betaFeedback

WorksMathematics

The Scott sobriety on countable Frechet-Urysohn dcpos and locally compact meet-continuous dcpos

Abstract

In this paper, we focus on the Scott Sobriety of countable dcpos and locally compact dcpos. Recently, Chen, Kou and Yang proved that every countable dcpo with its Scott topology first-countable is sober endowed with the Scott topology. A natural question is whether a countable dcpo whose Scott topology is a Frechet-Urysohn sapce must be sober. We show that a countable dcpo with condition FG is Scott sober if its Scott topology is Frechet-Urysohn. Besides, a countable Scott non sober Frechet-Urysohn dcpo is constructed. He and Zhang gave two sufficient conditions for the open problem whether every locally compact dcpo is Scott sober. One of their sufficient conditions is that any locally compact meet-continuous coherent dcpo is Scott sober. We further prove that each locally compact meet-continuous dcpo is always sober with respect to the Scott topology.

Keywords. Scott topology, sobriety, Fréchet–Urysohn space, countable dcpo, weak one-step closure, meet-continuity, local compactness.

Mathematics Subject Classification: 54D10; 54D55; 06B35

Introduction

Sobriety is probably the most important and useful property of non-Hausdorff topological spaces. In Domain theory, a classical result is that every quasicontinuous dcpos and in particular every continuous dcpos is sober with respect to the Scott topology ([4]). Countability alone gives some different examples. Johnstone constructed a countable dcpo with a non-sober Scott space ([8]). Miao, Xi, Li and Zhao later answered a question of Jung by constructing a countable complete lattice with non-sober Scott topology. Their construction even admits a distributive lattice order ([10]).

Countability nevertheless has a strong consequence for convergence. The Scott space of every countable dcpo is sequential. Indeed, a sequentially closed subset is lower, as the constant sequence at pp converges to each point below pp. It also contains the supremum of each of its directed subsets, since every countable directed set has an increasing cofinal sequence converging to its supremum. Chen, Kou and Yang obtained a positive sobriety result under a stronger assumption. They proved that first countability of ΣP\Sigma P forces a countable dcpo PP to be quasicontinuous and hence Scott sober ([3]). These two results single out the Fréchet–Urysohn property. First countability implies this property, which in turn implies sequentiality. Its distinctive requirement is that each individual point of cl⁡A\operatorname{cl} A be the limit of a sequence from AA. A related theorem of He concerns the square of a space. An ω\omega-well-filtered coherent dd-space XX is sober if X×XX \times X is Fréchet–Urysohn ([5]). Here we ask whether the Fréchet–Urysohn property of ΣP\Sigma P itself suffices when PP is countable.

We answer this question negatively. The countable dcpo constructed in Section 4 has an irreducible Fréchet-Urysohn Scott space and no greatest element. Its whole space is therefore a closed irreducible set without a generic point. To verify the Fréchet-Urysohn property we compute the Scott closure of an arbitrary subset. One sequence in the construction, (ar1)r≥1(a^{1}_{r})_{r\geq1}, converges to every point of the space.

The positive result depends on the order structure of the common upper bounds. Suppose that for every x,y∈Px,y \in P, there is a finite set Fx,y⊆PF_{x,y} \subseteq P such that

↑x∩↑y=↑Fx,y.\uparrow x \cap\uparrow y = \uparrow F_{x,y}.

Miao, Yuan and Li used this finite generation property in their study of join continuity and well-filteredness ([11]). Here its role is to recover directed approximation from a single convergent sequence. Given an→xa_n \to x, countability allows us to pass to a subsequence on which, for each p∈Pp \in P, the assertion p≤anp \leq a_n has a constant truth value from some index onwards. Finite generation makes the set of eventual lower bounds directed. This yields weak one-step closure in the sense of Miao, Li and Zhao ([9]). We then prove that every countable dcpo with weak one-step closure is Scott sober. In that proof, an enumeration of an irreducible closed set without a greatest element would yield an increasing branch with no upper bound in the set, contrary to Scott closedness. Thus the Fréchet-Urysohn property suffices under FG. In particular, the counterexample constructed by Miao, Xi, Li and Zhao cannot be Fréchet-Urysohn, since binary joins give ↑x∩↑y=↑(x∨y)\uparrow x \cap\uparrow y = \uparrow(x \vee y) ([10]).

A second question arises when the cardinality restriction is lifted. Xu and Zhao asked whether local compactness of ΣP\Sigma P implies sobriety for an arbitrary dcpo PP ([13]). The countable case has an affirmative answer. In fact, He, Dong and Wang proved that a countable dcpo with core-compact Scott space is quasicontinuous ([6]). For arbitrary cardinality, He and Zhang proved sobriety when ΣP\Sigma P is locally compact and second countable. They also proved it when PP is meet-continuous and ΣP\Sigma P is locally compact and coherent ([7]).

Can coherence be removed from the latter result?

Chen constructed a meet-continuous dcpo with a compact second-countable core-compact dcpo that is neither locally compact nor sober ([1]). This gives a negative answer to the related question of Jia recorded in ([13]). Theorem 5.18 answers Question 3.14 affirmatively for meet-continuous dcpos of arbitrary cardinality. He and Zhang use coherence to make ↑a∩↑b\uparrow a \cap\uparrow b compact ([7]). This allows them to take infima along the decreasing columns of a triangular array. We instead work inside each of a sequence of prescribed Scott-compact sets KnK_n. Meet-continuity supplies common lower bounds for finite families in suitably chosen neighbourhoods. By successive relatively closed refinements and a finitely branching tree argument, we obtain points pn∈Knp_n \in K_n with pn≤pn+1p_n \leq p_{n+1} whenever Kn+1⊆int⁡(↑Kn)K_{n+1} \subseteq\operatorname{int}(\uparrow K_n) for all nn. Inside an irreducible Scott-closed set FF, this selection puts a point in every countable intersection of nonempty relatively open sets. We can then adapt the minimal closed set argument of He and Zhang to show that the points with a countable neighbourhood base in FF form a dense directed set. Its supremum is the greatest element of FF.

Preliminaries

Write N+={1,2,…}\mathbb{N}_+ = \{1,2,\ldots\} and N0={0,1,2,…}\mathbb{N}_0 = \{0,1,2,\ldots\}. A countable set is a finite or countably infinite set. For a subset AA of a partially ordered set PP, put

↑A={y∈P∣∃a∈A,a≤y},↓A={y∈P∣∃a∈A,y≤a}.\uparrow A = \{y \in P \mid\exists a \in A, a \leq y\}, \qquad\downarrow A = \{y \in P \mid\exists a \in A, y \leq a\}.

For a singleton, we write ↑x\uparrow x and ↓x\downarrow x. A set AA is upper if A=↑AA = \uparrow A and lower if A=↓AA = \downarrow A.

A subset DD of PP is directed if it is nonempty and every pair of its elements has an upper bound in DD. A dcpo is a partially ordered set in which every directed subset DD admits a supremum, denoted by ⋁D\bigvee D. A subset E⊆DE \subseteq D is cofinal in DD if each d∈Dd \in D lies below some e∈Ee \in E.

Let PP be a dcpo. Its Scott topology σ(P)\sigma(P) consists of the upper sets UU with the following property. For each directed D⊆PD \subseteq P, if ⋁D∈U\bigvee D \in U, then D∩U≠∅D \cap U \ne\varnothing. We denote the space (P,σ(P))(P,\sigma(P)) by ΣP\Sigma P. The Scott closed sets are precisely the lower sets containing the suprema of their directed subsets.

A nonempty subset AA of a space XX is irreducible if A⊆C1∪C2A \subseteq C_1 \cup C_2 with C1,C2C_1,C_2 closed in XX implies A⊆C1A \subseteq C_1 or A⊆C2A \subseteq C_2. Equivalently, if two open sets meet AA, then their intersection meets AA. A T0T_0 space is sober if each nonempty irreducible closed set is the closure of a unique point.

A space XX is Fréchet–Urysohn if for each A⊆XA \subseteq X and x∈cl⁡XAx \in\operatorname{cl}_X A there exists a sequence of points of AA converging to xx. If xn→xx_n \to x in ΣP\Sigma P, then xn→yx_n \to y for every y≤xy \leq x, since every open neighbourhood of yy contains xx. If xn→xx_n \to x and xn≤ynx_n \leq y_n for all nn, then also yn→xy_n \to x, since any open set containing xnx_n contains yny_n.

In a topological space XX, a GδG_\delta set is a countable intersection of open sets and a subset is dense precisely when it meets every nonempty open set.

Definition 2.1 ([9]). A dcpo PP has weak one-step closure if for every A⊆PA \subseteq P and x∈cl⁡Ax \in\operatorname{cl} A, there is a directed D⊆↓AD \subseteq\mathord{\downarrow} A with x≤⋁Dx \leq\bigvee D.

