
where F i , E i F_i,E_i F i , E i are closed in 2 ω 2^\omega 2 ω . Since Q 0 ⊆ H Q_0\subseteq H Q 0 ⊆ H and Q 1 ⊆ X ∖ H Q_1\subseteq X\setminus H Q 1 ⊆ X ∖ H , we have
F i ∩ Q 1 = ∅ , E i ∩ Q 0 = ∅ ( i ∈ ω ) . F_i\cap Q_1=\varnothing,\qquad E_i\cap Q_0=\varnothing\quad(i\in\omega). F i ∩ Q 1 = ∅ , E i ∩ Q 0 = ∅ ( i ∈ ω ) .
Recursively choose 0 = t 0 < t 1 < ⋯ 0=t_0<t_1<\cdots 0 = t 0 < t 1 < ⋯ so that, putting I n = [ t n , t n + 1 ) I_n=[t_n,t_{n+1}) I n = [ t n , t n + 1 ) , for every x ∈ 2 ω x\in2^\omega x ∈ 2 ω ,
x ↾ I n ≡ 1 ⟹ x ∉ ⋃ i ≤ n F i , x ↾ I n ≡ 0 ⟹ x ∉ ⋃ i ≤ n E i . (2) \begin{aligned}
x\mathbin{\upharpoonright}I_n\equiv1&\Longrightarrow x\notin\bigcup_{i\leq n}F_i,\\
x\mathbin{\upharpoonright}I_n\equiv0&\Longrightarrow x\notin\bigcup_{i\leq n}E_i. \tag*{(2)}
\end{aligned} x ↾ I n ≡ 1 x ↾ I n ≡ 0 ⟹ x ∈ / i ≤ n ⋃ F i , ⟹ x ∈ / i ≤ n ⋃ E i . (2)
To see that this is possible, assume we have defined t n t_n t n . We show how to define t n + 1 t_{n+1} t n + 1 . For each u ∈ 2 t n u \in2^{t_n} u ∈ 2 t n , let q u 0 = u ⌢ 0 q_u^0=u^\frown0 q u 0 = u ⌢ 0 and q u 1 = u ⌢ 1 q_u^1=u^\frown1 q u 1 = u ⌢ 1 , where 0 \mathbf{0} 0 and 1 \mathbf{1} 1 are the constant (infinite) sequences with values 0 and 1, respectively. Since q u 1 ∈ Q 1 q_u^1 \in Q_1 q u 1 ∈ Q 1 , we have q u 1 ∉ ⋃ i ≤ n F i q_u^1 \notin\bigcup_{i\le n}F_i q u 1 ∈ / ⋃ i ≤ n F i . Similarly, q u 0 ∉ ⋃ i ≤ n E i q_u^0 \notin\bigcup_{i\le n}E_i q u 0 ∈ / ⋃ i ≤ n E i . As these unions are closed, choose integers k u 0 , k u 1 > t n k_u^0,k_u^1>t_n k u 0 , k u 1 > t n such that
[ q u 1 ↾ k u 1 ] ∩ ⋃ i ≤ n F i = ∅ , [ q u 0 ↾ k u 0 ] ∩ ⋃ i ≤ n E i = ∅ . [q_u^1\mathbin{\upharpoonright} k_u^1]\cap\bigcup_{i\le n}F_i=\varnothing,\qquad[q_u^0\mathbin{\upharpoonright} k_u^0]\cap\bigcup_{i\le n}E_i=\varnothing. [ q u 1 ↾ k u 1 ] ∩ i ≤ n ⋃ F i = ∅ , [ q u 0 ↾ k u 0 ] ∩ i ≤ n ⋃ E i = ∅ .
There are only finitely many u ∈ 2 t n u\in2^{t_n} u ∈ 2 t n , so we may define
t n + 1 = max { k u ε : u ∈ 2 t n , ε ∈ { 0 , 1 } } . t_{n+1}=\max\{k_u^\varepsilon:u\in2^{t_n},\ \varepsilon\in\{0,1\}\}. t n + 1 = max { k u ε : u ∈ 2 t n , ε ∈ { 0 , 1 }} .
If x ↾ I n = 1 x\mathbin{\upharpoonright} I_n=1 x ↾ I n = 1 , let u = x ↾ t n u=x\mathbin{\upharpoonright} t_n u = x ↾ t n . Then x ↾ t n + 1 ⊆ q u 1 x\mathbin{\upharpoonright} t_{n+1}\subseteq q_u^1 x ↾ t n + 1 ⊆ q u 1 . Since k u 1 ≤ t n + 1 k_u^1\le t_{n+1} k u 1 ≤ t n + 1 , we have x ∈ [ q u 1 ↾ k u 1 ] x\in[q_u^1\mathbin{\upharpoonright} k_u^1] x ∈ [ q u 1 ↾ k u 1 ] , hence x ∉ ⋃ i ≤ n F i x\notin\bigcup_{i\le n}F_i x ∈ / ⋃ i ≤ n F i . Similarly, if x ↾ I n = 0 x\mathbin{\upharpoonright} I_n=0 x ↾ I n = 0 , then x ∉ ⋃ i ≤ n E i x\notin\bigcup_{i\le n}E_i x ∈ / ⋃ i ≤ n E i , proving (2).
Choose B ∈ B B\in\mathcal B B ∈ B that block-splits ( I n ) n ∈ ω (I_n)_{n\in\omega} ( I n ) n ∈ ω , and put x = χ B x=\chi_B x = χ B . Then x ↾ I n = 0 x\mathbin{\upharpoonright} I_n=0 x ↾ I n = 0 for infinitely many n n n and x ↾ I n = 1 x\mathbin{\upharpoonright} I_n=1 x ↾ I n = 1 for infinitely many n n n . Fix i ∈ ω i\in\omega i ∈ ω . Since the set of indices n n n for which x ↾ I n = 1 x\mathbin{\upharpoonright} I_n=1 x ↾ I n = 1 is infinite, choose such an n ≥ i n\ge i n ≥ i . By (2), x ∉ ⋃ j ≤ n F j x\notin\bigcup_{j\le n}F_j x ∈ / ⋃ j ≤ n F j , and in particular x ∉ F i x\notin F_i x ∈ / F i . Likewise, x ∉ E i x\notin E_i x ∈ / E i . As i i i was arbitrary, x ∉ ⋃ i ∈ ω ( F i ∪ E i ) ⊇ X x\notin\bigcup_{i\in\omega}(F_i\cup E_i)\supseteq X x ∈ / ⋃ i ∈ ω ( F i ∪ E i ) ⊇ X , a contradiction.
Corollary 3.5. If b < s \mathfrak b<\mathfrak s b < s , then there is a weak λ \lambda λ -set which is not a λ \lambda λ -set.
Proof. By Proposition 2.10, choose a separable metrizable space X X X of cardinality b \mathfrak b b which is not a λ \lambda λ -set. Since ∣ X ∣ = b < s |X|=\mathfrak b<\mathfrak s ∣ X ∣ = b < s , Corollary 3.3 shows that X X X is a weak λ \lambda λ -set.
A weak λ \lambda λ -set which is not a λ \lambda λ -set under p = c \mathfrak p=\mathfrak c p = c
In this section, we show that p = c \mathfrak p=\mathfrak c p = c implies the existence of weak λ \lambda λ -sets that are not λ \lambda λ -sets. This complements Corollary 3.5, which establishes the same conclusion under b < s \mathfrak b<\mathfrak s b < s .
This result was recently obtained by Osipov and Lipin [18 Theorem 1.1] , whose proof uses both Hindman’s finite-unions theorem and Bell’s theorem. Our proof, obtained independently, also uses Hindman’s theorem but does not require Bell’s theorem.
In our argument, the finite-unions theorem yields the compact covers needed for the separators. We first construct a decreasing sequence of block sequences and then choose points along this sequence.
Throughout this section, we identify P ( ω ) \mathcal P(\omega) P ( ω ) with 2 ω 2^\omega 2 ω via characteristic functions and equip [ ω ] < ω [\omega]^{<\omega} [ ω ] < ω and [ ω ] ω [\omega]^\omega [ ω ] ω with the corresponding subspace topologies.
We prove Theorem D by constructing a witness of cardinality c \mathfrak c c in 2 ω 2^\omega 2 ω .
Definition 4.1. Let I ∈ [ ω ] ω I\in[\omega]^\omega I ∈ [ ω ] ω . A block sequence indexed by I I I is a sequence A = ( a i ) i ∈ I A=(a_i)_{i\in I} A = ( a i ) i ∈ I of nonempty finite subsets of ω \omega ω such that max a i < min a j \max a_i<\min a_j max a i < min a j whenever i , j ∈ I i,j\in I i , j ∈ I and i < j i<j i < j . We define:
(i) the set of all finite unions of blocks of A A A , FU ( A ) = { ⋃ i ∈ F a i : F ∈ [ I ] < ω ∖ { ∅ } } \operatorname{FU}(A)=\{\bigcup_{i\in F}a_i:F\in[I]^{<\omega}\setminus\{\varnothing\}\} FU ( A ) = { ⋃ i ∈ F a i : F ∈ [ I ] < ω ∖ { ∅ }} ;
(ii) the set of all infinite unions of blocks of A A A , [ A ] ∞ = { ⋃ i ∈ J a i : J ∈ [ I ] ω } [A]_\infty=\{\bigcup_{i\in J}a_i:J\in[I]^\omega\} [ A ] ∞ = { ⋃ i ∈ J a i : J ∈ [ I ] ω } ;
(iii) the saturation of A A A , S ( A ) = { r ∪ x : r ∈ [ ω ] < ω , x ∈ [ A ] ∞ } S(A)=\{r\cup x:r\in[\omega]^{<\omega},\ x\in[A]_\infty\} S ( A ) = { r ∪ x : r ∈ [ ω ] < ω , x ∈ [ A ] ∞ } .
For J ∈ [ I ] ω J\in[I]^\omega J ∈ [ I ] ω , we write A ↾ J = ( a j ) j ∈ J A\mathbin{\upharpoonright}J=(a_j)_{j\in J} A ↾ J = ( a j ) j ∈ J . For m ∈ ω m\in\omega m ∈ ω , we abbreviate A ≥ m = A ↾ ( I ∖ m ) A_{\ge m}=A\mathbin{\upharpoonright}(I\setminus m) A ≥ m = A ↾ ( I ∖ m ) .
Definition 4.2. Let A A A and B = ( b j ) j ∈ J B=(b_j)_{j\in J} B = ( b j ) j ∈ J be block sequences. We say that B B B is a condensation of A A A , and write B ≼ A B\preccurlyeq A B ≼ A , if b j ∈ FU ( A ) b_j\in\operatorname{FU}(A) b j ∈ FU ( A ) for every j ∈ J j\in J j ∈ J .
Write B ≼ ∗ A B\preccurlyeq^* A B ≼ ∗ A if B ≥ m ≼ A B_{\ge m}\preccurlyeq A B ≥ m ≼ A for some m ∈ ω m\in\omega m ∈ ω . Equivalently, b j ∈ FU ( A ) b_j\in\operatorname{FU}(A) b j ∈ FU ( A ) for all but finitely many j ∈ J j\in J j ∈ J .
Both ⪯ \preceq ⪯ and ⪯ ∗ \preceq^{*} ⪯ ∗ are easily seen to be transitive.
Figure 2. A condensation B ⪯ A B \preceq A B ⪯ A .
Discarding finitely many initial blocks does not change the saturation:
S ( A ≥ m ) = S ( A ) ( ∀ m ∈ ω ) . S(A_{\geq m}) = S(A) \qquad(\forall m \in\omega). S ( A ≥ m ) = S ( A ) ( ∀ m ∈ ω ) .
Indeed, those blocks can be absorbed into the finite set r r r in the definition of S ( A ) S(A) S ( A ) .
Lemma 4.3. Let A A A and B B B be block sequences.
(i) If B ⪯ A B \preceq A B ⪯ A , then FU ( B ) ⊆ FU ( A ) \operatorname{FU}(B) \subseteq\operatorname{FU}(A) FU ( B ) ⊆ FU ( A ) .
(ii) If B ⪯ A B \preceq A B ⪯ A , then [ B ] ∞ ⊆ [ A ] ∞ [B]_{\infty} \subseteq[A]_{\infty} [ B ] ∞ ⊆ [ A ] ∞ .
(iii) If B ⪯ ∗ A B \preceq^{*} A B ⪯ ∗ A , then S ( B ) ⊆ S ( A ) S(B) \subseteq S(A) S ( B ) ⊆ S ( A ) .
Proof. Write A = ( a i ) i ∈ I A = (a_i)_{i\in I} A = ( a i ) i ∈ I and B = ( b j ) j ∈ J B = (b_j)_{j\in J} B = ( b j ) j ∈ J .
(i) The set FU ( A ) \operatorname{FU}(A) FU ( A ) is closed under nonempty finite unions. As each b j b_j b j is in FU ( A ) \operatorname{FU}(A) FU ( A ) , every nonempty finite union of blocks of B B B is in FU ( A ) \operatorname{FU}(A) FU ( A ) , therefore FU ( B ) ⊆ FU ( A ) \operatorname{FU}(B) \subseteq\operatorname{FU}(A) FU ( B ) ⊆ FU ( A ) .
(ii) Let x ∈ [ B ] ∞ x \in[B]_{\infty} x ∈ [ B ] ∞ . Write x = ⋃ j ∈ K b j x = \bigcup_{j\in K} b_j x = ⋃ j ∈ K b j for some K ∈ [ J ] ω K \in[J]^{\omega} K ∈ [ J ] ω . For each j ∈ K j \in K j ∈ K , choose a nonempty finite F j ⊆ I F_j \subseteq I F j ⊆ I such that b j = ⋃ i ∈ F j a i b_j = \bigcup_{i\in F_j} a_i b j = ⋃ i ∈ F j a i . Let H = ⋃ j ∈ K F j H = \bigcup_{j\in K} F_j H = ⋃ j ∈ K F j . The sets F j F_j F j are pairwise disjoint, so H H H is infinite. Then:
x = ⋃ j ∈ K b j = ⋃ j ∈ K ⋃ i ∈ F j a i = ⋃ i ∈ H a i ∈ [ A ] ∞ . x = \bigcup_{j\in K} b_j = \bigcup_{j\in K}\bigcup_{i\in F_j} a_i = \bigcup_{i\in H} a_i \in[A]_{\infty}. x = j ∈ K ⋃ b j = j ∈ K ⋃ i ∈ F j ⋃ a i = i ∈ H ⋃ a i ∈ [ A ] ∞ .
For (iii), choose m ∈ ω m \in\omega m ∈ ω with B ≥ m ⪯ A B_{\geq m} \preceq A B ≥ m ⪯ A . By (ii), [ B ≥ m ] ∞ ⊆ [ A ] ∞ [B_{\geq m}]_{\infty} \subseteq[A]_{\infty} [ B ≥ m ] ∞ ⊆ [ A ] ∞ , so S ( B ≥ m ) ⊆ S ( A ) S(B_{\geq m}) \subseteq S(A) S ( B ≥ m ) ⊆ S ( A ) . Thus it suffices to see that S ( B ) ⊆ S ( B ≥ m ) S(B) \subseteq S(B_{\geq m}) S ( B ) ⊆ S ( B ≥ m ) . Let x ∈ S ( B ) x \in S(B) x ∈ S ( B ) . Fix r ∈ [ ω ] < ω r \in[\omega]^{<\omega} r ∈ [ ω ] < ω and K ∈ [ J ] ω K \in[J]^{\omega} K ∈ [ J ] ω with x = r ∪ ⋃ j ∈ K b j x = r \cup\bigcup_{j\in K} b_j x = r ∪ ⋃ j ∈ K b j . Then:
x = r ∪ ⋃ j ∈ K b j = ( r ∪ ⋃ j ∈ K ∩ m b j ) ∪ ⋃ j ∈ K ∖ m b j ∈ S ( B ≥ m ) . x = r \cup\bigcup_{j\in K} b_j = \left(r \cup\bigcup_{j\in K\cap m} b_j\right) \cup\bigcup_{j\in K\setminus m} b_j \in S(B_{\geq m}). x = r ∪ j ∈ K ⋃ b j = r ∪ j ∈ K ∩ m ⋃ b j ∪ j ∈ K ∖ m ⋃ b j ∈ S ( B ≥ m ) .
Lemma 4.4. Let δ < p \delta< \mathfrak{p} δ < p be an ordinal and let ( A ξ ) ξ < δ (A_{\xi})_{\xi<\delta} ( A ξ ) ξ < δ be a ⪯ ∗ \preceq^{*} ⪯ ∗ -decreasing family of block sequences. Then there exists a block sequence B B B such that B ⪯ ∗ A ξ B \preceq^{*} A_{\xi} B ⪯ ∗ A ξ for every ξ < δ \xi< \delta ξ < δ .
Proof. For m ∈ ω m \in\omega m ∈ ω , put T m = { s ∈ [ ω ] < ω ∖ { ∅ } : min s ≥ m } T_m = \{s \in[\omega]^{<\omega} \setminus\{\varnothing\} : \min s \geq m\} T m = { s ∈ [ ω ] < ω ∖ { ∅ } : min s ≥ m } . Let
F = { FU ( A ξ ) : ξ < δ } ∪ { T m : m ∈ ω } . \mathcal{F} = \{\operatorname{FU}(A_{\xi}) : \xi< \delta\} \cup\{T_m : m \in\omega\}. F = { FU ( A ξ ) : ξ < δ } ∪ { T m : m ∈ ω } .
Write N = [ ω ] < ω ∖ { ∅ } N = [\omega]^{<\omega} \setminus\{\varnothing\} N = [ ω ] < ω ∖ { ∅ } . Then F \mathcal{F} F is a collection of subsets of the infinite countable set N N N .
Claim 1. F \mathcal{F} F is centered. Proof of the claim. The sets T m T_m T m are decreasing for m ∈ ω m \in\omega m ∈ ω and clearly infinite. Thus it suffices to show that given a nonempty finite H ⊆ δ H \subseteq\delta H ⊆ δ and m ∈ ω m \in\omega m ∈ ω ,
T m ∩ ⋂ ξ ∈ H FU ( A ξ ) T_m \cap\bigcap_{\xi\in H} \operatorname{FU}(A_\xi) T m ∩ ξ ∈ H ⋂ FU ( A ξ )
is infinite.
