GiCaWaAGT.dvi @ Applied General Topology c© Universidad Politécnica de Valencia Volume 7, No. 2, 2006 pp. 211-231 On resolutions of linearly ordered spaces Agata Caserta, Alfio Giarlotta∗ and Stephen Watson Abstract. We define an extended notion of resolution of topological spaces, where the resolving maps are partial instead of total. To show the usefulness of this notion, we give some examples and list several properties of resolutions by partial maps. In particular, we focus our attention on order resolutions of linearly ordered sets. Let X be a set endowed with a Hausdorff topology τ and a (not necessarily related) linear order �. A unification of X is a pair (Y, ı), where Y is a LOTS and ı : X →֒ Y is an injective, order-preserving and open-in-the-range function. We exhibit a canonical unification (Y, ı) of (X, �, τ ) such that Y is an order resolution of a GO-space (X, �, τ ∗), whose topology τ ∗ refines τ . We prove that (Y, ı) is the unique minimum unification of X. Further, we explicitly describe the canonical unification of an order resolution. 2000 AMS Classification: 54F05, 06F30, 46A40, 54A10 Keywords: Resolution, lexicographic ordering, GO-space, linearly ordered topological space, pseudo-jump, TO-embedding, unification. 1. Resolutions by partial maps Let (X,τ) be a topological space; if the topology is understood, we denote it by X. By neighborhood of a point we mean open neighborhood. For each point x ∈ X, the set of all neighborhoods of x is denoted by τ(x). Similarly, if B is a base for X, then B(x) denotes the set of all basic neighborhoods of x. In this paper topological spaces are assumed to be Hausdorff. A chain is a linearly ordered set. If (X,�) is a chain, then its reverse chain is (X,�∗), where x �∗ y if and only if y � x for each x,y ∈ X; to simplify notation, we denote this chain by X∗. If A and B are subchains of (X,�), the notation A ≺ B stands for a ≺ b for each a ∈ A and b ∈ B; in particular, if A = {a}, we simplify notation and write a ≺ B. ∗Corresponding author. 212 A. Caserta, A. Giarlotta and S. Watson A topological space (X,τ) is orderable if there exists a linear order � on X such that the order topology τ� coincides with τ. A linearly ordered topological space (for short, a LOTS ) is a chain endowed with the order topology; we denote a LOTS by (X,�,τ�). A topological space (X,τ) is suborderable if there exists an orderable topo- logical space (Y,σ) such that (X,τ) embeds homeomorphically into (Y,σ). It is known that a topological space (X,τ) is suborderable if and only if there exists a total order on X (called a compatible order) such that (i) the original topology is finer than the order topology, and (ii) each point of X has a local base consisting of (possibly degenerate) intervals. A generalized ordered space (for short, a GO-space) is a suborderable space endowed with a compatible or- der. The class of GO-spaces is known to coincide with the class of topological subspaces of LOTS. In the sequel we assume without loss of generality that a GO-space X is a subspace of the LOTS in which it embeds. For a recent survey on LOTS and GO-spaces, see [3] (Section F-7) and references therein. Now we define the notion of a resolution of a family of topological spaces. This elegant and fundamental idea was introduced by Fedorc̆uk in 1968 (see [2]) and extensively studied by Watson in 1992 (see [4]). Definition 1.1. Let (X,τ) be a topological space, (Yx,τx)x∈X a family of topological spaces and (fx : X(x) → Yx)x∈X a family of continuous maps, where X(x) is an open subset of X \{x}. We endow the set ⋃ x∈X{x}×Yx with a topology τ⊗ induced by a base B. For each point x ∈ X, neighborhood U ⊆ X of x and open set V ⊆ Yx, we define basic open sets U ⊗x V in B as follows: U ⊗x V := {x}×V ∪ ⋃ { {x′}×Yx′ : x ′ ∈ (U ∩f−1x V ) } . The topological space ( ⋃ x∈X{x}×Yx,τ⊗ ) is called the resolution of X at each x ∈ X into Yx by the map fx; we denote it by ⊗ x∈X (Yx,fx). The space X is the global space and the Yx’s are the local spaces. Without loss of generality, we assume that the global space X is non-trivial; in fact, if X = {x}, then the resolution is homeomorphic to Yx. Our defini- tion slightly extends the classical notion of resolution, in which the resolving functions fx are defined on the whole set X \ {x}. For sake of clarity, we refer to the classical notion as a resolution by total functions. On the other hand, the resolving functions fx described in Definition 1.1 can be thought of as partial functions fx : X \{x}→ Yx; we call the associated topological space a resolution by partial functions or simply a resolution. In this section we give some examples and list several properties of reso- lutions. We start by showing that ⊗ x∈X (Yx,fx) is a well-defined topological space. Lemma 1.2. The family B is a base for a topology on the set ⋃ x∈X{x}×Yx. We need a technical result, which we prove first. Lemma 1.3. For each (x′,y′) ∈ U ⊗x V , there exists U ′⊗x′ V ′ ∈B such that (x′,y′) ∈ U′ ⊗x′ V ′ ⊆ U ⊗x V . On resolutions of linearly ordered spaces 213 Proof. The result is obvious if x′ = x. Assume that x′ 6= x. Then (x′,y′) ∈ U ⊗x V implies that (x ′,y′) ∈ ⋃ { {w}×Yw : w ∈ (U ∩f −1 x V ) } , whence x′ ∈ U ∩ f−1x V and y ′ ∈ Yx′ . Set U ′ := U ∩ f−1x V and V ′ := Yx′ . Note that (x′,y′) ∈ U′⊗x′ V ′ ∈B, because fx is continuous on an open subset of X\{x} and so U′ is a neighborhood of x′. Furthermore, we have: U′⊗x′ V ′ ⊆ ⋃ { {w}×Yw : w ∈ U ∩f −1 x V } ⊆ U ⊗x V. Thus U′⊗x′ V ′ satisfies the claim. � By Lemma 1.3, we can assume without loss of generality that a basic neigh- borhood of (x,y) is of the type U ⊗x V , where U ⊆ X is a neighborhood of x and V ⊆ Yx is an open set. Proof of Lemma 1.2. It suffices to show that for any two basic open sets U1 ⊗x1 V1 and U2 ⊗x2 V2 in B, if (x ′,y′) ∈ (U1 ⊗x1 V1)∩(U2 ⊗x2 V2), then there exists U′⊗x′ V ′ ∈B such that (x′,y′) ∈ U′⊗x′ V ′ ⊆ (U1 ⊗x1 V1)∩(U2 ⊗x2 V2). For each i ∈ {1, 2}, if (x′,y′) ∈ Ui ⊗xi Vi, then Lemma 1.3 implies that there exists U′i ⊗x′ V ′ i such that (x ′,y′) ∈ U′i ⊗x′ V ′ i ⊆ Ui ⊗xi Vi. Set U ′ := U′1 ∩U ′ 2 and V ′ := V ′1 ∩V ′ 2 . Then, we