Presheaf-Based Universal Extensions of Quantale-Enriched Specialization Semilattices

Presheaf-Based Universal Extensions of Quantale-Enriched Specialization Semilattices Joaquim Reizi Higuchi November 29, 2025 Abstract We fix a commutative unital quantale Qand introduce Q-specialization semilattices with 0: Q-valued preorders equipped with a compatible join-semilattice structure. For each such Swe construct a canonical universal extension S7−→ U(S) inside the presheaf Q-category [Sop, Q]: the object U(S) is defined as the full sub-Q-category of Q-presheaves consisting of suitable Q-ideals, and carries a natural structure of principal additive Q-specialization semilattice with 0. We show that the enriched Yoneda embedding yS:S→U(S) exhibits Uas a reflection of the category QSpecSL0of Q-specialization semilattices with 0 into the full subcategory QAddPrinSL0of principal additive ones (Theorem A). Our main structural result (Theorem B) identifies U(S) with a semilattice-compatible core of the Isbell completion I(S) of the underlying Q-category (S, d). On the presheaf side we consider the MacNeille–Isbell nucleus JSand the Q-ideal closure cSon [Sop, Q], and we show that the composite NS=JS◦cSis a nucleus whose fixpoints form a full sub-Q-category I(S)add ⊆I(S). We prove that U(S) is canonically isomorphic, as a Qspecialization semilattice with 0, to I(S)add. In the Lawvere case Q= [0,∞] the universal extension U(S) coincides, as a metric space, with the Isbell completion and hence with the injective hull (tight span) of a metric space, now endowed with a canonical principal additive semilattice structure (Corollary C). Thus we obtain tight spans with structure. For Q= 2 we recover Lipparini’s universal extension of specialization semilattices with 0, and for frame-valued quantales Q= Ω(X) we obtain a sheaf-like bundle of local ideal completions over a topological space X. Keywords: quantales; enriched categories; specialization semilattices; presheaves; universal constructions; Isbell completion; tight span; frames and locales; many-valued logics. MSC (2020): 18D20; 06F07; 54F05; 54E35; 03B50. 1 Introduction Specialization orders and semilattice structures arise naturally throughout general topology, lattice theory, and domain theory. Given a topological space X, the specialization preorder associated with its topology records which points lie in the closure of which singletons. When this preorder is compatible with a join-semilattice structure, one obtains specialization semilattices, which provide a convenient algebraic framework for studying closure operators, continuous lattices, and related order-theoretic phenomena. In particular, Lipparini introduced and studied specialization semilattices with 0 and their universal extensions, showing that every such structure admits a canonical reflection into a principal additive specialization semilattice, constructed via an ideal completion and equipped with a suitable closure operator. Conceptually, 1 Lipparini’s construction explains how to “freely add principal generators” to a given specialization semilattice while preserving the interaction between order and finite joins. In parallel, the theory of quantales and quantale-enriched categories has developed into a unifying language for a wide range of mathematical phenomena. Lawvere observed that generalized metric spaces can be viewed as categories enriched over a suitable quantale of weights, while frames and locales can be seen as quantale-based encodings of topological or logical structure. In this enriched setting, presheaf constructions and universal completions such as the Isbell/MacNeille completion play a central role: they provide canonical ways to complete a given Q-category by freely adding weighted colimits and limits. In the Lawvere case Q= [0,∞], the Isbell completion of a generalized metric space recovers the injective hull or tight span of the underlying metric space; in the Boolean case Q= 2 it recovers the MacNeille completion of a preorder. Despite these parallel developments, there is at present no general framework that fully combines the algebraic viewpoint of specialization semilattices with the topological and metric viewpoints of quantale-enriched category theory. On the one hand, Lipparini’s theory is essentially “crisp”: it is formulated for ordinary preorders and Boolean-valued specialization relations, and it does not see the enriched structure underlying generalized metrics, fuzzy orders, or frame-valued specializations. On the other hand, the standard enriched completions such as the Isbell completion are “purely enriched”: they do not take into account additional semilattice structure on the underlying objects, and therefore do not provide a direct counterpart to Lipparini’s universal extension. The aim of this paper is to provide a best-of-both-worlds framework that integrates these two perspectives. We fix a commutative unital quantale Qand introduce a notion of Q-specialization semilattice with 0, which treats a specialization semilattice as a Q-enriched preorder equipped with a compatible join-semilattice structure. For each such Swe construct a canonical universal extension S7−→ U(S) using presheaves and the enriched Yoneda embedding, and we show that this extension can be identified with a semilattice-compatible core inside the Isbell completion of the underlying Q-category. In the Boolean case Q= 2 this recovers Lipparini’s universal extension; in the Lawvere metric case Q= [0,∞] it recovers the tight span, but now equipped with a natural principal additive semilattice structure. More concretely, we proceed as follows. We regard a Q-specialization semilattice (S, ∨, d, 0) as a Q-valued preorder d:S×S→Qtogether with a join-semilattice structure (S, ∨,0) satisfying axioms that lift Lipparini’s conditions to the enriched level. We then consider the presheaf Q-category [Sop, Q] and single out, inside it, a full sub-Q-category of Q-ideals, that is, Qpresheaves φ:S→Qsatisfying a normalisation at 0 and a ⊗-subadditivity condition along finite joins. We define U(S) := IdQ(S)⊆[Sop, Q] to be the Q-category of such Q-ideals, equipped with the induced