Let η = max H \eta= \max H η = max H . For every ξ ∈ H \xi\in H ξ ∈ H , choose m ξ m_\xi m ξ with ( A η ) ≥ m ξ ⪯ A ξ (A_\eta)_{\geq m_\xi} \preceq A_\xi ( A η ) ≥ m ξ ⪯ A ξ . Choose M ≥ max { m ξ : ξ ∈ H } M \geq\max\{m_\xi: \xi\in H\} M ≥ max { m ξ : ξ ∈ H } such that every block of ( A η ) ≥ M (A_\eta)_{\geq M} ( A η ) ≥ M has minimum at least m m m . Then ( A η ) ≥ M ⪯ A ξ (A_\eta)_{\geq M} \preceq A_\xi ( A η ) ≥ M ⪯ A ξ for every ξ ∈ H \xi\in H ξ ∈ H .
Thus, FU ( ( A η ) ≥ M ) ⊆ FU ( A ξ ) \operatorname{FU}((A_\eta)_{\geq M}) \subseteq\operatorname{FU}(A_\xi) FU (( A η ) ≥ M ) ⊆ FU ( A ξ ) for every ξ ∈ H \xi\in H ξ ∈ H . Also, given x ∈ FU ( ( A η ) ≥ M ) x \in\operatorname{FU}((A_\eta)_{\geq M}) x ∈ FU (( A η ) ≥ M ) , we have min x ≥ m \min x \geq m min x ≥ m , so x ∈ T m x \in T_m x ∈ T m . As FU ( ( A η ) ≥ M ) \operatorname{FU}((A_\eta)_{\geq M}) FU (( A η ) ≥ M ) is infinite, this proves the claim.
Since ∣ F ∣ ≤ ∣ δ ∣ + ℵ 0 < p |\mathcal{F}| \leq|\delta| + \aleph_0 < \mathfrak{p} ∣ F ∣ ≤ ∣ δ ∣ + ℵ 0 < p , choose an infinite pseudointersection P P P of F \mathcal{F} F . Since P ⊆ ∗ T m P \subseteq^* T_m P ⊆ ∗ T m for every m m m , we can recursively choose b n ∈ P b_n \in P b n ∈ P with max b n < min b n + 1 \max b_n < \min b_{n+1} max b n < min b n + 1 . Put B = ( b n ) n ∈ ω B = (b_n)_{n \in\omega} B = ( b n ) n ∈ ω . For each ξ < δ \xi< \delta ξ < δ , the inclusion P ⊆ ∗ FU ( A ξ ) P \subseteq^* \operatorname{FU}(A_\xi) P ⊆ ∗ FU ( A ξ ) implies that all but finitely many b n b_n b n belong to FU ( A ξ ) \operatorname{FU}(A_\xi) FU ( A ξ ) . Thus B ⪯ ∗ A ξ B \preceq^* A_\xi B ⪯ ∗ A ξ .
We use Hindman’s finite-unions theorem, which we state below.
Proposition 4.5 [10 Corollary 3.3] . For every block sequence A A A indexed by a set I ∈ [ ω ] ω I \in[\omega]^\omega I ∈ [ ω ] ω and every finite coloring d : FU ( A ) → L d : \operatorname{FU}(A) \to L d : FU ( A ) → L , where L L L is a finite set, there exists a block sequence B ⪯ A B \preceq A B ⪯ A indexed by I I I such that d d d is constant on FU ( B ) \operatorname{FU}(B) FU ( B ) .
For the next lemma, we need an almost disjoint family of cardinality c \mathfrak{c} c . An almost disjoint family on an infinite countable set N N N is a collection A ⊆ [ N ] ω \mathcal{A} \subseteq[N]^\omega A ⊆ [ N ] ω such that ∣ a ∩ b ∣ < ω |a \cap b| < \omega ∣ a ∩ b ∣ < ω for every distinct a , b ∈ A a,b \in\mathcal{A} a , b ∈ A . It is well-known that there are almost disjoint families of cardinality c \mathfrak{c} c . For instance, for each z ∈ 2 ω z \in2^\omega z ∈ 2 ω , let a z = { z ↾ n : n ∈ ω } a_z = \{z \mathbin{\upharpoonright} n : n \in\omega\} a z = { z ↾ n : n ∈ ω } . Then { a z : z ∈ 2 ω } \{a_z : z \in2^\omega\} { a z : z ∈ 2 ω } is an almost disjoint family of cardinality c \mathfrak{c} c on 2 < ω 2^{<\omega} 2 < ω . A bijection of 2 < ω 2^{<\omega} 2 < ω with N N N gives an almost disjoint family of cardinality c \mathfrak{c} c on N N N .
Lemma 4.6. Let A A A be a block sequence and let E ⊆ [ ω ] ω E \subseteq[\omega]^\omega E ⊆ [ ω ] ω have cardinality less than c \mathfrak{c} c . There exists a subsequence B B B of A A A such that S ( B ) ∩ E = ∅ S(B) \cap E = \varnothing S ( B ) ∩ E = ∅ .
Proof. Write A = ( a i ) i ∈ I A = (a_i)_{i \in I} A = ( a i ) i ∈ I , where I ∈ [ ω ] ω I \in[\omega]^\omega I ∈ [ ω ] ω . Let A ⊆ [ I ] ω \mathcal{A} \subseteq[I]^\omega A ⊆ [ I ] ω be an almost disjoint family of cardinality c \mathfrak{c} c . For distinct a , b ∈ A a,b \in\mathcal{A} a , b ∈ A , the unions ⋃ i ∈ a a i \bigcup_{i \in a} a_i ⋃ i ∈ a a i and ⋃ i ∈ b a i \bigcup_{i \in b} a_i ⋃ i ∈ b a i have finite intersection, so S ( A ↾ a ) ∩ S ( A ↾ b ) = ∅ S(A \mathbin{\upharpoonright} a) \cap S(A \mathbin{\upharpoonright} b) = \varnothing S ( A ↾ a ) ∩ S ( A ↾ b ) = ∅ . Consequently, ∣ { a ∈ A : S ( A ↾ a ) ∩ E ≠ ∅ } ∣ ≤ ∣ E ∣ < c |\{a \in\mathcal{A} : S(A \mathbin{\upharpoonright} a) \cap E \neq\varnothing\}| \leq|E| < \mathfrak{c} ∣ { a ∈ A : S ( A ↾ a ) ∩ E = ∅ } ∣ ≤ ∣ E ∣ < c . Choose a ∈ A a \in\mathcal{A} a ∈ A with S ( A ↾ a ) ∩ E = ∅ S(A \mathbin{\upharpoonright} a) \cap E = \varnothing S ( A ↾ a ) ∩ E = ∅ and put B = A ↾ a B = A \mathbin{\upharpoonright} a B = A ↾ a .
The next lemma says that whenever [ ω ] < ω [\omega]^{<\omega} [ ω ] < ω is colored with two colors, every block sequence has a condensation whose saturation is covered by countably many compact sets whose intersections with [ ω ] < ω [\omega]^{<\omega} [ ω ] < ω are nonempty and monochromatic.
Lemma 4.7. Let A A A be a block sequence and let c : [ ω ] < ω → 2 c : [\omega]^{<\omega} \to2 c : [ ω ] < ω → 2 . There exist B ⪯ A B \preceq A B ⪯ A and a countable family K \mathcal{K} K of compact subsets of P ( ω ) ≈ 2 ω \mathcal{P}(\omega) \approx2^\omega P ( ω ) ≈ 2 ω such that:
( I ) (I) ( I ) S ( B ) ⊆ ⋃ K S(B) \subseteq\bigcup\mathcal{K} S ( B ) ⊆ ⋃ K ;
( I I ) (II) ( I I ) for every K ∈ K K \in\mathcal{K} K ∈ K , K ∩ [ ω ] < ω ≠ ∅ K \cap[\omega]^{<\omega} \neq\varnothing K ∩ [ ω ] < ω = ∅ ;
( I I I ) (III) ( I I I ) for every K ∈ K K \in\mathcal{K} K ∈ K , c ∣ K ∩ [ ω ] < ω c|_{K \cap[\omega]^{<\omega}} c ∣ K ∩ [ ω ] < ω is constant;
( I V ) (IV) ( I V ) for every K ∈ K K \in\mathcal{K} K ∈ K , K ∩ [ ω ] ω ⊆ S ( B ) K \cap[\omega]^\omega\subseteq S(B) K ∩ [ ω ] ω ⊆ S ( B ) .
Proof. Enumerate [ ω ] ω = { z j : j ∈ ω } [\omega]^\omega= \{z_j : j \in\omega\} [ ω ] ω = { z j : j ∈ ω } . We recursively construct sequences ( R n ) n ∈ ω (R_n)_{n \in\omega} ( R n ) n ∈ ω and ( b n ) n ∈ ω (b_n)_{n \in\omega} ( b n ) n ∈ ω such that, for every n ∈ ω n \in\omega n ∈ ω :
(i) R n R_n R n is a block sequence;
(ii) b n ∈ FU ( R n ) b_n \in\operatorname{FU}(R_n) b n ∈ FU ( R n ) ;
(iii) R 0 = A R_0=A R 0 = A ;
(iv) max b n < min b n + 1 \max b_n<\min b_{n+1} max b n < min b n + 1 ;
(v) R n + 1 ⪯ R n R_{n+1}\preceq R_n R n + 1 ⪯ R n , and max b n < min s \max b_n<\min s max b n < min s for every block s s s of R n + 1 R_{n+1} R n + 1 ;
(vi) For every j ≤ n j\le n j ≤ n and F ⊆ n F\subseteq n F ⊆ n , the map
v ⟼ c ( z j ∪ ⋃ i ∈ F b i ∪ b n ∪ v ) v\longmapsto c\left(z_j\cup\bigcup_{i\in F}b_i\cup b_n\cup v\right) v ⟼ c ( z j ∪ i ∈ F ⋃ b i ∪ b n ∪ v )
is constant on FU ( R n + 1 ) ∪ { ∅ } \operatorname{FU}(R_{n+1})\cup\{\varnothing\} FU ( R n + 1 ) ∪ { ∅ } .
Start with R 0 = A R_0=A R 0 = A . Assume that ( R i : i ≤ n ) (R_i:i\le n) ( R i : i ≤ n ) and ( b i : i < n ) (b_i:i<n) ( b i : i < n ) have been chosen. We show how to choose b n b_n b n and R n + 1 R_{n+1} R n + 1 . Put Q n = ( n + 1 ) × P ( n ) Q_n=(n+1)\times\mathcal{P}(n) Q n = ( n + 1 ) × P ( n ) and define
d n : FU ( R n ) ⟶ 2 Q n and d n ( v ) = ( c ( z j ∪ ⋃ i ∈ F b i ∪ v ) ) ( j , F ) ∈ Q n . d_n:\operatorname{FU}(R_n)\longrightarrow2^{Q_n}
\qquad\text{and}\qquad
d_n(v)=\left(c\left(z_j\cup\bigcup_{i\in F}b_i\cup v\right)\right)_{(j,F)\in Q_n}. d n : FU ( R n ) ⟶ 2 Q n and d n ( v ) = ( c ( z j ∪ i ∈ F ⋃ b i ∪ v ) ) ( j , F ) ∈ Q n .
Let I I I be the index set of R n R_n R n . Since 2 Q n 2^{Q_n} 2 Q n is finite, apply Hindman’s finite-unions theorem (Proposition 4.5) to R n R_n R n and d n d_n d n to obtain a condensation H n = ( h i ) i ∈ I ⪯ R n H_n=(h_i)_{i\in I}\preceq R_n H n = ( h i ) i ∈ I ⪯ R n such that d n d_n d n is constant on FU ( H n ) \operatorname{FU}(H_n) FU ( H n ) . Put i ∗ = min I i_*=\min I i ∗ = min I and define
b n = h i ∗ and R n + 1 = H n ↾ ( I ∖ { i ∗ } ) . b_n=h_{i_*}
\qquad\text{and}\qquad
R_{n+1}=H_n\mathbin{\upharpoonright}(I\setminus\{i_*\}). b n = h i ∗ and R n + 1 = H n ↾ ( I ∖ { i ∗ }) .
Clearly, all the required properties are satisfied.
Now let B = ( b n ) n ∈ ω B=(b_n)_{n\in\omega} B = ( b n ) n ∈ ω . Then B B B is a block sequence, B ⪯ A B\preceq A B ⪯ A , and B ≥ n + 1 ⪯ R n + 1 B_{\ge n+1}\preceq R_{n+1} B ≥ n + 1 ⪯ R n + 1 for every n ∈ ω n\in\omega n ∈ ω . For j ≤ n j\le n j ≤ n and F ⊆ n F\subseteq n F ⊆ n , put
K j , F , n = { z j ∪ ⋃ i ∈ F b i ∪ b n ∪ ⋃ k ∈ J b k : J ⊆ { k ∈ ω : k > n } } . (3) K_{j,F,n}=\left\{z_j\cup\bigcup_{i\in F}b_i\cup b_n\cup\bigcup_{k\in J}b_k:J\subseteq\{k\in\omega:k>n\}\right\}. \tag*{(3)} K j , F , n = { z j ∪ i ∈ F ⋃ b i ∪ b n ∪ k ∈ J ⋃ b k : J ⊆ { k ∈ ω : k > n } } . (3)
Claim 1. Each set K j , F , n K_{j,F,n} K j , F , n is compact.
Proof of the claim. Fix n ∈ ω n\in\omega n ∈ ω , j ≤ n j\le n j ≤ n , and F ⊆ n F\subseteq n F ⊆ n . Put T n = { k ∈ ω : k > n } T_n=\{k\in\omega:k>n\} T n = { k ∈ ω : k > n } and u = z j ∪ ⋃ i ∈ F b i ∪ b n u=z_j\cup\bigcup_{i\in F}b_i\cup b_n u = z j ∪ ⋃ i ∈ F b i ∪ b n . Define
Φ j , F , n : 2 T n ⟶ P ( ω ) ≅ 2 ω , Φ j , F , n ( t ) = u ∪ ⋃ k ∈ T n t ( k ) = 1 b k . \Phi_{j,F,n}:2^{T_n}\longrightarrow\mathcal{P}(\omega)\cong2^\omega,
\qquad
\Phi_{j,F,n}(t)=u\cup\bigcup_{\substack{k\in T_n\\t(k)=1}}b_k. Φ j , F , n : 2 T n ⟶ P ( ω ) ≅ 2 ω , Φ j , F , n ( t ) = u ∪ k ∈ T n t ( k ) = 1 ⋃ b k .
The range of Φ j , F , n \Phi_{j,F,n} Φ j , F , n is K j , F , n K_{j,F,n} K j , F , n . We check that Φ j , F , n \Phi_{j,F,n} Φ j , F , n is continuous, which then guarantees that K j , F , n K_{j,F,n} K j , F , n is compact. For brevity, write Φ = Φ j , F , n \Phi=\Phi_{j,F,n} Φ = Φ j , F , n . The sets
V m 1 = { x ⊆ ω : m ∈ x } and V m 0 = { x ⊆ ω : m ∉ x } , V_m^1=\{x\subseteq\omega:m\in x\}
\qquad\text{and}\qquad
V_m^0=\{x\subseteq\omega:m\notin x\}, V m 1 = { x ⊆ ω : m ∈ x } and V m 0 = { x ⊆ ω : m ∈ / x } ,
for m ∈ ω m\in\omega m ∈ ω form a subbasis for the topology of P ( ω ) \mathcal{P}(\omega) P ( ω ) . As for each m ∈ ω m\in\omega m ∈ ω , V m 0 = P ( ω ) ∖ V m 1 V_m^0=\mathcal{P}(\omega)\setminus V_m^1 V m 0 = P ( ω ) ∖ V m 1 , it suffices to check that the preimages of the sets V m 1 V_m^1 V m 1 are clopen sets.
Case 1: m ∈ u m\in u m ∈ u . Then Φ ( t ) ∈ V m 1 \Phi(t)\in V_m^1 Φ ( t ) ∈ V m 1 for every t ∈ 2 T n t\in2^{T_n} t ∈ 2 T n , so Φ − 1 [ V m 1 ] = 2 T n \Phi^{-1}[V_m^1]=2^{T_n} Φ − 1 [ V m 1 ] = 2 T n , which is clopen.
Case 2: m ∈ b k ∖ u m\in b_k\setminus u m ∈ b k ∖ u for some k ∈ T n k\in T_n k ∈ T n . We claim that Φ − 1 [ V m 1 ] = { t ∈ 2 T n : t ( k ) = 1 } \Phi^{-1}[V_m^1]=\{t\in2^{T_n}:t(k)=1\} Φ − 1 [ V m 1 ] = { t ∈ 2 T n : t ( k ) = 1 } , which is clopen. If t ( k ) = 1 t(k)=1 t ( k ) = 1 , then b k ⊆ Φ ( t ) b_k\subseteq\Phi(t) b k ⊆ Φ ( t ) , so m ∈ Φ ( t ) m\in\Phi(t) m ∈ Φ ( t ) . Conversely, if m ∈ Φ ( t ) m\in\Phi(t) m ∈ Φ ( t ) , then m ∉ u m \notin u m ∈ / u implies that there exists k ′ ∈ T n k' \in T_n k ′ ∈ T n with t ( k ′ ) = 1 t(k')=1 t ( k ′ ) = 1 and m ∈ b k ′ m \in b_{k'} m ∈ b k ′ . Since the blocks are pairwise disjoint, k = k ′ k=k' k = k ′ , so t ( k ) = 1 t(k)=1 t ( k ) = 1 .
