Introduction
Domain theory constitutes a framework within which order and topology admit a profound integration. The Scott topology is a canonical topology induced by the order structure of a poset and plays a fundamental role in domain theory. Sobriety and well-filteredness are two kinds of important topological properties. The study of sobriety and well-filteredness for the Scott spaces clarifies two complementary aspects of the interaction between order and topology. The Scott topology on a continuous dcpo possesses favourable properties. If we extend these properties to general dcpos, it requires some topological conditions that remain useful without Hausdorff separation axiom. Well-filteredness is a desirable candidate. In this paper, we focus on well-filtered spaces.
The distinction between well-filteredness and sobriety emerged from a series of counterexamples and representation results. Johnstone constructed a dcpo whose Scott topology is not sober and not well-filtered([10]). As we know, every sober space is well-filtered([2, 3]). Kou showed that there is a dcpo which is well-filtered but non sober for the Scott topology and thus answered a question posed by Heckmann ([12, 4]). Xi and Zhao later study the models for well-filtered T 1 T_1 T 1 spaces and produce new non sober examples ([21]). Xi and Lawson proved that every Lawson-compact dcpo and hence every complete lattice is Scott well-filtered([20]). In addition, Lawson, Wu and Xi proved that every core-compact well-filtered T 0 T_0 T 0 space is sober ([13]). Topological Rudin’s lemma extended by Heckmann and Keimel provides a useful tool for constructing an irreducible closed set([5]). Shen, Xi, Xu and Zhao constructed well-filtered reflections and proved that arbitrary ordinary topological products of well-filtered spaces are well-filtered([18]). Furthermore, Xu, Shen, Xi and Zhao developed Rudin sets and well-filtered determined sets (or WD sets) to characterize well-filteredness and describe its reflections([22]). Thus the Rudin sets and WD sets are useful tools for studying well-filtered spaces.
The product preservation of well-filteredness and sobriety is already nontrivial for the Scott topology. For two posets P P P and Q Q Q , the Scott topology Σ ( P × Q ) \Sigma(P \times Q) Σ ( P × Q ) can be strictly finer than the product topology Σ P × Σ Q \Sigma P \times\Sigma Q Σ P × Σ Q . Consequently, the ordinary product theorems with respect to sobriety and well-filteredness do not apply directly to this Scott product problem. So Xu and Zhao propose four questions for well-filtered spaces and sober spaces (Questions 4.6, 4.7, 4.8 and 4.9 in[24 ] ). Recently, Miao, Xi, Jia, Li and Zhao construct two Scott sober dcpos whose order product is not Scott sober([14]). Later, He obtained two Scott sober complete lattices even two co-frames whose order product is not sober for the Scott topology([8]). In this paper, we solve the question 4.7 and question 4.9. Specifically, the first result is that for every family of Scott well-filtered dcpos { P i ∣ i ∈ I } \{P_i \mid i \in I\} { P i ∣ i ∈ I } , Σ ∏ i ∈ I P i \Sigma\prod_{i\in I}P_i Σ ∏ i ∈ I P i is well-filtered. The proof for the above result analyzes projections of Rudin sets and finite coordinate replacements.
The collection K ( X ) K(X) K ( X ) consists of all the nonempty compact saturated subsets of X X X ordered by the reverse inclusion. The upper Vietoris topology of the Smyth power space and the Scott topology of K ( X ) K(X) K ( X ) need not coincide. Xu, Wen and Xi proved that well-filteredness of X X X implies that Σ K ( X ) \Sigma K(X) Σ K ( X ) is well-filtered([23]). However, for two dcpos P P P and Q Q Q , K ( P × Q ) K(P \times Q) K ( P × Q ) is not equal to K ( P ) × K ( Q ) K(P) \times K(Q) K ( P ) × K ( Q ) . So the well-filteredness of Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) is unknown under the condition that Σ K ( P ) \Sigma K(P) Σ K ( P ) and Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are well-filtered. We first prove that for dcpos P P P and Q Q Q , Σ K ( P ) \Sigma K(P) Σ K ( P ) and Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are well-filtered if Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) is well-filtered. Furthermore, we also show that if Σ K ( P ) \Sigma K(P) Σ K ( P ) is well-filtered and Q Q Q a R B RB R B -domain such that K ( P × Q ) K(P \times Q) K ( P × Q ) is a dcpo, then Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) is well-filtered.
The WD sets as a useful tool for studying well-filtered spaces are defined by using all well-filtered spaces and continuous mappings. Thus it is difficult to check whether a closed subset of a topological space is a WD set. It is essential to give a characterization for WD sets of a topological space X X X by just using the information on X X X . Motivated by the open filters viewpoint of the Hofmann–Mislove theorem, we prove that, for every closed A ⊆ X A \subseteq X A ⊆ X , A ∈ W D ( X ) A \in\mathrm{WD}(X) A ∈ WD ( X ) if and only if p A ∈ H X p_A \in\mathcal{H}_X p A ∈ H X , where H X \mathcal{H}_X H X is a subfamily of all proper Scott open filters on O ( X ) \mathcal{O}(X) O ( X ) with some rules.
Preliminaries
Let P P P be a poset. For each A ⊆ P A \subseteq P A ⊆ P , set ↑ A = { x ∈ P ∣ ∃ a ∈ A , a ≤ x } \mathord{\uparrow}A=\{x\in P\mid\exists a\in A,a\leq x\} ↑ A = { x ∈ P ∣ ∃ a ∈ A , a ≤ x } and ↓ A = { x ∈ P ∣ ∃ a ∈ A , x ≤ a } \mathord{\downarrow}A=\{x\in P\mid\exists a\in A,x\leq a\} ↓ A = { x ∈ P ∣ ∃ a ∈ A , x ≤ a } . We write ↓ { x } \mathord{\downarrow}\{x\} ↓ { x } as ↓ x \mathord{\downarrow}x ↓ x and ↑ { x } \mathord{\uparrow}\{x\} ↑ { x } as ↑ x \mathord{\uparrow}x ↑ x for each x ∈ P x\in P x ∈ P . A subset D D D of P P P is directed provided that D D D is nonempty and ∀ a , b ∈ D \forall a,b\in D ∀ a , b ∈ D , ↑ a ∩ ↑ b ∩ D ≠ ∅ \mathord{\uparrow}a\cap\mathord{\uparrow}b\cap D\ne\varnothing ↑ a ∩ ↑ b ∩ D = ∅ . Dually, a nonempty subset D ⊆ P D\subseteq P D ⊆ P is filtered if ∀ a , b ∈ D \forall a,b\in D ∀ a , b ∈ D , ↓ a ∩ ↓ b ∩ D ≠ ∅ \mathord{\downarrow}a\cap\mathord{\downarrow}b\cap D\ne\varnothing ↓ a ∩ ↓ b ∩ D = ∅ . An element x x x is an upper bound for a subset A A A if a ≤ x a\leq x a ≤ x for each a ∈ A a\in A a ∈ A . A poset P P P is called a directed complete poset (dcpo, for short) if every directed subset of P P P has a supremum. For each pair of x , y ∈ P x,y\in P x , y ∈ P , we say that x x x is way below y y y , written x ≪ y x\ll y x ≪ y , if for each directed set D ⊆ P D\subseteq P D ⊆ P whose supremum exists, y ≤ ⋁ D y\leq\bigvee D y ≤ ⋁ D implies ↑ x ∩ D ≠ ∅ \mathord{\uparrow}x\cap D\ne\varnothing ↑ x ∩ D = ∅ . An element p ∈ P p\in P p ∈ P is compact if p ≪ p p\ll p p ≪ p . A subset U ⊆ P U\subseteq P U ⊆ P is Scott open if U = ↑ U U=\mathord{\uparrow}U U = ↑ U and for each directed subset D D D with ⋁ D \bigvee D ⋁ D existing, ⋁ D ∈ U \bigvee D\in U ⋁ D ∈ U implies D ∩ U ≠ ∅ D\cap U\ne\varnothing D ∩ U = ∅ . The set of Scott open sets of P P P forms the Scott topology σ ( P ) \sigma(P) σ ( P ) . We write Σ P \Sigma P Σ P for ( P , σ ( P ) ) (P,\sigma(P)) ( P , σ ( P )) . Let P , Q P,Q P , Q be dcpos. Then the mapping f : Σ P ⟶ Σ Q f:\Sigma P\longrightarrow\Sigma Q f : Σ P ⟶ Σ Q is continuous (or called Scott continuous) if and only if f f f is monotone and for each directed set D ⊆ P D\subseteq P D ⊆ P , f ( ⋁ D ) = ⋁ d ∈ D f ( d ) f(\bigvee D)=\bigvee_{d\in D}f(d) f ( ⋁ D ) = ⋁ d ∈ D f ( d ) .
Let X X X be a topological space. We write cl X ( A ) \operatorname{cl}_{X}(A) cl X ( A ) for the closure of A A A . A subset A A A of X X X is called a saturated set if A A A is just the intersection of all open sets containing A A A . A subset F F F of X X X is irreducible if F F F is nonempty and for every pair of closed sets A , B ⊆ X A,B\subseteq X A , B ⊆ X , F ⊆ A ∪ B F\subseteq A\cup B F ⊆ A ∪ B implies F ⊆ A F\subseteq A F ⊆ A or F ⊆ B F\subseteq B F ⊆ B . A T 0 T_{0} T 0 space X X X is sober if every irreducible closed set is just the closure of a singleton. For a T 0 T_{0} T 0 space X X X , the specialization order ≤ \leq ≤ on X X X is defined by x ≤ y x\leq y x ≤ y if and only if x ∈ cl X ( { y } ) x\in\operatorname{cl}_{X}(\{y\}) x ∈ cl X ({ y }) . Unless otherwise stated, throughout the paper, whenever an order-theoretic concept is mentioned in the context of a T 0 T_{0} T 0 space X X X , it is to be interpreted with respect to the specialization order on X X X . A subset A A A of a topological space X X X is saturated if A = ↑ A A=\mathord{\uparrow}A A = ↑ A under the specialization order. We use K ( X ) K(X) K ( X ) to denote the set of all nonempty compact saturated sets of X X X with the reverse inclusion order and O ( X ) \mathcal{O}(X) O ( X ) (Γ ( X ) \Gamma(X) Γ ( X ) ) to denote the lattice of all open subsets (closed subsets) of $X ordered by inclusion. A topological space X X X is well-filtered if for each filtered family F \mathcal{F} F of compact saturated subsets of X X X and each open set U U U of X X X , ⋂ F ⊆ U \bigcap\mathcal{F}\subseteq U ⋂ F ⊆ U implies F ⊆ U F\subseteq U F ⊆ U for some F ∈ F F\in\mathcal{F} F ∈ F . A dcpo P P P is called Scott well-filtered if the dcpo P P P endowed with the Scott topology is well-filtered. A topological space X X X is a retract of a topological space Y Y Y if there are two continuous maps f : X ⟶ Y f:X\longrightarrow Y f : X ⟶ Y and g : Y ⟶ X g:Y\longrightarrow X g : Y ⟶ X such that g ∘ f = id X g\circ f=\operatorname{id}_{X} g ∘ f = id X .
The Scott well-filteredness of the arbitrary product
In this section, we show that the product of Scott well-filtered dcpos is still Scott well-filtered. Thus this gives positive answers to Questions 4.7 and 4.9 posed by Xu and Zhao([24]).
Definition 3.1. ([22 ] ) Let X X X be a T 0 T_{0} T 0 space. A nonempty closed subset A A A of X X X is called a Rudin set if there exists a filtered family K ⊆ K ( X ) \mathcal{K}\subseteq K(X) K ⊆ K ( X ) such that A A A is minimal in
M ( K ) = { B ∈ Γ ( X ) ∣ ∀ K ∈ K , B ∩ K ≠ ∅ } . M(\mathcal{K})=\{B\in\Gamma(X)\mid\forall K\in\mathcal{K},\ B\cap K\neq\varnothing\}. M ( K ) = { B ∈ Γ ( X ) ∣ ∀ K ∈ K , B ∩ K = ∅ } .
where the order on M ( K ) M(\mathcal{K}) M ( K ) is the inclusion order. The symbol RD ( X ) \operatorname{RD}(X) RD ( X ) denotes the collection of all closed Rudin sets of $X.
In the following, we call the filtered family K ⊆ K ( X ) \mathcal{K}\subseteq K(X) K ⊆ K ( X ) a witness of the set A ∈ RD ( X ) A\in\operatorname{RD}(X) A ∈ RD ( X ) .
Lemma 3.2. Let K \mathcal{K} K be a witnesses of a closed Rudin set A A A . Then for every K 0 ∈ K K_{0}\in\mathcal{K} K 0 ∈ K ,
cl X ( A ∩ K 0 ) = A . \operatorname{cl}_{X}(A\cap K_{0})=A. cl X ( A ∩ K 0 ) = A . Proof. Choose any K ∈ K K\in\mathcal{K} K ∈ K . By the filteredness of K \mathcal{K} K , there exists K 1 ∈ K K_{1}\in\mathcal{K} K 1 ∈ K with K 1 ⊆ K 0 ∩ K K_{1}\subseteq K_{0}\cap K K 1 ⊆ K 0 ∩ K . It follows from A ∈ M ( K ) A\in M(\mathcal{K}) A ∈ M ( K ) that we can select a ∈ A ∩ K 1 a\in A\cap K_{1} a ∈ A ∩ K 1 . This implies a ∈ A ∩ K 0 ∩ K a\in A\cap K_{0}\cap K a ∈ A ∩ K 0 ∩ K and thus
K ∩ cl X ( A ∩ K 0 ) ≠ ∅ . K\cap\operatorname{cl}_{X}(A\cap K_{0})\neq\varnothing. K ∩ cl X ( A ∩ K 0 ) = ∅ . So cl X ( A ∩ K 0 ) ∈ M ( K ) \operatorname{cl}_{X}(A\cap K_{0})\in M(\mathcal{K}) cl X ( A ∩ K 0 ) ∈ M ( K ) and clearly cl X ( A ∩ K 0 ) ⊆ A \operatorname{cl}_{X}(A\cap K_{0})\subseteq A cl X ( A ∩ K 0 ) ⊆ A . The minimality of A A A gives cl X ( A ∩ K 0 ) = A \operatorname{cl}_{X}(A\cap K_{0})=A cl X ( A ∩ K 0 ) = A .
Theorem 3.3. ([22 ] ) Let X X X be a T 0 T_{0} T 0 space. Then the following are equivalent:
X X X is well-filtered;
For every closed Rudin set A A A , A = cl X ( { a } ) A=\operatorname{cl}_{X}(\{a\}) A = cl X ({ a }) for some a ∈ X a\in X a ∈ X .
Lemma 3.4. Let R = ∏ i ∈ I P i R=\prod_{i\in I}P_{i} R = ∏ i ∈ I P i be the product of a family consisting of dcpos and A ∈ RD ( ∑ R ) A\in\operatorname{RD}(\sum R) A ∈ RD ( ∑ R ) .
For each i ∈ I i\in I i ∈ I , set A i = π i ( A ) A_{i}=\pi_{i}(A) A i = π i ( A ) . Then the following statements hold.
A i ∈ RD ( ∑ P i ) A_{i}\in\operatorname{RD}(\sum P_{i}) A i ∈ RD ( ∑ P i ) ;
If a = ( a i ) i ∈ I ∈ A a=(a_{i})_{i\in I}\in A a = ( a i ) i ∈ I ∈ A and b j ∈ A j b_{j}\in A_{j} b j ∈ A j , then x = [ a , b j ] ∈ A x=[a,b_{j}]\in A x = [ a , b j ] ∈ A , where x = ( x i ) i ∈ I x=(x_{i})_{i\in I} x = ( x i ) i ∈ I with x i = a i x_{i}=a_{i} x i = a i (i ∈ I ∖ { j } i\in I\setminus\{j\} i ∈ I ∖ { j } ) and x j = b j x_{j}=b_{j} x j = b j ;
If a ∈ A a\in A a ∈ A and for any finite F ⊆ I F\subseteq I F ⊆ I , b j ∈ A j b_{j}\in A_{j} b j ∈ A j for j ∈ F j\in F j ∈ F , then y = [ a , F ] ∈ A y=[a,F]\in A y = [ a , F ] ∈ A , where y = ( y i ) i ∈ I y=(y_{i})_{i\in I} y = ( y i ) i ∈ I with y i = a i y_{i}=a_{i} y i = a i (i ∈ I ∖ F i\in I\setminus F i ∈ I ∖ F ) and y j = b j y_{j}=b_{j} y j = b j (j ∈ F j\in F j ∈ F ).
Proof. (1) As A ∈ RD ( Σ P ) A \in\operatorname{RD}(\Sigma P) A ∈ RD ( Σ P ) , there exists a filtered family K = { K λ ∣ λ ∈ Λ } \mathcal{K} = \{K_\lambda\mid\lambda\in\Lambda\} K = { K λ ∣ λ ∈ Λ } consisting of nonempty compact saturated subsets of Σ R \Sigma R Σ R such that A A A is a minimal element in M ( K ) M(\mathcal{K}) M ( K ) .
Fix i ∈ I i \in I i ∈ I and set
Q = ∏ j ∈ I ∖ { i } P j , T = P i . Q = \prod_{j \in I \setminus\{i\}} P_j,\qquad T = P_i. Q = j ∈ I ∖ { i } ∏ P j , T = P i .
Then Q Q Q and T T T are dcpos. Define a mapping φ i : R ⟶ Q × T \varphi_i : R \longrightarrow Q \times T φ i : R ⟶ Q × T by
φ i ( x ) = ( x ∣ I ∖ { i } , x i ) , \varphi_i(x) = (x|_{I \setminus\{i\}}, x_i), φ i ( x ) = ( x ∣ I ∖ { i } , x i ) ,
where x ∣ I ∖ { i } x|_{I \setminus\{i\}} x ∣ I ∖ { i } denotes the element obtained from x x x by deleting its i i i -th coordinate. One can check that φ i \varphi_i φ i is an order isomorphism. Moreover, φ i \varphi_i φ i preserves the directed suprema coordinatewise. Consequently, φ i \varphi_i φ i and its converse mapping are Scott continuous. Set B = φ i ( A ) B = \varphi_i(A) B = φ i ( A ) and let
π Q : Q × T ⟶ Q , π T : Q × T ⟶ T \pi_Q : Q \times T \longrightarrow Q,\qquad\pi_T : Q \times T \longrightarrow T π Q : Q × T ⟶ Q , π T : Q × T ⟶ T
be two projections. Thus π T ( B ) = A i \pi_T(B) = A_i π T ( B ) = A i . Furthermore, π Q \pi_Q π Q and π T \pi_T π T are Scott continuous. Take
L λ = φ i ( K λ ) . L_\lambda= \varphi_i(K_\lambda). L λ = φ i ( K λ ) .
Each L λ L_\lambda L λ is a nonempty saturated compact set in Σ ( Q × T ) \Sigma(Q \times T) Σ ( Q × T ) because φ i \varphi_i φ i is an order isomorphism. The collection { L λ ∣ λ ∈ Λ } \{L_\lambda\mid\lambda\in\Lambda\} { L λ ∣ λ ∈ Λ } is a filtered family as { λ ∣ λ ∈ Λ } \{\lambda\mid\lambda\in\Lambda\} { λ ∣ λ ∈ Λ } is filtered.
Claim 1. { L λ ∣ λ ∈ Λ } \{L_\lambda\mid\lambda\in\Lambda\} { L λ ∣ λ ∈ Λ } is a witness family for the closed set B B B .
Since φ i \varphi_i φ i is an order isomorphism and A A A is Scott closed in R R R , B B B is closed in Σ ( Q × T ) \Sigma(Q \times T) Σ ( Q × T ) . Suppose C ⊆ B C \subseteq B C ⊆ B is a Scott closed subset of Q × T Q \times T Q × T which meets each L λ L_\lambda L λ . Then φ i − 1 ( C ) ⊆ A \varphi_i^{-1}(C) \subseteq A φ i − 1 ( C ) ⊆ A is a closed subset of R R R and meets each K λ K_\lambda K λ . The minimality of A A A implies that φ i − 1 ( C ) = A \varphi_i^{-1}(C) = A φ i − 1 ( C ) = A . As a consequence, C = B C = B C = B and hence B ∈ RD ( Σ ( Q × T ) ) B \in\operatorname{RD}(\Sigma(Q \times T)) B ∈ RD ( Σ ( Q × T )) . This proves Claim 1.
