betaFeedback

WorksMathematics

A co-sober Alexandrov space whose Scott power space is not co-sober

Abstract

We show that the Scott power construction does not preserve co-sobriety. For a dcpo PP, let Fin⁡(P)\operatorname{Fin}(P) be the poset of nonempty finitely generated upper sets ordered by reverse inclusion. We prove that co-sobriety of ΣFin⁡(P)Σ\operatorname{Fin}(P) implies co-sobriety of ΣPΣP. The proof adapts the point-saturation and union argument used for full compact saturated power posets; the finite step follows from a result of Heckmann and Keimel. Applying this transfer to a dcpo with a non-co-sober Scott space yields a co-sober Alexandrov space whose Scott power space is not co-sober. We also construct a countable example for which the Smyth power poset is countable. In the countable example, the same poset is co-sober under the upper Vietoris topology but not under the Scott topology.

Introduction

In [1 Definition 8.9 and Theorem 8.10], in order to provide an alternative approach to the theory of compactly generated spaces and establish a dual form of the Hofmann–Mislove Theorem, Escardó, Lawson and Simpson introduced the notion of co-sober spaces. A nonempty compact saturated subset is called kk-irreducible if it is not the union of two proper compact saturated subsets. A T0T_0-space is co-sober if every kk-irreducible compact saturated subset is the saturation of a point. Despite the formal duality with sobriety, the two notions behave differently; in particular, a sober space need not be co-sober [6 Example 2.9]; see also [7 Example 2.8].

In an invited academic presentation1, Xu asked whether the Scott power space of a co-sober space is co-sober. In this paper, we give a negative answer to this question.

For a T0T_0 space XX, let Q(X)\mathcal{Q}(X) be the set of its nonempty compact saturated subsets, ordered by reverse inclusion. The Scott power space of XX is

Pσ(X)=ΣQ(X),P_{\sigma}(X)=\Sigma\mathcal{Q}(X),

where Q(X)\mathcal{Q}(X) is endowed with its Scott topology. We study whether this construction preserves co-sobriety. The Scott topology on Q(X)\mathcal{Q}(X) can differ from the upper Vietoris topology, so the two power constructions must be treated separately.

Our first result shows that the Scott power construction does not preserve co-sobriety.

Theorem 1.1. There exists a T0T_0 Alexandrov co-sober space YY such that Pσ(Y)\mathcal{P}_{\sigma}(Y) is not co-sober.

The construction can be made countable.

Theorem 1.2. There exists a space YY as in Theorem 1.1 such that both YY and Q(Y)\mathcal{Q}(Y) are countable. For this example, the space Q(Y)\mathcal{Q}(Y) endowed with the upper Vietoris topology is co-sober.

The proof of Theorem 1.1 uses a dcpo PP whose Scott space is sober but not co-sober, constructed by Shen, Wu, Xi and Zhao [6 Example 2.9]. We show that non-co-sobriety of ΣP\Sigma P passes to ΣFin⁡(P)\Sigma\operatorname{Fin}(P), where

Fin⁡(P)={↑F:∅≠F⊆P is finite}\operatorname{Fin}(P)=\{\mathord{\uparrow}F:\varnothing\ne F\subseteq P\text{ is finite}\}

is ordered by reverse inclusion. If Y=A(P)Y=\mathcal{A}(P) is the Alexandrov space of the same poset, then YY is co-sober and Q(Y)=Fin⁡(P)\mathcal{Q}(Y)=\operatorname{Fin}(P).

Zhao, He and Wang proved related reflection results for full compact saturated power posets [10 Theorem 3.13 and Corollary 3.16]. Their argument uses point-saturation and union maps. We restrict these maps to the finite poset Fin⁡(P)\operatorname{Fin}(P). The additional input is

U∈σ(P)⟹□U={H∈Fin⁡(P):H⊆U}∈σ(Fin⁡(P)),U\in\sigma(P)\Longrightarrow\mathord{\Box}U=\{H\in\operatorname{Fin}(P):H\subseteq U\}\in\sigma(\operatorname{Fin}(P)),

which follows from Heckmann and Keimel’s finite Rudin result [4 Corollary 3.9].

For the countable version, we start with a countable non-sober complete lattice, whose existence was proved by Miao, Xi, Li and Zhao [5 Corollary 3.6]. An argument using the lower topology supplies a nonprincipal co-sobriety witness. We then realize the lower space as a saturated subspace of the Scott space of a countable dcpo and apply the finite transfer.

Scott power spaces have also been studied in connection with sobriety and well-filteredness [8]. In a 2026 preprint, Xu, Xu, Ji and Qiu constructed a co-sober space whose upper Vietoris Smyth power space is not co-sober [9]. Their result concerns a different topology.

Preliminaries

Let PP be a poset. For A⊆PA\subseteq P, write

↑A={y∈P:(∃x∈A) x≤y},↓A={y∈P:(∃x∈A) y≤x}.\mathord{\uparrow}A=\{y\in P:(\exists x\in A)\ x\le y\},\qquad\mathord{\downarrow}A=\{y\in P:(\exists x\in A)\ y\le x\}.