Case 3: m ∉ u ∪ ⋃ k ∈ T n b k m \notin u \cup\bigcup_{k\in T_n} b_k m ∈ / u ∪ ⋃ k ∈ T n b k . Then m ∉ Φ ( t ) m \notin\Phi(t) m ∈ / Φ ( t ) for every t ∈ 2 T n t\in2^{T_n} t ∈ 2 T n , so Φ − 1 [ V m 1 ] = ∅ \Phi^{-1}[V^1_m]=\varnothing Φ − 1 [ V m 1 ] = ∅ , which is clopen.
Thus the preimage of every subbasic open set is open, so Φ \Phi Φ is continuous. Since 2 T n 2^{T_n} 2 T n is compact, its image K j , F , n K_{j,F,n} K j , F , n is compact.
Put K = { K j , F , n : n ∈ ω , j ≤ n , F ⊆ n } \mathcal{K}=\{K_{j,F,n}:n\in\omega,\ j\le n,\ F\subseteq n\} K = { K j , F , n : n ∈ ω , j ≤ n , F ⊆ n } .
K \mathcal{K} K is countable, and, by the Claim, each element of K \mathcal{K} K is compact. We verify (I)–(IV).
(I) Let x ∈ S ( B ) x\in S(B) x ∈ S ( B ) and write x = r ∪ ⋃ k ∈ J b k x=r\cup\bigcup_{k\in J}b_k x = r ∪ ⋃ k ∈ J b k , where r ∈ [ ω ] < ω r\in[\omega]^{<\omega} r ∈ [ ω ] < ω and J ∈ [ ω ] ω J\in[\omega]^\omega J ∈ [ ω ] ω . Choose j j j with r = z j r=z_j r = z j . Since J J J is infinite, choose n ∈ J n\in J n ∈ J with n ≥ j n\ge j n ≥ j , and put F = J ∩ n F=J\cap n F = J ∩ n and L = J ∖ ( n + 1 ) L=J\setminus(n+1) L = J ∖ ( n + 1 ) . Then F ⊆ n F\subseteq n F ⊆ n , L ⊆ { k ∈ ω : k > n } L\subseteq\{k\in\omega:k>n\} L ⊆ { k ∈ ω : k > n } , and
x = z j ∪ ⋃ i ∈ F b i ∪ b n ∪ ⋃ k ∈ L b k ∈ K j , F , n . x=z_j\cup\bigcup_{i\in F}b_i\cup b_n\cup\bigcup_{k\in L}b_k\in K_{j,F,n}. x = z j ∪ i ∈ F ⋃ b i ∪ b n ∪ k ∈ L ⋃ b k ∈ K j , F , n .
Thus S ( B ) ⊆ ⋃ K S(B)\subseteq\bigcup\mathcal{K} S ( B ) ⊆ ⋃ K .
For the remaining properties, fix n ∈ ω n\in\omega n ∈ ω , j ≤ n j\le n j ≤ n , and F ⊆ n F\subseteq n F ⊆ n , and put u = z j ∪ ⋃ i ∈ F b i ∪ b n u=z_j\cup\bigcup_{i\in F}b_i\cup b_n u = z j ∪ ⋃ i ∈ F b i ∪ b n .
(II) Taking J = ∅ J=\varnothing J = ∅ in (3) gives u ∈ K j , F , n ∩ [ ω ] < ω u\in K_{j,F,n}\cap[\omega]^{<\omega} u ∈ K j , F , n ∩ [ ω ] < ω , so this intersection is nonempty.
(III) Let x ∈ K j , F , n ∩ [ ω ] < ω x\in K_{j,F,n}\cap[\omega]^{<\omega} x ∈ K j , F , n ∩ [ ω ] < ω and write x = u ∪ ⋃ k ∈ J b k x=u\cup\bigcup_{k\in J}b_k x = u ∪ ⋃ k ∈ J b k , where J ⊆ { k ∈ ω : k > n } J\subseteq\{k\in\omega:k>n\} J ⊆ { k ∈ ω : k > n } . Since the blocks are nonempty and pairwise disjoint, J J J is finite. The relation B ≥ n + 1 ≼ R n + 1 B_{\ge n+1}\preccurlyeq R_{n+1} B ≥ n + 1 ≼ R n + 1 implies that
v = ⋃ k ∈ J b k ∈ FU ( R n + 1 ) ∪ { ∅ } . v=\bigcup_{k\in J}b_k\in\operatorname{FU}(R_{n+1})\cup\{\varnothing\}. v = k ∈ J ⋃ b k ∈ FU ( R n + 1 ) ∪ { ∅ } .
As x = u ∪ v x=u\cup v x = u ∪ v , we have c ( x ) = c ( u ) c(x)=c(u) c ( x ) = c ( u ) by property (vi). Hence c ↾ ( K j , F , n ∩ [ ω ] < ω ) c\mathbin{\upharpoonright}(K_{j,F,n}\cap[\omega]^{<\omega}) c ↾ ( K j , F , n ∩ [ ω ] < ω ) is constant.
(IV) Let x ∈ K j , F , n ∩ [ ω ] ω x\in K_{j,F,n}\cap[\omega]^\omega x ∈ K j , F , n ∩ [ ω ] ω and write x = u ∪ ⋃ k ∈ J b k x=u\cup\bigcup_{k\in J}b_k x = u ∪ ⋃ k ∈ J b k , where J ⊆ { k ∈ ω : k > n } J\subseteq\{k\in\omega:k>n\} J ⊆ { k ∈ ω : k > n } . Since u u u and all the blocks are finite, J J J is infinite. Therefore
x = z j ∪ ⋃ k ∈ F ∪ { n } ∪ J b k ∈ S ( B ) . x=z_j\cup\bigcup_{k\in F\cup\{n\}\cup J}b_k\in S(B). x = z j ∪ k ∈ F ∪ { n } ∪ J ⋃ b k ∈ S ( B ) .
This proves K j , F , n ∩ [ ω ] ω ⊆ S ( B ) K_{j,F,n}\cap[\omega]^\omega\subseteq S(B) K j , F , n ∩ [ ω ] ω ⊆ S ( B ) and completes the proof.
Lemma 4.8. For every block sequence A A A , the set { e x : x ∈ [ A ] ∞ } \{e_x:x\in[A]^\infty\} { e x : x ∈ [ A ] ∞ } is unbounded in ( ω ω , ≤ ∗ ) (\omega^\omega,\le^*) ( ω ω , ≤ ∗ ) .
Proof. Write A = ( a i ) i ∈ I A=(a_i)_{i\in I} A = ( a i ) i ∈ I and fix g ∈ ω ω g\in\omega^\omega g ∈ ω ω . Choose ( i k : k ∈ ω ) (i_k:k\in\omega) ( i k : k ∈ ω ) and ( m k : k ∈ ω ) (m_k:k\in\omega) ( m k : k ∈ ω ) such that, for every k ∈ ω k\in\omega k ∈ ω :
i k ∈ I i_k\in I i k ∈ I ;
i k + 1 > i k i_{k+1}>i_k i k + 1 > i k ;
m k = ∑ j < k ∣ a i j ∣ m_k=\sum_{j<k}|a_{i_j}| m k = ∑ j < k ∣ a i j ∣ ;
min a i k > g ( m k ) \min a_{i_k}>g(m_k) min a i k > g ( m k ) .
Such choices are possible because the minima of the blocks tend to infinity.
Put x = ⋃ k ∈ ω a i k ∈ [ A ] ∞ x=\bigcup_{k\in\omega}a_{i_k}\in[A]^\infty x = ⋃ k ∈ ω a i k ∈ [ A ] ∞ . Given k ∈ ω k\in\omega k ∈ ω , by construction, exactly m k m_k m k elements of x x x precede min a i k \min a_{i_k} min a i k , so e x ( m k ) = min a i k > g ( m k ) e_x(m_k)=\min a_{i_k}>g(m_k) e x ( m k ) = min a i k > g ( m k ) . Thus g < ∗ e x g<^*e_x g < ∗ e x , as required.
We also record the following elementary property of F σ F_\sigma F σ sets.
Lemma 4.9. Let M M M be a metrizable space and let P 0 P_0 P 0 , P 1 P_1 P 1 be F σ F_\sigma F σ subsets of M M M with M = P 0 ∪ P 1 M=P_0\cup P_1 M = P 0 ∪ P 1 . There exist disjoint F σ F_\sigma F σ subsets H 0 H_0 H 0 , H 1 H_1 H 1 of M M M such that
M = H 0 ∪ H 1 and H i ⊆ P i ( i ∈ { 0 , 1 } ) . M=H_0\cup H_1 \qquad\text{and}\qquad H_i\subseteq P_i \quad(i\in\{0,1\}). M = H 0 ∪ H 1 and H i ⊆ P i ( i ∈ { 0 , 1 }) . Proof. For each i ∈ { 0 , 1 } i\in\{0,1\} i ∈ { 0 , 1 } , write P i = ⋃ n ∈ ω C i , n P_i=\bigcup_{n\in\omega}C_{i,n} P i = ⋃ n ∈ ω C i , n , where ( C i , n ) n ∈ ω (C_{i,n})_{n\in\omega} ( C i , n ) n ∈ ω is an increasing sequence of closed subsets of M M M . We may assume that C 0 , 0 = C 1 , 0 = ∅ C_{0,0}=C_{1,0}=\varnothing C 0 , 0 = C 1 , 0 = ∅ . Define
H 0 = ⋃ n ∈ ω ( C 0 , n + 1 ∖ C 1 , n ) and H 1 = ⋃ n ∈ ω ( C 1 , n + 1 ∖ C 0 , n + 1 ) . H_0=\bigcup_{n\in\omega}(C_{0,n+1}\setminus C_{1,n}) \qquad\text{and}\qquad H_1=\bigcup_{n\in\omega}(C_{1,n+1}\setminus C_{0,n+1}). H 0 = n ∈ ω ⋃ ( C 0 , n + 1 ∖ C 1 , n ) and H 1 = n ∈ ω ⋃ ( C 1 , n + 1 ∖ C 0 , n + 1 ) . Each difference of closed sets is F σ F_\sigma F σ in the metrizable space M M M , so both H i H_i H i are F σ F_\sigma F σ , and clearly H i ⊆ P i H_i\subseteq P_i H i ⊆ P i . It is now straightforward to check that H 0 ∩ H 1 = ∅ H_0\cap H_1=\varnothing H 0 ∩ H 1 = ∅ and M = H 0 ∪ H 1 M=H_0\cup H_1 M = H 0 ∪ H 1 .
Combining Lemma 4.9 with Proposition 2.10 gives the following criterion.
Lemma 4.10. Let X X X be a separable metrizable space. Suppose that, for every pair of disjoint countable subsets Y , Z ⊆ X Y,Z\subseteq X Y , Z ⊆ X , there is a countable family F \mathcal{F} F of closed subsets of X X X such that
∣ X ∖ ⋃ F ∣ < b |X\setminus\bigcup\mathcal{F}|<\mathfrak{b} ∣ X ∖ ⋃ F ∣ < b ;
for every F ∈ F F\in\mathcal{F} F ∈ F , either F ∩ Y = ∅ F\cap Y=\varnothing F ∩ Y = ∅ or F ∩ Z = ∅ F\cap Z=\varnothing F ∩ Z = ∅ .
Then X X X is a weak λ \lambda λ -set.
Proof. Fix disjoint countable subsets Y , Z ⊆ X Y,Z\subseteq X Y , Z ⊆ X and choose F \mathcal{F} F as in the hypothesis. Put
T = ( X ∖ ⋃ F ) ∪ Y ∪ Z . T=\left(X\setminus\bigcup\mathcal{F}\right)\cup Y\cup Z. T = ( X ∖ ⋃ F ) ∪ Y ∪ Z . Since ∣ T ∣ < b |T|<\mathfrak{b} ∣ T ∣ < b , Proposition 2.10 implies that T T T is a λ \lambda λ -set. Thus Y = T ∩ G Y=T\cap G Y = T ∩ G for some G δ G_\delta G δ subset G G G of X X X . Define
P 0 = Y ∪ { F ∈ F : F ∩ Z = ∅ } , P_0=Y\cup\{F\in\mathcal{F}:F\cap Z=\varnothing\}, P 0 = Y ∪ { F ∈ F : F ∩ Z = ∅ } , P 1 = ( X ∖ G ) ∪ { F ∈ F : F ∩ Y = ∅ } . P_1=(X\setminus G)\cup\{F\in\mathcal{F}:F\cap Y=\varnothing\}. P 1 = ( X ∖ G ) ∪ { F ∈ F : F ∩ Y = ∅ } . Both sets are F σ F_\sigma F σ in X X X . Since T ⊆ Y ∪ ( X ∖ G ) T\subseteq Y\cup(X\setminus G) T ⊆ Y ∪ ( X ∖ G ) and every member of F \mathcal{F} F is contained in P 0 P_0 P 0 or P 1 P_1 P 1 , it follows that { P 0 , P 1 } \{P_0,P_1\} { P 0 , P 1 } is a cover of X X X in which P 0 P_0 P 0 and P 1 P_1 P 1 are F σ F_\sigma F σ sets. Apply Lemma 4.9 to obtain a partition X = H 0 ∪ H 1 X=H_0\cup H_1 X = H 0 ∪ H 1 into F σ F_\sigma F σ sets with H i ⊆ P i H_i\subseteq P_i H i ⊆ P i . Then Y ∩ H 1 = ∅ Y\cap H_1=\varnothing Y ∩ H 1 = ∅ and Z ∩ H 0 = ∅ Z\cap H_0=\varnothing Z ∩ H 0 = ∅ , so Y ⊆ H 0 Y\subseteq H_0 Y ⊆ H 0 and Z ⊆ H 1 Z\subseteq H_1 Z ⊆ H 1 . Each H i H_i H i is also G δ G_\delta G δ , since its complement is F σ F_\sigma F σ . Hence X X X is a weak λ \lambda λ -set.
We are now ready to prove Theorem D.
Proof of Theorem D. Assume p = c \mathfrak{p}=\mathfrak{c} p = c . Enumerate 2 [ ω ] < ω × [ [ ω ] ≤ ω ] ≤ ω 2^{[\omega]^{<\omega}}\times[[\omega]^{\leq\omega}]^{\leq\omega} 2 [ ω ] < ω × [[ ω ] ≤ ω ] ≤ ω as { ( c α , E α ) : α < c } \{(c_\alpha,E_\alpha):\alpha<\mathfrak{c}\} {( c α , E α ) : α < c } . First recursively construct sequences ( A α ) α < c (A_\alpha)_{\alpha<\mathfrak{c}} ( A α ) α < c and ( K α ) α < c (\mathcal{K}_\alpha)_{\alpha<\mathfrak{c}} ( K α ) α < c such that, for every α < c \alpha<\mathfrak{c} α < c :
A α A_\alpha A α is a block sequence;
A α ≼ ∗ A ξ A_\alpha\preccurlyeq^* A_\xi A α ≼ ∗ A ξ for every ξ < α \xi<\alpha ξ < α ;
K α \mathcal{K}_\alpha K α is a countable family of compact subsets of 2 ω 2^\omega 2 ω ;
S ( A α ) ∩ E α = ∅ S(A_\alpha)\cap E_\alpha=\varnothing S ( A α ) ∩ E α = ∅ ;
S ( A α ) ⊆ ⋃ K α S(A_\alpha)\subseteq\bigcup\mathcal{K}_\alpha S ( A α ) ⊆ ⋃ K α ;
(vi) for every K ∈ K α K \in\mathcal{K}_{\alpha} K ∈ K α ,
∅ ≠ K ∩ [ ω ] < ω , c α ↾ ( K ∩ [ ω ] < ω ) is constant , K ∩ [ ω ] ω ⊆ S ( A α ) . \varnothing\ne K \cap[\omega]^{<\omega}, \qquad c_{\alpha} \restriction(K \cap[\omega]^{<\omega}) \text{ is constant}, \qquad K \cap[\omega]^{\omega} \subseteq S(A_{\alpha}). ∅ = K ∩ [ ω ] < ω , c α ↾ ( K ∩ [ ω ] < ω ) is constant , K ∩ [ ω ] ω ⊆ S ( A α ) .
Assume that ( A ξ ) ξ < α (A_{\xi})_{\xi<\alpha} ( A ξ ) ξ < α and ( K ξ ) ξ < α (K_{\xi})_{\xi<\alpha} ( K ξ ) ξ < α have been chosen. We show how to choose A α A_{\alpha} A α and K α K_{\alpha} K α . If α = 0 \alpha= 0 α = 0 , let D = ( { n } ) n ∈ ω D = (\{n\})_{n\in\omega} D = ({ n } ) n ∈ ω . If α > 0 \alpha> 0 α > 0 , apply Lemma 4.4 to obtain a block sequence D ⪯ ∗ A ξ D \preceq^{*} A_{\xi} D ⪯ ∗ A ξ for every ξ < α \xi< \alpha ξ < α . This is possible because the previously chosen block sequences are ⪯ ∗ \preceq^{*} ⪯ ∗ -decreasing and α < p \alpha< \mathfrak{p} α < p .
Use Lemma 4.6 to choose a subsequence C C C of D D D such that S ( C ) ∩ E α = ∅ S(C) \cap E_{\alpha} = \varnothing S ( C ) ∩ E α = ∅ . Apply Lemma 4.7 to C C C and c α c_{\alpha} c α to obtain a block sequence A α ⪯ C A_{\alpha} \preceq C A α ⪯ C and a family K α \mathcal{K}_{\alpha} K α satisfying (iii), (v), and (vi). Since A α ⪯ C ⪯ D A_{\alpha} \preceq C \preceq D A α ⪯ C ⪯ D , transitivity gives (ii). By Lemma 4.3, S ( A α ) ⊆ S ( C ) S(A_{\alpha}) \subseteq S(C) S ( A α ) ⊆ S ( C ) , so (iv) holds as well. This ends the recursive construction.
This fixes the block sequences and covers independently of the points. Now enumerate ω ω = { g α : α < c } \omega^{\omega} = \{g_{\alpha} : \alpha< \mathfrak{c}\} ω ω = { g α : α < c } . For each α < c \alpha< \mathfrak{c} α < c , use Lemma 4.8 to choose x α ∈ [ A α ] ∞ x_{\alpha} \in[A_{\alpha}]_{\infty} x α ∈ [ A α ] ∞ such that g α ≺ ∞ e x α g_{\alpha} \prec^{\infty} e_{x_{\alpha}} g α ≺ ∞ e x α . Put
X = [ ω ] < ω ∪ { x α : α < c } . X = [\omega]^{<\omega} \cup\{x_{\alpha} : \alpha< \mathfrak{c}\}. X = [ ω ] < ω ∪ { x α : α < c } .