For each λ \lambda λ , set
S λ = B ∩ L λ , H λ = π Q ( S λ ) and J λ = π T ( S λ ) . S_\lambda= B \cap L_\lambda,\qquad H_\lambda= \pi_Q(S_\lambda)\ \text{and}\ J_\lambda= \pi_T(S_\lambda). S λ = B ∩ L λ , H λ = π Q ( S λ ) and J λ = π T ( S λ ) .
As B B B is closed in Σ ( Q × T ) \Sigma(Q \times T) Σ ( Q × T ) and L λ L_\lambda L λ is compact, S λ S_\lambda S λ is also compact. By the Scott continuity of π Q \pi_Q π Q and π T \pi_T π T , H λ H_\lambda H λ and J λ J_\lambda J λ are nonempty compact subset of Σ Q \Sigma Q Σ Q and Σ T \Sigma T Σ T , respectively. By Lemma 3.2, we have
cl Σ ( Q × T ) ( S λ ) = B . \operatorname{cl}_{\Sigma(Q \times T)}(S_\lambda) = B. cl Σ ( Q × T ) ( S λ ) = B .
For each λ \lambda λ , define
E λ = { t ∈ T ∣ ∃ q ∈ H λ , ( q , t ) ∈ B } . E_\lambda= \{t \in T \mid\exists q \in H_\lambda,\ (q,t) \in B\}. E λ = { t ∈ T ∣ ∃ q ∈ H λ , ( q , t ) ∈ B } .
Claim 2. E λ E_\lambda E λ is closed in Σ T \Sigma T Σ T .
Let t ′ ≤ t ∈ E λ t' \leq t \in E_\lambda t ′ ≤ t ∈ E λ and ( q , t ) ∈ B (q,t) \in B ( q , t ) ∈ B with q ∈ H λ q \in H_\lambda q ∈ H λ . Then ( q , t ′ ) ≤ ( q , t ) (q,t') \leq(q,t) ( q , t ′ ) ≤ ( q , t ) and further ( q , t ′ ) ∈ B (q,t') \in B ( q , t ′ ) ∈ B because B B B is lower. So t ′ ∈ E λ t' \in E_\lambda t ′ ∈ E λ and therefore E λ E_\lambda E λ is also lower.
Suppose D ⊆ E λ D \subseteq E_\lambda D ⊆ E λ is directed and take t ^ = sup D \hat{t} = \sup D t ^ = sup D . For each d ∈ D d \in D d ∈ D , put
F d = { q ∈ H λ : ( q , d ) ∈ B } . F_d = \{q \in H_\lambda: (q,d) \in B\}. F d = { q ∈ H λ : ( q , d ) ∈ B } .
Then F d F_d F d is nonempty as d ∈ E λ d \in E_\lambda d ∈ E λ . It is not difficult to check that the mapping f d : Q ⟶ Q × T f_d : Q \longrightarrow Q \times T f d : Q ⟶ Q × T is Scott continuous, where f d f_d f d is defined by
∀ a ∈ Q , f d ( a ) = ( a , d ) . \forall a \in Q,\ f_d(a) = (a,d). ∀ a ∈ Q , f d ( a ) = ( a , d ) .
Consequently,
f d − 1 ( B ) = { q ∈ Q ∣ ( q , d ) ∈ B } f_d^{-1}(B) = \{q \in Q \mid(q,d) \in B\} f d − 1 ( B ) = { q ∈ Q ∣ ( q , d ) ∈ B }
is Scott closed in Q Q Q . So F d = H λ ∩ f d − 1 ( B ) F_d=H_\lambda\cap f_d^{-1}(B) F d = H λ ∩ f d − 1 ( B ) is closed in the compact subspace H λ H_\lambda H λ . Choose any d 1 , … , d n ∈ D d_1,\ldots,d_n\in D d 1 , … , d n ∈ D . There is a d 0 ∈ D d_0\in D d 0 ∈ D larger than each d k d_k d k (1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n ) as D D D is directed. Since B B B is lower, we have
∅ ≠ F d 0 ⊆ ⋂ k = 1 n F d k . \varnothing\ne F_{d_0}\subseteq\bigcap_{k=1}^{n}F_{d_k}. ∅ = F d 0 ⊆ k = 1 ⋂ n F d k .
The compactness of H λ H_\lambda H λ gives that we can select q ^ ∈ ⋂ d ∈ D F d ⊆ H λ \hat{q}\in\bigcap_{d\in D}F_d\subseteq H_\lambda q ^ ∈ ⋂ d ∈ D F d ⊆ H λ . Now, { ( q ^ , d ) ∣ d ∈ D } ⊆ B \{(\hat{q},d)\mid d\in D\}\subseteq B {( q ^ , d ) ∣ d ∈ D } ⊆ B is directed and
⋁ d ∈ D ( q ^ , d ) = ( q ^ , t ^ ) . \bigvee_{d\in D}(\hat{q},d)=(\hat{q},\hat{t}). d ∈ D ⋁ ( q ^ , d ) = ( q ^ , t ^ ) .
It follows from the Scott closedness of B B B that ( q ^ , t ^ ) ∈ B (\hat{q},\hat{t})\in B ( q ^ , t ^ ) ∈ B . Then we deduce that t ^ ∈ E λ \hat{t}\in E_\lambda t ^ ∈ E λ . Hence, E λ E_\lambda E λ is Scott closed in T T T .
Claim 3. Each A i A_i A i is closed in Σ T \Sigma T Σ T .
By the definition of these sets, we can check that
J λ ⊆ E λ ⊆ A i . J_\lambda\subseteq E_\lambda\subseteq A_i. J λ ⊆ E λ ⊆ A i . Then cl Σ T ( J λ ) ⊆ cl Σ T ( A i ) \operatorname{cl}_{\Sigma T}(J_\lambda)\subseteq\operatorname{cl}_{\Sigma T}(A_i) cl Σ T ( J λ ) ⊆ cl Σ T ( A i ) .
As
S λ ⊆ π T − 1 ( π T ( S λ ) ) ⊆ π T − 1 ( cl Σ T ( J λ ) ) , S_\lambda\subseteq\pi_T^{-1}(\pi_T(S_\lambda))\subseteq\pi_T^{-1}(\operatorname{cl}_{\Sigma T}(J_\lambda)), S λ ⊆ π T − 1 ( π T ( S λ )) ⊆ π T − 1 ( cl Σ T ( J λ )) , we know
B = cl Σ ( Q × T ) ( S λ ) ⊆ π T − 1 ( cl Σ T ( J λ ) ) . B=\operatorname{cl}_{\Sigma(Q\times T)}(S_\lambda)\subseteq\pi_T^{-1}(\operatorname{cl}_{\Sigma T}(J_\lambda)). B = cl Σ ( Q × T ) ( S λ ) ⊆ π T − 1 ( cl Σ T ( J λ )) . Hence A i = π T ( B ) ⊆ cl Σ T ( J λ ) A_i=\pi_T(B)\subseteq\operatorname{cl}_{\Sigma T}(J_\lambda) A i = π T ( B ) ⊆ cl Σ T ( J λ ) . This shows that cl Σ T ( J λ ) = cl Σ T ( A i ) \operatorname{cl}_{\Sigma T}(J_\lambda)=\operatorname{cl}_{\Sigma T}(A_i) cl Σ T ( J λ ) = cl Σ T ( A i ) .
By Claim 2, E λ E_\lambda E λ is Scott closed in T T T . Consequently, we have
cl Σ T ( A i ) = cl Σ T ( J λ ) ⊆ E λ ⊆ A i ⊆ cl Σ T ( A i ) . \operatorname{cl}_{\Sigma T}(A_i)=\operatorname{cl}_{\Sigma T}(J_\lambda)\subseteq E_\lambda\subseteq A_i\subseteq\operatorname{cl}_{\Sigma T}(A_i). cl Σ T ( A i ) = cl Σ T ( J λ ) ⊆ E λ ⊆ A i ⊆ cl Σ T ( A i ) . This yields that for every λ \lambda λ ,
E λ = A i = cl Σ T ( A i ) . E_\lambda=A_i=\operatorname{cl}_{\Sigma T}(A_i). E λ = A i = cl Σ T ( A i ) . This proves Claim 3.
Claim 4. Each A i A_i A i is a closed Rudin set in Σ T \Sigma T Σ T .
Set
M λ = ↑ T J λ = { t ∈ T ∣ ∃ s ∈ S λ , π T ( s ) ≤ t } . M_\lambda=\mathord{\uparrow_T}J_\lambda=\{t\in T\mid\exists s\in S_\lambda,\ \pi_T(s)\le t\}. M λ = ↑ T J λ = { t ∈ T ∣ ∃ s ∈ S λ , π T ( s ) ≤ t } . Then each M λ M_\lambda M λ is a nonempty compact saturated set as J λ J_\lambda J λ is compact. If L ν ⊆ L λ ∩ L μ L_\nu\subseteq L_\lambda\cap L_\mu L ν ⊆ L λ ∩ L μ , then
S ν ⊆ S λ ∩ S μ and thus M ν ⊆ M λ ∩ M μ . S_\nu\subseteq S_\lambda\cap S_\mu\text{ and thus }M_\nu\subseteq M_\lambda\cap M_\mu. S ν ⊆ S λ ∩ S μ and thus M ν ⊆ M λ ∩ M μ . So the family A = { M λ ∣ λ ∈ Λ } \mathcal{A}=\{M_\lambda\mid\lambda\in\Lambda\} A = { M λ ∣ λ ∈ Λ } is a filtered family of nonempty compact saturated set in Σ T \Sigma T Σ T . Moreover, for each λ ∈ Λ \lambda\in\Lambda λ ∈ Λ ,
A i ∩ M λ ⊇ J λ ≠ ∅ . A_i\cap M_\lambda\supseteq J_\lambda\ne\varnothing. A i ∩ M λ ⊇ J λ = ∅ . In other words, the closed set A i A_i A i belongs to M ( A ) M(\mathcal{A}) M ( A ) .
Let C ⊆ A i C\subseteq A_i C ⊆ A i be a closed subset in Σ T \Sigma T Σ T such that C C C meets each M λ M_\lambda M λ . Choose c ∈ C ∩ M λ c\in C\cap M_\lambda c ∈ C ∩ M λ . Then there is s ∈ S λ s\in S_\lambda s ∈ S λ with π T ( s ) ≤ c \pi_T(s)\le c π T ( s ) ≤ c . As C C C is lower, we obtain π T ( s ) ∈ C \pi_T(s)\in C π T ( s ) ∈ C . As a result,
s ∈ B ∩ π T − 1 ( C ) ∩ L λ . s\in B\cap\pi_T^{-1}(C)\cap L_\lambda. s ∈ B ∩ π T − 1 ( C ) ∩ L λ .
This implies that B ∩ π T − 1 ( C ) B \cap\pi_T^{-1}(C) B ∩ π T − 1 ( C ) is a closed subset of B B B that meets each L λ L_\lambda L λ . By Claim 1, the minimality of B B B gives B ∩ π T − 1 ( C ) = B B \cap\pi_T^{-1}(C) = B B ∩ π T − 1 ( C ) = B . This yields that A i = π T ( B ) ⊆ C A_i = \pi_T(B) \subseteq C A i = π T ( B ) ⊆ C and therefore C = A i C = A_i C = A i . This proves Claim 4.
(2) Define a mapping π − i : R ⟶ Q \pi_{-i}: R \longrightarrow Q π − i : R ⟶ Q by π − i ( r ) = r ∣ I ∖ { i } \pi_{-i}(r) = r|_{I\setminus\{i\}} π − i ( r ) = r ∣ I ∖ { i } .
Claim 5: φ i ( A ) = π Q ( B ) × A i = π − i ( A ) × A i \varphi_i(A) = \pi_Q(B) \times A_i = \pi_{-i}(A) \times A_i φ i ( A ) = π Q ( B ) × A i = π − i ( A ) × A i .
Fix t ∈ A i t \in A_i t ∈ A i and set
B t = { q ∈ Q ∣ ( q , t ) ∈ B } . B_t = \{q \in Q \mid(q,t) \in B\}. B t = { q ∈ Q ∣ ( q , t ) ∈ B } .
Then B t = f t − 1 ( B ) B_t = f_t^{-1}(B) B t = f t − 1 ( B ) is Scott closed in Q Q Q as f t f_t f t is Scott continuous. As for every λ \lambda λ , t ∈ A i = E λ t \in A_i = E_\lambda t ∈ A i = E λ , there exists q λ ∈ H λ q_\lambda\in H_\lambda q λ ∈ H λ satisfying ( q λ , t ) ∈ B (q_\lambda,t) \in B ( q λ , t ) ∈ B . So q λ ∈ B t q_\lambda\in B_t q λ ∈ B t and besides there is u λ ∈ T u_\lambda\in T u λ ∈ T with
( q λ , u λ ) ∈ S λ = B ∩ L λ . (q_\lambda,u_\lambda) \in S_\lambda= B \cap L_\lambda. ( q λ , u λ ) ∈ S λ = B ∩ L λ .
So we know for each λ ∈ Λ \lambda\in\Lambda λ ∈ Λ ,
( q λ , u λ ) ∈ B ∩ π Q − 1 ( B t ) ∩ L λ . (q_\lambda,u_\lambda) \in B \cap\pi_Q^{-1}(B_t) \cap L_\lambda. ( q λ , u λ ) ∈ B ∩ π Q − 1 ( B t ) ∩ L λ .
Now, the closed subset B ∩ π Q − 1 ( B t ) ⊆ B B \cap\pi_Q^{-1}(B_t) \subseteq B B ∩ π Q − 1 ( B t ) ⊆ B meets each L λ L_\lambda L λ . By Claim 1, B ∩ π Q − 1 ( B t ) = B B \cap\pi_Q^{-1}(B_t) = B B ∩ π Q − 1 ( B t ) = B and hence each q ∈ π Q ( B ) q \in\pi_Q(B) q ∈ π Q ( B ) belongs to B t B_t B t . As a consequence, ( q , t ) ∈ B (q,t) \in B ( q , t ) ∈ B for every q ∈ π Q ( B ) q \in\pi_Q(B) q ∈ π Q ( B ) . Since t ∈ A i t \in A_i t ∈ A i is arbitrary, this indicates that π Q ( B ) × A i ⊆ B \pi_Q(B) \times A_i \subseteq B π Q ( B ) × A i ⊆ B . It is not difficult to verify that B ⊆ π Q ( B ) × A i B \subseteq\pi_Q(B) \times A_i B ⊆ π Q ( B ) × A i and π Q ( B ) = π − i ( A ) \pi_Q(B) = \pi_{-i}(A) π Q ( B ) = π − i ( A ) . Therefore, we have
φ i ( A ) = π Q ( B ) × A i = π − i ( A ) × A i . \varphi_i(A) = \pi_Q(B) \times A_i = \pi_{-i}(A) \times A_i. φ i ( A ) = π Q ( B ) × A i = π − i ( A ) × A i .
This Claim is proved.
Claim 6: Let a ∈ A a \in A a ∈ A and b i ∈ A i b_i \in A_i b i ∈ A i . Then [ a , b i ] ∈ A [a,b_i] \in A [ a , b i ] ∈ A , where [ a , b i ] ∈ R [a,b_i] \in R [ a , b i ] ∈ R is defined by
( [ a , b i ] ) j = { b i , j = i , a j , j ≠ i . ([a,b_i])_j =
\begin{cases}
b_i, & j=i,\\
a_j, & j\ne i.
\end{cases} ([ a , b i ] ) j = { b i , a j , j = i , j = i .
As π − i ( a ) ∈ π − i ( A ) \pi_{-i}(a) \in\pi_{-i}(A) π − i ( a ) ∈ π − i ( A ) , by Claim 5, we get
φ i ( [ a , b i ] ) = ( π − i ( a ) , b i ) ∈ φ i ( A ) . \varphi_i([a,b_i]) = (\pi_{-i}(a),b_i) \in\varphi_i(A). φ i ([ a , b i ]) = ( π − i ( a ) , b i ) ∈ φ i ( A ) .
Then [ a , b i ] ∈ A [a,b_i] \in A [ a , b i ] ∈ A because φ i \varphi_i φ i is an order isomorphism. The assertion (2) is valid.
(3) Fix a ∈ A a \in A a ∈ A and a finite set F ⊆ I F \subseteq I F ⊆ I with b i ∈ A i b_i \in A_i b i ∈ A i for each i ∈ F i \in F i ∈ F . Define
[ a , F ] j = { b j , j ∈ F , a j , j ∉ F . [a,F]_j =
\begin{cases}
b_j, & j \in F,\\
a_j, & j \notin F.
\end{cases} [ a , F ] j = { b j , a j , j ∈ F , j ∈ / F .
F = ∅ F = \varnothing F = ∅ means [ a , F ] = a [a,F] = a [ a , F ] = a . So, we consider the situation F ≠ ∅ F \ne\varnothing F = ∅ and enumerate F F F as { i 1 , … , i n } \{i_1,\ldots,i_n\} { i 1 , … , i n } . Set
a ( 0 ) = a , a ( k ) = [ a ( k − 1 ) , b i k ] ( 1 ≤ k ≤ n ) . a^{(0)} = a,\quad a^{(k)} = [a^{(k-1)},b_{i_k}] \quad(1 \le k \le n). a ( 0 ) = a , a ( k ) = [ a ( k − 1 ) , b i k ] ( 1 ≤ k ≤ n ) .
By induction, each a ( k ) a^{(k)} a ( k ) ( 0 ≤ k ≤ n ) (0 \le k \le n) ( 0 ≤ k ≤ n ) belongs to A A A and hence [ a , F ] = a ( n ) ∈ A [a,F] = a^{(n)} \in A [ a , F ] = a ( n ) ∈ A . The assertion (3) holds.
Theorem 3.5. Let { P i ∣ i ∈ I } \{P_i \mid i \in I\} { P i ∣ i ∈ I } be a family of dcpos such that each Σ P i \Sigma P_i Σ P i is well-filtered. Then Σ ∏ i ∈ I P i \Sigma\prod_{i\in I} P_i Σ ∏ i ∈ I P i is well-filtered.
Proof. By Theorem 3.3, it suffices to prove that each A ∈ RD ( Σ R ) A \in\operatorname{RD}(\Sigma R) A ∈ RD ( Σ R ) is a principal ideal. By Lemma 3.4, A i = π i ( A ) A_i = \pi_i(A) A i = π i ( A ) is a closed Rudin set in Σ P i \Sigma P_i Σ P i . By Theorem 3.3, there is a unique p i ∈ P i p_i \in P_i p i ∈ P i such that A i = ↓ p i A_i = \mathord{\downarrow}p_i A i = ↓ p i . Set p = ( p i ) i ∈ I ∈ R p = (p_i)_{i\in I} \in R p = ( p i ) i ∈ I ∈ R . Then for each a ∈ A a \in A a ∈ A and i ∈ I i \in I i ∈ I , a i ∈ A i = ↓ p i a_i \in A_i = \mathord{\downarrow}p_i a i ∈ A i = ↓ p i . As a result, a ∈ ↓ p a \in\mathord{\downarrow}p a ∈ ↓ p . This yields that p p p is larger than any elements coming from A A A . It remains to show that p ∈ A p \in A p ∈ A .
As A ≠ ∅ A \ne\varnothing A = ∅ , we can select a b = ( b i ) i ∈ I ∈ A b = (b_i)_{i\in I} \in A b = ( b i ) i ∈ I ∈ A . Let Fin ( I ) \operatorname{Fin}(I) Fin ( I ) be the collection of all finite subsets of I I I , containing the empty set. For each F ∈ Fin ( I ) F \in\operatorname{Fin}(I) F ∈ Fin ( I ) , define
b i F = { p i , i ∈ F , b i , i ∉ F . b_i^F =
\begin{cases}
p_i, & i \in F,\\
b_i, & i \notin F.