We write ↑x\mathord{\uparrow}x and ↓x\mathord{\downarrow}x for the sets generated by singletons.

A subset U⊆PU\subseteq P is Scott open if it is an upper set and, whenever a directed set D⊆PD\subseteq P has a supremum with ⋁D∈U\bigvee D\in U, one has D∩U≠∅D\cap U\ne\varnothing. This definition applies to arbitrary posets; no dcpo assumption is imposed. The Scott topology is denoted by σ(P)\sigma(P) and the resulting space by ΣP\Sigma P. Its specialization order is the original order on PP [2, 3].

For a topological space XX, a subset is saturated if it is an upper set in the specialization preorder. Throughout the paper, members of Q(X)\mathcal{Q}(X) are nonempty. A set K∈Q(X)K\in\mathcal{Q}(X) is kk-irreducible if

K=K1∪K2,K1,K2∈Q(X),(1)K=K_1\cup K_2,\quad K_1,K_2\in\mathcal{Q}(X), \tag*{(1)}

@page 2/10 starts_mid=1 ends_mid=1 where Q(X)\mathcal{Q}(X) is endowed with its Scott topology. We study whether this construction preserves co-sobriety. The Scott topology on Q(X)\mathcal{Q}(X) can differ from the upper Vietoris topology, so the two power constructions must be treated separately.

Our first result shows that the Scott power construction does not preserve co-sobriety.

Theorem 1.1. There exists a T0T_0 Alexandrov co-sober space YY such that Pσ(Y)\mathcal{P}_{\sigma}(Y) is not co-sober.

The construction can be made countable.

Theorem 1.2. There exists a space YY as in Theorem 1.1 such that both YY and Q(Y)\mathcal{Q}(Y) are countable. For this example, the space Q(Y)\mathcal{Q}(Y) endowed with the upper Vietoris topology is co-sober.

The proof of Theorem 1.1 uses a dcpo PP whose Scott space is sober but not co-sober, constructed by Shen, Wu, Xi and Zhao [6 Example 2.9]. We show that non-co-sobriety of ΣP\Sigma P passes to ΣFin⁡(P)\Sigma\operatorname{Fin}(P), where

Fin⁡(P)={↑F:∅≠F⊆P is finite}\operatorname{Fin}(P)=\{\mathord{\uparrow}F:\varnothing\ne F\subseteq P\text{ is finite}\}

is ordered by reverse inclusion. If Y=A(P)Y=\mathcal{A}(P) is the Alexandrov space of the same poset, then YY is co-sober and Q(Y)=Fin⁡(P)\mathcal{Q}(Y)=\operatorname{Fin}(P).

Zhao, He and Wang proved related reflection results for full compact saturated power posets [10 Theorem 3.13 and Corollary 3.16]. Their argument uses point-saturation and union maps. We restrict these maps to the finite poset Fin⁡(P)\operatorname{Fin}(P). The additional input is

U∈σ(P)⟹□U={H∈Fin⁡(P):H⊆U}∈σ(Fin⁡(P)),U\in\sigma(P)\Longrightarrow\mathord{\Box}U=\{H\in\operatorname{Fin}(P):H\subseteq U\}\in\sigma(\operatorname{Fin}(P)),

which follows from Heckmann and Keimel’s finite Rudin result [4 Corollary 3.9].

For the countable version, we start with a countable non-sober complete lattice, whose existence was proved by Miao, Xi, Li and Zhao [5 Corollary 3.6]. An argument using the lower topology supplies a nonprincipal co-sobriety witness. We then realize the lower space as a saturated subspace of the Scott space of a countable dcpo and apply the finite transfer.

Scott power spaces have also been studied in connection with sobriety and well-filteredness [8]. In a 2026 preprint, Xu, Xu, Ji and Qiu constructed a co-sober space whose upper Vietoris Smyth power space is not co-sober [9]. Their result concerns a different topology.

Preliminaries

Let PP be a poset. For A⊆PA\subseteq P, write

↑A={y∈P:(∃x∈A) x≤y},↓A={y∈P:(∃x∈A) y≤x}.\mathord{\uparrow}A=\{y\in P:(\exists x\in A)\ x\le y\},\qquad\mathord{\downarrow}A=\{y\in P:(\exists x\in A)\ y\le x\}.

We write ↑x\mathord{\uparrow}x and ↓x\mathord{\downarrow}x for the sets generated by singletons.

A subset U⊆PU\subseteq P is Scott open if it is an upper set and, whenever a directed set D⊆PD\subseteq P has a supremum with ⋁D∈U\bigvee D\in U, one has D∩U≠∅D\cap U\ne\varnothing. This definition applies to arbitrary posets; no dcpo assumption is imposed. The Scott topology is denoted by σ(P)\sigma(P) and the resulting space by ΣP\Sigma P. Its specialization order is the original order on PP [2, 3].

