Introduction
Power spaces form one of the central objects of domain theory, because they supply semantic models for programming languages endowed with nondeterministic choice [1 , 2 , 3 ] . Classically, the upper power space of a space X X X is constructed from the nonempty compact saturated subsets of X X X , and it is again a domain whenever X X X is a continuous domain; the construction is then well behaved both algebraically and order-theoretically. When X X X is an arbitrary topological space, however, compact saturated sets no longer retain these properties, and the hypotheses underlying the construction must be identified individually. The hypotheses that prove to be relevant are refinement properties of compact sets with respect to open covers, and they can be formulated without any reference to order theory.
The prototype of such a property is the following question. Let X X X be a topological space, let K ⊆ X K \subseteq X K ⊆ X be compact and let U 1 , U 2 ⊆ X U_1,U_2 \subseteq X U 1 , U 2 ⊆ X be open with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . In what sense can K K K be decomposed into two parts L 1 ⊆ U 1 L_1 \subseteq U_1 L 1 ⊆ U 1 and L 2 ⊆ U 2 L_2 \subseteq U_2 L 2 ⊆ U 2 subject to a prescribed smallness requirement? If the pieces are required to be compact, one speaks of the Wilker property W; if they are only required to be Lindelöf, of the weak Wilker property W w W_{\mathrm{w}} W w . In both properties the decomposition is effected by subsets of the ambient space. If, in addition, the pieces are required to lie inside K K K , one obtains the inner weak Wilker property W i W_{\mathrm{i}} W i and the inner compact refinement property W i n W_{\mathrm{in}} W in ; in that case the requirement K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 forces K = L 1 ∪ L 2 K=L_1\cup L_2 K = L 1 ∪ L 2 , so that the decomposition is a genuine decomposition of K K K itself. The terminology goes back to the study of Wilker spaces and of the sobriety of function spaces carried out in [4 ] , where the four properties were introduced and compared for the first time.
The interest in these properties is not merely combinatorial. They are equivalent to continuity statements for the binary union operation on lattices of open sets, namely
μ : O L ( X ) × O L ( X ) ⟶ O k ( X ) , μ ( U , V ) = U ∪ V , \mu:\mathcal{O}_{L}(X)\times\mathcal{O}_{L}(X)\longrightarrow\mathcal{O}_{k}(X),\qquad\mu(U,V)=U\cup V, μ : O L ( X ) × O L ( X ) ⟶ O k ( X ) , μ ( U , V ) = U ∪ V ,
where O L ( X ) \mathcal{O}_{L}(X) O L ( X ) and O k ( X ) \mathcal{O}_{k}(X) O k ( X ) carry the topologies generated by the sets Φ ( L ) = { U ∈ O ( X ) : L ⊆ U } \Phi(L)=\{U\in\mathcal{O}(X):L\subseteq U\} Φ ( L ) = { U ∈ O ( X ) : L ⊆ U } for L L L Lindelöf and for L L L compact respectively. The weak Wilker property is exactly the continuity of μ \mu μ , and the inner weak Wilker property is exactly the continuity of all the restrictions μ K \mu_K μ K obtained by replacing X X X with a compact subspace K K K [4 ] . Continuity of such operations is in turn the technical heart of the theory of power spaces [1 , 2 , 5 ] , and it is also the key to the sobriety of function spaces with the compact-open topology: the property S introduced in [6 ] , which characterizes sobriety of C k ( X , Y ) C_k(X,Y) C k ( X , Y ) for sober Y Y Y under a mild order-theoretic hypothesis, is a closure condition that can only be verified through precisely this kind of refinement argument.
Three groups of questions arising from this circle of ideas remain open, and they constitute the object of the present paper.
Permanence. Which of the four properties are inherited by subspaces, by closed subspaces, by retracts, by topological sums? A property of compact sets that is not inherited by closed subspaces is of limited utility, since closed subspaces arise naturally in the constructions of power spaces.
Separation. Are the four properties distinct? In particular, is W i \mathrm{W}_{i} W i strictly weaker than W \mathrm{W} W , that is, is there a space in which every compact set can be decomposed into Lindelöf pieces lying inside it, but not into compact pieces? Without such examples the hierarchy collapses and the inner conditions lose their meaning.
Preservation and reflection. Do the properties survive the passage to products, to the power spaces P S ( X ) \mathrm{P}_{S}(X) P S ( X ) and P H ( X ) \mathrm{P}_{H}(X) P H ( X ) , and to the function space C k ( X , S ) \mathrm{C}_{k}(X,S) C k ( X , S ) ? Do they entail sobriety? Questions of this type determine whether the classes W \mathrm{W} W , W w \mathrm{W}_{w} W w , W i \mathrm{W}_{i} W i can serve as a working category of spaces, for instance whether they are reflective in the category of T 0 T_{0} T 0 spaces.
Our answer to the first group is complete for subspaces, retracts and sums. The inner properties W i \mathrm{W}_{i} W i and W i n \mathrm{W}_{in} W in are hereditary for arbitrary subspaces, while W \mathrm{W} W and W w \mathrm{W}_{w} W w are hereditary for closed subspaces only, and we show by example that the latter cannot be improved: there is a locally supercompact space in W ∖ W i \mathrm{W}\setminus\mathrm{W}_{i} W ∖ W i , so W w \mathrm{W}_{w} W w fails to be hereditary for arbitrary subspaces. All four properties are preserved by continuous retracts, and W w \mathrm{W}_{w} W w , W i \mathrm{W}_{i} W i , W i n \mathrm{W}_{in} W in are finitely additive over topological sums. A useful criterion is that, for KC-spaces, the weak and the inner weak Wilker properties coincide, and that a space all of whose compact subspaces are Hausdorff belongs to W i n \mathrm{W}_{in} W in ; both criteria reduce the verification of the inner conditions to compact Hausdorff pieces, where normality is available.
For the second group we use Kovár’s countably infinite compact Fréchet–Urysohn hyperconnected KC-space M 0 M_{0} M 0 [7 ] . Since M 0 M_{0} M 0 is countable, every subset of it is Lindelöf, which gives W i \mathrm{W}_{i} W i ; hyperconnectedness together with the KC property forbids a decomposition into two proper closed pieces, which excludes W \mathrm{W} W . The same argument applies to the space Q + = Q ∪ { ∞ } \mathbb{Q}^{+}=\mathbb{Q}\cup\{\infty\} Q + = Q ∪ { ∞ } obtained by adding to Q \mathbb{Q} Q a point whose neighbourhoods are the complements of compact subsets of Q \mathbb{Q} Q . Thus W i \mathrm{W}_{i} W i is strictly weaker than W \mathrm{W} W , and the two properties are separated by countably infinite compact KC-spaces.
The third group presents the principal difficulties, and our results in this direction are predominantly negative. Compactness is not preserved by products in this context, and we exploit this in two ways. First, following the cofinitе modification of compact T 1 T_{1} T 1 spaces [8 ] , we construct compact T 1 T_{1} T 1 spaces P P P and Q Q Q , modelled on a grid of cardinality ℵ 2 \aleph_{2} ℵ 2 augmented by two points, such that every subspace of P P P and of Q Q Q lies in W i n \mathrm{W}_{in} W in , while P × Q P\times Q P × Q fails even W w \mathrm{W}_{w} W w ; the obstruction arises from a cardinality argument showing that no two Lindelöf pieces cover the grid. Since P ⊔ Q P\sqcup Q P ⊔ Q then belongs to W i n \mathrm{W}_{in} W in but its square does not, the full subcategories determined by W w \mathrm{W}_{w} W w and by W i \mathrm{W}_{i} W i are not reflective in the category of T 0 T_{0} T 0 spaces. Second, we combine the same grid with a hereditarily compact T 1 T_{1} T 1 topology and with a one-point compactification, and obtain a hereditarily compact T 1 T_{1} T 1 space E E E and a compact Hausdorff space Y Y Y such that E × Y ∈ W w ∖ W i E\times Y\in\mathrm{W}_{w}\setminus\mathrm{W}_{i} E × Y ∈ W w ∖ W i ; this shows that the product theorem for W w \mathrm{W}_{w} W w is sharp, and that W i \mathrm{W}_{i} W i is not preserved by products even when both factors satisfy strong compactness conditions. We also show that the Vietoris constructions do not preserve W i \mathrm{W}_{i} W i : the upper Vietoris space P S ( Z ) \mathrm{P}_{S}(Z) P S ( Z ) and the lower Vietoris space P H ( Z ) \mathrm{P}_{H}(Z) P H ( Z ) of a compact T 1 T_{1} T 1 space Z ∈ W i n Z\in\mathrm{W}_{in} Z ∈ W in may both fail W w \mathrm{W}_{w} W w , and this failure persists over locally compact Hausdorff spaces. Finally, we prove that O k ( M 0 ) ≅ C k ( M 0 , S ) \mathcal{O}_{k}(M_{0})\cong\mathrm{C}_{k}(M_{0},S) O k ( M 0 ) ≅ C k ( M 0 , S ) is not sober, and that C k ( M 0 , S ) \mathrm{C}_{k}(M_{0},S) C k ( M 0 , S ) fails property S S S ; this gives a concrete counterexample in the theory of sobriety of function spaces, and it shows that the sobriety hypothesis in the characterization given in [6 ] cannot be omitted.
Taken together, the results show that the four Wilker-type properties behave like a genuine hierarchy of topological invariants: they are stable under the constructions that preserve compactness, they collapse over KC-spaces, and they are separated by explicit countable compact examples. The counterexamples indicate that the obstructions are of a cardinality-theoretic rather than of an order-theoretic nature, and this is the point of view adopted throughout.
Preliminaries
We collect in this section the notions and the elementary facts that are used repeatedly in the sequel. For the standard material on topology we refer to Engelking [9 ] , and for domain theory and the theory of continuous lattices to Gierz et al. [2 ] and Abramsky and Jung [10 ] . Throughout, a space is a topological space and maps are continuous; we write O ( X ) \mathcal{O}(X) O ( X ) for the family of all open subsets of X X X and cl A \operatorname{cl}A cl A , int A \operatorname{int}A int A for the closure and the interior of A ⊆ X A\subseteq X A ⊆ X .
Separation, specialization and saturation
Definition 2.1. A space X X X is T 0 T_0 T 0 if distinct points have distinct closures, and T 1 T_1 T 1 if every singleton is closed. On a T 0 T_0 T 0 space the specialization order is defined by
x ≤ y ⟺ x ∈ cl { y } . x \leq y \quad\Longleftrightarrow\quad x \in\operatorname{cl}\{y\}. x ≤ y ⟺ x ∈ cl { y } . It is a partial order, and every open set is upward closed with respect to it. A subset A ⊆ X A \subseteq X A ⊆ X is saturated if A = ↑ A A = \mathord{\uparrow} A A = ↑ A , where ↑ A = { y : ∃ a ∈ A , a ≤ y } \mathord{\uparrow} A = \{y : \exists a \in A,\ a \leq y\} ↑ A = { y : ∃ a ∈ A , a ≤ y } ; equivalently, A A A is an intersection of open sets. The saturation of an arbitrary subset A ⊆ X A \subseteq X A ⊆ X is
sat A = ⋂ { U ∈ O ( X ) : A ⊆ U } , \operatorname{sat} A = \bigcap\{U \in\mathcal{O}(X) : A \subseteq U\}, sat A = ⋂ { U ∈ O ( X ) : A ⊆ U } , so that sat A \operatorname{sat} A sat A is the smallest saturated set containing A A A .
Thus a subset of a T 0 T_0 T 0 space is saturated exactly when it is an intersection of open sets, and sat A \operatorname{sat} A sat A is saturated for every A A A . In a T 1 T_1 T 1 space the specialization order is discrete and all subsets are saturated, so that saturation carries no information; it becomes meaningful only for T 0 T_0 T 0 spaces that are not T 1 T_1 T 1 , which is precisely the situation in the study of power spaces.
Definition 2.2. Let X X X be a space.
X X X is Hausdorff if distinct points have disjoint open neighbourhoods, and normal if disjoint closed sets have disjoint open neighbourhoods.
X X X is KC if every compact subset of X X X is closed.
X X X is hyperconnected if any two nonempty open subsets of X X X intersect.
X X X is Fréchet–Urysohn if for every A ⊆ X A \subseteq X A ⊆ X and every x ∈ cl A x \in\operatorname{cl} A x ∈ cl A there is a sequence ( a n ) (a_n) ( a n ) in A A A converging to $x.
Every Hausdorff space is KC, and every KC space is T 1 T_1 T 1 ; neither implication reverses. In a hyperconnected space the complement of a proper closed set is a nonempty open set, so the union of two proper closed sets is proper. This remark is used in §3 to exclude the Wilker property.
Compact and Lindelöf subsets
Definition 2.3. A subset K K K of a space X X X is compact if every cover of K K K by open subsets of X X X has a finite subcover, and Lindelöf if every such cover has a countable subcover. We write K ( X ) \mathcal{K}(X) K ( X ) for the family of all nonempty compact subsets of X X X and L ( X ) \mathcal{L}(X) L ( X ) for the family of all nonempty Lindelöf subsets of X X X . For A ⊆ X A \subseteq X A ⊆ X we put L ( A ) = { L ∈ L ( X ) : L ⊆ A } \mathcal{L}(A) = \{L \in\mathcal{L}(X) : L \subseteq A\} L ( A ) = { L ∈ L ( X ) : L ⊆ A } and K ( A ) = { K ∈ K ( X ) : K ⊆ A } \mathcal{K}(A) = \{K \in\mathcal{K}(X) : K \subseteq A\} K ( A ) = { K ∈ K ( X ) : K ⊆ A } , so that L ( A ) \mathcal{L}(A) L ( A ) and K ( A ) \mathcal{K}(A) K ( A ) are the families of nonempty Lindelöf, respectively compact, subsets of the subspace A A A .
Compactness and Lindelöfness are intrinsic properties of the subspace K K K , but it is convenient to test them with covers by open sets of the ambient space, as in Definition 2.3; the two readings agree. The following permanence facts will be used without further mention.
Lemma 2.4. Let X X X be a space and let B ⊆ X B \subseteq X B ⊆ X .
The union of finitely many compact subsets of X X X is compact, and the union of countably many Lindelöf subsets of X X X is Lindelöf.
If K ⊆ B K \subseteq B K ⊆ B is compact in the subspace B B B , then K K K is compact in X X X ; conversely, if K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and K ⊆ B K \subseteq B K ⊆ B , then K K K is compact in the subspace B B B . The same two statements hold with “compact” replaced by “Lindelöf”.
If B B B is closed in X X X and L ∈ L ( X ) L \in\mathcal{L}(X) L ∈ L ( X ) , then L ∩ B L \cap B L ∩ B is Lindelöf as a subspace of L L L , hence L ∩ B ∈ L ( X ) L \cap B \in\mathcal{L}(X) L ∩ B ∈ L ( X ) whenever L ∩ B ≠ ∅ L \cap B \ne\varnothing L ∩ B = ∅ .
If f : X → Y f : X \to Y f : X → Y is continuous, then f ( K ) ∈ K ( Y ) f(K) \in\mathcal{K}(Y) f ( K ) ∈ K ( Y ) for every K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and f ( L ) ∈ L ( Y ) f(L) \in\mathcal{L}(Y) f ( L ) ∈ L ( Y ) for every L ∈ L ( X ) L \in\mathcal{L}(X) L ∈ L ( X ) .
sat L \operatorname{sat} L sat L is Lindelöf for every L ∈ L ( X ) L \in\mathcal{L}(X) L ∈ L ( X ) , and Φ ( L ) = Φ ( sat L ) \Phi(L) = \Phi(\operatorname{sat} L) Φ ( L ) = Φ ( sat L ) for every L ⊆ X L \subseteq X L ⊆ X , where Φ ( E ) = { U ∈ O ( X ) : E ⊆ U } \Phi(E) = \{U \in\mathcal{O}(X) : E \subseteq U\} Φ ( E ) = { U ∈ O ( X ) : E ⊆ U } .
Proof. (i) is standard. For (ii), let W \mathcal{W} W be a cover of K K K by sets open in B B B , say W = B ∩ U W W = B \cap U_W W = B ∩ U W with U W ∈ O ( X ) U_W \in\mathcal{O}(X) U W ∈ O ( X ) . Since K K K is compact in B B B , finitely many members cover K K K , and these are open in X X X ; this gives the first claim. Conversely, if V \mathcal{V} V is a cover of K K K by sets open in B B B and K K K is compact in X X X , choose for each V ∈ V V \in\mathcal{V} V ∈ V an open set U V U_V U V of X X X with V = B ∩ U V V = B \cap U_V V = B ∩ U V ; the family { U V } \{U_V\} { U V } covers K K K , so finitely many of them, say U V 1 , … , U V n U_{V_1},\ldots,U_{V_n} U V 1 , … , U V n , cover K K K , and then V 1 , … , V n V_1,\ldots,V_n V 1 , … , V n cover K K K . The Lindelöf version is proved in the same way, with “finite” replaced by “countable”.
(iii) The set L ∩ B L \cap B L ∩ B is closed in the subspace L L L , because B B B is closed in X X X ; a closed subset of a Lindelöf space is Lindelöf, and Lindelöfness is intrinsic by (ii).
(iv) is the standard fact that continuous images of compact, respectively Lindelöf, spaces are compact, respectively Lindelöf.
(v) Let G \mathcal{G} G be a cover of sat L \operatorname{sat} L sat L by open sets. It covers L L L , so countably many members G 1 , G 2 , … G_1,G_2,\ldots G 1 , G 2 , … cover L L L ; their union is an open set containing L L L , hence it contains sat L \operatorname{sat} L sat L by the definition of saturation. Thus G 1 , G 2 , … G_1,G_2,\ldots G 1 , G 2 , … cover sat L \operatorname{sat} L sat L . Finally, L ⊆ sat L L \subseteq\operatorname{sat} L L ⊆ sat L gives Φ ( sat L ) ⊆ Φ ( L ) \Phi(\operatorname{sat} L) \subseteq\Phi(L) Φ ( sat L ) ⊆ Φ ( L ) , and if L ⊆ U L \subseteq U L ⊆ U with U U U open, then sat L ⊆ U \operatorname{sat} L \subseteq U sat L ⊆ U , so Φ ( L ) ⊆ Φ ( sat L ) \Phi(L) \subseteq\Phi(\operatorname{sat} L) Φ ( L ) ⊆ Φ ( sat L ) . □
Power spaces and function spaces
For a space X X X we consider the hyperspaces of closed, respectively of compact saturated, subsets of X X X :
P H ( X ) = { F ⊆ X : F = cl F , F ≠ ∅ } , \mathcal{P}_{H}(X)=\{F\subseteq X:F=\operatorname{cl}F,\ F\ne\varnothing\}, P H ( X ) = { F ⊆ X : F = cl F , F = ∅ } ,
P S ( X ) = { C ∈ K ( X ) : C = sat C } . \mathcal{P}_{S}(X)=\{C\in\mathcal{K}(X):C=\operatorname{sat}C\}. P S ( X ) = { C ∈ K ( X ) : C = sat C } .
For U ∈ O ( X ) U\in\mathcal{O}(X) U ∈ O ( X ) put
⟨ U ⟩ = { F ∈ P H ( X ) : F ∩ U ≠ ∅ } , \langle U\rangle=\{F\in\mathcal{P}_{H}(X):F\cap U\ne\varnothing\}, ⟨ U ⟩ = { F ∈ P H ( X ) : F ∩ U = ∅ } ,
□ U = { C ∈ P S ( X ) : C ⊆ U } . \Box U=\{C\in\mathcal{P}_{S}(X):C\subseteq U\}. □ U = { C ∈ P S ( X ) : C ⊆ U } .
Definition 2.5. The family { ⟨ U ⟩ : U ∈ O ( X ) } \{\langle U\rangle:U\in\mathcal{O}(X)\} {⟨ U ⟩ : U ∈ O ( X )} is a subbase for the lower Vietoris topology on P H ( X ) \mathcal{P}_{H}(X) P H ( X ) , and the family { □ U : U ∈ O ( X ) } \{\Box U:U\in\mathcal{O}(X)\} { □ U : U ∈ O ( X )} is a base for the upper Vietoris topology on P S ( X ) \mathcal{P}_{S}(X) P S ( X ) [2 , 5 ] . The resulting spaces are denoted by P H ( X ) \mathcal{P}_{H}(X) P H ( X ) and P S ( X ) \mathcal{P}_{S}(X) P S ( X ) as well. For spaces X , Y X,Y X , Y we write C k ( X , Y ) C_k(X,Y) C k ( X , Y ) for the set of all continuous maps f : X → Y f:X\to Y f : X → Y endowed with the compact-open topology [11 ] , a subbase of which is given by
[ K , V ] = { f ∈ C ( X , Y ) : f ( K ) ⊆ V } , [K,V]=\{f\in C(X,Y):f(K)\subseteq V\}, [ K , V ] = { f ∈ C ( X , Y ) : f ( K ) ⊆ V } , K ∈ K ( X ) K\in\mathcal{K}(X) K ∈ K ( X ) , V ∈ O ( Y ) V\in\mathcal{O}(Y) V ∈ O ( Y ) .
The Sierpiński space is S = { 0 , 1 } S=\{0,1\} S = { 0 , 1 } with { 1 } \{1\} { 1 } as its only nonempty proper open set; it is the space of truth values, and O ( X ) \mathcal{O}(X) O ( X ) is in bijection with C ( X , S ) C(X,S) C ( X , S ) through the characteristic functions χ U \chi_U χ U .
Note that the members of P S ( X ) \mathcal{P}_{S}(X) P S ( X ) are compact by definition, a property on which the utility of the upper Vietoris topology rests: the sets □ U \Box U □ U form a base encoding the relation of inclusion.
Definition 2.6. A nonempty subset F F F of a space T T T is irreducible if F ∩ G i ≠ ∅ F\cap G_i\ne\varnothing F ∩ G i = ∅ for i = 1 , 2 i=1,2 i = 1 , 2 with G 1 , G 2 ∈ O ( T ) G_1,G_2\in\mathcal{O}(T) G 1 , G 2 ∈ O ( T ) implies F ∩ G 1 ∩ G 2 ≠ ∅ F\cap G_1\cap G_2\ne\varnothing F ∩ G 1 ∩ G 2 = ∅ . A system of irreducible sets is a family H H H that assigns to each space T T T a collection H ( T ) H(T) H ( T ) of irreducible subsets of T T T containing all singletons and closed under continuous images. A point t ∈ T t\in T t ∈ T is a generic point of F ∈ H ( T ) F\in H(T) F ∈ H ( T ) if cl F = cl { t } \operatorname{cl}F=\operatorname{cl}\{t\} cl F = cl { t } , and T T T is H H H -sober if every F ∈ H ( T ) F\in H(T) F ∈ H ( T ) has a unique generic point. Taking for H H H the system of all irreducible sets one recovers the usual notion of sobriety [12 ] .
Definition 2.7. Let X , Y X,Y X , Y be spaces and let τ \tau τ be a topology on C ( X , Y ) C(X,Y) C ( X , Y ) whose specialization order coincides with the pointwise order. Following [6 ] , we say that τ \tau τ satisfies property S S S if for every F ∈ H ( C ( X , Y ) , τ ) F\in H(C(X,Y),\tau) F ∈ H ( C ( X , Y ) , τ ) and every g ∈ C ( X , Y ) g\in C(X,Y) g ∈ C ( X , Y ) the identities
g − 1 ( V ) = ⋃ f ∈ F f − 1 ( V ) for all V ∈ O ( Y ) (1) g^{-1}(V)=\bigcup_{f\in F}f^{-1}(V)\quad\text{for all }V\in\mathcal{O}(Y) \tag*{(1)} g − 1 ( V ) = f ∈ F ⋃ f − 1 ( V ) for all V ∈ O ( Y ) (1) imply g ∈ cl τ F g\in\operatorname{cl}_{\tau}F g ∈ cl τ F .
Thus property S S S is a closure test: a function whose open preimages are unions of the corresponding preimages of the members of an irreducible family must lie in the closure of that family. For sober Y Y Y and for topologies finer than the pointwise one, this test is equivalent to sobriety of ( C ( X , Y ) , τ ) (C(X,Y),\tau) ( C ( X , Y ) , τ ) [6 ] ; we return to this equivalence in § 4 .