\end{cases} b i F = { p i , b i , i ∈ F , i ∈ / F .
As each p i ∈ A i p_i \in A_i p i ∈ A i and b ∈ A b \in A b ∈ A , by Lemma 3.4(3), we know b F = ( b i F ) i ∈ I ∈ A b^F = (b_i^F)_{i\in I} \in A b F = ( b i F ) i ∈ I ∈ A for every F ∈ Fin ( I ) F \in\operatorname{Fin}(I) F ∈ Fin ( I ) . So each b F ≤ p b^F \leq p b F ≤ p . Since each p i p_i p i is the largest element in A i A_i A i , we deduce that b F ≤ b G b^F \leq b^G b F ≤ b G if F ⊆ G F \subseteq G F ⊆ G . Thus the set
D = { b F ∣ F ∈ Fin ( I ) } D = \{b^F \mid F \in\operatorname{Fin}(I)\} D = { b F ∣ F ∈ Fin ( I )}
is directed and contained in A A A . Clearly, p p p is an upper bound of D D D . For each i ∈ I i \in I i ∈ I , choose F ~ = { i } \widetilde{F} = \{i\} F = { i } . Then b F ~ ∈ D b^{\widetilde{F}} \in D b F ∈ D and hence p i = π i ( b F ~ ) ∈ π i ( D ) p_i = \pi_i(b^{\widetilde{F}}) \in\pi_i(D) p i = π i ( b F ) ∈ π i ( D ) . Consequently, we obtain
sup R D = ( ⋁ P i π i ( D ) ) i ∈ I = ( p i ) i ∈ I = p . \sup_R D = \left(\bigvee_{P_i} \pi_i(D)\right)_{i\in I} = (p_i)_{i\in I} = p. R sup D = ( P i ⋁ π i ( D ) ) i ∈ I = ( p i ) i ∈ I = p .
Since A A A is Scott closed and D ⊆ A D \subseteq A D ⊆ A is directed, we have p = sup D ∈ A p = \sup D \in A p = sup D ∈ A . Now, we can conclude that p p p is a largest element of A A A . Therefore, A = ↓ p A = \mathord{\downarrow}p A = ↓ p . By Theorem 3.3, Σ R \Sigma R Σ R is well-filtered.
Lemma 3.6. ([21 ] ) A retract of a well-filtered space is well-filtered.
Corollary 3.7. Let R = ∏ i ∈ I P i R = \prod_{i\in I} P_i R = ∏ i ∈ I P i be a nonempty order product of dcpos. Then the following statements are equivalent:
Σ R \Sigma R Σ R is well-filtered;
Each Σ P i \Sigma P_i Σ P i is well-filtered.
Proof. (1) ⟹ \Longrightarrow ⟹ (2) Since each Σ P i \Sigma P_i Σ P i is a retract of Σ R \Sigma R Σ R , by Lemma 3.6, we have Σ P i \Sigma P_i Σ P i is well-filtered. (2) ⟹ \Longrightarrow ⟹ (1) It is straightforward by using Theorem 3.5.
The Scott well-filteredness of the product for Smyth powerdomain
In this section, we consider when Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) is well-filtered under the assumption that Σ K ( P ) \Sigma K(P) Σ K ( P ) and Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are well-filtered.
Lemma 4.1. Let P , Q P, Q P , Q be a pair of dcpos and H ∈ K ( P ) H \in K(P) H ∈ K ( P ) . If C ⊆ P × Q C \subseteq P \times Q C ⊆ P × Q is Scott closed, then C ( H ) C(H) C ( H ) is Scott closed in Q Q Q , where
C ( H ) = π Q ( C ∩ ( H × Q ) ) = { q ∈ Q ∣ ∃ p ∈ H , ( p , q ) ∈ C } . C(H) = \pi_Q(C \cap(H \times Q)) = \{q \in Q \mid\exists p \in H,\ (p,q) \in C\}. C ( H ) = π Q ( C ∩ ( H × Q )) = { q ∈ Q ∣ ∃ p ∈ H , ( p , q ) ∈ C } .
Proof. Claim 1: C ( H ) C(H) C ( H ) is a lower set. Suppose q ∈ C ( H ) q \in C(H) q ∈ C ( H ) and q 0 ≤ q q_0 \leq q q 0 ≤ q . Then there exists p ∈ H p \in H p ∈ H with ( p , q ) ∈ C (p,q) \in C ( p , q ) ∈ C . As C ⊆ P × Q C \subseteq P \times Q C ⊆ P × Q is a lower set and ( p , q 0 ) ≤ ( p , q ) (p,q_0) \leq(p,q) ( p , q 0 ) ≤ ( p , q ) , we have ( p , q 0 ) ∈ C (p,q_0) \in C ( p , q 0 ) ∈ C . As a result, q 0 ∈ C ( H ) q_0 \in C(H) q 0 ∈ C ( H ) . This proves Claim 1.
Claim 2: C ( H ) C(H) C ( H ) is closed under directed suprema in Q Q Q . Let D ⊆ C ( H ) D \subseteq\mathcal{C}(H) D ⊆ C ( H ) be a directed subset and q 1 = sup D q_1 = \sup D q 1 = sup D . For every d ∈ D d \in D d ∈ D , take
F d = { h ∈ H ∣ ( h , d ) ∈ C } = H ∩ C d F_d = \{h \in H \mid(h,d) \in C\} = H \cap C_d F d = { h ∈ H ∣ ( h , d ) ∈ C } = H ∩ C d
where C d = { x ∈ P ∣ ( x , d ) ∈ C } C_d = \{x \in P \mid(x,d) \in C\} C d = { x ∈ P ∣ ( x , d ) ∈ C } . Then F d F_d F d is nonempty as d ∈ C ( H ) d \in\mathcal{C}(H) d ∈ C ( H ) . One can check that C d C_d C d is Scott closed in P P P . Choose any d 1 , … , d n ∈ D d_1,\ldots,d_n \in D d 1 , … , d n ∈ D . The directedness of D D D provides d 0 ∈ D d_0 \in D d 0 ∈ D with d k ≤ d 0 d_k \leq d_0 d k ≤ d 0 for each k k k . If p ∈ F d 0 p \in F_{d_0} p ∈ F d 0 , then ( p , d 0 ) ∈ C (p,d_0) \in C ( p , d 0 ) ∈ C . Since C C C is a lower set, each ( p , d k ) (p,d_k) ( p , d k ) belongs to C C C . In other words,
∅ ≠ F d 0 ⊆ ⋂ k = 1 n F d k . \varnothing\neq F_{d_0} \subseteq\bigcap_{k=1}^{n} F_{d_k}. ∅ = F d 0 ⊆ k = 1 ⋂ n F d k .
As H H H is compact, we can choose a p 0 ∈ ⋂ d ∈ D F d p_0 \in\bigcap_{d \in D} F_d p 0 ∈ ⋂ d ∈ D F d . Equivalently, ( p 0 , d ) ∈ C (p_0,d) \in C ( p 0 , d ) ∈ C for each d ∈ D d \in D d ∈ D . Now, the directed set { ( p 0 , d ) ∣ d ∈ D } \{(p_0,d) \mid d \in D\} {( p 0 , d ) ∣ d ∈ D } is contained in C C C and further
⋁ d ∈ D ( p 0 , d ) = ( p 0 , q 1 ) . \bigvee_{d \in D} (p_0,d) = (p_0,q_1). d ∈ D ⋁ ( p 0 , d ) = ( p 0 , q 1 ) .
It follows from the Scott closedness of C C C that ( p 0 , q 1 ) ∈ C (p_0,q_1) \in C ( p 0 , q 1 ) ∈ C . As p 0 ∈ H p_0 \in H p 0 ∈ H , this yields that q 1 ∈ C [ H ] q_1 \in C[H] q 1 ∈ C [ H ] . This proves Claim 2.
By these two claims, C ( H ) C(H) C ( H ) is Scott closed in Q Q Q . □
Remark 4.2. In Lemma 4.1, it suffices to assume that H H H is a compact subset of Σ P \Sigma P Σ P . The saturation of H H H is useless.
Corollary 4.3. Let H H H be a compact subset in Σ P \Sigma P Σ P and U ∈ σ ( P × Q ) U \in\sigma(P \times Q) U ∈ σ ( P × Q ) . Then
V H ( U ) = { q ∈ Q ∣ H × { q } ⊆ U } V_H(U) = \{q \in Q \mid H \times\{q\} \subseteq U\} V H ( U ) = { q ∈ Q ∣ H × { q } ⊆ U }
is Scott open in Q Q Q .
Proof. Set C = ( P × Q ) ∖ U C = (P \times Q) \setminus U C = ( P × Q ) ∖ U . One can check that
Q ∖ V H ( U ) = { q ∈ Q ∣ ∃ p ∈ H , ( p , q ) ∈ C } = C ( H ) . Q \setminus V_H(U) = \{q \in Q \mid\exists p \in H,\ (p,q) \in C\} = C(H). Q ∖ V H ( U ) = { q ∈ Q ∣ ∃ p ∈ H , ( p , q ) ∈ C } = C ( H ) .
By Lemma 4.1 and Remark 4.2, C ( H ) C(H) C ( H ) is Scott closed in Q Q Q . Hence, V H ( U ) ∈ σ ( Q ) V_H(U) \in\sigma(Q) V H ( U ) ∈ σ ( Q ) . □
Proposition 4.4. Let H H H be a compact subset in Σ P \Sigma P Σ P and K K K be a compact subset in Σ Q \Sigma Q Σ Q . Then H × K H \times K H × K is compact in Σ ( P × Q ) \Sigma(P \times Q) Σ ( P × Q ) . Furthermore, if H , K H,K H , K are nonempty and saturated, then H × K H \times K H × K is also nonempty compact and saturated in Σ ( P × Q ) \Sigma(P \times Q) Σ ( P × Q ) .
Proof. This proposition holds if either H H H or K K K is empty. So we suppose that H H H and K K K are nonempty. Let { U i ∣ i ∈ I } ⊆ σ ( P × Q ) \{U_i \mid i \in I\} \subseteq\sigma(P \times Q) { U i ∣ i ∈ I } ⊆ σ ( P × Q ) be an open cover of H × K H \times K H × K . For every finite subset F ⊆ I F \subseteq I F ⊆ I , set
U F = ⋃ i ∈ F U i and V F = { q ∈ Q ∣ H × { q } ⊆ U F } . U_F = \bigcup_{i \in F} U_i \text{ and } V_F = \{q \in Q \mid H \times\{q\} \subseteq U_F\}. U F = i ∈ F ⋃ U i and V F = { q ∈ Q ∣ H × { q } ⊆ U F } .
By Corollary 4.3, V F ∈ σ ( Q ) V_F \in\sigma(Q) V F ∈ σ ( Q ) . Clearly, V F ⊆ V G V_F \subseteq V_G V F ⊆ V G if F ⊆ G F \subseteq G F ⊆ G .
Fix q ∈ K q \in K q ∈ K . For each i ∈ I i \in I i ∈ I , it follows from U i ∈ σ ( P × Q ) U_i \in\sigma(P \times Q) U i ∈ σ ( P × Q ) that the set
W i , q = { p ∈ P ∣ ( p , q ) ∈ U i } W_{i,q} = \{p \in P \mid(p,q) \in U_i\} W i , q = { p ∈ P ∣ ( p , q ) ∈ U i }
is Scott open in Σ P \Sigma P Σ P . As all U i U_i U i cover H × K H \times K H × K , the collection { W i , q ∣ i ∈ I } \{W_{i,q} \mid i \in I\} { W i , q ∣ i ∈ I } covers the compact set H H H for each q ∈ K q \in K q ∈ K . The compactness of H H H gives that there is a finite F q ⊆ I F_q \subseteq I F q ⊆ I such that
H ⊆ ⋃ i ∈ F q W i , q . H \subseteq\bigcup_{i \in F_q} W_{i,q}. H ⊆ i ∈ F q ⋃ W i , q .
Equivalently, H × { q } ⊆ U F q H \times\{q\} \subseteq U_{F_q} H × { q } ⊆ U F q and therefore q ∈ V F q q \in V_{F_q} q ∈ V F q . Consequently, { V F ∣ F ⊆ I finite } \{V_F \mid F \subseteq I\text{ finite}\} { V F ∣ F ⊆ I finite } forms an open cover of K K K . By the compactness of K K K , there exist finite subsets F 1 , … , F m F_1,\ldots,F_m F 1 , … , F m contained in I I I satisfying
K ⊆ V F 1 ∪ ⋯ ∪ V F m . K \subseteq V_{F_1} \cup\cdots\cup V_{F_m}. K ⊆ V F 1 ∪ ⋯ ∪ V F m .
Put F ∗ = F 1 ∪ ⋯ ∪ F m F^* = F_1 \cup\cdots\cup F_m F ∗ = F 1 ∪ ⋯ ∪ F m . Then F ∗ F^* F ∗ is finite and satisfies
K ⊆ V F ∗ and H × K ⊆ U F ∗ . K \subseteq V_{F^*}\quad\text{and}\quad H \times K \subseteq U_{F^*}. K ⊆ V F ∗ and H × K ⊆ U F ∗ .
This shows that H × K H \times K H × K is compact in Σ ( P × Q ) \Sigma(P \times Q) Σ ( P × Q ) . The saturation of H × K H \times K H × K is valid if H H H and K K K are saturated.
By Proposition 4.4, for two dcpos P P P and Q Q Q , K ( P ) × K ( Q ) K(P) \times K(Q) K ( P ) × K ( Q ) can be order embedded into K ( P × Q ) K(P \times Q) K ( P × Q ) by the product operator. The following example shows that K ( P ) × K ( Q ) K(P) \times K(Q) K ( P ) × K ( Q ) is not equal to K ( P × Q ) K(P \times Q) K ( P × Q ) in general.
Example 4.5. Let P = N ∪ { ∞ } P = \mathbb{N} \cup\{\infty\} P = N ∪ { ∞ } be the dcpo ordered by
0 ≤ 1 ≤ ⋯ ≤ n ≤ n + 1 ≤ ⋯ ≤ ∞ . 0 \le1 \le\cdots\le n \le n+1 \le\cdots\le\infty. 0 ≤ 1 ≤ ⋯ ≤ n ≤ n + 1 ≤ ⋯ ≤ ∞. Then P P P is a complete lattice and hence Σ P \Sigma P Σ P is well-filtered. One can verify that K ( P ) = { ↑ x ∣ x ∈ P } K(P) = \{\uparrow x \mid x \in P\} K ( P ) = { ↑ x ∣ x ∈ P } . Consider A = ↑ P × P ( 1 , 0 ) ∪ ↑ P × P ( 0 , 1 ) A = \uparrow_{P\times P}(1,0) \cup\uparrow_{P\times P}(0,1) A = ↑ P × P ( 1 , 0 ) ∪ ↑ P × P ( 0 , 1 ) . Clearly, A ∈ K ( P × P ) A \in K(P \times P) A ∈ K ( P × P ) but A ∉ K ( P ) × K ( P ) A \notin K(P) \times K(P) A ∈ / K ( P ) × K ( P ) . So we have K ( P ) × K ( P ) ≠ K ( P × P ) K(P) \times K(P) \ne K(P \times P) K ( P ) × K ( P ) = K ( P × P ) . Furthermore, K ( P × P ) K(P \times P) K ( P × P ) is not isomorphic to K ( P ) × K ( P ) K(P) \times K(P) K ( P ) × K ( P ) . Note that K ( P ) K(P) K ( P ) is order isomorphic to P P P . Choose any three elements α = ( a , b ) \alpha= (a,b) α = ( a , b ) , β = ( c , d ) \beta= (c,d) β = ( c , d ) and γ = ( e , f ) \gamma= (e,f) γ = ( e , f ) in P × P P \times P P × P . Then we have
α ∧ β ∧ γ = ( min { a , c , e } , min { b , d , f } ) = min { α ∧ β , α ∧ γ , β ∧ γ } . \begin{aligned}
\alpha\wedge\beta\wedge\gamma
&= (\min\{a,c,e\},\min\{b,d,f\})\\
&= \min\{\alpha\wedge\beta,\alpha\wedge\gamma,\beta\wedge\gamma\}.
\end{aligned} α ∧ β ∧ γ = ( min { a , c , e } , min { b , d , f }) = min { α ∧ β , α ∧ γ , β ∧ γ } . Set B = ↑ P × P ( 0 , 2 ) B = \uparrow_{P\times P}(0,2) B = ↑ P × P ( 0 , 2 ) , C = ↑ P × P ( 1 , 1 ) C = \uparrow_{P\times P}(1,1) C = ↑ P × P ( 1 , 1 ) and D = ↑ P × P ( 2 , 0 ) D = \uparrow_{P\times P}(2,0) D = ↑ P × P ( 2 , 0 ) in K ( P × P ) K(P \times P) K ( P × P ) . Then B B B , C C C and D D D belong to K ( P × P ) K(P \times P) K ( P × P ) and
B ∧ C ∧ D = B ∪ C ∪ D ∉ { B ∧ C , B ∧ D , C ∧ D } . B \wedge C \wedge D = B \cup C \cup D \notin\{B \wedge C,B \wedge D,C \wedge D\}. B ∧ C ∧ D = B ∪ C ∪ D ∈ / { B ∧ C , B ∧ D , C ∧ D } . This means that K ( P × P ) ≇ P × P K(P \times P) \not\cong P \times P K ( P × P ) ≅ P × P . Therefore, K ( P × P ) ≇ K ( P ) × K ( P ) K(P \times P) \not\cong K(P) \times K(P) K ( P × P ) ≅ K ( P ) × K ( P ) .
Lemma 4.6. ([2 ] ) Let X X X be a T 0 T_0 T 0 space and A ⊆ K ( X ) \mathcal{A} \subseteq K(X) A ⊆ K ( X ) a nonempty family. Then ⋁ K ( X ) A \bigvee_{K(X)} \mathcal{A} ⋁ K ( X ) A exists if and only if ⋂ A ∈ K ( X ) \bigcap\mathcal{A} \in K(X) ⋂ A ∈ K ( X ) . In this case, ⋁ K ( X ) A = ⋂ A \bigvee_{K(X)} \mathcal{A} = \bigcap\mathcal{A} ⋁ K ( X ) A = ⋂ A .
Lemma 4.7. ([2 ] ) Let X X X be a well-filtered T 0 T_0 T 0 space and K \mathcal{K} K a filtered family consisting of nonempty compact saturated subsets of X X X . Then ⋂ K \bigcap\mathcal{K} ⋂ K is also a nonempty compact saturated set.
Theorem 4.8. Let P P P and Q Q Q be dcpos. If Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) is well-filtered, then Σ K ( P ) \Sigma K(P) Σ K ( P ) and Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are well-filtered.
Proof. We first prove that K ( P ) K(P) K ( P ) and K ( Q ) K(Q) K ( Q ) are dcpos. Fix a q 0 ∈ Q q_0 \in Q q 0 ∈ Q , define a mapping e P : K ( P ) ⟶ K ( P × Q ) e_P: K(P) \longrightarrow K(P \times Q) e P : K ( P ) ⟶ K ( P × Q ) by e P ( A ) = A × ↑ Q q 0 e_P(A) = A \times\uparrow_Q q_0 e P ( A ) = A × ↑ Q q 0 . By Proposition 4.4, e P e_P e P is well defined. Let { A i ∣ i ∈ I } \{A_i \mid i \in I\} { A i ∣ i ∈ I } be a nonempty filtered family in K ( P ) K(P) K ( P ) . Then { e P ( A i ) ∣ i ∈ I } \{e_P(A_i) \mid i \in I\} { e P ( A i ) ∣ i ∈ I } forms a directed subset of K ( P × Q ) K(P \times Q) K ( P × Q ) . Since Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) is well-filtered, K ( P × Q ) K(P \times Q) K ( P × Q ) is a dcpo. By Lemma 4.6 and Lemma 4.7, we can fix a B ∈ K ( P × Q ) B \in K(P \times Q) B ∈ K ( P × Q ) such that
B = ⋂ i ∈ I e P ( A i ) = ⋂ i ∈ I ( A i × ↑ Q q 0 ) = ( ⋂ i ∈ I A i ) × ↑ Q q 0 . B = \bigcap_{i\in I}e_P(A_i)=\bigcap_{i\in I}(A_i\times\uparrow_Qq_0)=\left(\bigcap_{i\in I}A_i\right)\times\uparrow_Qq_0. B = i ∈ I ⋂ e P ( A i ) = i ∈ I ⋂ ( A i × ↑ Q q 0 ) = ( i ∈ I ⋂ A i ) × ↑ Q q 0 .
