PF-well-filtered spaces

We show that in the Smyth power space Q(X), for open sets \mathcal{U} and \mathcal{V},\mathcal{U}\ll\mathcal{V} if and only if \mathcal{U}=\emptyset.

Let \mathcal{U} be a nonempty open set. Then there exists an open set W\neq\emptyset such that \emptyset\neq\Box W\subseteq\mathcal{U}. Consider \Phi_W=\{\Box O\mid\ O\in\mathcal{O}(X), W\nsubseteq O\}.

Claim 1: \Phi_W is an open cover of Q(X).

Assume that K\in Q(X), then W\nsubseteq K because K has an empty interior. Take w\in W\setminus K. Since K is saturated, we can choose an open set O such that K\subseteq O and w\notin O. Thus K\in\Box O\in\Phi_W.

Claim 2: There is no finite subfamily of \Phi_W that can cover \Box W.

Given \{\Box O_1,...,\Box O_m\} is a finite subfamily of \Phi_W. Then for each j\in\{1,...,m\}, take w_j\in W\setminus O_j. Let L=\uparrow_X\{w_1,...,w_m\}\in Q(X). Then L\subseteq W because W is an open set. As a result, L\in\Box W. But L\notin\bigcup_{j=1}^m\Box O_j due to the fact that w_j\in L\setminus O_j.

To conclude, \mathcal{U}\nll Q(X) and then \mathcal{U}\ll\mathcal{V} if and only if \mathcal{U}=\emptyset, which is sufficient to show that Q(X) is open well-filtered.

By Lemma 3.1, Q(E) is open well-filtered. Now we prove E is not open well-filtered.

For n\ge 3, let N_n=(1-\frac{1}{n}, 1] and b_n=\frac{1}{2}+\frac{1}{n}. Define H_n=N_n\times[0,b_n), U_n=E\cap H_n and G=E\cap(I\times[0,\frac{1}{2})). Clearly, U_n and G are open sets in E for each n.

In order to prove that \{U_n\}_{n\ge 3} is a \ll-filtered family, we define the map \rho: \mathcal{O}(Y)\to \mathcal{O}(E), which sends each G to the intersection G\cap E.

Claim 1: \rho is an order-isomorphism between two open set lattices.

We only need to show that for G, H\in\mathcal{O}(Y), G\cap E\subseteq H\cap E implies G\subseteq H. Suppose G\nsubseteq H, take (x,t)\in G\setminus H. Consider G_x=\{s\in J\mid (x,s)\in G\} and H_x=\{s\in J\mid (x,s)\in H\}. Note that the map f from J to I\times J which sends each s to (x,s) is continuous and G_x=f^{-1}(G), H_x=f^{-1}(H). Thus they are open sets in J and we can take g,h\in I such that G_x=[0,g) and H_x=[0,h). Since (x,t)\in G\setminus H, t\in G_x but t\notin H_x, that is, h\le t<g. Therefore, h<g. Now consider x: If x\in A, take s\in (h,g)\cap\mathbb{Q}; If x\in I\setminus A, take s\in (h,g)\cap(\mathbb{R}\setminus\mathbb{Q}). No matter what the situation is, we can guarantee that (x,s)\in E. But h<s<g implies s\in G_x\setminus H_x, whence (x,s)\in G\setminus H. To conclude, G\cap E\nsubseteq H\cap E, a contradiction.

Claim 2: H_{n+1}\ll H_n for each n\ge 3.

This is obvious because H_{n+1}\subseteq [1-\frac{1}{n+1}, 1]\times [0,b_{n+1}]\subseteq H_n and [1-\frac{1}{n+1}, 1]\times [0,b_{n+1}] is compact in Y.

Note that U_n=E\cap H_n, then U_{n+1}\ll U_n for each n\ge 3 by Claims 1 and 2, which implies \{U_n\}_{n\ge 3} is a \ll-filtered family.

It is evident that \bigcap_{n\ge 3} U_n=E\cap(\{1\}\times[0,\frac{1}{2}])=\{1\}\times([0,\frac{1}{2})\setminus\mathbb{Q})\subseteq G, but U_n\nsubseteq G for each n because A is dense and the "height" of U_n is strictly bigger than \frac{1}{2}.

For every c\in C, since X is locally compact, there exists H_c\in Q(X) such that c\in int_X H_c\subseteq H_c\subseteq U. Then \{int_X H_c\}_{c\in C} is an open cover of C and C\subseteq\bigcup_{i=1}^n int_X H_{c_i} because C is compact. Let L=\bigcup_{i=1}^n H_{c_i}\in Q(X) and it is easy to see C\subseteq int_X L\subseteq L\subseteq U.

Every PF-sober space is PF-well-filtered.