Core compactness and supercompactness
Definition 2.8. Let U U U , V V V be open subsets of a space X X X . We say that V V V is way below U U U , and write V ≪ U V \ll U V ≪ U , if every cover of U U U by open sets contains a finite subfamily covering V V V . The space X X X is core compact if for every U ∈ O ( X ) U \in\mathcal{O}(X) U ∈ O ( X ) and every x ∈ U x \in U x ∈ U there is V ∈ O ( X ) V \in\mathcal{O}(X) V ∈ O ( X ) with x ∈ V ≪ U x \in V \ll U x ∈ V ≪ U . A space is supercompact if every open cover of it contains a member that covers the whole space, and locally supercompact if every point has a supercompact open neighbourhood.
The relation ≪ \ll ≪ is transitive and stable under finite unions: if V 1 ≪ U V_1 \ll U V 1 ≪ U and V 2 ≪ U V_2 \ll U V 2 ≪ U , then V 1 ∪ V 2 ≪ U V_1 \cup V_2 \ll U V 1 ∪ V 2 ≪ U , because a finite subcover for V 1 V_1 V 1 together with a finite subcover for V 2 V_2 V 2 covers V 1 ∪ V 2 V_1 \cup V_2 V 1 ∪ V 2 . Core compactness can therefore be read as an approximation property: every open set is the union of the open sets way below it. It is equivalent to the continuity of the lattice O ( X ) \mathcal{O}(X) O ( X ) [2 , 13 , 14 ] , a fact we use in § 3 .
Definition 2.9. A space X X X is coherent if the intersection of two compact saturated subsets of X X X is again compact, and co-sober if every nonempty compact saturated set that is not the union of two proper compact saturated subsets is the saturation of some point [4 ] .
Coherence is the order-theoretic counterpart of the requirement that the family of compact saturated sets be closed under binary intersections, while co-sobriety is the dual of sobriety, obtained by replacing irreducible closed sets with compact saturated sets that are not unions of two proper such sets.
The lattice of open sets as a topological space
For E ⊆ X E \subseteq X E ⊆ X put Φ ( E ) = { U ∈ O ( X ) : E ⊆ U } \Phi(E) = \{U \in\mathcal{O}(X) : E \subseteq U\} Φ ( E ) = { U ∈ O ( X ) : E ⊆ U } and consider the families
B k = { Φ ( K ) : K ∈ K ( X ) } , B L = { Φ ( L ) : L ∈ L ( X ) } . \mathcal{B}_{k} = \{\Phi(K) : K \in\mathcal{K}(X)\}, \qquad\mathcal{B}_{L} = \{\Phi(L) : L \in\mathcal{L}(X)\}. B k = { Φ ( K ) : K ∈ K ( X )} , B L = { Φ ( L ) : L ∈ L ( X )} .
The members of B k \mathcal{B}_{k} B k and of B L \mathcal{B}_{L} B L are the “upper” sets determined by compact, respectively Lindelöf, tests. By Lemma 2.4(i) the union of two compact sets is compact and the union of two Lindelöf sets is Lindelöf, so that both families are closed under finite intersections; this is the content of the next lemma, which also records the specialization order of the resulting spaces.
Lemma 2.10. Let X X X be a space and E E E , F ⊆ X F \subseteq X F ⊆ X . Then Φ ( E ) ∩ Φ ( F ) = Φ ( E ∪ F ) \Phi(E) \cap\Phi(F) = \Phi(E \cup F) Φ ( E ) ∩ Φ ( F ) = Φ ( E ∪ F ) . Consequently B k \mathcal{B}_{k} B k and B L \mathcal{B}_{L} B L are bases for topologies on O ( X ) \mathcal{O}(X) O ( X ) ; the resulting spaces are denoted by O k ( X ) \mathcal{O}_{k}(X) O k ( X ) and O L ( X ) \mathcal{O}_{L}(X) O L ( X ) . In both of them the specialization order is inclusion, and cl { U } = { V ∈ O ( X ) : V ⊆ U } \operatorname{cl}\{U\} = \{V \in\mathcal{O}(X) : V \subseteq U\} cl { U } = { V ∈ O ( X ) : V ⊆ U } for every U ∈ O ( X ) U \in\mathcal{O}(X) U ∈ O ( X ) .
Proof. For W ∈ O ( X ) W \in\mathcal{O}(X) W ∈ O ( X ) we have W ∈ Φ ( E ) ∩ Φ ( F ) W \in\Phi(E) \cap\Phi(F) W ∈ Φ ( E ) ∩ Φ ( F ) if and only if E ⊆ W E \subseteq W E ⊆ W and F ⊆ W F \subseteq W F ⊆ W , that is, if and only if E ∪ F ⊆ W E \cup F \subseteq W E ∪ F ⊆ W ; this proves the identity. Since Φ ( ∅ ) = O ( X ) \Phi(\varnothing) = \mathcal{O}(X) Φ ( ∅ ) = O ( X ) , the members of B k \mathcal{B}_{k} B k cover O ( X ) \mathcal{O}(X) O ( X ) , and the same holds for B L \mathcal{B}_{L} B L . If K 1 , K 2 ∈ K ( X ) K_1, K_2 \in\mathcal{K}(X) K 1 , K 2 ∈ K ( X ) , then Φ ( K 1 ) ∩ Φ ( K 2 ) = Φ ( K 1 ∪ K 2 ) ∈ B k \Phi(K_1) \cap\Phi(K_2) = \Phi(K_1 \cup K_2) \in\mathcal{B}_{k} Φ ( K 1 ) ∩ Φ ( K 2 ) = Φ ( K 1 ∪ K 2 ) ∈ B k because K 1 ∪ K 2 K_1 \cup K_2 K 1 ∪ K 2 is compact, so B k \mathcal{B}_{k} B k is closed under finite intersections and hence a base; the argument for B L \mathcal{B}_{L} B L is the same, with Lemma 2.4(i) replacing compactness.
For the specialization order, let U , V ∈ O ( X ) U,V \in\mathcal{O}(X) U , V ∈ O ( X ) . A basic neighbourhood of U U U is a set Φ ( K ) \Phi(K) Φ ( K ) with K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and K ⊆ U K \subseteq U K ⊆ U , and Φ ( K ) \Phi(K) Φ ( K ) contains V V V exactly when K ⊆ V K \subseteq V K ⊆ V . Hence U ∈ cl { V } U \in\operatorname{cl}\{V\} U ∈ cl { V } if and only if K ⊆ V K \subseteq V K ⊆ V for every compact K ⊆ U K \subseteq U K ⊆ U . Taking K = { x } K = \{x\} K = { x } for x ∈ U x \in U x ∈ U shows that U ⊆ V U \subseteq V U ⊆ V is necessary; it is also sufficient, since K ⊆ U ⊆ V K \subseteq U \subseteq V K ⊆ U ⊆ V for every K ⊆ U K \subseteq U K ⊆ U . Thus U ≤ V U \leq V U ≤ V if and only if U ⊆ V U \subseteq V U ⊆ V , and the computation also gives cl { U } = { V : V ⊆ U } \operatorname{cl}\{U\} = \{V : V \subseteq U\} cl { U } = { V : V ⊆ U } . □ \square □
The two spaces O k ( X ) \mathcal{O}_{k}(X) O k ( X ) and O L ( X ) \mathcal{O}_{L}(X) O L ( X ) carry the same underlying set O ( X ) \mathcal{O}(X) O ( X ) but different topologies, and O L ( X ) \mathcal{O}_{L}(X) O L ( X ) has the finer one because K ( X ) ⊆ L ( X ) \mathcal{K}(X) \subseteq\mathcal{L}(X) K ( X ) ⊆ L ( X ) . They are the spaces on which the union operation of the introduction is defined; we write μ \mu μ for the binary operation μ ( U , V ) = U ∪ V \mu(U,V) = U \cup V μ ( U , V ) = U ∪ V and, for K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , for its restriction μ K : O L ( K ) × O L ( K ) → O k ( K ) \mu_{K} : \mathcal{O}_{L}(K) \times\mathcal{O}_{L}(K) \to\mathcal{O}_{k}(K) μ K : O L ( K ) × O L ( K ) → O k ( K ) .
Wilker-Type Properties and Their Heredity
The four properties studied in this paper differ only in the class of sets that is allowed to witness a decomposition of a compact set along a binary open cover. We first give the unified definition, then describe the elementary relations between the four properties, and finally study their behaviour under subspaces, retracts, sums and products.
Definition 3.1. Let X X X be a space. For K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and U 1 , U 2 ∈ O ( X ) U_1, U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 consider the condition
( ∃ L 1 , L 2 ∈ C ) K ⊆ L 1 ∪ L 2 and L j ⊆ U j ( j = 1 , 2 ) . (2) (\exists L_1,L_2 \in\mathcal{C})\quad K \subseteq L_1 \cup L_2 \quad\text{and}\quad L_j \subseteq U_j\ (j = 1,2). \tag*{(2)} ( ∃ L 1 , L 2 ∈ C ) K ⊆ L 1 ∪ L 2 and L j ⊆ U j ( j = 1 , 2 ) . (2)
If (2) holds for all such K , U 1 , U 2 K,U_1,U_2 K , U 1 , U 2 , then, according to the choice of the class C \mathcal{C} C , the space X X X is said to have the property listed in Table 1 . The corresponding classes of spaces are denoted by W W W , W ω W_{\omega} W ω , W i W_i W i and W i n W_{\mathrm{in}} W in .
Table 1. The four Wilker-type properties, according to the class C C C of admissible witnesses in condition (2).
When C = L ( K ) C = \mathcal{L}(K) C = L ( K ) or C = K ( K ) C = \mathcal{K}(K) C = K ( K ) , the witnesses must lie inside K K K , and then the inclusion K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 of (2) is equivalent to the identity K = L 1 ∪ L 2 K = L_1 \cup L_2 K = L 1 ∪ L 2 . This is the reason for the word “inner”: the decomposition is a decomposition of K K K itself and not only of a neighbourhood of K K K .
Remark 3.2. Since every compact set is Lindelöf , W ⊆ W w W \subseteq W_{\mathrm{w}} W ⊆ W w , and since L ( K ) ⊆ L ( X ) \mathcal{L}(K) \subseteq\mathcal{L}(X) L ( K ) ⊆ L ( X ) and K ( K ) ⊆ L ( K ) \mathcal{K}(K) \subseteq\mathcal{L}(K) K ( K ) ⊆ L ( K ) , we have
W i n ⊆ W ∩ W i ⊆ W w . W_{\mathrm{in}} \subseteq W \cap W_{\mathrm{i}} \subseteq W_{\mathrm{w}}. W in ⊆ W ∩ W i ⊆ W w .
The properties W W W and W i W_{\mathrm{i}} W i are not comparable: Proposition 3.9 below produces a space in W i ∖ W W_{\mathrm{i}} \setminus W W i ∖ W , and Proposition 4.4 produces a space in W ∖ W i W \setminus W_{\mathrm{i}} W ∖ W i . Hence the only inclusions among the four classes are the two displayed above, together with the inclusion W i n ⊆ W w W_{\mathrm{in}} \subseteq W_{\mathrm{w}} W in ⊆ W w that follows from them.
We now show that W i W_{\mathrm{i}} W i can be tested on compact subspaces only. This is the technical reason why the inner weak Wilker property is much better behaved than W w W_{\mathrm{w}} W w .
Proposition 3.3. For a space X X X the following are equivalent:
X ∈ W i X \in W_{\mathrm{i}} X ∈ W i ;
every subspace B ⊆ X B \subseteq X B ⊆ X belongs to W i W_{\mathrm{i}} W i ;
every subspace B ⊆ X B \subseteq X B ⊆ X belongs to W w W_{\mathrm{w}} W w ;
every compact subspace K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) belongs to W w W_{\mathrm{w}} W w .
Proof. (1) ⇒ \Rightarrow ⇒ (2). Assume X ∈ W i X \in W_{\mathrm{i}} X ∈ W i and let B ⊆ X B \subseteq X B ⊆ X be a subspace. Take K ∈ K ( B ) K \in\mathcal{K}(B) K ∈ K ( B ) and open sets U 1 , U 2 ∈ O ( X ) U_1, U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ ( B ∩ U 1 ) ∪ ( B ∩ U 2 ) K \subseteq(B \cap U_1) \cup(B \cap U_2) K ⊆ ( B ∩ U 1 ) ∪ ( B ∩ U 2 ) ; such a pair exists for every binary open cover of K K K in B B B , since each open set of B B B is of the form B ∩ U B \cap U B ∩ U . Then K K K is compact in X X X by Lemma 2.4(ii), and K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . So W i W_{\mathrm{i}} W i supplies L 1 , L 2 ∈ L ( X ) L_1, L_2 \in\mathcal{L}(X) L 1 , L 2 ∈ L ( X ) with L j ⊆ K ∩ U j L_j \subseteq K \cap U_j L j ⊆ K ∩ U j and K = L 1 ∪ L 2 K = L_1 \cup L_2 K = L 1 ∪ L 2 . From L j ⊆ K ⊆ B L_j \subseteq K \subseteq B L j ⊆ K ⊆ B we obtain L j ⊆ B ∩ U j L_j \subseteq B \cap U_j L j ⊆ B ∩ U j , and L j L_j L j is Lindelöf as a subspace of B B B by Lemma 2.4(ii). Hence B ∈ W i B \in W_{\mathrm{i}} B ∈ W i .
(2) ⇒ \Rightarrow ⇒ (3). This is immediate from W i ⊆ W w W_{\mathrm{i}} \subseteq W_{\mathrm{w}} W i ⊆ W w .
(3) ⇒ \Rightarrow ⇒ (4). A compact subspace is a subspace, so it belongs to W w W_{\mathrm{w}} W w .
(4) ⇒ \Rightarrow ⇒ (1). Assume that every compact subspace of X X X belongs to W w W_{\mathrm{w}} W w , and take K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , U 1 , U 2 ∈ O ( X ) U_1, U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . The sets K ∩ U 1 K \cap U_1 K ∩ U 1 and K ∩ U 2 K \cap U_2 K ∩ U 2 are open in the subspace K K K and cover it, so, since K ∈ W w K \in W_{\mathrm{w}} K ∈ W w , there are L 1 , L 2 ∈ L ( K ) L_1, L_2 \in\mathcal{L}(K) L 1 , L 2 ∈ L ( K ) with L j ⊆ K ∩ U j L_j \subseteq K \cap U_j L j ⊆ K ∩ U j and K = L 1 ∪ L 2 K = L_1 \cup L_2 K = L 1 ∪ L 2 . By Lemma 2.4(ii) each L j L_j L j is Lindelöf as a subspace of X X X , and L j ⊆ U j L_j \subseteq U_j L j ⊆ U j . Hence X ∈ W i X \in W_{\mathrm{i}} X ∈ W i . □
Proposition 3.4. The properties W i W_{\mathrm{i}} W i and W i n W_{\mathrm{in}} W in are hereditary with respect to arbitrary subspaces.
Proof. The statement for W i W_{\mathrm{i}} W i is part of Proposition 3.3. For W i n W_{\mathrm{in}} W in , let X ∈ W i n X \in W_{\mathrm{in}} X ∈ W in , let B ⊆ X B \subseteq X B ⊆ X and let K ∈ K ( B ) K \in\mathcal{K}(B) K ∈ K ( B ) , U 1 , U 2 ∈ O ( X ) U_1, U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ ( B ∩ U 1 ) ∪ ( B ∩ U 2 ) K \subseteq(B \cap U_1) \cup(B \cap U_2) K ⊆ ( B ∩ U 1 ) ∪ ( B ∩ U 2 ) . Then K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 , so W i n W_{\mathrm{in}} W in gives F 1 , F 2 ∈ K ( X ) F_1, F_2 \in\mathcal{K}(X) F 1 , F 2 ∈ K ( X ) with F j ⊆ K ∩ U j F_j \subseteq K \cap U_j F j ⊆ K ∩ U j and K = F 1 ∪ F 2 K = F_1 \cup F_2 K = F 1 ∪ F 2 . Since F j ⊆ K ⊆ B F_j \subseteq K \subseteq B F j ⊆ K ⊆ B , each F j F_j F j is compact in the subspace B B B by Lemma 2.4(ii), and F j ⊆ B ∩ U j F_j \subseteq B \cap U_j F j ⊆ B ∩ U j . Hence B ∈ W i n B \in W_{\mathrm{in}} B ∈ W in . □
The corresponding statement for W W W and W w W_{\mathrm{w}} W w is false for arbitrary subspaces, but it holds for closed ones.
Proposition 3.5. The properties W W W and W w W_{\mathrm{w}} W w are hereditary with respect to closed subspaces.
Proof. Let B B B be a closed subspace of X X X , let X ∈ W w X \in W_{\mathrm{w}} X ∈ W w and let K ∈ K ( B ) K \in\mathcal{K}(B) K ∈ K ( B ) , U 1 , U 2 ∈ O ( X ) U_1, U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ ( B ∩ U 1 ) ∪ ( B ∩ U 2 ) K \subseteq(B \cap U_1) \cup(B \cap U_2) K ⊆ ( B ∩ U 1 ) ∪ ( B ∩ U 2 ) . Then K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 , so there are L 1 , L 2 ∈ L ( X ) L_1, L_2 \in\mathcal{L}(X) L 1 , L 2 ∈ L ( X ) with L j ⊆ U j L_j \subseteq U_j L j ⊆ U j and K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 . Put L j ′ = L j ∩ B L'_j = L_j \cap B L j ′ = L j ∩ B . Since B B B is closed in X X X , the set L j ′ L'_j L j ′ is closed in the Lindelöf space L j L_j L j , hence Lindelöf by Lemma 2.4(ii), and L j ′ ⊆ B ∩ U j L'_j \subseteq B \cap U_j L j ′ ⊆ B ∩ U j . Moreover K ⊆ B K \subseteq B K ⊆ B gives
K = K ∩ ( L 1 ∪ L 2 ) ⊆ ( L 1 ∩ B ) ∪ ( L 2 ∩ B ) = L 1 ′ ∪ L 2 ′ . K = K \cap(L_1 \cup L_2) \subseteq(L_1 \cap B) \cup(L_2 \cap B) = L'_1 \cup L'_2. K = K ∩ ( L 1 ∪ L 2 ) ⊆ ( L 1 ∩ B ) ∪ ( L 2 ∩ B ) = L 1 ′ ∪ L 2 ′ .
Therefore B ∈ W w B \in W_{\mathrm{w}} B ∈ W w . The argument for W W W is the same, with “compact” in place of “Lindelöf”: a closed subset of a compact space is compact. □
That the restriction to closed subspaces in Proposition 3.5 cannot be removed is shown by Proposition 4.4, where a locally supercompact space in W ∖ W i W \setminus W_{\mathrm{i}} W ∖ W i is constructed; its subspace M M M belongs to W W W but not to W w W_{\mathrm{w}} W w , although the ambient space does.
Proposition 3.6. The properties W W W , W w W_{\mathrm{w}} W w and W i W_{\mathrm{i}} W i are preserved by continuous retracts: if e : X → Y e:X\to Y e : X → Y is an embedding and r : Y → X r:Y\to X r : Y → X is continuous map with r ∘ e = id X r\circ e=\operatorname{id}_{X} r ∘ e = id X , then Y ∈ W Y\in W Y ∈ W implies X ∈ W X\in W X ∈ W for each W ∈ { W , W w , W i } W\in\{W,W_{\mathrm{w}},W_{\mathrm{i}}\} W ∈ { W , W w , W i } .
Proof. Let e : X → Y e:X\to Y e : X → Y be an embedding with retraction r r r , so that r ( e ( x ) ) = x r(e(x))=x r ( e ( x )) = x for all x ∈ X x\in X x ∈ X .
Assume first that Y ∈ W w Y\in W_{\mathrm{w}} Y ∈ W w , and take K ∈ K ( X ) K\in\mathcal{K}(X) K ∈ K ( X ) , U 1 , U 2 ∈ O ( X ) U_{1},U_{2}\in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K\subseteq U_{1}\cup U_{2} K ⊆ U 1 ∪ U 2 . Since e e e is continuous, e ( K ) e(K) e ( K ) is compact in Y Y Y ; since r r r is continuous, the sets W j = r − 1 ( U j ) W_{j}=r^{-1}(U_{j}) W j = r − 1 ( U j ) are open in Y Y Y . For k ∈ K k\in K k ∈ K we have r ( e ( k ) ) = k ∈ U 1 ∪ U 2 r(e(k))=k\in U_{1}\cup U_{2} r ( e ( k )) = k ∈ U 1 ∪ U 2 , hence e ( k ) ∈ W 1 ∪ W 2 e(k)\in W_{1}\cup W_{2} e ( k ) ∈ W 1 ∪ W 2 ; therefore e ( K ) ⊆ W 1 ∪ W 2 e(K)\subseteq W_{1}\cup W_{2} e ( K ) ⊆ W 1 ∪ W 2 . As Y ∈ W w Y\in W_{\mathrm{w}} Y ∈ W w , there are L 1 , L 2 ∈ L ( Y ) L_{1},L_{2}\in\mathcal{L}(Y) L 1 , L 2 ∈ L ( Y ) with L j ⊆ W j L_{j}\subseteq W_{j} L j ⊆ W j and e ( K ) ⊆ L 1 ∪ L 2 e(K)\subseteq L_{1}\cup L_{2} e ( K ) ⊆ L 1 ∪ L 2 . Put L j ′ = r ( L j ) L'_{j}=r(L_{j}) L j ′ = r ( L j ) . Then L j ′ L'_{j} L j ′ is Lindelöf, being a continuous image of L j L_{j} L j , and L j ⊆ r − 1 ( U j ) L_{j}\subseteq r^{-1}(U_{j}) L j ⊆ r − 1 ( U j ) gives L j ′ ⊆ U j L'_{j}\subseteq U_{j} L j ′ ⊆ U j . Finally
K = r ( e ( K ) ) ⊆ r ( L 1 ∪ L 2 ) = L 1 ′ ∪ L 2 ′ . K=r(e(K))\subseteq r(L_{1}\cup L_{2})=L'_{1}\cup L'_{2}. K = r ( e ( K )) ⊆ r ( L 1 ∪ L 2 ) = L 1 ′ ∪ L 2 ′ .
Hence X ∈ W w X\in W_{\mathrm{w}} X ∈ W w .
Assume next that Y ∈ W i Y\in W_{\mathrm{i}} Y ∈ W i . With K , U 1 , U 2 , W 1 , W 2 K,U_{1},U_{2},W_{1},W_{2} K , U 1 , U 2 , W 1 , W 2 as above we have e ( K ) ⊆ W 1 ∪ W 2 e(K)\subseteq W_{1}\cup W_{2} e ( K ) ⊆ W 1 ∪ W 2 , so there are L 1 , L 2 ∈ L ( Y ) L_{1},L_{2}\in\mathcal{L}(Y) L 1 , L 2 ∈ L ( Y ) with L j ⊆ e ( K ) ∩ W j L_{j}\subseteq e(K)\cap W_{j} L j ⊆ e ( K ) ∩ W j and e ( K ) = L 1 ∪ L 2 e(K)=L_{1}\cup L_{2} e ( K ) = L 1 ∪ L 2 . Putting L j ′ = r ( L j ) L'_{j}=r(L_{j}) L j ′ = r ( L j ) we obtain, as before, that L j ′ L'_{j} L j ′ is Lindelöf, and now
L j ′ ⊆ r ( e ( K ) ∩ r − 1 ( U j ) ) ⊆ r ( e ( K ) ) ∩ U j = K ∩ U j . L'_{j}\subseteq r(e(K)\cap r^{-1}(U_{j}))\subseteq r(e(K))\cap U_{j}=K\cap U_{j}. L j ′ ⊆ r ( e ( K ) ∩ r − 1 ( U j )) ⊆ r ( e ( K )) ∩ U j = K ∩ U j .
while K = r ( e ( K ) ) = L 1 ′ ∪ L 2 ′ K=r(e(K))=L'_{1}\cup L'_{2} K = r ( e ( K )) = L 1 ′ ∪ L 2 ′ . Hence X ∈ W i X\in W_{\mathrm{i}} X ∈ W i .