For any dcpo, if D⊆↓AD \subseteq\mathord{\downarrow} A is directed, then cl⁡A\operatorname{cl} A contains DD, its supremum and all points below that supremum. Thus the definition is equivalent to the identity

cl⁡A={x∈P:x≤⋁D for some directed D⊆↓A}(A⊆P).\operatorname{cl} A = \{x \in P : x \leq\bigvee D \text{ for some directed } D \subseteq\mathord{\downarrow} A\} \qquad(A \subseteq P).

Definition 2.2. A poset PP satisfies the finite generation condition for intersections of principal upper sets if

for each x,y∈P,↑x∩↑y=↑Fx,yfor some finite Fx,y⊆P.(FG)\text{for each } x,y \in P,\quad\mathord{\uparrow}x \cap\mathord{\uparrow}y = \mathord{\uparrow}F_{x,y}\quad\text{for some finite } F_{x,y} \subseteq P. \qquad(\mathrm{FG})

Lemma 2.3. Let FF be a Scott closed subset of a dcpo PP. The induced order makes FF a dcpo, and σ(F)\sigma(F) is the subspace topology on FF from ΣP\Sigma P. If PP has weak one-step closure, then FF has weak one-step closure.

Proof. For a directed D⊆FD \subseteq F, Scott closedness gives ⋁PD∈F\bigvee_P D \in F. This is also the least upper bound in FF. Consequently the trace on FF of each Scott-closed subset of PP is Scott closed in FF. Conversely, a Scott-closed subset BB of FF is lower in PP since FF is lower in PP. The equality of the two suprema shows that BB contains the supremum in PP of each directed subset of BB. This proves that BB is Scott closed in PP and establishes the assertion about topologies.

Suppose now that PP has weak one-step closure. Take A⊆FA \subseteq F and x∈cl⁡ΣFAx \in\operatorname{cl}_{\Sigma F} A. Then x∈cl⁡ΣPAx \in\operatorname{cl}_{\Sigma P} A, so some directed D⊆↓PAD \subseteq\mathord{\downarrow}_P A has x≤⋁PDx \leq\bigvee_P D. Lowerness gives ↓PA=↓FA\mathord{\downarrow}_P A = \mathord{\downarrow}_F A, and the equality ⋁PD=⋁FD\bigvee_P D = \bigvee_F D proves the last assertion. ◁

A sufficient condition for Fréchet-Urysohn countable dcpo becoming sober

We begin with a property of Scott-convergent sequences that applies even when the dcpo is uncountable.

Lemma 3.1. Suppose xn→xx_n \to x in the Scott space of a dcpo PP. Every Scott-open neighbourhood UU of xx contains an element uu for which u≤xnu \leq x_n for infinitely many indices nn.

Proof. Suppose each u∈Uu \in U lies below only finitely many terms. Applying this to xx, we can discard finitely many terms and reindex so that x≰xnx \nleq x_n for every nn. The lower set

C=(P∖U)∪⋃n≥1↓xnC = (P \setminus U) \cup\bigcup_{n \geq1} \mathord{\downarrow}x_n

contains all the remaining terms and omits xx. We prove that CC is Scott closed. Let D⊆CD \subseteq C be directed and put d=⋁Dd = \bigvee D. If d∉Ud \notin U, then d∈Cd \in C. Otherwise take d0∈D∩Ud_0 \in D \cap U. The set D0=D∩↓d0D_0 = D \cap\mathord{\downarrow}d_0 is directed and cofinal in DD, so ⋁D0=d\bigvee D_0 = d. Every z∈D0z \in D_0 belongs to C∩UC \cap U and hence lies below some xnx_n. For this index d0≤z≤xnd_0 \leq z \leq x_n. Therefore

D0⊆⋃{n:d0≤xn}↓xn.D_0 \subseteq\bigcup_{\{n:d_0\leq x_n\}} \mathord{\downarrow}x_n.

The right-hand side is a finite union of closed sets and is contained in CC. It contains ⋁D0\bigvee D_0, proving Scott closedness. The open neighbourhood P∖CP \setminus C of xx now contradicts convergence.

Lemma 3.2. Let xn→xx_n \to x in the Scott space of a countable dcpo PP. There is a subsequence (xnk)(x_{n_k}) such that for each p∈Pp \in P the truth value of p≤xnkp \le x_{n_k} is constant for all sufficiently large kk. Its set of eventual lower bounds

L={p∈P:p≤xnk for all sufficiently large k}L = \{p \in P : p \le x_{n_k}\text{ for all sufficiently large }k\}

is lower and satisfies x∈cl⁡Lx \in\operatorname{cl} L.

Proof. Enumerate PP as p1,p2,…p_1,p_2,\ldots, repeating elements if it is finite. Put I0=N+I_0 = \mathbb{N}_+. For k≥1k \ge1, one of the two possible truth values of pk≤xnp_k \le x_n occurs on an infinite subset IkI_k of Ik−1I_{k-1}. Choose such an IkI_k, then choose strictly increasing indices nk∈Ikn_k \in I_k. For k≥ik \ge i we have nk∈Iin_k \in I_i, which gives the asserted constancy.

Transitivity makes LL lower. For any Scott-open neighbourhood UU of xx, apply Lemma 3.1 to the convergent subsequence. It supplies p∈Up \in U which lies below infinitely many xnkx_{n_k}. The stabilized comparison with this particular pp must therefore be true. Hence p∈U∩Lp \in U \cap L, proving x∈cl⁡Lx \in\operatorname{cl} L.

Proposition 3.3. Let PP be a countable dcpo satisfying FGFG. If ΣP\Sigma P is Fréchet–Urysohn, then PP has weak one-step closure.

Proof. Fix A⊆PA \subseteq P and x∈cl⁡Ax \in\operatorname{cl} A, and take a sequence in AA converging to xx. Pass to the subsequence given by Lemma 3.2, denoting it by (an)(a_n) and its set of eventual lower bounds by LL. We have L⊆↓AL \subseteq\mathord{\downarrow} A and x∈cl⁡Lx \in\operatorname{cl} L, so LL is nonempty.

For p,q∈Lp,q \in L, all sufficiently late ana_n lie in ↑p∩↑q=↑Fp,q\mathord{\uparrow}p \cap\mathord{\uparrow}q = \mathord{\uparrow}F_{p,q}. The finite set Fp,qF_{p,q} is therefore nonempty. At least one member ff lies below infinitely many ana_n. By the choice of the subsequence, f≤anf \le a_n eventually, so f∈Lf \in L. As ff is a common upper bound of p,qp,q, this proves directedness. For any directed subset DD of a dcpo,

cl⁡D=↓(⋁D).\operatorname{cl} D = \mathord{\downarrow}\left(\bigvee D\right).

Indeed, the ideal on the right is closed and contains DD, while every closed set containing DD contains ⋁D\bigvee D and its lower bounds. Taking D=LD = L gives x≤⋁Lx \le\bigvee L, as required.

Lemma 3.4. If a countable dcpo PP has weak one-step closure, then ΣP\Sigma P is Fréchet–Urysohn.

Proof. For x∈cl⁡Ax \in\operatorname{cl} A choose a directed D⊆↓AD \subseteq\mathord{\downarrow} A with x≤⋁Dx \le\bigvee D. Enumerate DD as e1,e2,…e_1,e_2,\ldots, using repetitions if necessary. Set d1=e1d_1 = e_1 and choose dn+1∈Dd_{n+1} \in D above both dnd_n and en+1e_{n+1}. The sequence (dn)(d_n) is increasing and cofinal in DD. For each nn choose an∈Aa_n \in A above dnd_n. Every Scott-open neighbourhood UU of xx contains ⋁D\bigvee D. Consequently some dnd_n is in UU, and upperness puts every later dkd_k and aka_k in UU. This proves an→xa_n \to x.

Lemma 3.5. Let PP be a countable dcpo with weak one-step closure. Suppose ΣP\Sigma P is irreducible and PP has no greatest element. Then for each nonempty Scott-open set UU there exists a finite subset E⊆UE \subseteq U such that int⁡(↑E)≠∅\operatorname{int}(\mathord{\uparrow}E) \ne\varnothing.

Proof. Write MM for the set of maximal elements of PP. Each x∈Px \in P lies below a maximal element. Indeed, in ↑x\mathord{\uparrow}x every nonempty chain has its supremum as an upper bound, and xx bounds the empty chain. Zorn’s lemma gives a maximal element of ↑x\mathord{\uparrow}x, which is also maximal in PP. Thus MM meets every nonempty Scott-open set and is dense. Fix m∈Mm \in M. The closed set cl⁡(M∖{m})∪↓m\operatorname{cl}(M \setminus\{m\}) \cup\mathord{\downarrow}m contains MM and hence is all of PP. The ideal ↓m\mathord{\downarrow}m is proper by the hypothesis on greatest elements. Irreducibility now implies cl⁡(M∖{m})=P\operatorname{cl}(M \setminus\{m\}) = P.