hom-structure and the pointwise semilattice structure. The enriched Yoneda embedding yS:S→[Sop, Q] factors through U(S), and we show that (U(S), yS) is universal among maps from Sinto principal additive Q-specialization semilattices with 0. From a categorical point of view, this construction produces a reflection of the category of Q-specialization semilattices with 0 into its full subcategory of principal additive ones. From an enriched point of view, it realises this reflection internally to the presheaf category via the enriched Yoneda embedding. Conceptually, U(S) plays for Q-specialization semilattices the same role that the Isbell completion plays for general Q-categories, but with the additional feature that it is built to respect the semilattice structure. The heart of the paper is a structural comparison between this presheaf-based universal extension and the Isbell completion of the underlying Q-category. Let I(S) denote the Isbell 2 completion of (S, d); it can be described as the full sub-Q-category of [Sop, Q] consisting of those presheaves that are fixed by the MacNeille–Isbell nucleus JS=⇑⇓. We exploit two nuclei on the presheaf category: the MacNeille nucleus JSand the Q-ideal closure cSdetermined by IdQ(S). Their composite NS:= JS◦cS is a nucleus on [Sop, Q] which, when restricted to the Isbell completion I(S), yields a canonical idempotent endofunctor NS:I(S)−→ I(S). We show that the full sub-Q-category of fixed points I(S)add := Fix(NS)⊆I(S) can be identified as a semilattice-compatible core of the Isbell completion: its objects are precisely those Isbell points whose specialization structure is generated by principal Q-ideals and whose crisp specialization relation is principal additive in Lipparini’s sense. Our main comparison theorem (Theorem B) states that the presheaf-based universal extension and this core of the Isbell completion agree: The universal extension U(S) is canonically isomorphic, as a Q-specialization semilattice with 0, to the semilattice-compatible core I(S)add of the Isbell completion of S. In other words, U(S) is not merely a convenient completion built inside a presheaf category: it is precisely the piece of the Isbell completion that still carries a natural principal additive semilattice structure. This identification explains, in a uniform and conceptual way, why known enriched completions such as tight spans and localic bundles admit natural semilattice structures and how they relate to ideal completions of specialization semilattices. A particularly transparent instance of our construction occurs in the Lawvere metric case. Let Q= [0,∞] with its standard Lawvere quantale structure, and let (S, d) be a generalized metric space. The Isbell completion I(S) is known to coincide with the injective hull (or tight span) of (S, d): it is the smallest hyperconvex metric space into which Sembeds isometrically. We show that in this situation U(S) identifies with I(S) as a metric space, and that the presheafbased construction endows the tight span with a canonical principal additive specialization semilattice structure. Thus, in the finite symmetric case, the tight span of a metric space is not only the minimal injective hull in the metric sense, but also the universal principal additive completion in the sense of specialization semilattices. We refer to this phenomenon as a tight span with structure. From this perspective, our results can be summarised as providing a semilattice-compatible core inside the Isbell completion, and showing that this core coincides with the presheaf-based universal extension U(S) of Sas a Q-specialization semilattice. This bridges Lipparini’s algebraic theory of specialization semilattices with the enriched theory of presheaves, nuclei, and Isbell completions. It also offers a unified language for a number of constructions that have so far been studied in different communities: universal extensions of specialization semilattices, injective hulls and tight spans of metric spaces, and localic completions in topology and logic. Main contributions. The main contributions of the paper can be summarised as follows. •We introduce the notion of a Q-specialization semilattice with 0 for a fixed commutative unital quantale Q, together with principal additive objects and appropriate morphisms. This yields categories QSpecSL0and QAddPrinSL0which extend Lipparini’s setting from Q= 2 to a general quantale. 3 •For each such Swe define U(S) as the full sub-Q-category of [Sop, Q] consisting of Q-ideals, and we show that U(S) carries a canonical structure of principal additive Q-specialization semilattice with 0. The enriched Yoneda embedding yS:S→U(S) is a morphism of Q-specialization semilattices with 0. •We prove that the assignment S7→ U(S) extends to a functor U:QSpecSL0−→ QAddPrinSL0 which is left adjoint to the inclusion J:QAddPrinSL0,→QSpecSL0. Equivalently, (U(S), yS) is characterised up to isomorphism by a universal property: every morphism from Sto a principal additive Q-specialization semilattice with 0 factors uniquely through yS(Theorem A). •We construct a MacNeille nucleus NSon the Isbell completion I(S) whose fixed points form a full sub-Q-category I(S)add ⊆I(S), and we identify I(S)add as the semilatticecompatible core of I(S). We show that there is a canonical isomorphism U(S)∼ =I(S)add in QSpecSL0(Theorem B), thereby relating the presheaf-based universal extension U(S) to the internal structure of the Isbell completion. •We specialise this framework to three classes of quantales: 1. For Q= 2 we recover Lipparini’s universal extension of specialization semilattices with 0 via ideal completion, now seen as an instance of the general Q-presheaf construction. 2. For the Lawvere quantale Q= [0,∞] we identify U(S) with the Isbell completion and, in the finite symmetric case, with the tight span, equipped with a canonical principal additive specialization semilattice structure (Corollary C). 