For a topological space XX, a subset is saturated if it is an upper set in the specialization preorder. Throughout the paper, members of Q(X)\mathcal{Q}(X) are nonempty. A set K∈Q(X)K\in\mathcal{Q}(X) is kk-irreducible if

K=K1∪K2,K1,K2∈Q(X),(2)K=K_1\cup K_2,\quad K_1,K_2\in\mathcal{Q}(X), \tag*{(2)}

implies K=K1K = K_1 or K=K2K = K_2. A T0T_0 space is co-sober if each kk-irreducible member of Q(X)\mathcal{Q}(X) is the saturation of a point [1 Definition 8.9].

The Smyth order on Q(X)\mathcal{Q}(X) is

K⊑L⟺L⊆K.K \sqsubseteq L \Longleftrightarrow L \subseteq K.

The upper Vietoris topology on Q(X)\mathcal{Q}(X) is generated by the sets

□U={K∈Q(X):K⊆U},U∈O(X).\Box U = \{K \in\mathcal{Q}(X) : K \subseteq U\},\quad U \in\mathcal{O}(X).

We denote the corresponding space by PS(X)P_S(X). For a poset PP, A(P)A(P) denotes the T0T_0 Alexandrov space whose open sets are all upper subsets of PP.

The next two lemmas collect elementary facts needed below.

Lemma 2.1. Let PP be a poset.

(1) Q(A(P))=Fin⁡(P)\mathcal{Q}(A(P)) = \operatorname{Fin}(P) as Smyth ordered sets.

(2) The space A(P)A(P) is co-sober.

(3) The upper Vietoris topology on Fin⁡(P)=Q(A(P))\operatorname{Fin}(P) = \mathcal{Q}(A(P)) is the Alexandrov topology of the Smyth order. Hence PS(A(P))P_S(A(P)) is co-sober.

Proof. The saturated subsets of A(P)A(P) are the upper subsets of PP. If an upper set KK is compact, then {↑x:x∈K}\{\mathord{\uparrow}x : x \in K\} is an open cover of KK, so K=↑FK = \mathord{\uparrow}F for some nonempty finite F⊆KF \subseteq K. Conversely, each principal upper set is compact: any open set containing its least element contains the whole principal upper set. A finite union of compact sets is compact. This proves (1).

If K=↑x1∪⋯∪↑xnK = \mathord{\uparrow}x_1 \cup\cdots\cup\mathord{\uparrow}x_n is kk-irreducible, repeated application of the defining property gives K=↑xiK = \mathord{\uparrow}x_i for some ii. This proves (2).

Every upper Vietoris open is upper in the Smyth order. Conversely, for H∈Fin⁡(P)H \in\operatorname{Fin}(P) the set HH is open in A(P)A(P) and

□H={K∈Fin⁡(P):K⊆H}=↑Fin⁡(P)H.\Box H = \{K \in\operatorname{Fin}(P) : K \subseteq H\} = \mathord{\uparrow}_{\operatorname{Fin}(P)}H.

Thus every principal Alexandrov open is upper Vietoris open, proving (3).

Lemma 2.2. Let XX be a topological space.

(1) The saturation of a compact subset of XX is compact.

(2) If AA is a saturated subspace of a T0T_0 space XX and K⊆AK \subseteq A is compact saturated and kk-irreducible in AA, then KK has the same properties in XX.

Proof. For (1), let an open cover of the saturation be given. A finite subfamily covers the original compact set. The union of this finite subfamily is open and contains the compact set, so it also contains its saturation.

For (2), compactness passes from the subspace to the ambient space. If x∈Kx \in K and x≤Xyx \leq_X y, saturation of AA gives y∈Ay \in A, and saturation of KK in AA gives y∈Ky \in K. Thus KK is saturated in XX. If K=K1∪K2K = K_1 \cup K_2 with Ki∈Q(X)K_i \in\mathcal{Q}(X), then Ki⊆K⊆AK_i \subseteq K \subseteq A. Each KiK_i is compact saturated in the subspace AA, so kk-irreducibility in AA gives K=KiK = K_i for some ii.

A finite transfer argument

Let PP be a dcpo and order Fin⁡(P)\operatorname{Fin}(P) by reverse inclusion. The poset Fin⁡(P)\operatorname{Fin}(P) need not be a dcpo. Accordingly, every Scott topological argument below concerns only directed families whose supremum exists. The next proposition records the consequence of Heckmann and Keimel’s Corollary 3.9 that is needed here.

Proposition 3.1. If UU is Scott open in PP, then □U\Box U is Scott open in Fin⁡(P)\operatorname{Fin}(P).

Proof. The set □U\Box U is upper in the Smyth order. Let H⊆Fin⁡(P)\mathcal{H} \subseteq\operatorname{Fin}(P) be directed, suppose that its supremum H∗H_{\ast} exists in Fin⁡(P)\operatorname{Fin}(P), and assume H∗⊆UH_{\ast} \subseteq U. First observe that

H∗=⋂H∈HH.(3)H_{\ast}=\bigcap_{H\in\mathcal{H}}H. \tag*{(3)}

Indeed, H∗⊆HH_{\ast}\subseteq H for every H∈HH\in\mathcal{H}. Conversely, if xx belongs to the intersection, then ↑x\mathord{\uparrow}x is an upper bound of H\mathcal{H} in the reverse inclusion order. The least upper bound property gives ↑x⊆H∗\mathord{\uparrow}x\subseteq H_{\ast}, hence x∈H∗x\in H_{\ast}.