Thus, { e x α : α < c } \{e_{x_{\alpha}} : \alpha< \mathfrak{c}\} { e x α : α < c } is unbounded in ( ω ω , ≤ ∗ ) (\omega^{\omega}, \leq^{*}) ( ω ω , ≤ ∗ ) . By the definition of b \mathfrak{b} b , this collection has cardinality at least b = c \mathfrak{b} = \mathfrak{c} b = c , so ∣ X ∣ = c |X| = \mathfrak{c} ∣ X ∣ = c . We now show that X X X is a weak λ \lambda λ -set but not a λ \lambda λ -set.
X X X is a weak λ \lambda λ -set. Let Y , Z ⊆ X Y, Z \subseteq X Y , Z ⊆ X be countable and disjoint. Put E = ( Y ∪ Z ) ∩ [ ω ] ω E = (Y \cup Z) \cap[\omega]^{\omega} E = ( Y ∪ Z ) ∩ [ ω ] ω and choose a coloring c : [ ω ] < ω → 2 c : [\omega]^{<\omega} \to2 c : [ ω ] < ω → 2 with value zero on Y ∩ [ ω ] < ω Y \cap[\omega]^{<\omega} Y ∩ [ ω ] < ω and one on Z ∩ [ ω ] < ω Z \cap[\omega]^{<\omega} Z ∩ [ ω ] < ω . Choose α \alpha α with ( c , E ) = ( c α , E α ) (c, E) = (c_{\alpha}, E_{\alpha}) ( c , E ) = ( c α , E α ) . For every β ≥ α \beta\geq\alpha β ≥ α , Property (ii) and the choice of x β x_{\beta} x β , together with Lemma 4.3, give
x β ∈ [ A β ] ∞ ⊆ S ( A β ) ⊆ S ( A α ) . (4) x_{\beta} \in[A_{\beta}]_{\infty} \subseteq S(A_{\beta}) \subseteq S(A_{\alpha}). \tag*{(4)} x β ∈ [ A β ] ∞ ⊆ S ( A β ) ⊆ S ( A α ) . (4)
Also by (vi), for every K ∈ K α K \in\mathcal{K}_{\alpha} K ∈ K α , c α ↾ ( K ∩ [ ω ] < ω ) c_{\alpha} \restriction(K \cap[\omega]^{<\omega}) c α ↾ ( K ∩ [ ω ] < ω ) is constant. Let ε ( K ) ∈ { 0 , 1 } \varepsilon(K) \in\{0,1\} ε ( K ) ∈ { 0 , 1 } be the unique value such that c α ↾ ( K ∩ [ ω ] < ω ) ≡ ε ( K ) c_{\alpha} \restriction(K \cap[\omega]^{<\omega}) \equiv\varepsilon(K) c α ↾ ( K ∩ [ ω ] < ω ) ≡ ε ( K ) .
By the choice of c α = c c_{\alpha} = c c α = c , K K K avoids Z ∩ [ ω ] < ω Z \cap[\omega]^{<\omega} Z ∩ [ ω ] < ω if ε ( K ) = 0 \varepsilon(K) = 0 ε ( K ) = 0 , and avoids Y ∩ [ ω ] < ω Y \cap[\omega]^{<\omega} Y ∩ [ ω ] < ω if ε ( K ) = 1 \varepsilon(K) = 1 ε ( K ) = 1 . Moreover, K ∩ [ ω ] ω ⊆ S ( A α ) K \cap[\omega]^{\omega} \subseteq S(A_{\alpha}) K ∩ [ ω ] ω ⊆ S ( A α ) and S ( A α ) ∩ E α = ∅ S(A_{\alpha}) \cap E_{\alpha} = \varnothing S ( A α ) ∩ E α = ∅ , so every K ∈ K α K \in\mathcal{K}_{\alpha} K ∈ K α avoids ( Y ∪ Z ) ∩ [ ω ] ω = E α (Y \cup Z) \cap[\omega]^{\omega} = E_{\alpha} ( Y ∪ Z ) ∩ [ ω ] ω = E α . Thus, for every K ∈ K α K \in\mathcal{K}_{\alpha} K ∈ K α ,
ε ( K ) = 0 ⟹ K ∩ Z = ∅ and ε ( K ) = 1 ⟹ K ∩ Y = ∅ . (5) \varepsilon(K) = 0 \Longrightarrow K \cap Z = \varnothing\qquad\text{and} \qquad\varepsilon(K) = 1 \Longrightarrow K \cap Y = \varnothing. \tag*{(5)} ε ( K ) = 0 ⟹ K ∩ Z = ∅ and ε ( K ) = 1 ⟹ K ∩ Y = ∅ . (5)
Put F = { K ∩ X : K ∈ K α } \mathcal{F} = \{K \cap X : K \in\mathcal{K}_{\alpha}\} F = { K ∩ X : K ∈ K α } . By Property (iii), F \mathcal{F} F is countable, and each of its members is closed in X X X . Equation (5) shows that each member of F \mathcal{F} F is disjoint from Y Y Y or from Z Z Z . By (4) and Property (v),
X ∖ ⋃ F ⊆ [ ω ] < ω ∪ { x ξ : ξ < α } . X \setminus\bigcup\mathcal{F} \subseteq[\omega]^{<\omega} \cup\{x_{\xi} : \xi< \alpha\}. X ∖ ⋃ F ⊆ [ ω ] < ω ∪ { x ξ : ξ < α } .
The set on the right has cardinality less than b = c \mathfrak{b} = \mathfrak{c} b = c . Since Y , Z Y, Z Y , Z were arbitrary, Lemma 4.10 implies that X X X is a weak λ \lambda λ -set.
X X X is not a λ \lambda λ -set. By Lemma 2.11, the unboundedness of { e x α : α < c } = { e x : x ∈ X ∖ [ ω ] < ω } \{e_{x_{\alpha}} : \alpha< \mathfrak{c}\} = \{e_x : x \in X \setminus[\omega]^{<\omega}\} { e x α : α < c } = { e x : x ∈ X ∖ [ ω ] < ω } implies that [ ω ] < ω [\omega]^{<\omega} [ ω ] < ω is not G δ G_{\delta} G δ in X X X . Hence X X X is not a λ \lambda λ -set.
We emphasize that the proof presented in this section is direct and does not rely on Bell’s theorem, in contrast to the approach adopted in [18 ] .
Weak λ-sets when s < b \mathfrak{s} < \mathfrak{b} s < b
The inequality b < s \mathfrak{b} < \mathfrak{s} b < s yields a weak λ \lambda λ -set which is not a λ \lambda λ -set. When s < b \mathfrak{s} < \mathfrak{b} s < b , the two classes coincide. The proof applies to arbitrary first countable spaces.
Recall that every weak λ \lambda λ -space is T 1 T_{1} T 1 .
Proposition 5.1. Assume s < b \mathfrak{s}<\mathfrak{b} s < b . Every first countable weak λ \lambda λ -space is a λ \lambda λ -space.
Proof. Let X X X be a first countable weak λ \lambda λ -space. Since X X X is T 1 T_1 T 1 , first countability makes every singleton, and hence every finite subset, G δ G_\delta G δ in X X X . Let D ⊆ X D\subseteq X D ⊆ X be countably infinite. For each d ∈ D d\in D d ∈ D , fix a decreasing open local base ( U ( d , n ) : n ∈ ω ) (U(d,n):n\in\omega) ( U ( d , n ) : n ∈ ω ) . Since X X X is T 1 T_1 T 1 , for every d ∈ D d\in D d ∈ D we have
⋂ n ∈ ω U ( d , n ) = { d } (6) \bigcap_{n\in\omega}U(d,n)=\{d\} \tag*{(6)} n ∈ ω ⋂ U ( d , n ) = { d } (6)
Fix a splitting family of subsets of D D D , S ⊆ P ( D ) \mathcal{S}\subseteq\mathcal{P}(D) S ⊆ P ( D ) , of cardinality s \mathfrak{s} s . For S ∈ S S\in\mathcal{S} S ∈ S , write S 1 = S S^1=S S 1 = S and S 0 = D ∖ S S^0=D\setminus S S 0 = D ∖ S . For each S ∈ S S\in\mathcal{S} S ∈ S , choose a partition X = H S 0 ∪ ˙ H S 1 X=H_S^0\mathbin{\dot\cup}H_S^1 X = H S 0 ∪ ˙ H S 1 into F σ F_\sigma F σ sets such that S i ⊆ H S i S^i\subseteq H_S^i S i ⊆ H S i for i < 2 i<2 i < 2 . This is possible because X X X is a weak λ \lambda λ -space. For ε < 2 \varepsilon<2 ε < 2 , write
H S ε = ⋃ n ∈ ω F S , n ε , H_S^\varepsilon=\bigcup_{n\in\omega}F_{S,n}^\varepsilon, H S ε = n ∈ ω ⋃ F S , n ε ,
where each F S , n ε F_{S,n}^\varepsilon F S , n ε is closed in X X X and the sequence is increasing in n n n .
Choose an increasing sequence ( D n ) n ∈ ω (D_n)_{n\in\omega} ( D n ) n ∈ ω of finite subsets of D D D with D = ⋃ n D n D=\bigcup_nD_n D = ⋃ n D n . For each S ∈ S S\in\mathcal{S} S ∈ S , choose q S ∈ ω ω q_S\in\omega^\omega q S ∈ ω ω such that, for every n ∈ ω n\in\omega n ∈ ω ,
U ( d , q S ( n ) ) ∩ F S , n 0 = ∅ for d ∈ D n ∩ S 1 , U ( d , q S ( n ) ) ∩ F S , n 1 = ∅ for d ∈ D n ∩ S 0 . (7) \begin{aligned}
U(d,q_S(n))\cap F_{S,n}^0&=\varnothing&&\text{for }d\in D_n\cap S^1,\\
U(d,q_S(n))\cap F_{S,n}^1&=\varnothing&&\text{for }d\in D_n\cap S^0. \tag*{(7)}
\end{aligned} U ( d , q S ( n )) ∩ F S , n 0 U ( d , q S ( n )) ∩ F S , n 1 = ∅ = ∅ for d ∈ D n ∩ S 1 , for d ∈ D n ∩ S 0 . (7)
Such a choice is possible because D n D_n D n is finite, F S , n 0 F_{S,n}^0 F S , n 0 avoids S 1 S^1 S 1 , F S , n 1 F_{S,n}^1 F S , n 1 avoids S 0 S^0 S 0 , and the local bases are decreasing.
As s < b \mathfrak{s}<\mathfrak{b} s < b , { q S : S ∈ S } \{q_S:S\in\mathcal{S}\} { q S : S ∈ S } is bounded in ( ω ω , ≤ ∗ ) (\omega^\omega,\leq^*) ( ω ω , ≤ ∗ ) . Choose h ∈ ω ω h\in\omega^\omega h ∈ ω ω such that q S ≤ ∗ h q_S\leq^*h q S ≤ ∗ h for every S ∈ S S\in\mathcal{S} S ∈ S . We may assume that for every n ∈ ω n\in\omega n ∈ ω , h ( n ) ≥ n h(n)\geq n h ( n ) ≥ n . Define
G = ⋂ m ∈ ω ⋃ n ≥ m ⋃ d ∈ D n U ( d , h ( n ) ) . (8) G=\bigcap_{m\in\omega}\bigcup_{n\geq m}\bigcup_{d\in D_n}U(d,h(n)). \tag*{(8)} G = m ∈ ω ⋂ n ≥ m ⋃ d ∈ D n ⋃ U ( d , h ( n )) . (8)
This is a G δ G_\delta G δ subset of X X X , and D ⊆ G D\subseteq G D ⊆ G . We will show that D = G D=G D = G .
Suppose towards a contradiction that G ∖ D ≠ ∅ G\setminus D\neq\varnothing G ∖ D = ∅ and fix x ∈ G ∖ D x\in G\setminus D x ∈ G ∖ D . By Equation (8) , there exists an infinite set N ⊆ ω N\subseteq\omega N ⊆ ω and points d n ∈ D n d_n\in D_n d n ∈ D n , for n ∈ N n\in N n ∈ N , such that
x ∈ ⋂ n ∈ N U ( d n , h ( n ) ) . x\in\bigcap_{n\in N}U(d_n,h(n)). x ∈ n ∈ N ⋂ U ( d n , h ( n )) .
For each d ∈ D d\in D d ∈ D , Equation (6) gives some M ′ ∈ ω M'\in\omega M ′ ∈ ω with x ∉ U ( d , M ′ ) x\notin U(d,M') x ∈ / U ( d , M ′ ) . If n ∈ N n\in N n ∈ N and n ≥ M ′ n\geq M' n ≥ M ′ , then h ( n ) ≥ n ≥ M ′ h(n)\geq n\geq M' h ( n ) ≥ n ≥ M ′ , so U ( d , h ( n ) ) ⊆ U ( d , M ′ ) U(d,h(n))\subseteq U(d,M') U ( d , h ( n )) ⊆ U ( d , M ′ ) . Hence x ∉ U ( d , h ( n ) ) x\notin U(d,h(n)) x ∈ / U ( d , h ( n )) , whereas x ∈ U ( d n , h ( n ) ) x\in U(d_n,h(n)) x ∈ U ( d n , h ( n )) , and therefore d n ≠ d d_n\neq d d n = d . Thus each point of D D D occurs only finitely often among the d n d_n d n , which implies that A = { d n : n ∈ N } ⊆ D A=\{d_n:n\in N\}\subseteq D A = { d n : n ∈ N } ⊆ D is infinite.
Choose S ∈ S S\in\mathcal{S} S ∈ S which splits A A A , and let ε < 2 \varepsilon<2 ε < 2 be such that x ∈ H S ε x\in H_S^\varepsilon x ∈ H S ε . Since the closed sets F S , n ε F_{S,n}^\varepsilon F S , n ε increase to H S ε H_S^\varepsilon H S ε and q S ≤ ∗ h q_S\leq^*h q S ≤ ∗ h , choose M ∈ ω M\in\omega M ∈ ω such that x ∈ F S , M ε x\in F_{S,M}^\varepsilon x ∈ F S , M ε and q S ( n ) ≤ h ( n ) q_S(n)\leq h(n) q S ( n ) ≤ h ( n ) for every n ≥ M n\geq M n ≥ M . For n ∈ N n\in N n ∈ N with n ≥ M n\geq M n ≥ M , we have x ∈ F S , n ε x\in F_{S,n}^\varepsilon x ∈ F S , n ε and
x ∈ U ( d n , h ( n ) ) ⊆ U ( d n , q S ( n ) ) . (9) x\in U(d_n,h(n))\subseteq U(d_n,q_S(n)). \tag*{(9)} x ∈ U ( d n , h ( n )) ⊆ U ( d n , q S ( n )) . (9)
Equation (7) then gives d n ∈ S ε d_n\in S^\varepsilon d n ∈ S ε : otherwise U ( d n , q S ( n ) ) U(d_n,q_S(n)) U ( d n , q S ( n )) would be disjoint from F S , n ε F_{S,n}^\varepsilon F S , n ε , contradicting Equation (9) . Consequently, A ∖ S ε ⊆ { d n : n ∈ N , n < M } A\setminus S^\varepsilon\subseteq\{d_n:n\in N,\ n<M\} A ∖ S ε ⊆ { d n : n ∈ N , n < M } , contradicting that S S S splits A A A .
Therefore G = D G=D G = D . Since D D D was arbitrary, X X X is a λ \lambda λ -space. □ \square □
Proof of Theorem B. Every weak λ \lambda λ -set is first countable, so the conclusion follows from Proposition 5.1. □ \square □
Weak lambda-sets in the Cohen model
In the previous sections, we saw that both p = c \mathfrak p=\mathfrak c p = c and b < s \mathfrak b<\mathfrak s b < s imply the existence of weak λ \lambda λ -sets which are not λ \lambda λ -sets. Proposition 5.1 gives the opposite conclusion when s < b \mathfrak s<\mathfrak b s < b . In this section, we obtain the same conclusion by adding ℵ 2 \aleph_2 ℵ 2 Cohen reals to a model of CH.
For a set A A A , let Fn ( A , 2 ) \operatorname{Fn}(A,2) Fn ( A , 2 ) denote the set of finite partial functions from A A A into 2 2 2 , ordered by reverse inclusion: p ≤ q p\le q p ≤ q if and only if p ⊇ q p\supseteq q p ⊇ q . For a cardinal κ \kappa κ , the Cohen forcing for adding κ \kappa κ many Cohen reals is C κ = Fn ( κ × ω , 2 ) \mathbb C_\kappa=\operatorname{Fn}(\kappa\times\omega,2) C κ = Fn ( κ × ω , 2 ) . The Cohen forcing for adding a single Cohen real is C = Fn ( ω , 2 ) \mathbb C=\operatorname{Fn}(\omega,2) C = Fn ( ω , 2 ) .
To prove Theorem C, we first determine how adding one Cohen real affects whether a ground-model set is a λ \lambda λ -set or a weak λ \lambda λ -set. We will need the following well-known result.
Lemma 6.1 ([7 Corollary 2.2] ). Let P \mathbb P P be a ccc forcing, and let X X X be a set. Suppose that p ∈ P p\in\mathbb P p ∈ P and D ˙ \dot D D ˙ is a P \mathbb P P -name such that
p ⊩ D ˙ ⊆ X ˇ and ∣ D ˙ ∣ ≤ ℵ 0 . p\Vdash\dot D\subseteq\check X\text{ and }|\dot D|\le\aleph_0. p ⊩ D ˙ ⊆ X ˇ and ∣ D ˙ ∣ ≤ ℵ 0 . Then there exists a countable A ⊆ X A\subseteq X A ⊆ X such that p ⊩ D ˙ ⊆ A ˇ p\Vdash\dot D\subseteq\check A p ⊩ D ˙ ⊆ A ˇ .