3. For Q= Ω(X), the frame of opens of a topological space X, we interpret U(S) as a sheaf-like bundle of local universal extensions over X, whose fibres are ideal completions of the crisp specialization semilattices determined by each point of X. Organisation of the paper. Section 2 recalls basic facts about quantales, Q-categories, and specialization semilattices with 0, including Lipparini’s universal extension in the Boolean case. In Section 3 we introduce Q-specialization semilattices with 0, principal additive objects, and the categories QSpecSL0and QAddPrinSL0. Section 4 develops the presheaf-based construction U(S) = IdQ(S) and shows that the enriched Yoneda embedding yS:S→U(S) is a morphism of Q-specialization semilattices. The universal property of Uis proved in Section 5, where we show that Uis left adjoint to the inclusion of principal additive objects (Theorem A). In Section 6 we compare U(S) with the Isbell completion I(S) and identify U(S) with the semilattice-compatible core I(S)add (Theorem B). Section 7 is devoted to examples and applications, focusing on the Lawvere metric case, tight spans with semilattice structure, frame-valued specialisation, and the recovery of Lipparini’s classical universal extension. We conclude in Section 8 with a discussion of non-commutative quantales, connections to quantale-valued logics and generalized algebraic structures, and several directions for future work. 2 Preliminaries In this section we briefly recall the basic notions from quantale theory and Q-enriched category theory that we will use throughout the paper, as well as the classical notion of specialization semilattice with 0 due to Lipparini. The reader familiar with quantales and enriched categories may wish to skim Sections 2.1 and 2.2 and focus on Section 2.3. 4 2.1 Commutative unital quantales We start by recalling the definition of a quantale. Definition 2.1. A(unital) quantale is a structure Q= (Q, ≤,⊗, k) consisting of a complete lattice (Q, ≤), a binary operation ⊗:Q×Q→Q, and a distinguished element k∈Qsuch that: 1. (Q, ⊗, k) is a monoid: (a⊗b)⊗c=a⊗(b⊗c), k ⊗a=a=a⊗k; 2. ⊗distributes over arbitrary joins in each variable: a⊗_ i bi=_ i (a⊗bi),_ i ai⊗b=_ i (ai⊗b) for all families {ai}i,{bi}i⊆Q. If, in addition, ⊗is commutative, we speak of a commutative unital quantale. A quantale is called integral if the monoidal unit kis the top element of the lattice (Q, ≤). Typical examples to keep in mind include: •the two-element Boolean algebra Q=2={0<1}with ⊗=∧and k= 1; •the Lawvere quantale Q= [0,∞] with the reversed order a≤Qbiff a≥b(in the usual order), monoidal product a⊗b=a+b(truncated at ∞), and unit k= 0; •the frame of opens Ω(X) of a topological space X, with order given by inclusion, monoidal product ⊗=∩, and unit k=X. A crucial feature of a quantale is the existence of an internal hom (or residuation) making ⊗a left adjoint in each variable. Definition 2.2. Let Q= (Q, ≤,⊗, k) be a quantale and fix a∈Q. The right residual of ais the map a⇒(−): Q→Q defined by a⇒c:= _{b∈Q|a⊗b≤c}(c∈Q). By completeness of (Q, ≤), the join above always exists. The definition implies the fundamental Galois connection a⊗b≤c⇐⇒ b≤a⇒c(∀a, b, c ∈Q). When Qis commutative, the right and left residuals coincide and we will write simply a⇒c for the unique element satisfying this adjointness equation. The residual will play a central role in the Q-enriched homs of the presheaf categories [Sop, Q], where it appears in the formula [Sop, Q](φ, ψ) = ^ x∈Sφ(x)⇒ψ(x), see Section 4. Throughout the rest of the paper we assume, unless explicitly stated otherwise, that Qis a fixed commutative unital quantale. In several results (notably in Section 5) we will furthermore assume that Qis integral. 5 2.2 Q-categories and Q-preorders We next recall the basic notions of Q-enriched category theory in the sense of Kelly. Since in this paper we are mainly interested in enriched preorders, we restrict attention to small Q-categories whose hom-objects are elements of Q. Definition 2.3. AQ-category Sconsists of •a set of objects, denoted Ob(S); •for each pair of objects a, b, an element S(a, b)∈Q(the hom-value); such that, for all a, b, c ∈Ob(S), (enriched identities) k≤ S(a, a), (enriched composition) S(a, b)⊗ S(b, c)≤ S(a, c). In the case where every hom-set is a mere truth value (i.e. a single element of Q), a Qcategory is completely determined by a map d:S×S−→ Q, where S= Ob(S), satisfying k≤d(a, a), d(a, b)⊗d(b, c)≤d(a, c). Such data is often called a Q-valued preorder on S, and we will freely switch between the notations (S, d) and S= (S, d) when convenient. Definition 2.4. Let (S, dS) and (T, dT) be Q-categories of this form. A Q-functor F: (S, dS)→ (T, dT) is a function F:S→Tsuch that dS(a, b)≤dT(F(a), F(b)) (∀a, b ∈S). Thus a Q-functor does not decrease the hom-value between any two objects. When Q= 2 we recover preorders and monotone maps; when Q= [0,∞] with the Lawvere quantale structure we obtain generalised metric spaces and non-expansive maps. The one-object Q-category Qitself plays a distinguished role: its single object ∗has Q(∗,∗) = Qas hom-value, and a Q-functor Sop →Qis precisely what we will call a Q-presheaf on S. Unfolding the definition, such a functor is a map φ:S→Qsatisfying dS(b, x)⊗φ(x)≤φ(b) (∀b, x ∈S), which is exactly the form of the presheaf condition we will use in Section 4. 2.3 Lipparini’s specialization semilattices with 0 We now recall the classical notion of specialization semilattice with 0 due to Lipparini, which provides the starting point and guiding example for our quantale-enriched generalisation. Definition 2.5. Aspecialization semilattice with 0 is a quadruple (S, ∨,⊑,0) such that: 1. (S, ∨,0) is a join-semilattice with least element 0; 2. ⊑is a preorder on S(reflexive and transitive); 3. the following compatibility axioms hold for all a, a1, b ∈S: 6 (S1) the semilattice order is finer than ⊑: a≤b=⇒a⊑b; (S2) if a⊑band b⊑cthen a⊑c; (S3) if a⊑band a1⊑bthen a∨a1⊑b; (S4) if a⊑0 then a= 0. Here, as usual, the underlying order ≤on Sis defined by a≤bif and only if a∨b=b. Axiom (S1) says that this order is contained in the specialization preorder ⊑, (S3) expresses the stability of ⊑under finite joins, and (S4) is a separation condition for the least element. These structures arise naturally when one considers specialization orders coming from closure operators or from topological spaces equipped with compatible semilattice operations. Lipparini showed that every specialization semilattice with 0 admits a canonical universal extension into a principal additive specialization semilattice with 0, obtained via an ideal completion and equipped with a closure operator Ksatisfying K(a∨b) = K(a)∨K(b), K(0) = 0. This extension has a universal property expressing that any homomorphism from Sinto a principal additive specialization semilattice with 0 factors uniquely through the canonical embedding S→Se. In the enriched setting of the present paper, the case Q= 2 will recover this construction exactly (see Section 7), while for more general quantales Qwe obtain a uniform generalisation via presheaves and the enriched Yoneda embedding. 