Conversely, given an irreducible set \mathcal{A}\subseteq Q(X) and define \mathcal{F}_{\mathcal{A}}=\{L\in Q(X) \mid \exists K\in\mathcal{A}, K\subseteq int_X L\}.

Since \mathcal{A}\neq\emptyset, take K\in\mathcal{A}. Then by Lemma 3.3, there exists L\in Q(X) such that K\subseteq int_X L\subseteq\ L\subseteq X. So L\in\mathcal{F}_{\mathcal{A}}, which implies \mathcal{F}_{\mathcal{A}}\neq\emptyset.

Given L_1, L_2\in\mathcal{F}_{\mathcal{A}}, then \mathcal{A}\cap\Box(int_X L_i)\neq\emptyset for i=1, 2. Due to the irreducibility of \mathcal{A}, we have \mathcal{A}\cap\Box(int_X L_1)\cap\Box(int_X L_2)\neq\emptyset. Take K_3 in this intersection and by Lemma 3.3, there exists L_3\in Q(X) such that K_3\subseteq int_X L_3\subseteq L_3\subseteq int_X L_1\cap int_X L_2. Thus L_3\in\mathcal{F}_{\mathcal{A}} with L_3\subseteq L_1\cap L_2. Therefore, \mathcal{F}_{\mathcal{A}} is filtered.

Claim: \bigcap\mathcal{F}_{\mathcal{A}}=\bigcap\mathcal{A}.

Assume z\notin\bigcap\mathcal{A}, then there exists K\in\mathcal{A} such that z\notin K. Because K is saturated, z\notin G for some G\in\mathcal{O}(X) and K\subseteq G. By Lemma 3.3, there exists L\in Q(X) such that K\subseteq int_X L\subseteq L\subseteq G. Hence, L\in\mathcal{F}_{\mathcal{A}} and z\notin L, which implies z\notin\bigcap\mathcal{F}_{\mathcal{A}}. To conclude, we have \bigcap\mathcal{F}_{\mathcal{A}}\subseteq\bigcap\mathcal{A}. The reverse inclusion is obvious.

Therefore, if \bigcap\mathcal{A}=\uparrow x\subseteq U\in\mathcal{O}(X), then \bigcap\mathcal{F}_{\mathcal{A}}=\uparrow x\subseteq U. Since X is PF-well-filtered, there exists L\in\mathcal{F}_{\mathcal{A}} satisfying L\subseteq U. Thus there exists K\in\mathcal{A} such that K\subseteq int_X L\subseteq L\subseteq U.

(\Rightarrow) For any C\in RD(X). Suppose C^{ul}\nsubseteq C, take x\in C^{ul}\setminus C. Let U=X\setminus C, then U\in\mathcal{O}(X) with x\in U and C^u\subseteq\uparrow x. Since C is a Rudin set, take \mathcal{K}, which witnesses this. For each K\in\mathcal{K}, define L_K=\uparrow(C\cap K) and \mathcal{L}=\{L_K\mid K\in\mathcal{K}\}. It is easy to see that \mathcal{L}\subseteq Q(X) is filtered.

Claim: \bigcap_{K\in\mathcal{K}} L_K=C^{u}.

Assume y\in\bigcap_{K\in\mathcal{K}} L_K, then for each K\in\mathcal{K}, y\ge c_K for some c_K\in C\cap K. Consider B=C\cap\downarrow y. Obviously, B is closed and B\subseteq C with c_K\in B\cap K\neq\emptyset. Thus C\subseteq\downarrow y by the minimality of C, which implies y\in C^u. The reverse inclusion is obvious.

Now define \mathcal{M}=\{M_K\mid K\in\mathcal{K}\}, where M_K=L_K\cup\uparrow x. It is evident that \mathcal{M}\subseteq Q(X) is filtered and \bigcap_{K\in\mathcal{K}} M_K=C^u\cup\uparrow x=\uparrow x\subseteq U. So M_{K_0}\subseteq U=X\setminus C for some K_0\in\mathcal{K} by the PF-well-filteredness of X. But \emptyset\neq C\cap K_0\subseteq L_{K_0}\subseteq M_{K_0}, which implies C\cap M_{K_0}\neq\emptyset, a contradiction.

Note that when C^u=\emptyset, we can also show that \bigcap_{K\in\mathcal{K}} L_K=\emptyset and the conclusion still holds.

(\Leftarrow) Let \mathcal{K}\subseteq Q(X) be filtered and \bigcap\mathcal{K}=\uparrow x\subseteq U\in\mathcal{O}(X). Suppose for each K\in\mathcal{K}, K\nsubseteq U. Set F=X\setminus U, it is closed. Thus K\cap F\neq\emptyset implies that there exists set C satisfying the condition of Lemma 3.5. Consequently, C\in RD(X) and C^u\subseteq\bigcap\mathcal{K}=\uparrow x due to the fact that C\cap K\neq\emptyset. As a result, x\in C^{ul}=C, which implies that x\in C\subseteq F=X\setminus U, a contradiction.