For each H∈HH\in\mathcal{H}, choose a nonempty finite set FHF_H with H=↑FHH=\mathord{\uparrow}F_H. On nonempty finite subsets of PP, write

F⪯SmG⟺↑G⊆↑F.F\preceq_{\mathrm{Sm}}G\Longleftrightarrow\mathord{\uparrow}G\subseteq\mathord{\uparrow}F.

Directedness of H\mathcal{H} under reverse inclusion implies that {FH:H∈H}\{F_H:H\in\mathcal{H}\} is directed for ⪯Sm\preceq_{\mathrm{Sm}}. By (1),

⋂H∈H↑FH=H∗⊆U.\bigcap_{H\in\mathcal{H}}\mathord{\uparrow}F_H=H_{\ast}\subseteq U.

Heckmann and Keimel’s Corollary 3.9 [4 Corollary 3.9] gives an H∈HH\in\mathcal{H} such that FH⊆UF_H\subseteq U. Since UU is upper, H=↑FH⊆UH=\mathord{\uparrow}F_H\subseteq U. Therefore H∩□U≠∅\mathcal{H}\cap\Box U\ne\varnothing.

Define

η:P⟶Fin⁡(P),η(x)=↑x.\eta:P\longrightarrow\operatorname{Fin}(P),\qquad\eta(x)=\mathord{\uparrow}x.

The map η\eta is monotone. If D⊆PD\subseteq P is directed, then η[D]\eta[D] is directed and its supremum exists in Fin⁡(P)\operatorname{Fin}(P), with

⋁Fin⁡(P)η[D]=⋂d∈D↑d=↑⋁D.\bigvee_{\operatorname{Fin}(P)}\eta[D]=\bigcap_{d\in D}\mathord{\uparrow}d=\mathord{\uparrow}\bigvee D.

Thus η\eta preserves directed suprema, and is Scott continuous as a map ΣP→ΣFin⁡(P)\Sigma P\to\Sigma\operatorname{Fin}(P).

For K∈Q(ΣP)K\in\mathcal{Q}(\Sigma P) and A⊆Fin⁡(P)A\subseteq\operatorname{Fin}(P), put

Λ(K)=↑Fin⁡(P)η[K],μ(A)=⋃A.\Lambda(K)=\mathord{\uparrow}_{\operatorname{Fin}(P)}\eta[K],\qquad\mu(A)=\bigcup A.

These are the finite analogues of the point-saturation and union maps used for full compact saturated power posets; compare [10 Theorem 3.13].

Proposition 3.2. Let PP be a dcpo.

  1. If K∈Q(ΣP)K\in\mathcal{Q}(\Sigma P), then Λ(K)∈Q(ΣFin⁡(P))\Lambda(K)\in\mathcal{Q}(\Sigma\operatorname{Fin}(P)) and μ(Λ(K))=K\mu(\Lambda(K))=K.

  2. If A∈Q(ΣFin⁡(P))A\in\mathcal{Q}(\Sigma\operatorname{Fin}(P)), then μ(A)∈Q(ΣP)\mu(A)\in\mathcal{Q}(\Sigma P).

  3. For K∈Q(ΣP)K\in\mathcal{Q}(\Sigma P), the set KK is kk-irreducible if and only if Λ(K)\Lambda(K) is kk-irreducible.

  4. For K∈Q(ΣP)K\in\mathcal{Q}(\Sigma P), the set KK is principal if and only if Λ(K)\Lambda(K) is principal in the Smyth order.

Proof. By Scott continuity of η\eta, the image η[K]\eta[K] is compact, and Λ(K)\Lambda(K) is its saturation in ΣFin⁡(P)\Sigma\operatorname{Fin}(P). Lemma 2.2(1) proves the first assertion in (1). If H∈Λ(K)H\in\Lambda(K), then H⊆↑x⊆KH\subseteq\mathord{\uparrow}x\subseteq K for some x∈Kx\in K, so μ(Λ(K))⊆K\mu(\Lambda(K))\subseteq K. Conversely, x∈↑x∈Λ(K)x\in\mathord{\uparrow}x\in\Lambda(K) for every x∈Kx\in K. Thus μ(Λ(K))=K\mu(\Lambda(K))=K.

For (2), μ(A)\mu(A) is nonempty and upper in PP. Let {Ui:i∈I}\{U_i:i\in I\} be a Scott open cover of μ(A)\mu(A). Each H=↑F∈Fin⁡(P)H=\mathord{\uparrow}F\in\operatorname{Fin}(P) is a finite union of principal upper sets. Each principal upper set is compact in

ΣP\Sigma P, because every Scott open set containing its least element contains the whole principal upper set. Thus HH is compact. Therefore, for every H∈AH \in\mathcal{A}, there is a finite set IH⊆II_H \subseteq I such that

H⊆VH:=⋃i∈IHUi.H \subseteq V_H := \bigcup_{i\in I_H} U_i.