Given UU, choose m∈U∩Mm \in U \cap M. Intersecting each open neighbourhood of mm with UU shows that m∈cl⁡((M∩U)∖{m})m \in\operatorname{cl}((M \cap U) \setminus\{m\}). By Lemma 3.4 there is a sequence from this set converging to $m. Each of its values vv occurs only finitely often, since P∖↓vP \setminus\mathord{\downarrow}v is a neighbourhood of mm. We may therefore take its terms m1,m2,…m_1,m_2,\ldots pairwise distinct.

Suppose for a contradiction that int⁡(↑E)=∅\operatorname{int}(\mathord{\uparrow}E)=\varnothing for every finite E⊆UE\subseteq U. Enumerate UU as u1,u2,…u_1,u_2,\ldots, put En={u1,…,un}E_n=\{u_1,\ldots,u_n\}, and let An=P∖↑EnA_n=P\setminus\mathord{\uparrow}E_n. Each AnA_n is lower and dense. Weak one-step closure supplies a directed subset of AnA_n with supremum above mnm_n. Its supremum is mnm_n by maximality. Since mn∈Um_n\in U, the subset meets UU. Its trace on UU is directed and cofinal, by directedness and upperness of UU. Call this trace DnD_n. We have

Dn⊆U∖↑En,⋁Dn=mn.D_n\subseteq U\setminus\mathord{\uparrow}E_n,\qquad\bigvee D_n=m_n.

Let B=⋃n≥1DnB=\bigcup_{n\geq1}D_n. The closed set cl⁡B\operatorname{cl}B contains every mnm_n, and hence their limit mm. Apply Lemma 3.4 once more to obtain bk∈Bb_k\in B converging to mm.

For each nn, only finitely many bkb_k belong to DnD_n, because Dn⊆↓mnD_n\subseteq\mathord{\downarrow}m_n and P∖↓mnP\setminus\mathord{\downarrow}m_n is a neighbourhood of mm. Fix ii. If ui≤bku_i\leq b_k and bk∈Dnb_k\in D_n, then n<in<i: otherwise ui∈Enu_i\in E_n would give bk∈↑Enb_k\in\mathord{\uparrow}E_n. It follows that

{k:ui≤bk}⊆⋃n<i{k:bk∈Dn}.\{k:u_i\leq b_k\}\subseteq\bigcup_{n<i}\{k:b_k\in D_n\}.

The set on the right is finite. Thus no uiu_i lies below infinitely many terms of (bk)(b_k), contradicting Lemma 3.1 for the neighbourhood UU of mm.

Theorem 3.6. Every countable dcpo with weak one-step closure has a sober Scott space.

Proof. Let FF be a nonempty irreducible Scott-closed subset of PP. By Lemma 2.3 we may work in FF with its induced order and Scott topology. Suppose FF has no greatest element. Enumerate FF as p1,p2,…p_1,p_2,\ldots, allowing repetitions. Starting with W0=FW_0=F, we construct nonempty open sets WnW_n and finite sets EnE_n. If Wn−1W_{n-1} has been chosen, the proper closed subset ↓Fpn\mathord{\downarrow}_F p_n has empty interior in FF, because every nonempty open set meets its nonempty open complement by irreducibility. Hence Un=Wn−1∖↓FpnU_n=W_{n-1}\setminus\mathord{\downarrow}_F p_n is nonempty and open. Lemma 3.5 gives a finite En⊆UnE_n\subseteq U_n with Wn=int⁡F(↑FEn)≠∅W_n=\operatorname{int}_F(\mathord{\uparrow}_F E_n)\ne\varnothing. In particular, En≠∅E_n\ne\varnothing. The set UnU_n is upper in FF, so Wn⊆↑FEn⊆UnW_n\subseteq\mathord{\uparrow}_F E_n\subseteq U_n. Consequently

En+1⊆↑FEn,En∩↓Fpn=∅.E_{n+1}\subseteq\mathord{\uparrow}_F E_n,\qquad E_n\cap\mathord{\downarrow}_F p_n=\varnothing.

Take the tree of finite increasing lists (e1,…,en)(e_1,\ldots,e_n) with ei∈Eie_i\in E_i, ordered by extension and rooted at the empty list. For every nn one can start from any en∈Ene_n\in E_n and choose predecessors backwards to obtain a list at level nn. Thus the tree has nodes at arbitrarily high levels. Each node has only finitely many immediate successors. König’s lemma gives an infinite branch (en)(e_n). Indeed, at each step choose a child having descendants at arbitrarily high levels, which exists since the number of children is finite. Let p=⋁nenp=\bigvee_n e_n in FF and take NN with p=pNp=p_N. Now eN∈ENe_N\in E_N and eN≤pNe_N\leq p_N, contradicting the construction. Therefore FF has a greatest element pp. Its lowerness in PP gives F=↓pF=\mathord{\downarrow}p, and antisymmetry gives uniqueness.

Theorem 3.7. If PP is a countable dcpo satisfying FG and ΣP\Sigma P is Fréchet–Urysohn, then ΣP\Sigma P is sober.

Proof. Proposition 3.3 gives weak one-step closure. Apply Theorem 3.6.

Corollary 3.8. Suppose PP is a countable dcpo in which each pair with a common upper bound has a least upper bound. If ΣP\Sigma P is Fréchet–Urysohn, then ΣP\Sigma P is sober.

Proof. If x,yx,y have a common upper bound, the singleton {x∨y}\{x\vee y\} generates ↑x∩↑y\mathord{\uparrow}x\cap\mathord{\uparrow}y. Otherwise the empty set generates this intersection. The hypotheses of Theorem 3.7 are satisfied.

Remark 3.9. The join assumption in Corollary 3.8 is strictly stronger than (FG). For example, let u,vu,v be incomparable and let s,ts,t be incomparable, with each of s,ts,t above both u,vu,v. Every upper set of this four-element poset is finitely generated. The common upper bounds of u,vu,v are exactly s,ts,t, so the pair has no least upper bound.

A countable non-sober Fréchet–Urysohn Scott space

We construct a countable dcpo whose Scott space is Fréchet–Urysohn and has a nonempty irreducible closed set without a generic point. This shows what can happen when the finite generation hypothesis in Theorem 3.7 is dropped.

Write N0={0,1,…}\mathbb{N}_0=\{0,1,\ldots\} and N+={1,2,…}\mathbb{N}_+=\{1,2,\ldots\}. Take distinct symbols

P={mjk:k∈N0, j∈N+}∪{dj,lk:k∈N0, j,l∈N+}∪{ark:k,r∈N+}.P=\{m_j^k:k\in\mathbb{N}_0,\ j\in\mathbb{N}_+\}\cup\{d_{j,l}^k:k\in\mathbb{N}_0,\ j,l\in\mathbb{N}_+\}\cup\{a_r^k:k,r\in\mathbb{N}_+\}.

In particular there is no aa-element at layer zero. Give PP the reflexive and transitive closure of the following relations.

(R1)dj,lk≤dj,l+1k≤mjk(k∈N0, j,l∈N+),(R2)dj,lk≤mrk+1(k∈N0, j,l,r∈N+, l≤r),(R3)dj,lk≤ark(k,j,l,r∈N+, l≤r),(R4)ark≤dj,lk−1(k,r,j,l∈N+, r≤j, r≤l).\begin{aligned} \text{(R1)}\quad&d_{j,l}^k\le d_{j,l+1}^k\le m_j^k &&(k\in\mathbb{N}_0,\ j,l\in\mathbb{N}_+),\\ \text{(R2)}\quad&d_{j,l}^k\le m_r^{k+1} &&(k\in\mathbb{N}_0,\ j,l,r\in\mathbb{N}_+,\ l\le r),\\ \text{(R3)}\quad&d_{j,l}^k\le a_r^k &&(k,j,l,r\in\mathbb{N}_+,\ l\le r),\\ \text{(R4)}\quad&a_r^k\le d_{j,l}^{k-1} &&(k,r,j,l\in\mathbb{N}_+,\ r\le j,\ r\le l). \end{aligned}

Denote the row {dj,lk:l∈N+}\{d_{j,l}^k:l\in\mathbb{N}_+\} by DjkD_j^k. The superscript of a dd- or aa-element is its layer. On these elements define the height by w(dj,lk)=lw(d_{j,l}^k)=l and w(ark)=rw(a_r^k)=r. Along an upward path avoiding the mm-elements, layers can only decrease and heights can only increase, in the weak sense. An (R4) step strictly decreases the layer. Inside one layer a path can ascend a single row and, at a positive layer, pass from the row to an aa-element. Leaving this aa-element requires an (R4) step. It follows that there is no cycle of nonidentity steps. Since no nonidentity step starts at an mm-element, the relation is a partial order and all the mjkm_j^k are maximal. The indexing makes PP countable.

Lemma 4.1. Suppose ds,lh≤mjkd_{s,l}^h\le m_j^k and (h,s)≠(k,j)(h,s)\ne(k,j). Then k≤h+1k\le h+1 and l≤jl\le j. If ds,lh≤zd_{s,l}^h\le z for a nonmaximal zz, then l≤w(z)l\le w(z).

