BeLuAGT.dvi @ Applied General Topology c© Universidad Politécnica de Valencia Volume 9, No. 2, 2008 pp. 281-292 Scott-representability of some spaces of Tall and Mǐskin Harold Bennett and David Lutzer ∗ Abstract. In this paper we show that a variation of a technique of Mǐskin and Tall yields a cocompact completely regular Moore space that is Scott-domain-representable and has a closed Gδ-subspace that is not Scott-domain-representable. This clarifies the general topology of Scott-domain-representable spaces and raises additional questions about Scott-domain representability in Moore spaces. 2000 AMS Classification: Primary 54E30; Secondary 54D70,06B35, 06F30, 54H12, 54D20 Keywords: domain, Scott-domain, Scott-domain-representable space, Moore space, complete Moore space, cocompact, Čech-complete, subcompact, Cho- quet complete. 1. Introduction A domain is a continuous poset (P, ⊑) in which each non-empty directed subset has a supremum. A Scott domain is a domain in which each nonempty bounded set has a supremum. (For more details, see Section 2.) Representing mathematical objects as the set of maximal elements of a domain or of a Scott domain is an idea that originated in theoretical computer science. Every domain carries a natural topology, called the Scott topology, and a topological space is said to be domain representable (respectively, Scott-domain- representable) if it is homeomorphic to the set of maximal elements of a do- main (respectively, a Scott domain) with the relative Scott topology. In recent years, topologists have come to see domain representability and Scott-domain representability as strong completeness properties associated with the Baire category theorem. For example, every subcompact regular space is domain- representable [4] and every domain-representable space is Choquet complete [8], and therefore a Baire space. (See Section 2 for definitions.) ∗Corresponding author. 282 H. Bennett and D. Lutzer The basic general topology of domain-representable spaces is fairly well un- derstood. For example, while domain-representability is an open-hereditary property, it is not closed-hereditary (because if X is any completely regular space that is not domain-representable, then the space obtained from βX by isolating all points of βX − X is domain-representable [3] and contains X as a closed subspace). Similarly, Scott-domain-representability is open-hereditary and not closed-hereditary (as can be seen by applying the same βX construc- tion described above). Further, any Gδ-subspace of a domain-representable space is domain-representable, as shown in [3], so it is natural to ask whether Gδ-subspaces of Scott-domain-representable spaces inherit Scott-domain rep- resentability. Among metrizable spaces, the answer is “Yes,” because if X is a Scott-domain representable metric space, then X is completely metrizable. Let Y be a Gδ-subset of X. Then Y is also completely metrizable so that a recent result of Kopperman, Kunzi, and Waszkiewicz [7] shows that Y is Scott-domain representable. The first goal of this paper is to show that, without metrizabil- ity, Scott-domain-representability is not inherited by (closed) Gδ-subspaces. Furthermore, our example is a Moore space, a particularly nicely-behaved type of generalized metric space. It was already known that the equivalence among metric spaces of essen- tially all strong completeness properties (complete metrizability, Scott-domain- representability, Čech-completeness, cocompactness, subcompactness, and domain-representability) breaks down outside of the metric space category. But there is still a rich theory of completeness in the wider class of Moore spaces, and results due to K. Martin, Tall, Rudin, Bennett, Lutzer, and Reed show that • among Moore spaces, domain-representability is equivalent to subcom- pactness [4] and is equivalent to Rudin-completeness [2] which is strictly weaker than Moore-completeness [6]; • for completely regular Moore spaces, Moore completeness