By Proposition 3.1, □VH\Box V_H is Scott open in Fin⁡(P)\operatorname{Fin}(P). These sets cover A\mathcal{A}. Compactness of A\mathcal{A} gives H1,…,Hm∈AH_1,\ldots,H_m \in\mathcal{A} with

A⊆□VH1∪⋯∪□VHm.\mathcal{A} \subseteq\Box V_{H_1} \cup\cdots\cup\Box V_{H_m}.

It follows that μ(A)\mu(\mathcal{A}) is covered by the finitely many UiU_i whose indices occur in IH1∪⋯∪IHmI_{H_1}\cup\cdots\cup I_{H_m}. Thus μ(A)\mu(\mathcal{A}) is compact saturated.

Suppose that KK is kk-irreducible and

Λ(K)=A∪B,A,B∈Q(ΣFin⁡(P)).\Lambda(K) = \mathcal{A} \cup\mathcal{B}, \quad\mathcal{A},\mathcal{B} \in\mathcal{Q}(\Sigma\operatorname{Fin}(P)).

Parts (1) and (2) give

K=μ(A)∪μ(B).K = \mu(\mathcal{A}) \cup\mu(\mathcal{B}).

Assume K=μ(A)K = \mu(\mathcal{A}). For x∈Kx \in K, choose Hx∈AH_x \in\mathcal{A} with x∈Hxx \in H_x. Since HxH_x is upper, ↑x⊆Hx\uparrow x \subseteq H_x, or Hx⊑↑xH_x \sqsubseteq\uparrow x. Saturation of A\mathcal{A} gives ↑x∈A\uparrow x \in\mathcal{A}. Hence η[K]⊆A\eta[K] \subseteq\mathcal{A}, and another use of saturation gives Λ(K)⊆A\Lambda(K) \subseteq\mathcal{A}. Therefore Λ(K)=A\Lambda(K) = \mathcal{A}.

Conversely, suppose that Λ(K)\Lambda(K) is kk-irreducible and K=K1∪K2K = K_1 \cup K_2 with Ki∈Q(ΣP)K_i \in\mathcal{Q}(\Sigma P). Then

Λ(K)=Λ(K1)∪Λ(K2).\Lambda(K) = \Lambda(K_1) \cup\Lambda(K_2).

Thus Λ(K)=Λ(Ki)\Lambda(K) = \Lambda(K_i) for some ii, and applying μ\mu gives K=KiK = K_i. This proves (3).

If K=↑xK = \uparrow x, then

Λ(K)=↑Fin⁡(P)η(x).\Lambda(K) = \uparrow_{\operatorname{Fin}(P)}\eta(x).

Conversely, suppose Λ(K)=↑Fin⁡(P)H0\Lambda(K) = \uparrow_{\operatorname{Fin}(P)}H_0. Every member of ↑Fin⁡(P)H0\uparrow_{\operatorname{Fin}(P)}H_0 is contained in H0H_0, and H0H_0 itself belongs to this set. Hence its union is H0H_0, and (1) gives K=H0K = H_0. Since H0∈Λ(K)H_0 \in\Lambda(K), there is x∈Kx \in K with H0⊆↑xH_0 \subseteq\uparrow x. Because x∈H0x \in H_0 and H0H_0 is upper, ↑x⊆H0\uparrow x \subseteq H_0. Hence K=↑xK = \uparrow x. This proves (4).

Corollary 3.3. For every dcpodcpo PP,

ΣFin⁡(P) co-sober⟹ΣP co-sober.\Sigma\operatorname{Fin}(P)\ \text{co-sober} \Longrightarrow\Sigma P\ \text{co-sober}.

Equivalently, non-co-sobriety of ΣP\Sigma P implies non-co-sobriety of ΣFin⁡(P)\Sigma\operatorname{Fin}(P).

Proof. If ΣP\Sigma P is not co-sober, choose a nonprincipal kk-irreducible member KK of Q(ΣP)\mathcal{Q}(\Sigma P). Proposition 3.2 shows that Λ(K)\Lambda(K) is a nonprincipal kk-irreducible member of Q(ΣFin⁡(P))\mathcal{Q}(\Sigma\operatorname{Fin}(P)).

Remark 3.4. The proof does not assert that Λ\Lambda is Scott continuous as a map between compact saturated power posets, or that Λ\Lambda and μ\mu form a topological retraction. Proposition 3.1 is used to prove compactness of μ(A)\mu(\mathcal{A}) and supplies the finite step not contained in the identity μΛ=id⁡\mu\Lambda= \operatorname{id}.

A counterexample

Proof of Theorem 1.1. Shen, Wu, Xi and Zhao constructed a dcpo PP such that ΣP\Sigma P is sober but not co-sober [6 Example 2.9]. Let

Y=A(P).Y = A(P).

By Lemma 2.1, YY is a T0T_0 co-sober space and Q(Y)=Fin⁡(P)Q(Y)=\operatorname{Fin}(P). Corollary 3.3 shows that

Pσ(Y)=ΣQ(Y)=ΣFin⁡(P)P_{\sigma}(Y)=\Sigma Q(Y)=\Sigma\operatorname{Fin}(P)

is not co-sober.