obtain: (x′,y′) ∈ U′ ⊗x′ V ′ = (U′1 ⊗x′ V ′ 1 ) ∩ (U ′ 2 ⊗x′ V ′ 2 ) ⊆ (U1 ⊗x1 V1) ∩ (U2 ⊗x2 V2). This proves the claim. � A smaller base for the resolution topology τ⊗ is the following. Lemma 1.4. Let (X,τ) and (Yx,τx)x∈X be topological spaces. Assume that BX is a base for τ and for each x ∈ X, Bx is a base for τx. The family B⊗ := {U ⊗x V : x ∈ X ∧ U ∈BX (x) ∧ V ∈Bx} is a base for ⊗ x∈X (Yx,fx). Proof. Let U ⊗x V be a basic open set in the resolution topology and z a point of U ⊗x V . By Lemma 1.3, we can assume that z = (x,y) for some y ∈ V . Select U′ ∈BX and V ′ ∈Bx such that x ∈ U ′ ⊆ U and y ∈ V ′ ⊆ V . Then the set U′⊗x V ′ ∈B⊗ is such that (x,y) ∈ U ′⊗x V ′ ⊆ U ⊗x V . It follows that B⊗ is a base for ⊗ x∈X (Yx,fx). � Some topological spaces can viewed as a resolution of other spaces only if we use partial resolving functions. In the next example we show that the closed unit interval is a resolution by partial maps of the unit circle. Example 1.5. Let S1 be the unit circle (having the origin (0, 0) as its center). Resolve each point s ∈ S1\{(1, 0)} into the one-point space {ys} by the constant function fs. Note that in order to resolve the point t = (1, 0) into a two-point discrete space {y−t ,y + t }, we cannot use a total function to map continuously S1 \ {t} onto {y−t ,y + t }. Let t ∗ = (−1, 0) be the antipodal point of t. Define ft : S 1 \ {t,t∗} → {y−t ,y + t } by ft(x,y) := y − t if y < 0 and ft(x,y) := y + t if y > 0. The resolution is (homeomorphic to) the unit interval [0, 1]. 214 A. Caserta, A. Giarlotta and S. Watson Another advantage of this extended notion of resolution is that a wide class of subspaces of a resolution by total functions can be seen as a resolution by partial functions. Lemma 1.6. Let Z := ⊗ x∈X (Yx,gx) be the resolution of X at each x ∈ X into Yx by a total map gx. Assume that W is a subspace of Z with the property that for each x ∈ X, there exists an open set Vx ⊆ Yx such that W ∩ ({x}×Yx) = {x}×Vx. Then W is a resolution ⊗ x∈X (Yx,fx) by partial maps fx. Proof. For each x ∈ X, let Vx be an open subset of Yx such that W ∩ ({x}× Yx) = {x}×Vx. Continuity of the total function gx : X\{x}→ Yx implies that the set X(x) := g−1x Vx is open in X \{x}. Denote by fx the partial maps on X \{x}, defined as gx ↾X(x). Then W = ⊗ x∈X (Yx,fx). � In the next examples, the symbol n denotes the discrete LOTS with exactly n elements, i.e., n = {0, 1, . . . ,n− 1}. Example 1.7. The double arrow space is the lexicographic product R ×lex 2 endowed with the order topology. This space can be seen as the resolution of R at each point x into the discrete space 2 by the function fx : R \{x} → 2, defined by fx(x ′) = 0 for x′ < x and fx(x ′′) = 1 for x′′ > x. The Sorgenfrey line is the subspace S = {(x, 1) : x ∈ R} of the double arrow space. The Sorgenfrey line can be trivially seen as a resolution by partial maps: the global space is R, the local spaces are all equal to 1 and the resolving functions are the constant maps gx : (x,→) → 1. Example 1.8. The Alexandroff duplicate is the space R×2, whose topology is such that the subspace R×{0} is homeomorphic to R and the subspace R×{1} is made of isolated points. The Michael line M is the usual space R with each irrational isolated. The Michael line is homeomorphic to the subspace (P ×{1}) ∪ (Q ×{0}) of the Alexandroff duplicate. The space M can also be viewed as a subspace of the following resolution of R. Resolve each rational into the space 1. Further, resolve each irrational x into the discrete space 3 by the function fx : R \ {x} → 3, defined by fx(x ′) = 0 for x′ < x and fx(x ′′) = 2 for x′′ > x. The Michael line is the subspace {(x, 1) : x ∈ R} of this resolution. According to Lemma 1.6, we can view the Michael line M as a resolution by partial maps. Resolve the global space R into 1 at each rational x by the constant (total) map fx : R \{x}→ 1. Further, resolve R at each irrational x into 2 by the empty map. Next we list some simple properties of resolutions; their proof is straightfor- ward and is omitted. Lemma 1.9. (Monotonicity) Let U,U1,U2 be neighborhoods of x ∈ X and V,V1,V2 open sets in Yx. We have: (i) if U1 ⊆ U2, then U1 ⊗x V ⊆ U2 ⊗x V ; On resolutions of linearly ordered spaces 215 (ii) if V1 ⊆ V2, then U ⊗x V1 ⊆ U ⊗x V2; (iii) if U1 ⊆ U2 and V1 ⊆ V2, then U1 ⊗x V1 ⊆ U2 ⊗x V2. Lemma 1.10. (Distributivity) Let (Ui)i∈I be a family of neighborhoods of x ∈ X and (Vj )j∈J a family of open sets in Yx. For each h ∈ I and k ∈ J, we have: (i) ( ⋂ i∈I Ui ) ⊗x Vk = ⋂ i∈I ( Ui ⊗x Vk ) ; (ii) Uh ⊗x ( ⋂ j∈J Vj ) = ⋂ j∈J ( Uh ⊗x Vj ) ; (iii) ( ⋂ i∈I Ui ) ⊗x ( ⋂ j∈J Vj ) = ⋂ i∈I ⋂ j∈J ( Ui ⊗x Vj ) ; (i’) ( ⋃ i∈I Ui ) ⊗x Vk = ⋃ i∈I ( Ui ⊗x Vk ) ; (ii’) Uh ⊗x ( ⋃ j∈J Vj ) = ⋃ j∈J ( Uh ⊗x Vj ) ; (iii’) ( ⋃ i∈I Ui ) ⊗x ( ⋃ j∈J Vj ) = ⋃ i∈I ⋃ j∈J ( Ui ⊗x Vj ) . Lemma 1.11. (Decomposability) Let U be a neighborhood of x ∈ X and V an open set in Yx. We have: U ⊗x V = (U ⊗x Yx) ∩ (X ⊗x V ) . Lemma 1.12. For each x ∈ X, we have: X ⊗x Yx = ⋃ {{x′}×Yx′ : x ′ ∈ domfx ∪{x}} . In particular, if domfx = X \{x}, then X ⊗x Yx = ⋃ x∈X{x}×Yx. Now we define the projections of the resolution on the global and the local spaces. Definition 1.13. The global projection is the function π : ⋃ x∈X{x}×Yx → X defined by π(x,y) := x for each (x,y) ∈ domπ. Further, for each x ∈ X, the local projection at x is the function πx : ⋃ x′∈X(x)∪{x}{x ′}×Yx′ → Yx defined as follows for each (x′,y′) ∈ domπx: πx(x ′,y′) : = { y′ if x′ = x fx(x ′) if x′ ∈ X(x). Note that for each x ∈ X such that domfx = X \ {x}, the domain of πx is⋃ x∈X{x}×Yx. If the resolution functions fx are total maps, then the global and local pro- jections are continuous (see [4], Theorem 6). The same holds also in the case that the resolution functions are partial maps and their domain is endowed with the subspace topology of τ⊗. The next lemma summarizes some related facts; its proof is easy and is omitted. Lemma 1.14. For each neighborhood U ⊆ X of x and open set V ⊆ Yx, we have: (i) π−1U = (U ⊗x Yx) ∪ ⋃ x′∈U\domfx {x′}×Yx′ ; (ii) if U \{x}⊆ domfx, then π −1U = U ⊗x Yx; (iii) π−1x V = X ⊗x V . Corollary 1.15. The global and the local projections are continuous functions. 