Lemma 6.2. Let X ⊆ 2 ω X\subseteq2^\omega X ⊆ 2 ω , and let D ⊆ X D\subseteq X D ⊆ X be countable. If D D D is not G δ G_\delta G δ in X X X , then
C ⊩ ∃ B ⊆ D ˇ ( no relatively F σ and G δ subset H of X ˇ satisfies H ∩ D ˇ = B ) . \mathbb C\Vdash\exists B\subseteq\check D\ \left(\text{no relatively }F_\sigma\text{ and }G_\delta\text{ subset }H\text{ of }\check X\text{ satisfies }H\cap\check D=B\right). C ⊩ ∃ B ⊆ D ˇ ( no relatively F σ and G δ subset H of X ˇ satisfies H ∩ D ˇ = B ) .
Proof. Work over a countable transitive model M M M of ZFC with X , D ∈ M X,D\in M X , D ∈ M satisfying the hypotheses there. Let G ′ ⊆ C G'\subseteq\mathbb C G ′ ⊆ C be generic over M M M . Since D D D is countably infinite in M M M , Fn ( ω , 2 ) \operatorname{Fn}(\omega,2) Fn ( ω , 2 ) and Fn ( D , 2 ) \operatorname{Fn}(D,2) Fn ( D , 2 ) are isomorphic in M M M . Thus there exists a Fn ( D , 2 ) \operatorname{Fn}(D,2) Fn ( D , 2 ) -generic filter G G G over M M M such that M [ G ] = M [ G ′ ] M[G]=M[G'] M [ G ] = M [ G ′ ] .
Let g = ⋃ G : D → 2 g=\bigcup G:D\to2 g = ⋃ G : D → 2 . Then g ∈ M [ G ] g\in M[G] g ∈ M [ G ] . Let B 0 = g − 1 [ { 0 } ] B_0=g^{-1}[\{0\}] B 0 = g − 1 [{ 0 }] and B 1 = g − 1 [ { 1 } ] B_1=g^{-1}[\{1\}] B 1 = g − 1 [{ 1 }] .
Working in M [ G ] M[G] M [ G ] , suppose that there exists H ⊆ X H\subseteq X H ⊆ X which is F σ F_\sigma F σ and G δ G_\delta G δ in X X X and satisfies H ∩ D = B 1 H\cap D=B_1 H ∩ D = B 1 . Choose G δ G_\delta G δ sets H 0 , H 1 ⊆ 2 ω H_0,H_1\subseteq2^\omega H 0 , H 1 ⊆ 2 ω such that
H 0 ∩ X = X ∖ H , H 1 ∩ X = H . H_0\cap X=X\setminus H,\qquad H_1\cap X=H. H 0 ∩ X = X ∖ H , H 1 ∩ X = H .
For each i < 2 i<2 i < 2 , choose a sequence U i = ( U n i : n ∈ ω ) U^i=(U^i_n:n\in\omega) U i = ( U n i : n ∈ ω ) of open subsets of 2 ω 2^\omega 2 ω such that
H i = ⋂ n ∈ ω U n i . H_i=\bigcap_{n\in\omega}U^i_n. H i = n ∈ ω ⋂ U n i .
Let U ˙ i ∈ M \dot U^i\in M U ˙ i ∈ M be a Fn ( D , 2 ) \operatorname{Fn}(D,2) Fn ( D , 2 ) -name such that U ˙ G i = U i \dot U^i_G=U^i U ˙ G i = U i for each i < 2 i<2 i < 2 . By the Maximum Principle in M M M , let g ˙ \dot g g ˙ be a Fn ( D , 2 ) \operatorname{Fn}(D,2) Fn ( D , 2 ) -name such that Fn ( D , 2 ) ⊩ g ˙ = ⋃ G ˙ \operatorname{Fn}(D,2)\Vdash\dot g=\bigcup\dot G Fn ( D , 2 ) ⊩ g ˙ = ⋃ G ˙ .
By the Truth Lemma, there exists p 0 ∈ G p_0\in G p 0 ∈ G such that
(i) p 0 ⊩ U ˙ 0 p_0\Vdash\dot U^0 p 0 ⊩ U ˙ 0 and U ˙ 1 \dot U^1 U ˙ 1 are sequences of open subsets of 2 ω 2^\omega 2 ω ;
(ii) p 0 ⊩ X ˇ ∩ ⋂ n ∈ ω U ˙ 0 ( n ) ∩ ⋂ n ∈ ω U ˙ 1 ( n ) = ∅ p_0\Vdash\check X\cap\bigcap_{n\in\omega}\dot U^0(n)\cap\bigcap_{n\in\omega}\dot U^1(n)=\varnothing p 0 ⊩ X ˇ ∩ ⋂ n ∈ ω U ˙ 0 ( n ) ∩ ⋂ n ∈ ω U ˙ 1 ( n ) = ∅ ;
(iii) p 0 ⊩ ∀ d ∈ D ˇ ∀ n ∈ ω ( g ˙ ( d ) = 0 ⇒ d ∈ U ˙ 0 ( n ) ) p_0\Vdash\forall d\in\check D\ \forall n\in\omega\ (\dot g(d)=0\Rightarrow d\in\dot U^0(n)) p 0 ⊩ ∀ d ∈ D ˇ ∀ n ∈ ω ( g ˙ ( d ) = 0 ⇒ d ∈ U ˙ 0 ( n )) ;
(iv) p 0 ⊩ ∀ d ∈ D ˇ ∀ n ∈ ω ( g ˙ ( d ) = 1 ⇒ d ∈ U ˙ 1 ( n ) ) p_0\Vdash\forall d\in\check D\ \forall n\in\omega\ (\dot g(d)=1\Rightarrow d\in\dot U^1(n)) p 0 ⊩ ∀ d ∈ D ˇ ∀ n ∈ ω ( g ˙ ( d ) = 1 ⇒ d ∈ U ˙ 1 ( n )) .
Working in M M M , for p ≤ p 0 p\le p_0 p ≤ p 0 , n ∈ ω n\in\omega n ∈ ω , and i < 2 i<2 i < 2 , let
V p , n , i = { x ∈ X : ( ∃ q ≤ p ) q ⊩ x ˇ ∈ U ˙ i ( n ˇ ) } . V_{p,n,i}=\{x\in X:(\exists q\le p)\ q\Vdash\check x\in\dot U^i(\check n)\}. V p , n , i = { x ∈ X : ( ∃ q ≤ p ) q ⊩ x ˇ ∈ U ˙ i ( n ˇ )} .
Each V p , n , i V_{p,n,i} V p , n , i is open in X X X . Indeed, for x ∈ V p , n , i x\in V_{p,n,i} x ∈ V p , n , i , choose q ≤ p q\le p q ≤ p forcing x ˇ ∈ U ˙ i ( n ˇ ) \check x\in\dot U^i(\check n) x ˇ ∈ U ˙ i ( n ˇ ) . Since q ⊩ q\Vdash q ⊩ “U ˙ i ( n ˇ ) \dot U^i(\check n) U ˙ i ( n ˇ ) is open”, there exist q 0 ≤ q q_0\le q q 0 ≤ q and t ∈ 2 < ω t\in2^{<\omega} t ∈ 2 < ω with t ⊆ x t\subseteq x t ⊆ x and
q 0 ⊩ X ˇ ∩ [ t ˇ ] ⊆ U ˙ i ( n ˇ ) . q_0\Vdash\check X\cap[\check t]\subseteq\dot U^i(\check n). q 0 ⊩ X ˇ ∩ [ t ˇ ] ⊆ U ˙ i ( n ˇ ) .
Then X ∩ [ t ] X \cap[t] X ∩ [ t ] is a neighborhood of x x x contained in V p , n , i V_{p,n,i} V p , n , i .
If d ∈ D ∖ dom ( p ) d \in D \setminus\operatorname{dom}(p) d ∈ D ∖ dom ( p ) , let q = p ∪ { ( d , i ) } q=p\cup\{(d,i)\} q = p ∪ {( d , i )} . Then q ≤ p q\leq p q ≤ p and q ⊩ g ˙ ( d ˇ ) = i ˇ q\Vdash\dot g(\check d)=\check i q ⊩ g ˙ ( d ˇ ) = i ˇ . Hence q ⊩ d ˇ ∈ U ˙ i ( n ˇ ) q\Vdash\check d\in\dot U^i(\check n) q ⊩ d ˇ ∈ U ˙ i ( n ˇ ) , so
D ∖ dom ( p ) ⊆ V p , n , i . D\setminus\operatorname{dom}(p)\subseteq V_{p,n,i}. D ∖ dom ( p ) ⊆ V p , n , i .
For x ∈ X ∖ D x\in X\setminus D x ∈ X ∖ D , by (ii) there exist p ≤ p 0 p\leq p_0 p ≤ p 0 , n ∈ ω n\in\omega n ∈ ω , and i < 2 i<2 i < 2 such that p ⊩ x ˇ ∉ U ˙ i ( n ˇ ) p\Vdash\check x\notin\dot U^i(\check n) p ⊩ x ˇ ∈ / U ˙ i ( n ˇ ) . So x ∉ V p , n , i x\notin V_{p,n,i} x ∈ / V p , n , i . Also x ∉ dom ( p ) x\notin\operatorname{dom}(p) x ∈ / dom ( p ) , since dom ( p ) ⊆ D \operatorname{dom}(p)\subseteq D dom ( p ) ⊆ D .
Working in M M M , for each p ≤ p 0 p\leq p_0 p ≤ p 0 and m ∈ ω m\in\omega m ∈ ω , let
W p , m = X ∩ ⋃ d ∈ dom ( p ) [ d ↾ m ] . W_{p,m}=X\cap\bigcup_{d\in\operatorname{dom}(p)}[d\mathbin{\upharpoonright}m]. W p , m = X ∩ d ∈ dom ( p ) ⋃ [ d ↾ m ] .
These sets are clopen in X X X and, since dom ( p ) \operatorname{dom}(p) dom ( p ) is finite, ⋂ m ∈ ω W p , m = dom ( p ) \bigcap_{m\in\omega}W_{p,m}=\operatorname{dom}(p) ⋂ m ∈ ω W p , m = dom ( p ) . The preceding observations give, in M M M ,
D = ⋂ { V p , n , i ∪ W p , m : p ≤ p 0 , n , m ∈ ω , i < 2 } . D=\bigcap\{V_{p,n,i}\cup W_{p,m}:p\leq p_0,\ n,m\in\omega,\ i<2\}. D = ⋂ { V p , n , i ∪ W p , m : p ≤ p 0 , n , m ∈ ω , i < 2 } .
Since Fn ( D , 2 ) \operatorname{Fn}(D,2) Fn ( D , 2 ) is countable in M M M , this is a countable intersection of open subsets of X X X in M M M . Thus D D D is G δ G_\delta G δ in X X X in M M M , a contradiction.
Corollary 6.3
Corollary 6.3. Let X ⊆ 2 ω X\subseteq2^\omega X ⊆ 2 ω .
If X X X is a λ \lambda λ -set, then C ⊩ X ˇ \mathbb C\Vdash\check X C ⊩ X ˇ is a λ \lambda λ -set.
If X X X is not a λ \lambda λ -set, then C ⊩ X ˇ \mathbb C\Vdash\check X C ⊩ X ˇ is not a weak λ \lambda λ -set.
Proof. For (i), we must see that for every p ∈ C p\in\mathbb C p ∈ C and C \mathbb C C -name D ˙ \dot D D ˙ such that p ⊩ D ˙ ⊆ X ˇ p\Vdash\dot D\subseteq\check X p ⊩ D ˙ ⊆ X ˇ and ∣ D ˙ ∣ ≤ ℵ 0 |\dot D|\leq\aleph_0 ∣ D ˙ ∣ ≤ ℵ 0 , we have p ⊩ p\Vdash p ⊩ “D ˙ \dot D D ˙ is relatively G δ G_\delta G δ in X ˇ \check X X ˇ ”.
By Lemma 6.1, there exists a countable A ⊆ X A\subseteq X A ⊆ X such that p ⊩ D ˙ ⊆ A ˇ p\Vdash\dot D\subseteq\check A p ⊩ D ˙ ⊆ A ˇ . Since X X X is a λ \lambda λ -set, A A A is relatively G δ G_\delta G δ in X X X , and hence p ⊩ p\Vdash p ⊩ “A ˇ \check A A ˇ is relatively G δ G_\delta G δ in X ˇ \check X X ˇ ”. As also p ⊩ ∣ A ˇ ∖ D ˙ ∣ ≤ ℵ 0 p\Vdash|\check A\setminus\dot D|\leq\aleph_0 p ⊩ ∣ A ˇ ∖ D ˙ ∣ ≤ ℵ 0 , it follows that p ⊩ p\Vdash p ⊩ “D ˙ \dot D D ˙ is relatively G δ G_\delta G δ in X ˇ \check X X ˇ ”.
For (ii), let D ⊆ X D\subseteq X D ⊆ X be countable and not relatively G δ G_\delta G δ in X X X . By Lemma 6.2,
C ⊩ ∃ B ⊆ D ˇ (no relatively F σ and G δ subset H of X ˇ satisfies H ∩ D ˇ = B ) . \mathbb C\Vdash\exists B\subseteq\check D\ \text{(no relatively }F_\sigma\text{ and }G_\delta\text{ subset }H\text{ of }\check X\text{ satisfies }H\cap\check D=B\text{)}. C ⊩ ∃ B ⊆ D ˇ (no relatively F σ and G δ subset H of X ˇ satisfies H ∩ D ˇ = B ) .
Thus, C ⊩ X ˇ \mathbb C\Vdash\check X C ⊩ X ˇ is not a weak λ \lambda λ -set.
A Luzin set is an uncountable L ⊆ 2 ω L\subseteq2^\omega L ⊆ 2 ω whose intersection with every meager subset of 2 ω 2^\omega 2 ω is countable. By [20 Theorem 4.4] , every weak λ \lambda λ -set contained in 2 ω 2^\omega 2 ω is perfectly meager, hence meager in 2 ω 2^\omega 2 ω . A Luzin set L L L is not meager: otherwise the Luzin property applied to L L L itself would make L L L countable. Thus no Luzin set is a weak λ \lambda λ -set.
The following result is due to Miller. It is also cited in the proof of Theorem 22 of Miller’s survey [16 ] .
Proposition 6.4 (Miller [15 pp. 577–578] ). Assume CH. Then:
C ω 2 ⊩ every X ⊆ 2 ω of size ℵ 2 contains a continuous injective image of a Luzin set. \mathbb C_{\omega_2}\Vdash\text{every }X\subseteq2^\omega\text{ of size }\aleph_2\text{ contains a continuous injective image of a Luzin set.} C ω 2 ⊩ every X ⊆ 2 ω of size ℵ 2 contains a continuous injective image of a Luzin set.
In Proposition 6.4, both the Luzin set and the set containing its continuous injective image are subsets of 2 ω 2^\omega 2 ω . In the following lemma, a weak λ \lambda λ -set may be any separable metrizable space.
Lemma 6.5. Assume that every X ⊆ 2 ω X\subseteq2^\omega X ⊆ 2 ω of size ℵ 2 \aleph_2 ℵ 2 contains a continuous injective image of a Luzin set. Then every weak λ \lambda λ -set has size at most ℵ 1 \aleph_1 ℵ 1 .
Proof. Suppose that a weak λ \lambda λ -set X X X has size at least ℵ 2 \aleph_{2} ℵ 2 . Identify X X X with a subspace of [ 0 , 1 ] ω [0,1]^{\omega} [ 0 , 1 ] ω and fix a continuous surjection π : 2 ω → [ 0 , 1 ] ω \pi:2^{\omega}\to[0,1]^{\omega} π : 2 ω → [ 0 , 1 ] ω . Choose Y ⊆ 2 ω Y\subseteq2^{\omega} Y ⊆ 2 ω of size ℵ 2 \aleph_{2} ℵ 2 such that π [ Y ] ⊆ X \pi[Y]\subseteq X π [ Y ] ⊆ X and π ↾ Y \pi\mathbin{\upharpoonright}Y π ↾ Y is injective. By hypothesis, there is a continuous injection f : L → Y f:L\to Y f : L → Y from a Luzin set L L L . Thus g = π ∘ f : L → X g=\pi\circ f:L\to X g = π ∘ f : L → X is continuous and injective. Pulling back relatively F σ F_{\sigma} F σ and G δ G_{\delta} G δ separators along g g g shows that L L L is a weak λ \lambda λ -set, contrary to the fact that no Luzin set is a weak λ \lambda λ -set.
Corollary 6.6. Assume CH. Then:
C ω 2 ⊩ every weak λ -set has size at most ℵ 1 . \mathbb{C}_{\omega_{2}}\Vdash\text{ every weak }\lambda\text{-set has size at most }\aleph_{1}. C ω 2 ⊩ every weak λ -set has size at most ℵ 1 . Proof. Apply Lemma 6.5 using Proposition 6.4.
The following lemma is well-known and follows from standard techniques involving the use of nice names; see [14 Lemma IV.3.10 and the proof of Lemma V.2.19] .
Lemma 6.7. Let p ∈ C ω 2 p\in\mathbb{C}_{\omega_{2}} p ∈ C ω 2 and let X ˙ \dot{X} X ˙ be a name such that
p ⊩ ∅ ≠ X ˙ ⊆ 2 ω and ∣ X ˙ ∣ ≤ ℵ 1 . p\Vdash\varnothing\ne\dot{X}\subseteq2^{\omega}\quad\text{and}\quad|\dot{X}|\leq\aleph_{1}. p ⊩ ∅ = X ˙ ⊆ 2 ω and ∣ X ˙ ∣ ≤ ℵ 1 . There exist S ⊆ ω 2 S\subseteq\omega_{2} S ⊆ ω 2 of size at most ℵ 1 \aleph_{1} ℵ 1 , containing all coordinates of p p p , and a Fn ( S × ω , 2 ) \operatorname{Fn}(S\times\omega,2) Fn ( S × ω , 2 ) -name Y ˙ \dot{Y} Y ˙ such that p ⊩ X ˙ = Y ˙ p\Vdash\dot{X}=\dot{Y} p ⊩ X ˙ = Y ˙ .