The case of W W W is obtained from the first part by reading “compact” for “Lindelöf” everywhere: the continuous image of a compact set is compact. □
Proposition 3.7. Let W ∈ { W w , W i , W i n } W\in\{W_{\mathrm{w}},W_{\mathrm{i}},W_{\mathrm{in}}\} W ∈ { W w , W i , W in } . For a family { X α } α ∈ I \{X_{\alpha}\}_{\alpha\in I} { X α } α ∈ I of nonempty spaces, the topological sum ⨆ α ∈ I X α \bigsqcup_{\alpha\in I}X_{\alpha} ⨆ α ∈ I X α belongs to W W W if and only if X α ∈ W X_{\alpha}\in W X α ∈ W for every α ∈ I \alpha\in I α ∈ I .
Proof. Necessity. If X = ⨆ α X α ∈ W X=\bigsqcup_{\alpha}X_{\alpha}\in W X = ⨆ α X α ∈ W , then each summand X α X_{\alpha} X α is a clopen subspace of X X X . For W = W w W=W_{\mathrm{w}} W = W w we apply Proposition 3.5 (closed subspaces), and for W = W i W=W_{\mathrm{i}} W = W i or W = W i n W=W_{\mathrm{in}} W = W in we apply Proposition 3.4 (arbitrary subspaces); in each case X α ∈ W X_{\alpha}\in W X α ∈ W .
Sufficiency. Assume X α ∈ W X_{\alpha}\in W X α ∈ W for all α \alpha α , and let X = ⨆ α X α X=\bigsqcup_{\alpha}X_{\alpha} X = ⨆ α X α . Let K ∈ K ( X ) K\in\mathcal{K}(X) K ∈ K ( X ) and let U 1 , U 2 ∈ O ( X ) U_{1},U_{2}\in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K\subseteq U_{1}\cup U_{2} K ⊆ U 1 ∪ U 2 . Since { X α } \{X_{\alpha}\} { X α } is a pairwise disjoint open cover of X X X , the compact set K K K meets only finitely many summands; write I K = { α 1 , … , α n } I_{K}=\{\alpha_{1},\ldots,\alpha_{n}\} I K = { α 1 , … , α n } for the finite set of indices with K ∩ X α ≠ ∅ K\cap X_{\alpha}\ne\varnothing K ∩ X α = ∅ . For each α ∈ I K \alpha\in I_{K} α ∈ I K the set K ∩ X α K\cap X_{\alpha} K ∩ X α is compact in the subspace X α X_{\alpha} X α , and
K ∩ X α ⊆ ( X α ∩ U 1 ) ∪ ( X α ∩ U 2 ) . K\cap X_{\alpha}\subseteq(X_{\alpha}\cap U_{1})\cup(X_{\alpha}\cap U_{2}). K ∩ X α ⊆ ( X α ∩ U 1 ) ∪ ( X α ∩ U 2 ) .
Since X α ∈ W X_{\alpha}\in W X α ∈ W , there are sets L 1 , α , L 2 , α L_{1,\alpha},L_{2,\alpha} L 1 , α , L 2 , α in X α X_{\alpha} X α , of the kind required by W W W , with L j , α ⊆ X α ∩ U j L_{j,\alpha}\subseteq X_{\alpha}\cap U_{j} L j , α ⊆ X α ∩ U j and K ∩ X α ⊆ L 1 , α ∪ L 2 , α K\cap X_{\alpha}\subseteq L_{1,\alpha}\cup L_{2,\alpha} K ∩ X α ⊆ L 1 , α ∪ L 2 , α ; when W ∈ { W i , W i n } W\in\{W_{\mathrm{i}},W_{\mathrm{in}}\} W ∈ { W i , W in } we may and do choose them inside K ∩ X α K\cap X_{\alpha} K ∩ X α .
If W = W w W=W_{\mathrm{w}} W = W w , put L j = ⋃ α ∈ I K L j , α L_{j}=\bigcup_{\alpha\in I_{K}}L_{j,\alpha} L j = ⋃ α ∈ I K L j , α . This is a Lindelöf subset of X X X by Lemma 2.4(i), because I K I_{K} I K is finite, and L j ⊆ U j L_{j}\subseteq U_{j} L j ⊆ U j ; moreover K ⊆ L 1 ∪ L 2 K\subseteq L_{1}\cup L_{2} K ⊆ L 1 ∪ L 2 . Hence X ∈ W w X\in W_{\mathrm{w}} X ∈ W w . If W = W i W=W_{\mathrm{i}} W = W i , the sets L j , α L_{j,\alpha} L j , α are Lindelöf and L j = ⋃ α ∈ I K L j , α ⊆ K ∩ U j L_{j}=\bigcup_{\alpha\in I_{K}}L_{j,\alpha}\subseteq K\cap U_{j} L j = ⋃ α ∈ I K L j , α ⊆ K ∩ U j is Lindelöf as a finite union of Lindelöf sets, so X ∈ W i X\in W_{\mathrm{i}} X ∈ W i . If W = W i n W=W_{\mathrm{in}} W = W in , the sets L j , α L_{j,\alpha} L j , α are compact subsets of K ∩ U j K\cap U_{j} K ∩ U j and their finite union is compact, so X ∈ W i n X\in W_{\mathrm{in}} X ∈ W in . □
The next two results are the technical core of the section: they show that over KC-spaces the weak and the inner weak Wilker properties coincide, and that the inner compact refinement property is automatic as soon as all compact subspaces are Hausdorff.
Proposition 3.8. Let X X X be a space.
If X X X is a KC-space, then X ∈ W w X\in W_{\mathrm{w}} X ∈ W w if and only if X ∈ W i X\in W_{\mathrm{i}} X ∈ W i .
If every compact subspace of X X X is Hausdorff, then X ∈ W i n X\in W_{\mathrm{in}} X ∈ W in .
Proof. (i) Let X X X be a KC-space with X ∈ W w X\in W_{\mathrm{w}} X ∈ W w , and take K ∈ K ( X ) K\in\mathcal{K}(X) K ∈ K ( X ) , U 1 , U 2 ∈ O ( X ) U_{1},U_{2}\in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K\subseteq U_{1}\cup U_{2} K ⊆ U 1 ∪ U 2 . By W w W_{\mathrm{w}} W w there are L 1 , L 2 ∈ L ( X ) L_{1},L_{2}\in\mathcal{L}(X) L 1 , L 2 ∈ L ( X ) with L j ⊆ U j L_{j}\subseteq U_{j} L j ⊆ U j and K ⊆ L 1 ∪ L 2 K\subseteq L_{1}\cup L_{2} K ⊆ L 1 ∪ L 2 . Since X X X is KC, the compact set K K K is closed in X X X , so L j ∩ K L_{j}\cap K L j ∩ K is a closed subset of the Lindelöf space L j L_{j} L j and hence Lindelöf. Moreover L j ∩ K ⊆ K ∩ U j L_{j}\cap K\subseteq K\cap U_{j} L j ∩ K ⊆ K ∩ U j , and
K = K ∩ ( L 1 ∪ L 2 ) = ( L 1 ∩ K ) ∪ ( L 2 ∩ K ) . K=K\cap(L_{1}\cup L_{2})=(L_{1}\cap K)\cup(L_{2}\cap K). K = K ∩ ( L 1 ∪ L 2 ) = ( L 1 ∩ K ) ∪ ( L 2 ∩ K ) .
Thus the sets L j ′ = L j ∩ K L'_{j}=L_{j}\cap K L j ′ = L j ∩ K are Lindelöf subsets of K K K contained in K ∩ U j K\cap U_{j} K ∩ U j and they cover K K K , which is exactly X ∈ W i X\in W_{\mathrm{i}} X ∈ W i . The converse is Remark 3.2. (ii) Assume that every compact subspace of X X X is Hausdorff, and take K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , U 1 , U 2 ∈ O ( X ) U_1,U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . Put C j = K ∖ U j C_j = K \setminus U_j C j = K ∖ U j ; these are closed subsets of the compact space K K K , and C 1 ∩ C 2 = ∅ C_1 \cap C_2 = \varnothing C 1 ∩ C 2 = ∅ because U 1 ∪ U 2 U_1 \cup U_2 U 1 ∪ U 2 covers K K K . By hypothesis K K K is compact Hausdorff, hence normal, so there are relatively open sets G 1 , G 2 G_1,G_2 G 1 , G 2 of K K K with C j ⊆ G j C_j \subseteq G_j C j ⊆ G j and G 1 ∩ G 2 = ∅ G_1 \cap G_2 = \varnothing G 1 ∩ G 2 = ∅ . Let F j = K ∖ G j F_j = K \setminus G_j F j = K ∖ G j . Each F j F_j F j is closed in the compact space K K K and therefore compact, and
F j ⊆ K ∖ C j = K ∩ U j , F_j \subseteq K \setminus C_j = K \cap U_j, F j ⊆ K ∖ C j = K ∩ U j ,
F 1 ∪ F 2 = ( K ∖ G 1 ) ∪ ( K ∖ G 2 ) = K ∖ ( G 1 ∩ G 2 ) = K . F_1 \cup F_2 = (K \setminus G_1) \cup(K \setminus G_2) = K \setminus(G_1 \cap G_2) = K. F 1 ∪ F 2 = ( K ∖ G 1 ) ∪ ( K ∖ G 2 ) = K ∖ ( G 1 ∩ G 2 ) = K .
Hence X ∈ W i n X \in W_{\mathrm{in}} X ∈ W in . □
We now separate W i W_i W i from W W W . The examples are countably infinite compact KC-spaces; in a countable space every subset is Lindelöf, so that W i W_i W i holds automatically, and the KC property together with hyperconnectedness prevents a decomposition into two proper closed pieces, which excludes W W W .
Proposition 3.9. There is a countably infinite compact KC-space M 0 M_0 M 0 with M 0 ∈ W i ∖ W M_0 \in W_i \setminus W M 0 ∈ W i ∖ W .
Proof. Let M 0 M_0 M 0 be Kovár’s countably infinite compact Fréchet–Urysohn hyperconnected KC-space [7 Example 2.2 and the following paragraph, p. 6] ; the construction is recalled in [4 before Lemma 3.14] .
M 0 ∈ W i M_0 \in W_i M 0 ∈ W i . Since M 0 M_0 M 0 is countable, every subset of it is countable, and every countable space is Lindelöf. Given K ∈ K ( M 0 ) K \in\mathcal{K}(M_0) K ∈ K ( M 0 ) and U 1 , U 2 ∈ O ( M 0 ) U_1,U_2 \in\mathcal{O}(M_0) U 1 , U 2 ∈ O ( M 0 ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 , put L j = K ∩ U j L_j = K \cap U_j L j = K ∩ U j . Each L j L_j L j is a subset of K K K , hence a Lindelöf subset of M 0 M_0 M 0 , and L j ⊆ K ∩ U j L_j \subseteq K \cap U_j L j ⊆ K ∩ U j ; moreover
L 1 ∪ L 2 = ( K ∩ U 1 ) ∪ ( K ∩ U 2 ) = K ∩ ( U 1 ∪ U 2 ) = K . L_1 \cup L_2 = (K \cap U_1) \cup(K \cap U_2) = K \cap(U_1 \cup U_2) = K. L 1 ∪ L 2 = ( K ∩ U 1 ) ∪ ( K ∩ U 2 ) = K ∩ ( U 1 ∪ U 2 ) = K . So the witnesses required by W i W_i W i exist.
M 0 ∉ W M_0 \notin W M 0 ∈ / W . Suppose that M 0 ∈ W M_0 \in W M 0 ∈ W . Choose distinct points a , b ∈ M 0 a,b \in M_0 a , b ∈ M 0 and put U 1 = M 0 ∖ { a } U_1 = M_0 \setminus\{a\} U 1 = M 0 ∖ { a } , U 2 = M 0 ∖ { b } U_2 = M_0 \setminus\{b\} U 2 = M 0 ∖ { b } . Both are open: for instance { a } \{a\} { a } is compact and hence closed, because M 0 M_0 M 0 is a KC-space, so its complement U 1 U_1 U 1 is open. Clearly U 1 ∪ U 2 = M 0 U_1 \cup U_2 = M_0 U 1 ∪ U 2 = M 0 , so the compact set K = M 0 K = M_0 K = M 0 satisfies K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . By W W W there are compact sets K 1 ⊆ U 1 K_1 \subseteq U_1 K 1 ⊆ U 1 and K 2 ⊆ U 2 K_2 \subseteq U_2 K 2 ⊆ U 2 with M 0 = K 1 ∪ K 2 M_0 = K_1 \cup K_2 M 0 = K 1 ∪ K 2 . Being compact in the KC-space M 0 M_0 M 0 , the sets K 1 K_1 K 1 and K 2 K_2 K 2 are closed; and K 1 ⊆ M 0 ∖ { a } K_1 \subseteq M_0 \setminus\{a\} K 1 ⊆ M 0 ∖ { a } gives a ∉ K 1 a \notin K_1 a ∈ / K 1 , so K 1 ≠ M 0 K_1 \ne M_0 K 1 = M 0 , and similarly K 2 ≠ M 0 K_2 \ne M_0 K 2 = M 0 . Hence M 0 ∖ K 1 M_0 \setminus K_1 M 0 ∖ K 1 and M 0 ∖ K 2 M_0 \setminus K_2 M 0 ∖ K 2 are nonempty open sets, while
( M 0 ∖ K 1 ) ∩ ( M 0 ∖ K 2 ) = M 0 ∖ ( K 1 ∪ K 2 ) = ∅ . (M_0 \setminus K_1) \cap(M_0 \setminus K_2) = M_0 \setminus(K_1 \cup K_2) = \varnothing. ( M 0 ∖ K 1 ) ∩ ( M 0 ∖ K 2 ) = M 0 ∖ ( K 1 ∪ K 2 ) = ∅ . contradicting the hyperconnectedness of M 0 M_0 M 0 . Therefore M 0 ∉ W M_0 \notin W M 0 ∈ / W , and M 0 ∈ W i ∖ W M_0 \in W_i \setminus W M 0 ∈ W i ∖ W . □
The space below is a more concrete example with the same effect; it is obtained from Q \mathbb{Q} Q by adding a point at infinity whose neighbourhoods are the complements of compact subsets of Q \mathbb{Q} Q .
Example 3.10. Let Q + = Q ∪ { ∞ } \mathbb{Q}^{+} = \mathbb{Q} \cup\{\infty\} Q + = Q ∪ { ∞ } , where Q \mathbb{Q} Q carries its usual topology and the neighbourhoods of ∞ \infty ∞ are the sets Q + ∖ C \mathbb{Q}^{+} \setminus C Q + ∖ C with C ∈ K ( Q ) C \in\mathcal{K}(\mathbb{Q}) C ∈ K ( Q ) . Then Q + \mathbb{Q}^{+} Q + is a countably infinite compact KC-space which is not Hausdorff, and Q + ∈ W i ∖ W \mathbb{Q}^{+} \in W_i \setminus W Q + ∈ W i ∖ W .
Proof. The family { Q + ∖ C : C ∈ K ( Q ) } \{\mathbb{Q}^{+} \setminus C : C \in\mathcal{K}(\mathbb{Q})\} { Q + ∖ C : C ∈ K ( Q )} is a neighbourhood base at ∞ \infty ∞ : it is closed under finite intersections because the union of two compact subsets of Q \mathbb{Q} Q is compact, and each of its members contains ∞ \infty ∞ .
Q + \mathbb{Q}^{+} Q + is compact. Let U \mathcal{U} U be an open cover of Q + \mathbb{Q}^{+} Q + and let U ∈ U U \in\mathcal{U} U ∈ U contain ∞ \infty ∞ . Then U U U contains some Q + ∖ C \mathbb{Q}^{+} \setminus C Q + ∖ C with C ∈ K ( Q ) C \in\mathcal{K}(\mathbb{Q}) C ∈ K ( Q ) , so C C C is covered by the remaining members of U \mathcal{U} U ; since C C C is compact in Q \mathbb{Q} Q , finitely many of them cover C C C . Together with U U U this gives a finite subcover.
Q + \mathbb{Q}^{+} Q + is a KC-space. For C ∈ K ( Q ) C \in\mathcal{K}(\mathbb{Q}) C ∈ K ( Q ) the set Q + ∖ C \mathbb{Q}^{+} \setminus C Q + ∖ C is open, so C C C is closed in Q + \mathbb{Q}^{+} Q + . Now let E E E be a compact subset of Q + \mathbb{Q}^{+} Q + . If ∞ ∉ E \infty\notin E ∞ ∈ / E , then E ⊆ Q E \subseteq\mathbb{Q} E ⊆ Q , and E E E is compact in the open subspace Q \mathbb{Q} Q of Q + \mathbb{Q}^{+} Q + , hence compact in Q \mathbb{Q} Q and therefore closed in Q \mathbb{Q} Q ; since Q \mathbb{Q} Q is open in Q + \mathbb{Q}^{+} Q + , it follows that E E E is closed in Q + \mathbb{Q}^{+} Q + . If ∞ ∈ E \infty\in E ∞ ∈ E , take x ∈ cl E ( E ∩ Q ) x \in\operatorname{cl}_{E}(E \cap\mathbb{Q}) x ∈ cl E ( E ∩ Q ) . As Q \mathbb{Q} Q is first countable, there are x n ∈ E ∩ Q x_n \in E \cap\mathbb{Q} x n ∈ E ∩ Q with x n → x x_n \to x x n → x . The set H = { x } ∪ { x n : n < ω } H = \{x\} \cup\{x_n : n < \omega\} H = { x } ∪ { x n : n < ω } is compact in Q \mathbb{Q} Q , being a convergent sequence together with its limit, so H H H is closed in Q + \mathbb{Q}^{+} Q + . Hence E ∩ H E \cap H E ∩ H is a closed subset of the compact set E E E and is therefore compact; it is contained in the Hausdorff space H H H , so it is closed in H H H . From x n ∈ E ∩ H x_n \in E \cap H x n ∈ E ∩ H and x n → x x_n \to x x n → x we obtain x ∈ cl H ( E ∩ H ) = E ∩ H ⊆ E x \in\operatorname{cl}_{H}(E \cap H) = E \cap H \subseteq E x ∈ cl H ( E ∩ H ) = E ∩ H ⊆ E . Consequently E ∩ Q E \cap\mathbb{Q} E ∩ Q is closed in Q \mathbb{Q} Q , and E E E is closed in Q + \mathbb{Q}^{+} Q + . Thus Q + \mathbb{Q}^{+} Q + is a KC-space, and it is countable by construction.
Q + \mathbb{Q}^{+} Q + is not Hausdorff. Suppose that 0 ∈ U 0 \in U 0 ∈ U and ∞ ∈ V \infty\in V ∞ ∈ V are disjoint open sets. Since V V V contains some Q + ∖ C \mathbb{Q}^{+} \setminus C Q + ∖ C with C ∈ K ( Q ) C \in\mathcal{K}(\mathbb{Q}) C ∈ K ( Q ) , and U ⊆ C U \subseteq C U ⊆ C . On the other hand U U U contains a rational neighbourhood of 0 0 0 , so there is ε > 0 \varepsilon> 0 ε > 0 with ( − ε , ε ) ∩ Q ⊆ U ⊆ C (-\varepsilon,\varepsilon) \cap\mathbb{Q} \subseteq U \subseteq C ( − ε , ε ) ∩ Q ⊆ U ⊆ C . The set C C C is compact as a subspace of Q \mathbb{Q} Q , hence compact and closed in R \mathbb{R} R , so [ − ε , ε ] = cl R ( ( − ε , ε ) ∩ Q ) ⊆ C ⊆ Q [-\varepsilon,\varepsilon] = \operatorname{cl}_{\mathbb{R}}((-\varepsilon,\varepsilon) \cap\mathbb{Q}) \subseteq C \subseteq\mathbb{Q} [ − ε , ε ] = cl R (( − ε , ε ) ∩ Q ) ⊆ C ⊆ Q , which is absurd because [ − ε , ε ] [-\varepsilon,\varepsilon] [ − ε , ε ] contains irrational numbers. Q + ∈ W i \mathbb{Q}^{+} \in W_i Q + ∈ W i . Since Q + \mathbb{Q}^{+} Q + is countable, every subset of it is Lindelöf, and the argument used for M 0 M_0 M 0 in the proof of Proposition 3.9 applies verbatim: for K ∈ K ( Q + ) K \in\mathcal{K}(\mathbb{Q}^{+}) K ∈ K ( Q + ) and K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 , the sets L j = K ∩ U j L_j = K \cap U_j L j = K ∩ U j are Lindelöf and satisfy L 1 ∪ L 2 = K L_1 \cup L_2 = K L 1 ∪ L 2 = K .
Q + ∉ W \mathbb{Q}^{+} \notin W Q + ∈ / W . Suppose that Q + ∈ W \mathbb{Q}^{+} \in W Q + ∈ W , and put U 1 = Q + ∖ { 0 } U_1 = \mathbb{Q}^{+} \setminus\{0\} U 1 = Q + ∖ { 0 } and U 2 = Q + ∖ { ∞ } = Q U_2 = \mathbb{Q}^{+} \setminus\{\infty\} = \mathbb{Q} U 2 = Q + ∖ { ∞ } = Q . Both are open: the singleton { 0 } \{0\} { 0 } is compact and hence closed, because Q + \mathbb{Q}^{+} Q + is a KC-space. Since U 1 ∪ U 2 = Q + U_1 \cup U_2 = \mathbb{Q}^{+} U 1 ∪ U 2 = Q + , the compact set K = Q + K = \mathbb{Q}^{+} K = Q + satisfies K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 , so by W W W there are compact sets K 1 ⊆ U 1 K_1 \subseteq U_1 K 1 ⊆ U 1 and K 2 ⊆ U 2 K_2 \subseteq U_2 K 2 ⊆ U 2 with Q + = K 1 ∪ K 2 \mathbb{Q}^{+} = K_1 \cup K_2 Q + = K 1 ∪ K 2 . Being compact in a KC-space, they are closed, and 0 ∉ K 1 0 \notin K_1 0 ∈ / K 1 , ∞ ∉ K 2 \infty\notin K_2 ∞ ∈ / K 2 , so both are proper closed subsets. Hence Q + ∖ K 1 \mathbb{Q}^{+} \setminus K_1 Q + ∖ K 1 and Q + ∖ K 2 \mathbb{Q}^{+} \setminus K_2 Q + ∖ K 2 are nonempty open sets containing 0 0 0 and ∞ \infty ∞ respectively, and their intersection is empty, a contradiction. Therefore Q + ∉ W \mathbb{Q}^{+} \notin W Q + ∈ / W , and Q + ∈ W i ∖ W \mathbb{Q}^{+} \in W_i \setminus W Q + ∈ W i ∖ W . □
The following theorem identifies W w W_w W w and W i W_i W i with continuity conditions for the union operation on the lattices of open sets; it is the reason why these properties are relevant for power spaces. The proof uses the fact, recorded in Lemma 2.4(v), that a basic neighbourhood of an open set may always be taken to be generated by a saturated Lindelöf set.
Theorem 3.11. Let X X X be a space. Then X ∈ W w X \in W_w X ∈ W w if and only if μ : O L ( X ) × O L ( X ) → O k ( X ) \mu: \mathcal{O}_L(X) \times\mathcal{O}_L(X) \to\mathcal{O}_k(X) μ : O L ( X ) × O L ( X ) → O k ( X ) , μ ( U , V ) = U ∪ V \mu(U,V) = U \cup V μ ( U , V ) = U ∪ V , is continuous. Moreover X ∈ W i X \in W_i X ∈ W i if and only if, for every K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , the restriction μ K : O L ( K ) × O L ( K ) → O k ( K ) \mu_K: \mathcal{O}_L(K) \times\mathcal{O}_L(K) \to\mathcal{O}_k(K) μ K : O L ( K ) × O L ( K ) → O k ( K ) , μ K ( U , V ) = U ∪ V \mu_K(U,V) = U \cup V μ K ( U , V ) = U ∪ V , is continuous.
Proof. We use the bases of Lemma 2.10. Observe first that Φ ( L ) = Φ ( sat L ) \Phi(L) = \Phi(\operatorname{sat} L) Φ ( L ) = Φ ( sat L ) and that sat L \operatorname{sat} L sat L is Lindelöf whenever L L L is, by Lemma 2.4(v); hence a basic open set of O k ( X ) \mathcal{O}_k(X) O k ( X ) or of O L ( X ) \mathcal{O}_L(X) O L ( X ) can always be written as Φ ( L ) \Phi(L) Φ ( L ) with L L L Lindelöf and saturated.