3Q-specialization semilattices Throughout this section we fix a commutative unital integral quantale Q= (Q, ≤,⊗, k). We recall that a Q-preorder on a set Sis a map d:S×S→Qsuch that k≤d(a, a), d(a, b)⊗d(b, c)≤d(a, c) (∀a, b, c ∈S), and that we write a⊑db:⇐⇒ k≤d(a, b) for the associated crisp specialization relation. 3.1 Definition We begin by lifting Lipparini’s notion of specialization semilattice with 0 to the quantaleenriched setting. Definition 3.1. AQ-specialization semilattice with 0 is a quadruple (S, ∨, d, 0) consisting of •a join-semilattice (S, ∨) with a distinguished least element 0 ∈S; •aQ-preorder d:S×S→Q; such that the following conditions hold for all a, a1, b ∈S: (QS1) If a≤bin the underlying semilattice order (i.e. a∨b=b), then k≤d(a, b), equivalently a⊑db. 7 (QS2) The relation ⊑dis compatible with the semilattice order in the sense that a⊑db, b ⊑dc=⇒a⊑dc, which is automatic from transitivity of d. (QS3) The specialization degrees are submultiplicative with respect to finite joins: d(a, b)⊗d(a1, b)≤d(a∨a1, b). Condition (QS1) states that the enriched specialization extends the underlying semilattice order; (QS3) says that if two elements specialise to bto degree d(a, b) and d(a1, b), then their join specialises to bat least to the composite degree d(a, b)⊗d(a1, b). Since 0 is the least element, (QS1) implies in particular that 0≤b⇒k≤d(0, b) (∀b∈S), so 0 is always “maximally specialised” to every element. Remark 3.2 (Boolean case).When Q= 2 = {0<1}with ⊗=∧and k= 1, a Q-preorder dis the same as an ordinary preorder ⊑via d(a, b) = 1 iff a⊑b. In this case the axioms above say: •(QS1) a≤b⇒a⊑b; •(QS2) ⊑is transitive and reflexive; •(QS3) if a⊑band a1⊑bthen a∨a1⊑b. Thus (S, ∨, d, 0) is a 2-specialization semilattice with 0 if and only if (S, ∨,⊑,0) is a specialization semilattice with 0 in the sense of Lipparini. In particular, our notion is a direct quantale-enriched lift of the classical one. We shall often write simply Q-specialization semilattice when the presence of 0 is understood. 3.2 Morphisms and the category QSpecSL0 We next define the morphisms between Q-specialization semilattices. As usual in enriched category theory, we adopt a lax inequality condition on hom-values. Definition 3.3. Let (S, ∨, dS,0) and (T, ∨, dT,0) be Q-specialization semilattices with 0. A Q-specialization semilattice homomorphism (or Q-SSL homomorphism) is a function f:S→T such that: 1. fis a semilattice homomorphism preserving 0: f(a∨a1) = f(a)∨f(a1), f(0) = 0; 2. fis lax Q-monotone with respect to dSand dT: dS(a, b)≤dTf(a), f(b)(∀a, b ∈S).(3.1) Condition (3.1) is the usual enriched functor inequality: the degree to which aspecializes to bin Sis bounded above by the degree to which f(a) specializes to f(b) in T. 8 Remark 3.4 (Strict vs. lax morphisms).One could consider a stricter notion of morphism requiring equality dS(a, b) = dTf(a), f(b) for all a, b ∈S. This would amount to an isometric embedding of enriched hom-values. For our purposes, however, the lax condition (3.1) is more natural: it already implies monotonicity with respect to the crisp specialization relations, and it is precisely the condition that arises from viewing Sand Tas Q-categories and fas a Q-functor. The universal property of the presheaf-based extension U(S) is most naturally formulated in this lax setting. Proposition 3.5. If f:S→Tis a Q-SSL homomorphism, then fis monotone with respect to the crisp specialization relations: a⊑dSb=⇒f(a)⊑dTf(b). Proof. If a⊑dSbthen k≤dS(a, b). By (3.1) we have dS(a, b)≤dT(f(a), f(b)), hence k≤dTf(a), f(b), which means f(a)⊑dTf(b). We also single out morphisms that are “injective and order-reflecting” on the crisp level. Definition 3.6. AQ-SSL homomorphism f:S→Tis called an embedding if 1. fis injective on the underlying sets; 2. freflects crisp specialization: f(a)⊑dTf(b) =⇒a⊑dSb(∀a, b ∈S). In particular, embeddings are injective morphisms that realise Sas a (crisp) sub-specialization semilattice of T. The Yoneda embeddings yS:S→U(S) constructed later will be embeddings in this sense. Definition 3.7. We write QSpecSL0 for the category whose objects are Q-specialization semilattices with 0 and whose morphisms are Q-SSL homomorphisms. It is straightforward to verify that composition of Q-SSL homomorphisms is again a Q-SSL homomorphism and that identities are such, so that QSpecSL0is indeed a category. 3.3 Principal additive structure We now recall and adapt the notion of principal additive structure from the introduction. The key point is that “principal” and “additive” are determined entirely by the crisp specialization relation. Let (S, ∨, d, 0) be a Q-specialization semilattice with 0, and let ⊑dbe the associated crisp specialization relation. For each a∈Swe define the down-set Sa:= {b∈S|b⊑da}. 