We are now ready to present a proof of Theorem C .
Proof of Theorem C . Work over a countable transitive model M M M of ZFC + + + CH. Put κ = ( ω 2 ) M \kappa=(\omega_{2})^{M} κ = ( ω 2 ) M , and let G G G be C κ \mathbb{C}_{\kappa} C κ -generic over M M M . Working in M [ G ] M[G] M [ G ] , fix a nonempty weak λ \lambda λ -set X X X . By Corollary 6.6, ∣ X ∣ ≤ ℵ 1 < c |X|\leq\aleph_{1}<\mathfrak{c} ∣ X ∣ ≤ ℵ 1 < c . Thus X X X is zero-dimensional, as observed in § 3 . Since X X X is separable and metrizable, we may assume that X ⊆ 2 ω X\subseteq2^{\omega} X ⊆ 2 ω .
Choose a C κ \mathbb{C}_{\kappa} C κ -name X ˙ ∈ M \dot{X}\in M X ˙ ∈ M with X ˙ G = X \dot{X}_{G}=X X ˙ G = X and p ∈ G p\in G p ∈ G such that p ⊩ ∅ ≠ X ˙ ⊆ 2 ω p\Vdash\varnothing\ne\dot{X}\subseteq2^{\omega} p ⊩ ∅ = X ˙ ⊆ 2 ω and p ⊩ ∣ X ˙ ∣ ≤ ℵ 1 p\Vdash|\dot{X}|\leq\aleph_{1} p ⊩ ∣ X ˙ ∣ ≤ ℵ 1 . Working in M M M , apply Lemma 6.7 to obtain S ⊆ κ S\subseteq\kappa S ⊆ κ of size at most ℵ 1 \aleph_{1} ℵ 1 and a Fn ( S × ω , 2 ) \operatorname{Fn}(S\times\omega,2) Fn ( S × ω , 2 ) -name Y ˙ \dot{Y} Y ˙ such that p ⊩ C κ X ˙ = Y ˙ p\Vdash_{\mathbb{C}_{\kappa}}\dot{X}=\dot{Y} p ⊩ C κ X ˙ = Y ˙ . Choose β ∈ κ ∖ S \beta\in\kappa\setminus S β ∈ κ ∖ S and put
P = Fn ( ( κ ∖ { β } ) × ω , 2 ) , Q = Fn ( { β } × ω , 2 ) . P=\operatorname{Fn}((\kappa\setminus\{\beta\})\times\omega,2),\qquad Q=\operatorname{Fn}(\{\beta\}\times\omega,2). P = Fn (( κ ∖ { β }) × ω , 2 ) , Q = Fn ({ β } × ω , 2 ) .
Let H = G ∩ P H=G\cap P H = G ∩ P and K = G ∩ Q K=G\cap Q K = G ∩ Q . By [14 Lemma V.2.18] , H H H is P P P -generic over M M M , K K K is Q Q Q -generic over M [ H ] M[H] M [ H ] , and M [ G ] = M [ H ] [ K ] M[G]=M[H][K] M [ G ] = M [ H ] [ K ] . Put N = M [ H ] N=M[H] N = M [ H ] . Since Y ˙ \dot{Y} Y ˙ is a P P P -name and p ∈ G p\in G p ∈ G , X = Y ˙ G = Y ˙ H ∈ N X=\dot{Y}_{G}=\dot{Y}_{H}\in N X = Y ˙ G = Y ˙ H ∈ N .
The poset Q Q Q is naturally isomorphic to C \mathbb{C} C . Assume, towards a contradiction, that N ⊨ X N\models X N ⊨ X is not a λ \lambda λ -set. Corollary 6.3(ii), applied in N N N , implies that X X X is not a weak λ \lambda λ -set in N [ K ] = M [ G ] N[K]=M[G] N [ K ] = M [ G ] , a contradiction. Hence N ⊨ X N\models X N ⊨ X is a λ \lambda λ -set, and Corollary 6.3(i), again applied in N N N , implies that X X X is a λ \lambda λ -set in M [ G ] M[G] M [ G ] .
Corollary 6.8. Assuming Con ( Z F C ) \operatorname{Con}(\mathrm{ZFC}) Con ( ZFC ) , the existence of a weak λ \lambda λ -set which is not a λ \lambda λ -set is independent of ZFC + b = s = ℵ 1 +\mathfrak{b}=\mathfrak{s}=\aleph_{1} + b = s = ℵ 1 .
Proof. Working in the metatheory, it is known that Con ( Z F C ) \operatorname{Con}(\mathrm{ZFC}) Con ( ZFC ) implies Con ( Z F C + C H ) \operatorname{Con}(\mathrm{ZFC}+\mathrm{CH}) Con ( ZFC + CH ) since CH holds in the constructible universe L L L (see e.g. [14 Theorem II.6.24] ). Since CH implies p = c \mathfrak{p}=\mathfrak{c} p = c and b = s = ℵ 1 \mathfrak{b}=\mathfrak{s}=\aleph_{1} b = s = ℵ 1 , under CH Theorem D gives a weak λ \lambda λ -set which is not a λ \lambda λ -set. So ZFC + b = s = ℵ 1 +\mathfrak{b}=\mathfrak{s}=\aleph_{1} + b = s = ℵ 1 does not imply that every weak λ \lambda λ -set is a λ \lambda λ -set.
On the other hand, working in ZFC + + + CH, Theorem C shows that
C ω 2 ⊩ every weak λ -set is a λ -set . \mathbb{C}_{\omega_{2}}\Vdash\text{ every weak }\lambda\text{-set is a }\lambda\text{-set}. C ω 2 ⊩ every weak λ -set is a λ -set . As C ω 2 \mathbb{C}_{\omega_{2}} C ω 2 also forces b = s = ℵ 1 \mathfrak{b}=\mathfrak{s}=\aleph_{1} b = s = ℵ 1 (see e.g. [2 Section 11.3] ), Con ( Z F C + C H ) \operatorname{Con}(\mathrm{ZFC}+\mathrm{CH}) Con ( ZFC + CH ) implies the consistency of ZFC + b = s = ℵ 1 +\mathfrak{b}=\mathfrak{s}=\aleph_{1} + b = s = ℵ 1 together with the assertion that every weak λ \lambda λ -set is a λ \lambda λ -set. Thus, assuming Con ( Z F C ) \operatorname{Con}(\mathrm{ZFC}) Con ( ZFC ) , the existence of a weak λ \lambda λ -set which is not a λ \lambda λ -set cannot be proved from ZFC + b = s = ℵ 1 +\mathfrak{b}=\mathfrak{s}=\aleph_{1} + b = s = ℵ 1 .
:::`
An almost-normal MAD family under CH
In this section we construct an almost-normal MAD family under the Continuum Hypothesis. This construction improves a previous result appearing in [4 ] , where the existence of an almost-normal almost disjoint family which fails to be normal is shown to follow from CH. An earlier unpublished version of that paper (see [5 ] ) claimed that it actually follows from CH that almost-normal MAD families exist. However, there was a gap in the argument. On the other hand, the main idea of the construction there was the use of good families of partitioners to obtain maximality. We now fix that gap and provide a correct proof of the existence of an almost-normal MAD family under CH using good families of partitioners.
We start with some notation, terminology and definitions. For each set D ⊆ ω D \subseteq\omega D ⊆ ω let D 0 = ω ∖ D D^0 = \omega\setminus D D 0 = ω ∖ D and D 1 = D D^1 = D D 1 = D . For any collection H ⊆ P ( ω ) \mathcal{H} \subseteq\mathcal{P}(\omega) H ⊆ P ( ω ) and any set C ⊆ ω C \subseteq\omega C ⊆ ω , we write
H C = { a ∈ H : ∣ a ∩ C ∣ = ω } . \mathcal{H}_C = \{a \in\mathcal{H} : |a \cap C| = \omega\}. H C = { a ∈ H : ∣ a ∩ C ∣ = ω } .
For each s ∈ Fn ( H , 2 ) s \in\operatorname{Fn}(\mathcal{H},2) s ∈ Fn ( H , 2 ) , define the Boolean combination determined by s s s as
H s : = ⋂ H ∈ dom ( s ) H s ( H ) . \mathcal{H}^s := \bigcap_{H \in\operatorname{dom}(s)} H^{s(H)}. H s := H ∈ dom ( s ) ⋂ H s ( H ) .
We define H ∅ = ω \mathcal{H}^{\varnothing} = \omega H ∅ = ω . Finally, if A , D ⊆ P ( ω ) \mathcal{A},\mathcal{D} \subseteq\mathcal{P}(\omega) A , D ⊆ P ( ω ) and s ∈ Fn ( D , 2 ) s \in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) , we let
A ( s ) = { a ∈ A : a ⊆ ∗ D s } . \mathcal{A}(s) = \{a \in\mathcal{A} : a \subseteq^* \mathcal{D}^s\}. A ( s ) = { a ∈ A : a ⊆ ∗ D s } .
The observation below follows directly from the definitions. Its proof is straightforward and left to the reader.
Definition 7.2. Let D \mathcal{D} D be a collection of partitioners for an almost disjoint family A \mathcal{A} A . We say that D \mathcal{D} D is good for A \mathcal{A} A if, for every s ∈ Fn ( D , 2 ) s \in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) ,
∣ A ( s ) ∣ < ω ⟹ D s = ∗ ⋃ A ( s ) . |\mathcal{A}(s)| < \omega\quad\Longrightarrow\quad\mathcal{D}^s =^* \bigcup\mathcal{A}(s). ∣ A ( s ) ∣ < ω ⟹ D s = ∗ ⋃ A ( s ) .
Note that if D \mathcal{D} D is good for A \mathcal{A} A , then every subfamily of D \mathcal{D} D is also good for A \mathcal{A} A . The next lemma gives a condition ensuring that a partitioner for a given partition of A \mathcal{A} A preserves goodness when added to a good family of partitioners D \mathcal{D} D .
Lemma 7.3. Let A \mathcal{A} A be an almost disjoint family, D \mathcal{D} D be a good family of partitioners for A \mathcal{A} A , and partition A = B ∪ ˙ E \mathcal{A} = \mathcal{B} \mathbin{\dot\cup} \mathcal{E} A = B ∪ ˙ E . For each s ∈ Fn ( D , 2 ) s \in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) , let
L s = D s ∖ ⋃ E ( s ) , R s = D s ∖ ⋃ B ( s ) L_s = \mathcal{D}^s \setminus\bigcup\mathcal{E}(s), \qquad R_s = \mathcal{D}^s \setminus\bigcup\mathcal{B}(s) L s = D s ∖ ⋃ E ( s ) , R s = D s ∖ ⋃ B ( s )
and set
L = B ∪ { L s : s ∈ Fn ( D , 2 ) , ∣ E ( s ) ∣ < ω } , R = E ∪ { R s : s ∈ Fn ( D , 2 ) , ∣ B ( s ) ∣ < ω } . \mathcal{L}=\mathcal{B}\cup\{L_s:s\in\operatorname{Fn}(\mathcal{D},2),\ |\mathcal{E}(s)|<\omega\},
\qquad
\mathcal{R}=\mathcal{E}\cup\{R_s:s\in\operatorname{Fn}(\mathcal{D},2),\ |\mathcal{B}(s)|<\omega\}. L = B ∪ { L s : s ∈ Fn ( D , 2 ) , ∣ E ( s ) ∣ < ω } , R = E ∪ { R s : s ∈ Fn ( D , 2 ) , ∣ B ( s ) ∣ < ω } .
Then L ⊥ R \mathcal{L}\perp\mathcal{R} L ⊥ R and every D ⊆ ω D\subseteq\omega D ⊆ ω that separates L \mathcal{L} L from R \mathcal{R} R also separates B \mathcal{B} B from E \mathcal{E} E and makes D ∪ { D } \mathcal{D}\cup\{D\} D ∪ { D } good for A \mathcal{A} A .
Proof. Since B \mathcal{B} B and E \mathcal{E} E are disjoint subfamilies of A \mathcal{A} A , B ⊥ E \mathcal{B}\perp\mathcal{E} B ⊥ E . Fix s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) with ∣ B ( s ) ∣ < ω |\mathcal{B}(s)|<\omega ∣ B ( s ) ∣ < ω and let a ∈ B a\in\mathcal{B} a ∈ B . Then either a ∩ D s a\cap D^s a ∩ D s is finite or a ∈ B ( s ) a\in\mathcal{B}(s) a ∈ B ( s ) , in both cases ∣ a ∩ R s ∣ < ω |a\cap R_s|<\omega ∣ a ∩ R s ∣ < ω . Similarly, for every s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) with ∣ E ( s ) ∣ < ω |\mathcal{E}(s)|<\omega ∣ E ( s ) ∣ < ω and every e ∈ E e\in\mathcal{E} e ∈ E , we have ∣ L s ∩ e ∣ < ω |L_s\cap e|<\omega ∣ L s ∩ e ∣ < ω .
It remains to consider L s ∩ R t L_s\cap R_t L s ∩ R t . Fix s , t ∈ Fn ( D , 2 ) s,t\in\operatorname{Fn}(\mathcal{D},2) s , t ∈ Fn ( D , 2 ) such that E ( s ) \mathcal{E}(s) E ( s ) and B ( t ) \mathcal{B}(t) B ( t ) are finite. If s , t s,t s , t are incompatible, then L s ∩ R t ⊆ D s ∩ D t = ∅ L_s\cap R_t\subseteq D^s\cap D^t=\varnothing L s ∩ R t ⊆ D s ∩ D t = ∅ . If they are compatible, then
A ( s ∪ t ) = E ( s ∪ t ) ∪ B ( s ∪ t ) ⊆ E ( s ) ∪ B ( t ) . \mathcal{A}(s\cup t)=\mathcal{E}(s\cup t)\cup\mathcal{B}(s\cup t)\subseteq\mathcal{E}(s)\cup\mathcal{B}(t). A ( s ∪ t ) = E ( s ∪ t ) ∪ B ( s ∪ t ) ⊆ E ( s ) ∪ B ( t ) .
The right-hand side is finite, so goodness gives
D s ∪ t = D s ∩ D t ⊆ ∗ ⋃ E ( s ) ∪ ⋃ B ( t ) . D^{s\cup t}=D^s\cap D^t\subseteq^*\bigcup\mathcal{E}(s)\cup\bigcup\mathcal{B}(t). D s ∪ t = D s ∩ D t ⊆ ∗ ⋃ E ( s ) ∪ ⋃ B ( t ) .
Hence L s ∩ R t ⊆ ∗ ⋃ E ( s ) ∪ ⋃ B ( t ) L_s\cap R_t\subseteq^*\bigcup\mathcal{E}(s)\cup\bigcup\mathcal{B}(t) L s ∩ R t ⊆ ∗ ⋃ E ( s ) ∪ ⋃ B ( t ) which from their definition implies that L s ∩ R t L_s\cap R_t L s ∩ R t is finite.
Let us now show that D ∪ { D } \mathcal{D}\cup\{D\} D ∪ { D } is a good family of partitioners for each D D D that separates L \mathcal{L} L from R \mathcal{R} R . For such a set, we also have that D D D separates B \mathcal{B} B from E \mathcal{E} E , so it is a partitioner for A \mathcal{A} A . It remains to show that goodness is preserved when adding D D D . If D ∈ D D\in\mathcal{D} D ∈ D , then D ∪ { D } = D \mathcal{D}\cup\{D\}=\mathcal{D} D ∪ { D } = D and we are done. Assume D ∉ D D\notin\mathcal{D} D ∈ / D and put D ′ = D ∪ { D } \mathcal{D}'=\mathcal{D}\cup\{D\} D ′ = D ∪ { D } . Fix s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) and put s i = s ∪ { ( D , i ) } s_i=s\cup\{(D,i)\} s i = s ∪ {( D , i )} for i ∈ 2 i\in2 i ∈ 2 . Since D D D separates B \mathcal{B} B from E \mathcal{E} E , we have
A ( s 1 ) = B ( s ) and A ( s 0 ) = E ( s ) . \mathcal{A}(s_1)=\mathcal{B}(s)
\qquad\text{and}\qquad
\mathcal{A}(s_0)=\mathcal{E}(s). A ( s 1 ) = B ( s ) and A ( s 0 ) = E ( s ) .
If B ( s ) \mathcal{B}(s) B ( s ) is finite, then clearly ⋃ B ( s ) ⊆ ∗ D ∩ D s = D ′ s 1 \bigcup\mathcal{B}(s)\subseteq^*D\cap D^s=D^{\prime s_1} ⋃ B ( s ) ⊆ ∗ D ∩ D s = D ′ s 1 , as each b ∈ B ( s ) b\in\mathcal{B}(s) b ∈ B ( s ) is almost contained in D ′ s 1 D^{\prime s_1} D ′ s 1 and B ( s ) \mathcal{B}(s) B ( s ) is finite. On the other hand, since R s ∈ R R_s\in\mathcal{R} R s ∈ R , we have that R s ∩ D = ( D s ∖ ⋃ B ( s ) ) ∩ D R_s\cap D=(D^s\setminus\bigcup\mathcal{B}(s))\cap D R s ∩ D = ( D s ∖ ⋃ B ( s )) ∩ D is finite, or equivalently, that D ′ s 1 = D s ∩ D ⊆ ∗ ⋃ B ( s ) D^{\prime s_1}=D^s\cap D\subseteq^*\bigcup\mathcal{B}(s) D ′ s 1 = D s ∩ D ⊆ ∗ ⋃ B ( s ) . Together, these inclusions give
D ′ s 1 = ∗ ⋃ B ( s ) = ⋃ A ( s 1 ) . D^{\prime s_1}=^*\bigcup\mathcal{B}(s)=\bigcup\mathcal{A}(s_1). D ′ s 1 = ∗ ⋃ B ( s ) = ⋃ A ( s 1 ) .