Consequently, π P ( B ) = ⋂ i ∈ I A i \pi_P(B)=\bigcap_{i\in I} A_i π P ( B ) = ⋂ i ∈ I A i is a nonempty compact saturated subset of Σ P \Sigma P Σ P as π P \pi_P π P is Scott continuous. By Lemma 4.6 and Lemma 4.7 again, ⋂ i ∈ I A i = ⋁ i ∈ I A i \bigcap_{i\in I} A_i=\bigvee_{i\in I} A_i ⋂ i ∈ I A i = ⋁ i ∈ I A i . Therefore K ( P ) K(P) K ( P ) is a dcpo.
Similarly, K ( Q ) K(Q) K ( Q ) is also a dcpo.
Define r P : K ( P × Q ) ⟶ K ( P ) r_P:K(P\times Q)\longrightarrow K(P) r P : K ( P × Q ) ⟶ K ( P ) by r P ( B ) = π P ( B ) r_P(B)=\pi_P(B) r P ( B ) = π P ( B ) . Then π P ( B ) \pi_P(B) π P ( B ) is a nonempty compact saturated subset of Σ P \Sigma P Σ P . Thus r P r_P r P is well defined. One can check that e P e_P e P is Scott continuous and r P ∘ e P = id K ( P ) r_P\circ e_P=\operatorname{id}_{K(P)} r P ∘ e P = id K ( P ) .
Let { B j ∣ j ∈ J } \{B_j\mid j\in J\} { B j ∣ j ∈ J } be a nonempty filtered family in K ( P × Q ) K(P\times Q) K ( P × Q ) . Clearly, π P ( ⋂ j ∈ J B j ) ⊆ ⋂ j ∈ J π P ( B j ) \pi_P\left(\bigcap_{j\in J}B_j\right)\subseteq\bigcap_{j\in J}\pi_P(B_j) π P ( ⋂ j ∈ J B j ) ⊆ ⋂ j ∈ J π P ( B j ) . Conversely, choose an arbitrary p ∈ ⋂ j ∈ J π P ( B j ) p\in\bigcap_{j\in J}\pi_P(B_j) p ∈ ⋂ j ∈ J π P ( B j ) and put
F j = { q ∈ Q ∣ ( p , q ) ∈ B j } . F_j=\{q\in Q\mid(p,q)\in B_j\}. F j = { q ∈ Q ∣ ( p , q ) ∈ B j } .
Each F j F_j F j is nonempty and saturated in Q Q Q . In addition,
F j = π Q ( B j ∩ ( ↓ P p × Q ) ) , F_j=\pi_Q(B_j\cap(\mathord{\downarrow_P}p\times Q)), F j = π Q ( B j ∩ ( ↓ P p × Q )) ,
because B j B_j B j is upper.
Since ↓ P p \mathord{\downarrow_P}p ↓ P p is Scott closed, ↓ P p × Q = π P − 1 ( ↓ P p ) \mathord{\downarrow_P}p\times Q=\pi_P^{-1}(\mathord{\downarrow_P}p) ↓ P p × Q = π P − 1 ( ↓ P p ) is Scott closed in P × Q P\times Q P × Q . So we have B j ∩ ( ↓ P p × Q ) B_j\cap(\mathord{\downarrow_P}p\times Q) B j ∩ ( ↓ P p × Q ) is compact and hence F j ∈ K ( Q ) F_j\in K(Q) F j ∈ K ( Q ) . The family { F λ : λ ∈ Λ } \{F_\lambda:\lambda\in\Lambda\} { F λ : λ ∈ Λ } is filtered as the family { B j ∣ j ∈ J } \{B_j\mid j\in J\} { B j ∣ j ∈ J } is filtered. As K ( Q ) K(Q) K ( Q ) is a dcpo, by Lemma 4.6, we have ⋂ j ∈ J F j \bigcap_{j\in J}F_j ⋂ j ∈ J F j is a nonempty compact saturated subset of Σ Q \Sigma Q Σ Q . Choose
q ∈ ⋂ j ∈ J F j . q\in\bigcap_{j\in J}F_j. q ∈ j ∈ J ⋂ F j .
This leads to ( p , q ) ∈ ⋂ j ∈ J B j (p,q)\in\bigcap_{j\in J}B_j ( p , q ) ∈ ⋂ j ∈ J B j and hence p ∈ π P ( ⋂ j ∈ J B j ) p\in\pi_P\left(\bigcap_{j\in J}B_j\right) p ∈ π P ( ⋂ j ∈ J B j ) . As a result, π P ( ⋂ j ∈ J B j ) = ⋂ j ∈ J π P ( B j ) \pi_P\left(\bigcap_{j\in J}B_j\right)=\bigcap_{j\in J}\pi_P(B_j) π P ( ⋂ j ∈ J B j ) = ⋂ j ∈ J π P ( B j ) . So r P r_P r P is Scott continuous. Consequently, Σ K ( P ) \Sigma K(P) Σ K ( P ) is a retract of Σ K ( P × Q ) \Sigma K(P\times Q) Σ K ( P × Q ) . By Lemma 3.6 , Σ K ( P ) \Sigma K(P) Σ K ( P ) is well-filtered.
Naturally, it arises the following problem:
Problem 4.9. Let P P P and Q Q Q be dcpos such that Σ K ( P ) \Sigma K(P) Σ K ( P ) , Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are well-filtered and K ( P × Q ) K(P\times Q) K ( P × Q ) is a dcpo. Is Σ K ( P × Q ) \Sigma K(P\times Q) Σ K ( P × Q ) well-filtered?
In the following, we will give a partial answer to the above problem (see Theorem 4.17).
Lemma 4.10. ([2 ] ) A closed subspace of a well-filtered space is well-filtered.
Lemma 4.11. ([8 ] ) Let P P P be a dcpo and A A A a nonempty Scott closed subset of P P P . Then A A A is a subdcpo and its Scott topology is the subspace topology inherited from Σ P \Sigma P Σ P .
Lemma 4.12. [22 ] Let K \mathcal{K} K be a nonempty family consisting of nonempty compact subsets of a topological space X X X and F ⊆ X F\subseteq X F ⊆ X a nonempty closed subset such that F ∈ M ( K ) F\in M(\mathcal{K}) F ∈ M ( K ) . Then there is a closed set A A A contained in F F F such that A A A is a minimal element of M ( K ) M(\mathcal{K}) M ( K ) . In particular, if K ⊆ K ( X ) \mathcal{K}\subseteq K(X) K ⊆ K ( X ) is filtered, then A A A is an irreducible closed subset of X X X .
Lemma 4.13. ([22]) Let f : X → Y f : X \to Y f : X → Y be a continuous map between T 0 T_{0} T 0 spaces X X X and Y Y Y . If A ∈ RD ( X ) A \in\operatorname{RD}(X) A ∈ RD ( X ) , then cl Y ( f ( A ) ) ∈ RD ( Y ) \operatorname{cl}_{Y}(f(A)) \in\operatorname{RD}(Y) cl Y ( f ( A )) ∈ RD ( Y ) .
Lemma 4.14. Let P P P be a dcpo and { q i ∣ i ∈ I } \{q_i \mid i \in I\} { q i ∣ i ∈ I } be a directed set consisting of Scott continuous self-maps on P P P such that ⋁ i ∈ I q i = id P \bigvee_{i \in I} q_i = \operatorname{id}_{P} ⋁ i ∈ I q i = id P . If there exist a Scott well-filtered dcpo W i W_i W i and two Scott continuous mappings f i : P → W i f_i : P \to W_i f i : P → W i and g i : W i → P g_i : W_i \to P g i : W i → P such that q i = g i ∘ f i q_i = g_i \circ f_i q i = g i ∘ f i for each i i i , then Σ P \Sigma P Σ P is well-filtered.
Proof. Let A ⊆ P A \subseteq P A ⊆ P be a closed Rudin set. By Lemma 4.13, cl Σ W i ( f i ( A ) ) \operatorname{cl}_{\Sigma W_i}(f_i(A)) cl Σ W i ( f i ( A )) is a closed Rudin set in Σ W i \Sigma W_i Σ W i . As Σ W i \Sigma W_i Σ W i is well-filtered, by Theorem 3.3, there is w i ∈ W i w_i \in W_i w i ∈ W i with
cl Σ W i ( f i ( A ) ) = ↓ W i w i . \operatorname{cl}_{\Sigma W_i}(f_i(A)) = \mathord{\downarrow}_{W_i} w_i. cl Σ W i ( f i ( A )) = ↓ W i w i .
Take a i = g i ( w i ) a_i = g_i(w_i) a i = g i ( w i ) . As g i g_i g i is an order preserving, we have
q i ( A ) = g i ( f i ( A ) ) ⊆ g i ( ↓ W i w i ) ⊆ ↓ P a i . q_i(A) = g_i(f_i(A)) \subseteq g_i(\mathord{\downarrow}_{W_i} w_i) \subseteq\mathord{\downarrow}_{P} a_i. q i ( A ) = g i ( f i ( A )) ⊆ g i ( ↓ W i w i ) ⊆ ↓ P a i .
Since g i g_i g i is Scott continuous, we know
a i ∈ g i ( cl Σ W i f i ( A ) ) ⊆ cl Σ P ( q i ( A ) ) . a_i \in g_i(\operatorname{cl}_{\Sigma W_i} f_i(A)) \subseteq\operatorname{cl}_{\Sigma P}(q_i(A)). a i ∈ g i ( cl Σ W i f i ( A )) ⊆ cl Σ P ( q i ( A )) .
As ↓ P a i \mathord{\downarrow}_{P} a_i ↓ P a i is Scott closed, we deduce that cl Σ P ( q i ( A ) ) = ↓ P a i \operatorname{cl}_{\Sigma P}(q_i(A)) = \mathord{\downarrow}_{P} a_i cl Σ P ( q i ( A )) = ↓ P a i . It follows from q i ≤ id P q_i \leq\operatorname{id}_{P} q i ≤ id P and the Scott closedness of A A A that q i ( A ) ⊆ A q_i(A) \subseteq A q i ( A ) ⊆ A . This implies a i ∈ A a_i \in A a i ∈ A .
Choose any q i , q j q_i, q_j q i , q j . We can select a q k q_k q k with q i , q j ≤ q k q_i, q_j \leq q_k q i , q j ≤ q k . Then we have q i ( A ) , q j ( A ) ⊆ ↓ P a k q_i(A), q_j(A) \subseteq\mathord{\downarrow}_{P} a_k q i ( A ) , q j ( A ) ⊆ ↓ P a k . As a consequence, ↓ P a i = cl Σ P ( q i ( A ) ) ⊆ ↓ a k \mathord{\downarrow}_{P} a_i = \operatorname{cl}_{\Sigma P}(q_i(A)) \subseteq\mathord{\downarrow} a_k ↓ P a i = cl Σ P ( q i ( A )) ⊆ ↓ a k and ↓ P a j = cl Σ P ( q j ( A ) ) ⊆ ↓ a k \mathord{\downarrow}_{P} a_j = \operatorname{cl}_{\Sigma P}(q_j(A)) \subseteq\mathord{\downarrow} a_k ↓ P a j = cl Σ P ( q j ( A )) ⊆ ↓ a k . This yields that { a i ∣ i ∈ I } \{a_i \mid i \in I\} { a i ∣ i ∈ I } is directed. Take a = ⋁ i ∈ I a i a = \bigvee_{i \in I} a_i a = ⋁ i ∈ I a i . Then a ∈ A a \in A a ∈ A since A A A is Scott closed. For every x ∈ A x \in A x ∈ A ,
x = ⋁ i ∈ I q i ( x ) ≤ ⋁ i ∈ I a i = a . x = \bigvee_{i \in I} q_i(x) \leq\bigvee_{i \in I} a_i = a. x = i ∈ I ⋁ q i ( x ) ≤ i ∈ I ⋁ a i = a .
Thus A = ↓ P a A = \mathord{\downarrow}_{P} a A = ↓ P a because A A A is lower. By Theorem 3.3, Σ P \Sigma P Σ P is well-filtered. □
Lemma 4.15. Let M M M be a dcpo such that Σ K ( M ) \Sigma K(M) Σ K ( M ) is well-filtered. If F F F is a finite poset, then Σ K ( M × F ) \Sigma K(M \times F) Σ K ( M × F ) is well-filtered.
Proof. Set
S = K ( M ) , and T = S ∪ { ∅ } , S = K(M), \text{ and } T = S \cup\{\varnothing\}, S = K ( M ) , and T = S ∪ { ∅ } ,
both ordered by the reverse inclusion. As Σ K ( M ) \Sigma K(M) Σ K ( M ) is well-filtered, S S S is a dcpo. Hence T T T is also a dcpo with a greatest element ⊤ T = ∅ \top_{T} = \varnothing ⊤ T = ∅ and further ⊤ T \top_{T} ⊤ T is a compact element. It is not difficult to check that
K ( T ) = { Q ∪ { ⊤ T } ∣ Q ∈ K ( S ) ∪ { ∅ } } K(T) = \{Q \cup\{\top_{T}\} \mid Q \in K(S) \cup\{\varnothing\}\} K ( T ) = { Q ∪ { ⊤ T } ∣ Q ∈ K ( S ) ∪ { ∅ }}
and
σ ( T ) = { A ∪ { ⊤ T } ∣ A ∈ σ ( S ) } ∪ { ∅ } . \sigma(T) = \{A \cup\{\top_{T}\} \mid A \in\sigma(S)\} \cup\{\varnothing\}. σ ( T ) = { A ∪ { ⊤ T } ∣ A ∈ σ ( S )} ∪ { ∅ } .
It follows from the well-filteredness of Σ S \Sigma S Σ S that Σ T \Sigma T Σ T is well-filtered. We enumerate F F F as { a 1 , a 2 , ⋯ , a n } \{a_1, a_2, \cdots, a_n\} { a 1 , a 2 , ⋯ , a n } . Let
B = T n , ⊤ B = ( ⊤ T , ⊤ T , ⋯ , ⊤ T ) ∈ B and B 0 = B ∖ { ⊤ B } . B = T^{n}, \quad\top_{B} = (\top_{T}, \top_{T}, \cdots, \top_{T}) \in B \text{ and } B_{0} = B \setminus\{\top_{B}\}. B = T n , ⊤ B = ( ⊤ T , ⊤ T , ⋯ , ⊤ T ) ∈ B and B 0 = B ∖ { ⊤ B } .
By Theorem 3.5, Σ B \Sigma B Σ B is well-filtered. The greatest element ⊤ B \top_{B} ⊤ B is compact in B B B because ⊤ T \top_{T} ⊤ T is compact in T T T . So { ⊤ B } \{\top_{B}\} { ⊤ B } is Scott open in B B B and hence B 0 B_{0} B 0 is Scott closed. By Lemma 4.10 and Lemma 4.11, Σ B 0 \Sigma B_{0} Σ B 0 is well-filtered. For D = ( D 1 , D 2 , ⋯ , D n ) ∈ B D = (D_{1}, D_{2}, \cdots, D_{n}) \in B D = ( D 1 , D 2 , ⋯ , D n ) ∈ B and k ∈ { 1 , ⋯ , n } k \in\{1, \cdots, n\} k ∈ { 1 , ⋯ , n } define
p D ( k ) = ⋃ { D m ∣ m ∈ { 1 , ⋯ , n } , a m ≤ F a k } . p_{D}(k) = \bigcup\{D_{m} \mid m \in\{1, \cdots, n\}, a_{m} \leq_{F} a_{k}\}. p D ( k ) = ⋃ { D m ∣ m ∈ { 1 , ⋯ , n } , a m ≤ F a k } .
Then each p D ( k ) ∈ T p_D(k) \in T p D ( k ) ∈ T . The mapping p : B ⟶ B p : B \longrightarrow B p : B ⟶ B is defined by
∀ D ∈ B , p ( D ) = ( p D ( 1 ) , p D ( 2 ) , ⋯ , p D ( n ) ) . \forall D \in B,\quad p(D) = (p_D(1),p_D(2),\cdots,p_D(n)). ∀ D ∈ B , p ( D ) = ( p D ( 1 ) , p D ( 2 ) , ⋯ , p D ( n )) .
Then we can check that p p p is an order preserving and p ( ⊤ B ) = ⊤ B p(\top_B)=\top_B p ( ⊤ B ) = ⊤ B .
Let { D i ∣ i ∈ I } ⊆ B \{D_i \mid i \in I\} \subseteq B { D i ∣ i ∈ I } ⊆ B be a directed subset. For convenience, put D i = ( D i , 1 , D i , 2 , ⋯ D i , n ) D_i=(D_{i,1},D_{i,2},\cdots D_{i,n}) D i = ( D i , 1 , D i , 2 , ⋯ D i , n ) for each i ∈ I i \in I i ∈ I . Then by Lemma 4.6,
p ( ⋁ i ∈ I D i ) = p ( ⋁ i ∈ I D i , 1 , ⋁ i ∈ I D i , 2 , ⋯ , ⋁ i ∈ I D i , n ) = p ( ⋂ i ∈ I D i , 1 , ⋂ i ∈ I D i , 2 , ⋯ , ⋂ i ∈ I D i , n ) = ( ⋃ a m ≤ F a 1 ⋂ i ∈ I D i , m , ⋃ a m ≤ F a 2 ⋂ i ∈ I D i , m , ⋯ , ⋃ a m ≤ F a n ⋂ i ∈ I D i , m ) \begin{aligned}
p\left(\bigvee_{i\in I}D_i\right)
&=p\left(\bigvee_{i\in I}D_{i,1},\bigvee_{i\in I}D_{i,2},\cdots,\bigvee_{i\in I}D_{i,n}\right)\\
&=p\left(\bigcap_{i\in I}D_{i,1},\bigcap_{i\in I}D_{i,2},\cdots,\bigcap_{i\in I}D_{i,n}\right)\\
&=\left(\bigcup_{a_m\le_F a_1}\bigcap_{i\in I}D_{i,m},\bigcup_{a_m\le_F a_2}\bigcap_{i\in I}D_{i,m},\cdots,\bigcup_{a_m\le_F a_n}\bigcap_{i\in I}D_{i,m}\right)
\end{aligned} p ( i ∈ I ⋁ D i ) = p ( i ∈ I ⋁ D i , 1 , i ∈ I ⋁ D i , 2 , ⋯ , i ∈ I ⋁ D i , n ) = p ( i ∈ I ⋂ D i , 1 , i ∈ I ⋂ D i , 2 , ⋯ , i ∈ I ⋂ D i , n ) = ( a m ≤ F a 1 ⋃ i ∈ I ⋂ D i , m , a m ≤ F a 2 ⋃ i ∈ I ⋂ D i , m , ⋯ , a m ≤ F a n ⋃ i ∈ I ⋂ D i , m )
and
⋁ i ∈ I p ( D i ) = ⋁ i ∈ I ( p D i ( 1 ) , p D i ( 2 ) , ⋯ , p D i ( n ) ) = ( ⋂ i ∈ I p D i ( 1 ) , ⋂ i ∈ I p D i ( 2 ) , ⋯ , ⋂ i ∈ I p D i ( n ) ) = ( ⋂ i ∈ I ⋃ a m ≤ F a 1 D i , m , ⋂ i ∈ I ⋃ a m ≤ F a 2 D i , m , ⋯ , ⋂ i ∈ I ⋃ a m ≤ F a n D i , m ) . \begin{aligned}
\bigvee_{i\in I}p(D_i)
&=\bigvee_{i\in I}(p_{D_i}(1),p_{D_i}(2),\cdots,p_{D_i}(n))\\
&=\left(\bigcap_{i\in I}p_{D_i}(1),\bigcap_{i\in I}p_{D_i}(2),\cdots,\bigcap_{i\in I}p_{D_i}(n)\right)\\
&=\left(\bigcap_{i\in I}\bigcup_{a_m\le_F a_1}D_{i,m},\bigcap_{i\in I}\bigcup_{a_m\le_F a_2}D_{i,m},\cdots,\bigcap_{i\in I}\bigcup_{a_m\le_F a_n}D_{i,m}\right).