Proof. A path to a nonmaximal element avoids every mm-element, so the height inequality follows along the path. For a path ending at mjkm_j^k, consider its last step. If it is (R2), it starts at a point di,tkd_{i,t}^k with t≤jt\le j. Before this step, comparison of the layers and the heights gives k−1≤hk-1\le h and l≤tl\le t. If the last step is (R1), the path has entered DjkD_j^k from another row, since the starting point is outside DjkD_j^k. Its first entry into DjkD_j^k must use (R4), from some ark+1a_r^{k+1} with r≤jr\le j. Consequently k≤hk\le h and l≤r≤jl\le r\le j.

Proposition 4.2. The poset PP is a dcpo. Every directed subset DD has a greatest element or has a cofinal subset in a row DjkD_j^k with unbounded heights. In the latter case ⋁D=mjk\bigvee D=m_j^k. In particular, ⋁Djk=mjk\bigvee D_j^k=m_j^k for all k∈N0k\in\mathbb{N}_0 and j∈N+j\in\mathbb{N}_+.

Proof. If mjk∈Dm_j^k\in D, directedness and maximality make it the greatest element of DD. Otherwise let hh be the least layer of a point of DD. If arh∈Da_r^h\in D, its only nonmaximal upper bound in a layer at least hh is arha_r^h itself. Given z∈Dz\in D, apply this observation to a common upper bound of zz and arha_r^h in DD. Thus arha_r^h is greatest.

It remains to consider a least layer hh containing only row elements of DD. Fix dj,l0h∈Dd_{j,l_0}^h\in D and set

D′={z∈D:dj,l0h≤z}.D'=\{z\in D:d_{j,l_0}^h\le z\}.

This set is directed and cofinal in DD. A point of D′D' lies at layer hh, since layer cannot increase along a path to it. At that layer the only possible targets of such a path are elements of DjhD_j^h and aa-elements. The latter have been excluded. Hence D′⊆DjhD'\subseteq D_j^h. If the heights in D′D' are bounded, the point of largest height is greatest in D′D' and, by cofinality, in DD. If they are unbounded, DD is cofinal in the row. It is bounded above by mjhm_j^h. Lemma 4.1 excludes a nonmaximal upper bound, since its height would bound all these heights. It also excludes a different maximal symbol mikm_i^k, since it would give the same bound. Thus mjhm_j^h is the only upper bound of D′D' and ⋁D=mjh\bigvee D=m_j^h. Taking D=DjkD=D_j^k gives the final assertion.

Lemma 4.3. Every nonempty Scott-open set contains a tail of (mjk)j≥1(m_j^k)_{j\ge1} for each k∈N0k \in\mathbb{N}_0 and a tail of (ark)r≥1(a_r^k)_{r\ge1} for each k∈N+k \in\mathbb{N}_+. In particular, ark→xa_r^k \to x as r→∞r \to\infty for every k∈N+k \in\mathbb{N}_+ and every x∈Px \in P.

Proof. Let UU be nonempty and Scott open. Every row element lies below the maximal element of its row, and ark≤dr,rk−1≤mrk−1a_r^k \le d_{r,r}^{k-1} \le m_r^{k-1}. Thus UU contains some mjkm_j^k. Since mjk=⋁Djkm_j^k = \bigvee D_j^k, some dj,lkd_{j,l}^k belongs to UU. Upperness and (R2) give mrk+1∈Um_r^{k+1} \in U for every r≥lr \ge l. When k≥1k \ge1, (R3) gives ark∈Ua_r^k \in U for every r≥lr \ge l. In particular alk∈Ua_l^k \in U, so (R4) and (R1) give di,lk−1,mik−1∈Ud_{i,l}^{k-1},m_i^{k-1} \in U for all i≥li \ge l.

Starting with the first maximal symbol in UU and repeating these steps upwards and downwards, we obtain a maximal symbol in UU at every layer. For the full tail at a positive layer kk, use (R2) from layer k−1k-1. For the full tail at layer zero, use the downward step from layer one. For the aa-tail at positive layer kk, use (R3) at that layer. Every open neighbourhood of xx now contains a tail of (ark)r≥1(a_r^k)_{r\ge1}, which gives the convergence.

Proposition 4.4. The Scott space ΣP\Sigma P is irreducible and non-sober.

Proof. Any two nonempty opens contain tails of the same sequence (ar1)r≥1(a_r^1)_{r\ge1} and therefore intersect. Hence PP itself is irreducible and closed. The distinct maximal points m10m_1^0 and m20m_2^0 preclude a greatest element. Since each point closure is a principal ideal, PP has no generic point.

Proposition 4.5. For a lower set L⊆PL \subseteq P let Jk={j∈N+:Djk⊆L}J_k = \{j \in\mathbb{N}_+ : D_j^k \subseteq L\} for k∈N0k \in\mathbb{N}_0. If some JkJ_k is unbounded, then cl⁡L=P\operatorname{cl} L = P. If every JkJ_k is bounded, then

cl⁡L=L∪⋃k∈N0j∈Jk↓mjk.\operatorname{cl} L = L \cup\bigcup_{\substack{k\in\mathbb{N}_0\\j\in J_k}} \downarrow m_j^k.

Proof. Suppose JkJ_k is unbounded. For any r≥1r \ge1, choose j∈Jkj \in J_k with j≥rj \ge r. Relation (R4) gives ark+1≤dj,rk∈La_r^{k+1} \le d_{j,r}^k \in L. Thus the entire sequence (ark+1)r≥1(a_r^{k+1})_{r\ge1} belongs to LL. It converges to every point by Lemma 4.3, so cl⁡L=P\operatorname{cl} L = P.

Now let every JkJ_k be bounded, hence finite. Write HH for the right-hand side of the displayed formula. It is lower and satisfies L⊆H⊆cl⁡LL \subseteq H \subseteq\operatorname{cl} L, because each mjkm_j^k with j∈Jkj \in J_k is the supremum of a row contained in LL.

We first show that enlarging LL to HH does not complete any previously incomplete row. Fix s∉Jhs \notin J_h. A missing element of DshD_s^h bounds the heights of all earlier elements of that row in LL, since LL is lower. Thus some b∈N0b \in\mathbb{N}_0 bounds the heights in Dsh∩LD_s^h \cap L, including when the intersection is empty. If ds,lh∈H∖Ld_{s,l}^h \in H \setminus L, then ds,lh≤mjkd_{s,l}^h \le m_j^k for a pair with j∈Jkj \in J_k and (k,j)≠(h,s)(k,j) \ne(h,s). Lemma 4.1 gives k≤h+1k \le h+1 and l≤jl \le j. Hence the heights in Dsh∩HD_s^h \cap H are bounded above by

max⁡({b,0}∪J0∪⋯∪Jh+1).\max\bigl(\{b,0\}\cup J_0\cup\cdots\cup J_{h+1}\bigr).

This maximum exists by finiteness. In particular Dsh⊈HD_s^h \not\subseteq H.

Let D⊆HD \subseteq H be directed. If DD has a greatest element, its supremum is already in HH. Otherwise Proposition 4.2 provides a cofinal subset of unbounded height in some DshD_s^h, with ⋁D=msh\bigvee D=m_s^h. Lowerness now gives Dsh⊆HD_s^h \subseteq H. The preceding paragraph forces s∈Jhs \in J_h, so msh∈Hm_s^h \in H. We have proved that HH is Scott closed. Since L⊆H⊆cl⁡LL \subseteq H \subseteq\operatorname{cl} L, the formula follows.

Theorem 4.6. The Scott space ΣP\Sigma P is Fréchet–Urysohn.

Proof. Let A⊆PA \subseteq P and x∈cl⁡Ax \in\operatorname{cl} A. All Scott-closed sets are lower, so L=↓AL=\downarrow A has cl⁡L=cl⁡A\operatorname{cl} L=\operatorname{cl} A. Define the sets JkJ_k from LL as in Proposition 4.5. If some JkJ_k is unbounded, the proof of that proposition gives ark+1∈La_r^{k+1} \in L for every rr. Choose br∈Ab_r \in A with ark+1≤bra_r^{k+1} \le b_r. Every Scott-open neighbourhood of xx contains ark+1a_r^{k+1} eventually, and upperness then puts brb_r there eventually. Thus $b_r \to x. Suppose next that all the JkJ_k are bounded. For x∈Lx \in L, take b∈Ab \in A above xx. The constant sequence at bb converges to xx. For x∉Lx \notin L, the closure formula gives x≤mjkx \le m_j^k for some j∈Jkj \in J_k. For each l≥1l \ge1 take bl∈Ab_l \in A above dj,lkd_{j,l}^k. Every Scott-open neighbourhood of xx contains mjkm_j^k. It therefore meets DjkD_j^k and contains a tail of that row. It also contains blb_l eventually. This produces a sequence in AA converging to xx in every case.