9 We write IdQ(S)⊆P(S) for the full sub-Q-category spanned by the Q-ideals, and we recall that we defined the universal extension of Sto be U(S) := IdQ(S), with underlying semilattice structure given by pointwise joins and least element. In particular, U(S) is a full sub-Q-category of P(S), and the enriched Yoneda embedding yS:S→P(S) lands in U(S). The Q-ideals form a closure system inside the complete lattice underlying P(S). Lemma 6.2. The class IdQ(S)of Q-ideals is closed under arbitrary meets in P(S)(taken pointwise). Equivalently, IdQ(S)is a full reflective subcategory of P(S). Proof. Let {φi}i∈I⊆IdQ(S) and define φ:= Vi∈Iφipointwise. Since Qis a complete lattice and ⊗preserves arbitrary joins in each variable, it preserves arbitrary meets in each variable when the order is reversed; thus (I0) and (I∨) for each φipass to (I0) and (I∨) for φ. The presheaf condition is similarly inherited from the φi, using the fact that dand ⇒are compatible with arbitrary meets. Hence φ∈IdQ(S). As a consequence, there is a nucleus on P(S) selecting the Q-ideal hull of each presheaf. Definition 6.3. The Q-ideal closure cS:P(S)−→ P(S) is defined by cS(φ) := ^{θ∈IdQ(S)|φ≤θ}, where the inequality is pointwise. Equivalently, cSis left adjoint to the inclusion IdQ(S),→P(S). Proposition 6.4. The map cSis a nucleus on P(S): it is inflationary, monotone, and idempotent. Moreover, its fixpoints are exactly the Q-ideals, Fix(cS) = IdQ(S). Proof. Inflationarity and monotonicity are immediate from the definition. Idempotence follows from the fact that IdQ(S) is a reflective full subcategory: once we have applied cSonce, applying it again does not change the object. The characterisation of fixpoints holds because cS(φ) = φ if and only if φis itself a Q-ideal. 6.3 The combined nucleus NSand its fixed points We now combine the MacNeille nucleus JSand the ideal nucleus cSinto a single nucleus on the presheaf category. Definition 6.5. The Isbell–ideal nucleus NS:P(S)−→ P(S) is defined as the composite NS:= JS◦cS. Lemma 6.6. The map NSis a nucleus on P(S), with Fix(NS) = Fix(JS)∩Fix(cS) = I(S)∩IdQ(S). 16 Proof. Since both JSand cSare nuclei on P(S), their composite NS=JS◦cSis inflationary and monotone. The idempotence of NSfollows from the general fact that the composite of two nuclei is again a nucleus (the proof is a straightforward calculation using idempotence and inflationarity of JSand cS). For the fixpoints, let us write C=cSand D=JSfor brevity. Suppose φ∈Fix(C)∩Fix(D), i.e. C(φ) = φand D(φ) = φ. Then NS(φ) = D(C(φ)) = D(φ) = φ, so φ∈Fix(NS). Conversely, if φ∈Fix(NS) then φ=D(C(φ)). Since Cis inflationary, we have φ≤C(φ), while Dis inflationary so C(φ)≤D(C(φ)) = φ. Hence C(φ) = φ, and therefore D(φ) = D(C(φ)) = φas well. Thus φ∈Fix(C)∩Fix(D), establishing the identity Fix(NS) = Fix(C)∩Fix(D). Substituting C=cS,D=JSand using Propositions 6.4 and the definition of I(S) we obtain Fix(NS) = I(S)∩IdQ(S). We now define the semilattice-compatible core of the Isbell completion as the full sub-Qcategory of I(S) consisting of the fixed points of NS. Definition 6.7. The semilattice-compatible core of the Isbell completion I(S) is the full subQ-category I(S)add := I(S)∩IdQ(S) = Fix(NS)⊆I(S) spanned by those Isbell-fixed presheaves which are also Q-ideals. Since I(S)add is defined as a full sub-Q-category of [Sop, Q], it inherits a Q-specialization semilattice structure with 0 from its ambient Q-category, obtained by restricting the pointwise join and least element. One checks that NSrestricts to a nucleus on I(S), so that I(S)add is again the fixpoint category of a nucleus and therefore reflective inside I(S). 6.4 The comparison functor and Theorem B We are now ready to relate the universal extension U(S) to the semilattice-compatible core I(S)add. Recall that U(S) = IdQ(S) is the full sub-Q-category of [Sop, Q] spanned by the Q-ideals. We define a comparison functor by restricting the MacNeille nucleus JS. Definition 6.8. The comparison functor ES:U(S)−→ I(S)add is defined by ES(φ) := JS(φ) for each φ∈U(S) = IdQ(S), viewed as an object of I(S)add via Lemma 6.6. Since U(S) and I(S)add are full subcategories of P(S), and JSis a Q-functor, it is immediate that ESis a Q-functor and a morphism of Q-specialization semilattices with 0. Moreover, ES is surjective on objects: if ψ∈I(S)add ⊆IdQ(S) is fixed by JSthen ES(ψ) = JS(ψ) = ψ. 17 Theorem 6.9 (Theorem B: comparison with Isbell completion).For each Q-specialization semilattice with 0 (S, ∨, d, 0), the comparison functor ES:U(S)−→ I(S)add is an isomorphism in QSpecSL0. Equivalently, the presheaf-based universal extension U(S)is canonically isomorphic, as a Q-specialization semilattice with 0, to the semilattice-compatible core I(S)add of the Isbell completion I(S). Proof sketch. We outline the main steps and refer to the general theory of nuclei on Q-categories for some of the technical details. Step 1: I(S)add is principal additive. By construction, I(S)add is a full sub-Q-category of U(S) and inherits the semilattice structure and principal additive closure from U(S). In particular, the crisp specialization relation on I(S)add is the restriction of that on U(S), and the principal additive closure operator on U(S) restricts to one on I(S)add. Thus I(S)add is an object of QAddPrinSL0, and the inclusion I(S)add ,→QSpecSL0 agrees with the inclusion J:QAddPrinSL0,→QSpecSL0on objects. Step 2: ESis a morphism in QSpecSL0.Since both U(S) and I(S)add are full sub-Q-categories of P(S) and the semilattice structure is pointwise, the MacNeille nucleus JSpreserves finite joins and 0, and it is a Q-functor. Hence ESis a Q-specialization semilattice homomorphism with 0, i.e. ESis a morphism in QSpecSL0. Step 3: universality of U(S)and I(S)add.By Theorem 5.1, the pair (U(S), yS) is initial among morphisms from Sinto principal additive Q-specialization semilattices with 0. On the other hand, the construction of I(S)add via the nucleus NSshows that (I(S)add, ηS), where ηS:SyS −→ P(S)NS −−→ I(S)add, enjoys the same universal property: for any T∈QAddPrinSL0and any morphism f:S→ J(T) in QSpecSL0, there exists a unique ¯ f:I(S)add →Tin QAddPrinSL0such that f= J(¯ f)◦ηS. This follows from the fact that NSis a nucleus whose fixpoints form a full reflective subcategory of U(S), and hence of QAddPrinSL0. Step 4: uniqueness of the reflection. The left adjoint to the inclusion functor J:QAddPrinSL0,→ QSpecSL0is unique up to canonical isomorphism. Since both S7−→ U(S) and S7−→ I(S)add realise this left adjoint on objects (by Steps 1–3), there exists a unique natural isomorphism αS:U(S)∼ = −−→ I(S)add in QAddPrinSL0whose composite with the inclusion I(S)add ,→QSpecSL0agrees with ES. In particular, ESis an isomorphism in QSpecSL0, as required. 