216 A. Caserta, A. Giarlotta and S. Watson Let U ⊆ X be a neighborhood of x and V ⊆ Yx an open set. The global section and the local section of U ⊗x V are defined, respectively, by globx(U,V ) := π −1U and locx(U,V ) := π −1 x V . Each basic open set is the intersection of its global and local section. Corollary 1.16. For each neighborhood U ⊆ X of x and open set V ⊆ Yx, we have U ⊗x V = π −1U ∩π−1x V = globx(U,V ) ∩ locx(U,V ). Proof. Since ⋃ {{x′}×Yx′ : x ′ ∈ U \ domfx} ∩ (X ⊗x V ) = {x} × V , the claim follows from Lemmas 1.11 and 1.14. � 2. Order resolutions In this section we focus our attention on particular types of resolutions, in which both the global space (X,�,τ�) and the local spaces (Yx,�x,τ�x )x∈X are LOTS. If the spaces Yx are compact, there is a standard way to define the resolving functions fx (called order maps in this setting); the resulting topological space is called an order resolution. In our definition we allow the resolving functions fx to be partially defined, in order to deal also with the cases in which some of the local spaces Yx have no maximum and/or no minimum. We use the following notation for intervals in X (rays are considered as particular intervals): • I is the family of all intervals in X (including X); further, I(x) := {I ∈I : x ∈ I}; • −→ I := {(x′,→) : x′ ∈ X}∪{X} and ←− I := {(←,x′′) : x′′ ∈ X}∪{X}; • −−→ I(x) := {I ∈ −→ I : x ∈ I} and ←−− I(x) := {I ∈ ←− I : x ∈ I}. Similarly, intervals in Yx are denoted as follows: • Ix: the family of all intervals in Yx (including Yx); • −→ Ix := {(y ′,→) : y′ ∈ Yx} and ←− Ix := {(←,y ′′) : y′′ ∈ Yx}. Definition 2.1. For each x ∈ X, let X(x) be the following subset of X \{x}: X(x) : =    X \{x} if ∃ minYx ∧ ∃ maxYx (←,x) if ∃ minYx ∧ ∄ maxYx (x,→) if ∄ minYx ∧ ∃ maxYx ∅ if ∄ minYx ∧ ∄ maxYx. The order map fx : X(x) → Yx is defined as follows for each x ′ ∈ X(x) (if any): fx(x ′) : = { min Yx if x ′ ≺ x max Yx if x ′ ≻ x. The space ( ⋃ x∈X{x}×Yx,τ⊗ ) is called the resolution of X at each x ∈ X into Yx by the order map fx and is denoted by ⊗Ord x∈X Yx or simply by ⊗ x∈X Yx. Also, we denote by B⊗ the following base (cf. Lemma 1.4) for the resolution topology τ⊗: B⊗ := {U ⊗x V : x ∈ X ∧ U ∈I(x) ∧ V ∈Ix} . On resolutions of linearly ordered spaces 217 Example 2.2. Resolve each ordinal α in ω1 into the half-open interval [0, 1) by the order map. Denote this resolution by ⊗ α∈ω1 [0, 1). For each successor ordinal α, a neighborhood base is given by sets of the type {α}×(a,b) and {α}× [0,b). Further, if α is a limit ordinal, then a neighborhood base is composed of all sets of the type {α}×(a,b) and {α}× [0,b)∪ (⋃ β∈(γ,α){β}× [0, 1) ) , where γ < α. Note that the resolution space described above has the same underlying set of the lexicographic product ω1 ×lex [0, 1), but its topology τ⊗ is finer than the order topology τ�lex . In fact, neighborhood bases at limit ordinals are the same for the two topologies, but at successor ordinals neighborhood bases for the resolution topology are strictly finer than for the order topology (cf. Example 2.10). Next we compute U ⊗x V in the case that both U and V are rays. The notation ⋃ (x,→){w} × Yw stands for ⋃ w∈(x,→){w} × Yw; a similar meaning have the other symbols. Lemma 2.3. Let (x′,→) and (←,x′′) be open rays in X containing x, and (y′,→) and (←,y′′) open rays in Yx. The following equalities hold: (x′,→) ⊗x (y ′,→) =    {x}× (y′,→) if ∄maxYx {x}× (y′,→) ∪ ⋃ (x,→) {w}×Yw if ∃maxYx (←,x′′) ⊗x (y ′,→) =    {x}× (y′,→) if ∄maxYx {x}× (y′,→) ∪ ⋃ (x,x′′) {w}×Yw if ∃maxYx (x′,→) ⊗x (←,y ′′) =    {x}× (←,y′′) if ∄minYx {x}× (←,y′′) ∪ ⋃ (x′,x) {w}×Yw if ∃minYx (←,x′′) ⊗x (←,y ′′) =    {x}× (←,y′′) if ∄minYx {x}× (←,y′′) ∪ ⋃ (←,x) {w}×Yw if ∃minYx X ⊗x (y ′,→) = (x′ →) ⊗x (y ′,→) X ⊗x (←,y ′′) = (←,x′′) ⊗x (←,y ′′) 218 A. Caserta, A. Giarlotta and S. Watson (x′,→) ⊗x Yx =    {x}×Yx if ∄minYx ∧ ∄maxYx ⋃ (x′,x] {w}×Yw if ∃minYx ∧ ∄maxYx ⋃ [x,→) {w}×Yw if ∄minYx ∧∃maxYx ⋃ (x′,→) {w}×Yw if ∃minYx ∧∃maxYx (←,x′′) ⊗x Yx =    {x}×Yx if ∄minYx ∧ ∄maxYx ⋃ (←,x] {w}×Yw if ∃minYx ∧ ∄maxYx ⋃ [x,x′′) {w}×Yw if ∄minYx ∧∃maxYx ⋃ (←,x′′) {w}×Yw if ∃minYx ∧∃maxYx and X ⊗x Yx =    {x}×Yx if ∄minYx ∧ ∄maxYx ⋃ (←,x] {w}×Yw if ∃minYx ∧ ∄maxYx ⋃ [x,→) {w}×Yw if ∄minYx ∧∃maxYx ⋃ X {w}×Yw if ∃minYx ∧∃maxYx. Proof. Straightforward from definition. � We use rays to define a subfamily S⊗ ⊆B⊗, which is a subbase for τ⊗. Set • −→ S := {U ⊗x V : x ∈ X ∧ U ∈ −−→ I(x) ∧ V ∈ −→ Ix}, • ←− S := {U ⊗x V : x ∈ X ∧ U ∈ ←−− I(x) ∧ V ∈ ←− Ix}, • S⊗ := −→ S ∪ ←− S . Lemma 2.4. For each U1 ∈ −−→ I(x), U2 ∈ ←−− I(x), V1 ∈ −→ Ix\{Yx} and V2 ∈ ←− Ix\{Yx}, we have: (i) U2 ⊗x V1 = (X ⊗x V1) ∩ (U2 ⊗x Yx) ⊆ X ⊗x V1 = U1 ⊗x V1; (ii) U1 ⊗x V2 = (U1 ⊗x Yx) ∩ (X ⊗x V2) ⊆ X ⊗x V2 = U2 ⊗x V2; On resolutions of linearly ordered spaces 219 (iii) (U1 ∩ U2) ⊗x (V1 ∩ V2) = (U1 ⊗x V2) ∩ (U2 ⊗x V1) = (U1 ⊗x V1) ∩ (U2 ⊗x V2) = {x}× (V1 ∩V2); (iv) (U1 ∩U2) ⊗x Yx = (U1 ⊗x Yx) ∩ (U2 ⊗x Yx). Proof. Parts (i) and (ii), as well as the second and third equality in (iii) follow from Lemma 2.3. The first equality in (iii) follows from part (i), part (ii) and Lemma 1.10 (iii). Finally, part (iv) is an immediate consequence of Lemma 1.10 (i). � Corollary 2.5. For each x ∈ X and U ⊗x V ∈B⊗, there exist U1 ⊗x V1 ∈ −→ S and U2 ⊗x V2 ∈ ←− S such that U ⊗x V = (U1 ⊗x V1) ∩ (U2 ⊗x V2). In particular, S⊗ is a subbase for τ⊗. Proof. Let x ∈ X. The open interval U ⊆ X containing x can have the following forms: (i) U = X; (ii) U = (x′,→), with x′ ≺ x; (iii) U = (←,x′′), with x ≺ x′′; (iv) U = (x′,x′′), with x′ ≺ x ≺ x′′. Similarly, we have four possibilities for the open interval V ⊆ Yx, namely: (i) V = Yx; (ii) V = (y ′,→); (iii) V = (←,y′′); (iv) V = (y′,y′′). In all sixteen cases, the claim follows from Lemma 2.4. � Since both the global space X and the local spaces Yx are LOTS, we can obtain another topological space as follows. Consider the reverse of all chains and endow them with the relative order topology. By applying the resolution operator to the LOTS X∗ and (Y ∗x )x∈X , we obtain a new topological space, the reverse order resolution ⊗Ord x∈X∗ Y ∗ x . To simplify notation, we denote this space by ⊗∗ x∈X Yx. We shall show that ⊗∗ x∈X Yx = ⊗ x∈X Yx (see Corollary 2.8). Note that if I = (y,z) is an open interval in Yx, then I ∗ ⊆ Y ∗x is the open interval (y,z)∗ = (z,y). Fix x ∈ X. Assume that Yx has a minimum element yx but no maximum element. The order map fx and the reverse order map f ∗ x are defined, respectively, as follows: fx : (←,x) → Yx , w 7−→ yx f∗x : (x,→) → Y ∗ x , w 7−→ yx. Thus, if I is the open interval [yx,y ′′) ⊆ Yx, then (fx) −1I = (←,x) ⊆ X and (f∗x ) −1 I∗ = (x,→) ⊆ X∗ whence (f∗x ) −1 I∗ = ( fx −1I )∗ . Similar considerations can be done in the case that Yx has a maximum but not a minimum, or has both a minimum and a maximum. The following lemma summarizes the above results. Lemma 2.6. Let x ∈ X and assume that Yx has a minimum (respectively, maximum) element yx. If I ⊆ Yx is the open interval [yx,y) (respectively, (y,yx]), then (f ∗ x ) −1 I∗ = ( fx −1I )∗ . Order resolution and reverse order resolution have the same basic open sets. 220 A. Caserta, A. Giarlotta and S. Watson Theorem 2.7. Each basic open set U ⊗x V in ⊗ x∈X Yx is equal to the basic open set U∗ ⊗∗x V ∗ in ⊗∗ x∈X Yx. Proof. Several cases have to be considered: (i) x = minX; (ii) x = maxX; (iii) x 6= minX and x 6= maxX. We examine only case (iii), since the others are similar. Let U ⊗x V be a basic open set in ⊗ x∈X Yx; without loss of generality, let U = (x′,x′′), where x′ ≺ x ≺ x′′. For V ⊆ Yx, we can assume that one of the following cases occurs: (a) V = [y,y′′), with y = minYx; (b) V = (y ′,y], with y = maxYx; (c) V = (y ′,y′′). In case (a), Lemma 2.6 yields U ⊗x V = (x ′,x′′) ⊗x [y,y ′′) = {x}× [y,y′′) ∪ ⋃ {{w}×Yw : w ∈ (x ′,x)} and U∗⊗∗xV ∗ = (x′,x′′)∗⊗∗x[y,y ′′)∗ = {x}×(y′′,y] ∪ ⋃ {{w}×Y ∗w : w ∈ (x,x ′)} . Therefore, U ⊗xV = U ∗⊗∗xV ∗, as claimed. Case (b) is similar to (a). For case (c), we have: U ⊗x V = (x ′,x′′) ⊗x (y ′,y′′) = {x}× (y′,y′′) and U∗⊗∗x V ∗ = (x′,x′′)∗ ⊗∗x (y ′,y′′)∗ = {x}× (y′′,y′) whence U ⊗xV = U ∗⊗∗xV ∗ also in this case. � Corollary 2.8. ⊗ x∈X Yx = ⊗∗ x∈X Yx. Proof. Since the underlying set is the same for both spaces, it suffices to show that their topologies coincide. Theorem 2.7 yields U ⊗x V = U ∗ ⊗∗x V ∗ for any basic open set U ⊗x V in ⊗ x∈X Yx. By duality, we obtain U ∗ ⊗∗x V ∗ = U∗∗ ⊗∗∗x V ∗∗ = U ⊗x V for any basic open set U ∗ ⊗∗x V ∗ in ⊗∗ x∈X Yx. The result follows. � In the last part of this section we study the relationship between two natural topologies defined on the set ⋃ x∈X{x}× Yx. Namely, we compare the order resolution ⊗ x∈X Yx and the chain ∑ x∈X Yx endowed with the order topology τΣ. Lemma 2.9. The resolution topology τ⊗ on ⊗ x∈X Yx is finer than the order topology τΣ on ∑ x∈X Yx. Proof. Let I be an open ray in ∑ x∈X Yx and (x,y) a point in I. To prove the claim we exhibit a neighborhood W ∈ τ⊗(x,y) such that W ⊆ I. By duality (see Theorem 2.7), it suffices to examine the case I = (←, (x′′,y′′)), where x � x′′. First assume that x ≺ x′′. Select U ∈ τ(x) such that x′′ /∈ U and set W := π−1U. Continuity of the global projection π (see Corollary 1.15) implies that W is a neighborhood of (x,y) in ⊗ x∈X Yx, which does not contain the point (x′′,y′′). Thus (x,y) ∈ W ⊆ I, as claimed. On resolutions of linearly ordered spaces 221 Next, let x = x′′ and y ≺x y ′′. Set U := X and V := (←,y′′). Note that fx −1V is equal either to the empty set (if x = minX or Yx has no minimum element) or to the open ray (←,x) (if x 6= minX and Yx has a minimum element). Thus we obtain: (x,y) ∈ U⊗xV = {x}×(←,y ′′)∪ ⋃ { {x′}×Yx′ : x ′ ∈ ( fx −1(←,y′′) ∩X )} ⊆ I. The open set W := U ⊗x V satisfies the claim. � The converse of Lemma 2.9 does not hold in general. In particular, the order resolution of a LOTS into LOTS is a GO-space that is not necessarily a LOTS. Example 2.10. Let X be the discrete LOTS 2 = {0, 1} and Y0 = Y1 the half-open interval [a,b) ⊆ R. We show that the resolution topology τ⊗ on⊗ x∈X Yx = ⊗ i∈2[a,b) is strictly finer than the order topology τΣ on the chain∑ x∈X Yx = 2 ×lex [a,b). Consider the point (1,a) ∈ 2 × [a,b), the open neighborhood {1} of 1 and the open set [a,b) = Y1. Observe that f0 is undefined and f1 : X(1) → [a,b) is defined by f1(0) = a. Thus, the basic open set {1}⊗1 [a,b) is the half-open interval [(1,a), (1,b)). On the other hand, any interval I ∈ τΣ satisfying (1,a) ∈ I ⊆ {1}⊗1 [a,b) must contain a subinterval of the type I ′ = ((0,y1), (1,y2)), where y1,y2 ∈ [a,b). Thus, {1}⊗1 [a,b) is not open in 2 ×lex [a,b). The global projection π : ⋃ x∈X{x}× Yx → X is continuous whenever its domain is endowed with the resolution topology. On the other hand, continuity of π is not ensured if its domain is endowed with the order topology. For exam- ple, π : (2 ×lex [a,b),τΣ) → 2 fails to be continuous, because (2 ×lex [a,b),τΣ) is homeomorphic to the connected interval [a,b). The local projections πx : ⋃ x′∈X(x)∪{x}{x ′}× Yx′ → Yx, with x ∈ X, are continuous if their domain is endowed with the subspace topology of the reso- lution (see Corollary 1.15). Next we show that they are continuous also if we endow their domain with the subspace topology of the LOTS (∑ x∈X Yx,τΣ ) . Before proving this fact, we mention a technical lemma. Lemma 2.11. Let V ⊆ Yx be an open set. The set {x} × V is open in(∑ x∈X Yx,τΣ ) if both of the following conditions are verified: (a) if V contains maxYx, then either x = maxX or x has an immediate successor x′′ and Yx′′ has a minimum; (b) if V contains minYx, then either x = minX or x has an immediate predecessor x′ and Yx′ has a maximum. In particular, {x}×V is open in (∑ x∈X Yx,τΣ ) whenever V contains neither maxYx nor minYx. Proof. Straightforward from definition. � Lemma 2.12. For each x ∈ X, the function πx is continuous with respect to the subspace topology of τΣ. 222 A. Caserta, A. Giarlotta and S. Watson Proof. Let V ⊆ Yx be an open set. Then π −1 x V = X ⊗x V , using Lemma 1.14. To prove the result, it suffices to show that X ⊗x V is open (in the subspace topology of τΣ) for V = (←,y). If Yx has no minimum element, then X ⊗x (← ,y) = {x}× (←,y) is open by Lemma 2.11. On the