If E ( s ) \mathcal{E}(s) E ( s ) is finite, a dual argument gives that D ′ s 0 = ∗ ⋃ A ( s 0 ) D^{\prime s_0}=^*\bigcup\mathcal{A}(s_0) D ′ s 0 = ∗ ⋃ A ( s 0 ) . We conclude that D ∪ { D } \mathcal{D}\cup\{D\} D ∪ { D } is good for A \mathcal{A} A .
The next lemma gives a sufficient condition for adding a separator containing a prescribed set C ⊆ ω C\subseteq\omega C ⊆ ω while preserving goodness.
Lemma 7.4. Let A \mathcal{A} A be countable and almost disjoint, let D \mathcal{D} D be a countable good family of partitioners for A \mathcal{A} A , and let C ⊆ ω C\subseteq\omega C ⊆ ω . Suppose that, for every s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) ,
∣ A C ( s ) ∣ < ω ⟹ C ∩ D s ⊆ ∗ ⋃ A C ( s ) . (10) |\mathcal{A}_C(s)|<\omega\quad\Longrightarrow\quad C\cap D^s\subseteq^*\bigcup\mathcal{A}_C(s). \tag*{(10)} ∣ A C ( s ) ∣ < ω ⟹ C ∩ D s ⊆ ∗ ⋃ A C ( s ) . (10) Then there exists a set D ⊆ ω D\subseteq\omega D ⊆ ω with C ⊆ D C\subseteq D C ⊆ D separating A C \mathcal{A}_C A C from A ∖ A C \mathcal{A}\setminus\mathcal{A}_C A ∖ A C such that D ∪ { D } \mathcal{D}\cup\{D\} D ∪ { D } is good for A \mathcal{A} A .
Proof. Let L , R \mathcal{L},\mathcal{R} L , R be the collections from Lemma 7.3 with B = A C \mathcal{B}=\mathcal{A}_C B = A C and E = A ∖ A C \mathcal{E}=\mathcal{A}\setminus\mathcal{A}_C E = A ∖ A C . They are orthogonal and, since A \mathcal{A} A and Fn ( D , 2 ) \operatorname{Fn}(\mathcal{D},2) Fn ( D , 2 ) are countable, both collections are countable. By the definition of A C \mathcal{A}_C A C , C C C meets every member of E \mathcal{E} E only finitely. For each s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) with ∣ A C ( s ) ∣ < ω |\mathcal{A}_C(s)|<\omega ∣ A C ( s ) ∣ < ω , (10) gives
∣ C ∩ ( D s ∖ ⋃ A C ( s ) ) ∣ < ω . \left|C\cap\left(D^s\setminus\bigcup\mathcal{A}_C(s)\right)\right|<\omega. C ∩ ( D s ∖ ⋃ A C ( s ) ) < ω . Thus { C } ⊥ R \{C\}\perp\mathcal{R} { C } ⊥ R , and hence ( { C } ∪ L ) ⊥ R (\{C\}\cup\mathcal{L})\perp\mathcal{R} ({ C } ∪ L ) ⊥ R . Lemma 2.3 gives a set D ⊇ C D\supseteq C D ⊇ C separating { C } ∪ L \{C\}\cup\mathcal{L} { C } ∪ L from R \mathcal{R} R . By Lemma 7.3 , D D D separates A C \mathcal{A}_C A C from its complement in A \mathcal{A} A and D ∪ { D } \mathcal{D}\cup\{D\} D ∪ { D } is good for A \mathcal{A} A .
Lemma 7.5. Suppose that D \mathcal{D} D is good for A \mathcal{A} A , that A ⊆ A ′ \mathcal{A} \subseteq\mathcal{A}' A ⊆ A ′ , and that A ′ \mathcal{A}' A ′ is almost disjoint. If every D ∈ D D \in\mathcal{D} D ∈ D is a partitioner for A ′ \mathcal{A}' A ′ , then D \mathcal{D} D is good for A ′ \mathcal{A}' A ′ .
Proof. Fix s ∈ Fn ( D , 2 ) s \in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) and suppose that A ′ ( s ) \mathcal{A}'(s) A ′ ( s ) is finite. Then A ( s ) \mathcal{A}(s) A ( s ) is finite, and goodness for A \mathcal{A} A gives D s = ∗ ⋃ A ( s ) \mathcal{D}^{s} =^{*} \bigcup\mathcal{A}(s) D s = ∗ ⋃ A ( s ) . Almost disjointness of A ′ \mathcal{A}' A ′ implies that b ∩ D s b \cap\mathcal{D}^{s} b ∩ D s is finite for each b ∈ A ′ ∖ A b \in\mathcal{A}' \setminus\mathcal{A} b ∈ A ′ ∖ A . Thus A ′ ( s ) = A ( s ) \mathcal{A}'(s) = \mathcal{A}(s) A ′ ( s ) = A ( s ) and D s = ∗ ⋃ A ′ ( s ) \mathcal{D}^{s} =^{*} \bigcup\mathcal{A}'(s) D s = ∗ ⋃ A ′ ( s ) .
We now include some terminology in order to simplify our coming recursive construction.
Definition 7.6. Let P ⊆ P ( ω ) × P ( ω ) \mathcal{P} \subseteq\mathcal{P}(\omega) \times\mathcal{P}(\omega) P ⊆ P ( ω ) × P ( ω ) be countable. We say that ( A , P ) (\mathcal{A},\mathcal{P}) ( A , P ) is a countable compatible system if:
(a) A \mathcal{A} A is a countable almost disjoint family,
(b) D = { D : ( C , D ) ∈ P } \mathcal{D} = \{D : (C,D) \in\mathcal{P}\} D = { D : ( C , D ) ∈ P } is a good family of partitioners for A \mathcal{A} A ;
(c) C ⊆ ∗ D C \subseteq^{*} D C ⊆ ∗ D for every ( C , D ) ∈ P (C,D) \in\mathcal{P} ( C , D ) ∈ P and
(d) for each ( C , D ) ∈ P (C,D) \in\mathcal{P} ( C , D ) ∈ P , A C = { a ∈ A : a ⊆ ∗ D } \mathcal{A}_{C} = \{a \in\mathcal{A} : a \subseteq^{*} D\} A C = { a ∈ A : a ⊆ ∗ D } .
Lemma 7.7. Let ( A , P ) (\mathcal{A},\mathcal{P}) ( A , P ) be a countable compatible system , and let X ∈ [ ω ] ω X \in[\omega]^{\omega} X ∈ [ ω ] ω be almost disjoint from A \mathcal{A} A . There exists a ∗ ∈ [ ω ] ω a_{*} \in[\omega]^{\omega} a ∗ ∈ [ ω ] ω such that ∣ a ∗ ∩ X ∣ = ω |a_{*} \cap X| = \omega ∣ a ∗ ∩ X ∣ = ω and ( A ∪ { a ∗ } , P ) (\mathcal{A} \cup\{a_{*}\},\mathcal{P}) ( A ∪ { a ∗ } , P ) is a countable compatible system.
Proof. Adding the pair ( ω , ω ) (\omega,\omega) ( ω , ω ) to P \mathcal{P} P does not affect compatibility, so we may assume that P ≠ ∅ \mathcal{P} \ne\varnothing P = ∅ . Enumerate P \mathcal{P} P as { ( C n , D n ) : n < ω } \{(C_n,D_n):n<\omega\} {( C n , D n ) : n < ω } , allowing repetitions if necessary, and put D = { D n : n < ω } \mathcal{D}=\{D_n:n<\omega\} D = { D n : n < ω } .
The hypotheses imply that A \mathcal{A} A is infinite: otherwise goodness applied to s = ∅ s=\varnothing s = ∅ would make ω \omega ω almost covered by finitely many members of A \mathcal{A} A , contrary to the choice of X X X . Enumerate A = { a j : j < ω } \mathcal{A}=\{a_j:j<\omega\} A = { a j : j < ω } . Starting with s 0 = ∅ s_0=\varnothing s 0 = ∅ , construct an increasing sequence ( s n ) n < ω (s_n)_{n<\omega} ( s n ) n < ω in Fn ( D , 2 ) \operatorname{Fn}(\mathcal{D},2) Fn ( D , 2 ) such that, for each n ∈ ω n\in\omega n ∈ ω ,
dom ( s n ) = { D i : i < n } and ∣ X ∩ D s n ∣ = ω . (11) \operatorname{dom}(s_n)=\{D_i:i<n\}\qquad\text{and}\qquad|X\cap\mathcal{D}^{s_n}|=\omega. \tag*{(11)} dom ( s n ) = { D i : i < n } and ∣ X ∩ D s n ∣ = ω . (11)
This can be easily done by picking i ∈ 2 i\in2 i ∈ 2 such that ( X ∩ D s n ) ∩ D n i (X\cap\mathcal{D}^{s_n})\cap D_n^i ( X ∩ D s n ) ∩ D n i is infinite and by defining s n + 1 ( D n ) = i s_{n+1}(D_n)=i s n + 1 ( D n ) = i . Define:
h = ⋃ n < ω s n : D → 2 h=\bigcup_{n<\omega}s_n:\mathcal{D}\to2 h = n < ω ⋃ s n : D → 2
For every n < ω n<\omega n < ω , the set A ( s n ) \mathcal{A}(s_n) A ( s n ) is infinite, otherwise goodness would give D s n = ∗ ⋃ A ( s n ) \mathcal{D}^{s_n}=^{*}\bigcup\mathcal{A}(s_n) D s n = ∗ ⋃ A ( s n ) , where the right-hand side is a finite union of members of A \mathcal{A} A , contradicting (11) and the fact that X X X is almost disjoint from A \mathcal{A} A .
Choose recursively finite nonempty sets F n F_n F n for n < ω n<\omega n < ω such that
(i) F n ⊆ D s n ∖ ( ⋃ j < n a j ∪ ⋃ m < n F m ) F_n\subseteq\mathcal{D}^{s_n}\setminus\left(\bigcup_{j<n}a_j\cup\bigcup_{m<n}F_m\right) F n ⊆ D s n ∖ ( ⋃ j < n a j ∪ ⋃ m < n F m ) ;
(ii) F n ∩ X ≠ ∅ F_n\cap X\ne\varnothing F n ∩ X = ∅ ;
(iii) for every i < n i<n i < n with h ( D i ) = 1 h(D_i)=1 h ( D i ) = 1 , F n ∩ C i ≠ ∅ F_n\cap C_i\ne\varnothing F n ∩ C i = ∅ .
By (11), X ∩ D s n X\cap\mathcal{D}^{s_n} X ∩ D s n is infinite. As X X X is almost disjoint from A \mathcal{A} A , we can choose l ∈ X l\in X l ∈ X satisfying item (i). Similarly, if i < n i<n i < n and h ( D i ) = 1 h(D_i)=1 h ( D i ) = 1 , choose b n ∈ A ( s n ) ∖ { a j : j < n } b_n\in\mathcal{A}(s_n)\setminus\{a_j:j<n\} b n ∈ A ( s n ) ∖ { a j : j < n } .
As b n ∈ A ( s n ) b_n\in\mathcal{A}(s_n) b n ∈ A ( s n ) and s n ( D i ) = h ( D i ) = 1 s_n(D_i)=h(D_i)=1 s n ( D i ) = h ( D i ) = 1 , we have that b n ⊆ ∗ D i b_n\subseteq^{*}D_i b n ⊆ ∗ D i , so compatibility ensures that ∣ b n ∩ C i ∩ D s n ∣ = ω |b_n\cap C_i\cap\mathcal{D}^{s_n}|=\omega ∣ b n ∩ C i ∩ D s n ∣ = ω . Thus we can pick l ∈ ( b n ∩ C i ∩ D s n ) ∖ ⋃ j < n a j l\in(b_n\cap C_i\cap\mathcal{D}^{s_n})\setminus\bigcup_{j<n}a_j l ∈ ( b n ∩ C i ∩ D s n ) ∖ ⋃ j < n a j out of ⋃ m < n F m \bigcup_{m<n}F_m ⋃ m < n F m .
With these witnesses we can build F n F_n F n . Let a ∗ = ⋃ n < ω F n a_*=\bigcup_{n<\omega}F_n a ∗ = ⋃ n < ω F n . The F n F_n F n are pairwise disjoint and each meets X X X , so ∣ a ∗ ∩ X ∣ = ω |a_*\cap X|=\omega ∣ a ∗ ∩ X ∣ = ω . For a fixed j < ω j<\omega j < ω , every F n F_n F n with n > j n>j n > j avoids a j a_j a j , and therefore ∣ a ∗ ∩ a j ∣ < ω |a_*\cap a_j|<\omega ∣ a ∗ ∩ a j ∣ < ω . Thus A ∪ { a ∗ } \mathcal{A}\cup\{a_*\} A ∪ { a ∗ } is almost disjoint.
Fix i < ω i<\omega i < ω . For every n > i n>i n > i , F n ⊆ D i h ( D i ) F_n\subseteq D_i^{h(D_i)} F n ⊆ D i h ( D i ) . If h ( D i ) = 1 h(D_i)=1 h ( D i ) = 1 , then a ∗ ⊆ ∗ D i a_*\subseteq^*D_i a ∗ ⊆ ∗ D i , and item (iii) gives infinitely many distinct points of a ∗ ∩ C i a_*\cap C_i a ∗ ∩ C i . If h ( D i ) = 0 h(D_i)=0 h ( D i ) = 0 , then ∣ a ∗ ∩ D i ∣ < ω |a_*\cap D_i|<\omega ∣ a ∗ ∩ D i ∣ < ω ; since C i ⊆ ∗ D i C_i\subseteq^*D_i C i ⊆ ∗ D i , also ∣ a ∗ ∩ C i ∣ < ω |a_*\cap C_i|<\omega ∣ a ∗ ∩ C i ∣ < ω . Thus every D i D_i D i remains a partitioner for A ∪ { a ∗ } \mathcal{A}\cup\{a_*\} A ∪ { a ∗ } and item (d) holds for a ∗ a_* a ∗ . Lemma 7.5 gives goodness for the extension, so the resulting system is compatible.
We now extend the almost disjoint family to force it to satisfy (10) for a prescribed set C ⊆ ω C\subseteq\omega C ⊆ ω , preserving compatibility with all pairs in a given countable compatible system.
Lemma 7.8. Let ( A , P ) (\mathcal{A},\mathcal{P}) ( A , P ) be a countable compatible system, and let C ⊆ ω C\subseteq\omega C ⊆ ω . Put D = { D : ( C ′ , D ) ∈ P } \mathcal{D}=\{D:(C',D)\in\mathcal{P}\} D = { D : ( C ′ , D ) ∈ P } . There exists a countable almost disjoint family A ′ \mathcal{A}' A ′ extending A \mathcal{A} A such that:
(a) ( A ′ , P ) (\mathcal{A}',\mathcal{P}) ( A ′ , P ) is a countable compatible system;
(b) for every s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) , if ∣ A C ′ ( s ) ∣ < ω |\mathcal{A}'_C(s)|<\omega ∣ A C ′ ( s ) ∣ < ω , then C ∩ D s ⊆ ∗ ⋃ A C ′ ( s ) C\cap D^s\subseteq^*\bigcup\mathcal{A}'_C(s) C ∩ D s ⊆ ∗ ⋃ A C ′ ( s ) .
Proof. Enumerate Fn ( D , 2 ) \operatorname{Fn}(\mathcal{D},2) Fn ( D , 2 ) as ( s k ) k < ω (s_k)_{k<\omega} ( s k ) k < ω in which every function occurs infinitely often. We construct an increasing sequence of countable almost disjoint families
A 0 = A ⊆ A 1 ⊆ ⋯ \mathcal{A}_0=\mathcal{A}\subseteq\mathcal{A}_1\subseteq\cdots A 0 = A ⊆ A 1 ⊆ ⋯
such that ( A k , P ) (\mathcal{A}_k,\mathcal{P}) ( A k , P ) is a countable compatible system for every k < ω k<\omega k < ω .
At stage k < ω k<\omega k < ω , suppose that A k \mathcal{A}_k A k has been constructed. Put
B k = ( A k ) C ( s k ) and X k = ( C ∩ D s k ) ∖ ⋃ B k . \mathcal{B}_k=(\mathcal{A}_k)_C(s_k)
\qquad\text{and}\qquad
X_k=(C\cap D^{s_k})\setminus\bigcup\mathcal{B}_k. B k = ( A k ) C ( s k ) and X k = ( C ∩ D s k ) ∖ ⋃ B k .
If B k \mathcal{B}_k B k is infinite or X k X_k X k is finite, set A k + 1 = A k \mathcal{A}_{k+1}=\mathcal{A}_k A k + 1 = A k . Otherwise, B k \mathcal{B}_k B k is finite and X k X_k X k is infinite. We have that X k X_k X k is almost disjoint from A k \mathcal{A}_k A k and so we can apply Lemma 7.7 to ( A k , P ) (\mathcal{A}_k,\mathcal{P}) ( A k , P ) and X k X_k X k . Let b k b_k b k be the resulting member and set A k + 1 = A k ∪ { b k } \mathcal{A}_{k+1}=\mathcal{A}_k\cup\{b_k\} A k + 1 = A k ∪ { b k } . The extended system is compatible and ∣ b k ∩ X k ∣ = ω |b_k\cap X_k|=\omega ∣ b k ∩ X k ∣ = ω with X k ⊆ C ∩ D s k X_k\subseteq C\cap D^{s_k} X k ⊆ C ∩ D s k . Since every member of D \mathcal{D} D is a partitioner for A k + 1 \mathcal{A}_{k+1} A k + 1 and b k ∈ A k + 1 b_k\in\mathcal{A}_{k+1} b k ∈ A k + 1 , we necessarily have that b k ⊆ ∗ D s k b_k\subseteq^*D^{s_k} b k ⊆ ∗ D s k . Thus b k ∈ ( A k + 1 ) C ( s k ) b_k\in(\mathcal{A}_{k+1})_C(s_k) b k ∈ ( A k + 1 ) C ( s k ) . Also b k ∉ A k b_k\notin\mathcal{A}_k b k ∈ / A k , since X k X_k X k is almost disjoint from A k \mathcal{A}_k A k .