is equivalent to Čech-completeness [2]; • there is a completely regular Moore space that is Čech-complete but not cocompact [11] and not Scott-domain-representable [5]; • if a Moore space is Scott-domain-representable, then it is completely regular and Moore-complete, Čech complete [8], and cocompact [7]; Additional equivalents of domain-representability among Moore spaces that involve the strong Choquet game are given in [5]. The second goal of this paper is to explore the role of Scott-domain representability in the class of completely regular Moore spaces and we show that a certain Moore space X0 (due to Mǐskin [10]) is Scott-domain-representable and contains a closed Gδ- subspace Z (due to Tall [11]) that is not Scott-domain representable. This example raises a natural question about completeness and representability in Moore spaces, namely: Question 1.1. Is Scott-domain-representability equivalent to cocompactness among completely regular Moore spaces? Scott-representability of some spaces of Tall and Miškin 283 Kopperman, Kunzi, and Waszkiewicz [7] have characterized Scott-domain- representability in any completely regular space as being a combination of cocompactness and a bi-topological condition (“pairwise complete regularity”), but is not yet clear how to apply their characterization in the Moore space con- text. A natural place to look for counterexamples to Question 1.1 is in Mǐskin’s construction of a cocompact Moore space, mentioned above. In Section 3 we show that some of Mǐskin’s spaces are Scott-domain-representable, but we do not know the answer to the following: Question 1.2. Is it true that each of Mǐskin’s spaces in [10] is Scott-domain- representable? In this paper we show that a certain Čech-complete Moore space constructed by Tall embeds as a closed subspace of a Scott-domain representable Moore space. To what extent is this a general phenomenon? More precisely, we have: Question 1.3. Does each completely regular, Čech-complete Moore space X embed in a Moore space Y (X) that is Scott-domain-representable? What if X is required to be a dense subspace of Y (X)? What if X is required to be a closed subspace? Basic definitions appear in Section 2. Section 3 gives the basic constructions due to Tall and Mǐskin, and shows that, with some additional restrictions, one of Mǐskin’s spaces is Scott-domain-representable and has a closed Gδ-subspace that is not. Throughout the paper, we reserve the symbols R, Q, and P for the usual sets of real, rational, and irrational numbers. 2. Basic definitions A space X is cocompact if it is T1 and has a collection C of closed subsets with the following two properties: a) if D is a centered 1 subcollection of C, then ⋂ D 6= ∅; b) if U is an open subset of X and x ∈ U , then some C ∈ C has x ∈ Int(C) ⊆ C ⊆ U . Note that the members of C might not be the closures of their interiors, even when the interiors are non-void. If one insists that members of C are the closures of their interiors, i.e., are regularly-closed sets, then one obtains a different notion called regular cocompactness. The Sorgenfrey line, for example, is cocompact but not regularly cocompact [2]. Cocompactness was introduced by de Groot and his colleagues [1]. Another strong completeness first studied by the Amsterdam school is subcompactness, where we say that a space X is subcompact if X has a base B with the property that ⋂ F 6= ∅ whenever F ⊆ B has the property that if B1, B2 ∈ F, then some B3 ∈ F has cl(B3) ⊆ B1 ∩ B2. 