The preceding argument uses no cardinality property of the dcpo in [6 Example 2.9]. The remainder of the paper gives a separate countable construction.

A countable witness in a lower topology

Miao, Xi, Li and Zhao proved that a countable non-sober complete lattice exists [5 Corollary 3.6]. Fix such a lattice LL. Since ΣL\Sigma L is T0T_0 and non-sober, there is a nonempty irreducible Scott closed set C⊆LC\subseteq L such that

C≠↓x for every x∈L.(4)C\ne\mathord{\downarrow}x\text{ for every }x\in L. \tag*{(4)}

Only the existence statement from [5 Corollary 3.6] is used; no witness from that paper is imported.

The lower topology ω(L)\omega(L) is generated by the subbasic sets

L∖↑a,a∈L.L\setminus\mathord{\uparrow}a,\quad a\in L.

Its specialization order is the reverse of the lattice order. Thus its saturated sets are precisely the lower subsets of LL, and the saturation of xx is ↓x\mathord{\downarrow}x.

We shall also use the elementary inclusion

ω(L)⊆σ(Lop⁡).(5)\omega(L)\subseteq\sigma(L^{\operatorname{op}}). \tag*{(5)}

Indeed, each lower subbasic set L∖↑aL\setminus\mathord{\uparrow}a is an upper set in Lop⁡L^{\operatorname{op}}. If a filtered set F⊆LF\subseteq L has meet ⋀F∉↑a\bigwedge F\notin\mathord{\uparrow}a, then some f∈Ff\in F satisfies a≰fa\not\leq f; otherwise aa would be a lower bound of FF and hence a≤⋀Fa\leq\bigwedge F. Thus L∖↑aL\setminus\mathord{\uparrow}a is Scott open in Lop⁡L^{\operatorname{op}}, and so is every lower open set.

Lemma 5.1. Let LL be a complete lattice and let K⊆LK\subseteq L be nonempty. Then KK is compact saturated in (L,ω(L))(L,\omega(L)) if and only if KK is Scott closed in LL.

Proof. Suppose first that KK is Scott closed. It is lower and therefore saturated. By Alexander’s subbase theorem, it is enough to consider a cover by lower subbasic opens. Let

K⊆⋃i∈I(L∖↑ai),K\subseteq\bigcup_{i\in I}(L\setminus\mathord{\uparrow}a_i),

and suppose that no finite subfamily covers KK. For every finite F⊆IF\subseteq I, choose kF∈Kk_F\in K with ai≤kFa_i\leq k_F for all i∈Fi\in F, and put

dF=⋁i∈Fai.d_F=\bigvee_{i\in F}a_i.

where d∅=⊥d_{\emptyset}=\bot. Since dF≤kFd_F\leq k_F and KK is lower, dF∈Kd_F\in K. The family {dF:F⊆I finite}\{d_F:F\subseteq I\text{ finite}\} is directed, and Scott closedness gives

⋁F⊆IF finitedF=⋁i∈Iai∈K.\bigvee_{\substack{F\subseteq I\\F\text{ finite}}}d_F=\bigvee_{i\in I}a_i\in K.

This element lies in every ↑ai\uparrow a_i, contradicting the cover. Hence KK is compact.

Conversely, suppose that KK is compact saturated. Then KK is lower. Let D⊆KD\subseteq K be directed and set s=⋁Ds=\bigvee D. If s∉Ks\notin K, then for every k∈Kk\in K there is d∈Dd\in D with d≰kd\nleq k; otherwise kk would be an upper bound of DD, and lowerness would give s∈Ks\in K. Hence

{L∖↑d:d∈D}\{L\setminus\uparrow d:d\in D\}

is an open cover of KK. A finite subcover corresponds to d1,…,dn∈Dd_1,\ldots,d_n\in D. Choose d∈Dd\in D above all did_i. Then d∈Kd\in K belongs to none of the selected complements, a contradiction. Thus s∈Ks\in K, and KK is Scott closed.

A nonempty closed set is irreducible exactly when it is not the union of two proper closed subsets. Lemma 5.1 and (4) give the following.

Corollary 5.2. The set CC is a nonprincipal kk-irreducible compact saturated subset of (L,ω(L))(L,\omega(L)).

A countable staircase dcpo

Fix a sequence a0,a1,…a_0,a_1,\ldots in which every element of LL occurs. For x∈Lx\in L and n∈Nn\in\mathbb{N}, define

Un(x)=⋂i≤nai≰x(L∖↑ai).U_n(x)=\bigcap_{\substack{i\leq n\\a_i\nleq x}}(L\setminus\uparrow a_i).

The empty intersection is LL. Each Un(x)U_n(x) is a lower open neighbourhood of xx, the sequence is decreasing, and

⋂n∈NUn(x)=↓x.\bigcap_{n\in\mathbb{N}}U_n(x)=\downarrow x.

Moreover, the sets Un(x)U_n(x) form a neighbourhood base at xx in the lower topology, since every basic lower open neighbourhood involves only finitely many subbasic complements.