The case of W w W_w W w . Assume X ∈ W w X \in W_w X ∈ W w and let Φ ( K ) \Phi(K) Φ ( K ) be a basic open set of O k ( X ) \mathcal{O}_k(X) O k ( X ) , with K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) . Then μ − 1 Φ ( K ) = { ( U , V ) : K ⊆ U ∪ V } \mu^{-1}\Phi(K) = \{(U,V): K \subseteq U \cup V\} μ − 1 Φ ( K ) = {( U , V ) : K ⊆ U ∪ V } . Given ( U , V ) (U,V) ( U , V ) in this set, W w W_w W w provides L 1 , L 2 ∈ L ( X ) L_1,L_2 \in\mathcal{L}(X) L 1 , L 2 ∈ L ( X ) with L j ⊆ U j L_j \subseteq U_j L j ⊆ U j (where U 1 = U U_1 = U U 1 = U , U 2 = V U_2 = V U 2 = V ) and K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 . Then ( U , V ) ∈ Φ ( L 1 ) × Φ ( L 2 ) (U,V) \in\Phi(L_1) \times\Phi(L_2) ( U , V ) ∈ Φ ( L 1 ) × Φ ( L 2 ) , and every ( U ′ , V ′ ) ∈ Φ ( L 1 ) × Φ ( L 2 ) (U',V') \in\Phi(L_1) \times\Phi(L_2) ( U ′ , V ′ ) ∈ Φ ( L 1 ) × Φ ( L 2 ) satisfies K ⊆ L 1 ∪ L 2 ⊆ U ′ ∪ V ′ K \subseteq L_1 \cup L_2 \subseteq U' \cup V' K ⊆ L 1 ∪ L 2 ⊆ U ′ ∪ V ′ . Hence Φ ( L 1 ) × Φ ( L 2 ) ⊆ μ − 1 Φ ( K ) \Phi(L_1) \times\Phi(L_2) \subseteq\mu^{-1}\Phi(K) Φ ( L 1 ) × Φ ( L 2 ) ⊆ μ − 1 Φ ( K ) , and μ \mu μ is continuous at ( U , V ) (U,V) ( U , V ) .
Conversely, assume that μ \mu μ is continuous, and take K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and U 1 , U 2 ∈ O ( X ) U_1,U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . Then ( U 1 , U 2 ) ∈ μ − 1 Φ ( K ) (U_1,U_2) \in\mu^{-1}\Phi(K) ( U 1 , U 2 ) ∈ μ − 1 Φ ( K ) , so by continuity there are L 1 , L 2 ∈ L ( X ) L_1,L_2 \in\mathcal{L}(X) L 1 , L 2 ∈ L ( X ) with ( U 1 , U 2 ) ∈ Φ ( L 1 ) × Φ ( L 2 ) (U_1,U_2) \in\Phi(L_1) \times\Phi(L_2) ( U 1 , U 2 ) ∈ Φ ( L 1 ) × Φ ( L 2 ) and Φ ( L 1 ) × Φ ( L 2 ) ⊆ μ − 1 Φ ( K ) \Phi(L_1) \times\Phi(L_2) \subseteq\mu^{-1}\Phi(K) Φ ( L 1 ) × Φ ( L 2 ) ⊆ μ − 1 Φ ( K ) . In particular L j ⊆ U j L_j \subseteq U_j L j ⊆ U j , and every pair of open sets G ⊇ L 1 G \supseteq L_1 G ⊇ L 1 , H ⊇ L 2 H \supseteq L_2 H ⊇ L 2 satisfies K ⊆ G ∪ H K \subseteq G \cup H K ⊆ G ∪ H . We claim that K ⊆ sat L 1 ∪ sat L 2 K \subseteq\operatorname{sat} L_1 \cup\operatorname{sat} L_2 K ⊆ sat L 1 ∪ sat L 2 . Otherwise there is x ∈ K ∖ ( sat L 1 ∪ sat L 2 ) x \in K \setminus(\operatorname{sat} L_1 \cup\operatorname{sat} L_2) x ∈ K ∖ ( sat L 1 ∪ sat L 2 ) ; since x ∉ sat L 1 = { G ∈ O ( X ) : L 1 ⊆ G } x \notin\operatorname{sat} L_1 = \{G \in\mathcal{O}(X): L_1 \subseteq G\} x ∈ / sat L 1 = { G ∈ O ( X ) : L 1 ⊆ G } , there is an open set G ⊇ L 1 G \supseteq L_1 G ⊇ L 1 with x ∉ G x \notin G x ∈ / G , and similarly there is an open set H ⊇ L 2 H \supseteq L_2 H ⊇ L 2 with x ∉ H x \notin H x ∈ / H . Then K ⊆ G ∪ H K \subseteq G \cup H K ⊆ G ∪ H but x ∉ G ∪ H x \notin G \cup H x ∈ / G ∪ H , a contradiction. Since sat L j ⊆ U j \operatorname{sat} L_j \subseteq U_j sat L j ⊆ U j and sat L j \operatorname{sat} L_j sat L j is Lindelöf, we conclude that X ∈ W w X \in W_w X ∈ W w .
The case of W i W_i W i . Assume X ∈ W i X \in W_i X ∈ W i , let K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and let Φ K ( F ) \Phi_K(F) Φ K ( F ) be a basic open set of O k ( K ) \mathcal{O}_k(K) O k ( K ) , where F ∈ K ( K ) F \in\mathcal{K}(K) F ∈ K ( K ) and Φ K ( F ) = { W ∈ O ( K ) : F ⊆ W } \Phi_K(F) = \{W \in\mathcal{O}(K): F \subseteq W\} Φ K ( F ) = { W ∈ O ( K ) : F ⊆ W } for F ⊆ K F \subseteq K F ⊆ K . Take ( U , V ) ∈ μ K − 1 Φ K ( F ) (U,V) \in\mu_K^{-1}\Phi_K(F) ( U , V ) ∈ μ K − 1 Φ K ( F ) , so that F ⊆ U ∪ V F \subseteq U \cup V F ⊆ U ∪ V . Choose U ′ , V ′ ∈ O ( X ) U',V' \in\mathcal{O}(X) U ′ , V ′ ∈ O ( X ) with U ′ ∩ K = U U' \cap K = U U ′ ∩ K = U , V ′ ∩ K = V V' \cap K = V V ′ ∩ K = V , then F ⊆ U ′ ∪ V ′ F \subseteq U' \cup V' F ⊆ U ′ ∪ V ′ , and since F ∈ K ( X ) F \in\mathcal{K}(X) F ∈ K ( X ) we may apply X ∈ W i X \in W_i X ∈ W i and obtain L 1 , L 2 ∈ L ( X ) L_1,L_2 \in\mathcal{L}(X) L 1 , L 2 ∈ L ( X ) with L j ⊆ F ∩ U j ′ L_j \subseteq F \cap U'_j L j ⊆ F ∩ U j ′ and F = L 1 ∪ L 2 F = L_1 \cup L_2 F = L 1 ∪ L 2 . Then L j ⊆ F ⊆ K L_j \subseteq F \subseteq K L j ⊆ F ⊆ K , so L j ∈ L ( K ) L_j \in\mathcal{L}(K) L j ∈ L ( K ) and L j ⊆ U j ′ ∩ K = U j L_j \subseteq U'_j \cap K = U_j L j ⊆ U j ′ ∩ K = U j . Thus ( U , V ) ∈ Φ K ( L 1 ) × Φ K ( L 2 ) (U,V) \in\Phi_K(L_1) \times\Phi_K(L_2) ( U , V ) ∈ Φ K ( L 1 ) × Φ K ( L 2 ) , and every ( U ′ ′ , V ′ ′ ) ∈ Φ K ( L 1 ) × Φ K ( L 2 ) (U'',V'') \in\Phi_K(L_1) \times\Phi_K(L_2) ( U ′′ , V ′′ ) ∈ Φ K ( L 1 ) × Φ K ( L 2 ) satisfies F = L 1 ∪ L 2 ⊆ U ′ ′ ∪ V ′ ′ F = L_1 \cup L_2 \subseteq U'' \cup V'' F = L 1 ∪ L 2 ⊆ U ′′ ∪ V ′′ , so that Φ K ( L 1 ) × Φ K ( L 2 ) ⊆ μ K − 1 Φ K ( F ) \Phi_K(L_1) \times\Phi_K(L_2) \subseteq\mu_K^{-1}\Phi_K(F) Φ K ( L 1 ) × Φ K ( L 2 ) ⊆ μ K − 1 Φ K ( F ) . Hence μ K \mu_K μ K is continuous.
Conversely, assume that μ K \mu_K μ K is continuous for every K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , and take K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , U 1 , U 2 ∈ O ( X ) U_1,U_2 \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . Put V j = K ∩ U j V_j = K \cap U_j V j = K ∩ U j ; then V 1 , V 2 V_1,V_2 V 1 , V 2 are open in K K K and K = V 1 ∪ V 2 K = V_1 \cup V_2 K = V 1 ∪ V 2 , so ( V 1 , V 2 ) ∈ μ K − 1 Φ K ( K ) (V_1,V_2) \in\mu_K^{-1}\Phi_K(K) ( V 1 , V 2 ) ∈ μ K − 1 Φ K ( K ) . By continuity there are L 1 , L 2 ∈ L ( K ) L_1,L_2 \in\mathcal{L}(K) L 1 , L 2 ∈ L ( K ) with L j ⊆ V j L_j \subseteq V_j L j ⊆ V j and Φ K ( L 1 ) × Φ K ( L 2 ) ⊆ μ K − 1 Φ K ( K ) \Phi_K(L_1) \times\Phi_K(L_2) \subseteq\mu_K^{-1}\Phi_K(K) Φ K ( L 1 ) × Φ K ( L 2 ) ⊆ μ K − 1 Φ K ( K ) . This means that K ⊆ W 1 ∪ W 2 K \subseteq W_1 \cup W_2 K ⊆ W 1 ∪ W 2 whenever W j W_j W j are open in K K K with W j ⊇ L j W_j \supseteq L_j W j ⊇ L j . Repeating the argument of the previous paragraph inside the space K K K , with sat K L = ⋂ { W ∈ O ( K ) : L ⊆ W } \operatorname{sat}_K L = \bigcap\{W \in\mathcal{O}(K): L \subseteq W\} sat K L = ⋂ { W ∈ O ( K ) : L ⊆ W } in place of sat L \operatorname{sat} L sat L , we obtain K ⊆ sat K L 1 ∪ sat K L 2 K \subseteq\operatorname{sat}_K L_1 \cup\operatorname{sat}_K L_2 K ⊆ sat K L 1 ∪ sat K L 2 . Now sat K L j \operatorname{sat}_K L_j sat K L j is Lindelöf as a subspace of K K K , hence as a subspace of X X X ; since V j ⊆ U j V_j \subseteq U_j V j ⊆ U j and K ⊆ sat K L 1 ∪ sat K L 2 K \subseteq\operatorname{sat}_K L_1 \cup\operatorname{sat}_K L_2 K ⊆ sat K L 1 ∪ sat K L 2 , we conclude that X ∈ W i X \in W_i X ∈ W i . □
Core compactness gives a sufficient condition for the weak Wilker property. The proof rests on the following standard interpolation property of the way-below relation; it holds because a core compact space has a continuous lattice of open sets [2 , 14 ] .
Lemma 3.12. Let X X X be a core compact space and let V ≪ U V \ll U V ≪ U be open subsets of X X X . Then there is an open set W W W with V ≪ W ≪ U V \ll W \ll U V ≪ W ≪ U .
Proof. For a core compact space the lattice O ( X ) \mathcal{O}(X) O ( X ) , ordered by inclusion, is continuous [2 ] , and in a continuous lattice the way-below relation is interpolative [2 Lemma I-1.20] : from V ≪ U V \ll U V ≪ U one obtains an open set W W W with V ≪ W ≪ U V \ll W \ll U V ≪ W ≪ U . Note that W ⊆ U W \subseteq U W ⊆ U , since W ≪ U W \ll U W ≪ U forces W ⊆ U W \subseteq U W ⊆ U . □
Theorem 3.13. Let X X X be a core compact space. Then every open subset of X X X is a union of Lindelöf open sets; more precisely, for every open cover U \mathcal{U} U of a compact set K ⊆ X K \subseteq X K ⊆ X there are finitely many Lindelöf open sets, each contained in a member of U \mathcal{U} U , that cover K K K . In particular X ∈ W w X \in W_w X ∈ W w .
Proof. Step 1: a point has arbitrarily small Lindelöf open neighbourhoods. Fix x ∈ U x \in U x ∈ U with U U U open. By core compactness there is V 0 ∈ O ( X ) V_0 \in\mathcal{O}(X) V 0 ∈ O ( X ) with x ∈ V 0 ≪ U x \in V_0 \ll U x ∈ V 0 ≪ U . Applying Lemma 3.12 repeatedly, we obtain open sets V 0 , V 1 , V 2 , … V_0,V_1,V_2,\ldots V 0 , V 1 , V 2 , … with
x ∈ V 0 ≪ V 1 ≪ V 2 ≪ ⋯ ≪ U . x \in V_0 \ll V_1 \ll V_2 \ll\cdots\ll U. x ∈ V 0 ≪ V 1 ≪ V 2 ≪ ⋯ ≪ U .
Put N = ⋃ n < ω V n N = \bigcup_{n<\omega} V_n N = ⋃ n < ω V n . Then N N N is open and x ∈ N ⊆ U x \in N \subseteq U x ∈ N ⊆ U . To see that N N N is Lindelöf, let G \mathcal{G} G be a cover of N N N by open sets. For fixed n n n , the family of finite unions of members of G \mathcal{G} G is directed and its union contains N N N , hence contains V n + 1 V_{n+1} V n + 1 ; since V n ≪ V n + 1 V_n \ll V_{n+1} V n ≪ V n + 1 , there is a finite G n ⊆ G \mathcal{G}_n \subseteq\mathcal{G} G n ⊆ G with V n ⊆ ⋃ G n V_n \subseteq\bigcup\mathcal{G}_n V n ⊆ ⋃ G n . Therefore N = ⋃ n < ω V n ⊆ ⋃ n ⋃ G n N = \bigcup_{n<\omega} V_n \subseteq\bigcup_n \bigcup\mathcal{G}_n N = ⋃ n < ω V n ⊆ ⋃ n ⋃ G n , and ⋃ n G n \bigcup_n \mathcal{G}_n ⋃ n G n is a countable subfamily of G \mathcal{G} G covering N N N . Thus N N N is a Lindelöf open set with x ∈ N ⊆ U x \in N \subseteq U x ∈ N ⊆ U .
Step 2: finitely many such sets cover a compact set. Let U \mathcal{U} U be an open cover of K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) . For each x ∈ K x \in K x ∈ K choose U x ∈ U U_x \in\mathcal{U} U x ∈ U with x ∈ U x x \in U_x x ∈ U x and, by Step 1, a Lindelöf open set N x N_x N x with x ∈ N x ⊆ U x x \in N_x \subseteq U_x x ∈ N x ⊆ U x . Since K K K is compact, there is a finite F ⊆ K F \subseteq K F ⊆ K with K ⊆ ⋃ x ∈ F N x K \subseteq\bigcup_{x\in F}N_x K ⊆ ⋃ x ∈ F N x . Grouping the finitely many sets N x N_x N x according to the member U x U_x U x of U \mathcal{U} U that contains them, and taking finite unions, we obtain finitely many Lindelöf open sets, each contained in a member of U \mathcal{U} U , whose union contains K K K .
Step 3: X ∈ W w X \in\mathsf{W}_w X ∈ W w . Given K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) and K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 , apply Step 2 to the two-element cover { U 1 , U 2 } \{U_1,U_2\} { U 1 , U 2 } : there are Lindelöf open sets L 1 ⊆ U 1 L_1 \subseteq U_1 L 1 ⊆ U 1 , L 2 ⊆ U 2 L_2 \subseteq U_2 L 2 ⊆ U 2 with K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 . Hence X ∈ W w X \in\mathsf{W}_w X ∈ W w . □ \square □
The last result of this section describes the behaviour of W w \mathsf{W}_w W w under products. Its hypothesis is that one factor belongs to W w \mathsf{W}_w W w and that the compact subspaces of the other factor are Hausdorff; under this assumption the product remains in W w \mathsf{W}_w W w , and if moreover the first factor is KC, then the product is KC and belongs to W i \mathsf{W}_i W i .
Theorem 3.14
Theorem 3.14. Let X ∈ W w X \in\mathsf{W}_w X ∈ W w and suppose that every compact subspace of Y Y Y is Hausdorff. Then X × Y ∈ W w X \times Y \in\mathsf{W}_w X × Y ∈ W w . If in addition X X X is a KC-space, then X × Y X \times Y X × Y is a KC-space and X × Y ∈ W i X \times Y \in\mathsf{W}_i X × Y ∈ W i .
Proof. Let K ∈ K ( X × Y ) K \in\mathcal{K}(X \times Y) K ∈ K ( X × Y ) and let U 1 , U 2 U_1,U_2 U 1 , U 2 be open subsets of X × Y X \times Y X × Y with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 ; we may assume K ≠ ∅ K \ne\varnothing K = ∅ . Put C = π Y ( K ) C = \pi_Y(K) C = π Y ( K ) , a compact subset of Y Y Y , and replace each U i U_i U i by U i ∩ ( X × C ) U_i \cap(X \times C) U i ∩ ( X × C ) , which is open in X × C X \times C X × C and which we still denote by U i U_i U i .
Fibres over C C C . Since every compact subspace of Y Y Y is Hausdorff, C C C is Hausdorff, hence so is X × C X \times C X × C . For y ∈ C y \in C y ∈ C the set K y = K ∩ ( X × { y } ) K_y = K \cap(X \times\{y\}) K y = K ∩ ( X × { y }) is closed in K K K , because X × { y } X \times\{y\} X × { y } is closed in X × C X \times C X × C , and is therefore compact. As K y ⊆ U 1 ∪ U 2 K_y \subseteq U_1 \cup U_2 K y ⊆ U 1 ∪ U 2 , finitely many open rectangles G r × H r G_r \times H_r G r × H r with G r × H r ⊆ U c ( r ) G_r \times H_r \subseteq U_{c(r)} G r × H r ⊆ U c ( r ) , c ( r ) ∈ { 1 , 2 } c(r) \in\{1,2\} c ( r ) ∈ { 1 , 2 } , cover K y K_y K y , and we may assume y ∈ H r y \in H_r y ∈ H r for all r r r . Set
V i = ⋃ c ( r ) = i G r , H = ⋂ r H r . V_i = \bigcup_{c(r)=i} G_r,\qquad H = \bigcap_r H_r. V i = c ( r ) = i ⋃ G r , H = r ⋂ H r .
Then y ∈ H y \in H y ∈ H , K y ⊆ ( V 1 ∪ V 2 ) × H K_y \subseteq(V_1 \cup V_2) \times H K y ⊆ ( V 1 ∪ V 2 ) × H , and V i × H ⊆ U i V_i \times H \subseteq U_i V i × H ⊆ U i for i = 1 , 2 i = 1,2 i = 1 , 2 .
Shrinking the base. Put D y = π Y ( K ∖ ( ( V 1 ∪ V 2 ) × C ) ) D_y = \pi_Y\bigl(K \setminus((V_1 \cup V_2) \times C)\bigr) D y = π Y ( K ∖ (( V 1 ∪ V 2 ) × C ) ) . The set D y D_y D y is compact, being a continuous image of a closed subset of the compact space K K K , and y ∉ D y y \notin D_y y ∈ / D y because K y ⊆ ( V 1 ∪ V 2 ) × C K_y \subseteq(V_1 \cup V_2) \times C K y ⊆ ( V 1 ∪ V 2 ) × C . As C C C is compact Hausdorff, it is regular, so there is an open set N y N_y N y of C C C with y ∈ N y y \in N_y y ∈ N y and B y = cl C N y ⊆ C ∖ D y B_y = \operatorname{cl}_C N_y \subseteq C \setminus D_y B y = cl C N y ⊆ C ∖ D y . Observe that B y ⊆ H B_y \subseteq H B y ⊆ H . The set T y = K ∩ ( X × B y ) T_y = K \cap(X \times B_y) T y = K ∩ ( X × B y ) is compact, being closed in K K K , and π X ( T y ) ⊆ V 1 ∪ V 2 \pi_X(T_y) \subseteq V_1 \cup V_2 π X ( T y ) ⊆ V 1 ∪ V 2 by the choice of B y B_y B y ; moreover V i × B y ⊆ U i V_i \times B_y \subseteq U_i V i × B y ⊆ U i .
Conclusion. Since π X ( T y ) \pi_X(T_y) π X ( T y ) is compact and X ∈ W w X \in\mathsf{W}_w X ∈ W w , there are A 1 ( y ) , A 2 ( y ) ∈ L ( X ) A_1(y),A_2(y) \in\mathcal{L}(X) A 1 ( y ) , A 2 ( y ) ∈ L ( X ) with π X ( T y ) ⊆ A 1 ( y ) ∪ A 2 ( y ) \pi_X(T_y) \subseteq A_1(y) \cup A_2(y) π X ( T y ) ⊆ A 1 ( y ) ∪ A 2 ( y ) and A i ( y ) ⊆ V i A_i(y) \subseteq V_i A i ( y ) ⊆ V i . The family { N y : y ∈ C } \{N_y : y \in C\} { N y : y ∈ C } covers the compact set C C C , so there is a finite J ⊆ C J \subseteq C J ⊆ C with C = ⋃ y ∈ J N y C = \bigcup_{y\in J}N_y C = ⋃ y ∈ J N y . Put
L i = ⋃ y ∈ J A i ( y ) × B y . L_i = \bigcup_{y\in J} A_i(y) \times B_y. L i = y ∈ J ⋃ A i ( y ) × B y .
Then L i ⊆ U i L_i \subseteq U_i L i ⊆ U i , because A i ( y ) ⊆ V i A_i(y) \subseteq V_i A i ( y ) ⊆ V i and V i × B y ⊆ U i V_i \times B_y \subseteq U_i V i × B y ⊆ U i , and K = ⋃ y ∈ J T y ⊆ L 1 ∪ L 2 K = \bigcup_{y\in J} T_y \subseteq L_1 \cup L_2 K = ⋃ y ∈ J T y ⊆ L 1 ∪ L 2 . It remains to note that L i L_i L i is Lindelöf: the product of a Lindelöf space and a compact space is Lindelöf, since for such a product A × B A \times B A × B one covers each slice { a } × B \{a\} \times B { a } × B by finitely many rectangles, intersects the corresponding first-coordinate neighbourhoods to obtain an open G a ∋ a G_a \ni a G a ∋ a , chooses a countable A 0 ⊆ A A_0 \subseteq A A 0 ⊆ A with A = ⋃ a ∈ A 0 G a A = \bigcup_{a\in A_0}G_a A = ⋃ a ∈ A 0 G a and collects the countably many finite subfamilies so obtained. Therefore L i L_i L i is a finite union of Lindelöf sets, hence Lindelöf, and X × Y ∈ W w X \times Y \in\mathsf{W}_w X × Y ∈ W w .