1A collection D is centered if ⋂ {Di : i ≤ n} 6= ∅ whenever {Di : i ≤ n} is a finite subcollection of D. 284 H. Bennett and D. Lutzer To define domain-representability and Scott-domain-representability, we be- gin with a poset (S, ⊑). A subset E ⊆ S is directed if for each e1, e2 ∈ E some e3 ∈ E has e1, e2 ⊑ e3. If sup(E) ∈ S whenever E is a nonempty directed sub- set of S, then S is a dcpo (“directed-complete partial order”). Given a, b ∈ S, we write a ≪ b to mean that whenever E ⊆ S is a directed set with b ⊑ sup(E), then some e ∈ E has a ⊑ e. The set ⇓(b) is defined to be {a ∈ S : a ≪ b}. In case ⇓(b) is directed and has b as its supremum for each b ∈ S, we say that S is continuous. If S is a continuous dcpo, then we say that S is a domain. If the domain S has the additional property that every nonempty bounded subset of S has a supremum in S, then we say that S is a Scott domain. Among domains, Scott domains are easily characterized: Lemma 2.1. A domain (S, ⊑) is a Scott domain if and only if sup({a, b}) exists whenever a, b ∈ S and a, b ⊑ c for some c ∈ S. Proof. To prove the nontrivial half of the lemma, suppose E is a nonempty bounded subset of S. Let f ∈ S be an upper bound for E. If e1, e2, e3 ∈ E, then sup({e1, e2}) ∈ S and f is an upper bound for {sup({e1, e2}), e3} in S so that sup(sup(e1, e2), e3) ∈ S. It is easy to show that sup(sup(e1, e2), e3) = sup(sup(ei, ej ), ek) for each permutation i, j, k of 1, 2, 3, so that the supremum of each three-element subset of the bounded set E is well-defined. Similarly, sup(F ) is a well-defined point of S for each non-empty finite subset F ⊆ E. Now let D := {sup(F ) : ∅ 6= F ⊆ E and |F | < ω}. Then D is a directed subset of S so that, S being a domain, sup(D) ∈ S. Clearly sup(D) = sup(E) as required. � Every poset (S, ⊑) can be endowed with a special topology called the Scott topology in which a set U is open if and only if it satisfies both (i) if x ⊑ y and x ∈ U , then y ∈ U , and (ii) if E ⊆ S is a nonempty directed set with sup(E) ∈ U , then E ∩ U 6= ∅. In a domain S, the collection of all sets ⇑(a) := {b ∈ S : a ≪ b} is a base for the Scott topology on S. The set of maximal elements of a domain S is denoted by max(S). If a topological space X is homeomorphic to the subspace max(S) of some domain S with the relative Scott topology, then we say that X is domain-representable. If S is a Scott-domain and X is homeomorphic to max(S), then we say that X is Scott-domain-representable. Kopperman, Kunzi, and Waszkiewicz [7] have characterized Scott-domain- representable spaces as being the cocompact spaces that also satisfy a certain bi-topological condition. A short, direct proof of cocompactness of any Scott- domain-representable space is possible and we give it here. A central tool is the following Interpolation Lemma [9]. Lemma 2.2. Suppose a ≪ c in a domain S. Then some b ∈ S has a ≪ b ≪ c. Lemma 2.3. Let S be a Scott domain. For each p ∈ S, let ↑(p) = {q ∈ S : p ⊑ q}. Then each set ↑(p) ∩ max(S) is a relatively closed subset of max(S). Proof. Suppose that x ∈ max(S) is a limit point of ↑(p) ∩ max(S). Then for each q ≪ x, ⇑(q) ∩ ↑(p) 6= ∅. Consequently p and q have a common extension, Scott-representability of some spaces of Tall and Miškin 285 so that r(p, q) := sup{p, q} is in S. Let E := {r(p, q) : q ≪ x}. We claim that E is a directed set. For suppose that r(p, q1), r(p, q2) ∈ E. Because ⇓(x) is directed, some q3 ∈ ⇓(x) has q1, q2 ⊑ q3. Then r(p, q3) ∈ E and r(p, qi) ⊑ r(p, q3) for i = 1, 2. Because S is a dcpo, sup(E) ∈ S so that some z ∈ max(S) has sup(E) ⊑ z. Recall that as a subspace of S, max(S) is a T1- space. Therefore, if z 6= x, then some q4 ≪ x has z 6∈ ⇑(q4). Because q4 ≪ x, Lemma 2.2 gives q5 ∈ S with q4 ≪ q5 ≪ x. But q5 ≪ x forces r(p, q5) ∈ E so that q4 ≪ q5 ⊑ r(p, q5) ⊑ sup(E) ⊑ z. Therefore z ∈ ⇑(q5) ⊆ ⇑(q4), contrary to our choice of q4. Therefore, p ⊑ sup(E) = z = x showing that x ∈ ↑(p) as required. � Our next result appears in [7]. We present an easy direct proof. Corollary 2.4. Suppose S is a Scott domain. Then the subspace of maximal elements of S is cocompact. Proof. First, the subspace max(S) of S is T1. Second, let C = {max(S) ∩ ↑(p) : p ∈ S}. In the light of Lemma 2.3, each member of C is a closed subset of max(S). To verify the first part of the cocompactness definition, suppose that D ⊆ C is a centered collection. Write D = {max(S) ∩ ↑(a) : a ∈ A}. Then, given any finite set F := {a1, · · · , ak} ⊆ A we know that ↑(a1) ∩ · · · ∩ ↑(ak) 6= ∅ because D is centered, so that sup(F ) ∈ S by Lemma 2.1. Let  := {sup(F ) : ∅ 6= F ⊆ A and |F | < ω}. Then  is directed, so sup(Â) ∈ S, say sup(Â) = b ∈ S. Then ∅ 6= ↑(b) ∩ max(S) ⊆ ⋂ D, as required. To verify the second part of the definition of cocompactness, it is enough to consider a point x in a basic open set max(S)∩⇑(q). The Interpolation Lemma 2.2 provides a point p ∈ S with q ≪ p ≪ x. Then ⇑(p) is a neighborhood of x with ⇑(p) ⊆ ↑(p) ⊆ ⇑(q) so that x is in the relative interior of ↑(p) ∩ max(S) which is contained in the closed set ↑(p)∩max(S) ⊆ max(S)∩⇑(q), as required to show that max(S) is cocompact. � 3. A variation of spaces of Tall and Mǐskin Tall and Mǐskin began their constructions with a countable subset of the plane that had uncountably many limit points on the x-axis. We need more control and so we replace that countable set by a binary tree T with ω-many levels and use its branch space Y in place of the limit points on the x-axis. This tree may be embedded in the upper half plane in such a way that its branch space corresponds in a natural way to an uncountable set (the Cantor set) on the x-axis. Therefore, the space we will construct is one of the spaces due to Mǐskin. Description of the space X and the subspace X0: a) The tree T : Let T be a binary tree with ω-many levels. Denote the unique minimal element of T by 0̄. The level of any d ∈ T in our tree is denoted by lv(d) and T (n) = {d ∈ T : lv(d) = n} so that T = ⋃ {T (n) : 0 ≤ n < ω}. 286 H. Bennett and D. Lutzer b) The branches of T : Let Y be the set of all branches of T , i.e., each y ∈ Y is a maximal linearly ordered subset of T . We let e(y, n) denote the unique element of the branch y that lies at level n of the tree T . Thus, for example, e(y, 0) = 0̄ for each y ∈ Y and if y1, y2 ∈ Y have e(y1, n) = e(y2, n), then e(y1, k) = e(y2, k) for each 0 ≤ k ≤ n. c) The space X: Let T ∗ := {(d, S) : d ∈ T, ∅ 6= S ⊆ Y }. The underlying set of our space is X = T ∗ ∪Y and the set X is topologized by isolating each point of T ∗ and by using the sets N (y, n) = {y} ∪ {(d, S) ∈ T ∗ : lv(d) ≥ n, d ∈ y, and y ∈ S} as basic neighborhoods of y ∈ Y . Equivalently, N (y, n) = {y} ∪ {(e(y, k), S) ∈ T ∗ : n ≤ k < ω and y ∈ S}. d) The subspace X0 = X − {(0̄, Y )}. This is a closed and open subspace of X and will be the space used in our example. However, X0 is not needed until the very end of Section 3. As in [10], the space X is a cocompact, Čech-complete, completely regular Moore space and therefore so is its closed and open subspace X0. The subspace X0 contains a closed (and hence Gδ) -subspace Z := {(d, S) ∈ T ∗ : (d, S) 6= (0̄, Y ), |S| < ω} ∪ Y that is homeomorphic to one of the spaces constructed by Tall in [11]. Consequently, Z is not cocompact and therefore is not Scott- domain-representable (by Corollary 2.4). What remains is to prove that X0 is Scott-domain representable. Our first step is to define the Scott domain that comes very close to representing X, and then to work around the “almost” part of that statement to show that X0 is Scott-domain representable. We begin by constructing our poset. The poset (S, ⊑) e) The sets I(B, k): Let ∅ 6= B ⊆ Y and let k ≥ 0. If |B| = 1, then let I(B, k) := N (y, k) where y is the unique point of B. If |B| ≥ 2, then let I(B, k) := {(d, S) ∈ T ∗ : lv(d) ≥ k, B ⊆ S, and ∀y ∈ B, d ∈ y}. Note that the condition d ∈ y is equivalent to e(y, lv(d)) = d, a fact that will be used later. f) Let S = {{t} : t ∈ X} ∪ {I(B, k) : I(B, k) 6= ∅, ∅ 6= B ⊆ Y, 0 ≤ k < ω} and let ⊑ denote reverse inclusion. Consequently, if t ∈ X, then I(B, k) ⊑ {t} means t ∈ I(B, k). Remark 3.1. If |B| ≥ 2, one can prove that I(B, k) = ⋂ {N (y, k) : y ∈ B}, and that was the way we initially thought of the sets I(B, k). However, that fact is not really needed in our construction. The next example illustrates how the sets I(B, k) can behave. It introduces special notations, and parts c), d), and e) will be very important tools in the proofs of later