Let

DL={x^:x∈L}∪{xn:x∈L, n∈N}.D_L=\{\hat{x}:x\in L\}\cup\{x_n:x\in L,\ n\in\mathbb{N}\}.

Define a relation on DLD_L by

xn≤xm⟺n≤m for the same label x,xn≤y^⟺y∈Un(x),x^≤y^⟺y≤Lx.\begin{aligned} x_n\leq x_m&\Longleftrightarrow n\leq m\text{ for the same label }x,\\ x_n\leq\hat{y}&\Longleftrightarrow y\in U_n(x),\\ \hat{x}\leq\hat{y}&\Longleftrightarrow y\leq_L x. \end{aligned}

We impose no other comparisons. Reflexivity and antisymmetry are immediate. Transitivity in the mixed cases follows from Um(x)⊆Un(x)U_m(x)\subseteq U_n(x) for n≤mn\leq m and from lowerness of Un(x)U_n(x). Thus this relation is a partial order, and DLD_L is countable.

Lemma 6.1. The poset DLD_L is a dcpo. Let E⊆DLE \subseteq D_L be directed.

(1) If EE contains a point x^\hat{x}, set

E#={x∈L:x^∈E}.E^\# = \{x \in L : \hat{x} \in E\}.

Then E#E^\# is filtered and

⋁E=⋀E#.\bigvee E = \bigwedge E^\#.

(2) If EE contains no point of the form x^\hat{x}, then all its elements lie on one staircase. If the occurring indices have a maximum mm, the supremum is xmx_m; if they are unbounded, the supremum is x^\hat{x}.

Proof. Suppose that EE contains x^0\hat{x}_0. For each e∈Ee \in E, directedness gives u∈Eu \in E above both ee and x^0\hat{x}_0. No staircase point lies above a point x^\hat{x}, so the point part of EE is cofinal. If x,y∈E#x,y \in E^\#, a common upper bound in EE has the form z^\hat{z} with z≤Lx,yz \leq_L x,y. Hence E#E^\# is filtered. If m=⋀E#m = \bigwedge E^\#, then m^\hat{m} is the least upper bound of the cofinal point part and therefore of EE.

Now suppose that EE contains only staircase points. Points with different labels have no common staircase upper bound, so directedness forces one label xx. If the set of indices is bounded, it has a maximum. If it is unbounded, every member of EE is below x^\hat{x}. Any upper bound must then have the form y^\hat{y}, and y∈Un(x)y \in U_n(x) for every index occurring in EE. Unboundedness and the decreasing property of the sets Un(x)U_n(x) give

y∈⋂n∈NUn(x)=↓x.y \in\bigcap_{n\in\mathbb{N}} U_n(x) = \mathord{\downarrow}x.

Thus x^≤y^\hat{x} \leq\hat{y}, proving that x^\hat{x} is the supremum.

Let

L^={x^:x∈L}⊆DL,e(x)=x^.\widehat{L} = \{\hat{x}: x \in L\} \subseteq D_L,\quad e(x) = \hat{x}.

The point set L^\widehat{L} is upper in DLD_L, hence saturated in ΣDL\Sigma D_L.

Proposition 6.2. The map

e:(L,ω(L))⟶L^⊆ΣDLe : (L,\omega(L)) \longrightarrow\widehat{L} \subseteq\Sigma D_L

is a homeomorphism onto the saturated subspace L^\widehat{L}.

Proof. The map is bijective. Let OO be Scott open in DLD_L and put

V=e−1(O∩L^).V = e^{-1}(O \cap\widehat{L}).

If x∈Vx \in V, the directed chain x0≤x1≤⋯x_0 \leq x_1 \leq\cdots has supremum x^\hat{x}. Scott inaccessibility gives xn∈Ox_n \in O for some nn. If y∈Un(x)y \in U_n(x), then xn≤y^x_n \leq\hat{y}, so y^∈O\hat{y} \in O. Hence Un(x)⊆VU_n(x) \subseteq V, and VV is lower open.

Conversely, let VV be lower open and define

OV={y^:y∈V}∪{xn:Un(x)⊆V}.O_V = \{\hat{y}: y \in V\} \cup\{x_n : U_n(x) \subseteq V\}.

The set OVO_V is upper. If y^∈OV\hat{y} \in O_V and y^≤z^\hat{y} \leq\hat{z}, then z≤Lyz \leq_L y, and lowerness of VV gives z∈Vz \in V. If xn∈OVx_n \in O_V and xn≤xmx_n \leq x_m, then Um(x)⊆Un(x)⊆VU_m(x) \subseteq U_n(x) \subseteq V. If xn∈OVx_n \in O_V and xn≤y^x_n \leq\hat{y}, then y∈Un(x)⊆Vy \in U_n(x) \subseteq V. These are all possible cases.

Let E⊆DLE \subseteq D_L be directed and suppose ⋁E∈OV\bigvee E \in O_V. If EE is contained in one staircase and its indices are bounded, its supremum is its maximum and belongs to E∩OVE \cap O_V. If its indices are unbounded, say on the xx staircase, then ⋁E=x^\bigvee E = \hat{x} and x∈Vx \in V. Choose NN with UN(x)⊆VU_N(x) \subseteq V. Some occurring index is at least NN, so E∩OV≠∅E \cap O_V \ne\varnothing. It remains to consider the case in which EE contains point elements. By Lemma 6.1,

⋁E=m^,m=⋀E♯.\bigvee E = \hat{m},\quad m = \bigwedge E^{\sharp}.