other hand, if Yx has a minimum element ymin, then π −1 x (←,y) = X ⊗x [ymin,y) = (←, (x,y)). � Next we introduce the notion of pseudojump. Definition 2.13. A pseudojump in the chain ∑ x∈X Yx is a jump (x ′,x′′) in X (i.e., a pair of consecutive points of X) with the property that either (a) ∃maxYx′ and ∄ minYx′′ , or (b) ∄ maxYx′ and ∃minYx′′ . The notion of pseudojump of a chain L = ∑ x∈X Yx obviously depends on the chosen representation of L as a sum of other chains. Consider, e.g., the isomorphic chains (0, 1) (which lacks pseudojumps), (0, 1/2) ⊕ [1/2, 1) (which has exactly one pseudojumps) and ∑ n∈ω ( 1 n+3 , 1 n+2 ] ⊕ (1/2, 1) (which has countably many pseudojumps). On the other hand, if we endow the chains with additional structure, then the notion becomes significant. The next theorem characterizes the order resolution as a LOTS. Theorem 2.14. Let (X,�,τ�) and {(Yx,�x,τ�x )}x∈X be LOTS. The follow- ing statements are equivalent: (i) the order topology τΣ on ∑ x∈X Yx is equal to the resolution topology τ⊗ on ⊗ x∈X Yx, i.e., ⊗ x∈X Yx is a LOTS; (ii) the global projection π : (∑ x∈X Yx,τΣ ) → (X,τ�) is continuous; (iii) the chain ∑ x∈X Yx has no pseudojumps. Proof. (i) ⇒ (ii). This implication follows from Corollary 1.15. (ii) ⇒ (i). By Lemma 2.9, it suffices to show that τ⊗ ⊆ τΣ. Let U ⊗x V be a basic open set for τ⊗. Since U ⊗x V = ( π−1U ) ∩ ( π−1x V ) by Corollary 1.16, the claim follows from hypothesis and Lemma 2.12. (ii) ⇒ (iii). We prove the contrapositive. Without loss of generality, assume that there exists a jump (x′,x′′) in X such that Yx′ has no maximum and Yx′′ has a minimum y ′′ min. Consider the nonempty open ray (x ′,→) ⊆ X. Since the set π−1(x′,→) = [(x′′,y′′min),→) is not open in (∑ x∈XYx,τΣ ) (cf. Example 2.10), it follows that π is not continuous with respect to τΣ. (iii) ⇒ (ii). Assume that ∑ x∈X Yx has no pseudojumps. It suffices to prove that π−1(x′,→) is open in τΣ for each x ′ ∈ X. Let (x,y) ∈ π−1(x′,→). We find an interval I ⊆ ∑ x∈X Yx such that (x,y) ∈ I ⊆ π −1(x′,→). If y 6= minYx, then there exists t ∈ Yx such that t ≺x y. Set I := ((x,t),→). Next assume that y = minYx. If (x ′,x) is a jump in X, then by hypothesis there exists y′max := maxYx′ . Thus I := ((x ′,y′max),→) satisfies the claim. On the other hand, if (x′,x) is not a jump in X, then we can select w ∈ (x′,x) and t ∈ Yw, and set I := ((w,t),→). � On resolutions of linearly ordered spaces 223 3. Unifications and resolutions By TO-space we mean a triple (X,�,τ) such that X is a nonempty set, � is a linear order on X and τ is a Hausdorff topology on X (not necessarily related to the order �). We describe how a TO-space can be canonically embedded into a LOTS that is an order resolution. Definition 3.1. Let (X,�,τ) and (Y,≤,σ) be TO-spaces. A function f : X → Y is a TO-homomorphism if f : (X,τ) → (Y,σ) is open in the range (i.e., open sets of X are mapped into open sets of f(X)) and f : (X,�) → (Y,≤) is order- preserving. In particular, a TO-embedding (respectively, a TO-isomorphism) is an injective (respectively, bijective) TO-homomorphism. Next we list some simple but useful properties of order-preserving maps between chains. Their proof is easy and is omitted. Lemma 3.2. Let f : X → Y be an order-preserving map between chains. We have: (i) the f-preimage of a convex subset of Y is a convex subset of X; (ii) if f is surjective, then the f-preimage of an open [closed, half-open] interval is an open [closed, half-open] interval; (iii) if X is a GO-space, Y is a LOTS and f is surjective, then f is con- tinuous; (iv) if X is a LOTS, Y is a GO-space and f is injective, then f is a TO- embedding; (v) if X and Y are LOTS and f is bijective, then f is a homeomorphism. As the next example shows, the hypothesis that Y is a LOTS is necessary in (iii) and (v). In particular, a TO-isomorphism may fail to be a homeomorphism. Example 3.3. Let X be the unit interval [0, 1] (with the order topology) and Y the topological sum [0, 1) ⊕{1} (with the natural order). The identity map is a TO-isomorphism but is not continuous. Note that X is a LOTS, whereas Y is a GO-space but not a LOTS. Let X be a TO-space. Observe that if f : X → Y is a TO-embedding of X into a LOTS, then its (partial) inverse f−1 : f(X) → X is a continuous and order-preserving map of a GO-space onto X. Vice versa, assume that g : Z → X is a continuous and order-preserving map of a GO-space onto X. Let Y be a LOTS such that Z ⊆ Y . Choose zx ∈ Z is such that g(zx) = x. The function h: X → Y , defined by h(x) := zx for each x ∈ X, is a TO-embedding of X into a LOTS. Definition 3.4. Let (X,�,τ) be a TO-space and ı: X →֒ Y is a TO-embedding of X into a LOTS (Y,≤,τ≤). The pair ((Y,≤,τ≤) , ı) is called a unification of (the topology and the order of) X; if there is no risk of confusion, we simplify the notation and write (Y,ı). Without loss of generality, we assume that a unification of a TO-space X is a pair (Y,ı) such that X is a subchain of Y and ı is the canonical inclusion. (Note that X and ı(X) have different topologies, in general.) 224 A. Caserta, A. Giarlotta and S. Watson A unification (Y,ı) of X is continuous if the TO-embedding ı is a continuous map. Further, (Y,ı) is a minimum unification if for any other unification (Z,ϕ) of X, there exists a TO-embedding ψ : Y →֒ Z such that ϕ = ψ ◦ ı. Example 3.5. Let X = 2 ×lex [0, 1) = [0, 1) ⊕ [0, 1) be a chain endowed with the resolution topology. The following LOTS (together with the canonical inclusions ıY , ıW and ıZ , respectively) are examples of unifications of X: Y = [0, 1] ⊕ [0, 1), W = [0, 2] ⊕ [0, 1) and Z = [0, 1) ⊕ ( 1 n+1 )n∈ω ⊕ [0, 1). Observe that (Y,ıY ) is a continuous minimum unification of X, whereas (W,ıW ) and (Z,ıZ ) are continuous unifications of X, which fail to be minimum. Note also that despite Y and Z are homeomorphic LOTS, the unifications (Y,ıY ) and (Z,ıZ ) are different. Example 3.6. Define on the set X = ω ∪{ω + 1}, endowed with the natural order ≤, a topology σ as follows. Let U be an ultrafilter on ω containing the cofinite filter. We define σ as the topology on X such that all natural numbers are isolated points and a system of σ-neighborhoods for the point ω + 1 is given by {U ∪{ω + 1} : U ∈U}. Then (X,≤,σ) is a TO-space that fails to be a GO- space, because its character χ(X,σ) is uncountable but its pseudo-character ψ(X,σ) is countable (see [1], Problem 3.12.4). A (minimum) unification for X is given by (Y,ı), where Y is the LOTS (ω + 2,≤,τ≤) and ı is the canonical embedding. Minimum unifications of a TO-space are essentially unique. To prove unique- ness, we first define the so-called canonical unification of a TO-space and then show that this is (up to a relabeling) its unique minimum unification. The canonical unification of a TO-space (X,�,τ) is defined in two steps: (i) we obtain a minimum (in the sense of Definition 3.7) refinement τ∗ of the topology τ such that (X,�,τ∗) is a GO-space; (ii) we embed the GO-space (X,�,τ∗) into a minimum (in the sense of Definition 3.9) LOTS that is an order resolution ⊗ x∈X Yx. Definition 3.7. Let (X,�,τ) be a TO-space. The GO-cone above X is the (nonempty) family of all GO-spaces (X,�,σ) such that σ refines τ. The GO- extension of X, denoted by (X,�,τ∗), is the minimum of the GO-cone above X, in the sense that if (X,�,σ) is another GO-space such that σ refines τ, then σ refines also τ∗. The GO-extension of X is well-defined. Lemma 3.8. For each TO-space (X,�,τ), the GO-cone above X has a mini- mum (X,�,τ∗). Proof. We define τ∗. Let x ∈ X. For each τ-neighborhood U of x, denote by C(x,U) the union of all �-convex subsets of U containing x. Let Bx := {C(x,U) : U ∈ τ(x)}. Then B∗ := ⋃ x∈X Bx is a base for a topology τ ∗ on X. The topology τ∗ refines both the original topology τ and the order topology τ� (because (X,τ) is Hausdorff). Further, (X,�,τ ∗) is a GO-space. On resolutions of linearly ordered spaces 225 Next we prove that (X,�,τ∗) is the minimum of the GO-cone above X. Let (X,�,σ) be a GO-space such that σ refines τ. Let B be a base for σ composed of convex sets and C(x,U) a basic open set in τ∗. For each x′ ∈ C(x,U), there exists B ∈B such that x′ ∈ B ⊆ U. Thus x′ ∈ B ⊆ C(x,U), hence C(x,U) is open in σ. This shows that σ refines τ∗. � Definition 3.9. Let (X,�,τ) be a GO-space. A completion of X is a pair (Y,ı), where Y is a LOTS and ı: X →֒ Y is an order-preserving homeomorphic embedding. A completion (Y,ı) is minimum if for any other completion (Z,ϕ), there exists an order-preserving homeomorphic embedding Ψ(ı,ϕ) : Y →֒ Z such that ϕ = Ψ(ı,ϕ) ◦ ı. Note that if X is a GO-space, then both notions of unification of X (as a TO- space) and completion of X make sense. We show that the unique minimum unification of X is indeed its unique minimum completion. Theorem 3.10. Let (X,�,σ) be a GO-space. (i) There exists a minimum completion ( ⊗ x∈X Yx, ı) of X such that the space ⊗ x∈X Yx is the order resolution of X at each point x into a chain Yx with either one, two or three points. (ii) The image ı(X) is topologically dense in ⊗ x∈X Yx. (iii) If (W,η) is another minimum completion of X, then the maps Ψ(ı,η) and Ψ(η,ı) are order-preserving homeomorphisms between ⊗ x∈X Yx and W such that Ψ(ı,η) = Ψ(η,ı)−1. Proof. For each x ∈ X, we define a chain Yx with either one, two or three points as follows. If x has a σ-neighborhood base consisting of open intervals, then set Yx := {yx}. On the other hand, if x has no neighborhood base consisting of open intervals, then we define Yx according to cases (I)-(VII) described below. Let x be a non-endpoint of X. If (I) x has an immediate predecessor and no immediate successor, set Yx := {yx,y + x }. If (II) x has an immediate successor and no immediate predecessor, set Yx := {y − x ,yx}. Further, if x has neither im- mediate predecessor nor immediate successor, it follows that a σ-neighborhood base at x has one of the following forms: (III) the family of all half-open inter- vals (a,x] containing a point different from x; (IV) the family of all half-open intervals [x,b) containing a point different from x; (V) the singleton {x}. Let Yx be the chain {yx,y + x } in case (III), {y − x ,yx} in case (IV), and {y − x ,yx,y + x } in case (V). Finally, if (VI) x is the minimum point of X and has no imme- diate successor, or (VII) x is the maximum point of X and has no immediate predecessor, set Yx := {yx,y + x } in case (VI), and Yx := {y − x ,yx} in case (VII). Let ⊗ x∈X Yx be the order resolution of the LOTS (X,�,τ�) at each point x into the LOTS Yx defined as above. By Theorem 2.14, the space ⊗ x∈X Yx is a LOTS. Further, the correspondence x 7→ (x,yx) gives a order-preserving homeomorphic embedding ı: (X,�,σ) → ⊗ x∈X Yx. Thus ( ⊗ x∈X Yx, ı) is a completion of X. Next we prove that it is minimum. Assume that ϕ: X →֒ Z is an order-preserving homeomorphic embedding of X into a LOTS. We define an order-preserving homeomorphic embedding 226 A. Caserta, A. Giarlotta and S. Watson Ψ = Ψ(ı,ϕ) : ⊗ x∈X Yx →֒ Z such that ϕ = Ψ ◦ ı. For each x ∈ X, denote zx := ϕ(x) and let Ψ(x,yx) := zx. To define Ψ for the points of ⊗ x∈X Yx that are not in the range of ı, we carry a case by case analysis. Since ϕ is a TO-embedding into a LOTS, in cases (I) and (VI) the point zx is isolated in the range of ϕ, whereas in case (III) the interval (←,zx] is open in the range of ϕ. It follows that there exists z+x ∈ Z that is the immediate successor of zx. Set Ψ(x,y + x ) := z + x . Further, cases (II), (IV) and (VII) are dual to, respectively, (I), (III) and (VI). Thus, we set Ψ(x,y−x ) := z − x , where z−x is the immediate predecessor of zx in Z. Finally, in case (V), a combination of the arguments given above yields that there exist z−x ,z + x ∈ Z, which are, respectively, the immediate predecessor and successor of zx. Set Ψ(x,y − x ) := z − x and Ψ(x,y+x ) := z + x . By construction, Ψ is an injective order-preserving map such that ϕ = Ψ ◦ ı. Thus Ψ is a TO-embedding by Lemma 3.2. Next we prove continuity of Ψ using continuity of ϕ. It suffices to show that for any open ray (←,z) ⊆ Z, the preimage Ψ−1(←,z) is open in ⊗ x∈X Yx. If z belongs to the image of Ψ, then Ψ−1(←,z) = (←, Ψ−1(z)) is open in ⊗ x∈X Yx. Now let (←,z) be such that z is not in the image of