lemmas in this section. Example 3.2. Let d′, d′′ be the two points of T (1) and recall that 0̄ is the unique point of T (0). Let y′, y′′ ∈ Y have e(y′, 1) = d′ and e(y′′, 1) = d′′ (so Scott-representability of some spaces of Tall and Miškin 287 that y′, y′′ are two branches of T that disagree at level 1 of the tree). Note that y′ ∩ y′′ = {(0̄, Y )} = T (0). Then a) A set of the form I(B, k) can be empty. For example, I({y′, y′′}, 1) = ∅ = I(Y, 1) because if (d, S) ∈ I({y′, y′′}, 1) then lv(d) ≥ 1 and e(y′, lv(d)) = d = e(y′′, lv(d)). Because T is a tree and lv(d) ≥ 1 we must have d′ = e(y′, 1) = e(y′′, 1) = d′′ so that y′ ∩ y′′ contains some element of T at or above level 1, which is false. b) I({y′, y′′}, 0) is the infinite set {(0̄, S) : y′, y′′ ∈ S ⊆ Y } and I(Y, 0) is the singleton set {(0̄, Y )}. c) For any I(B, k), if (d, S) ∈ I(B, k) then (d̂, S) ∈ I(B, k) where d̂ is the unique predecessor of d in level k of the tree T . d) For any I(B, k), if (d, S) ∈ I(B, k) then (d, B) ∈ I(B, k) and (d, S′) ∈ I(B, k) whenever S ⊆ S′ ⊆ Y . In particular (d, Y ) ∈ I(B, k). e) From b), c) and d), the only way that |I(B, k)| = 1 is for k = 0 and B = Y , and then I(Y, 0) = {(0̄, Y )}. Lemma 3.3. For any B ⊆ Y and any k ≥ 0, |I(B, k)∩Y | ≤ 1. If |B| ≥ 2 then I(B, k) ⊆ T ∗ and π1[I(B, k)] is finite, where π1 : T ∗ → T is first coordinate projection. Proof. The first two assertions follow directly from the definition of the sets I(B, k), so we prove only the final assertion. Because |B| ≥ 2 we may choose distinct y1, y2 ∈ B. Then there is some integer L such that e(y1, L) 6= e(y2, L) so that e(y1, j) 6= e(y2, j) for each j ≥ L. Therefore, if (d, S) ∈ I(B, k), we know that d ∈ T and lv(d) < L, and there are only finitely many such points. � Lemma 3.4. The maximal elements of S are the singleton sets {x} where x ∈ X. Lemma 3.5. If ∅ 6= B1 ⊆ B2 ⊆ Y with and k1 ≤ k2, then I(B1, k1) ⊑ I(B2, k2). Furthermore if I(B1, k1) ⊑ I(B2, k2) 6= ∅, then B1 ⊆ B2 and k1 ≤ k2. Proof. First suppose that B1 ⊆ B2 and k1 ≤ k2. If |B2| = 1 then B1 = B2. Let y be the unique point of B2. Then k1 ≤ k2 gives I(B2, k2) = N (y, k2) ⊆ N (y, k1) = I(B1, k1) and hence I(B1, k1) ⊑ I(B2, k2). In case B2 has at least two points, then I(B2, k2) ⊆ T ∗ so that each element of I(B2, k2) has the form (d, S) where lv(d) ≥ k2 and d ∈ B2. Hence lv(d) ≥ k2 ≥ k1 and d ∈ y for each y ∈ B2. Because B1 ⊆ B2, we have (d, S) ∈ I(B1, k1), as required. To prove the second claim, note that I(B1, k1) ⊑ I(B2, k2) gives I(B2, k2) ⊆ I(B1, k1) because ⊑ is reverse inclusion. Now fix any (d, S) ∈ I(B2, k2). Then lv(d) ≥ k2, B2 ⊆ S, and d ∈ y for all y ∈ B2. Let d̂ be the unique predecessor of d at level k2 of the tree T . Then (see Example 3.2), (d̂, S) ∈ I(B2, k2) ⊆ I(B1, k1) so that k2 = lv(d̂) ≥ k1. Thus k1 ≤ k2. Next, Example 3.2 shows that since (d, S) ∈ I(B2, k2), (d, B2) ∈ I(B2, k2) ⊆ I(B1, k1) so that B1 ⊆ B2, as required. � 288 H. Bennett and D. Lutzer Lemma 3.6. Let E := {I(Bα, kα) : α ∈ A} be a directed subset of (S, ⊑) that contains no maximal element of itself. Let C = ⋃ {Bα : α ∈ A}. a) If |C| = 1 then the set {kα : α ∈ A} is unbounded, and sup(E) = {y} where y is the unique point of C. (Note that in this case, y ∈ Y .) b) If |C| ≥ 2, then {kα : α ∈ A} is bounded and sup(E) = I(C, L) where L = max{kα : α ∈ A}. Proof. In case (a), it is clear that {y} is an upper bound for E, and that no other {z} for z ∈ Y can be an upper bound for E. In addition, each Bα = {y}. If the set {kα : α ∈ A} is bounded, let kβ be its largest member. Then I(Bβ , kβ ) is the maximal member of E, contrary to hypothesis. Therefore {kα : α ∈ A} is unbounded, and now it is clear that sup E = {y}. To prove (b), fix distinct y1, y2 ∈ C and choose αi ∈ A with yi ∈ Bαi for i = 1, 2. Using directedness of E, find β ∈ A with I(Bαi , kαi ) ⊑ I(Bβ , kβ ). Then I(Bβ , kβ ) 6= ∅ so that by Lemma 3.5 yi ∈ Bαi ⊆ Bβ . According to Lemma 3.3, the set F := π1[I(Bβ , kβ )] is finite. Next, we claim that some d ∈ F has d ∈ π1[I(Bα, kα)] for each α ∈ A. For contradiction, suppose that corresponding to each d ∈ F there is some γ(d) ∈ A with d 6∈ π1[I(Bγ(d), kγ(d))]. Directedness of E provides some η ∈ A with I(Bβ , kβ ) ⊑ I(Bη, kη) and such that I(Bγ(d), kγ(d)) ⊑ I(Bη, kη) for each of the finitely many d ∈ F . Choose any (d̄, S) ∈ I(Bη, kη). Then I(Bβ , kβ ) ⊑ I(Bη, kη) yields I(Bη, kη) ⊆ I(Bβ , kβ) so that d̄ ∈ π1[I(Bβ , kβ )] = F . Be- cause d̄ ∈ F we know that γ(d̄) is defined and d̄ 6∈ π1[I(Bγ(d̄), kγ(d̄)]. Because I(Bγ(d̄), kγ(b̄)) ⊑ I(Bη, kη) we have (d̄, S) ∈ I(Bη, kη) ⊆ I(Bγd̄ , kγb̄ ) and that is impossible because we know that d̄ 6∈ π1[I(Bγ d̄ , kγ(d̄))]. At this stage of the argument, we know that there is some d0 ∈ F with d0 ∈ π1[I(Bα, kα)] for each α ∈ A. Then for some Sα ⊆ Y we have (d0, Sα) ∈ I(Bα, kα). Because Bα ⊆ C, part (c) of Example 3.2 shows that (d0, C) ∈ I(Bα, kα). Consequently lv(d0) ≥ kα and we conclude that lv(d0) is an upper bound for the set {kα : α ∈ A}. Let L be the largest member of the set {kα : α ∈ A}. Note that lv(d0) ≥ L. Next we claim that (d0, C) ∈ I(C, L). Consider the membership criteria for I(C, L). We already know that lv(d0) ≥ L and obviously C ⊆ C, so all we must show is that d0 ∈ y for each y ∈ C. Fix any y ∈ C. Then there is some α ∈ A with y ∈ Bα. From above we know that (d0, C) ∈ I(Bα, kα) so that y ∈ Bα gives d0 ∈ y as required. Now we know that I(C, L) 6= ∅ so that I(C, L) ∈ S. According to Lemma 3.5, I(C, L) is an upper bound for E. To complete the proof that I(C, L) = sup(E), we consider any upper bound G ∈ S for E and we will show that I(C, L) ⊑ G. With I(Bβ , kβ ) as defined in the second paragraph of this proof, we have I(Bβ , kβ ) ⊑ G so that G ⊆ I(Bβ , kβ ). Hence G ⊆ I(Bβ , kβ ) ⊆ T ∗ so that either G has the form G = I(H, m) or else G = {(e, S)} ∈ max S. In the first case, Lemma 3.5 shows that I(Bα, kα) ⊑ I(H, m) implies Bα ⊆ H and kα ≤ m for each α ∈ A, so that C ⊆ H and L = max{kα : α ∈ A} ≤ m. Hence I(C, L) ⊑ I(H, m) = G, as claimed. In the second case, where G = {(e, S)}, we will show that (e, S) ∈ I(C, L). Note Scott-representability of some spaces of Tall and Miškin 289 that I(Bα, kα) ⊑ G = {(e, S)} gives (e, S) ∈ I(Bα, kα) so that lv(e) ≥ kα and Bα ⊆ S for each α and therefore C ⊆ S and lv(e) ≥ max{kα : α ∈ A} = L. Furthermore, if y ∈ C then y ∈ Bα for some α ∈ A so that (e, S) ∈ I(Bα, kα) guarantees that e ∈ y. Therefore, I(C, L) ⊑ G, as required. to show that I(C, L) = sup(E). � Lemma 3.7. In S, we have I(B1, k1) ≪ I(B2, k2) if and only if B1 is a finite set, B1 ⊆ B2, and k1 ≤ k2. Proof. First suppose I(B1, k1) ≪ I(B2, k2). Then I(B1, k1) ⊑ I(B2, k2) so that B1 ⊆ B2 and k1 ≤ k2. We let F be the collection of all finite subsets of B2. Then E := {I(F, k2) : F ∈ F} is a directed subset of S and I(B2, k2) = sup E so that I(B1, k1) ≪ I(B2, k2) gives I(B1, k1) ⊑ I(F1, k2) for some F1 ∈ F, showing that B1 ⊆ F1. Since F1 is finite, so is B1. For the converse, suppose that B1 is a finite set and B1 ⊆ B2 and k1 ≤ k2 (so that I(B1, k1) ⊑ I(B2, k2)), and suppose that E = {I(Bα, kα) : α ∈ A} is a directed subset of S with I(B2, k2) ⊑ sup(E). If E contains a maximal element of itself, there is nothing to prove, so assume that E contains no maximal element. Let C := ⋃ {Bα : α ∈ A}. There are several cases to consider. In case |C| ≥ 2, Lemma 3.6 gives I(B1, k1) ⊑ I(B2, k2) ⊑ sup E = I(C, L) where L is the largest member of the bounded set {kα : α ∈ A}, say L = kγ for some γ ∈ A. Then I(B1, k1) ⊑ I(B2, K2) ⊑ I(C, L) gives B1 ⊆ B2 ⊆ C. Therefore, each y in the finite set B1 is a point of C = ⋃ {Bα : α ∈ A}, so we may find α(y) ∈ A with y ∈ Bα(y). Directedness of the collection E allows us to find β ∈ A with I(Bα(y), kα(y)) ⊑ I(Bβ , kβ) for each y in the finite set B1 and therefore y ∈ Bα(y) ⊆ Bβ . Therefore B1 ⊆ Bβ. Once again using directedness, find δ ∈ A with I(Bγ , kγ ), I(Bβ , kβ ) ⊑ I(Bδ, kδ). Then B1 ⊆ Bβ ⊆ Bδ and k1 ≤ max{kα : α ∈ A} = L = kγ ≤ kδ ≤ L. Therefore I(B1, k1) ⊑ I(Bδ, kδ) ∈ E as