Let A ′ = ⋃ k < ω A k \mathcal{A}'=\bigcup_{k<\omega}\mathcal{A}_k A ′ = ⋃ k < ω A k . This set is countable and almost disjoint. Every D ∈ D D\in\mathcal{D} D ∈ D is a partitioner for A ′ \mathcal{A}' A ′ and the condition in item (d) holds for every member of A ′ \mathcal{A}' A ′ , since each member belongs to some A k \mathcal{A}_k A k . Lemma 7.5, applied to A ⊆ A ′ \mathcal{A}\subseteq\mathcal{A}' A ⊆ A ′ , shows that D \mathcal{D} D is good for A ′ \mathcal{A}' A ′ . Hence ( A ′ , P ) (\mathcal{A}',\mathcal{P}) ( A ′ , P ) is a countable compatible system.
Fix s ∈ Fn ( D , 2 ) s\in\operatorname{Fn}(\mathcal{D},2) s ∈ Fn ( D , 2 ) and suppose that A C ′ ( s ) \mathcal{A}'_C(s) A C ′ ( s ) is finite. Choose k k k with s k = s s_k=s s k = s and A C ′ ( s ) ⊆ A k \mathcal{A}'_C(s)\subseteq\mathcal{A}_k A C ′ ( s ) ⊆ A k . Then B k = A C ′ ( s ) \mathcal{B}_k=\mathcal{A}'_C(s) B k = A C ′ ( s ) . If X k X_k X k were infinite, the construction would add b k ∈ A C ′ ( s ) ∖ A k b_k\in\mathcal{A}'_C(s)\setminus\mathcal{A}_k b k ∈ A C ′ ( s ) ∖ A k , a contradiction. Thus X k = ( C ∩ D s ) ∖ ⋃ A C ′ ( s ) X_k=(C\cap D^s)\setminus\bigcup\mathcal{A}'_C(s) X k = ( C ∩ D s ) ∖ ⋃ A C ′ ( s ) is finite, proving item (b). □ \square □
Combining the previous results gives the following extension property.
Proposition 7.9. Let ( A , P ) (\mathcal{A},\mathcal{P}) ( A , P ) be a countable compatible system and C ⊆ ω C\subseteq\omega C ⊆ ω . There exist a countable almost disjoint extension A ⊆ A ′ \mathcal{A}\subseteq\mathcal{A}' A ⊆ A ′ and a set D ⊆ ω D\subseteq\omega D ⊆ ω with C ⊆ D C\subseteq D C ⊆ D such that ( A ′ , P ∪ { ( C , D ) } ) (\mathcal{A}',\mathcal{P}\cup\{(C,D)\}) ( A ′ , P ∪ {( C , D )}) is a countable compatible system.
Proof. Put D = { D ′ : ( C ′ , D ′ ) ∈ P } \mathcal{D}=\{D':(C',D')\in\mathcal{P}\} D = { D ′ : ( C ′ , D ′ ) ∈ P } . Let A ′ \mathcal{A}' A ′ be the extension given by Lemma 7.8. By item (b) and Lemma 7.4, there is a set D ⊆ ω D\subseteq\omega D ⊆ ω with C ⊆ D C\subseteq D C ⊆ D separating A C ′ \mathcal{A}'_C A C ′ from A ′ ∖ A C ′ \mathcal{A}'\setminus\mathcal{A}'_C A ′ ∖ A C ′ and making D ∪ { D } \mathcal{D}\cup\{D\} D ∪ { D } good for A ′ \mathcal{A}' A ′ . □ \square □
We are now ready to prove Theorem E.
Proof of Theorem E. By CH, fix an enumeration P ( ω ) = { C α : α < ω 1 } \mathcal{P}(\omega)=\{C_\alpha:\alpha<\omega_1\} P ( ω ) = { C α : α < ω 1 } . Let A 0 = { a n : n < ω } \mathcal{A}_0=\{a_n:n<\omega\} A 0 = { a n : n < ω } be a partition of ω \omega ω into infinite sets.
We recursively construct an increasing chain of countable almost disjoint families A α \mathcal{A}_\alpha A α and sets D α ⊆ ω D_\alpha\subseteq\omega D α ⊆ ω satisfying the following.
(i) D α = { D ξ : ξ < α } D_\alpha=\{D_\xi:\xi<\alpha\} D α = { D ξ : ξ < α } is a good family of partitioners for A α \mathcal{A}_\alpha A α ;
(ii) C ξ ⊆ D ξ C_\xi\subseteq D_\xi C ξ ⊆ D ξ for every ξ < α \xi<\alpha ξ < α ;
(iii) for each ξ < α \xi<\alpha ξ < α , ( A α ) C ξ = { a ∈ A α : a ⊆ ∗ D ξ } (\mathcal{A}_\alpha)_{C_\xi}=\{a\in\mathcal{A}_\alpha:a\subseteq^*D_\xi\} ( A α ) C ξ = { a ∈ A α : a ⊆ ∗ D ξ }
Suppose that α < ω 1 \alpha<\omega_1 α < ω 1 and that the construction has been carried out through stage α \alpha α . We show how to extend the construction to stage α + 1 \alpha+1 α + 1 .
Apply Proposition 7.9 to ( A α , { ( C ξ , D ξ ) : ξ < α } ) (\mathcal{A}_\alpha,\{(C_\xi,D_\xi):\xi<\alpha\}) ( A α , {( C ξ , D ξ ) : ξ < α }) and C α C_\alpha C α to obtain a countable almost disjoint family A α + 1 \mathcal{A}_{\alpha+1} A α + 1 extending A α \mathcal{A}_\alpha A α and a set D α ⊇ C α D_\alpha\supseteq C_\alpha D α ⊇ C α such that D α + 1 = D α ∪ { D α } D_{\alpha+1}=D_\alpha\cup\{D_\alpha\} D α + 1 = D α ∪ { D α } is good for A α + 1 \mathcal{A}_{\alpha+1} A α + 1 and, for every ξ ≤ α \xi\leq\alpha ξ ≤ α , ( A α + 1 ) C ξ = { a ∈ A α + 1 : a ⊆ ∗ D ξ } (\mathcal{A}_{\alpha+1})_{C_\xi}=\{a\in\mathcal{A}_{\alpha+1}:a\subseteq^*D_\xi\} ( A α + 1 ) C ξ = { a ∈ A α + 1 : a ⊆ ∗ D ξ } . Thus, all three requirements hold at stage α + 1 \alpha+1 α + 1 .
If γ < ω 1 \gamma<\omega_1 γ < ω 1 is a limit ordinal, define A γ = ⋃ α < γ A α \mathcal{A}_\gamma=\bigcup_{\alpha<\gamma}\mathcal{A}_\alpha A γ = ⋃ α < γ A α . This family is countable and almost disjoint. For each ξ < γ \xi<\gamma ξ < γ , D ξ D_\xi D ξ is a partitioner for A γ \mathcal{A}_\gamma A γ , since every member of A γ \mathcal{A}_\gamma A γ belongs to some A α \mathcal{A}_\alpha A α with ξ < α < γ \xi<\alpha<\gamma ξ < α < γ . Moreover, condition (iii) holds, since
( A γ ) C ξ = { a ∈ A γ : ∣ a ∩ C ξ ∣ = ω } = ⋃ ξ < α < γ { a ∈ A α : ∣ a ∩ C ξ ∣ = ω } = ⋃ ξ < α < γ { a ∈ A α : a ⊆ ∗ D ξ } = { a ∈ A γ : a ⊆ ∗ D ξ } . \begin{aligned}
(\mathcal{A}_\gamma)_{C_\xi}=\{a\in\mathcal{A}_\gamma:|a\cap C_\xi|=\omega\}=\bigcup_{\xi<\alpha<\gamma}\{a\in\mathcal{A}_\alpha:|a\cap C_\xi|=\omega\} \\
&=\bigcup_{\xi<\alpha<\gamma}\{a\in\mathcal{A}_\alpha:a\subseteq^*D_\xi\}=\{a\in\mathcal{A}_\gamma:a\subseteq^*D_\xi\}.
\end{aligned} ( A γ ) C ξ = { a ∈ A γ : ∣ a ∩ C ξ ∣ = ω } = ξ < α < γ ⋃ { a ∈ A α : ∣ a ∩ C ξ ∣ = ω } = ξ < α < γ ⋃ { a ∈ A α : a ⊆ ∗ D ξ } = { a ∈ A γ : a ⊆ ∗ D ξ } .
To see that (i) holds, fix s ∈ Fn ( D γ , 2 ) s\in\operatorname{Fn}(D_\gamma,2) s ∈ Fn ( D γ , 2 ) . Choose β < γ \beta<\gamma β < γ with dom ( s ) ⊆ D β \operatorname{dom}(s)\subseteq D_\beta dom ( s ) ⊆ D β . Since D β D_\beta D β is good for A β \mathcal{A}_\beta A β and every member of D β D_\beta D β is a partitioner for A γ \mathcal{A}_\gamma A γ , Lemma 7.5 implies that D β D_\beta D β is good for A γ \mathcal{A}_\gamma A γ . Hence, if ∣ A γ ( s ) ∣ < ω |\mathcal{A}_\gamma(s)|<\omega ∣ A γ ( s ) ∣ < ω , then D γ s = D β s = ∗ ⋃ A γ ( s ) D_\gamma^s=D_\beta^s=^*\bigcup\mathcal{A}_\gamma(s) D γ s = D β s = ∗ ⋃ A γ ( s ) . This completes the recursive construction.
Set A = ⋃ α < ω 1 A α \mathcal{A}=\bigcup_{\alpha<\omega_1}\mathcal{A}_\alpha A = ⋃ α < ω 1 A α . We claim that A \mathcal{A} A is a MAD family. Indeed, given X ∈ [ ω ] ω X\in[\omega]^\omega X ∈ [ ω ] ω , choose α < ω 1 \alpha<\omega_1 α < ω 1 such that X = C α X=C_\alpha X = C α , and put s = { ( D α , 1 ) } ∈ Fn ( D α + 1 , 2 ) s=\{(D_\alpha,1)\}\in\operatorname{Fn}(D_{\alpha+1},2) s = {( D α , 1 )} ∈ Fn ( D α + 1 , 2 ) . If no member of A α + 1 \mathcal{A}_{\alpha+1} A α + 1 met X X X infinitely, then, by compatibility, A α + 1 ( s ) = ( A α + 1 ) C α = ∅ \mathcal{A}_{\alpha+1}(s)=(\mathcal{A}_{\alpha+1})_{C_\alpha}=\varnothing A α + 1 ( s ) = ( A α + 1 ) C α = ∅ . Goodness applied to s s s , with D α + 1 s = D α D_{\alpha+1}^s=D_\alpha D α + 1 s = D α , would then make D α D_\alpha D α finite, contradicting X ⊆ D α X\subseteq D_\alpha X ⊆ D α . Thus some member of A α + 1 \mathcal{A}_{\alpha+1} A α + 1 , and hence of A \mathcal{A} A , meets X X X infinitely.
Finally, fix C ⊆ ω C\subseteq\omega C ⊆ ω and choose α < ω 1 \alpha<\omega_1 α < ω 1 with C = C α C=C_\alpha C = C α . Given a ∈ A a\in\mathcal{A} a ∈ A , we can pick γ < ω 1 \gamma<\omega_1 γ < ω 1 such that γ > α \gamma>\alpha γ > α and a ∈ A γ a\in\mathcal{A}_\gamma a ∈ A γ . At stage γ \gamma γ , D α D_\alpha D α is a partitioner and ( A γ ) C α = { a ∈ A γ : a ⊆ ∗ D α } (\mathcal{A}_\gamma)_{C_\alpha}=\{a\in\mathcal{A}_\gamma:a\subseteq^*D_\alpha\} ( A γ ) C α = { a ∈ A γ : a ⊆ ∗ D α } gives
a ⊆ ∗ D α ⟺ ∣ a ∩ C ∣ = ω . a\subseteq^*D_\alpha\quad\Longleftrightarrow\quad|a\cap C|=\omega. a ⊆ ∗ D α ⟺ ∣ a ∩ C ∣ = ω .
Since a ∈ A a\in\mathcal{A} a ∈ A was arbitrary, D α D_\alpha D α separates A C \mathcal{A}_C A C from A ∖ A C \mathcal{A}\setminus\mathcal{A}_C A ∖ A C . Proposition 2.5 now implies that Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is almost-normal.
Theorem E gives the consistency, relative to ZFC, of the existence of an almost-normal MAD family. We now turn to its independence.
No almost-normal MAD families under OCAT _T T + a p > ω 1 \mathfrak{ap}>\omega_1 ap > ω 1
In [4 ] , the author proved that the Proper Forcing Axiom (PFA) implies that no almost-normal MAD family exists. We obtain the same conclusion under OCAT + a p > ω 1 _T+\mathfrak{ap}>\omega_1 T + ap > ω 1 , where OCAT _T T is Todorčević’s Open Coloring Axiom, as stated in [14 Definition V.6.25] . The cardinal a p \mathfrak{ap} ap is the least size of an almost disjoint family A \mathcal{A} A with a subset B ⊆ A \mathcal{B}\subseteq\mathcal{A} B ⊆ A for which no X ⊆ ω X \subseteq\omega X ⊆ ω meets every member of B \mathcal{B} B infinitely and every member of A ∖ B \mathcal{A} \setminus\mathcal{B} A ∖ B only finitely [3 p. 14] .
A partially ordered set P \mathbb{P} P is said to be σ \sigma σ -centered if it can be expressed as a countable union of centered sets. Luzin gaps in the sense used here were introduced by Todorčević [24 ] . Given an uncountable almost disjoint family A \mathcal{A} A , one defines its splitting partial order, denoted by P A \mathbb{P}_{\mathcal{A}} P A . We will not define P A \mathbb{P}_{\mathcal{A}} P A nor Luzin gaps in this paper as we will not use these definitions directly. The interested reader may consult e.g. [8 Definition 5 and 12] . We use only the following statements:
If A \mathcal{A} A is MAD, then P A \mathbb{P}_{\mathcal{A}} P A is not σ \sigma σ -centered [8 Theorem 9(2)(b)] .
Under O C A T \mathrm{OCA}_{\mathsf{T}} OCA T , P A \mathbb{P}_{\mathcal{A}} P A is either σ \sigma σ -centered or fails the ccc [8 Theorem 9(3)(b)] .
Under O C A T \mathrm{OCA}_{\mathsf{T}} OCA T , if P A \mathbb{P}_{\mathcal{A}} P A fails the ccc, then A \mathcal{A} A contains a 2-Luzin gap [8 Theorem 9(1)(c)] .
The two sides of a 2-Luzin gap are disjoint subfamilies of cardinality ω 1 \omega_1 ω 1 which cannot be separated [12 p. 1365] .
From this, the following proposition follows immediately.
Proposition 8.1. Assume O C A T \mathrm{OCA}_{\mathsf{T}} OCA T . Every MAD family A \mathcal{A} A contains disjoint subfamilies B , C ∈ [ A ] ℵ 1 \mathcal{B},\mathcal{C} \in[\mathcal{A}]^{\aleph_1} B , C ∈ [ A ] ℵ 1 which cannot be separated.
Proof. If A \mathcal{A} A is MAD, (i) and (ii) imply that P A \mathbb{P}_{\mathcal{A}} P A fails the ccc. Now apply (iii) and (iv).
We can now prove Theorem F.
Proof of Theorem F. Suppose that A \mathcal{A} A is MAD and Ψ ( A ) \Psi(\mathcal{A}) Ψ ( A ) is almost-normal. Fix B , C \mathcal{B},\mathcal{C} B , C as in Proposition 8.1. Since ∣ B ∪ C ∣ = ω 1 < a p |\mathcal{B} \cup\mathcal{C}| = \omega_1 < \mathfrak{ap} ∣ B ∪ C ∣ = ω 1 < ap , there is X ⊆ ω X \subseteq\omega X ⊆ ω which weakly separates B \mathcal{B} B from C \mathcal{C} C . Thus B ⊆ A X \mathcal{B} \subseteq\mathcal{A}_X B ⊆ A X and C ⊆ A ∖ A X \mathcal{C} \subseteq\mathcal{A} \setminus\mathcal{A}_X C ⊆ A ∖ A X . However, by Proposition 2.5 there exists a partitioner separating A X \mathcal{A}_X A X from A ∖ A X \mathcal{A} \setminus\mathcal{A}_X A ∖ A X , and hence separating B \mathcal{B} B from C \mathcal{C} C , a contradiction.
Since p ≤ a p \mathfrak{p} \leq\mathfrak{ap} p ≤ ap [3 p. 14] , the conclusion also holds under O C A T + p > ω 1 \mathrm{OCA}_{\mathsf{T}} + \mathfrak{p} > \omega_1 OCA T + p > ω 1 , and in particular under O C A T + M A ω 1 \mathrm{OCA}_{\mathsf{T}} + \mathrm{MA}_{\omega_1} OCA T + MA ω 1 . As M A + O C A T + c = ℵ 2 \mathrm{MA} + \mathrm{OCA}_{\mathsf{T}} + \mathfrak{c} = \aleph_2 MA + OCA T + c = ℵ 2 is consistent with ZFC [23 ] , the above observation, together with the results of § 7 , shows the following.
Corollary 8.2. The existence of almost-normal MAD families is independent of ZFC.
Acknowledgements
The first author was financed, in part, by the São Paulo Research Foundation (FAPESP), Brasil. Process Number 2025/07302-0.
The second author was financed, in part, by the São Paulo Research Foundation (FAPESP), Brasil. Process Number 2025/07302-0.
The third author was financed, in part, by the São Paulo Research Foundation (FAPESP), Brasil. Process Number 2025/08683-7.
Use of artificial intelligence
The main mathematical results presented in this paper had the first draft of their proofs generated by OpenAI’s GPT-6 Astra. The authors carefully reviewed and verified the correctness of these proofs, rewriting them as necessary to ensure their accuracy and clarity. OpenAI’s GPT-6.1 Sol and OpenAI’s GPT-6.1 Terra aided the authors with subsequent revisions of the proofs, grammar, and LaTeX formatting.