Proposition 4.7. Put x=a11x = a_1^1 and y=a21y = a_2^1. Their common upper bounds form the set

↑x∩↑y=⋃j≥2↑dj,20={dj,l0:j,l≥2}∪{mj0:j≥2}∪{mj1:j≥2}.\uparrow x \cap\uparrow y = \bigcup_{j \ge2} \uparrow d_{j,2}^0 = \{d_{j,l}^0 : j,l \ge2\} \cup\{m_j^0 : j \ge2\} \cup\{m_j^1 : j \ge2\}.

This set has no finite generating subset.

Proof. The first strict step from ar1a_r^1 uses (R4), reaching a row element dj,l0d_{j,l}^0 with j,l≥rj,l \ge r. At layer zero a path can continue up that row, end at mj0m_j^0, or use (R2) to end at mt1m_t^1 with t≥lt \ge l. All these destinations are attained, so

↑ar1={ar1}∪{dj,l0:j,l≥r}∪{mj0:j≥r}∪{mj1:j≥r}.\uparrow a_r^1 = \{a_r^1\} \cup\{d_{j,l}^0 : j,l \ge r\} \cup\{m_j^0 : j \ge r\} \cup\{m_j^1 : j \ge r\}.

In particular, a11a_1^1 and a21a_2^1 are incomparable. Intersect the cases r=1r = 1 and r=2r = 2. In the intersection each row starts at dj,20d_{j,2}^0, and all the displayed maximal elements lie above at least one such row element. This gives both equalities. Moreover, distinct layer-zero rows have incomparable row elements. Thus each dj,20d_{j,2}^0 for j≥2j \ge2 is minimal in the intersection. If ↑E=↑x∩↑y\uparrow E = \uparrow x \cap\uparrow y, then EE is contained in the intersection and must contain every minimal element of it. The infinitely many distinct points dj,20d_{j,2}^0 rule out a finite EE.

Corollary 4.8. There exists a countable dcpo with a Fréchet–Urysohn, non-sober Scott space. Such a dcpo can fail weak one-step closure.

Proof. Proposition 4.2, Proposition 4.4, and Theorem 4.6 verify the assertions about PP and its Scott space. Weak one-step closure for PP would give sobriety by Theorem 3.6.

Every meet-continuous locally compact dcpo is Scott sober

The preceding results use the countability of the underlying order. We now consider a sufficient condition involving the order and its Scott topology that applies to dcpos of arbitrary cardinality. We work in ZFC. Compactness means that every open cover has a finite subcover. A compact neighbourhood of a point is a compact set whose interior contains the point. We use two consequences of compactness in subspaces. A closed subset of a compact space is compact. In a nonempty compact space, a family of closed sets has nonempty intersection whenever every finite subfamily does, since otherwise the complements form an open cover with no finite subcover. This condition on finite subfamilies is the finite intersection property.

Definition 5.1. A Scott space ΣP\Sigma P is locally compact if for every x∈U∈σ(P)x \in U \in\sigma(P) there are a Scott-open set VV and a compact saturated set KK such that

x∈V⊆K⊆U.(1)x \in V \subseteq K \subseteq U. \tag*{(1)}

A set is saturated if it is an intersection of open sets. A dcpo PP is meet-continuous if for every x∈Px \in P and directed D⊆PD \subseteq P,

x≤⋁D⟹x∈cl⁡ΣP(↓x∩↓D).(2)x \le\bigvee D \quad\Longrightarrow\quad x \in\operatorname{cl}_{\Sigma P}(\downarrow x \cap\downarrow D). \tag*{(2)}

Equivalently, if x≤⋁Dx \le\bigvee D, every Scott-open neighbourhood of xx contains a point below both xx and some d∈Dd \in D.

We use subscripts in ↑XA\mathord{\uparrow}_{X}A, ↓XA\mathord{\downarrow}_{X}A, int⁡XA\operatorname{int}_{X}A and cl⁡XA\operatorname{cl}_{X}A to indicate the ambient order or topology when it is not clear from the argument.

Lemma 5.2. For x∈Px \in P, the principal ideal ↓x\mathord{\downarrow}x is Scott closed and equals cl⁡ΣP{x}\operatorname{cl}_{\Sigma P}\{x\}. The saturation of every A⊆PA \subseteq P is ↑A\mathord{\uparrow}A. In particular, ↑x\mathord{\uparrow}x is the intersection of all Scott-open neighbourhoods of xx.

Proof. The set ↓x\mathord{\downarrow}x is lower and closed under directed suprema. Every Scott-closed set containing xx contains ↓x\mathord{\downarrow}x, which proves the first assertion. Every open set containing AA contains ↑A\mathord{\uparrow}A. If y∉↑Ay \notin\mathord{\uparrow}A, then P∖↓yP \setminus\mathord{\downarrow}y is open, contains AA, and omits yy. This gives the reverse inclusion for the saturation.

It follows that the specialisation order of ΣP\Sigma P, given by x≤spyx \leq_{\mathrm{sp}} y if x∈cl⁡ΣP{y}x \in\operatorname{cl}_{\Sigma P}\{y\}, is the original order. Thus the Scott space is T0T_{0}. For a Scott-closed set FF, finding its generic point amounts to finding its greatest element.

Lemma 5.3. If PP is meet-continuous, then ↑(W∩↓b)\mathord{\uparrow}(W \cap\mathord{\downarrow}b) is Scott open for each W∈σ(P)W \in\sigma(P) and b∈Pb \in P.

Proof. The set is upper. Suppose it contains ⋁D\mathord{\bigvee}D for a directed DD. There is w∈Ww \in W with w≤b,⋁Dw \leq b,\mathord{\bigvee}D. By (2), there exist u∈Wu \in W and d∈Dd \in D with u≤w,du \leq w,d. Thus d∈↑(W∩↓b)d \in\mathord{\uparrow}(W \cap\mathord{\downarrow}b), as required.

For a dcpo XX and a subset A⊆XA \subseteq X, put

hX(A)=int⁡ΣX(↑XA).(3)h_{X}(A) = \operatorname{int}_{\Sigma X}(\mathord{\uparrow}_{X}A). \tag*{(3)}

We write h=hPh = h_{P} when working in PP. This operation is monotone, h(∅)=∅h(\varnothing) = \varnothing, and h(A)⊆↑Ah(A) \subseteq\mathord{\uparrow}A. The selection result used below applies also to compact sets which need not be saturated or nested.

To choose the points simultaneously, we need neighbourhoods in which finitely many points have a common lower bound. Consider first the coordinatewise order on the finite product PrP^{r}.

Lemma 5.4. Let PP be meet-continuous. For W∈σ(P)W \in\sigma(P) and r≥1r \geq1, the set

Γr(W)={(y1,…,yr)∈Pr:there is w∈W with w≤yi for every i}(4)\Gamma_{r}(W) = \{(y_{1},\ldots,y_{r}) \in P^{r} : \text{there is } w \in W \text{ with } w \leq y_{i} \text{ for every } i\} \tag*{(4)}

is Scott open in PrP^{r}.

Proof. The same lower bound works for any larger tuple, so Γr(W)\Gamma_{r}(W) is upper. Let D⊆PrD \subseteq P^{r} be directed with supremum in Γr(W)\Gamma_{r}(W). For the iith projection πi\pi_{i}, put ti=⋁πi[D]t_{i} = \bigvee\pi_{i}[D]. Directed suprema in PrP^{r} are coordinatewise. Hence there is w0∈Ww_{0} \in W below every tit_{i}. For i=1,…,ri = 1,\ldots,r, apply (2) to wi−1≤tiw_{i-1} \leq t_{i} and πi[D]\pi_{i}[D]. It yields wi∈Ww_{i} \in W and δi∈D\delta_{i} \in D with wi≤wi−1,πi(δi)w_{i} \leq w_{i-1},\pi_{i}(\delta_{i}). Choose a common upper bound δ∈D\delta\in D of the finitely many δi\delta_{i}. The point wrw_{r} lies below every coordinate of δ\delta. Thus δ∈Γr(W)\delta\in\Gamma_{r}(W).

Lemma 5.5. If SS is a dcpo and ΣP\Sigma P is locally compact, the Scott topology on S×PS \times P is the product of the Scott topologies on its factors.

Proof. The projections preserve directed suprema and are monotone, so all open rectangles are Scott open. For the converse let BB be Scott open in S×PS \times P and let (s,t)∈B(s,t) \in B. Its sections Ba={b∈P:(a,b)∈B}B_{a} = \{b \in P : (a,b) \in B\} are Scott open in PP, since b↦(a,b)b \mapsto(a,b) is monotone and preserves directed suprema. Moreover a≤a′a \leq a' implies Ba⊆Ba′B_{a} \subseteq B_{a'}. For every directed A⊆SA \subseteq S,

B⋁A=⋃a∈ABa.B \mathord{\bigvee} A = \bigcup_{a \in A} B_{a}.