The KC case. Assume now that X X X is a KC-space. We first note that every compact subset of Y Y Y is closed in Y Y Y : if D ⊆ Y D \subseteq Y D ⊆ Y is compact and z ∈ cl Y D z \in\operatorname{cl}_Y D z ∈ cl Y D , then D ∪ { z } D \cup\{z\} D ∪ { z } is compact, being a subset of the compact set cl Y D \operatorname{cl}_Y D cl Y D , and it is Hausdorff by hypothesis, so D D D is closed in D ∪ { z } D \cup\{z\} D ∪ { z } and z ∈ D z \in D z ∈ D . In particular C C C is closed in Y Y Y , and K K K is a closed subset of X × C X \times C X × C . To see that K K K is closed in X × C X \times C X × C , let ( x , y ) ∉ K (x,y) \notin K ( x , y ) ∈ / K . Since X X X is KC, it is T 1 T_1 T 1 , so { x } \{x\} { x } is closed in X X X and K x = π Y ( K ∩ ( { x } × C ) ) K_x=\pi_Y(K\cap(\{x\}\times C)) K x = π Y ( K ∩ ({ x } × C )) is compact; it is closed in the compact Hausdorff space C C C . Choose a closed neighbourhood B B B of y y y in C C C with B ∩ K x = ∅ B\cap K_x=\varnothing B ∩ K x = ∅ , and put D = π X ( K ∩ ( X × B ) ) D=\pi_X(K\cap(X\times B)) D = π X ( K ∩ ( X × B )) . Then D D D is compact, and it is closed in X X X because X X X is KC; moreover x ∉ D x\notin D x ∈ / D . Hence
( x , y ) ∈ ( X ∖ D ) × int C B ⊆ ( X × C ) ∖ K , (x,y)\in(X\setminus D)\times\operatorname{int}_C B\subseteq(X\times C)\setminus K, ( x , y ) ∈ ( X ∖ D ) × int C B ⊆ ( X × C ) ∖ K ,
which shows that K K K is closed. Thus X × Y X\times Y X × Y is a KC-space, and Proposition 3.8 (i) gives X × Y ∈ W i X\times Y\in W_i X × Y ∈ W i . □
Counterexamples in Power Spaces and Function Spaces
The heredity results of §3 show that the inner properties are stable, but they say nothing about products and about the hyperspace constructions. In this section we show that all of these can fail, and we use the failures to answer a question raised in [4 ] and to produce a non-sober function space.
The basic building block is a compact T 1 T_1 T 1 space whose topology is so large that no two Lindelöf pieces can cover a binary open cover.
Lemma 4.1. Let A A A be a set with ∣ A ∣ = ℵ 2 |A|=\aleph_2 ∣ A ∣ = ℵ 2 , put D = A × A D=A\times A D = A × A and M = D ⊔ { p , q } M=D\sqcup\{p,q\} M = D ⊔ { p , q } . Declare every point of D D D isolated, and take at p p p and q q q the neighbourhood bases
N p ( F ) = { p } ∪ ( ( A ∖ F ) × A ) , N q ( G ) = { q } ∪ ( A × ( A ∖ G ) ) , N_p(F)=\{p\}\cup((A\setminus F)\times A),\qquad N_q(G)=\{q\}\cup(A\times(A\setminus G)), N p ( F ) = { p } ∪ (( A ∖ F ) × A ) , N q ( G ) = { q } ∪ ( A × ( A ∖ G )) , where F , G F,G F , G range over the finite subsets of A A A . Then M M M is compact and T 1 T_1 T 1 , and for U = M ∖ { p } U=M\setminus\{p\} U = M ∖ { p } , V = M ∖ { q } V=M\setminus\{q\} V = M ∖ { q } there are no L U ∈ L ( U ) L_U\in\mathcal{L}(U) L U ∈ L ( U ) and L V ∈ L ( V ) L_V\in\mathcal{L}(V) L V ∈ L ( V ) with M ⊆ L U ∪ L V M\subseteq L_U\cup L_V M ⊆ L U ∪ L V . In particular M ∉ W w M\notin W_w M ∈ / W w .
Proof. M M M is T 1 T_1 T 1 and compact. For d ∈ D d\in D d ∈ D the set M ∖ { d } M\setminus\{d\} M ∖ { d } is a union of the open set D ∖ { d } D\setminus\{d\} D ∖ { d } and of suitable members of the two neighbourhood bases, hence open; and M ∖ { p } = N q ( ∅ ) M\setminus\{p\}=N_q(\varnothing) M ∖ { p } = N q ( ∅ ) , M ∖ { q } = N p ( ∅ ) M\setminus\{q\}=N_p(\varnothing) M ∖ { q } = N p ( ∅ ) are open. So all singletons are closed. For compactness, let U \mathcal{U} U be an open cover of M M M and choose U p , U q ∈ U U_p,U_q\in\mathcal{U} U p , U q ∈ U with p ∈ U p p\in U_p p ∈ U p , q ∈ U q q\in U_q q ∈ U q . There are finite F , G ⊆ A F,G\subseteq A F , G ⊆ A with N p ( F ) ⊆ U p N_p(F)\subseteq U_p N p ( F ) ⊆ U p and N q ( G ) ⊆ U q N_q(G)\subseteq U_q N q ( G ) ⊆ U q , and
M ∖ ( N p ( F ) ∪ N q ( G ) ) = F × G M\setminus(N_p(F)\cup N_q(G))=F\times G M ∖ ( N p ( F ) ∪ N q ( G )) = F × G
is finite; hence finitely many members of U \mathcal{U} U cover M M M .
A counting condition. In U = M ∖ { p } = D ∪ { q } U=M\setminus\{p\}=D\cup\{q\} U = M ∖ { p } = D ∪ { q } the column A × { b } A\times\{b\} A × { b } is a closed discrete subset, because its complement U ∖ ( A × { b } ) = N q ( { b } ) U\setminus(A\times\{b\})=N_q(\{b\}) U ∖ ( A × { b }) = N q ({ b }) is open. A closed discrete subset of a Lindelöf space is countable, so for every L U ∈ L ( U ) L_U\in\mathcal{L}(U) L U ∈ L ( U ) and every b ∈ A b\in A b ∈ A the set B b = { a : ( a , b ) ∈ B } B_b=\{a:(a,b)\in B\} B b = { a : ( a , b ) ∈ B } is at most countable, where B = L U ∩ D B=L_U\cap D B = L U ∩ D . Symmetrically, for L V ∈ L ( V ) L_V\in\mathcal{L}(V) L V ∈ L ( V ) and E = L V ∩ D E=L_V\cap D E = L V ∩ D every row E a = { b : ( a , b ) ∈ E } E_a=\{b:(a,b)\in E\} E a = { b : ( a , b ) ∈ E } is at most countable. Thus
( ∀ b ∈ A ) ∣ B b ∣ ≤ ℵ 0 , ( ∀ a ∈ A ) ∣ E a ∣ ≤ ℵ 0 . (3) (\forall b\in A)\ |B_b|\leq\aleph_0,\qquad(\forall a\in A)\ |E_a|\leq\aleph_0. \tag*{(3)} ( ∀ b ∈ A ) ∣ B b ∣ ≤ ℵ 0 , ( ∀ a ∈ A ) ∣ E a ∣ ≤ ℵ 0 . (3)
No cover by two Lindelöf pieces. Suppose that B , E ⊆ D B,E\subseteq D B , E ⊆ D satisfy (3). Choose I ⊆ A I\subseteq A I ⊆ A with ∣ I ∣ = ℵ 1 |I|=\aleph_1 ∣ I ∣ = ℵ 1 . Since
∣ ⋃ a ∈ I E a ∣ ≤ ℵ 1 ⋅ ℵ 0 = ℵ 1 < ∣ A ∣ , \left|\bigcup_{a\in I}E_a\right|\leq\aleph_1\cdot\aleph_0=\aleph_1<|A|, a ∈ I ⋃ E a ≤ ℵ 1 ⋅ ℵ 0 = ℵ 1 < ∣ A ∣ ,
there is b 0 ∈ A ∖ ⋃ a ∈ I E a b_0\in A\setminus\bigcup_{a\in I}E_a b 0 ∈ A ∖ ⋃ a ∈ I E a ; and since ∣ B b 0 ∣ ≤ ℵ 0 < ∣ I ∣ |B_{b_0}|\leq\aleph_0<|I| ∣ B b 0 ∣ ≤ ℵ 0 < ∣ I ∣ , there is a 0 ∈ I ∖ B b 0 a_0\in I\setminus B_{b_0} a 0 ∈ I ∖ B b 0 . Then ( a 0 , b 0 ) ∈ D (a_0,b_0)\in D ( a 0 , b 0 ) ∈ D while ( a 0 , b 0 ) ∉ B (a_0,b_0)\notin B ( a 0 , b 0 ) ∈ / B , because a 0 ∉ B b 0 a_0\notin B_{b_0} a 0 ∈ / B b 0 , and ( a 0 , b 0 ) ∉ E (a_0,b_0)\notin E ( a 0 , b 0 ) ∈ / E , because b 0 ∉ E a 0 b_0\notin E_{a_0} b 0 ∈ / E a 0 and a 0 ∈ I a_0\in I a 0 ∈ I . Hence B ∪ E ≠ D B\cup E\neq D B ∪ E = D .
Now assume that L U ∈ L ( U ) L_U\in\mathcal{L}(U) L U ∈ L ( U ) , L V ∈ L ( V ) L_V\in\mathcal{L}(V) L V ∈ L ( V ) and M ⊆ L U ∪ L V M\subseteq L_U\cup L_V M ⊆ L U ∪ L V . Put B = L U ∩ D B=L_U\cap D B = L U ∩ D and E = L V ∩ D E=L_V\cap D E = L V ∩ D ; by the counting condition B B B and E E E satisfy (3), so there is d ∈ D ∖ ( B ∪ E ) d\in D\setminus(B\cup E) d ∈ D ∖ ( B ∪ E ) . But d ∈ M ⊆ L U ∪ L V d\in M\subseteq L_U\cup L_V d ∈ M ⊆ L U ∪ L V , and L U ⊆ U = D ∪ { q } L_U\subseteq U=D\cup\{q\} L U ⊆ U = D ∪ { q } , L V ⊆ V = D ∪ { p } L_V\subseteq V=D\cup\{p\} L V ⊆ V = D ∪ { p } force d ∈ L U ∩ D = B d\in L_U\cap D=B d ∈ L U ∩ D = B or d ∈ L V ∩ D = E d\in L_V\cap D=E d ∈ L V ∩ D = E , a contradiction. Thus the open cover { U , V } \{U,V\} { U , V } of M M M admits no Lindelöf refinement, and M ∉ W w M\notin W_w M ∈ / W w . □
The cofinite modification of this space produces the two spaces P P P and Q Q Q whose product fails W w W_w W w although every subspace of each belongs to W i n W_{in} W in .
Theorem 4.2. There are compact T 1 T_1 T 1 spaces P P P and Q Q Q such that every subspace of P P P and every subspace of Q Q Q belongs to W i n W_{in} W in , while P × Q ∉ W w P\times Q\notin W_w P × Q ∈ / W w .
Proof. Let M M M be the space of Lemma 4.1 and let τ = O ( M ) \tau=\mathcal{O}(M) τ = O ( M ) . For a ∈ { p , q } a\in\{p,q\} a ∈ { p , q } put
τ a = { G ∈ τ : a ∉ G } ∪ { M ∖ F : F ⊆ M finite } , \tau_a=\{G\in\tau:a\notin G\}\cup\{M\setminus F:F\subseteq M\text{ finite}\}, τ a = { G ∈ τ : a ∈ / G } ∪ { M ∖ F : F ⊆ M finite } ,
and set P = ( M , τ p ) P=(M,\tau_p) P = ( M , τ p ) , Q = ( M , τ q ) Q=(M,\tau_q) Q = ( M , τ q ) ; this construction follows the cofinite modification of compact T 1 T_1 T 1 spaces studied in [8 ] . τ a \tau_a τ a is a topology. It contains ∅ \varnothing ∅ and M M M , and it is closed under finite intersections and arbitrary unions: the intersection of two members of τ \tau τ avoiding a a a avoids a a a , the intersection of M ∖ F M \setminus F M ∖ F with any member of τ a \tau_a τ a is in τ \tau τ , and it avoids a a a when the member does, while the intersection of two cofinite sets is cofinite. Moreover τ a ⊆ τ \tau_a \subseteq\tau τ a ⊆ τ , because M M M is T 1 T_1 T 1 .
The punctured space. On M ∖ { a } M \setminus\{a\} M ∖ { a } the topology τ a \tau_a τ a coincides with τ \tau τ : every τ a \tau_a τ a -open set restricts to a τ \tau τ -open set, since τ a ⊆ τ \tau_a \subseteq\tau τ a ⊆ τ , and every τ \tau τ -open set is τ a \tau_a τ a -open because it avoids a a a . Hence M ∖ { a } M \setminus\{a\} M ∖ { a } is the space D ∪ { the point other than a } D \cup\{\text{the point other than }a\} D ∪ { the point other than a } with the topology τ \tau τ , and it is Hausdorff: points of D D D are isolated, and N q ( G ) N_q(G) N q ( G ) separates q q q from every point outside it.
Subspaces of P P P lie in W i n \mathrm{W}_{\mathrm{in}} W in . Let S ⊆ M S \subseteq M S ⊆ M . If a ∈ S a \in S a ∈ S , then every relatively open set of S S S containing a a a is of the form S ∩ ( M ∖ F ) S \cap(M \setminus F) S ∩ ( M ∖ F ) with F F F finite, because the other generators of τ a \tau_a τ a avoid a a a ; hence S ∖ O ⊆ F S \setminus O \subseteq F S ∖ O ⊆ F is finite, and S S S is compact. Now let K K K be compact in S S S and let K ⊆ ( S ∩ O 1 ) ∪ ( S ∩ O 2 ) K \subseteq(S \cap O_1) \cup(S \cap O_2) K ⊆ ( S ∩ O 1 ) ∪ ( S ∩ O 2 ) with O 1 , O 2 ∈ τ a O_1,O_2 \in\tau_a O 1 , O 2 ∈ τ a . If a ∈ K ∩ O 1 a \in K \cap O_1 a ∈ K ∩ O 1 , put K 1 = K ∩ O 1 K_1 = K \cap O_1 K 1 = K ∩ O 1 and K 2 = K ∖ O 1 K_2 = K \setminus O_1 K 2 = K ∖ O 1 . Then K ∖ O 1 K \setminus O_1 K ∖ O 1 is finite by the observation above applied to S = K S = K S = K , so K 2 K_2 K 2 is finite and compact, while K 1 = K ∖ K 2 K_1 = K \setminus K_2 K 1 = K ∖ K 2 is closed in the T 1 T_1 T 1 space K K K and hence compact. Moreover K 1 ⊆ O 1 K_1 \subseteq O_1 K 1 ⊆ O 1 and K 2 ⊆ K ∖ O 1 ⊆ O 2 K_2 \subseteq K \setminus O_1 \subseteq O_2 K 2 ⊆ K ∖ O 1 ⊆ O 2 , and K 1 ∪ K 2 = K K_1 \cup K_2 = K K 1 ∪ K 2 = K ; the case a ∈ K ∩ O 2 a \in K \cap O_2 a ∈ K ∩ O 2 is symmetric. If a ∉ K a \notin K a ∈ / K , then K K K is a compact subset of the Hausdorff space M ∖ { a } M \setminus\{a\} M ∖ { a } , hence compact Hausdorff; applying the argument of Proposition 3.8(ii) inside K K K we obtain compact F 1 ⊆ K ∩ O 1 F_1 \subseteq K \cap O_1 F 1 ⊆ K ∩ O 1 , F 2 ⊆ K ∩ O 2 F_2 \subseteq K \cap O_2 F 2 ⊆ K ∩ O 2 with K = F 1 ∪ F 2 K = F_1 \cup F_2 K = F 1 ∪ F 2 . In both cases the witnesses lie in S ∩ O j S \cap O_j S ∩ O j , so S ∈ W i n S \in\mathrm{W}_{\mathrm{in}} S ∈ W in .
P × Q ∉ W w P \times Q \notin\mathrm{W}_{\mathrm{w}} P × Q ∈ / W w . Since τ p , τ q ⊆ τ \tau_p,\tau_q \subseteq\tau τ p , τ q ⊆ τ , the diagonal map δ : M → P × Q \delta\colon M \to P \times Q δ : M → P × Q , δ ( x ) = ( x , x ) \delta(x) = (x,x) δ ( x ) = ( x , x ) , is continuous, and δ ( M ) \delta(M) δ ( M ) is compact. With U = M ∖ { p } U = M \setminus\{p\} U = M ∖ { p } and V = M ∖ { q } V = M \setminus\{q\} V = M ∖ { q } we have δ ( M ) ⊆ ( U × Q ) ∪ ( P × V ) \delta(M) \subseteq(U \times Q) \cup(P \times V) δ ( M ) ⊆ ( U × Q ) ∪ ( P × V ) : if x ≠ p x \ne p x = p then x ∈ U x \in U x ∈ U , and if x = p x = p x = p then x ∈ V x \in V x ∈ V . Suppose that L 1 , L 2 L_1,L_2 L 1 , L 2 are Lindelöf subsets of P × Q P \times Q P × Q with L 1 ⊆ U × Q L_1 \subseteq U \times Q L 1 ⊆ U × Q , L 2 ⊆ P × V L_2 \subseteq P \times V L 2 ⊆ P × V and δ ( M ) ⊆ L 1 ∪ L 2 \delta(M) \subseteq L_1 \cup L_2 δ ( M ) ⊆ L 1 ∪ L 2 . Then π P ( L 1 ) ⊆ U \pi_P(L_1) \subseteq U π P ( L 1 ) ⊆ U and π Q ( L 2 ) ⊆ V \pi_Q(L_2) \subseteq V π Q ( L 2 ) ⊆ V , both projections are Lindelöf, being continuous images, and for x ∈ M x \in M x ∈ M the point δ ( x ) \delta(x) δ ( x ) lies in L 1 L_1 L 1 or in L 2 L_2 L 2 , so x ∈ π P ( L 1 ) x \in\pi_P(L_1) x ∈ π P ( L 1 ) or x ∈ π Q ( L 2 ) x \in\pi_Q(L_2) x ∈ π Q ( L 2 ) ; thus M ⊆ π P ( L 1 ) ∪ π Q ( L 2 ) M \subseteq\pi_P(L_1) \cup\pi_Q(L_2) M ⊆ π P ( L 1 ) ∪ π Q ( L 2 ) , contradicting Lemma 4.1. □
Corollary 4.3. Let Z = P ⊔ Q Z = P \sqcup Q Z = P ⊔ Q be the topological sum of the spaces of Theorem 4.2. Then Z Z Z is a compact T 1 T_1 T 1 space, Z ∈ W i n Z \in\mathrm{W}_{\mathrm{in}} Z ∈ W in and Z 2 ∉ W w Z^2 \notin\mathrm{W}_{\mathrm{w}} Z 2 ∈ / W w . Consequently the full subcategories of the category of T 0 T_0 T 0 spaces determined by W w \mathrm{W}_{\mathrm{w}} W w and by W i \mathrm{W}_{\mathrm{i}} W i are not reflective.
Proof. Both P P P and Q Q Q are compact T 1 T_1 T 1 , hence so is their sum, and Z ∈ W i n Z \in\mathrm{W}_{\mathrm{in}} Z ∈ W in by Proposition 3.7. The subspace P × Q P \times Q P × Q is clopen in Z 2 Z^2 Z 2 , because P P P and Q Q Q are clopen in Z Z Z ; if Z 2 Z^2 Z 2 belonged to W w \mathrm{W}_{\mathrm{w}} W w , then P × Q P \times Q P × Q would belong to W w \mathrm{W}_{\mathrm{w}} W w by Proposition 3.5, contradicting Theorem 4.2.
For the last assertion, let W \mathrm{W} W be one of W w , W i \mathrm{W}_{\mathrm{w}},\mathrm{W}_{\mathrm{i}} W w , W i and suppose that the inclusion of the full subcategory of W \mathrm{W} W -spaces into the category of T 0 T_0 T 0 spaces has a reflection η : Z 2 → T \eta\colon Z^2 \to T η : Z 2 → T . Since Z ∈ W i n ⊆ W Z \in\mathrm{W}_{\mathrm{in}} \subseteq\mathrm{W} Z ∈ W in ⊆ W , each projection π j : Z 2 → Z \pi_j\colon Z^2 \to Z π j : Z 2 → Z factors through η \eta η , say π j = f j ∘ η \pi_j = f_j \circ\eta π j = f j ∘ η with f j : T → Z f_j\colon T \to Z f j : T → Z . Put r = ( f 1 , f 2 ) : T → Z 2 r = (f_1,f_2)\colon T \to Z^2 r = ( f 1 , f 2 ) : T → Z 2 . Then r ∘ η = ( π 1 , π 2 ) = id Z 2 r \circ\eta= (\pi_1,\pi_2) = \operatorname{id}_{Z^2} r ∘ η = ( π 1 , π 2 ) = id Z 2 and ( η ∘ r ) ∘ η = η = id T ∘ η (\eta\circ r) \circ\eta= \eta= \operatorname{id}_T \circ\eta ( η ∘ r ) ∘ η = η = id T ∘ η , so η ∘ r = id T \eta\circ r = \operatorname{id}_T η ∘ r = id T by the uniqueness part of the reflection property. Hence Z 2 ≅ T ∈ W Z^2 \cong T \in\mathrm{W} Z 2 ≅ T ∈ W , which is impossible because Z 2 ∉ W w Z^2 \notin\mathrm{W}_{\mathrm{w}} Z 2 ∈ / W w and W i ⊆ W w \mathrm{W}_{\mathrm{i}} \subseteq\mathrm{W}_{\mathrm{w}} W i ⊆ W w . □
The next example shows that W w \mathrm{W}_{\mathrm{w}} W w is not hereditary for arbitrary subspaces, and that W ∖ W i \mathrm{W} \setminus\mathrm{W}_{\mathrm{i}} W ∖ W i is nonempty.
Proposition 4.4. There is a locally supercompact space S S S with S ∈ W ∖ W i S \in\mathrm{W} \setminus\mathrm{W}_{\mathrm{i}} S ∈ W ∖ W i . In particular W w \mathrm{W}_{\mathrm{w}} W w is not hereditary with respect to arbitrary subspaces.
Proof. Let M M M be the space of Lemma 4.1, let I = O ( M ) I = \mathcal{O}(M) I = O ( M ) and let S = S I S = S^I S = S I be the product of I I I many copies of the Sierpiński space, which we represent by characteristic functions of subsets of I I I . For finite F ⊆ I F \subseteq I F ⊆ I put
[ F ] = { T ⊆ I : F ⊆ T } . [F] = \{T \subseteq I : F \subseteq T\}. [ F ] = { T ⊆ I : F ⊆ T } .
The family of all [ F ] [F] [ F ] is a base for the topology of S S S .
S S S is locally supercompact. The point F F F belongs to [ F ] [F] [ F ] , and every open cover of [ F ] [F] [ F ] contains a member O O O with F ∈ O F \in O F ∈ O ; since O O O is a union of basic sets, it contains some [ F ′ ] [F'] [ F ′ ] with F ′ ⊆ F F' \subseteq F F ′ ⊆ F , and then [ F ] ⊆ [ F ′ ] ⊆ O [F] \subseteq[F'] \subseteq O [ F ] ⊆ [ F ′ ] ⊆ O . Hence [ F ] [F] [ F ] is supercompact, and it is an open neighbourhood of F F F .
S ∈ W S \in\mathrm{W} S ∈ W . Let K ⊆ S K \subseteq S K ⊆ S be compact and let K ⊆ O 1 ∪ O 2 K \subseteq O_1 \cup O_2 K ⊆ O 1 ∪ O 2 with O i O_i O i open. For each x ∈ K x \in K x ∈ K choose i ( x ) ∈ { 1 , 2 } i(x) \in\{1,2\} i ( x ) ∈ { 1 , 2 } and a basic set [ F x ] ⊆ O i ( x ) [F_x] \subseteq O_{i(x)} [ F x ] ⊆ O i ( x ) . Finitely many of these cover K K K ; grouping them according to i ( x ) i(x) i ( x ) and taking unions, we obtain compact sets L i ⊆ O i L_i \subseteq O_i L i ⊆ O i with K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 . Hence S ∈ W S \in\mathrm{W} S ∈ W .
S ∉ W i S \notin\mathrm{W}_{\mathrm{i}} S ∈ / W i . The map e : M → S e\colon M \to S e : M → S , e ( x ) = { G ∈ I : x ∈ G } e(x) = \{G \in I : x \in G\} e ( x ) = { G ∈ I : x ∈ G } , satisfies e − 1 ( { { G } } ) = G e^{-1}(\{\{G\}\}) = G e − 1 ({{ G }}) = G for G ∈ I G \in I G ∈ I , and the sets G G G generate the topology of M M M ; since M M M is T 0 T_0 T 0 , e e e is an embedding. By Lemma 4.1 we have M ∉ W w M \notin\mathrm{W}_{\mathrm{w}} M ∈ / W w , hence e ( M ) ∉ W w e(M) \notin\mathrm{W}_{\mathrm{w}} e ( M ) ∈ / W w and a fortiori e ( M ) ∉ W i e(M) \notin\mathrm{W}_{\mathrm{i}} e ( M ) ∈ / W i . Since W i \mathrm{W}_{\mathrm{i}} W i is hereditary for arbitrary subspaces by Proposition 3.4, S ∉ W i S \notin\mathrm{W}_{\mathrm{i}} S ∈ / W i . □
:::
The product theorem for W w W_w W w (Theorem 3.14) requires one factor to be in W w W_w W w and the compact subspaces of the other to be Hausdorff. The following example shows that even in the presence of a compact Hausdorff factor the product may leave W i W_i W i .