It is clear by Propositions 3.6 and 3.7.

For any A\in RD(Y), we have \overline{\xi(A)}\in RD([X,Y]) by Lemma 4.2. Since [X,Y] is a PF-well-filtered space, \overline{\xi(A)}=\overline{\xi(A)}^{ul}.

Claim: For any y\in A^{ul}, \xi_y\in\overline{\xi(A)}^{ul}.

Assume that g\in\overline{\xi(A)}^u, then \xi(A)\subseteq\overline{\xi(A)}, which implies \xi(A)\subseteq\downarrow g. For each x\in X, given a\in A, then \xi_a\le g. It follows that a\le g(x). Thus g(x)\in A^u. So y\le g(x). Therefore, \xi_y\le g. We conclude that \xi_y\in\overline{\xi(A)}^{ul}.

Hence, \xi_y\in\overline{\xi(A)}. Given x_0\in X, note that \xi(A)\subseteq F_{x_0}^{-1}(A) and F_{x_0} is continuous, we have \overline{\xi(A)}\subseteq F_{x_0}^{-1}(A); as a result, y\in A, which implies A^{ul}=A. Consequently, Y is a PF-well-filtered space by Proposition 3.6.

It is easy to see that PFSOB contains a non-T_1 space. Then the reflective hull of PFSOB exists by Lemma 5.2. Let I=[0,1] endowed with the Scott topology under the usual order. It is clear that I is sober; consequently, it is PF-sober. For any T_0 space X, define Y=I^{\mathcal{O}(X)}. Then Y is sober and e: X\to Y sends each x to e(x), where

for every U\in\mathcal{O}(X) is a topological embedding since e(U)=e(X)\cap\{z\in Y\mid z_U>\frac{1}{2}\}.

Claim: For each p\in Y, A_p=Y\setminus \{p\} endowed with the relative topology from Y is PF-sober.

Take C\in Irr_c(A_p), we need to prove that C_{A_p}^{ul}=C. Since the inclusion map from A_p to Y is continuous and Y is sober, \overline{C}^Y=\downarrow_Y q for some q\in Y. Note that C is closed in A_p, thus C=A_p\cap\downarrow_Y q. Consider the following three situations:

(1) If q\neq p, then q\in A_p and C=A_p\cap\downarrow_Y q=\downarrow_{A_p}q is a cut in A_p.

(2) If q=p=\top, then C=A_p\cap\downarrow_Y q=A_p is a cut in A_p.

(3) If q=p\neq\top, take U\in\mathcal{O}(X) such that p_U<1. Define

for each n\ge 1. Since q_n>p, we have q_n\in A_p and \{q_n\mid n\ge 1\}\subseteq C_{A_p}^u. It is clear that \bigcap_{n\ge 1}\downarrow_Y q_n=\downarrow_Y p. Consequently, C_{A_p}^{ul}\subseteq\bigcap_{n\ge 1}\downarrow_Y q_n=\downarrow_Y p. So C_{A_p}^{ul}\subseteq\downarrow_Y p\cap A_p=\downarrow_Y q\cap A_p=C, which implies that C is a cut in A_p.

Suppose \mathcal{K}\subseteq TOP_0 is a reflective category with PFSOB\subseteq\mathcal{K}. Given a T_0 space X and let \eta: X\to KX be the K-reflection. Since Y\in PFSOB, there exists a unique continuous map f: KX\to Y such that f\circ\eta=e. For each p\in Y\setminus e(X), the map e_p: X\to A_p is a topological embedding, and A_p\in\mathcal{K} by the Claim above. Thus there exists a unique continuous map f_p: KX\to A_p such that f_p\circ\eta=e_p. Let i_p be the inclusion from A_p to Y. Then i_p\circ f_p=f by the K-reflection \eta and the object Y. Therefore, f(KX)\subseteq f_p(KX)\subseteq A_p for each p\in Y\setminus e(X), which implies that f(KX)\subseteq\bigcap_{p\in Y\setminus e(X)} Y\setminus\{p\}=e(X). Let e^{-1}: e(X)\to X denote the inverse homeomorphism. Then r=e^{-1}\circ f: KX\to X is continuous. Thus r\circ\eta=e^{-1}\circ f\circ\eta=e^{-1}\circ e=id_X. Note that (\eta\circ r)\circ\eta=\eta\circ (r\circ\eta)=\eta. So \eta\circ r=id_{KX} by the K-reflection \eta and the object KX. We conclude that X is homeomorphic to KX with the continuous map r and \eta. Therefore, X\in\mathcal{K}, which implies \mathcal{K}=TOP_0.

This is an immediate consequence of Theorem 5.3.