\end{aligned} i ∈ I ⋁ p ( D i ) = i ∈ I ⋁ ( p D i ( 1 ) , p D i ( 2 ) , ⋯ , p D i ( n )) = ( i ∈ I ⋂ p D i ( 1 ) , i ∈ I ⋂ p D i ( 2 ) , ⋯ , i ∈ I ⋂ p D i ( n ) ) = ( i ∈ I ⋂ a m ≤ F a 1 ⋃ D i , m , i ∈ I ⋂ a m ≤ F a 2 ⋃ D i , m , ⋯ , i ∈ I ⋂ a m ≤ F a n ⋃ D i , m ) .
We claim that for each 1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n ,
⋃ a m ≤ F a k ⋂ i ∈ I D i , m = ⋂ i ∈ I ⋃ a m ≤ F a k D i , m . \bigcup_{a_m\le_F a_k}\bigcap_{i\in I}D_{i,m}
=
\bigcap_{i\in I}\bigcup_{a_m\le_F a_k}D_{i,m}. a m ≤ F a k ⋃ i ∈ I ⋂ D i , m = i ∈ I ⋂ a m ≤ F a k ⋃ D i , m .
Choose any r ∈ ⋃ a m ≤ F a k ⋂ i ∈ I D i , m r\in\bigcup_{a_m\le_F a_k}\bigcap_{i\in I}D_{i,m} r ∈ ⋃ a m ≤ F a k ⋂ i ∈ I D i , m . Select a m 0 ∈ { 1 , 2 , ⋯ , n } m_0\in\{1,2,\cdots,n\} m 0 ∈ { 1 , 2 , ⋯ , n } with a m 0 ≤ F a k a_{m_0}\le_F a_k a m 0 ≤ F a k such that
r ∈ ⋂ i ∈ I D i , m 0 . r\in\bigcap_{i\in I}D_{i,m_0}. r ∈ i ∈ I ⋂ D i , m 0 .
As for each i ∈ I i\in I i ∈ I , D i , m 0 D_{i,m_0} D i , m 0 belongs to ⋃ a m ≤ F a k D i , m \bigcup_{a_m\le_F a_k}D_{i,m} ⋃ a m ≤ F a k D i , m , we have r ∈ ⋂ i ∈ I ⋃ a m ≤ F a k D i , m r\in\bigcap_{i\in I}\bigcup_{a_m\le_F a_k}D_{i,m} r ∈ ⋂ i ∈ I ⋃ a m ≤ F a k D i , m . This yields
⋃ a m ≤ F a k ⋂ i ∈ I D i , m ⊆ ⋂ i ∈ I ⋃ a m ≤ F a k D i , m . \bigcup_{a_m\le_F a_k}\bigcap_{i\in I}D_{i,m}
\subseteq
\bigcap_{i\in I}\bigcup_{a_m\le_F a_k}D_{i,m}. a m ≤ F a k ⋃ i ∈ I ⋂ D i , m ⊆ i ∈ I ⋂ a m ≤ F a k ⋃ D i , m .
Conversely, let s ∈ ⋂ i ∈ I ⋃ a m ≤ F a k D i , m s\in\bigcap_{i\in I}\bigcup_{a_m\le_F a_k}D_{i,m} s ∈ ⋂ i ∈ I ⋃ a m ≤ F a k D i , m . Then for each i ∈ I i\in I i ∈ I , we can fix a m i ≤ a k m_i\le a_k m i ≤ a k with s ∈ D i , m i s\in D_{i,m_i} s ∈ D i , m i . Assume that s ∉ ⋃ a m ≤ F a k ⋂ i ∈ I D i , m s\notin\bigcup_{a_m\le_F a_k}\bigcap_{i\in I}D_{i,m} s ∈ / ⋃ a m ≤ F a k ⋂ i ∈ I D i , m . Put
X = { p ∣ 1 ≤ p ≤ n , a p ≤ F a k } . X=\{p\mid1\le p\le n,\ a_p\le_F a_k\}. X = { p ∣ 1 ≤ p ≤ n , a p ≤ F a k } .
It follows that for each p ∈ X p\in X p ∈ X , s ∉ ⋂ i ∈ I D i , p s\notin\bigcap_{i\in I}D_{i,p} s ∈ / ⋂ i ∈ I D i , p . Select a i p ∈ I i_p\in I i p ∈ I with s ∉ D i p , p s\notin D_{i_p,p} s ∈ / D i p , p . As { D i ∣ i ∈ I } \{D_i\mid i\in I\} { D i ∣ i ∈ I } is directed and X X X is finite, there is a j ∈ I j\in I j ∈ I such that D j ⊆ ⋂ p ∈ X D i p D_j\subseteq\bigcap_{p\in X}D_{i_p} D j ⊆ ⋂ p ∈ X D i p . For m j ∈ X m_j\in X m j ∈ X , we have s ∈ D j , m j s\in D_{j,m_j} s ∈ D j , m j and s ∉ D i m j , m j s\notin D_{i_{m_j},m_j} s ∈ / D i m j , m j . It follows from D j ⊆ ⋂ p ∈ X D i p D_j\subseteq\bigcap_{p\in X}D_{i_p} D j ⊆ ⋂ p ∈ X D i p that D j , m j ⊆ ⋂ p ∈ X D i p , m j D_{j,m_j}\subseteq\bigcap_{p\in X}D_{i_p,m_j} D j , m j ⊆ ⋂ p ∈ X D i p , m j . As a result, s ∈ ⋂ p ∈ X D i p , m j s\in\bigcap_{p\in X}D_{i_p,m_j} s ∈ ⋂ p ∈ X D i p , m j , a contradiction. So we know s ∈ ⋃ a m ≤ F a k ⋂ i ∈ I D i , m s\in\bigcup_{a_m\le_F a_k}\bigcap_{i\in I}D_{i,m} s ∈ ⋃ a m ≤ F a k ⋂ i ∈ I D i , m . Thus
⋂ i ∈ I ⋃ a m ≤ F a k D i , m ⊆ ⋃ a m ≤ F a k ⋂ i ∈ I D i , m . \bigcap_{i\in I}\bigcup_{a_m\le_F a_k}D_{i,m}
\subseteq
\bigcup_{a_m\le_F a_k}\bigcap_{i\in I}D_{i,m}. i ∈ I ⋂ a m ≤ F a k ⋃ D i , m ⊆ a m ≤ F a k ⋃ i ∈ I ⋂ D i , m .
Hence p p p is Scott continuous. For each E = ( E 1 , E 2 , … , E n ) ∈ B E=(E_1,E_2,\ldots,E_n)\in B E = ( E 1 , E 2 , … , E n ) ∈ B and each 1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n ,
p p ( E ) ( k ) = ⋃ { p E ( m ) ∣ a m ≤ F a k } = ⋃ a m ≤ F a k ⋃ a q ≤ F a m E q = ⋃ a q ≤ F a k E q = p E ( k ) . p_p(E)(k)=\bigcup\{p_E(m)\mid a_m\le_F a_k\}=\bigcup_{a_m\le_F a_k}\bigcup_{a_q\le_F a_m}E_q=\bigcup_{a_q\le_F a_k}E_q=p_E(k). p p ( E ) ( k ) = ⋃ { p E ( m ) ∣ a m ≤ F a k } = a m ≤ F a k ⋃ a q ≤ F a m ⋃ E q = a q ≤ F a k ⋃ E q = p E ( k ) .
This means that p p p is idempotent.
Set
A = { G ∈ B 0 ∣ if a i ≤ F a j , then G i ⊆ G j } . \mathcal{A}=\{G\in B_0\mid\text{if }a_i\le_F a_j,\text{ then }G_i\subseteq G_j\}. A = { G ∈ B 0 ∣ if a i ≤ F a j , then G i ⊆ G j } .
Let G = ( G 1 , G 2 , … , G n ) ∈ B 0 G=(G_1,G_2,\ldots,G_n)\in B_0 G = ( G 1 , G 2 , … , G n ) ∈ B 0 with p ( G ) = G p(G)=G p ( G ) = G . Suppose a j ≤ F a k a_j\le_F a_k a j ≤ F a k . Then we have
G j ⊆ ⋃ a m ≤ F a k G m = p G ( k ) = G k . G_j\subseteq\bigcup_{a_m\le_F a_k}G_m=p_G(k)=G_k. G j ⊆ a m ≤ F a k ⋃ G m = p G ( k ) = G k .
So G ∈ A G\in\mathcal{A} G ∈ A .
Suppose G ∈ A G\in\mathcal{A} G ∈ A . Then for each 1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n , G m ⊆ G k G_m\subseteq G_k G m ⊆ G k whenever a m ≤ F a k a_m\le_F a_k a m ≤ F a k . This yields that
p G ( k ) = ⋃ a m ≤ F a k G m = G k . p_G(k)=\bigcup_{a_m\le_F a_k}G_m=G_k. p G ( k ) = a m ≤ F a k ⋃ G m = G k .
Consequently, p ( G ) = G p(G)=G p ( G ) = G .
We conclude that A = { G ∈ B 0 ∣ p ( G ) = G } \mathcal{A}=\{G\in B_0\mid p(G)=G\} A = { G ∈ B 0 ∣ p ( G ) = G } . Since p p p is idempotent, we deduce that { p ( G ) ∣ G ∈ B 0 } = A \{p(G)\mid G\in B_0\}=\mathcal{A} { p ( G ) ∣ G ∈ B 0 } = A .
Suppose D \mathcal{D} D is directed in A \mathcal{A} A and set H = ⋁ B D H=\bigvee_B\mathcal{D} H = ⋁ B D . Since ⊤ B \top_B ⊤ B is compact in B B B , H = ⋁ B D = ⋁ B 0 D ∈ B 0 H=\bigvee_B\mathcal{D}=\bigvee_{B_0}\mathcal{D}\in B_0 H = ⋁ B D = ⋁ B 0 D ∈ B 0 . By the Scott continuity of p p p , we obtain
p ( H ) = p ( ⋁ B 0 D ) = ⋁ B 0 { p ( D ) ∣ D ∈ D } = ⋁ B 0 D = H . p(H)=p\left(\bigvee_{B_0}\mathcal{D}\right)=\bigvee_{B_0}\{p(D)\mid D\in\mathcal{D}\}=\bigvee_{B_0}\mathcal{D}=H. p ( H ) = p ( B 0 ⋁ D ) = B 0 ⋁ { p ( D ) ∣ D ∈ D } = B 0 ⋁ D = H .
Hence H ∈ A H\in\mathcal{A} H ∈ A . As a result, A \mathcal{A} A is a subdcpo of B 0 B_0 B 0 .
Let i : A → B 0 i:\mathcal{A}\to B_0 i : A → B 0 be the inclusion mapping. Then i i i is Scott continuous. Set q = p ∣ B 0 q=p|_{B_0} q = p ∣ B 0 as the mapping p ∣ B 0 : B 0 → A p|_{B_0}:B_0\to\mathcal{A} p ∣ B 0 : B 0 → A . Also, q q q is Scott continuous. Moreover, q ∘ i = i d A q\circ i=id_{\mathcal{A}} q ∘ i = i d A .
Since Σ B 0 \Sigma_{B_0} Σ B 0 is well-filtered, by Lemma 4.14, Σ A \Sigma_{\mathcal{A}} Σ A is well-filtered.
Finally, we verify that K ( M × F ) K(M\times F) K ( M × F ) and A \mathcal{A} A are order isomorphic.
For Q ∈ K ( M × F ) Q\in K(M\times F) Q ∈ K ( M × F ) and 1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n , put
Q k = { x ∈ M ∣ ( x , a k ) ∈ Q } . Q_k=\{x\in M\mid(x,a_k)\in Q\}. Q k = { x ∈ M ∣ ( x , a k ) ∈ Q } .
Since Q Q Q is upper, we have Q k = π M ( Q ∩ ( M × ↓ F a k ) ) Q_k=\pi_M(Q\cap(M\times\mathord{\downarrow_F}a_k)) Q k = π M ( Q ∩ ( M × ↓ F a k )) . Clearly, the projections π M \pi_M π M and π F \pi_F π F are Scott continuous. As ↓ F a k \mathord{\downarrow_F}a_k ↓ F a k is closed in Σ F \Sigma_F Σ F , we get
M × ↓ F a k = π F − 1 ( ↓ F a k ) M\times\mathord{\downarrow_F}a_k=\pi_F^{-1}(\mathord{\downarrow_F}a_k) M × ↓ F a k = π F − 1 ( ↓ F a k )
is Scott closed in M × F M\times F M × F . Thus Q ∩ ( M × ↓ F a k ) Q\cap(M\times\mathord{\downarrow_F}a_k) Q ∩ ( M × ↓ F a k ) is compact in Σ ( M × F ) \Sigma(M\times F) Σ ( M × F ) . The Scott continuity of π M \pi_M π M gives Q k Q_k Q k is compact in Σ M \Sigma M Σ M . Moreover, Q k Q_k Q k is upper in M M M . So Q k ∈ T Q_k\in\mathcal{T} Q k ∈ T .
If a j ≤ F a k a_j\le_F a_k a j ≤ F a k , then we have Q j ⊆ Q k Q_j\subseteq Q_k Q j ⊆ Q k as Q Q Q is upper. Since Q Q Q is nonempty, at least one of Q 1 , … , Q n Q_1,\ldots,Q_n Q 1 , … , Q n is nonempty. So we can define a mapping Φ : K ( M × F ) ⟶ A \Phi:K(M\times F)\longrightarrow\mathcal{A} Φ : K ( M × F ) ⟶ A by
∀ Q ∈ K ( M × F ) , Φ ( Q ) = ( Q 1 , … , Q n ) . \forall Q\in K(M\times F),\quad\Phi(Q)=(Q_1,\ldots,Q_n). ∀ Q ∈ K ( M × F ) , Φ ( Q ) = ( Q 1 , … , Q n ) .
For each G = ( G 1 , … , G n ) ∈ A G=(G_1,\ldots,G_n)\in\mathcal{A} G = ( G 1 , … , G n ) ∈ A , put
C G = { ( x , a k ) ∣ 1 ≤ k ≤ n , x ∈ G k } . C_G=\{(x,a_k)\mid1\le k\le n,\ x\in G_k\}. C G = {( x , a k ) ∣ 1 ≤ k ≤ n , x ∈ G k } .
Clearly, C G ⊆ ⋃ k = 1 n ( G k × ↑ F a k ) C_G \subseteq\bigcup_{k=1}^{n}(G_k \times\uparrow_F a_k) C G ⊆ ⋃ k = 1 n ( G k × ↑ F a k ) . Conversely, suppose ( x , a j ) ∈ G k × ↑ F a k (x,a_j) \in G_k \times\uparrow_F a_k ( x , a j ) ∈ G k × ↑ F a k . Then x ∈ G k x \in G_k x ∈ G k and a k ≤ F a j a_k \le_F a_j a k ≤ F a j . G ∈ A G \in\mathcal{A} G ∈ A gives G k ⊆ G j G_k \subseteq G_j G k ⊆ G j . So x ∈ G j x \in G_j x ∈ G j and therefore ( x , a j ) ∈ C G (x,a_j) \in C_G ( x , a j ) ∈ C G . This shows that
C G = ⋃ k = 1 n ( G k × ↑ F a k ) . C_G = \bigcup_{k=1}^{n}(G_k \times\uparrow_F a_k). C G = k = 1 ⋃ n ( G k × ↑ F a k ) .
Since each G k G_k G k is an upper set in M M M , each G k × ↑ F a k G_k \times\uparrow_F a_k G k × ↑ F a k is also upper in M × F M \times F M × F . So C G C_G C G is upper. Besides, C G C_G C G is nonempty because G ∈ B 0 G \in B_0 G ∈ B 0 .
For each 1 ≤ k ≤ n 1 \le k \le n 1 ≤ k ≤ n , define a mapping j k : M ⟶ M × F j_k : M \longrightarrow M \times F j k : M ⟶ M × F by j k ( x ) = ( x , a k ) j_k(x) = (x,a_k) j k ( x ) = ( x , a k ) . Then j k j_k j k preserves the directed superma and therefore is Scott continuous. The compactness of G k G_k G k gives
j k ( G k ) = G k × { a k } j_k(G_k) = G_k \times\{a_k\} j k ( G k ) = G k × { a k }
is compact in Σ ( M × F ) \Sigma(M \times F) Σ ( M × F ) . So its saturation
↑ M × F ( G k × { a k } ) = G k × ↑ F a k , \uparrow_{M \times F}(G_k \times\{a_k\}) = G_k \times\uparrow_F a_k, ↑ M × F ( G k × { a k }) = G k × ↑ F a k ,
is also compact in Σ ( M × F ) \Sigma(M \times F) Σ ( M × F ) . As a consequence, C G C_G C G is a finite union of compact sets and is therefore compact. In other words, C G ∈ K ( M × F ) C_G \in K(M \times F) C G ∈ K ( M × F ) and hence the mapping
Ψ : A ⟶ K ( M × F ) , Ψ ( G ) = C G \Psi: \mathcal{A} \longrightarrow K(M \times F), \qquad\Psi(G) = C_G Ψ : A ⟶ K ( M × F ) , Ψ ( G ) = C G
is well defined.
By construction, for every G ∈ A G \in\mathcal{A} G ∈ A and every 1 ≤ k ≤ n 1 \le k \le n 1 ≤ k ≤ n ,
{ x ∈ M ∣ ( x , a k ) ∈ C G } = G k . \{x \in M \mid(x,a_k) \in C_G\} = G_k. { x ∈ M ∣ ( x , a k ) ∈ C G } = G k .
Hence Φ ( Ψ ( G ) ) = G \Phi(\Psi(G)) = G Φ ( Ψ ( G )) = G . Conversely, for each C ∈ K ( M × F ) C \in K(M \times F) C ∈ K ( M × F ) ,
C Φ ( C ) = { ( x , a k ) ∣ 1 ≤ k ≤ n , x ∈ C k } = C . C_{\Phi(C)} = \{(x,a_k) \mid1 \le k \le n,\ x \in C_k\} = C. C Φ ( C ) = {( x , a k ) ∣ 1 ≤ k ≤ n , x ∈ C k } = C .
Therefore Φ \Phi Φ and Ψ \Psi Ψ are mutually inverse bijections.
Furthermore, for every C , E ∈ K ( M × F ) C,E \in K(M \times F) C , E ∈ K ( M × F ) ,
C ⊇ E ⟺ C k ⊇ E k for every 1 ≤ k ≤ n . C \supseteq E \Longleftrightarrow C_k \supseteq E_k \text{ for every } 1 \le k \le n. C ⊇ E ⟺ C k ⊇ E k for every 1 ≤ k ≤ n .
Consequently, Φ \Phi Φ is an order isomorphism and hence
K ( M × F ) ≅ A . K(M \times F) \cong\mathcal{A}. K ( M × F ) ≅ A .
Therefore, Σ K ( M × F ) \Sigma K(M \times F) Σ K ( M × F ) is well-filtered.
Lemma 4.16. Let A A A , B B B , R R R be dcpos such that Σ B \Sigma B Σ B is well-filtered and K ( A × B ) , K ( A × R ) K(A \times B),K(A \times R) K ( A × B ) , K ( A × R ) are dcpos. Suppose f : B → R f : B \to R f : B → R is Scott continuous. Define a mapping Φ f : K ( A × B ) ⟶ K ( A × R ) \Phi_f : K(A \times B) \longrightarrow K(A \times R) Φ f : K ( A × B ) ⟶ K ( A × R ) by
Φ f ( C ) = ↑ A × R { ( a , f ( b ) ) ∣ ( a , b ) ∈ C } . \Phi_f(C) = \uparrow_{A \times R}\{(a,f(b)) \mid(a,b) \in C\}. Φ f ( C ) = ↑ A × R {( a , f ( b )) ∣ ( a , b ) ∈ C } . Then Φ f \Phi_f Φ f is Scott continuous.