Indeed, if b∈B⋁Ab \in B \mathord{\bigvee} A, the directed set {(a,b):a∈A}\{(a,b) : a \in A\} has supremum in BB and therefore meets BB. Upperness gives the opposite inclusion.

By local compactness take t∈V⊆K⊆Bst \in V \subseteq K \subseteq B_s, where VV is open and KK compact saturated. The set R={a∈S:K⊆Ba}R = \{a \in S : K \subseteq B_a\} is upper. If VA∈RV_A \in R, then the sections BaB_a, a∈Aa \in A, cover KK. Finitely many suffice, and directedness gives one member of AA whose section contains all of KK. Thus RR is Scott open, and (s,t)∈R×V⊆B(s,t) \in R \times V \subseteq B.

Lemma 5.6. Under the hypotheses of Theorem 5.17, let y∈W∈σ(P)y \in W \in\sigma(P) and r≥1r \ge1. There is a Scott-open set TT, with y∈T⊆Wy \in T \subseteq W, such that every nonempty subset of TT of size at most rr has a common lower bound in WW.

Proof. Lemmas 5.4 and 5.5 show that Γr(W)\Gamma_r(W) is open in the ordinary product (ΣP)r(\Sigma P)^r, by induction on rr. It contains (y,…,y)(y,\ldots,y). Choose a product neighbourhood T1×⋯×TrT_1 \times\cdots\times T_r of this tuple inside Γr(W)\Gamma_r(W), and put T=W∩⋂i=1rTiT = W \cap\bigcap_{i=1}^r T_i. Every rr-tuple in TrT^r has a common lower bound in WW. Repeating an entry gives the assertion for nonempty subsets with fewer than rr elements.

In the rest of the proof PP satisfies the hypotheses of Theorem 5.17. We first show that if x∈h(E)x \in h(E), then under a suitable restriction on NN one can remove NN from EE and retain xx in the interior of the upward closure.

Lemma 5.7. If E⊆PE \subseteq P and x∈h(E)x \in h(E), there is an open neighbourhood VV of xx with V⊆↑(E∩↓v)V \subseteq\mathord{\uparrow}(E \cap\mathord{\downarrow}v) for every v∈Vv \in V. If x∉h(N)x \notin h(N) for N⊆PN \subseteq P, there is a relatively closed E′⊆EE' \subseteq E such that E′∩N=∅E' \cap N = \varnothing and x∈h(E′)x \in h(E').

Proof. Apply Lemma 5.6 to x∈h(E)x \in h(E) with r=2r = 2, obtaining VV. For any u,v∈Vu,v \in V there is w∈h(E)w \in h(E) below them both. Since w∈↑Ew \in\mathord{\uparrow}E, some e∈Ee \in E satisfies e≤we \leq w. Hence u∈↑(E∩↓v)u \in\mathord{\uparrow}(E \cap\mathord{\downarrow}v), proving the first assertion. If x∉h(N)x \notin h(N), the neighbourhood VV contains a point v∉↑Nv \notin\mathord{\uparrow}N. The set E′=E∩↓vE' = E \cap\mathord{\downarrow}v is relatively closed by Lemma 5.2, is disjoint from NN, and satisfies x∈V⊆h(E′)x \in V \subseteq h(E').

Lemma 5.8. The operation hh preserves finite unions.

Proof. For x∈h(A∪B)∖h(A)x \in h(A \cup B) \setminus h(A), apply Lemma 5.7 with E=A∪BE = A \cup B and N=AN = A. The resulting E′E' is contained in BB, so x∈h(B)x \in h(B). This proves h(A∪B)⊆h(A)∪h(B)h(A \cup B) \subseteq h(A) \cup h(B). Monotonicity gives equality. Induction and h(∅)=∅h(\varnothing) = \varnothing give the general assertion.

Lemma 5.9. Let DD be a nonempty Scott-compact set and U\mathcal{U} a relatively open cover of DD. Then

h(D)=⋃{h(C):C is closed in D, C⊆U for some U∈U}.(5)h(D) = \bigcup\{h(C) : C\text{ is closed in }D,\ C \subseteq U\text{ for some }U \in\mathcal{U}\}. \tag*{(5)}

Proof. Monotonicity puts the right-hand side in h(D)h(D). Suppose some x∈h(D)x \in h(D) does not belong to that union. Take a finite subcover U1,…,UmU_1,\ldots,U_m and put Aj=D∖UjA_j = D \setminus U_j. If BB is closed in DD and x∈h(B)x \in h(B), we claim x∈h(B∩Aj)x \in h(B \cap A_j) for every jj. Otherwise Lemma 5.7 supplies a relatively closed B′⊆BB' \subseteq B disjoint from B∩AjB \cap A_j with x∈h(B′)x \in h(B'). But B′B' is closed in DD and contained in UjU_j, against the choice of xx. Starting from B=DB = D and intersecting successively with the closed sets A1,…,AmA_1,\ldots,A_m would give x∈h(⋂jAj)=h(∅)x \in h(\bigcap_j A_j) = h(\varnothing). This proves the identity.

In particular, if a compact set GG lies in h(D)h(D), finitely many of the open sets h(C)h(C) in (5.5) cover GG. The preceding identity concerns an open cover of h(D)h(D) in PP.

We next apply the refinement lemma successively to a finite list of compact sets. It is important that each new set is closed relative to the original set in the same position.

Lemma 5.10. Let r≥1r \ge1, m≥0m \ge0, and let L0,…,LmL_0,\ldots,L_m be nonempty Scott-compact sets with Li+1⊆h(Li)L_{i+1} \subseteq h(L_i) for i<mi < m. For 0≤j≤m0 \leq j \leq m, denote by Dj\mathcal{D}_j the family of lists (D0,…,Dj)(D_0,\ldots,D_j) with the following properties.

  1. D0=L0D_0 = L_0 and each DiD_i is a nonempty closed subset of LiL_i.

  1. Di+1⊆h(Di)D_{i+1} \subseteq h(D_i) for all i<ji < j.

  1. For i<ji<j, each nonempty subset of Di+1D_{i+1} of size at most rr has a common lower bound in DiD_i.

For every j≤mj\le m one has

h(Lj)=⋃(D0,…,Dj)∈Djh(Dj).(6)h(L_j)=\bigcup_{(D_0,\ldots,D_j)\in\mathcal{D}_j}h(D_j). \tag*{(6)}

In particular, the last set in the list may be chosen so that z∈h(Dm)z\in h(D_m) for any prescribed z∈h(Lm)z\in h(L_m).

Proof. For j=0j=0 take D0=L0D_0=L_0. Assume the assertion at j<mj<m. For each y∈Lj+1⊆h(Lj)y\in L_{j+1}\subseteq h(L_j), fix a list in Dj\mathcal{D}_j such that y∈h(Dj)y\in h(D_j). Lemma 5.6 gives a neighbourhood Ty⊆h(Dj)T_y\subseteq h(D_j) of yy in which each nonempty collection of at most rr points has a lower bound in h(Dj)h(D_j). That lower bound dominates an element of DjD_j, which serves as a lower bound in DjD_j.

The traces of the TyT_y cover Lj+1L_{j+1}. For any z∈h(Lj+1)z\in h(L_{j+1}), Lemma 5.9 yields a closed subset Dj+1D_{j+1} of Lj+1L_{j+1}, contained in some TyT_y, with z∈h(Dj+1)z\in h(D_{j+1}). In particular Dj+1D_{j+1} is nonempty. Appending it to the list chosen for that yy satisfies all three requirements. This proves one inclusion in (5.6) at j+1j+1. The other follows from Dj+1⊆Lj+1D_{j+1}\subseteq L_{j+1}. Induction completes the proof. □

We shall use the following compactness argument to pass from the finite lists to an increasing sequence.

Lemma 5.11. Let each KiK_i be a nonempty Scott-compact set. For 0≤i≤k0\le i\le k let Fi(k)F_i^{(k)} be a nonempty closed subset of KiK_i such that Fi(k+1)⊆Fi(k)F_i^{(k+1)}\subseteq F_i^{(k)}. Assume that for i<ki<k every nonempty subset of Fi+1(k)F_{i+1}^{(k)} of size at most kk has a lower bound in Fi(k)F_i^{(k)}. Put Ei=⋂k≥iFi(k)E_i=\bigcap_{k\ge i}F_i^{(k)}. Each EiE_i is nonempty, and one can choose pi∈Eip_i\in E_i below every point of Ei+1E_{i+1}. Any such choices satisfy pi∈Kip_i\in K_i and pi≤pi+1p_i\le p_{i+1}.

Proof. For fixed ii the nested nonempty closed sets Fi(k)F_i^{(k)} have nonempty intersection by compactness of KiK_i. Consider the closed subsets of KiK_i in the family

{Fi(k):k≥i}∪{Ki∩↓y:y∈Ei+1}.\{F_i^{(k)}: k\ge i\}\cup\{K_i\cap\downarrow y:y\in E_{i+1}\}.