Theorem 4.5. There are a hereditarily compact T 1 T_1 T 1 space E E E and a compact Hausdorff space Y Y Y such that E × Y ∈ W w ∖ W i E \times Y \in W_w \setminus W_i E × Y ∈ W w ∖ W i .
Proof. Let A A A , D D D and M M M be as in Lemma 4.1 . On the underlying set of M M M take as a subbase the family
{ N p ( F ) , N q ( G ) : F , G ⊆ A finite } , \{N_p(F), N_q(G) : F,G \subseteq A\text{ finite}\}, { N p ( F ) , N q ( G ) : F , G ⊆ A finite } ,
and denote the resulting space by E E E .
E E E is T 1 T_1 T 1 . For d = ( a , b ) ∈ D d=(a,b)\in D d = ( a , b ) ∈ D we have E ∖ { d } = N p ( { a } ) ∪ N q ( { b } ) E\setminus\{d\}=N_p(\{a\})\cup N_q(\{b\}) E ∖ { d } = N p ({ a }) ∪ N q ({ b }) , and E ∖ { p } = N q ( ∅ ) E\setminus\{p\}=N_q(\varnothing) E ∖ { p } = N q ( ∅ ) , E ∖ { q } = N p ( ∅ ) E\setminus\{q\}=N_p(\varnothing) E ∖ { q } = N p ( ∅ ) ; all these sets are open.
E E E is hereditarily compact. A basic open set of E E E contained in D D D is of the form R = ( A ∖ F ) × ( A ∖ G ) R=(A\setminus F)\times(A\setminus G) R = ( A ∖ F ) × ( A ∖ G ) with F , G F,G F , G finite. Let T ⊆ D T\subseteq D T ⊆ D be nonempty and let U \mathcal{U} U be a cover of T T T by open sets. Choose a point t ∈ T t\in T t ∈ T and a basic set R R R with t ∈ T ∩ R ⊆ U t\in T\cap R\subseteq U t ∈ T ∩ R ⊆ U for some U ∈ U U\in\mathcal{U} U ∈ U . Then
T ∖ R ⊆ ⋃ a ∈ F ( T ∩ ( { a } × A ) ) ∪ ⋃ b ∈ G ( T ∩ ( A × { b } ) ) , T\setminus R\subseteq\bigcup_{a\in F}(T\cap(\{a\}\times A))\cup\bigcup_{b\in G}(T\cap(A\times\{b\})), T ∖ R ⊆ a ∈ F ⋃ ( T ∩ ({ a } × A )) ∪ b ∈ G ⋃ ( T ∩ ( A × { b })) ,
and each of the finitely many pieces on the right carries the cofinite topology, hence is compact; together with U U U this gives a finite subcover of T T T . An arbitrary T ⊆ E T\subseteq E T ⊆ E adds at most the two points p , q p,q p , q to a subset of D D D , and is still compact: a member of a cover that contains p p p contains N p ( F ) N_p(F) N p ( F ) for some finite F F F , and what remains of T T T is contained in the finite union of rows F × A F\times A F × A , a compact set in the cofinite topology.
E ∈ W i n E\in W_{in} E ∈ W in . Since every subspace of E E E is compact, for K ⊆ E K\subseteq E K ⊆ E and open O 1 , O 2 O_1,O_2 O 1 , O 2 of E E E with K ⊆ O 1 ∪ O 2 K\subseteq O_1\cup O_2 K ⊆ O 1 ∪ O 2 the sets F 1 = K ∩ O 1 F_1=K\cap O_1 F 1 = K ∩ O 1 and F 2 = K ∖ O 1 F_2=K\setminus O_1 F 2 = K ∖ O 1 are compact, and they satisfy F j ⊆ O j F_j\subseteq O_j F j ⊆ O j and F 1 ∪ F 2 = K F_1\cup F_2=K F 1 ∪ F 2 = K .
The product. Let Y = D ⊔ { ∞ } Y=D\sqcup\{\infty\} Y = D ⊔ { ∞ } be the one-point compactification of the discrete space D D D ; then Y Y Y is compact Hausdorff. Define h : M → E × Y h:M\to E\times Y h : M → E × Y by h ( d ) = ( d , d ) h(d)=(d,d) h ( d ) = ( d , d ) for d ∈ D d\in D d ∈ D , h ( p ) = ( p , ∞ ) h(p)=(p,\infty) h ( p ) = ( p , ∞ ) and h ( q ) = ( q , ∞ ) h(q)=(q,\infty) h ( q ) = ( q , ∞ ) . The first coordinate is continuous because the topology of E E E is coarser than that of M M M ; the second coordinate pulls back an open O ⊆ D O\subseteq D O ⊆ D to O O O and Y ∖ F Y\setminus F Y ∖ F , for finite F ⊆ D F\subseteq D F ⊆ D , to M ∖ F M\setminus F M ∖ F .
Moreover h − 1 ( E × { d } ) = { d } h^{-1}(E\times\{d\})=\{d\} h − 1 ( E × { d }) = { d } and h − 1 ( N a ( F ) × Y ) = N a ( F ) h^{-1}(N_a(F)\times Y)=N_a(F) h − 1 ( N a ( F ) × Y ) = N a ( F ) for a = p , q a=p,q a = p , q , so the preimages of a subbase of E × Y E\times Y E × Y form a subbase of M M M and h h h is an embedding.
Since E ∈ W i n ⊆ W w E\in W_{in}\subseteq W_w E ∈ W in ⊆ W w and every compact subspace of Y Y Y is Hausdorff, Theorem 3.14 gives E × Y ∈ W w E\times Y\in W_w E × Y ∈ W w . On the other hand h ( M ) ≅ M ∉ W w h(M)\cong M\notin W_w h ( M ) ≅ M ∈ / W w by Lemma 4.1 , so h ( M ) ∉ W i h(M)\notin W_i h ( M ) ∈ / W i ; were E × Y E\times Y E × Y in W i W_i W i , its subspace h ( M ) h(M) h ( M ) would be in W i W_i W i by Proposition 3.4 . Hence E × Y ∉ W i E\times Y\notin W_i E × Y ∈ / W i . □
We now turn to function spaces. Recall that C k ( M 0 , S ) C_k(M_0,S) C k ( M 0 , S ) is homeomorphic to O k ( M 0 ) O_k(M_0) O k ( M 0 ) , through the map U ↦ χ U U\mapsto\chi_U U ↦ χ U .
Theorem 4.6. Let M 0 M_0 M 0 be the countably infinite compact KC-space of Proposition 3.9 , with M 0 ∈ W i M_0\in W_i M 0 ∈ W i . Then O k ( M 0 ) ≅ C k ( M 0 , S ) O_k(M_0)\cong C_k(M_0,S) O k ( M 0 ) ≅ C k ( M 0 , S ) is not sober. Moreover, taking for H H H the system of all irreducible sets, the family { M 0 ∖ { x } : x ∈ M 0 } \{M_0\setminus\{x\}:x\in M_0\} { M 0 ∖ { x } : x ∈ M 0 } is an irreducible open cover of M 0 M_0 M 0 no member of which contains M 0 M_0 M 0 , so M 0 M_0 M 0 is not H H H -supercompact in the sense of [4 Definition 5.1] , and C k ( M 0 , S ) C_k(M_0,S) C k ( M 0 , S ) does not satisfy property S.
Proof. Put U x = M 0 ∖ { x } U_x=M_0\setminus\{x\} U x = M 0 ∖ { x } and E = { U x : x ∈ M 0 } ⊆ O k ( M 0 ) \mathcal{E}=\{U_x:x\in M_0\}\subseteq O_k(M_0) E = { U x : x ∈ M 0 } ⊆ O k ( M 0 ) .
E \mathcal{E} E is irreducible. Let C , D ∈ K ( M 0 ) C,D\in\mathcal{K}(M_0) C , D ∈ K ( M 0 ) with C , D ≠ M 0 C,D\ne M_0 C , D = M 0 . Since M 0 M_0 M 0 is a KC-space, C C C and D D D are closed; since M 0 M_0 M 0 is hyperconnected and M 0 ∖ C M_0\setminus C M 0 ∖ C , M 0 ∖ D M_0\setminus D M 0 ∖ D are nonempty open sets, they intersect, so C ∪ D ≠ M 0 C\cup D\ne M_0 C ∪ D = M 0 . Therefore
E ∩ Φ ( C ) = { U x : x ∈ M 0 ∖ C } ≠ ∅ , \mathcal{E}\cap\Phi(C)=\{U_x:x\in M_0\setminus C\}\ne\varnothing, E ∩ Φ ( C ) = { U x : x ∈ M 0 ∖ C } = ∅ ,
E ∩ Φ ( C ) ∩ Φ ( D ) = { U x : x ∈ M 0 ∖ ( C ∪ D ) } ≠ ∅ . \mathcal{E}\cap\Phi(C)\cap\Phi(D)=\{U_x:x\in M_0\setminus(C\cup D)\}\ne\varnothing. E ∩ Φ ( C ) ∩ Φ ( D ) = { U x : x ∈ M 0 ∖ ( C ∪ D )} = ∅ .
As the sets Φ ( C ) \Phi(C) Φ ( C ) form a base of O k ( M 0 ) O_k(M_0) O k ( M 0 ) , this says exactly that E \mathcal{E} E is irreducible.
The closure of E \mathcal{E} E . A point U ∈ O ( M 0 ) U\in O(M_0) U ∈ O ( M 0 ) belongs to cl E \operatorname{cl}\mathcal{E} cl E if and only if every basic neighbourhood Φ ( K ) \Phi(K) Φ ( K ) of U U U meets E \mathcal{E} E , that is, if and only if for every K ∈ K ( M 0 ) K\in\mathcal{K}(M_0) K ∈ K ( M 0 ) with K ⊆ U K\subseteq U K ⊆ U there is x ∈ M 0 x\in M_0 x ∈ M 0 with K ⊆ M 0 ∖ { x } K\subseteq M_0\setminus\{x\} K ⊆ M 0 ∖ { x } . The last condition means K ≠ M 0 K\ne M_0 K = M 0 , so it holds for all K ⊆ U K\subseteq U K ⊆ U precisely when U ≠ M 0 U\ne M_0 U = M 0 . Hence
cl E = O ( M 0 ) ∖ { M 0 } . \operatorname{cl}\mathcal{E}=O(M_0)\setminus\{M_0\}. cl E = O ( M 0 ) ∖ { M 0 } .
Since points of M 0 M_0 M 0 are compact, Lemma 2.10 gives cl { U } = { W ∈ O ( M 0 ) : W ⊆ U } \operatorname{cl}\{U\}=\{W\in O(M_0):W\subseteq U\} cl { U } = { W ∈ O ( M 0 ) : W ⊆ U } for every U U U .
O k ( M 0 ) O_k(M_0) O k ( M 0 ) is not sober. Suppose that cl E = cl { G } \operatorname{cl}\mathcal{E}=\operatorname{cl}\{G\} cl E = cl { G } for some G G G . Then G ∈ cl { G } = cl E G\in\operatorname{cl}\{G\}=\operatorname{cl}\mathcal{E} G ∈ cl { G } = cl E , so G ≠ M 0 G\ne M_0 G = M 0 ; on the other hand each U ∈ cl E U\in\operatorname{cl}\mathcal{E} U ∈ cl E satisfies U ∈ cl { G } U\in\operatorname{cl}\{G\} U ∈ cl { G } , that is, U ⊆ G U\subseteq G U ⊆ G . Since M 0 M_0 M 0 has at least two points, every x ∈ M 0 x \in M_0 x ∈ M 0 lies in some U y U_y U y with y ≠ x y \ne x y = x , so M 0 = ⋃ cl E ⊆ G M_0 = \bigcup\operatorname{cl} \mathcal{E} \subseteq G M 0 = ⋃ cl E ⊆ G , which forces G = M 0 G = M_0 G = M 0 , a contradiction. Thus the irreducible closed set cl E \operatorname{cl} \mathcal{E} cl E has no generic point. Property S S S fails. The map χ : O k ( M 0 ) → C k ( M 0 , S ) \chi: \mathcal{O}_k(M_0) \to C_k(M_0,\mathbb{S}) χ : O k ( M 0 ) → C k ( M 0 , S ) , χ ( U ) = 1 U \chi(U) = 1_U χ ( U ) = 1 U , is a homeomorphism, since χ − 1 ( [ K , { 1 } ] ) = Φ ( K ) \chi^{-1}([K,\{1\}]) = \Phi(K) χ − 1 ([ K , { 1 }]) = Φ ( K ) for K ∈ K ( M 0 ) K \in\mathcal{K}(M_0) K ∈ K ( M 0 ) . Put F = χ ( E ) F = \chi(\mathcal{E}) F = χ ( E ) and g = 1 M 0 g = 1_{M_0} g = 1 M 0 . Then F F F is irreducible in C k ( M 0 , S ) C_k(M_0,\mathbb{S}) C k ( M 0 , S ) , and
⋃ f ∈ F f − 1 ( { 1 } ) = ⋃ x ∈ M 0 U x = M 0 = g − 1 ( { 1 } ) , \bigcup_{f \in F} f^{-1}(\{1\}) = \bigcup_{x \in M_0} U_x = M_0 = g^{-1}(\{1\}), f ∈ F ⋃ f − 1 ({ 1 }) = x ∈ M 0 ⋃ U x = M 0 = g − 1 ({ 1 }) ,
while the preimages of the only other open set ∅ \varnothing ∅ are empty. Thus the hypothesis (1) of property S S S is satisfied. But g ∈ [ M 0 , { 1 } ] g \in[M_0,\{1\}] g ∈ [ M 0 , { 1 }] , an open set, and [ M 0 , { 1 } ] ∩ F = ∅ [M_0,\{1\}] \cap F = \varnothing [ M 0 , { 1 }] ∩ F = ∅ because no U x U_x U x equals M 0 M_0 M 0 ; hence g ∉ cl F g \notin\operatorname{cl} F g ∈ / cl F and property S S S fails. □
The next proposition is the positive counterpart of the previous result: for sober Y Y Y and for topologies with the pointwise specialization order, property S S S is equivalent to sobriety.
Proposition 4.7. The following equivalence is due to Xu, Bao and Zhang [6 ] . Let Y Y Y be an H H H -sober space and let τ \tau τ be a topology on C ( X , Y ) C(X,Y) C ( X , Y ) finer than the topology of pointwise convergence. If the specialization order of τ \tau τ is the pointwise order, that is, f ≤ τ g f \leq_\tau g f ≤ τ g if and only if f ( x ) ≤ Y g ( x ) f(x) \leq_Y g(x) f ( x ) ≤ Y g ( x ) for all x ∈ X x \in X x ∈ X , then ( C ( X , Y ) , τ ) (C(X,Y),\tau) ( C ( X , Y ) , τ ) is H H H -sober if and only if τ \tau τ satisfies property S S S .
Proof. Let F ∈ H ( C ( X , Y ) , τ ) F \in H(C(X,Y),\tau) F ∈ H ( C ( X , Y ) , τ ) . For x ∈ X x \in X x ∈ X the evaluation e x : ( C ( X , Y ) , τ ) → Y e_x : (C(X,Y),\tau) \to Y e x : ( C ( X , Y ) , τ ) → Y , e x ( f ) = f ( x ) e_x(f) = f(x) e x ( f ) = f ( x ) , is continuous, so e x ( F ) ∈ H ( Y ) e_x(F) \in H(Y) e x ( F ) ∈ H ( Y ) and has a generic point g ( x ) g(x) g ( x ) by hypothesis. Define g : X → Y g : X \to Y g : X → Y by letting g ( x ) g(x) g ( x ) be this generic point. For V ∈ O ( Y ) V \in\mathcal{O}(Y) V ∈ O ( Y ) we have
g ( x ) ∈ V ⟺ V ∩ e x ( F ) ≠ ∅ ⟺ x ∈ ⋃ f ∈ F f − 1 ( V ) , g(x) \in V \Longleftrightarrow V \cap e_x(F) \ne\varnothing\Longleftrightarrow x \in\bigcup_{f \in F} f^{-1}(V), g ( x ) ∈ V ⟺ V ∩ e x ( F ) = ∅ ⟺ x ∈ f ∈ F ⋃ f − 1 ( V ) ,
so g − 1 ( V ) g^{-1}(V) g − 1 ( V ) is open and g g g is continuous. From e x ( F ) ⊆ cl Y { g ( x ) } e_x(F) \subseteq\operatorname{cl}_Y\{g(x)\} e x ( F ) ⊆ cl Y { g ( x )} we obtain f ( x ) ≤ g ( x ) f(x) \leq g(x) f ( x ) ≤ g ( x ) for all f ∈ F f \in F f ∈ F and x ∈ X x \in X x ∈ X , so F ⊆ cl τ { g } F \subseteq\operatorname{cl}_\tau\{g\} F ⊆ cl τ { g } by the order condition. If property S S S holds, then g ∈ cl τ F g \in\operatorname{cl}_\tau F g ∈ cl τ F , so cl τ { g } ⊆ cl τ F \operatorname{cl}_\tau\{g\} \subseteq\operatorname{cl}_\tau F cl τ { g } ⊆ cl τ F and therefore cl τ F = cl τ { g } \operatorname{cl}_\tau F = \operatorname{cl}_\tau\{g\} cl τ F = cl τ { g } . Uniqueness of the generic point follows from the T 0 T_0 T 0 property.
Conversely, let h h h be a generic point of F F F and let g ∈ C ( X , Y ) g \in C(X,Y) g ∈ C ( X , Y ) satisfy (1). For x ∈ X x \in X x ∈ X and V ∈ O ( Y ) V \in\mathcal{O}(Y) V ∈ O ( Y ) we have, since e x − 1 ( V ) e_x^{-1}(V) e x − 1 ( V ) is an open neighbourhood of h h h and h ∈ cl τ F h \in\operatorname{cl}_\tau F h ∈ cl τ F ,
x ∈ h − 1 ( V ) ⟺ h ∈ e x − 1 ( V ) ⟺ F ∩ e x − 1 ( V ) ≠ ∅ x \in h^{-1}(V) \Longleftrightarrow h \in e_x^{-1}(V) \Longleftrightarrow F \cap e_x^{-1}(V) \ne\varnothing x ∈ h − 1 ( V ) ⟺ h ∈ e x − 1 ( V ) ⟺ F ∩ e x − 1 ( V ) = ∅
⟺ x ∈ ⋃ f ∈ F f − 1 ( V ) , \Longleftrightarrow x \in\bigcup_{f \in F} f^{-1}(V), ⟺ x ∈ f ∈ F ⋃ f − 1 ( V ) ,
hence g − 1 ( V ) = h − 1 ( V ) g^{-1}(V) = h^{-1}(V) g − 1 ( V ) = h − 1 ( V ) for all V V V . As Y Y Y is T 0 T_0 T 0 , this forces g = h ∈ cl τ F g = h \in\operatorname{cl}_\tau F g = h ∈ cl τ F , which is (1). □
Both hypotheses in Proposition 4.7 are needed. That the sobriety of Y Y Y cannot be dropped is shown by the following example, in which the order condition does hold and the closure test is satisfied, yet the function space is not sober.
Example 4.8. Let X = { ∗ } X = \{\ast\} X = { ∗ } and Y = N ⊔ { ∞ } Y = \mathbb{N} \sqcup\{\infty\} Y = N ⊔ { ∞ } , and put T n = { m ∈ N : m ≥ n } ∪ { ∞ } T_n = \{m \in\mathbb{N}: m \geq n\} \cup\{\infty\} T n = { m ∈ N : m ≥ n } ∪ { ∞ } for n ∈ N n \in\mathbb{N} n ∈ N . Let σ = { ∅ } ∪ { T n : n ∈ N } \sigma= \{\varnothing\} \cup\{T_n : n \in\mathbb{N}\} σ = { ∅ } ∪ { T n : n ∈ N } , τ = σ ∪ { T n ∖ { ∞ } : n ∈ N } \tau= \sigma\cup\{T_n \setminus\{\infty\}: n \in\mathbb{N}\} τ = σ ∪ { T n ∖ { ∞ } : n ∈ N } .
Then C k ( X , Y ) C_k(X,Y) C k ( X , Y ) is homeomorphic to ( Y , τ ) (Y,\tau) ( Y , τ ) , its specialization order is the usual order, the closure test (1) holds, and ( Y , τ ) (Y,\tau) ( Y , τ ) is not sober.
Proof. The family σ \sigma σ is a topology, and τ \tau τ is obtained from it by adding the sets T n ∖ { ∞ } = { m : m ≥ n } T_n \setminus\{\infty\} = \{m : m \geq n\} T n ∖ { ∞ } = { m : m ≥ n } ; it is again a topology. The nonempty closed sets of ( Y , σ ) (Y,\sigma) ( Y , σ ) are the finite initial segments { 0 , … , n − 1 } \{0,\ldots,n-1\} { 0 , … , n − 1 } and Y Y Y itself, each of them the closure of its greatest element.
Since X X X is a singleton, C ( X , Y ) C(X,Y) C ( X , Y ) may be identified with Y Y Y , and (1) reads: if g ∈ V ⟺ F ∩ V ≠ ∅ g \in V \Longleftrightarrow F \cap V \ne\varnothing g ∈ V ⟺ F ∩ V = ∅ for every V ∈ τ V \in\tau V ∈ τ , then g ∈ cl τ F g \in\operatorname{cl}_\tau F g ∈ cl τ F . For g = m < ∞ g = m < \infty g = m < ∞ the condition gives F ∩ T m + 1 = ∅ F \cap T_{m+1} = \varnothing F ∩ T m + 1 = ∅ and F ∩ T m ≠ ∅ F \cap T_m \ne\varnothing F ∩ T m = ∅ , hence m ∈ F m \in F m ∈ F and g ∈ cl τ F g \in\operatorname{cl}_\tau F g ∈ cl τ F ; for g = ∞ g = \infty g = ∞ it gives F ∩ T n ≠ ∅ F \cap T_n \ne\varnothing F ∩ T n = ∅ for all n n n , and since the sets T n T_n T n form a neighbourhood base at ∞ \infty ∞ , this means ∞ ∈ cl τ F \infty\in\operatorname{cl}_\tau F ∞ ∈ cl τ F . So the closure test holds.
Any two nonempty τ \tau τ -open sets contain a common element of N \mathbb{N} N , so Y Y Y is irreducible as a closed set. But cl τ { m } = { 0 , … , m } \operatorname{cl}_\tau\{m\} = \{0,\ldots,m\} cl τ { m } = { 0 , … , m } and cl τ { ∞ } = { ∞ } \operatorname{cl}_\tau\{\infty\} = \{\infty\} cl τ { ∞ } = { ∞ } , so no point has Y Y Y as its closure and ( Y , τ ) (Y,\tau) ( Y , τ ) is not sober. Finally, the specialization order of τ \tau τ is the usual order on Y Y Y , which is the pointwise order, as cl τ { m } = { 0 , … , m } \operatorname{cl}_\tau\{m\} = \{0,\ldots,m\} cl τ { m } = { 0 , … , m } shows. □
We pass to the Vietoris constructions. The following elementary lemma is used repeatedly; it says that the union of a compact, respectively Lindelöf, family of compact saturated sets is again compact, respectively Lindelöf.
Lemma 4.9. Let C \mathcal{C} C be one of K , L \mathcal{K},\mathcal{L} K , L and let X X X be a space. If E ∈ C ( P S ( X ) ) \mathcal{E} \in\mathcal{C}(\mathrm{P}_{\mathrm{S}}(X)) E ∈ C ( P S ( X )) , then ⋃ E ∈ C ( X ) \bigcup\mathcal{E} \in\mathcal{C}(X) ⋃ E ∈ C ( X ) .