Proof. Define a mapping h = id A × f : A × B ⟶ A × R h = \operatorname{id}_A \times f : A \times B \longrightarrow A \times R h = id A × f : A × B ⟶ A × R by h ( a , b ) = ( a , f ( b ) ) h(a,b) = (a,f(b)) h ( a , b ) = ( a , f ( b )) . Let { ( a i , b i ) : i ∈ I } \{(a_i,b_i): i \in I\} {( a i , b i ) : i ∈ I } be a directed subset of A × B A \times B A × B . Then by the Scott continuity of f f f , we have
h ( ⋁ i ∈ I ( a i , b i ) ) = h ( ⋁ i ∈ I a i , ⋁ i ∈ I b i ) = ( ⋁ i ∈ I a i , f ( ⋁ i ∈ I b i ) ) = ( ⋁ i ∈ I a i , ⋁ i ∈ I f ( b i ) ) = ⋁ i ∈ I h ( a i , b i ) . \begin{aligned}
h\left(\bigvee_{i\in I}(a_i,b_i)\right)=h\left(\bigvee_{i\in I}a_i,\bigvee_{i\in I}b_i\right) \\
&=\left(\bigvee_{i\in I}a_i,f\left(\bigvee_{i\in I}b_i\right)\right) \\
&=\left(\bigvee_{i\in I}a_i,\bigvee_{i\in I}f(b_i)\right) \\
&=\bigvee_{i\in I}h(a_i,b_i).
\end{aligned} h ( i ∈ I ⋁ ( a i , b i ) ) = h ( i ∈ I ⋁ a i , i ∈ I ⋁ b i ) = ( i ∈ I ⋁ a i , f ( i ∈ I ⋁ b i ) ) = ( i ∈ I ⋁ a i , i ∈ I ⋁ f ( b i ) ) = i ∈ I ⋁ h ( a i , b i ) .
Thus h h h is Scott continuous.
Let C ∈ K ( A × B ) C \in K(A \times B) C ∈ K ( A × B ) . Then Φ f ( C ) = ↑ A × R h ( C ) \Phi_f(C)=\mathord{\uparrow}_{A\times R}h(C) Φ f ( C ) = ↑ A × R h ( C ) is a nonempty compact saturated subset of Σ ( A × R ) \Sigma(A \times R) Σ ( A × R ) . Hence Φ f \Phi_f Φ f is well defined.
Moreover, if C , D ∈ K ( A × B ) C,D \in K(A \times B) C , D ∈ K ( A × B ) with C ⊇ D C \supseteq D C ⊇ D , then
Φ f ( C ) = ↑ A × R h ( C ) ⊇ ↑ A × R h ( D ) = Φ f ( D ) . \Phi_f(C)=\mathord{\uparrow}_{A\times R}h(C)\supseteq\mathord{\uparrow}_{A\times R}h(D)=\Phi_f(D). Φ f ( C ) = ↑ A × R h ( C ) ⊇ ↑ A × R h ( D ) = Φ f ( D ) .
Consequently, Φ f \Phi_f Φ f is an order preserving. For C ∈ K ( A × B ) C \in K(A \times B) C ∈ K ( A × B ) and a ∈ A a \in A a ∈ A , set
C a = { b ∈ B ∣ ( a , b ) ∈ C } . C_a=\{b \in B\mid(a,b)\in C\}. C a = { b ∈ B ∣ ( a , b ) ∈ C } .
We claim that C a = π B ( C ∩ ( ↓ A a × B ) ) C_a=\pi_B(C\cap(\mathord{\downarrow}_A a\times B)) C a = π B ( C ∩ ( ↓ A a × B )) .
Clearly C a ⊆ π B ( C ∩ ( ↓ A a × B ) ) C_a\subseteq\pi_B(C\cap(\mathord{\downarrow}_A a\times B)) C a ⊆ π B ( C ∩ ( ↓ A a × B )) . Conversely, if b b b belongs to π B ( C ∩ ( ↓ A a × B ) ) \pi_B(C\cap(\mathord{\downarrow}_A a\times B)) π B ( C ∩ ( ↓ A a × B )) , then there exists a ′ ≤ A a a'\leq_A a a ′ ≤ A a such that ( a ′ , b ) ∈ C (a',b)\in C ( a ′ , b ) ∈ C . As C C C is upper and ( a ′ , b ) ≤ ( a , b ) (a',b)\leq(a,b) ( a ′ , b ) ≤ ( a , b ) , we obtain ( a , b ) ∈ C (a,b)\in C ( a , b ) ∈ C and therefore b ∈ C a b\in C_a b ∈ C a . As a result, π B ( C ∩ ( ↓ A a × B ) ) ⊆ C a \pi_B(C\cap(\mathord{\downarrow}_A a\times B))\subseteq C_a π B ( C ∩ ( ↓ A a × B )) ⊆ C a . This proves that C a = π B ( C ∩ ( ↓ A a × B ) ) C_a=\pi_B(C\cap(\mathord{\downarrow}_A a\times B)) C a = π B ( C ∩ ( ↓ A a × B )) .
Since ↓ A a \mathord{\downarrow}_A a ↓ A a is Scott closed and π A \pi_A π A is Scott continuous, we get
↓ A a × B = π A − 1 ( ↓ A a ) \mathord{\downarrow}_A a\times B=\pi_A^{-1}(\mathord{\downarrow}_A a) ↓ A a × B = π A − 1 ( ↓ A a )
is Scott closed in A × B A\times B A × B . Hence C ∩ ( ↓ A a × B ) C\cap(\mathord{\downarrow}_A a\times B) C ∩ ( ↓ A a × B ) is also compact in Σ ( A × B ) \Sigma(A\times B) Σ ( A × B ) . It follows from the Scott continuity of π B \pi_B π B that C a = π B ( C ∩ ( ↓ A a × B ) ) C_a=\pi_B(C\cap(\mathord{\downarrow}_A a\times B)) C a = π B ( C ∩ ( ↓ A a × B )) is compact in Σ B \Sigma B Σ B . Furthermore, C a C_a C a is an upper set in B B B because C C C is upper in A × B A\times B A × B .
For every a ∈ A a\in A a ∈ A and r ∈ R r\in R r ∈ R , one can check that
( a , r ) ∈ Φ f ( C ) ⟺ C a ∩ f − 1 ( ↓ R r ) ≠ ∅ . (a,r)\in\Phi_f(C)\Longleftrightarrow C_a\cap f^{-1}(\mathord{\downarrow}_R r)\ne\varnothing. ( a , r ) ∈ Φ f ( C ) ⟺ C a ∩ f − 1 ( ↓ R r ) = ∅ .
Indeed, suppose that ( a , r ) ∈ Φ f ( C ) (a,r)\in\Phi_f(C) ( a , r ) ∈ Φ f ( C ) . Then there exists ( a ′ , b ) ∈ C (a',b)\in C ( a ′ , b ) ∈ C such that
a ′ ≤ A a and f ( b ) ≤ R r . a'\leq_A a\quad\text{and}\quad f(b)\leq_R r. a ′ ≤ A a and f ( b ) ≤ R r .
Since C C C is upper, ( a , b ) ∈ C (a,b)\in C ( a , b ) ∈ C and hence b ∈ C a ∩ f − 1 ( ↓ R r ) b\in C_a\cap f^{-1}(\mathord{\downarrow}_R r) b ∈ C a ∩ f − 1 ( ↓ R r ) . Conversely, if b ∈ C a ∩ f − 1 ( ↓ R r ) b\in C_a\cap f^{-1}(\mathord{\downarrow}_R r) b ∈ C a ∩ f − 1 ( ↓ R r ) , then ( a , b ) ∈ C (a,b)\in C ( a , b ) ∈ C and f ( b ) ≤ R r f(b)\leq_R r f ( b ) ≤ R r . Hence
( a , f ( b ) ) ∈ h ( C ) and ( a , f ( b ) ) ≤ ( a , r ) . (a,f(b))\in h(C)\quad\text{and}\quad(a,f(b))\leq(a,r). ( a , f ( b )) ∈ h ( C ) and ( a , f ( b )) ≤ ( a , r ) .
This implies ( a , r ) ∈ Φ f ( C ) (a,r)\in\Phi_f(C) ( a , r ) ∈ Φ f ( C ) .
Finally, we prove that Φ f \Phi_f Φ f preserves directed suprema. Let { C λ ∣ λ ∈ Λ } \{C_\lambda\mid\lambda\in\Lambda\} { C λ ∣ λ ∈ Λ } be a directed subset of K ( A × B ) K(A \times B) K ( A × B ) . As K ( A × B ) K(A \times B) K ( A × B ) is a dcpo, by Lemma 4.6, we have that
C = ⋂ λ ∈ Λ C λ = ⋁ λ ∈ Λ K ( A × B ) C λ . C = \bigcap_{\lambda\in\Lambda} C_\lambda= \bigvee_{\lambda\in\Lambda}^{K(A \times B)} C_\lambda. C = λ ∈ Λ ⋂ C λ = λ ∈ Λ ⋁ K ( A × B ) C λ .
belongs to K ( A × B ) K(A \times B) K ( A × B ) .
Since C ⊆ C λ C \subseteq C_\lambda C ⊆ C λ for every λ ∈ Λ \lambda\in\Lambda λ ∈ Λ , we have Φ f ( C ) ⊆ Φ f ( C λ ) \Phi_f(C) \subseteq\Phi_f(C_\lambda) Φ f ( C ) ⊆ Φ f ( C λ ) and hence Φ f ( C ) ⊆ ⋂ λ ∈ Λ Φ f ( C λ ) \Phi_f(C) \subseteq\bigcap_{\lambda\in\Lambda} \Phi_f(C_\lambda) Φ f ( C ) ⊆ ⋂ λ ∈ Λ Φ f ( C λ ) . Conversely, take any ( a , r ) ∈ ⋂ λ ∈ Λ Φ f ( C λ ) (a,r) \in\bigcap_{\lambda\in\Lambda} \Phi_f(C_\lambda) ( a , r ) ∈ ⋂ λ ∈ Λ Φ f ( C λ ) and let H = f − 1 ( ↓ R r ) H = f^{-1}(\mathord{\downarrow}R r) H = f − 1 ( ↓ R r ) . The Scott continuity of f f f gives that H H H is Scott closed in B B B .
By the characterization established above, for every λ ∈ Λ \lambda\in\Lambda λ ∈ Λ , ( C λ ) a ∩ H ≠ ∅ (C_\lambda)_a \cap H \ne\varnothing ( C λ ) a ∩ H = ∅ . In particular, every ( C λ ) a (C_\lambda)_a ( C λ ) a is nonempty and hence for every λ ∈ Λ \lambda\in\Lambda λ ∈ Λ , ( C λ ) a ∈ K ( B ) (C_\lambda)_a \in K(B) ( C λ ) a ∈ K ( B ) . Moreover, { ( C λ ) a ∣ λ ∈ Λ } \{(C_\lambda)_a \mid\lambda\in\Lambda\} {( C λ ) a ∣ λ ∈ Λ } forms a filtered family. Assume that H ∩ ( ⋂ λ ∈ Λ ( C λ ) a ) = ∅ H \cap\left(\bigcap_{\lambda\in\Lambda}(C_\lambda)_a\right) = \varnothing H ∩ ( ⋂ λ ∈ Λ ( C λ ) a ) = ∅ . Equivalently, ⋂ λ ∈ Λ ( C λ ) a ⊆ B ∖ H \bigcap_{\lambda\in\Lambda}(C_\lambda)_a \subseteq B \setminus H ⋂ λ ∈ Λ ( C λ ) a ⊆ B ∖ H . Since Σ B \Sigma B Σ B is well-filtered, there exists λ 0 ∈ Λ \lambda_0 \in\Lambda λ 0 ∈ Λ satisfying ( C λ 0 ) a ⊆ B ∖ H (C_{\lambda_0})_a \subseteq B \setminus H ( C λ 0 ) a ⊆ B ∖ H . This contradicts ( C λ 0 ) a ∩ H ≠ ∅ (C_{\lambda_0})_a \cap H \ne\varnothing ( C λ 0 ) a ∩ H = ∅ . So we obtain H ∩ ( ⋂ λ ∈ Λ ( C λ ) a ) ≠ ∅ H \cap\left(\bigcap_{\lambda\in\Lambda}(C_\lambda)_a\right) \ne\varnothing H ∩ ( ⋂ λ ∈ Λ ( C λ ) a ) = ∅ .
In addition, we can check that
C a = { b ∈ B : ( a , b ) ∈ C } = { b ∈ B : ( a , b ) ∈ C λ for every λ ∈ Λ } = ⋂ λ ∈ Λ ( C λ ) a . \begin{aligned}
C_a &= \{b \in B : (a,b) \in C\}\\
&= \{b \in B : (a,b) \in C_\lambda\text{ for every } \lambda\in\Lambda\}\\
&= \bigcap_{\lambda\in\Lambda}(C_\lambda)_a.
\end{aligned} C a = { b ∈ B : ( a , b ) ∈ C } = { b ∈ B : ( a , b ) ∈ C λ for every λ ∈ Λ } = λ ∈ Λ ⋂ ( C λ ) a .
This yields that C a ∩ f − 1 ( ↓ R r ) = C a ∩ H ≠ ∅ C_a \cap f^{-1}(\mathord{\downarrow}R r) = C_a \cap H \ne\varnothing C a ∩ f − 1 ( ↓ R r ) = C a ∩ H = ∅ . By the characterization above again, we obtain ( a , r ) ∈ Φ f ( C ) (a,r) \in\Phi_f(C) ( a , r ) ∈ Φ f ( C ) . Consequently, we know
Φ f ( ⋂ λ ∈ Λ C λ ) = ⋂ λ ∈ Λ Φ f ( C λ ) . \Phi_f\left(\bigcap_{\lambda\in\Lambda} C_\lambda\right) = \bigcap_{\lambda\in\Lambda} \Phi_f(C_\lambda). Φ f ( λ ∈ Λ ⋂ C λ ) = λ ∈ Λ ⋂ Φ f ( C λ ) .
Clearly, the subfamily { Φ f ( C λ ) ∣ λ ∈ Λ } \{\Phi_f(C_\lambda) \mid\lambda\in\Lambda\} { Φ f ( C λ ) ∣ λ ∈ Λ } is directed in K ( A × B ) K(A \times B) K ( A × B ) . By Lemma 4.6 and the above equality, we have
Φ f ( ⋁ λ ∈ Λ C λ ) = Φ f ( ⋂ λ ∈ Λ C λ ) = ⋂ λ ∈ Λ Φ f ( C λ ) = ⋁ λ ∈ Λ Φ f ( C λ ) . \Phi_f\left(\bigvee_{\lambda\in\Lambda} C_\lambda\right) = \Phi_f\left(\bigcap_{\lambda\in\Lambda} C_\lambda\right) = \bigcap_{\lambda\in\Lambda} \Phi_f(C_\lambda) = \bigvee_{\lambda\in\Lambda} \Phi_f(C_\lambda). Φ f ( λ ∈ Λ ⋁ C λ ) = Φ f ( λ ∈ Λ ⋂ C λ ) = λ ∈ Λ ⋂ Φ f ( C λ ) = λ ∈ Λ ⋁ Φ f ( C λ ) .
Therefore Φ f \Phi_f Φ f is Scott continuous. □
Definition 4.17. ([11 ] ) An R B RB R B -domain L L L is a dcpo that carries a directed set consisting of Scott continuous mappings f i : L → L f_i : L \to L f i : L → L , i ∈ I i \in I i ∈ I such that each image set f i ( L ) f_i(L) f i ( L ) is finite and ⋁ i ∈ I f i = id L \bigvee_{i \in I} f_i = \operatorname{id}_L ⋁ i ∈ I f i = id L .
Lemma 4.18. ([23 ] ) Let X X X be a well-filtered T 0 T_0 T 0 space. Then Σ K ( X ) \Sigma K(X) Σ K ( X ) is well-filtered.
Theorem 4.20. Let M M M and N N N be dcpos such that Σ K ( M ) \Sigma K(M) Σ K ( M ) is well-filtered and K ( M × N ) K(M \times N) K ( M × N ) is a dcpo. If There exist a directed set I I I and Scott continuous maps d i : N → N d_i : N \to N d i : N → N , i ∈ I i \in I i ∈ I , satisfying
(i) each d i ( N ) d_i(N) d i ( N ) is finite and d i ≤ id d_i \leq\operatorname{id} d i ≤ id ;
(ii) d i ≤ d j d_i \leq d_j d i ≤ d j if i ≤ j i \leq j i ≤ j ;
(iii) ⋁ i ∈ I d i = id \bigvee_{i \in I} d_i = \operatorname{id} ⋁ i ∈ I d i = id .
Then Σ K ( M × N ) \Sigma K(M \times N) Σ K ( M × N ) is well-filtered.
Proof. By Definition 4.17, N N N is an R B RB R B -domain. Hence by Remark 4.19, Σ N \Sigma N Σ N is well-filtered. For each i ∈ I i \in I i ∈ I , set N i = d i ( N ) N_i = d_i(N) N i = d i ( N ) . Let d ˉ i \bar{d}_i d ˉ i be the corestriction of the mapping d i d_i d i and j i : N i → N j_i : N_i \to N j i : N i → N the inclusion mapping. Then for each i i i , both d ˉ i \bar{d}_i d ˉ i and j i j_i j i are Scott continuous with d i = j i ∘ d ˉ i d_i = j_i \circ\bar{d}_i d i = j i ∘ d ˉ i .
Set Q = K ( M × N ) Q = K(M \times N) Q = K ( M × N ) and for each i ∈ I i \in I i ∈ I , Q i = K ( M × N i ) Q_i = K(M \times N_i) Q i = K ( M × N i ) . By Lemma 4.15 , Q i Q_i Q i is a dcpo and Σ Q i \Sigma Q_i Σ Q i is well-filtered. Define two mappings
p i : Q ⟶ Q i , p i ( C ) = ↑ M × N i ( i d M × d ˉ i ) ( C ) , r i : Q i ⟶ Q , r i ( E ) = ↑ M × N ( i d M × j i ) ( E ) . \begin{aligned}
p_i &: Q \longrightarrow Q_i,\qquad p_i(C) = \uparrow_{M \times N_i}(id_M \times\bar{d}_i)(C),\\
r_i &: Q_i \longrightarrow Q,\qquad r_i(E) = \uparrow_{M \times N}(id_M \times j_i)(E).
\end{aligned} p i r i : Q ⟶ Q i , p i ( C ) = ↑ M × N i ( i d M × d ˉ i ) ( C ) , : Q i ⟶ Q , r i ( E ) = ↑ M × N ( i d M × j i ) ( E ) .
By Lemma 4.16 , each p i p_i p i and r i r_i r i are Scott continuous. Let q i = r i ∘ p i q_i = r_i \circ p_i q i = r i ∘ p i . Then every q i q_i q i is also Scott continuous.
Choose any C ∈ Q C \in Q C ∈ Q and ( x , y ) ∈ C (x,y) \in C ( x , y ) ∈ C . It follows from d i ( y ) ≤ N y d_i(y) \leq_N y d i ( y ) ≤ N y that ( x , d i ( y ) ) ≤ ( x , y ) (x,d_i(y)) \leq(x,y) ( x , d i ( y )) ≤ ( x , y ) . So we have C ⊆ q i ( C ) C \subseteq q_i(C) C ⊆ q i ( C ) . If i ≤ j i \leq j i ≤ j , then d i ( y ) ≤ N d j ( y ) d_i(y) \leq_N d_j(y) d i ( y ) ≤ N d j ( y ) for every y ∈ N y \in N y ∈ N . So q j ( C ) ⊆ q i ( C ) q_j(C) \subseteq q_i(C) q j ( C ) ⊆ q i ( C ) and hence
q i ≤ q j ≤ i d Q ( i ≤ j ) . q_i \leq q_j \leq id_Q \qquad(i \leq j). q i ≤ q j ≤ i d Q ( i ≤ j ) .