7 Examples and applications In this final section we illustrate the abstract machinery developed above in three concrete settings. The Boolean case Q= 2 recovers Lipparini’s universal extension of specialization semilattices with 0. The Lawvere metric case Q= [0,∞] yields tight spans with semilattice structure, as anticipated in the Introduction. Finally, for frame-valued quantales Q= Ω(X) we obtain a sheaf-like bundle of local completions over a topological space X. 18 7.1 The Boolean case Q= 2 We first briefly revisit the case Q= 2, which provides a useful sanity check and conceptual guide. Here Q=2={0<1}is the two-element Boolean algebra with ⊗=∧and k= 1. A Q-preorder d:S×S→2 is the same as an ordinary preorder ⊑on Svia d(a, b) = 1 iff a⊑b. A 2-specialization semilattice with 0 (S, ∨, d, 0) is precisely a specialization semilattice with 0 in Lipparini’s sense: the axioms (QS1)–(QS3) reduce to a≤b⇒a⊑b, a ⊑b, a1⊑b⇒a∨a1⊑b. A 2-presheaf φ:S→2 satisfying the presheaf condition is the same as a down-set for ⊑, and the conditions (I0) and (I∨) say that this down-set is non-empty (contains 0) and closed under finite joins. Thus Q-ideals are exactly the usual ideals of S, and the universal extension U(S) = Id2(S) can be identified with the poset of all ideals of S, ordered by inclusion and equipped with the join given by ideal-theoretic sum. The enriched Yoneda embedding yS:S→U(S) sends each a∈Sto the principal ideal I(a) = {x∈S|x⊑a}, and Theorem 6.9 specialises to Lipparini’s universal extension: U(S) is the canonical principal additive specialization semilattice with 0 receiving a universal morphism from S, and its underlying poset is obtained by an ideal completion. 7.2 The Lawvere metric case Q= [0,∞] We now turn to the Lawvere quantale and explain how the abstract construction recovers, and enriches, tight spans and injective hulls of metric spaces. The Lawvere quantale and enriched metrics Let [0,∞] denote the set of extended non-negative reals. We equip it with the Lawvere quantale structure Qmet =[0,∞],≤Qmet ,⊗, k:= [0,∞],≥,+,0, that is, the order is reversed (so a≤Qmet biff a≥bin the usual order), the monoidal product is ordinary addition, and the unit kis 0. AQmet-preorder d:S×S→[0,∞] satisfies 0≤Qmet d(x, x), d(x, z)≤Qmet d(x, y)⊗d(y, z), which unpack as d(x, x)=0, d(x, z)≤d(x, y) + d(y, z) in the usual order. Thus a Qmet-preorder is exactly a generalised (Lawvere) metric on S. In this subsection we regard (S, d) primarily as a Lawvere metric space. Any additional semilattice structure (S, ∨,0) can be adjoined independently; it plays no restrictive role in the metric analysis that follows. 19 Q-presheaves and Q-ideals as potentials AQmet-presheaf on Sis a function φ:S→[0,∞] such that d(b, x) + φ(x)≥φ(b) (∀b, x ∈S).(7.1) Thus φis a 1-Lipschitz potential: the value at bcannot exceed the value at xplus the distance from bto x. The internal hom of Qmet is given by a⇒c:= max{0, c −a}, and the enriched hom in P(S) = [Sop, Qmet] between two presheaves φ, ψ is P(S)(φ, ψ) = ^ x∈Sφ(x)⇒ψ(x)= sup x∈S max{0, ψ(x)−φ(x)} in the usual order. This is the familiar Lawvere distance of ψabove φ. The Q-ideal conditions (I0) and (I∨) specialise to: 1. φ(0) = 0 (normalisation at the distinguished point 0); 2. subadditivity along joins: φ(a∨a1)≤φ(a) + φ(a1) (∀a, a1∈S). Thus a Qmet-ideal is a 1-Lipschitz potential normalised at 0 and subadditive with respect to the join-semilattice structure on S. The universal extension U(S) = IdQmet (S) can therefore be identified with a full subspace of the function space [0,∞]Sendowed with the Lawvere metric dU(φ, ψ) = sup x∈S max{0, ψ(x)−φ(x)}. If a symmetric metric is desired, one can pass to the associated symmetrisation dsym U(φ, ψ) := max{dU(φ, ψ), dU(ψ, φ)}= sup x∈Sψ(x)−φ(x). The pointwise join endows U(S) with a join-semilattice structure: (φ∨ψ)(x) := max{φ(x), ψ(x)}, and 0U(S)is given by the potential which is 0 at the basepoint and ∞elsewhere (or by a suitable normalised choice when the metric space has no distinguished 0). Lawvere case of Theorem B and Corollary C We now specialise Theorem 6.9 to the Lawvere quantale and recover the tight span with semilattice structure. Proposition 7.1. Let (S, d)be a Lawvere metric space. Then, as Lawvere metric spaces, the universal extension U(S)and the Isbell completion I(S)are isometric. In particular, U(S)is hyperconvex and injective in the usual metric sense. 20 Proof sketch. By Theorem 6.9, U(S) is isomorphic, as a Qmet-category, to the semilatticecompatible core I(S)add of the Isbell completion I(S). On the other hand, the underlying Qmet-category structure of I(S) is exactly the usual Lawvere metric of the Isbell completion of (S, d), and the inclusion I(S)add ,→I(S) is an isometry. Hence U(S) is isometric to a full subspace of I(S) which contains the image of Sunder the Yoneda embedding and is closed under the relevant weighted limits and colimits. The general theory of Isbell completions (see, e.g., [4, 2]) then implies that this subspace coincides with I(S) as a metric space. In the finite symmetric case we can sharpen this to an identification with the tight span. Corollary 7.2 (Tight span with structure).Let (S, d)be a finite metric space (symmetric and satisfying the usual separation axioms), regarded as a Lawvere metric space via Qmet. Then: 1. As a metric space, U(S)is isometric to the tight span (injective hull) of (S, d). 