Proof. Let ⋃ E ⊆ ⋃ G \bigcup\mathcal{E} \subseteq\bigcup\mathcal{G} ⋃ E ⊆ ⋃ G with G ⊆ O ( X ) \mathcal{G} \subseteq\mathcal{O}(X) G ⊆ O ( X ) . Each C ∈ E C \in\mathcal{E} C ∈ E is compact, so from C ⊆ ⋃ G C \subseteq\bigcup\mathcal{G} C ⊆ ⋃ G we obtain a finite V ⊆ G \mathcal{V} \subseteq\mathcal{G} V ⊆ G with C ⊆ ⋃ V C \subseteq\bigcup\mathcal{V} C ⊆ ⋃ V , that is, C ∈ □ ( ⋃ V ) C \in\square(\bigcup\mathcal{V}) C ∈ □ ( ⋃ V ) . Hence E ⊆ ⋃ V ∈ [ G ] < ω □ ( ⋃ V ) \mathcal{E} \subseteq\bigcup_{\mathcal{V} \in[\mathcal{G}]^{<\omega}} \square(\bigcup\mathcal{V}) E ⊆ ⋃ V ∈ [ G ] < ω □ ( ⋃ V ) , and the sets □ ( ⋃ V ) \square(\bigcup\mathcal{V}) □ ( ⋃ V ) are open in P S ( X ) \mathrm{P}_{\mathrm{S}}(X) P S ( X ) . Since E \mathcal{E} E is compact, respectively Lindelöf, finitely many, respectively countably many, of them cover E \mathcal{E} E ; the corresponding finite, respectively countable, subfamilies of G \mathcal{G} G then cover ⋃ E \bigcup\mathcal{E} ⋃ E . □ \square □
Proposition 4.10. Let X X X be a space. Then P S ( X ) ∈ W w \mathrm{P}_{\mathrm{S}}(X) \in\mathrm{W}_{w} P S ( X ) ∈ W w implies X ∈ W w X \in\mathrm{W}_{w} X ∈ W w , and P S ( X ) ∈ W i \mathrm{P}_{\mathrm{S}}(X) \in\mathrm{W}_{i} P S ( X ) ∈ W i or P H ( X ) ∈ W i \mathrm{P}_{\mathrm{H}}(X) \in\mathrm{W}_{i} P H ( X ) ∈ W i implies X ∈ W i X \in\mathrm{W}_{i} X ∈ W i .
Proof. The maps s : X → P S ( X ) s: X \to\mathrm{P}_{\mathrm{S}}(X) s : X → P S ( X ) , s ( x ) = ↑ x s(x) = \mathord{\uparrow}x s ( x ) = ↑ x , and h : X → P H ( X ) h: X \to\mathrm{P}_{\mathrm{H}}(X) h : X → P H ( X ) , h ( x ) = cl { x } h(x) = \operatorname{cl}\{x\} h ( x ) = cl { x } , are well defined: ↑ x \mathord{\uparrow}x ↑ x is compact because every open set containing x x x contains ↑ x \mathord{\uparrow}x ↑ x , and it is saturated, while cl { x } \operatorname{cl}\{x\} cl { x } is a nonempty closed set. They satisfy s − 1 ( □ U ) = U s^{-1}(\square U) = U s − 1 ( □ U ) = U and h − 1 ( ⟨ U ⟩ ) = U h^{-1}(\langle U\rangle) = U h − 1 (⟨ U ⟩) = U : indeed ↑ x ⊆ U \mathord{\uparrow}x \subseteq U ↑ x ⊆ U if and only if x ∈ U x \in U x ∈ U , and cl { x } ∩ U ≠ ∅ \operatorname{cl}\{x\} \cap U \ne\varnothing cl { x } ∩ U = ∅ if and only if x ∈ U x \in U x ∈ U , because open sets are upward closed for the specialization order. Since X X X is T 0 T_{0} T 0 , both maps are injective, hence embeddings.
The statements about W i \mathrm{W}_{i} W i follow from Proposition 3.4 : if P S ( X ) ∈ W i \mathrm{P}_{\mathrm{S}}(X) \in\mathrm{W}_{i} P S ( X ) ∈ W i , then its subspace s ( X ) ≅ X s(X) \cong X s ( X ) ≅ X belongs to W i \mathrm{W}_{i} W i , and similarly for P H ( X ) \mathrm{P}_{\mathrm{H}}(X) P H ( X ) .
Now assume P S ( X ) ∈ W w \mathrm{P}_{\mathrm{S}}(X) \in\mathrm{W}_{w} P S ( X ) ∈ W w , and take K ∈ K ( X ) K \in\mathcal{K}(X) K ∈ K ( X ) , U 1 , U 2 ∈ O ( X ) U_{1}, U_{2} \in\mathcal{O}(X) U 1 , U 2 ∈ O ( X ) with K ⊆ U 1 ∪ U 2 K \subseteq U_{1} \cup U_{2} K ⊆ U 1 ∪ U 2 . Then s ( K ) s(K) s ( K ) is compact and s ( K ) ⊆ □ U 1 ∪ □ U 2 s(K) \subseteq\square U_{1} \cup\square U_{2} s ( K ) ⊆ □ U 1 ∪ □ U 2 , so there are E i ∈ L ( P S ( X ) ) E_{i} \in\mathcal{L}(\mathrm{P}_{\mathrm{S}}(X)) E i ∈ L ( P S ( X )) with E i ⊆ □ U i E_{i} \subseteq\square U_{i} E i ⊆ □ U i and s ( K ) ⊆ E 1 ∪ E 2 s(K) \subseteq E_{1} \cup E_{2} s ( K ) ⊆ E 1 ∪ E 2 . For x ∈ K x \in K x ∈ K the set ↑ x = s ( x ) \mathord{\uparrow}x = s(x) ↑ x = s ( x ) lies in E 1 E_{1} E 1 or in E 2 E_{2} E 2 , and in either case x ∈ ⋃ E i x \in\bigcup E_{i} x ∈ ⋃ E i ; hence K ⊆ ⋃ E 1 ∪ ⋃ E 2 K \subseteq\bigcup E_{1} \cup\bigcup E_{2} K ⊆ ⋃ E 1 ∪ ⋃ E 2 , and ⋃ E i ⊆ U i \bigcup E_{i} \subseteq U_{i} ⋃ E i ⊆ U i . By Lemma 4.9 both unions are Lindelöf, so X ∈ W w X \in\mathrm{W}_{w} X ∈ W w . □ \square □
Lemma 4.11. Let A ⊆ X A \subseteq X A ⊆ X and B ⊆ Y B \subseteq Y B ⊆ Y be compact and let O ⊆ X × Y O \subseteq X \times Y O ⊆ X × Y be open with A × B ⊆ O A \times B \subseteq O A × B ⊆ O . Then there are open sets G ⊇ A G \supseteq A G ⊇ A in X X X and H ⊇ B H \supseteq B H ⊇ B in Y Y Y with G × H ⊆ O G \times H \subseteq O G × H ⊆ O .
Proof. Fix a ∈ A a \in A a ∈ A . For each b ∈ B b \in B b ∈ B choose an open rectangle G a , b × H a , b ⊆ O G_{a,b} \times H_{a,b} \subseteq O G a , b × H a , b ⊆ O with ( a , b ) ∈ G a , b × H a , b (a,b) \in G_{a,b} \times H_{a,b} ( a , b ) ∈ G a , b × H a , b . The sets H a , b H_{a,b} H a , b cover the compact set B B B , so finitely many of them, say H a , b 1 , … , H a , b m H_{a,b_{1}}, \ldots, H_{a,b_{m}} H a , b 1 , … , H a , b m , cover B B B . Put G a = ⋃ i = 1 m G a , b i G_{a} = \bigcup_{i=1}^{m} G_{a,b_{i}} G a = ⋃ i = 1 m G a , b i and H a = ⋃ i = 1 m H a , b i H_{a} = \bigcup_{i=1}^{m} H_{a,b_{i}} H a = ⋃ i = 1 m H a , b i . Then G a G_{a} G a is an open neighbourhood of a a a , H a ⊇ B H_{a} \supseteq B H a ⊇ B is open, and G a × H a ⊆ O G_{a} \times H_{a} \subseteq O G a × H a ⊆ O : for y ∈ H a y \in H_{a} y ∈ H a there is i i i with y ∈ H a , b i y \in H_{a,b_{i}} y ∈ H a , b i , and x ∈ G a ⊆ G a , b i x \in G_{a} \subseteq G_{a,b_{i}} x ∈ G a ⊆ G a , b i gives ( x , y ) ∈ G a , b i × H a , b i ⊆ O (x,y) \in G_{a,b_{i}} \times H_{a,b_{i}} \subseteq O ( x , y ) ∈ G a , b i × H a , b i ⊆ O .
The sets G a G_{a} G a cover the compact set A A A , so finitely many of them, say G a 1 , … , G a n G_{a_{1}}, \ldots, G_{a_{n}} G a 1 , … , G a n , cover A A A . Put G = ⋃ j = 1 n G a j G = \bigcup_{j=1}^{n} G_{a_{j}} G = ⋃ j = 1 n G a j and H = ⋂ j = 1 n H a j H = \bigcap_{j=1}^{n} H_{a_{j}} H = ⋂ j = 1 n H a j . Then G ⊇ A G \supseteq A G ⊇ A and H ⊇ B H \supseteq B H ⊇ B are open, and G × H ⊆ O G \times H \subseteq O G × H ⊆ O : if x ∈ G a j x \in G_{a_{j}} x ∈ G a j and y ∈ H ⊆ H a j y \in H \subseteq H_{a_{j}} y ∈ H ⊆ H a j , then ( x , y ) ∈ G a j × H a j ⊆ O (x,y) \in G_{a_{j}} \times H_{a_{j}} \subseteq O ( x , y ) ∈ G a j × H a j ⊆ O . □ \square □
Theorem 4.12. Let W ∈ { W , W w } \mathrm{W} \in\{\mathrm{W}, \mathrm{W}_{w}\} W ∈ { W , W w } and let X , Y X, Y X , Y be spaces with P S ( X ⊔ Y ) ∈ W \mathrm{P}_{\mathrm{S}}(X \sqcup Y) \in\mathrm{W} P S ( X ⊔ Y ) ∈ W . Then X × Y ∈ W X \times Y \in\mathrm{W} X × Y ∈ W .
Proof. We may assume X , Y ≠ ∅ X,Y \ne\varnothing X , Y = ∅ , the other case being trivial. Put T = X ⊔ Y T = X \sqcup Y T = X ⊔ Y and
K = P S ( T ) ∖ ( □ X ∪ □ Y ) . \mathcal{K} = \mathrm{P}_{\mathrm{S}}(T) \setminus(\square X \cup\square Y). K = P S ( T ) ∖ ( □ X ∪ □ Y ) .
Since X , Y X,Y X , Y are clopen in T T T , every C ∈ P S ( T ) C \in\mathrm{P}_{\mathrm{S}}(T) C ∈ P S ( T ) satisfies C = ( C ∩ X ) ∪ ( C ∩ Y ) C = (C \cap X) \cup(C \cap Y) C = ( C ∩ X ) ∪ ( C ∩ Y ) with C ∩ X C \cap X C ∩ X saturated in X X X and C ∩ Y C \cap Y C ∩ Y saturated in Y Y Y . Hence C ∈ K C \in\mathcal{K} C ∈ K if and only if C ∩ X ≠ ∅ C \cap X \ne\varnothing C ∩ X = ∅ and C ∩ Y ≠ ∅ C \cap Y \ne\varnothing C ∩ Y = ∅ , and in that case C ∩ X ∈ P S ( X ) C \cap X \in\mathrm{P}_{\mathrm{S}}(X) C ∩ X ∈ P S ( X ) and C ∩ Y ∈ P S ( Y ) C \cap Y \in\mathrm{P}_{\mathrm{S}}(Y) C ∩ Y ∈ P S ( Y ) .
K \mathcal{K} K is homeomorphic to P S ( X ) × P S ( Y ) \mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) P S ( X ) × P S ( Y ) . Define q : K → P S ( X ) × P S ( Y ) q: \mathcal{K} \to\mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) q : K → P S ( X ) × P S ( Y ) by q ( C ) = ( C ∩ X , C ∩ Y ) q(C) = (C \cap X, C \cap Y) q ( C ) = ( C ∩ X , C ∩ Y ) and p : P S ( X ) × P S ( Y ) → K p: \mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) \to\mathcal{K} p : P S ( X ) × P S ( Y ) → K by p ( A , B ) = A ⊔ B p(A,B) = A \sqcup B p ( A , B ) = A ⊔ B . Both maps are well defined and mutually inverse. For open G ⊆ X G \subseteq X G ⊆ X and H ⊆ Y H \subseteq Y H ⊆ Y ,
q − 1 ( □ G × □ H ) = { C ∈ K : C ∩ X ⊆ G , C ∩ Y ⊆ H } = K ∩ □ ( G ∪ H ) , q^{-1}(\square G \times\square H) = \{C \in\mathcal{K}: C \cap X \subseteq G,\ C \cap Y \subseteq H\}
= \mathcal{K} \cap\square(G \cup H), q − 1 ( □ G × □ H ) = { C ∈ K : C ∩ X ⊆ G , C ∩ Y ⊆ H } = K ∩ □ ( G ∪ H ) ,
which is open in K \mathcal{K} K , so q q q is continuous. Conversely, for open W ⊆ T W \subseteq T W ⊆ T write W = ( W ∩ X ) ∪ ( W ∩ Y ) W = (W \cap X) \cup(W \cap Y) W = ( W ∩ X ) ∪ ( W ∩ Y ) ; then q ( K ∩ □ W ) = □ ( W ∩ X ) × □ ( W ∩ Y ) q(\mathcal{K} \cap\square W) = \square(W \cap X) \times\square(W \cap Y) q ( K ∩ □ W ) = □ ( W ∩ X ) × □ ( W ∩ Y ) , so q q q maps basic open sets to basic open sets and is an open map. Being a continuous open bijection, q q q is a homeomorphism.
Since □ X \square X □ X and □ Y \square Y □ Y are basic open sets of P S ( T ) \mathrm{P}_{\mathrm{S}}(T) P S ( T ) , the set K \mathcal{K} K is closed in P S ( T ) \mathrm{P}_{\mathrm{S}}(T) P S ( T ) . Both W \mathrm{W} W and W w \mathrm{W}_{w} W w are hereditary for closed subspaces by Proposition 3.5 , so K ∈ W \mathcal{K} \in\mathrm{W} K ∈ W , and hence P S ( X ) × P S ( Y ) ∈ W \mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) \in\mathrm{W} P S ( X ) × P S ( Y ) ∈ W .
From P S ( X ) × P S ( Y ) \mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) P S ( X ) × P S ( Y ) to X × Y X \times Y X × Y . The map m : P S ( X ) × P S ( Y ) → P S ( X × Y ) m: \mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) \to\mathrm{P}_{\mathrm{S}}(X \times Y) m : P S ( X ) × P S ( Y ) → P S ( X × Y ) , m ( A , B ) = A × B m(A,B) = A \times B m ( A , B ) = A × B , is well defined and continuous. To see the latter, take a basic open set □ O \square O □ O of P S ( X × Y ) \mathrm{P}_{\mathrm{S}}(X \times Y) P S ( X × Y ) and ( A , B ) ∈ m − 1 ( □ O ) (A,B) \in m^{-1}(\square O) ( A , B ) ∈ m − 1 ( □ O ) , so that A × B ⊆ O A \times B \subseteq O A × B ⊆ O . By Lemma 4.11 there are open G ⊇ A G \supseteq A G ⊇ A and H ⊇ B H \supseteq B H ⊇ B with G × H ⊆ O G \times H \subseteq O G × H ⊆ O ; then ( A , B ) ∈ □ G × □ H (A,B) \in\square G \times\square H ( A , B ) ∈ □ G × □ H and every ( A ′ , B ′ ) ∈ □ G × □ H (A',B') \in\square G \times\square H ( A ′ , B ′ ) ∈ □ G × □ H satisfies A ′ × B ′ ⊆ G × H ⊆ O A' \times B' \subseteq G \times H \subseteq O A ′ × B ′ ⊆ G × H ⊆ O . Hence □ G × □ H ⊆ m − 1 ( □ O ) \square G \times\square H \subseteq m^{-1}(\square O) □ G × □ H ⊆ m − 1 ( □ O ) .
Now let e : X × Y → P S ( X ) × P S ( Y ) e: X \times Y \to\mathrm{P}_{\mathrm{S}}(X) \times\mathrm{P}_{\mathrm{S}}(Y) e : X × Y → P S ( X ) × P S ( Y ) , e ( x , y ) = ( ↑ x , ↑ y ) e(x,y) = (\mathord{\uparrow}x,\mathord{\uparrow}y) e ( x , y ) = ( ↑ x , ↑ y ) . For open O ⊆ X × Y O \subseteq X \times Y O ⊆ X × Y the set O O O is upward closed, so ( m ∘ e ) − 1 ( □ O ) = { ( x , y ) : ↑ x × ↑ y ⊆ O } = O (m \circ e)^{-1}(\square O) = \{(x,y): \mathord{\uparrow}x \times\mathord{\uparrow}y \subseteq O\} = O ( m ∘ e ) − 1 ( □ O ) = {( x , y ) : ↑ x × ↑ y ⊆ O } = O ; in particular m ∘ e m \circ e m ∘ e is continuous. Let K ∈ K ( X × Y ) K \in\mathcal{K}(X \times Y) K ∈ K ( X × Y )
::: and let K ⊆ O 1 ∪ O 2 K \subseteq O_1 \cup O_2 K ⊆ O 1 ∪ O 2 with O i O_i O i open. Then e ( K ) e(K) e ( K ) is compact and e ( K ) ⊆ m − 1 ( □ O 1 ) ∪ m − 1 ( □ O 2 ) e(K) \subseteq m^{-1}(\Box O_1) \cup m^{-1}(\Box O_2) e ( K ) ⊆ m − 1 ( □ O 1 ) ∪ m − 1 ( □ O 2 ) , so by P S ( X ) × P S ( Y ) ∈ W P_S(X) \times P_S(Y) \in\mathsf{W} P S ( X ) × P S ( Y ) ∈ W there are E i ∈ C ( P S ( X ) × P S ( Y ) ) E_i \in\mathcal{C}(P_S(X) \times P_S(Y)) E i ∈ C ( P S ( X ) × P S ( Y )) with E i ⊆ m − 1 ( □ O i ) E_i \subseteq m^{-1}(\Box O_i) E i ⊆ m − 1 ( □ O i ) and e ( K ) ⊆ E 1 ∪ E 2 e(K) \subseteq E_1 \cup E_2 e ( K ) ⊆ E 1 ∪ E 2 , where C = K \mathcal{C} = \mathcal{K} C = K if W = W \mathsf{W} = \mathsf{W} W = W and C = L \mathcal{C} = \mathcal{L} C = L if W = W w \mathsf{W} = \mathsf{W}_w W = W w . Put
L i = ⋃ m ( E i ) = ⋃ ( A , B ) ∈ E i ( A × B ) . L_i = \bigcup m(E_i) = \bigcup_{(A,B)\in E_i}(A \times B). L i = ⋃ m ( E i ) = ( A , B ) ∈ E i ⋃ ( A × B ) .
Each ( A , B ) ∈ E i (A,B) \in E_i ( A , B ) ∈ E i satisfies A × B ⊆ O i A \times B \subseteq O_i A × B ⊆ O i , so L i ⊆ O i L_i \subseteq O_i L i ⊆ O i ; the family m ( E i ) m(E_i) m ( E i ) is compact, respectively Lindelöf, because m m m is continuous, and its members are compact, so Lemma 4.9 shows that L i L_i L i is compact, respectively Lindelöf. Finally, for ( x , y ) ∈ K (x,y) \in K ( x , y ) ∈ K the pair e ( x , y ) e(x,y) e ( x , y ) lies in E i E_i E i for some i i i , whence ( x , y ) ∈ ↑ x × ↑ y ⊆ L i (x,y) \in\mathord{\uparrow}x \times\mathord{\uparrow}y \subseteq L_i ( x , y ) ∈ ↑ x × ↑ y ⊆ L i . Thus K ⊆ L 1 ∪ L 2 K \subseteq L_1 \cup L_2 K ⊆ L 1 ∪ L 2 and X × Y ∈ W X \times Y \in\mathsf{W} X × Y ∈ W . □ \square □
Corollary 4.13. There is a countably infinite compact T 1 T_1 T 1 space T T T with T ∈ W T \in\mathsf{W} T ∈ W and P S ( T ) ∉ W P_S(T) \notin\mathsf{W} P S ( T ) ∈ / W ; hence the answer to [4 Question 8.5 (W1)] is negative.
Proof. By [4 Theorem 3.16] there are countably infinite compact T 1 T_1 T 1 spaces X , Y X,Y X , Y with X , Y ∈ W X,Y \in\mathsf{W} X , Y ∈ W but X × Y ∉ W X \times Y \notin\mathsf{W} X × Y ∈ / W . Let T = X ⊔ Y T = X \sqcup Y T = X ⊔ Y ; it is countable, and it is compact and T 1 T_1 T 1 because both summands are clopen, compact and T 1 T_1 T 1 . We verify that T ∈ W T \in\mathsf{W} T ∈ W . Let K ∈ K ( T ) K \in\mathcal{K}(T) K ∈ K ( T ) and let U 1 , U 2 ∈ O ( T ) U_1,U_2 \in\mathcal{O}(T) U 1 , U 2 ∈ O ( T ) with K ⊆ U 1 ∪ U 2 K \subseteq U_1 \cup U_2 K ⊆ U 1 ∪ U 2 . Since X , Y X,Y X , Y are clopen, K ∩ X K \cap X K ∩ X and K ∩ Y K \cap Y K ∩ Y are compact and K = ( K ∩ X ) ∪ ( K ∩ Y ) K = (K \cap X) \cup(K \cap Y) K = ( K ∩ X ) ∪ ( K ∩ Y ) ; moreover K ∩ X ⊆ ( U 1 ∩ X ) ∪ ( U 2 ∩ X ) K \cap X \subseteq(U_1 \cap X) \cup(U_2 \cap X) K ∩ X ⊆ ( U 1 ∩ X ) ∪ ( U 2 ∩ X ) , so X ∈ W X \in\mathsf{W} X ∈ W gives compact sets F j , X ⊆ U j ∩ X F_{j,X} \subseteq U_j \cap X F j , X ⊆ U j ∩ X with K ∩ X ⊆ F 1 , X ∪ F 2 , X K \cap X \subseteq F_{1,X} \cup F_{2,X} K ∩ X ⊆ F 1 , X ∪ F 2 , X , and similarly Y ∈ W Y \in\mathsf{W} Y ∈ W gives compact F j , Y ⊆ U j ∩ Y F_{j,Y} \subseteq U_j \cap Y F j , Y ⊆ U j ∩ Y with K ∩ Y ⊆ F 1 , Y ∪ F 2 , Y K \cap Y \subseteq F_{1,Y} \cup F_{2,Y} K ∩ Y ⊆ F 1 , Y ∪ F 2 , Y . Put F j = F j , X ∪ F j , Y F_j = F_{j,X} \cup F_{j,Y} F j = F j , X ∪ F j , Y . Each F j F_j F j is compact, being a finite union of compact subsets of T T T ; moreover F j ⊆ U j F_j \subseteq U_j F j ⊆ U j and
K ⊆ ( F 1 , X ∪ F 2 , X ) ∪ ( F 1 , Y ∪ F 2 , Y ) = F 1 ∪ F 2 . K \subseteq(F_{1,X} \cup F_{2,X}) \cup(F_{1,Y} \cup F_{2,Y}) = F_1 \cup F_2. K ⊆ ( F 1 , X ∪ F 2 , X ) ∪ ( F 1 , Y ∪ F 2 , Y ) = F 1 ∪ F 2 .
So T ∈ W T \in\mathsf{W} T ∈ W . Finally, Theorem 4.12 with W = W \mathsf{W} = \mathsf{W} W = W and T = X ⊔ Y T = X \sqcup Y T = X ⊔ Y says that P S ( T ) ∈ W P_S(T) \in\mathsf{W} P S ( T ) ∈ W would force X × Y ∈ W X \times Y \in\mathsf{W} X × Y ∈ W ; since X × Y ∉ W X \times Y \notin\mathsf{W} X × Y ∈ / W , we obtain P S ( T ) ∉ W P_S(T) \notin\mathsf{W} P S ( T ) ∈ / W . □ \square □
The next theorem shows that the upper and the lower Vietoris constructions both fail W w \mathsf{W}_w W w over a compact T 1 T_1 T 1 space of the previous type, even though the space itself lies in W i n \mathsf{W}_{in} W in .
Theorem 4.14. There is a compact T 1 T_1 T 1 space Z ∈ W i n Z \in\mathsf{W}_{in} Z ∈ W in with P S ( Z ) ∉ W w P_S(Z) \notin\mathsf{W}_w P S ( Z ) ∈ / W w and P H ( Z ) ∉ W w P_H(Z) \notin\mathsf{W}_w P H ( Z ) ∈ / W w .
Proof. Take Z = P ⊔ Q Z = P \sqcup Q Z = P ⊔ Q from Corollary 4.3 and write S P , S Q S_P,S_Q S P , S Q for the two copies in Z Z Z of a set S ⊆ M S \subseteq M S ⊆ M . Define t : M → P S ( Z ) t:M \to P_S(Z) t : M → P S ( Z ) by t ( x ) = { x P , x Q } t(x) = \{x_P,x_Q\} t ( x ) = { x P , x Q } and put T = t ( M ) \mathcal{T} = t(M) T = t ( M ) . Since Z Z Z is T 1 T_1 T 1 , the set t ( x ) t(x) t ( x ) is a nonempty closed set, hence a compact saturated set. For open G , H ⊆ M G,H \subseteq M G , H ⊆ M ,
t − 1 ( □ ( G P ⊔ H Q ) ) = G ∩ H , t − 1 ( ⟨ G P ⊔ H Q ⟩ ) = G ∪ H . (4) t^{-1}(\Box(G_P \sqcup H_Q)) = G \cap H,\qquad t^{-1}(\langle G_P \sqcup H_Q\rangle) = G \cup H. \tag*{(4)} t − 1 ( □ ( G P ⊔ H Q )) = G ∩ H , t − 1 (⟨ G P ⊔ H Q ⟩) = G ∪ H . (4)
The right-hand sides are open in M M M , so T \mathcal{T} T is compact both in P S ( Z ) P_S(Z) P S ( Z ) and in P H ( Z ) P_H(Z) P H ( Z ) .
The upper Vietoris topology. Let U = M ∖ { p } U = M \setminus\{p\} U = M ∖ { p } and V = M ∖ { q } V = M \setminus\{q\} V = M ∖ { q } . By (4), T ⊆ □ ( U P ⊔ Q ) ∪ □ ( P ⊔ V Q ) \mathcal{T} \subseteq\Box(U_P \sqcup Q) \cup\Box(P \sqcup V_Q) T ⊆ □ ( U P ⊔ Q ) ∪ □ ( P ⊔ V Q ) . Suppose that E 1 , E 2 ∈ L ( P S ( Z ) ) E_1,E_2 \in\mathcal{L}(P_S(Z)) E 1 , E 2 ∈ L ( P S ( Z )) satisfy T ⊆ E 1 ∪ E 2 \mathcal{T} \subseteq E_1 \cup E_2 T ⊆ E 1 ∪ E 2 with E 1 ⊆ □ ( U P ⊔ Q ) E_1 \subseteq\Box(U_P \sqcup Q) E 1 ⊆ □ ( U P ⊔ Q ) and E 2 ⊆ □ ( P ⊔ V Q ) E_2 \subseteq\Box(P \sqcup V_Q) E 2 ⊆ □ ( P ⊔ V Q ) . Put
B = { x : x P ∈ ⋃ E 1 } , E = { x : x Q ∈ ⋃ E 2 } . B = \{x : x_P \in\bigcup E_1\},\qquad E = \{x : x_Q \in\bigcup E_2\}. B = { x : x P ∈ ⋃ E 1 } , E = { x : x Q ∈ ⋃ E 2 } .
By Lemma 4.9 the set ⋃ E i \bigcup E_i ⋃ E i is Lindelöf in Z Z Z ; since P P P and Q Q Q are clopen in Z Z Z , the trace P ∩ ⋃ E 1 P \cap\bigcup E_1 P ∩ ⋃ E 1 is closed in ⋃ E 1 \bigcup E_1 ⋃ E 1 and hence Lindelöf, and B B B is its homeomorphic image under x P ↦ x x_P \mapsto x x P ↦ x . Thus B ∈ L ( U ) B \in\mathcal{L}(U) B ∈ L ( U ) and, symmetrically, E ∈ L ( V ) E \in\mathcal{L}(V) E ∈ L ( V ) . Now t ( x ) ∈ E i t(x) \in E_i t ( x ) ∈ E i implies x ∈ B x \in B x ∈ B for i = 1 i=1 i = 1 and x ∈ E x \in E x ∈ E for i = 2 i=2 i = 2 , so T ⊆ E 1 ∪ E 2 \mathcal{T} \subseteq E_1 \cup E_2 T ⊆ E 1 ∪ E 2 gives M = B ∪ E M = B \cup E M = B ∪ E , contradicting Lemma 4.1. Hence P S ( Z ) ∉ W w P_S(Z) \notin\mathsf{W}_w P S ( Z ) ∈ / W w .
The lower Vietoris topology. Here T ⊆ ⟨ U P ⟩ ∪ ⟨ V Q ⟩ \mathcal{T} \subseteq\langle U_P\rangle\cup\langle V_Q\rangle T ⊆ ⟨ U P ⟩ ∪ ⟨ V Q ⟩ . Suppose that E 1 ⊆ ⟨ U P ⟩ E_1 \subseteq\langle U_P\rangle E 1 ⊆ ⟨ U P ⟩ and E 2 ⊆ ⟨ V Q ⟩ E_2 \subseteq\langle V_Q\rangle E 2 ⊆ ⟨ V Q ⟩ are Lindelöf subsets of P H ( Z ) P_H(Z) P H ( Z ) with T ⊆ E 1 ∪ E 2 \mathcal{T} \subseteq E_1 \cup E_2 T ⊆ E 1 ∪ E 2 , and put B = D ∩ t − 1 ( E 1 ) B = D \cap t^{-1}(E_1) B = D ∩ t − 1 ( E 1 ) , E = D ∩ t − 1 ( E 2 ) E = D \cap t^{-1}(E_2) E = D ∩ t − 1 ( E 2 ) .
Fix b ∈ A b \in A b ∈ A and consider the column C b = A × { b } ⊆ U C_b = A \times\{b\} \subseteq U C b = A × { b } ⊆ U . Since
⟨ U P ⟩ = ⟨ ( U ∖ C b ) P ⟩ ∪ ⋃ d ∈ C b { ⟨ d P ⟩ } , \langle U_P\rangle= \langle(U \setminus C_b)_P\rangle\cup\bigcup_{d\in C_b}\{\langle d_P\rangle\}, ⟨ U P ⟩ = ⟨( U ∖ C b ) P ⟩ ∪ d ∈ C b ⋃ {⟨ d P ⟩} ,
and E 1 E_1 E 1 is Lindelöf, there is a countable J b ⊆ C b J_b \subseteq C_b J b ⊆ C b such that the sets on the right with d d d ranging over J b J_b J b still cover E 1 E_1 E 1 . If d ∈ B ∩ C b d \in B \cap C_b d ∈ B ∩ C b , then t ( d ) ∈ E 1 t(d) \in E_1 t ( d ) ∈ E 1 ; but t ( d ) t(d) t ( d ) is disjoint from ( U ∖ C b ) P (U \setminus C_b)_P ( U ∖ C b ) P , and t ( d ) ∩ { c P } ≠ ∅ t(d) \cap\{c_P\} \ne\varnothing t ( d ) ∩ { c P } = ∅ only for c = d c=d c = d . Hence d ∈ J b d \in J_b d ∈ J b , and B ∩ C b ⊆ J b B \cap C_b \subseteq J_b B ∩ C b ⊆ J b is countable. Symmetrically, using the rows R a = { a } × A R_a = \{a\} \times A R a = { a } × A and the identity ⟨ V Q ⟩ = ⟨ ( V ∖ R a ) Q ⟩ ∪ ⋃ d ∈ R a { ⟨ d Q ⟩ } \langle V_Q\rangle= \langle(V \setminus R_a)_Q\rangle\cup\bigcup_{d\in R_a}\{\langle d_Q\rangle\} ⟨ V Q ⟩ = ⟨( V ∖ R a ) Q ⟩ ∪ ⋃ d ∈ R a {⟨ d Q ⟩} , we get ∣ E ∩ R a ∣ ≤ ℵ 0 |E \cap R_a| \le\aleph_0 ∣ E ∩ R a ∣ ≤ ℵ 0 for every a ∈ A a \in A a ∈ A .
Thus B B B and E E E satisfy the counting condition (3) of Lemma 4.1, while T ⊆ E 1 ∪ E 2 \mathcal{T} \subseteq E_1 \cup E_2 T ⊆ E 1 ∪ E 2 gives D = B ∪ E D = B \cup E D = B ∪ E , again a contradiction. Hence P H ( Z ) ∉ W w P_H(Z) \notin\mathsf{W}_w P H ( Z ) ∈ / W w . □ \square □
The failure is not an artefact of pathological spaces: it persists over locally compact Hausdorff spaces.
Proposition 4.15. There is a locally compact Hausdorff space L L L such that P H ( L ) ∉ W i P_{\mathrm{H}}(L) \notin W_i P H ( L ) ∈ / W i and P S ( L ) ∉ W i P_{\mathrm{S}}(L) \notin W_i P S ( L ) ∈ / W i . Hence neither P H P_{\mathrm{H}} P H nor P S P_{\mathrm{S}} P S preserves W i W_i W i , even over locally compact Hausdorff spaces.
Proof. Keep I = O ( M ) I = \mathcal{O}(M) I = O ( M ) , S = S I S = S^I S = S I and the basic open sets [ F ] [F] [ F ] from Proposition 4.4 , so that S ∉ W i S \notin W_i S ∈ / W i . Let D ∗ = I ⊔ { ∗ } D^* = I \sqcup\{\ast\} D ∗ = I ⊔ { ∗ } be discrete and let I + = I ⊔ { ∞ } I^+ = I \sqcup\{\infty\} I + = I ⊔ { ∞ } be the one-point compactification of the discrete space I I I ; both spaces are locally compact Hausdorff, and both belong to W i n W_{\mathrm{in}} W in because all their compact subspaces are finite, hence Hausdorff, so that Proposition 3.8 (ii) applies to D ∗ D^* D ∗ and to I + I^+ I + .
Define h : S → P H ( D ∗ ) h:S \to P_{\mathrm{H}}(D^*) h : S → P H ( D ∗ ) by h ( T ) = T ∪ { ∗ } h(T)=T\cup\{\ast\} h ( T ) = T ∪ { ∗ } and s : S → P S ( I + ) s:S \to P_{\mathrm{S}}(I^+) s : S → P S ( I + ) by s ( T ) = { ∞ } ∪ ( I ∖ T ) s(T)=\{\infty\}\cup(I\setminus T) s ( T ) = { ∞ } ∪ ( I ∖ T ) . The set s ( T ) s(T) s ( T ) is closed in the compact Hausdorff space I + I^+ I + , hence compact and saturated, and both maps are injective. For open O O O in the respective underlying space,
h − 1 ( ⟨ O ⟩ ) = { S , ∗ ∈ O , ⋃ i ∈ O [ { i } ] , ∗ ∉ O , h^{-1}(\langle O\rangle)=
\begin{cases}
S, & \ast\in O,\\
\bigcup_{i\in O}[\{i\}], & \ast\notin O,
\end{cases} h − 1 (⟨ O ⟩) = { S , ⋃ i ∈ O [{ i }] , ∗ ∈ O , ∗ ∈ / O ,
s − 1 ( □ O ) = { [ I ∖ O ] , ∞ ∈ O , ∅ , ∞ ∉ O . s^{-1}(\Box O)=
\begin{cases}
[I\setminus O], & \infty\in O,\\
\varnothing, & \infty\notin O.
\end{cases} s − 1 ( □ O ) = { [ I ∖ O ] , ∅ , ∞ ∈ O , ∞ ∈ / O .
These sets are open in S S S : in the first case because S S S is open, in the second because a union of basic open sets is open, in the third because [ I ∖ O ] [I\setminus O] [ I ∖ O ] is basic, and in the fourth trivially. The choices O = { i } O=\{i\} O = { i } and O = I + ∖ F O=I^+\setminus F O = I + ∖ F with F F F finite yield all coordinate tests, whence both maps are embeddings. Since S ∉ W i S\notin W_i S ∈ / W i , the subspaces h ( S ) ≅ S h(S)\cong S h ( S ) ≅ S and s ( S ) ≅ S s(S)\cong S s ( S ) ≅ S do not belong to W i W_i W i ; by Proposition 3.4 , neither P H ( D ∗ ) P_{\mathrm{H}}(D^*) P H ( D ∗ ) nor P S ( I + ) P_{\mathrm{S}}(I^+) P S ( I + ) belongs to W i W_i W i . □
The last result of this section concerns a countable metrizable space whose lower Vietoris space is well behaved but not co-sober. We write Σ O ( N ) \Sigma\mathcal{O}(N) Σ O ( N ) for the lattice O ( N ) \mathcal{O}(N) O ( N ) ordered by inclusion and endowed with the Scott topology.
Proposition 4.16. There is a countable metrizable space N N N such that P H ( N ) P_{\mathrm{H}}(N) P H ( N ) is second countable, coherent and inner weak Wilker, but neither co-sober nor Wilker; moreover the union operation Σ O ( N ) × Σ O ( N ) → Σ O ( N ) \Sigma\mathcal{O}(N)\times\Sigma\mathcal{O}(N)\to\Sigma\mathcal{O}(N) Σ O ( N ) × Σ O ( N ) → Σ O ( N ) is discontinuous for the product topology.
Proof. By [15 ] there is a countable metrizable space N N N for which P H ( N ) P_{\mathrm{H}}(N) P H ( N ) is not co-sober; fix such an N N N and put T = P H ( N ) T=P_{\mathrm{H}}(N) T = P H ( N ) .
T T T is second countable. Let B \mathcal{B} B be a countable base of N N N . For a family V ⊆ O ( N ) \mathcal{V}\subseteq\mathcal{O}(N) V ⊆ O ( N ) we have ⟨ ⋃ V ⟩ = ⋃ V ∈ V ⟨ V ⟩ \langle\bigcup\mathcal{V}\rangle=\bigcup_{V\in\mathcal{V}}\langle V\rangle ⟨ ⋃ V ⟩ = ⋃ V ∈ V ⟨ V ⟩ , so { ⟨ B ⟩ : B ∈ B } \{\langle B\rangle:B\in\mathcal{B}\} {⟨ B ⟩ : B ∈ B } is a countable base of T T T . Hence every subspace of T T T is second countable and therefore Lindelöf, and for K ⊆ T K\subseteq T K ⊆ T compact and K ⊆ U 1 ∪ U 2 K\subseteq U_1\cup U_2 K ⊆ U 1 ∪ U 2 the sets L j = K ∩ U j L_j=K\cap U_j L j = K ∩ U j are Lindelöf and satisfy L 1 ∪ L 2 = K L_1\cup L_2=K L 1 ∪ L 2 = K ; thus T ∈ W i T\in W_i T ∈ W i .
T T T is coherent. Let u : T × T → T u:T\times T\to T u : T × T → T , u ( F , G ) = F ∪ G u(F,G)=F\cup G u ( F , G ) = F ∪ G . The specialization order of T T T is inclusion: if F ⊈ G F\not\subseteq G F ⊆ G , then ⟨ N ∖ G ⟩ \langle N\setminus G\rangle ⟨ N ∖ G ⟩ contains F F F but not G G G . Hence u − 1 ( ⟨ V ⟩ ) = ( ⟨ V ⟩ × T ) ∪ ( T × ⟨ V ⟩ ) u^{-1}(\langle V\rangle)=(\langle V\rangle\times T)\cup(T\times\langle V\rangle) u − 1 (⟨ V ⟩) = (⟨ V ⟩ × T ) ∪ ( T × ⟨ V ⟩) for open V V V , and u ( A × B ) = A ∩ B u(A\times B)=A\cap B u ( A × B ) = A ∩ B for saturated families A , B ⊆ T A,B\subseteq T A , B ⊆ T , because both are upward closed and u ( F , F ) = F u(F,F)=F u ( F , F ) = F . If A A A and B B B are compact, then by continuity of u u u their intersection A ∩ B A\cap B A ∩ B is compact [16 ] ; this is coherence of T T T .
T ∉ W T\notin W T ∈ / W . By [4 ] , Proposition 4.8 and Theorem 7.16 a coherent Wilker space is co-sober, and the union operation on Σ O ( N ) \Sigma\mathcal{O}(N) Σ O ( N ) is continuous exactly when T ∈ W T\in W T ∈ W . Since T T T is not co-sober, T ∉ W T\notin W T ∈ / W , and the union operation on Σ O ( N ) \Sigma\mathcal{O}(N) Σ O ( N ) is discontinuous. □
Conclusions and Open Problems
The four Wilker-type properties form a hierarchy
W i n ⊆ W ∩ W i ⊆ W w , W_{\mathrm{in}}\subseteq W\cap W_i\subseteq W_w, W in ⊆ W ∩ W i ⊆ W w ,
in which both inclusions displayed on the left can be strict, and in which W W W and W i W_i W i are incomparable. The properties are exactly the continuity conditions for the union operation on the lattices O L ( X ) \mathcal{O}_L(X) O L ( X ) and O k ( X ) \mathcal{O}_k(X) O k ( X ) of open sets, and it is this characterization that accounts for their relevance to power spaces and to function spaces.
As far as permanence is concerned, the situation is now essentially complete. The inner properties W i W_i W i and W i n W_{\mathrm{in}} W in are hereditary for arbitrary subspaces, while W W W and W w W_w W w are hereditary for closed subspaces only, the counterexample being a locally supercompact space in W ∖ W i W\setminus W_i W ∖ W i . All four properties are preserved by continuous retracts, and W w W_w W w , W i W_i W i , W i n W_{\mathrm{in}} W in are finitely additive over topological sums. Over KC-spaces the weak and the inner weak Wilker properties coincide, and a space whose compact subspaces are Hausdorff always belongs to W i n W_{\mathrm{in}} W in ; these two criteria reduce most verifications to compact Hausdorff pieces. Core compactness implies W w W_{\mathrm{w}} W w , and the product of a W w W_{\mathrm{w}} W w -space with a space whose compact subspaces are Hausdorff is again in W w W_{\mathrm{w}} W w , and even in W i W_{\mathrm{i}} W i when the first factor is KC.
The negative results show that the properties are not stable under the constructions most relevant to power spaces. The cofinite modification of the grid space of cardinality ℵ 2 \aleph_{2} ℵ 2 produces compact T 1 T_{1} T 1 spaces P P P , Q Q Q , all of whose subspaces lie in W i n W_{\mathrm{in}} W in , with P × Q ∉ W w P \times Q \notin W_{\mathrm{w}} P × Q ∈ / W w , hence the subcategories determined by W w W_{\mathrm{w}} W w and by W i W_{\mathrm{i}} W i are not reflective. The same grid, combined with a hereditarily compact T 1 T_{1} T 1 topology and with a one-point compactification, gives a hereditarily compact T 1 T_{1} T 1 space E E E and a compact Hausdorff space Y Y Y with E × Y ∈ W w ∖ W i E \times Y \in W_{\mathrm{w}} \setminus W_{\mathrm{i}} E × Y ∈ W w ∖ W i , so that W i W_{\mathrm{i}} W i is not productive even for factors with strong separation properties. The Vietoris constructions exhibit analogous failures: for a compact T 1 T_{1} T 1 space Z ∈ W i n Z \in W_{\mathrm{in}} Z ∈ W in both P S ( Z ) \mathcal{P}_{S}(Z) P S ( Z ) and P H ( Z ) \mathcal{P}_{H}(Z) P H ( Z ) may fail W w W_{\mathrm{w}} W w , and this failure already occurs over locally compact Hausdorff spaces. Finally, for Kovár’s space M 0 M_{0} M 0 the homeomorphic spaces O k ( M 0 ) ≅ C k ( M 0 , S ) \mathcal{O}_{k}(M_{0}) \cong C_{k}(M_{0},S) O k ( M 0 ) ≅ C k ( M 0 , S ) are not sober, and C k ( M 0 , S ) C_{k}(M_{0},S) C k ( M 0 , S ) fails property S S S ; the irreducible family of co-singleton open sets has no generic point, which is precisely the obstruction.
Several questions remain. The first two concern the finer structure of the hierarchy, the next two the behaviour under the hyperspace constructions, and the last two the relation with sobriety.
Question 5.1. Is W i n = W i W_{\mathrm{in}} = W_{\mathrm{i}} W in = W i over K C KC K C -spaces? By Proposition 3.8 the weak and the inner weak Wilker properties coincide there, and W i n ⊆ W i W_{\mathrm{in}} \subseteq W_{\mathrm{i}} W in ⊆ W i always holds; it is not known whether the inclusion can be strict when all compact subsets are closed. Equivalently, does every K C KC K C -space in W i W_{\mathrm{i}} W i have the property that closed subsets of its compact subspaces can be separated by compact pieces?
Question 5.2. Is there a compact Hausdorff space in W i ∖ W W_{\mathrm{i}} \setminus W W i ∖ W ? The two examples of § 3 , namely M 0 M_{0} M 0 and Q + Q^{+} Q + , are compact, countable and K C KC K C , but neither is Hausdorff; countability supplies the Lindelöf witnesses, and it remains open whether an analogous conclusion can be obtained under a separation axiom. A positive answer would show that the Wilker property is not a consequence of compactness together with separation.
Question 5.3. Is W i W_{\mathrm{i}} W i preserved by finite products of W i W_{\mathrm{i}} W i -spaces? Theorem 3.14 requires one factor to be in W w W_{\mathrm{w}} W w and the compact subspaces of the other to be Hausdorff, and Theorem 4.5 shows that W w W_{\mathrm{w}} W w alone does not suffice for the inner property. The question is whether two factors possessing the inner weak Wilker property admit a product theorem stronger than that available for a W w W_{\mathrm{w}} W w -factor and a compact Hausdorff factor.
Question 5.4. Does P S ( X ) ∈ W i \mathcal{P}_{S}(X) \in W_{\mathrm{i}} P S ( X ) ∈ W i imply X ∈ W i X \in W_{\mathrm{i}} X ∈ W i , and likewise for P H \mathcal{P}_{H} P H ? Proposition 4.10 gives the converse implication, from the hyperspace to the space. The corresponding statement for W W W is also open: Corollary 4.13 exhibits a countable compact T 1 T_{1} T 1 space T ∈ W T \in W T ∈ W with P S ( T ) ∉ W \mathcal{P}_{S}(T) \notin W P S ( T ) ∈ / W , but no space X X X with P S ( X ) ∈ W \mathcal{P}_{S}(X) \in W P S ( X ) ∈ W and X ∉ W X \notin W X ∈ / W is known.
Question 5.5. Can the order-theoretic hypothesis in Proposition 4.7 be weakened? Example 4.8 shows that sobriety of Y Y Y cannot be dropped, and the equivalence itself uses the pointwise description of the specialization order in an essential way. It remains open whether property S S S , together with sobriety of Y Y Y , already forces the function space to be sober for topologies strictly finer than the compact-open one.
Question 5.6. Is the countability assumption in Corollary 4.13 necessary? The construction rests on the existence of countable compact T 1 T_{1} T 1 spaces with non-Wilker product, and it is not known whether an uncountable example with P S ( T ) ∉ W \mathcal{P}_{S}(T) \notin W P S ( T ) ∈ / W and T ∈ W T \in W T ∈ W can be obtained without it.
Declaration of generative AI and AI-assisted technologies in the manuscript preparation process
During the revision of this manuscript, ChatGPT (OpenAI) was used to assist with checking mathematical arguments, clarifying definitions and proofs, organizing the exposition, and language editing. The authors take responsibility for the content of the manuscript.