Since I I I is directed, { q i ∣ i ∈ I } \{q_i \mid i \in I\} { q i ∣ i ∈ I } is directed under the pointwise order.
Fix C ∈ Q C \in Q C ∈ Q . We show that
⋂ i ∈ I q i ( C ) = C . \bigcap_{i \in I} q_i(C) = C. i ∈ I ⋂ q i ( C ) = C .
Clearly, C ⊆ ⋂ i ∈ I q i ( C ) C \subseteq\bigcap_{i \in I} q_i(C) C ⊆ ⋂ i ∈ I q i ( C ) . Suppose W W W is Scott open subset of M × N M \times N M × N containing C C C . For each i ∈ I i \in I i ∈ I , take
W i = ( i d M × d i ) − 1 ( W ) . W_i = (id_M \times d_i)^{-1}(W). W i = ( i d M × d i ) − 1 ( W ) .
Each W i W_i W i is a Scott open subset of M × N M \times N M × N since i d M × d i id_M \times d_i i d M × d i is Scott continuous. ∀ i , j ∈ I \forall i,j \in I ∀ i , j ∈ I with i < j i < j i < j , choose any ( a , b ) ∈ W i (a,b) \in W_i ( a , b ) ∈ W i i.e. ( a , d i ( b ) ) ∈ W (a,d_i(b)) \in W ( a , d i ( b )) ∈ W . As d i ≤ d j d_i \leq d_j d i ≤ d j , ( a , d i ( b ) ) ≤ ( a , d j ( b ) ) (a,d_i(b)) \leq(a,d_j(b)) ( a , d i ( b )) ≤ ( a , d j ( b )) . It follows from W ∈ σ ( M × N ) W \in\sigma(M \times N) W ∈ σ ( M × N ) that ( a , d j ( b ) ) ∈ W (a,d_j(b)) \in W ( a , d j ( b )) ∈ W . So ( a , b ) ∈ W j (a,b) \in W_j ( a , b ) ∈ W j and therefore W i ⊆ W j W_i \subseteq W_j W i ⊆ W j . Take any ( x , y ) ∈ C (x,y) \in C ( x , y ) ∈ C , the condition (iii) gives
( x , y ) = sup i ∈ I ( x , d i ( y ) ) . (x,y) = \sup_{i \in I}(x,d_i(y)). ( x , y ) = i ∈ I sup ( x , d i ( y )) .
Since ( x , y ) ∈ C ⊆ W (x,y) \in C \subseteq W ( x , y ) ∈ C ⊆ W and W ∈ σ ( M × N ) W \in\sigma(M \times N) W ∈ σ ( M × N ) , there is ( x , d i ( y ) ) ∈ W (x,d_i(y)) \in W ( x , d i ( y )) ∈ W , or equivalently ( x , y ) ∈ W i (x,y) \in W_i ( x , y ) ∈ W i . This means that
C ⊆ ⋃ i ∈ I W i . C \subseteq\bigcup_{i \in I} W_i. C ⊆ i ∈ I ⋃ W i .
By the compactness of C C C , there exists i 1 , i 2 , … , i n ∈ I i_1,i_2,\ldots,i_n \in I i 1 , i 2 , … , i n ∈ I such that C ⊆ ⋃ k = 1 n W i k C \subseteq\bigcup_{k=1}^{n} W_{i_k} C ⊆ ⋃ k = 1 n W i k . The directedness of I I I provides i 0 i_0 i 0 being an upper bound of { i 1 , i 2 , … , i n } \{i_1,i_2,\ldots,i_n\} { i 1 , i 2 , … , i n } . As a consequence, C ⊆ W i 0 C \subseteq W_{i_0} C ⊆ W i 0 . In other words, ( i d M × d i 0 ) ( C ) ⊆ W (id_M \times d_{i_0})(C) \subseteq W ( i d M × d i 0 ) ( C ) ⊆ W . As W W W is upper, we have
⋂ i ∈ I q i ( C ) ⊆ q i 0 ( C ) = r i 0 ( p i 0 ( C ) ) = ↑ M × N ( i d M × j i 0 ) ( ↑ M × N i 0 ( i d M × d ˉ i 0 ) ( C ) ) ⊆ W . \bigcap_{i \in I}q_i(C) \subseteq q_{i_0}(C) = r_{i_0}(p_{i_0}(C)) = \uparrow_{M \times N}(id_M \times j_{i_0})\left(\uparrow_{M \times N_{i_0}}(id_M \times\bar{d}_{i_0})(C)\right) \subseteq W. i ∈ I ⋂ q i ( C ) ⊆ q i 0 ( C ) = r i 0 ( p i 0 ( C )) = ↑ M × N ( i d M × j i 0 ) ( ↑ M × N i 0 ( i d M × d ˉ i 0 ) ( C ) ) ⊆ W .
Since C C C is saturated, we conclude that ⋂ i ∈ I q i ( C ) ⊆ C \bigcap_{i \in I} q_i(C) \subseteq C ⋂ i ∈ I q i ( C ) ⊆ C . This shows that ⋂ i ∈ I q i ( C ) = C \bigcap_{i \in I} q_i(C) = C ⋂ i ∈ I q i ( C ) = C . By Lemma 4.6, we have ⋁ i ∈ I q i = i d Q \bigvee_{i \in I} q_i = id_Q ⋁ i ∈ I q i = i d Q . By Lemma 4.14 , Σ Q = Σ K ( M × N ) \Sigma Q = \Sigma K(M \times N) Σ Q = Σ K ( M × N ) is well-filtered.
Remark 4.21. In Theorem 4.20 , the dcpo N N N is actually a R B RB R B -domain. Consequently, if M M M is a dcpo such that Σ K ( M ) \Sigma K(M) Σ K ( M ) is well-filtered and N N N an R B RB R B -domain, then Σ K ( M × N ) \Sigma K(M \times N) Σ K ( M × N ) is well-filtered whenever K ( M × N ) K(M \times N) K ( M × N ) is a dcpo.
An intrinsic characterization of WD sets by open filters
In this section, we will present a characterization of WD sets in a T 0 T_0 T 0 space.
Definition 5.1. ([22 ] ) A subset A A A of a T 0 T_0 T 0 space X X X is called a well-filtered determined set , or simply a WD set if for every well-filtered space Y Y Y and every continuous mapping f : X → Y f:X \to Y f : X → Y , there is a unique y ∈ Y y \in Y y ∈ Y such that cl Y ( f ( A ) ) = cl Y ( { y } ) \operatorname{cl}_Y(f(A)) = \operatorname{cl}_Y(\{y\}) cl Y ( f ( A )) = cl Y ({ y }) .
Definition 5.2. Let X X X be a T 0 T_0 T 0 space. A proper filter F ⊆ O ( X ) \mathcal{F} \subseteq\mathcal{O}(X) F ⊆ O ( X ) is a filtered upper set with X ∈ F X \in\mathcal{F} X ∈ F and ∅ ∉ F \varnothing\notin\mathcal{F} ∅ ∈ / F . We denote by SF ( X ) \operatorname{SF}(X) SF ( X ) the set of all proper Scott open filters on O ( X ) \mathcal{O}(X) O ( X ) .
For A ⊆ X A \subseteq X A ⊆ X , define
p A = { U ∈ O ( X ) ∣ U ∩ A ≠ ∅ } . p_A = \{U \in\mathcal{O}(X) \mid U \cap A \ne\varnothing\}. p A = { U ∈ O ( X ) ∣ U ∩ A = ∅ } .
For x ∈ X x \in X x ∈ X , we write p { x } p_{\{x\}} p { x } as p x p_x p x .
Lemma 5.3. Let X X X be a T 0 T_0 T 0 space and A ⊆ X A \subseteq X A ⊆ X . Then the following statements hold:
p A = p cl X ( A ) p_A = p_{\operatorname{cl}_X(A)} p A = p cl X ( A ) ;
If A A A is nonempty, then p A p_A p A is a nonempty Scott open set in O ( X ) \mathcal{O}(X) O ( X ) .
p A ∈ SF ( X ) p_A \in\operatorname{SF}(X) p A ∈ SF ( X ) if and only if cl X ( A ) \operatorname{cl}_X(A) cl X ( A ) is an irreducible subset of X X X .
Proof. (1) The first assertion holds since an open set U ∈ O ( X ) U \in\mathcal{O}(X) U ∈ O ( X ) meets A A A if and only if U ∩ cl X ( A ) U \cap\operatorname{cl}_X(A) U ∩ cl X ( A ) is nonempty.
(2) Clearly, p A p_A p A is upper and nonempty in O ( X ) \mathcal{O}(X) O ( X ) . Let { U i ∣ i ∈ I } \{U_i \mid i \in I\} { U i ∣ i ∈ I } be a subset of O ( X ) \mathcal{O}(X) O ( X ) with ⋃ i ∈ I U i ∈ p A \bigcup_{i \in I} U_i \in p_A ⋃ i ∈ I U i ∈ p A . Then there is U i U_i U i such that U i ∩ A ≠ ∅ U_i \cap A \ne\varnothing U i ∩ A = ∅ . This proves the Scott openness of p A p_A p A .
(3) Let cl X ( A ) \operatorname{cl}_X(A) cl X ( A ) be an irreducible subset of X X X . Then A A A is also irreducible. Choose any U , V ∈ p A U,V \in p_A U , V ∈ p A . Then U ∩ A ≠ ∅ U \cap A \ne\varnothing U ∩ A = ∅ and V ∩ A ≠ ∅ V \cap A \ne\varnothing V ∩ A = ∅ . The irreducibility of A A A gives U ∩ V ∈ p A U \cap V \in p_A U ∩ V ∈ p A . This yields that p A p_A p A is filtered. As A A A is nonempty, by the second assertion, p A p_A p A is a proper Scott open set. So p A ∈ SF ( X ) p_A \in\operatorname{SF}(X) p A ∈ SF ( X ) .
Conversely, suppose M , N M,N M , N are open in X X X such that M ∩ A ≠ ∅ M \cap A \ne\varnothing M ∩ A = ∅ and N ∩ A ≠ ∅ N \cap A \ne\varnothing N ∩ A = ∅ . In other words, M , N ∈ p A M,N \in p_A M , N ∈ p A . As p A p_A p A is filtered and upper, we deduce that M ∩ N ∈ p A M \cap N \in p_A M ∩ N ∈ p A . So A A A is irreducible in X X X . The irreducibility of cl X ( A ) \operatorname{cl}_X(A) cl X ( A ) is obtained immediately.
Definition 5.4. Let X X X be a T 0 T_0 T 0 space. The collection H X \mathcal{H}_X H X is the smallest subfamily of SF ( X ) \operatorname{SF}(X) SF ( X ) satisfying the following conditions:
p x ∈ H X p_x \in\mathcal{H}_X p x ∈ H X for every x ∈ X x \in X x ∈ X ;
If { F i ∣ i ∈ I } ⊆ H X \{ \mathcal{F}_i \mid i \in I\} \subseteq\mathcal{H}_X { F i ∣ i ∈ I } ⊆ H X is directed under inclusion, then ⋃ i ∈ I F i ∈ H X \bigcup_{i \in I} \mathcal{F}_i \in\mathcal{H}_X ⋃ i ∈ I F i ∈ H X .
If { F i : i ∈ I } ⊆ H X \{\mathcal{F}_i : i \in I\} \subseteq\mathcal{H}_X { F i : i ∈ I } ⊆ H X is nonempty and ⋂ i ∈ I F i ∈ σ ( O ( X ) ) \bigcap_{i \in I} \mathcal{F}_i \in\sigma(\mathcal{O}(X)) ⋂ i ∈ I F i ∈ σ ( O ( X )) , then ⋂ i ∈ I F i ∈ H X \bigcap_{i \in I} \mathcal{F}_i \in\mathcal{H}_X ⋂ i ∈ I F i ∈ H X .
A subset K K K of a T 0 T_0 T 0 space X X X is supercompact if each cover U ⊆ O ( X ) \mathcal{U} \subseteq\mathcal{O}(X) U ⊆ O ( X ) of K K K contains an element U ∈ U U \in\mathcal{U} U ∈ U containing K K K ([7 ] ).
Lemma 5.6. ([7 ] ) A saturated subset K K K of a T 0 T_0 T 0 space X X X is supercompact if and only if K = ↑ x K = \mathord{\uparrow}x K = ↑ x for some x ∈ X x \in X x ∈ X .
Lemma 5.7. Let X X X be a T 0 T_{0} T 0 space and Y Y Y a well-filtered space. Fix a continuous mapping f : X ⟶ Y f:X\longrightarrow Y f : X ⟶ Y and define a subfamily R f R_{f} R f of SF ( X ) \operatorname{SF}(X) SF ( X ) by
F ∈ R f ⟺ ∃ K F ∈ K ( Y ) s.t. ∀ V ∈ O ( Y ) , ( f − 1 ( V ) ∈ F iff K F ⊆ V ) . \mathcal{F}\in R_{f}\Longleftrightarrow\exists K_{\mathcal{F}}\in K(Y)\ \text{s.t.}\ \forall V\in\mathcal{O}(Y),\ (f^{-1}(V)\in\mathcal{F}\ \text{iff}\ K_{\mathcal{F}}\subseteq V). F ∈ R f ⟺ ∃ K F ∈ K ( Y ) s.t. ∀ V ∈ O ( Y ) , ( f − 1 ( V ) ∈ F iff K F ⊆ V ) . Then R f R_{f} R f satisfies the three rules in Definition 5.4.
Proof. For any F ∈ R f \mathcal{F}\in R_{f} F ∈ R f , we have
K F = ⋂ { V ∈ O ( Y ) ∣ K F ⊆ V } = ⋂ { V ∈ O ( Y ) ∣ f − 1 ( V ) ∈ F } . K_{\mathcal{F}}=\bigcap\{V\in\mathcal{O}(Y)\mid K_{\mathcal{F}}\subseteq V\}=\bigcap\{V\in\mathcal{O}(Y)\mid f^{-1}(V)\in\mathcal{F}\}. K F = ⋂ { V ∈ O ( Y ) ∣ K F ⊆ V } = ⋂ { V ∈ O ( Y ) ∣ f − 1 ( V ) ∈ F } .
because K F K_{\mathcal{F}} K F is saturated. Consequently, K F 2 ⊆ K F 1 K_{\mathcal{F}_{2}}\subseteq K_{\mathcal{F}_{1}} K F 2 ⊆ K F 1 if F 1 ⊆ F 2 \mathcal{F}_{1}\subseteq\mathcal{F}_{2} F 1 ⊆ F 2 .
(1) For each x ∈ X x\in X x ∈ X , take K x = ↑ Y f ( x ) K_{x}=\uparrow_{Y}f(x) K x = ↑ Y f ( x ) . Then K x ∈ K ( Y ) K_{x}\in K(Y) K x ∈ K ( Y ) . ∀ U ∈ O ( Y ) \forall U\in\mathcal{O}(Y) ∀ U ∈ O ( Y ) , we have
K x ⊆ U ⟺ f ( x ) ∈ U ⟺ x ∈ f − 1 ( U ) ⟺ f − 1 ( U ) ∈ p x . K_{x}\subseteq U\Longleftrightarrow f(x)\in U\Longleftrightarrow x\in f^{-1}(U)\Longleftrightarrow f^{-1}(U)\in p_{x}. K x ⊆ U ⟺ f ( x ) ∈ U ⟺ x ∈ f − 1 ( U ) ⟺ f − 1 ( U ) ∈ p x .
Thus each p x ∈ R f p_{x}\in R_{f} p x ∈ R f .
(2) Let { F i ∣ i ∈ I } ⊆ R f \{\mathcal{F}_{i}\mid i\in I\}\subseteq R_{f} { F i ∣ i ∈ I } ⊆ R f be a directed subfamily. Then for each i i i , there exists a K F i ∈ K ( Y ) K_{\mathcal{F}_{i}}\in K(Y) K F i ∈ K ( Y ) such that for V ∈ O ( Y ) V\in\mathcal{O}(Y) V ∈ O ( Y ) , K F i ⊆ V K_{\mathcal{F}_{i}}\subseteq V K F i ⊆ V if and only if f − 1 ( V ) ∈ F i f^{-1}(V)\in\mathcal{F}_{i} f − 1 ( V ) ∈ F i . One can check that ⋃ i ∈ I F i \bigcup_{i\in I}\mathcal{F}_{i} ⋃ i ∈ I F i is a proper Scott open filter in O ( X ) \mathcal{O}(X) O ( X ) . Put K = ⋂ i ∈ I K F i K=\bigcap_{i\in I}K_{\mathcal{F}_{i}} K = ⋂ i ∈ I K F i . As { F i ∣ i ∈ I } ⊆ R f \{\mathcal{F}_{i}\mid i\in I\}\subseteq R_{f} { F i ∣ i ∈ I } ⊆ R f is directed, we know that { K F i ∣ i ∈ I } \{K_{\mathcal{F}_{i}}\mid i\in I\} { K F i ∣ i ∈ I } is a filtered family consisting of nonempty compact saturated sets of Y Y Y . By Lemma 4.7, K K K is also a nonempty compact saturated subset of Y Y Y . For each H ∈ O ( Y ) H\in\mathcal{O}(Y) H ∈ O ( Y ) , by the well-filteredness of Y Y Y , we deduce that
K ⊆ H ⟺ ∃ i ∈ I , K F i ⊆ H ⟺ ∃ i ∈ I , f − 1 ( H ) ∈ F i ⟺ f − 1 ( H ) ∈ ⋃ i ∈ I F i . \begin{aligned}
K\subseteq H&\Longleftrightarrow\exists i\in I,\ K_{\mathcal{F}_{i}}\subseteq H\\
&\Longleftrightarrow\exists i\in I,\ f^{-1}(H)\in\mathcal{F}_{i}\\
&\Longleftrightarrow f^{-1}(H)\in\bigcup_{i\in I}\mathcal{F}_{i}.
\end{aligned} K ⊆ H ⟺ ∃ i ∈ I , K F i ⊆ H ⟺ ∃ i ∈ I , f − 1 ( H ) ∈ F i ⟺ f − 1 ( H ) ∈ i ∈ I ⋃ F i .
Therefore, ⋃ i ∈ I F i ∈ R f \bigcup_{i\in I}\mathcal{F}_{i}\in R_{f} ⋃ i ∈ I F i ∈ R f .
(3) Suppose { A i : i ∈ I } ⊆ R f \{\mathcal{A}_{i}:i\in I\}\subseteq R_{f} { A i : i ∈ I } ⊆ R f is nonempty such that A = ⋂ i ∈ I A i \mathcal{A}=\bigcap_{i\in I}\mathcal{A}_{i} A = ⋂ i ∈ I A i is Scott open in O ( X ) \mathcal{O}(X) O ( X ) . Then for every i i i , we can fix a K A i ∈ K ( Y ) K_{\mathcal{A}_{i}}\in K(Y) K A i ∈ K ( Y ) satisfying for every W ∈ O ( Y ) W\in\mathcal{O}(Y) W ∈ O ( Y ) , K A i ⊆ W K_{\mathcal{A}_{i}}\subseteq W K A i ⊆ W if and only if f − 1 ( W ) ∈ A i f^{-1}(W)\in\mathcal{A}_{i} f − 1 ( W ) ∈ A i . Take K = ⋃ i ∈ I K A i K=\bigcup_{i\in I}K_{\mathcal{A}_{i}} K = ⋃ i ∈ I K A i . For each E ∈ O ( Y ) E\in\mathcal{O}(Y) E ∈ O ( Y )
K ⊆ E ⟺ ∀ i ∈ I , K A i ⊆ E ⟺ ∀ i ∈ I , f − 1 ( E ) ∈ A i ⟺ f − 1 ( E ) ∈ A . \begin{aligned}
K\subseteq E&\Longleftrightarrow\forall i\in I,\ K_{\mathcal{A}_{i}}\subseteq E\\
&\Longleftrightarrow\forall i\in I,\ f^{-1}(E)\in\mathcal{A}_{i}\\
&\Longleftrightarrow f^{-1}(E)\in\mathcal{A}.
\end{aligned} K ⊆ E ⟺ ∀ i ∈ I , K A i ⊆ E ⟺ ∀ i ∈ I , f − 1 ( E ) ∈ A i ⟺ f − 1 ( E ) ∈ A .