Ψ. Without loss of generality, assume that ϕ−1(←,z) 6= ∅. Let (x,y) ∈ Ψ−1(←,z). We claim that either (x,y) has an immediate successor (x′′,y′′), or there exists (x′′,y′′) ∈⊗ x∈X Yx such that (x,y) ≺ (x ′′,y′′) and Ψ(x′′,y′′) � z. Continuity of Ψ follows from the claim, since (x,y) ∈ (←, (x′′,y′′)) ⊆ Ψ−1(←,z). To prove the claim, assume by contradiction that (x,y) has no immediate successor and for each (x′′,y′′) ∈ ⊗ x∈X Yx, if (x,y) ≺ (x ′′,y′′) then z ≺ Ψ(x′′,y′′). Then x has no immediate successor and ı(x) = (x,yx) is such that yx is the maximum of Yx. It follows that ϕ −1(←,z) = (←,x] is not open in X, which contradicts the continuity of ϕ. This finishes the proof of (i). To prove (ii), observe that any two points in ⊗ x∈X Yx \ ı(X) cannot be consecutive. Furthermore, it is easy to show that all points in ⊗ x∈X Yx \ ı(X) are not isolated in τ⊗. It follows that ı(X) = ⊗ x∈X Yx. Finally, assume that (W,η) is another minimum completion of X. We show that the compositions Ψ(η,ı)◦Ψ(ı,η) and Ψ(ı,η)◦Ψ(η,ı) are the identity maps on ⊗ x∈X Yx and W , respectively. This will prove (iii). By hypothesis, there exist order-preserving homeomorphic embedding Ψ(ı,η) and Ψ(η,ı) such that η = Ψ(ı,η) ◦ ı and ı = Ψ(η,ı) ◦η. Thus the composition Ψ(η,ı) ◦ Ψ(ı,η) : ⊗ x∈X Yx →֒ ⊗ x∈X Yx is an order-preserving homeomorphic embedding, whose restriction to ı(X) is the canonical inclusion of ı(X) into⊗ x∈X Yx. By (ii), it follows that Ψ(η,ı) ◦ Ψ(ı,η) is the identity on ⊗ x∈X Yx. To prove that Ψ(ı,η) ◦ Ψ(η,ı) is the identity on W , we first show that η(X) is topologically dense in W . By way of contradiction, assume that there exists w ∈ W and an open interval (a,b) ⊆ W such that w ∈ (a,b) and (a,b) ∩ η(X) = ∅. It follows that (Ψ(a), Ψ(b)) is an open interval containing Ψ(w), which does not intersect ı(X). This contradicts property (ii). Now the equality Ψ(ı,η) ◦ Ψ(η,ı) = idW follows by an argument similar to that given above. � On resolutions of linearly ordered spaces 227 Note that the topology of ⊗ x∈X Yx is the same as its order topology (see Theorem 2.14). An immediate consequence of Theorem 3.10 is Corollary 3.11. For each GO-space (X,�,σ), the pair ( ⊗ x∈X Yx, ı) is the unique (up to a relabeling) minimum completion. Remark 3.12. For any GO-space (X,�,σ), a completion (Y,ı) of X is also a unification of X. The proof of Theorem 3.10 yields that the unique (up to a rela- beling) minimum completion ( ⊗ x∈X Yx, ı) of X has the property that for each TO-embedding ϕ: X →֒ Z of X into a LOTS, there exists a TO-embedding Ψ(ı,ϕ) : ⊗ x∈X Yx →֒ Z such that ϕ = Ψ(ı,ϕ) ◦ ı. Thus the minimum comple- tion of X is also a minimum unification of X (indeed, the minimum unification of X, cf. Corollary 3.15). Definition 3.13. Let (X,�,τ) be a TO-space. Using Lemma 3.8 and Theo- rem 3.10, we define the canonical unification ( ⊗̂ x∈XYx, ı̂) of X as follows. The LOTS ⊗̂ x∈XYx is an order resolution obtained as in the proof of Theorem 3.10: the global space is the GO-extension (X,�,τ∗) of (X,�,τ), whereas the local spaces Yx are the discrete LOTS with either one, two or three points. The TO- embedding ı̂ is the composition of the identity idτ ∗ τ : (X,�,τ) → (X,�,τ ∗) and the homeomorphic embedding ı: (X,�,τ∗) → ⊗̂ x∈XYx defined in the proof of Theorem 3.10. Theorem 3.14. Let (X,�,τ) be a TO-space. The canonical unification of X is a minimum unification. Further, for any other minimum unification (W,η̂) of X, there exist order-preserving homeomorphisms ψ : ⊗̂ x∈XYx → W and χ: W → ⊗̂ x∈XYx such that ψ ◦ ı̂ = χ, χ◦ ı̂ = ψ and ψ −1 = χ. Proof. First we show that ( ⊗̂ x∈XYx, ı̂) is a minimum unification. Let (Z,ϕ̂) be a unification of X. We define a TO-embedding ψ : ⊗̂ x∈XYx → Z such that ϕ̂ = ψ ◦ ı̂. Note that ϕ̂(X) ⊆ Z is a GO-space. Let σ be the topology on X such that ϕ̂ gives a homeomorphism between (X,σ) and ϕ̂(X). Thus (X,σ) is an element of the GO-cone above (X,τ). Further, we have ϕ̂ = ϕ ◦ idστ , where id σ τ : (X,τ) → (X,σ) is the identity map and ϕ: (X,σ) → Z is the homeomorphic embedding defined as ϕ̂. By definition of GO-extension of (X,�,τ), the identities id τ ∗ τ : (X,τ) → (X,τ ∗), id σ τ ∗ : (X,τ ∗) → (X,σ) and idστ : (X,τ) → (X,σ) are order-preserving open maps. Set ϕ′ := ϕ ◦ idστ ∗ . Then ϕ ′ : (X,�,τ∗) →֒ Z is a TO-embedding. By Theorem 3.10, there exists a TO-embedding Ψ(ı,ϕ′) : ⊗̂ x∈XYx → Z such that ϕ′ = Ψ(ı,ϕ′) ◦ ı. It follows that ϕ̂ = ϕ′ ◦ idτ ∗ τ = Ψ(ı,ϕ ′) ◦ ı̂. Thus ψ := Ψ(ı,ϕ′) satisfies the claim. Now let (W,η̂) be another minimum unification of X. Using the same no- tation as above (and as in Theorem 3.10), one can show that the maps ψ := Ψ(ı,η′) : ⊗̂ x∈X Yx → W and χ := Ψ(η,ı ′) : W → ⊗̂ x∈X Yx 228 A. Caserta, A. Giarlotta and S. Watson are TO-embeddings such that ψ◦ ı̂ = χ and χ◦ ı̂ = ψ. Theorem 3.10 (iii) yields that ψ and χ are order-preserving homeomorphisms such that ψ−1 = χ. � Remark 3.12 can now be restated as follows. Corollary 3.15. The canonical unification of a GO-space is its unique (up to a relabeling) minimum unification and minimum completion. We conclude the paper by describing explicitly the minimum unification (and completion) of an order resolution. Let (X,�) and (Yx,�x)x∈X be chains. Endow the set ⋃ x∈X{x}×Yx with the lexicographic order �lex and the order resolution topology τ⊗. We denote this GO-space by (⊗ x∈X Yx,�lex ) . Definition 3.16. For each x ∈ X, define a chain (Y +x ,� + x ) such that Yx ⊆ Y + x and �+x extends �x in the following way: (i) if x has an immediate predecessor x′ and an immediate successor x′′ such that ∃maxYx′ , ∄ minYx, ∄ maxYx and ∃minYx′′ , then set Y + x := Yx ∪{y0,y1} and y0 ≺ + x y ≺ + x y1 for all y ∈ Yx; (ii) if (i) does not hold and x has an immediate predecessor x′ such that ∃maxYx′ but ∄ minYx, then set Y + x := Yx ∪{y0} and y0 ≺ + x y for all y ∈ Yx; (iii) if (i) does not hold and x has an immediate successor x′′ such that ∃minYx′′ but ∄ maxYx, then set Y + x := Yx ∪{y1} and y ≺ + x y1 for all y ∈ Yx; (iv) in all other cases, set Y +x := Yx. Endow the chain ∑ x∈X Y + x with the order topology τ + Σ . We call the LOTS(∑ x∈X Y + x ,τ + Σ ) the lexicographic completion of the