2. The pointwise join on U(S)induces a canonical structure of principal additive specialization semilattice with 0on the tight span, and under this structure U(S)is the universal principal additive completion of (S, d)in QSpecSL0. Proof sketch. By standard results of Isbell and Dress (see [2, 1, 4]), the Isbell completion of a finite symmetric metric space (S, d) is isometric to its tight span (injective hull). Proposition 7.1 identifies U(S) with this completion as a Lawvere metric space. Transporting the principal additive structure of U(S) (given by Theorem 5.1) along this isometry yields a canonical principal additive specialization semilattice structure on the tight span. The universal property of U(S) as a reflection into principal additive objects (Theorem 5.1) then yields the claimed universal property of the tight span in QSpecSL0. Thus in the Lawvere case our construction produces a tight span with structure: the usual tight span of (S, d), now equipped with a canonical principal additive specialization semilattice structure which is universal among such structures compatible with the metric. A two-point example We conclude the Lawvere case with a simple but illuminating example that already exhibits the interaction between metric and semilattice structures. Let S={0, a}with metric d(0,0) = d(a, a)=0, d(0, a) = d(a, 0) = r > 0. We view (S, d) as a Lawvere metric space over Qmet. AQmet-presheaf φ:S→[0,∞] is determined by the values φ(0) and φ(a), which we denote by t0and ta, subject to the presheaf inequalities d(b, x) + φ(x)≥φ(b) (b, x ∈ {0, a}). Writing these out gives: b=x= 0 : 0 + t0≥t0(trivial), b=x=a: 0 + ta≥ta(trivial), b= 0, x =a:r+ta≥t0, b=a, x =0: r+t0≥ta. Imposing the normalisation φ(0) = 0 (condition (I0)) we obtain t0= 0 and thus ta≤rfrom the last inequality, while the third inequality is automatic. Therefore every Q-ideal is of the form φt(0) = 0, φt(a) = t 21 for a unique t∈[0, r], and conversely each t∈[0, r] defines a Q-ideal. The universal extension U(S) can thus be identified with the closed interval [0, r]: U(S)∼ ={φt|t∈[0, r]}. The Lawvere metric on U(S) is dU(φt, φs) = supmax(0, φs(0) −φt(0)),max(0, φs(a)−φt(a))= max{0, s −t}, and the symmetrisation gives the usual Euclidean metric |s−t|on [0, r]. The pointwise join on U(S) is (φt∨φs)(0) = 0,(φt∨φs)(a) = max{t, s}, so the semilattice structure on [0, r] is given by the usual pointwise maximum: t∨s:= max{t, s},0U(S):= 0. Geometrically, the Yoneda embedding sends 0 and ato the endpoints 0 and rof the interval [0, r], while U(S) fills in all points at distance tfrom 0 and r−tfrom a. This is precisely the classical tight span of a two-point metric space, now equipped with a natural principal additive semilattice structure in which joins are computed by taking the larger of two coordinates. 7.3 Frame-valued specialization We briefly sketch the case of frame-valued quantales, which yields a localic bundle of universal extensions. Let Xbe a topological space and let Q= Ω(X) be its frame of open sets, regarded as a commutative unital quantale under inclusion and finite intersections. A Q-valued relation d:S×S→Ω(X) assigns to each pair (a, b) of elements of San open set d(a, b)⊆X, which can be interpreted as the “region of X” where bspecializes to a. For each point x∈Xwe obtain an ordinary specialization preorder on Sby declaring a⊑xb:⇐⇒ x∈d(a, b). AQ-presheaf φ:S→Ω(X) then assigns to each a∈San open set φ(a), subject to the condition d(b, x)∩φ(x)⊆φ(b) (∀b, x ∈S). Conditions (I0) and (I∨) require that X⊆φ(0), φ(a∨a1)⊇φ(a)∩φ(a1), so that φis a kind of “open-valued ideal” on S. For each x∈Xwe can evaluate a Q-presheaf φat xto obtain a 2-valued presheaf φx(a) :=    1 if x∈φ(a), 0 otherwise, which is an ideal of the crisp specialization semilattice (S, ∨,⊑x,0). The universal extension U(S) therefore assembles, over each point x∈X, the ideal completion of the fibre Sxtogether with coherence conditions expressing the locality of these completions in the topology of X. More precisely, Theorem 6.9 identifies U(S) = IdQ(S) with a semilattice-compatible core I(S)add of the Isbell completion of the Q-category (S, d). The frame-valued interpretation of (S, d) as a bundle of specialisation orders over Xthen transfers to U(S) and I(S)add, yielding a sheaf-like bundle of local ideal completions. We leave a detailed localic analysis of this situation to future work; here the example serves mainly to illustrate that the presheaf-based universal extension U(S) behaves well in topological as well as metric contexts. 22 8 Discussion and further directions The main purpose of this paper has been to unify two lines of work that have so far developed largely independently: Lipparini’s algebraic theory of specialization semilattices with 0 and the enriched category-theoretic theory of quantale-valued preorders, presheaves, and Isbell completions. In this final section we briefly summarise our contributions and outline several directions for future research. 8.1 Summary of the results For a fixed commutative unital integral quantale Qwe introduced the notion of a Q-specialization semilattice with 0, treating a specialization semilattice as a Q-valued preorder endowed with a compatible join-semilattice structure. We defined the category QSpecSL0of such structures and its full subcategory QAddPrinSL0of principal additive objects. Our first main result (Theorem 5.1, Theorem A) constructs, for each Q-specialization semilattice S, a universal extension S7−→ U(S) as a full sub-Q-category of the presheaf category [Sop, Q], consisting of Q-ideals. The enriched Yoneda embedding yS:S→U(S) exhibits U(S) as a reflection of QSpecSL0into QAddPrinSL0: every morphism from Sinto a principal additive Q-specialization semilattice with 0 factors uniquely through yS. The second main result (Theorem 6.9, Theorem B) compares this presheaf-based universal extension with the Isbell completion I(S) of the underlying Q-category (S, d). On the presheaf side we considered two nuclei on [Sop, Q]: the MacNeille–Isbell nucleus JS=⇑⇓ and the Q-ideal closure cS. Their composite NS:= JS◦cS is again a nucleus, whose fixpoints define a full sub-Q-category I(S)add ⊆I(S) that we interpret as the semilattice-compatible core of the Isbell completion. We proved that U(S) is canonically isomorphic, as a Q-specialization semilattice with 0, to I(S)add. Thus U(S) is not only constructed inside a presheaf category but can be recognised as an intrinsic part of the Isbell completion. Our third main contribution (Corollary 7.2, Corollary C) specialises this comparison to the Lawvere quantale Q= [0,∞]. In this case Q-ideals are 1-Lipschitz potentials which are normalised and subadditive along joins, and the universal extension U(S) coincides, as a metric space, with the Isbell completion of the Lawvere metric space (S, d). In the finite symmetric case this is the classical tight span (injective hull) of (S, d), now equipped with a canonical principal additive specialization semilattice structure. We refer to this structure as a tight span with structure. The Boolean case Q= 2 recovers Lipparini’s universal extension via ideal completion, while the frame-valued case Q= Ω(X) yields a localic bundle of ideal completions varying over a space X. 8.2 Non-commutative quantales and asymmetric specialisation Throughout the paper we have assumed that the base quantale Q= (Q, ≤,⊗, k) is commutative. This hypothesis enters crucially in the formulation of (QS3), where we require d(a, b)⊗d(a1, b)≤d(a∨a1, b), and in the definition of Q-ideals, where the subadditivity condition is expressed using ⊗in a symmetric way. It is natural to ask to what extent the theory can be extended to noncommutative quantales. 