required. The remaining case is where |C| = 1, say C = {z}. Then Bα = {z} for each α ∈ A. Because E contains no maximal element of itself, Lemma 3.6 shows that sup E = {z} and that {kα : α ∈ A} is unbounded. Choose µ ∈ A with kµ > k1. Then I(Bµ, kµ) = N (z, kµ) ⊆ N (z, k1) = I(B1, k1) so that I(B1, k1) ⊑ I(Bµ, kµ) ∈ E as required. � Lemma 3.8. Suppose S ∈ S and y ∈ Y . Then S ≪ {y} if and only if S = I({y}, k) for some k ≥ 0. Proof. Suppose S = I({y}, k). By Lemma 3.7, I({y}, k) ≪ I({y}, k) ⊑ {y}, so we know that S = I({y}, k) ≪ {y}. For the converse, suppose S ∈ S has S ≪ {y}. Then S ⊑ {y} so that y ∈ S. By Lemma 3.3, either S = I({y}, k) or else S = {y}. If S = {y} let E := {I({y}, k) : k ≥ 0}. This is a directed set in S with sup E = {y} and yet no member I({y}, k) ∈ E has S = {y} ⊑ I({y}, k). Therefore, S must have the form S = I({y}, k) as claimed. � 290 H. Bennett and D. Lutzer Lemma 3.9. For t ∈ X − Y, {t} ≪ {t} provided t 6= (0̄, Y ). Proof. Write t = (d, S) with (d, S) 6= (0̄, Y ). To show that {t} ≪ {t}, suppose {t} ⊑ sup E where E is a directed subset of S. Maximality of {t} in S (see Lemma 3.4) shows that sup(E) = {t}. If E contains a maximal member, there is nothing to prove, so for contra- diction, suppose E contains no maximal member of itself. Then the collection E must be of the form E = {I(Bα, kα) : α ∈ A}. Write C = ⋃ {Bα : α ∈ A}. If |C| = 1, then C = {y} ⊆ Y , so that Lemma 3.6 shows sup E = {y} and hence {y} = {t}. That is impossible because y ∈ Y and t ∈ X − Y . Therefore |C| ≥ 2. Because |C| ≥ 2, from Lemma 3.6 we know that the set {kα : α ∈ A} is bounded and sup E = I(C, L) where L is the maximal element of the bounded set {kα : α ∈ A}. Then {t} = sup(E) = I(C, L) so that I(C, L) is a singleton. Part (e) of Example 3.2 shows that the set I(C, L) can be a singleton if and only if C = Y and L = 0, and then I(C, L) = {(0̄, Y )}, forcing us to conclude that t = (0̄, Y ), which is false. This contradiction completes the proof of the lemma. � Corollary 3.10. The poset (S, ⊑) is continuous. Proof. Consider any element S ∈ S. If S ≪ S, then S ∈ ⇓(S), so that ⇓(S) is directed with sup(⇓(S)) = S. So suppose S ≪ S is false. Then Lemmas 3.8 and 3.9 show that one of the following three statements must be true: (i) S = I(B, k) where B is infinite, or (ii) S = {y} for some y ∈ Y , or (iii) S = {(0̄, Y )}. If S = I(B, k) where B is infinite, let F be the collection of all finite subsets of B. Then, by Lemma 3.7, ⇓(I(B, k)) = {I(F, j) : j ≤ k, F ∈ F}, which is directed and has I(B, k) as its supremum, as required. In case S = {y} for some y ∈ Y , then ⇓(S) = {I({y}, k) : k ≥ 1} which is also directed and has supremum S = {y}, as required. The case where S = {(0̄, Y )} is actually a special case of item (i) because {(0̄, Y )} = I(Y, 0) as noted in Example 3.2, above. � Lemma 3.11. (S, ⊑) is a Scott domain. Proof. Suppose U1, U2 ∈ S have a common extension. We may assume that neither Ui is maximal in S (so that Ui = I(Bi, ki) for i = 1, 2) and that neither of U1, U2 is contained in the other. Then there is some (d, S) ∈ I(B1, k1) ∩ I(B2, k2). Let C = B1 ∪ B2. Because neither of U1, U2 is contained in the other, |C| ≥ 2 and (d, S) ∈ I(C, max(k1, k2)) yields I(C, max(k1, k2)) 6= ∅ so that I(C, max(k1, k2)) ∈ S. Clearly I(C, max(k1, k2)) is an upper bound for U1 and U2. To show that I(C, max(k1, k2)) is the least upper bound of U1 = I(B1, k1) and U2 = I(B2, k2), consider any upper bound U3 ∈ S for U1 and U2. From Ui ⊑ U3 we obtain U3 ⊆ U1∩U2. Because |C| ≥ 2 we know that U3 ⊆ U1∩U2 ⊆ Scott-representability of some spaces of Tall and Miškin 291 X − Y , so that U3 cannot have the form {y} for some y ∈ Y . Therefore either U3 = I(D, j) for some D and some j, or else U3 = {(d̂, Ŝ)} ∈ max(S). In the first case Bi ⊆ D and j ≥ ki for i = 1, 2 so that C ⊆ D and max(k1, k2) ≤ j and therefore (see Lemma 3.5) I(C, max(k1, k2)) ⊑ U3. In the second case, where U3 = {(d̂, Ŝ)} ∈ max(S), for i = 1, 2 we know that (d̂, Ŝ) ∈ I(Bi, ki) so that