Take a finite subfamily, and let BB consist of the finitely many points yy specified by it. Choose an integer kk at least i+1i+1, at least ∣B∣|B|, and at least every stage index in the subfamily. As B⊆Fi+1(k)B\subseteq F_{i+1}^{(k)}, some a∈Fi(k)a\in F_i^{(k)} is below BB if BB is nonempty. For B=∅B=\varnothing, choose any a∈Fi(k)a\in F_i^{(k)}. In either case aa belongs to all the selected sets. This is the finite intersection property, so compactness gives Ei∩⋂y∈Ei+1↓y≠∅E_i\cap\bigcap_{y\in E_{i+1}}\downarrow y\ne\varnothing. Choose pip_i from this intersection for each ii. Then pi+1∈Ei+1p_{i+1}\in E_{i+1} implies pi≤pi+1p_i\le p_{i+1}. □

Theorem 5.12. Let PP be meet-continuous and ΣP\Sigma P locally compact. Suppose that (Kn)n∈N0(K_n)_{n\in\mathbb{N}_0} is a sequence of nonempty Scott-compact subsets of PP such that Kn+1⊆h(Kn)K_{n+1}\subseteq h(K_n) for all nn. There are points pn∈Knp_n\in K_n such that pn≤pn+1p_n\le p_{n+1} for every nn.

Proof. We construct a rooted tree with a finite nonempty set of nodes at each level kk. A node there carries a list (F0,…,Fk)(F_0,\ldots,F_k) of nonempty sets closed in K0,…,KkK_0,\ldots,K_k, respectively. We require the following.

  1. For each node and i<ki<k, Fi+1⊆h(Fi)F_{i+1}\subseteq h(F_i).

  1. For each node and i<ki<k, every nonempty subset of Fi+1F_{i+1} of size at most kk has a common lower bound in FiF_i.

  1. Over all nodes at level kk, the sets h(Fk)h(F_k) cover Kk+1K_{k+1}.

A child’s first k+1k+1 coordinates will be contained in the corresponding coordinates of its parent. The root (K0)(K_0) has all the required properties since K1⊆h(K0)K_1 \subseteq h(K_0).

Suppose level kk has been constructed. To each list at this level apply Lemma 5.10 with r=k+1r=k+1. All its refined lists (D0,…,Dk)(D_0,\ldots,D_k) retain a record of their parent. Their last open sets h(Dk)h(D_k) cover the parent’s h(Fk)h(F_k). Over all parents these opens therefore cover Kk+1K_{k+1}. For each y∈Kk+1y\in K_{k+1}, choose such a list with y∈h(Dk)y\in h(D_k). Lemma 5.6 gives a neighbourhood Ty⊆h(Dk)T_y\subseteq h(D_k) of yy such that every nonempty subset of TyT_y of size at most k+1k+1 has a lower bound in h(Dk)h(D_k), hence one in DkD_k.

Apply Lemma 5.9 to Kk+1K_{k+1} and the relative cover by the Ty∩Kk+1T_y\cap K_{k+1}. For each g∈h(Kk+1)g\in h(K_{k+1}), there is a nonempty relatively closed C⊆Kk+1C\subseteq K_{k+1} contained in some TyT_y, with g∈h(C)g\in h(C). Adjoin CC to the particular list chosen for this yy. Such a list satisfies the first two conditions at level k+1k+1. Its old coordinates are closed in their respective KiK_i by transitivity of relative closedness. The resulting h(C)h(C) cover h(Kk+1)h(K_{k+1}) and hence Kk+2K_{k+2}. By compactness, finitely many suffice to cover Kk+2K_{k+2}. Keep the associated nodes and their parent edges. This gives a finite nonempty level k+1k+1 satisfying the third condition.

The tree has an infinite branch. Indeed, the root has descendants at arbitrarily large levels, and a node with this property has a child with the same property because the tree is finitely branching. Write (F0(k),…,Fk(k))(F_0^{(k)},\ldots,F_k^{(k)}) for the lists on this branch. The parent relation gives nesting in each coordinate, and the second condition gives the finite lower bounds required in Lemma 5.11. Its conclusion yields the desired points pi∈Kip_i\in K_i. □

We continue to assume that PP is meet-continuous and ΣP\Sigma P is locally compact. When FF is Scott closed, write hF(A)=int⁡F(↑FA)h_F(A)=\operatorname{int}_F(\mathord{\uparrow}_F A) for A⊆FA\subseteq F, using the induced order on FF.

Lemma 5.13. Let FF be a nonempty Scott-closed subset of a dcpo PP. Its induced order is a dcpo and its Scott topology is the subspace topology. Local compactness of ΣP\Sigma P and meet-continuity of PP each pass to FF.

Proof. For the first two assertions apply Lemma 2.3. Let x∈U∩Fx\in U\cap F, where UU is Scott open in PP. If ΣP\Sigma P is locally compact, choose an open VV and a compact saturated KK with x∈V⊆K⊆Ux\in V\subseteq K\subseteq U. Then F∩KF\cap K is closed in KK, hence compact, and is upper in FF. It contains the relative open neighbourhood F∩VF\cap V of xx and is contained in F∩UF\cap U. For meet-continuity, let D⊆FD\subseteq F be directed with x≤F⋁PD=⋁PDx\leq_F \bigvee_P D=\bigvee_P D. Meet-continuity of PP gives u∈Uu\in U below both xx and some d∈Dd\in D. Lowerness of FF puts uu in U∩FU\cap F, as required. □

Proposition 5.14. Let FF be a nonempty irreducible Scott-closed subset of PP. For every sequence (On)n∈N0(O_n)_{n\in\mathbb{N}_0} of nonempty relatively open subsets of FF, there is an increasing sequence (pn)(p_n) in FF with pn∈Onp_n\in O_n for each nn. In particular, ⋁npn∈⋂nOn\bigvee_n p_n\in\bigcap_n O_n.

Proof. Work in ΣF\Sigma F by Lemma 5.13. First choose a compact saturated K0⊆O0K_0\subseteq O_0 with nonempty interior. If KnK_n has been chosen, irreducibility gives a point of On+1∩int⁡FKnO_{n+1}\cap\operatorname{int}_F K_n. Choose a compact saturated neighbourhood Kn+1K_{n+1} of this point within the intersection. Thus Ki⊆OiK_i\subseteq O_i and Ki+1⊆int⁡FKi=hF(Ki)K_{i+1}\subseteq\operatorname{int}_F K_i=h_F(K_i). Theorem 5.12 gives pi∈Kip_i\in K_i with pi≤pi+1p_i\leq p_{i+1}. Each OiO_i is upper in FF, so the supremum belongs to every OiO_i. □

Corollary 5.15. If PP is meet-continuous and ΣP\Sigma P is locally compact and first countable, then ΣP\Sigma P is sober.

Proof. Let a,ba,b belong to a nonempty irreducible Scott-closed set FF. Restrict countable open neighbourhood bases at a,ba,b to FF and denote the resulting bases by (Un),(Vn)(U_n),(V_n). Each Un∩VnU_n\cap V_n is nonempty by irreducibility. Proposition 5.14 gives a point cc in their intersection for all nn. For any open local base at aa the intersection of its members is ↑Fa\uparrow_F a. Indeed, if a≰za\nleq z then F∖↓zF\setminus\downarrow z is a neighbourhood of aa omitting zz. The same applies to bb, so a,b≤ca,b\leq c. Thus FF is directed, contains its supremum, and is the principal ideal of that supremum. □

Lemma 5.16. Let FF be nonempty, irreducible and Scott closed in PP. Given nonempty relatively open subsets HnH_n of FF, n∈N0n \in\mathbb{N}_0, some b∈⋂nHnb \in\bigcap_n H_n has a countable open neighbourhood base in FF.

Proof. Put W−1=FW_{-1}=F. Since Wn−1∩HnW_{n-1} \cap H_n is nonempty by irreducibility, local compactness gives a nonempty relatively open WnW_n and a compact saturated QnQ_n in FF with

Wn⊆Qn⊆Wn−1∩Hn.W_n \subseteq Q_n \subseteq W_{n-1} \cap H_n.

In particular Qn+1⊆Wn⊆QnQ_{n+1} \subseteq W_n \subseteq Q_n. Among the closed subsets of FF meeting every QnQ_n choose an inclusion-minimal one, BB. Such sets exist because FF is one. For a nonempty chain of them and fixed nn, the traces on QnQ_n are closed and have the finite intersection property. Compactness gives a common point in QnQ_n. Thus the intersection of the chain meets each QnQ_n, and Zorn's lemma applies.

Whenever an open OO of FF meets BB, the proper closed subset B∖OB \setminus O of BB is closed in FF. Hence for some jj we have

∅≠B∩Qj⊆O.\varnothing\ne B \cap Q_j \subseteq O.