23 A first step would be to distinguish left and right versions of (QS3), for example: (QS3L)d(a, b)⊗d(a1, b)≤d(a∨a1, b), (QS3R)d(a1, b)⊗d(a, b)≤d(a∨a1, b), and to define left and right Q-ideals accordingly. One could then consider bi-ideals satisfying both conditions and study the resulting full subcategories. However, it is not clear whether a satisfactory reflection theorem still holds in this setting, or whether the presheaf-based construction U(S) admits a simple description. Problem 8.1 (Non-commutative bases).Develop a theory of Q-specialization semilattices for non-commutative quantales. In particular: •Formulate appropriate left/right analogues of (QS3) and the Q-ideal conditions which admit enough examples; •Determine whether a presheaf-based universal extension U(S)still exists and whether it can be characterised as the fixpoints of a suitable composite nucleus on an enriched presheaf category; •Identify non-commutative examples (e.g. convolution quantales on non-abelian groups, endomorphism quantales) where such a theory yields genuinely new completions. We expect that asymmetric specialization behaviours (for instance, left and right notions of approximation) may be naturally modelled by non-commutative quantales, but a systematic analysis remains to be undertaken. 8.3 Logical semantics and canonical models Quantales and quantale-enriched categories play an important role in the semantics of nonclassical logics: many-valued logics, fuzzy logics, and substructural logics often admit semantics where elements of Qrepresent degrees of truth or resources, and ⊗models a form of conjunction or composition. In this setting a Q-specialization semilattice (S, ∨, d, 0) can be interpreted as a space of “states” or “theories” with a Q-valued consequence relation: d(a, b) measures the degree to which bfollows from a, while finite joins represent the combination of information. From this viewpoint, the universal extension S7−→ U(S) may be regarded as constructing a canonical model or completion of a given semantic structure: its elements are Q-valued ideals (generalised theories) satisfying stability conditions that reflect both the enriched and the semilattice structure. The comparison with the Isbell completion situates this canonical model inside a well-understood enriched completion. Problem 8.2 (Quantale-valued logics and canonical frames).Investigate the role of U(S)as a canonical model in concrete logical systems whose semantics is based on a quantale Q. For instance: •For a given fuzzy or substructural logic with truth-value quantale Q, interpret Sas a semilattice of theories or filters and describe U(S)as a canonical Kripke-style structure; 24 •Compare U(S)with other canonical constructions (e.g. canonical frames in algebraic logic, coalgebraic canonical models) and determine whether U(S)enjoys completeness or representation properties for the logic; •Explore whether the nucleus NShas a direct logical interpretation (e.g. as a closure operation on semantic valuations) and whether fixed points of NScorrespond to canonical or saturated models. Such investigations could connect the present work with the extensive literature on quantalevalued logics and provide new tools for constructing canonical models in many-valued and substructural settings. 8.4 Higher-categorical refinements and 2-monadic structure Our construction S7→ U(S) has been formulated at the 1-categorical level: we defined a reflection U⊣J:QSpecSL0⇄QAddPrinSL0 between ordinary categories. However, both the enriched presheaf construction and the Isbell completion are naturally 2-categorical and interact with enriched natural transformations and higher cells. It is therefore natural to ask whether the reflector Uadmits a 2-categorical or monadic refinement. One possible approach is to view QSpecSL0as a 2-category whose 1-cells are Q-specialization homomorphisms and whose 2-cells are suitable enriched natural transformations or lax transformations. The presheaf 2-monad P:QCat →QCat, S 7→ [Sop, Q], is well understood in this setting, and the Isbell completion can be seen as a composite of presheaf and copresheaf constructions together with a MacNeille nucleus. Problem 8.3 (2-monadic structure of U).Study whether the reflector Uextends to a 2-monad or Kock–Z¨oberlein doctrine on an appropriate 2-category of Q-specialization semilattices. In particular: •Describe Uas induced by a composite of the presheaf 2-monad and a suitable 2-nucleus (in the sense of enriched MacNeille completions); •Determine whether algebras for this 2-monad can be described as “cocomplete” or “injective” Q-specialization semilattices, and whether U(S)is the free such algebra on S; •Analyse how this structure interacts with change-of-base of quantales and with enriched adjunctions between different Q-specialization categories. Clarifying the higher-categorical status of Uwould place the present reflection alongside other well-known 2-monadic completions (such as presheaf completions and Cauchy completions) and could reveal further structural properties of principal additive Q-specialization semilattices. Taken together, these directions suggest that the presheaf-based universal extension U(S) is not only a convenient technical tool, but also a robust organising principle: it provides a natural semilattice-compatible core inside enriched completions, and it promises to connect algebraic, metric, topological, and logical constructions in a common quantale-enriched language. 25