lv(d̂) ≥ ki, Bi ⊆ Ŝ, and that for each y ∈ Bi, y ∈ d̂. Hence I(C, max(k1, k2)) ⊑ U3. Therefore I(C, max(k1, k2)) = sup(U1, U2) as required. � There is a natural-looking function that sends each x ∈ X to the element {x} ∈ S. This mapping is 1-1, onto, and continuous from X to max(S), and it is tempting to think that the function is an a homeomorphism from X onto max(S). Unfortunately, it is not. The point (0̄, Y ) ∈ X is isolated in X, but the point {(0̄, Y )} is not an isolated point of max(S). We are lucky that (0̄, Y ) is the only “bad” point for the natural mapping. Recall that X0 = X − {(0̄, Y )}. Then we have: Lemma 3.12. The function h : X0 → max(S)−{{(0̄, Y )}} given by h(t) = {t} is a homeomorphism from X0 onto the open subspace max(S) − {{(0̄, Y )}} of max(S) with the relative Scott topology. Proof. Clearly the function h is 1-1 and h[X0] = max(S)−{{(0̄, Y )}}. To prove that h is continuous, it is enough to consider what happens at non-isolated points of X0, i.e., at points y ∈ Y . Suppose h(y) ∈ ⇑(p) ∩ max(S) where p ∈ S. Then Lemma 3.8 guarantees that p = I({y}, k) = N (y, k) for some k. We claim that that h[N (y, k + 1)] ⊆ ⇑(p). Apply Lemma 3.9 to show that if (d, S) ∈ N (y, k+1) then (d, S) 6= (0̄, Y ) so that h((d, S)) = {(d, S)} ≪ {(d, S)}. Then note that p ⊑ {(d, S)} ≪ {(d, S)} so that h(d, S) ∈ ⇑(p) as required. To prove that h is an open mapping onto max(S) −{{(0̄, Y )}}, the first step is to recall Lemma 3.9 which shows that if t ∈ X − Y with t 6= (0̄, Y ), i.e., if t is an isolated point of X0, then in S, {t} ≪ {t} so that h(t) = {t} is an isolated point of max(S). Second, consider any non-isolated point y ∈ X0 and note that for k ≥ 1, h[N (y, k)] = max(S) ∩ ⇑(I({y}, k). Therefore h is an open mapping onto max(S) − {{(0̄, Y )}} as required. � Our next lemma shows that X0 is Scott-domain-representable. Lemma 3.13. The subspace X0 = X −{(0̄, Y )} is Scott-domain-representable. Proof. Because S is a Scott domain, we know that its subspace max(S) is Scott-domain-representable. It is easy to check that for any domain D, the subspace max(D) is T1. Therefore we see that for our Scott domain S, the set max(S)−{{(0̄, Y )}} is an open subspace of the Scott-domain-representable space max(S). Now recall that any non-empty, relatively open subset of a Scott-domain representable space is also Scott-domain representable, and that completes the proof. � 292 H. Bennett and D. Lutzer References [1] J. Aarts, J. deGroot and R. McDowell, Cocompactness, Nieuw Archief voor Wiskungid 36 (1970), 2–15. [2] J. Aarts and D. Lutzer, Completeness Properties Designed for Recognizing Baire Spaces, Dissertationes Mathematicae 116 (1974), 1–45. [3] H. Bennett and D. Lutzer, Domain representable spaces, Fundamenta Mathematicae 189 (2006), 255–268. [4] H. Bennett and D. Lutzer, Domain representability of certain complete spaces, Houston J. Math, to appear. [5] H. Bennett, D. Lutzer and G. M. Reed, Domain representability and the Choquet game in Moore and BCO-spaces, Topology and its Applications, to appear. [6] M. E. Estill, Concerning abstract spaces, Duke Mathematics Journal 17 (1950), 317–327. [7] R. Kopperman, H. Kunzi and P. Waszkiewicz, Bounded complete models of topological spaces, Topology and its Applications 139 (2004), 285–297. [8] K. Martin, Topological games in domain theory, Topology and its Applications 129 (2003), 177–186. [9] K. Martin, M. Mislove and G.M. Reed, Topology and Domain Theory, pp. 371-394 in Recent Progress in General Topology II, ed. by M Husak and J. van Mill, Elsevier, Amsterdam, 2002. [10] V. Mǐskin, The Amsterdam properties in Moore spaces, Colloq. Math Soc. Janos Bolyai 41 (1983), 427–439. [11] F. Tall, A counterexample in the theories of compactness and metrization, Indag. Math. 35 (1973), 471–474. Received July 2007 Accepted October 2007 Harold Bennett (bennett@math.ttu.edu) Texas Tech University, Lubbock, TX 79409, USA. David Lutzer (lutzer@math.wm.edu) College of William & Mary, Williamsburg, VA 23187, USA.