Next, we prove that K K K is compact in Y Y Y . Let { G j ∣ j ∈ J } ⊆ O ( Y ) \{G_{j}\mid j\in J\}\subseteq\mathcal{O}(Y) { G j ∣ j ∈ J } ⊆ O ( Y ) be an open cover of K K K . Then we have f − 1 ( ⋃ j ∈ J G j ) ∈ A f^{-1}(\bigcup_{j\in J}G_{j})\in\mathcal{A} f − 1 ( ⋃ j ∈ J G j ) ∈ A . The family
B = { f − 1 ( ⋃ j ∈ J ~ G j ) ∣ J ~ ⊆ J is finite } \mathcal{B}=\left\{f^{-1}\left(\bigcup_{j\in\widetilde{J}}G_{j}\right)\mid\widetilde{J}\subseteq J\text{ is finite}\right\} B = ⎩ ⎨ ⎧ f − 1 j ∈ J ⋃ G j ∣ J ⊆ J is finite ⎭ ⎬ ⎫
is directed in O ( X ) \mathcal{O}(X) O ( X ) and further ⋃ B = f − 1 ( ⋃ j ∈ J G j ) \bigcup\mathcal{B}=f^{-1}(\bigcup_{j\in J}G_{j}) ⋃ B = f − 1 ( ⋃ j ∈ J G j ) . The Scott openness of A \mathcal{A} A gives that there exists a finite set J ^ ⊆ J \widehat{J}\subseteq J J ⊆ J such that f − 1 ( ⋃ j ∈ J ^ G j ) ∈ A f^{-1}(\bigcup_{j\in\widehat{J}}G_{j})\in\mathcal{A} f − 1 ( ⋃ j ∈ J G j ) ∈ A . This yields K ⊆ ⋃ j ∈ J ^ G j K\subseteq\bigcup_{j\in\widehat{J}}G_{j} K ⊆ ⋃ j ∈ J G j . This proves the compactness of K K K . Hence we conclude that A ∈ R f \mathcal{A}\in R_{f} A ∈ R f . □ \square □
Let X X X be a T 0 T_0 T 0 space. Define
P = { C ∣ C is a nonempty closed subset of X such that p C ∈ H X } . \mathscr{P}=\{C\mid C\text{ is a nonempty closed subset of }X\text{ such that }p_C\in\mathcal{H}_X\}. P = { C ∣ C is a nonempty closed subset of X such that p C ∈ H X } .
For each U ∈ O ( X ) U\in\mathcal{O}(X) U ∈ O ( X ) , set ◊ U = { C ∈ P ∣ C ∩ U ≠ ∅ } \Diamond U=\{C\in\mathscr{P}\mid C\cap U\ne\varnothing\} ◊ U = { C ∈ P ∣ C ∩ U = ∅ } and τ = { ◊ U ∣ U ∈ O ( X ) } \tau=\{\Diamond U\mid U\in\mathcal{O}(X)\} τ = { ◊ U ∣ U ∈ O ( X )} . By Lemma 5.3 (3), each element of P \mathscr{P} P is irreducible in X X X . This indicates that τ \tau τ forms a topology on P \mathscr{P} P . Moreover, each closed subset of ( P , τ ) (\mathscr{P},\tau) ( P , τ ) has the form □ V = { C ∈ P ∣ C ⊆ V } \Box V=\{C\in\mathscr{P}\mid C\subseteq V\} □ V = { C ∈ P ∣ C ⊆ V } for some closed subset V V V of X X X . Let C , D ∈ P C,D\in\mathscr{P} C , D ∈ P with C ⊈ D C\not\subseteq D C ⊆ D . Then C ∈ ◊ ( X ∖ D ) C\in\Diamond(X\setminus D) C ∈ ◊ ( X ∖ D ) and D ∉ ◊ ( X ∖ D ) D\notin\Diamond(X\setminus D) D ∈ / ◊ ( X ∖ D ) . Hence ( P , τ ) (\mathscr{P},\tau) ( P , τ ) is T 0 T_0 T 0 . Besides, for each A ⊆ P \mathcal{A}\subseteq\mathscr{P} A ⊆ P , we can check that cl τ ( A ) = □ ( cl X ( ⋃ A ) ) \operatorname{cl}_{\tau}(\mathcal{A})=\Box(\operatorname{cl}_{X}(\bigcup\mathcal{A})) cl τ ( A ) = □ ( cl X ( ⋃ A )) .
Lemma 5.8. Let B ⊆ P \mathcal{B}\subseteq\mathscr{P} B ⊆ P be a nonempty compact subset of ( P , τ ) (\mathscr{P},\tau) ( P , τ ) . Then H B = ⋂ B ∈ B p B \mathcal{H}_{\mathcal{B}}=\bigcap_{B\in\mathcal{B}}p_B H B = ⋂ B ∈ B p B belongs to SF ( X ) \operatorname{SF}(X) SF ( X ) .
Proof. Clearly, H B \mathcal{H}_{\mathcal{B}} H B is a proper filter of O ( X ) \mathcal{O}(X) O ( X ) . We assert that H B \mathcal{H}_{\mathcal{B}} H B is Scott open in O ( X ) \mathcal{O}(X) O ( X ) . Suppose { U i ∣ i ∈ I } ⊆ O ( X ) \{U_i\mid i\in I\}\subseteq\mathcal{O}(X) { U i ∣ i ∈ I } ⊆ O ( X ) is directed with ⋃ i ∈ I U i ∈ H B \bigcup_{i\in I}U_i\in\mathcal{H}_{\mathcal{B}} ⋃ i ∈ I U i ∈ H B . Then ( ⋃ i ∈ I U i ) ∩ B (\bigcup_{i\in I}U_i)\cap B ( ⋃ i ∈ I U i ) ∩ B is nonempty for each B ∈ B B\in\mathcal{B} B ∈ B . This means that { ◊ U i ∣ i ∈ I } ⊆ τ \{\Diamond U_i\mid i\in I\}\subseteq\tau { ◊ U i ∣ i ∈ I } ⊆ τ covers B \mathcal{B} B . The compactness of B \mathcal{B} B gives that
B ⊆ ( ◊ U i 1 ) ∪ ( ◊ U i 2 ) ∪ ⋯ ∪ ( ◊ U i n ) . \mathcal{B}\subseteq(\Diamond U_{i_1})\cup(\Diamond U_{i_2})\cup\cdots\cup(\Diamond U_{i_n}). B ⊆ ( ◊ U i 1 ) ∪ ( ◊ U i 2 ) ∪ ⋯ ∪ ( ◊ U i n ) .
As { U i ∣ i ∈ I } \{U_i\mid i\in I\} { U i ∣ i ∈ I } is directed, we can choose a U i ∗ U_{i_*} U i ∗ such that U i 1 ∪ U i 2 ∪ ⋯ ∪ U i n ⊆ U i ∗ U_{i_1}\cup U_{i_2}\cup\cdots\cup U_{i_n}\subseteq U_{i_*} U i 1 ∪ U i 2 ∪ ⋯ ∪ U i n ⊆ U i ∗ . As a result, B ⊆ ◊ U i ∗ \mathcal{B}\subseteq\Diamond U_{i_*} B ⊆ ◊ U i ∗ . This yields that U i ∗ ∈ H B U_{i_*}\in\mathcal{H}_{\mathcal{B}} U i ∗ ∈ H B . The Scott openness of H B \mathcal{H}_{\mathcal{B}} H B is proved. Therefore, H B ∈ SF ( X ) \mathcal{H}_{\mathcal{B}}\in\operatorname{SF}(X) H B ∈ SF ( X ) .
□ \square □
Lemma 5.9. Let X X X be a T 0 T_0 T 0 space. Then ( P , τ ) (\mathscr{P},\tau) ( P , τ ) is well-filtered.
Proof. Suppose { C j ∣ j ∈ J } \{\mathcal{C}_j\mid j\in J\} { C j ∣ j ∈ J } is a filtered family consisting of compact saturated subsets and D \mathcal{D} D is a closed subset in ( P , τ ) (\mathscr{P},\tau) ( P , τ ) such that each C j ∩ D ≠ ∅ \mathcal{C}_j\cap\mathcal{D}\ne\varnothing C j ∩ D = ∅ . By Lemma 4.12, there is a minimal closed subset E \mathcal{E} E contained in D \mathcal{D} D that meets each C j \mathcal{C}_j C j . Set
E = cl X ( ⋃ E ) , E j = C j ∩ E and H j = ⋂ F ∈ E j p F . E=\operatorname{cl}_{X}\left(\bigcup\mathcal{E}\right),\qquad\mathcal{E}_j=\mathcal{C}_j\cap\mathcal{E}\quad\text{and}\quad H_j=\bigcap_{F\in\mathcal{E}_j}p_F. E = cl X ( ⋃ E ) , E j = C j ∩ E and H j = F ∈ E j ⋂ p F .
As E \mathcal{E} E is closed and C j \mathcal{C}_j C j is compact, we know that E j \mathcal{E}_j E j is also compact. By Lemma 5.8, each H j H_j H j is a proper Scott open filter in O ( X ) \mathcal{O}(X) O ( X ) . As H X \mathcal{H}_X H X satisfies the third rule in Definition 5.4 and for each F ∈ E j F\in\mathcal{E}_j F ∈ E j , p F p_F p F belongs to H X \mathcal{H}_X H X , we know that each H j H_j H j belongs to H X \mathcal{H}_X H X . As { C j ∣ j ∈ J } \{\mathcal{C}_j\mid j\in J\} { C j ∣ j ∈ J } is filtered, we deduce that { H j ∣ j ∈ J } \{H_j\mid j\in J\} { H j ∣ j ∈ J } is directed.
Choose any C ∈ ⋃ j ∈ J H j C\in\bigcup_{j\in J}H_j C ∈ ⋃ j ∈ J H j . Then we can fix a j j j such that C C C meets each element D ∈ E j D\in\mathcal{E}_j D ∈ E j . Select a D j ∈ E j D_j\in\mathcal{E}_j D j ∈ E j . Then D j ⊆ ⋃ E ⊆ E D_j\subseteq\bigcup\mathcal{E}\subseteq E D j ⊆ ⋃ E ⊆ E and hence
∅ ≠ C ∩ D j ⊆ C ∩ E . \varnothing\ne C\cap D_j\subseteq C\cap E. ∅ = C ∩ D j ⊆ C ∩ E .
This leads to C ∈ p E C\in p_E C ∈ p E . So we have ⋃ j ∈ J H j ⊆ p E \bigcup_{j\in J}H_j\subseteq p_E ⋃ j ∈ J H j ⊆ p E .
Suppose G ∈ p E G\in p_E G ∈ p E or equivalently, G ∈ O ( X ) G\in\mathcal{O}(X) G ∈ O ( X ) and G ∩ E ≠ ∅ G\cap E\ne\varnothing G ∩ E = ∅ . This implies G ∩ ( ⋃ E ) ≠ ∅ G\cap(\bigcup\mathcal{E})\ne\varnothing G ∩ ( ⋃ E ) = ∅ . Take an E 0 ∈ E E_0\in\mathcal{E} E 0 ∈ E with G ∩ E 0 ≠ ∅ G\cap E_0\ne\varnothing G ∩ E 0 = ∅ . In other words, E 0 ∈ E ∩ ◊ G E_0\in\mathcal{E}\cap\Diamond G E 0 ∈ E ∩ ◊ G and E ∖ ( ◊ G ) \mathcal{E}\setminus(\Diamond G) E ∖ ( ◊ G ) is a proper closed subset of E \mathcal{E} E . By the minimality of E \mathcal{E} E , there exists a j 1 j_1 j 1 such that C j 1 ∩ ( E ∖ ( ◊ G ) ) = ∅ \mathcal{C}_{j_1}\cap(\mathcal{E}\setminus(\Diamond G))=\varnothing C j 1 ∩ ( E ∖ ( ◊ G )) = ∅ . As a consequence, E j 1 = C j 1 ∩ E ⊆ ◊ G \mathcal{E}_{j_1}=\mathcal{C}_{j_1}\cap\mathcal{E}\subseteq\Diamond G E j 1 = C j 1 ∩ E ⊆ ◊ G and therefore G ∈ H j 1 G\in H_{j_1} G ∈ H j 1 . Thus p E p_E p E is contained in ⋃ j ∈ J H j \bigcup_{j\in J}H_j ⋃ j ∈ J H j .
So we get p E = ⋃ j ∈ J H j p_E=\bigcup_{j\in J}H_j p E = ⋃ j ∈ J H j . As H X \mathcal{H}_X H X satisfies the second rule in Definition 5.4, p E ∈ H X p_E\in\mathcal{H}_X p E ∈ H X .
Clearly, E E E is nonempty and closed in X X X . This yields that E ∈ P E\in\mathscr{P} E ∈ P . By Lemma 5.3(3), E E E is also irreducible in X X X . Note that E = cl τ ( E ) = □ E = cl τ ( { E } ) \mathcal{E}=\operatorname{cl}_{\tau}(\mathcal{E})=\Box E=\operatorname{cl}_{\tau}(\{E\}) E = cl τ ( E ) = □ E = cl τ ({ E }) . This means that E ∈ E ⊆ P E\in\mathcal{E}\subseteq\mathscr{P} E ∈ E ⊆ P . As E = cl τ ( { E } ) \mathcal{E}=\operatorname{cl}_{\tau}(\{E\}) E = cl τ ({ E }) meets each C j \mathcal{C}_j C j and C j \mathcal{C}_j C j is saturated, we deduce that E E E belongs to each C j \mathcal{C}_j C j . As a result, E ∈ D ∩ ⋂ j ∈ J C j E\in\mathcal{D}\cap\bigcap_{j\in J}\mathcal{C}_j E ∈ D ∩ ⋂ j ∈ J C j . Therefore, ( P , τ ) (\mathscr{P},\tau) ( P , τ ) is well-filtered.
□ \square □
Theorem 5.10. Let X X X be a T 0 T_0 T 0 space and A ⊆ X A \subseteq X A ⊆ X a closed subset. Then A A A is a WD set of X X X if and only if p A ∈ H X p_A \in\mathcal{H}_X p A ∈ H X .
Proof. Let p A ∈ H X p_A \in\mathcal{H}_X p A ∈ H X . Then by Lemma 5.3, A A A is an irreducible closed subset of X X X . Suppose Y Y Y is a well-filtered space and f : X ⟶ Y f:X \longrightarrow Y f : X ⟶ Y is continuous. By Lemma 5.7 and the minimality of H X \mathcal{H}_X H X , H X ⊆ R f \mathcal{H}_X \subseteq R_f H X ⊆ R f . So p A ∈ R f p_A \in R_f p A ∈ R f and hence there is a K A ∈ K ( Y ) K_A \in\mathcal{K}(Y) K A ∈ K ( Y ) such that for each V ∈ O ( Y ) V \in\mathcal{O}(Y) V ∈ O ( Y ) ,
K A ⊆ V ⟺ f − 1 ( V ) ∈ p A ⟺ f ( A ) ∩ V ≠ ∅ . \begin{aligned}
K_A \subseteq V &\Longleftrightarrow f^{-1}(V) \in p_A \\
&\Longleftrightarrow f(A) \cap V \ne\varnothing.
\end{aligned} K A ⊆ V ⟺ f − 1 ( V ) ∈ p A ⟺ f ( A ) ∩ V = ∅ . Suppose { V j ∣ j ∈ J } ⊆ O ( Y ) \{V_j \mid j \in J\} \subseteq\mathcal{O}(Y) { V j ∣ j ∈ J } ⊆ O ( Y ) covers K A K_A K A . Then we have f ( A ) ∩ ( ⋃ j ∈ J V j ) ≠ ∅ f(A) \cap\left(\bigcup_{j \in J} V_j\right) \ne\varnothing f ( A ) ∩ ( ⋃ j ∈ J V j ) = ∅ . So we can fix a j 0 ∈ J j_0 \in J j 0 ∈ J with f ( A ) ∩ V j 0 ≠ ∅ f(A) \cap V_{j_0} \ne\varnothing f ( A ) ∩ V j 0 = ∅ . As a consequence, K A ⊆ V j 0 K_A \subseteq V_{j_0} K A ⊆ V j 0 . Hence K A K_A K A is supercompact in Y Y Y . By Lemma 5.6, K A = ↑ Y y K_A = \mathord{\uparrow}_Y y K A = ↑ Y y for some y ∈ Y y \in Y y ∈ Y . Now, we have that for each V ~ ∈ O ( Y ) \widetilde{V} \in\mathcal{O}(Y) V ∈ O ( Y ) ,
y ∈ V ~ ⟺ K A ⊆ V ~ ⟺ f ( A ) ∩ V ~ ≠ ∅ . y \in\widetilde{V} \Longleftrightarrow K_A \subseteq\widetilde{V} \Longleftrightarrow f(A) \cap\widetilde{V} \ne\varnothing. y ∈ V ⟺ K A ⊆ V ⟺ f ( A ) ∩ V = ∅ . So we have c l Y f ( A ) = c l Y { y } cl_Y f(A) = cl_Y\{y\} c l Y f ( A ) = c l Y { y } . Therefore, A A A is a WD set.
Conversely, let A A A be a WD set of X X X . By Lemma 5.9, ( P , τ ) (\mathcal{P},\tau) ( P , τ ) is well-filtered. Define a mapping η : X ⟶ P \eta:X \longrightarrow\mathcal{P} η : X ⟶ P by η ( x ) = c l X ( { x } ) \eta(x)=cl_X(\{x\}) η ( x ) = c l X ({ x }) . Then for each U ∈ O ( X ) U \in\mathcal{O}(X) U ∈ O ( X ) , we have η − 1 ( ◊ U ) = { x ∈ X ∣ c l X ( { x } ) ∩ U ≠ ∅ } = U \eta^{-1}(\Diamond U)=\{x \in X \mid cl_X(\{x\}) \cap U \ne\varnothing\}=U η − 1 ( ◊ U ) = { x ∈ X ∣ c l X ({ x }) ∩ U = ∅ } = U . So η \eta η is continuous. As A A A is a WD set, there exists C 0 ∈ P C_0 \in\mathcal{P} C 0 ∈ P such that
c l τ ( η ( A ) ) = c l τ ( { C 0 } ) . cl_\tau(\eta(A))=cl_\tau(\{C_0\}). c l τ ( η ( A )) = c l τ ({ C 0 }) . Then we obtain
A = c l X ( A ) = c l X ( ⋃ η ( A ) ) = ⋃ c l τ ( η ( A ) ) = ⋃ c l τ ( { C 0 } ) = ⋃ □ C 0 = C 0 . A=cl_X(A)=cl_X\left(\bigcup\eta(A)\right)=\bigcup cl_\tau(\eta(A))=\bigcup cl_\tau(\{C_0\})=\bigcup\Box C_0=C_0. A = c l X ( A ) = c l X ( ⋃ η ( A ) ) = ⋃ c l τ ( η ( A )) = ⋃ c l τ ({ C 0 }) = ⋃ □ C 0 = C 0 . As C 0 ∈ P C_0 \in\mathcal{P} C 0 ∈ P , we know p A = p C 0 ∈ H X p_A=p_{C_0}\in\mathcal{H}_X p A = p C 0 ∈ H X . The proof is complete.
Further Work
In the following, we list some problems that are no answers.
Problem 6.1. Let P P P and Q Q Q be dcpos such that
Σ K ( P ) \Sigma K(P) Σ K ( P ) and Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are well-filtered;
K ( P × Q ) K(P \times Q) K ( P × Q ) is a dcpo.
Is Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) well-filtered? In particular, the other conditions keep unchanged and Q Q Q is a continuous dcpo. Is Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) well-filtered?
Problem 6.2. Let P P P and Q Q Q be dcpos such that
Σ K ( P ) \Sigma K(P) Σ K ( P ) and Σ K ( Q ) \Sigma K(Q) Σ K ( Q ) are sober;
K ( P × Q ) K(P \times Q) K ( P × Q ) is a dcpo.
Is Σ K ( P × Q ) \Sigma K(P \times Q) Σ K ( P × Q ) sober?