For two such open sets take the larger of the two indices. This shows that their intersection meets BB, so BB is irreducible. It is also closed in PP. Each B∩WnB \cap W_n is a nonempty open subset of BB, as it contains B∩Qn+1B \cap Q_{n+1}. Proposition 5.14 applied to BB gives b∈B∩⋂nWnb \in B \cap\bigcap_n W_n. Thus b∈Qn∩Hnb \in Q_n \cap H_n for every nn. Since ↓Fb\mathord{\downarrow}_F b is closed, lies in BB and meets every QnQ_n, minimality implies

B=↓Fb.B=\mathord{\downarrow}_F b.

Now put Vn=↑F(Wn∩↓Fb)V_n=\mathord{\uparrow}_F(W_n \cap\mathord{\downarrow}_F b). By Lemma 5.3 in FF, these are open neighbourhoods of bb. If OO is any open neighbourhood of bb, take jj as above. Then Wj∩↓Fb⊆Qj∩B⊆OW_j \cap\mathord{\downarrow}_F b \subseteq Q_j \cap B \subseteq O. Since OO is upper, Vj⊆OV_j \subseteq O. This proves the base assertion. In particular ↑Fb=⋂nVn\mathord{\uparrow}_F b=\bigcap_n V_n is a GδG_\delta subset of FF.

Theorem 5.17. Let PP be a meet-continuous locally compact dcpo with respect to the Scott topology. Then ΣP\Sigma P is sober.

Proof. Fix a nonempty irreducible Scott-closed set FF. Let DD be the set of points of FF having a countable open neighbourhood base in FF. By applying Lemma 5.16 to a constant sequence of nonempty open sets, we see that DD is dense in FF. Given a,b∈Da,b \in D, interleave the members of countable bases at aa and bb. The same lemma gives c∈Dc \in D belonging to every one of these open sets. The neighbourhood calculation in Corollary 5.15 gives a,b≤ca,b \leq c. Thus DD is nonempty and directed.

Let p=⋁PDp=\bigvee_P D. Since FF is Scott closed, p∈Fp \in F. The closed set F∩↓PpF \cap\mathord{\downarrow}_P p contains the dense subset DD, and therefore contains FF. Conversely ↓Pp⊆F\mathord{\downarrow}_P p \subseteq F by lowerness. We have proved F=↓Pp=cl⁡ΣP{p}F=\mathord{\downarrow}_P p=\operatorname{cl}_{\Sigma P}\{p\}. If F=↓PqF=\mathord{\downarrow}_P q as well, then p≤qp \leq q and q≤pq \leq p, so p=qp=q.

Corollary 5.18. For every p∈Pp \in P and countable family of Scott-open neighbourhoods (Un)(U_n) of pp, there is a point b≤pb \leq p in ⋂nUn\bigcap_n U_n at which ΣP\Sigma P is first countable. Moreover, the set

Dp={b≤p:ΣP has a countable local base at b}D_p=\{b \leq p:\Sigma P\text{ has a countable local base at }b\}

is directed and has supremum pp.

Proof. Take F=↓PpF=\mathord{\downarrow}_P p. Every nonempty relative open contains pp, so FF is irreducible. By Lemma 5.16, some b∈F∩⋂nUnb \in F \cap\bigcap_n U_n has a relative countable base (Vj)(V_j). Write Vj=F∩GjV_j=F \cap G_j with GjG_j Scott open in PP. Because ↓Pb⊆F\mathord{\downarrow}_P b \subseteq F, we have Gj∩↓Pb=Vj∩↓PbG_j \cap\mathord{\downarrow}_P b=V_j \cap\mathord{\downarrow}_P b. The sets ↑P(Gj∩↓Pb)\mathord{\uparrow}_P(G_j \cap\mathord{\downarrow}_P b) are open neighbourhoods of bb by Lemma 5.3. For any ambient open OO containing bb, choose jj with Vj⊆O∩FV_j \subseteq O \cap F. Upperness of OO now gives ↑P(Gj∩↓Pb)⊆O\mathord{\uparrow}_P(G_j \cap\mathord{\downarrow}_P b) \subseteq O. This proves first countability at bb in ΣP\Sigma P.

Taking all UnU_n equal shows that DpD_p meets every neighbourhood of pp. For a,b∈Dpa,b \in D_p, every Scott-open neighbourhood of either point contains pp. Apply the first assertion to the interleaved local bases at aa and bb to find a common upper bound in DpD_p. Hence DpD_p is directed, with q=⋁Dp≤pq=\bigvee D_p \leq p. If p≰qp \not\leq q, the open P∖↓PqP \setminus\mathord{\downarrow}_P q would contain pp and miss DpD_p, a contradiction. Thus q=pq=p.

This is the approximation property obtained for locally compact sober web spaces in [2, Lemma 3.7.

The following two examples separate the hypotheses of Theorem 5.17 from first countability and coherence.

Example 5.19. Let ω1\omega_1 be the least uncountable ordinal, and order P=[0,ω1]P = [0,\omega_1] by the ordinal order. This chain is a dcpo. Suppose x≤⋁Dx \leq\bigvee D for a directed set DD. If some d∈Dd \in D dominates xx, then xx is itself below xx and dd. Otherwise D⊆↓xD \subseteq\mathord{\downarrow}x and ⋁D=x\bigvee D = x. Every Scott-open neighbourhood of xx then contains some member of DD. These two cases prove meet-continuity.

Every nonempty upper set has a least member α\alpha. If α\alpha is a nonzero limit ordinal, its predecessors form a directed set with supremum α\alpha disjoint from ↑α\mathord{\uparrow}\alpha. If α=β+1\alpha= \beta+ 1, a directed set missing ↑α\mathord{\uparrow}\alpha has supremum at most β\beta. Consequently the nonempty Scott opens are precisely PP and the upper sets generated by successor ordinals. Each is compact because any open cover has a member containing its least point. Thus ΣP\Sigma P is locally compact.

Every neighbourhood of ω1\omega_1 contains a tail (β,ω1](\beta,\omega_1] with β<ω1\beta< \omega_1. Given countably many neighbourhoods, choose one such βn\beta_n for each. The ordinal (⋁nβn)+1(\bigvee_n \beta_n) + 1 is still countable and belongs to all of them. Since the intersection of all neighbourhoods of ω1\omega_1 is {ω1}\{\omega_1\}, the top point has no countable local base. Its Scott space is sober by Theorem 5.17.

Example 5.20. Take two incomparable points a,ba,b below an infinite antichain MM. More precisely, order P={a,b}∪MP = \{a,b\} \cup M by a<ma < m and b<mb < m for m∈Mm \in M, with no other strict inequalities. If a directed set meets MM, the point it contains there is greatest. A directed set disjoint from MM is a singleton. Hence PP is a dcpo, every upper set is Scott open, and (2) holds by using xx itself whenever x≤max⁡Dx \leq\max D. The principal upper sets are compact open neighbourhoods of their generators. The Scott space is therefore locally compact. But ↑a∩↑b=M\mathord{\uparrow}a \cap\mathord{\uparrow}b = M, and the cover of MM by Scott-open singletons has no finite subcover. In particular this Scott space is not coherent. See also [12 Section 1.5].

Chen’s example is a meet-continuous dcpo whose Scott space is compact, second countable and core-compact, but fails both local compactness and sobriety [1 Theorem 6.4]. Thus core-compactness cannot replace local compactness in Theorem 5.17.

AI Declaration For more readable, the authors just use AI tools to assist with English language editing. The author independently verified the mathematical statements, proofs and references. The author takes full responsibility for the manuscript.

References

  1. [1]Y. Chen, A non-sober core-compact meet-continuous dcpo, preprint, arXiv:2610.22033 2026.arxiv.org/abs/2610.22033
  2. [6]Z. He, Z. Dong and K. Wang, Sobriety of Scott topologies under countability conditions, preprint, arXiv:2609.24001v2, 2026.arxiv.org/abs/2609.24001
  3. [7]Z. He and Y. Zhang, The sobriety for locally compact dcpos, preprint, arXiv:2610.17759, 2026.arxiv.org/abs/2610.17759
  4. [9]H. Miao, Q. Li and D. Zhao, One-step closure, weak one-step closure and meet continuity, Electronic Notes in Theoretical Informatics and Computer Science, 2 (2023), article 8, 8 pp.DOI
  5. [10]H. Miao, X. Xi, Q. Li and D. Zhao, Not every countable complete distributive lattice is sober, Mathematical Structures in Computer Science, 33 (2023), no. 9, 809–831.DOI
  6. [11]H. Miao, Z. Yuan and Q. Li, A discussion of well-filteredness and sobriety, Topology and its Applications, 291 (2021), 107450.DOI
  7. [12]R. Tix, K. Keimel and G. Plotkin, Semantic domains for combining probability and non-determinism, Electronic Notes in Theoretical Computer Science, 222 (2009), 3–99.
  8. [13]X. Xu and D. Zhao, Some open problems on well-filtered spaces and sober spaces, Topology and its Applications, 301 (2021), 107540.arxiv.org/abs/2005.00539

Paper details

vlnc:W_g3r4rg63
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