family (Yx,�x)x∈X . Open intervals in this LOTS are denoted by ((x′,y′), (x′′,y′′))+; a similar notation is used for the other types of intervals and for rays. (Note that the lexicographic completion of a family of chains is obtained by inserting a jump per pseudo- jump, thus eliminating all pseudojumps in the representation of the sum.) The set ⋃ x∈X{x}×Yx can be endowed with two topologies (apart from the resolution topology): the order topology τΣ and the subspace topology of τ + Σ (inherited from the lexicographic completion of the family (Yx)x∈X ). Note that∑ x∈XYx is an open dense subspace of (∑ x∈XY + x ,τ + Σ ) . In the next result we list some sets that are always open in the subspace topology (but possibly fail to be open in the order topology). Lemma 3.17. For each x ∈ X, the following sets are open in the space(⋃ x∈X{x}×Yx,τ + Σ ) : (i) A = ⋃ w∈(x,→){w}×Yw; (ii) B = ⋃ w∈[x,→){w}× Yw, if either ∄minYx, or ∃minYx and x has an immediate predecessor; (iii) C = ⋃ w∈[x,x′′){w}× Yw, if either ∄minYx, or ∃minYx and x has an immediate predecessor. (If x = maxX, then [x,x′′) = {x}.) On resolutions of linearly ordered spaces 229 Proof. For (i), note that A is always open in τΣ (and hence in τ + Σ ), except for the following case: Yx has no maximum, x has an immediate successor x ′′ and Yx′′ has a minimum y ′′ min. In this case, there exists ymax := maxY + x ∈ Y + x \Yx. It follows that A = [(x′′,y′′min),→) = ((x,ymax),→) + ∩ ( ⋃ x∈X {x}×Yx ) is open in τ+Σ . To prove (ii), first assume that Yx has no minimum and let (s,t) be a point of B. If s = x, then we can select y ∈ Yx such that y ≺x t and so (s,t) ∈ ((s,y),→) ⊆ B. If s 6= x, then (s,t) ∈ ((x,y),→) ⊆ B, where y is an arbitrary point of Yx. Thus B is open in this case. Next assume that Yx has a minimum ymin and x has an immediate predeces- sor x′. If Yx′ has no maximum, then there exists y ′ max := maxY + x′ ∈ Y + x′ \Yx′ . Since B = [(x,ymin),→) = ((x ′,y′max),→) + ∩ ( ⋃ x∈X {x}×Yx ) it follows that B is open. On the other hand, if Yx′ has a maximum, then B is open in the order topology τΣ and thus the same holds in the subspace topology of τ+Σ . To prove (iii), observe that C = A∩B, where A := ⋃ w∈(←,x′′){w}×Yw and B := ⋃ w∈[x,→){w}× Yw. Therefore, the claim follows from (the dual of) (i) and (ii). � Lemma 3.18. The inclusion ı⊗ : (⊗ x∈X Yx,�lex ) →֒ (∑ x∈XY + x ,τ + Σ ) is an order-preserving homeomorphic embedding. Proof. It suffices to prove that ı⊗ is a homeomorphic embedding. First we show that ı⊗ is continuous. Let S = (←, (x,y)) + ∩ (⋃ x∈X{x}×Yx ) be a subbasic open set in ı⊗ (⊗ x∈X Yx ) ; we show that ı−1⊗ (S) = S is open in ⊗ x∈X Yx. If (x,y) is such that y ∈ Yx, then Lemma 2.9 implies that S = (←, (x,y)) is open in τ⊗. Next, assume that y ∈ Y + x \ Yx; thus, either y = minY + x or y = maxY +x . If y = minY + x , then x has an immediate predecessor x ′ and Yx′ has a maximum y′max. By Lemma ??, S = (←, (x ′,y′max)] = π −1(←,x) is open in τ⊗. On the other hand, if y = minY + x , then S is the open ray (←, (x ′′,y′′min)), where x′′ is the immediate successor of x and y′′min is the minimum of Yx′′ . Next we show that ı−1⊗ : ı⊗ (⊗ x∈X Yx ) → ⊗ x∈X Yx is continuous. Let U⊗x V be a subbasic open set in S⊗; we prove that U⊗xV is open in ı⊗ (⊗ x∈X Yx ) ⊆∑ x∈X Y + x . By Lemma 2.3 and duality, it suffices to examine the following two cases: (a) U ⊗x V = X ⊗x (y,→) for some y ∈ Yx; (b) U ⊗x V = (x ′,→)⊗x Yx for some x′ ≺ x. For (a), Lemma 2.3 yields that X ⊗x (y,→) is equal to either {x}× (y,→) (if Yx has no maximum) or ((x,y),→) (if Yx has a maximum); in both cases the claim holds. For (b), the claim follows from Lemma 2.3, Lemma 2.11 and Lemma 3.17. � 230 A. Caserta, A. Giarlotta and S. Watson Theorem 3.19. ((∑ x∈XY + x ,τ + Σ ) , ı⊗ ) is the minimum unification of the GO- space (⊗ x∈X Yx,�lex ) . Proof. Since Lemma 3.18 implies that ((∑ x∈X Y + x ,τ + Σ ) , ı⊗ ) is a unification of (⊗ x∈X Yx,�lex ) , it suffices to show that it is minimum. Let (Z,ϕ) be a unification of ⊗ x∈X Yx. We define a TO-embedding ψ : ∑ x∈XY + x → Z such that ϕ = ψ ◦ ı⊗. Let (x,y) ∈ ∑ x∈XY + x . If (x,y) ∈ ⋃ x∈X{x}× Yx, then let ψ(x,y) := ϕ(ı−1⊗ (x,y)). If (x,y) ∈ ∑ x∈XY + x \ (⋃ x∈X{x}×Yx ) , then either (i) x has an immediate predecessor x′, Yx′ has a maximum and Yx has no minimum, or (ii) x has an immediate successor x′′, Yx′′ has a minimum and Yx has no maximum. We define ψ in case (ii) only, since (i) is dual to (ii). Observe that y ∈ Y +x \ Yx is the maximum of Y +x . If we denote y ′′ min := minYx′′ , then ((x,y),→) + = [(x′′,y′′min),→) + is an open ray in ∑ x∈XY + x . Thus the set [(x′′,y′′min),→) = ı −1 ⊗ ( [(x′′,y′′min),→) + ∩ ı⊗ (⊗x∈XYx) ) is open in ⊗ x∈X Yx. Since ϕ is a TO-embedding, the sets A := ϕ(←, (x ′′,y′′min)) and B := ϕ[(x′′,y′′min),→) are open in ϕ (⊗ x∈X Yx ) ; further, A ≺ minB = ϕ(x′′,y′′min). Note that there exists z0 ∈ Z such that A ≺ z0 ≺ minB, since otherwise B would fail to be open in ϕ (⊗ x∈X Yx ) ⊆ Z. Select such a z0 ∈ Z and define ψ(x,y) := z0. This completes the definition of ψ in case (ii). It is immediate to check that ψ is injective and order-preserving. Thus ψ is a TO-embedding by Lemma 3.2. � References [1] R. Engelking, General Topology (Heldermann Verlag, Berlin, 1989). [2] V. V. Fedorc̆uk, Bicompacta with noncoinciding dimensionalities, Soviet Math. Doklady, 9/5 (1968), 1148–1150. [3] K. P. Hart, J. Nagata and J.E. Vaughan (Eds.), Encyclopedia of General Topology (North-Holland, Amsterdam, 2004). [4] S. Watson, The Construction of Topological Spaces: Planks and Resolutions, in M. Hus̄ek and J. van Mill (eds.), Recent Progress in General Topology, 673–757 (North- Holland, Amsterdam, 1992). Received May 2005 Accepted June 2006 A. Caserta (agata.caserta@unina2.it) Department of Mathematics, Seconda Università degli Studi di Napoli, Caserta 81100, Italia On resolutions of linearly ordered spaces 231 A. Giarlotta (giarlott@unict.it) Department of Economics and Quantitative Methods, Università di Catania, Catania 95129, Italia S. Watson (watson@hilbert.math.yorku.ca) Department of Mathematics and Statistics, York University, Toronto M3J1P3, Canada