The set E♯E^{\sharp} is filtered in LL, hence directed in LopL^{\mathrm{op}}, and its supremum in LopL^{\mathrm{op}} is mm. Since m∈Vm \in V and V∈ω(L)⊆σ(Lop)V \in\omega(L) \subseteq\sigma(L^{\mathrm{op}}) by (3), Scott inaccessibility gives some y∈E♯∩Vy \in E^{\sharp} \cap V. Thus y^∈E∩OV\hat{y} \in E \cap O_V. Hence OVO_V is Scott open. By construction, OV∩L^=e[V]O_V \cap\hat{L} = e[V], which proves equality of the two topologies.

Corollary 6.3. The Scott space ΣDL\Sigma D_L is not co-sober.

Proof. By Corollary 5.2 and Proposition 6.2, C^=e[C]\hat{C} = e[C] is compact saturated and kk-irreducible in the point subspace. Lemma 2.2(2) gives the same properties in ΣDL\Sigma D_L.

The set C^\hat{C} is not principal. If C^=↑d\hat{C} = \uparrow d and d=xnd = x_n is a staircase point, then d∈↑dd \in\uparrow d but d∉L^d \notin\hat{L}. If d=x^d = \hat{x}, then

↑DLx^={y^:y≤Lx},\uparrow_{D_L}\hat{x} = \{\hat{y}: y \leq_L x\},

so C=↓xC = \downarrow x, contradicting (2).

The countable counterexample

Proof of Theorem 1.2. Let

Yc=A(DL).Y_c = A(D_L).

The space YcY_c is countable, T0T_0, and co-sober by Lemma 2.1. Moreover,

Q(Yc)=Fin⁡(DL),Q(Y_c) = \operatorname{Fin}(D_L),

which is countable because DLD_L is countable. Corollary 6.3 and Corollary 3.3 imply that

Pσ(Yc)=ΣFin⁡(DL)P_{\sigma}(Y_c) = \Sigma\operatorname{Fin}(D_L)

is not co-sober. Lemma 2.1(3) shows that the upper Vietoris space PS(Yc)P_S(Y_c) is the Alexandrov space of the same Smyth ordered set Fin⁡(DL)\operatorname{Fin}(D_L) and is therefore co-sober.

Hence the upper Vietoris and Scott topologies have different co-sobriety behavior on the same countable Smyth power poset.

Concluding remarks

Theorem 1.1 shows that the Scott power construction does not preserve co-sobriety. The finite transfer uses the point-saturation and union maps from the full compact saturated setting and Heckmann and Keimel’s finite Rudin result at the compactness step. The separate staircase construction yields a countable example and, for the same countable Smyth power poset, distinguishes the upper Vietoris topology from the Scott topology.

References

  1. [1]M. Escardó, J. D. Lawson, A. Simpson, Comparing Cartesian closed categories of (core) compactly generated spaces, Topology Appl. 143 (2004), 105–145.DOI
  2. [2]G. Gierz, K. H. Hofmann, K. Keimel, J. D. Lawson, M. Mislove, D. S. Scott, Continuous Lattices and Domains, Encyclopedia of Mathematics and its Applications, vol. 93, Cambridge University Press, Cambridge, 2003.DOI
  3. [3]J. Goubault-Larrecq, Non-Hausdorff Topology and Domain Theory: Selected Topics in Point-Set Topology, New Mathematical Monographs, vol. 22, Cambridge University Press, Cambridge, 2013.
  4. [4]R. Heckmann, K. Keimel, Quasicontinuous domains and the Smyth powerdomain, Electron. Notes Theor. Comput. Sci. 298 (2013), 215–232.DOI
  5. [5]H. Miao, X. Xi, Q. Li, D. Zhao, Not every countable complete distributive lattice is sober, Math. Struct. Comput. Sci. 33 (9) (2023), 809–831.DOI
  6. [6]C. Shen, G. Wu, X. Xi, D. Zhao, Sober Scott spaces are not always co-sober, Topology Appl. 282 (2020) 107316.DOI
  7. [7]X. Wen, X. Xu, Sober is not always co-sober, Topology Appl. 250 (2018) 48–52.DOI
  8. [8]X. Xu, X. Wen, X. Xi, Scott topology on Smyth power posets, Math. Struct. Comput. Sci. 33 (9) (2023), 832–867.DOI
  9. [9]X. Xu, F. Xu, H. Ji, D. Qiu, Smyth power space of a co-sober space is not always co-sober, arXiv:2608.02047 [math.GN] (2026).arxiv.org/abs/2608.02047
  10. [10]B. Zhao, Z. He, K. Wang, On co-sober spaces, Houston J. Math. 49 (4) (2023) 971–988.DOI

Paper details

vlnc:W_vhcy8bxp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents