Skip to content

Quantaloid

A category with complete ordered hom-sets in which typed composition preserves arbitrary joins separately in each argument.

Version
v1 · 2026-10-31 · History
Domain-specific #
14559
Domain group
Formal Sciences
Origin domain
Mathematics
Subdomain
Enriched Category Theory → Mathematics

Core Idea

A quantaloid is a category with a complete order on each set of arrows between a fixed source and target. Arrows may be joined when they are parallel, and composition preserves arbitrary such joins separately in either argument. This includes the empty join. Stubbe gives the concise definition as a category enriched in Sup, the category of complete join-lattices and join-preserving maps.[1] Its objects can form a proper class under a stated foundation, while every individual hom remains a set. Neither the number of objects nor an application to enriched categories is part of the defining test.[1]

The definition has two unlike realizations. In Rel, arrows are binary relations and joins are unions. In a one-object cost quantaloid, arrows are extended nonnegative numbers ordered in reverse, so joins are numerical infima and composition is addition.[2][1] Each is a whole quantaloid, rather than a metaphor for combining information.

How would you explain it like I'm…

 

No faithful explanation at this level. All three generators agree that any five-year-old picture (roads between towns, 'putting things together') collapses a quantaloid into a vague combining-information metaphor or a single concrete map, losing that it is a category whose hom-sets are complete join-lattices with composition preserving arbitrary (including empty) joins.

Merge-and-Chain Arrows

Imagine dots with arrows between them, where you can follow one arrow and then another to make a longer trip. In a Quantaloid, the arrows between the same two dots can always be lumped together into one combined arrow, and there is even an 'empty lump' with nothing in it. The key rule is that lumping and chaining get along: chaining with a lump is the same as chaining with each arrow and then lumping the results. One example is 'who is connected to whom' lists, where lumping means putting lists together. A very different example uses distances, where lumping means picking the smallest and chaining means adding.

Join-Ordered Categories

A Quantaloid is a category, a system of objects and arrows that can be composed, with extra order structure on each collection of arrows that share a start and an end. Every family of such parallel arrows, even an infinite or empty family, has a join, a least upper bound in that order. Composition must preserve these joins in each argument separately: composing f with the join of some g's equals the join of the composites. In the category of relations, the arrows are relations and the join is union. In a one-object cost example, arrows are nonnegative numbers (or infinity) ordered backwards, so the join is the infimum and composition is addition. These are two complete instances of the same definition, not metaphors.

 

A Quantaloid is a category in which each hom-set (the arrows from a fixed source to a fixed target) is a complete lattice, so arbitrary joins of parallel arrows exist, including the empty join (a bottom arrow). Composition must preserve these arbitrary joins in each argument separately: f composed with the join of a family equals the join of f composed with each member, and likewise on the other side. Equivalently, it is a category enriched in Sup, the category of complete join-lattices and join-preserving maps. The collection of objects may be a proper class, but each individual hom must be a set; how many objects there are is not part of the definition. In Rel, objects are sets, arrows are binary relations, joins are unions and composition is relational composition. In a one-object cost quantaloid, arrows are extended nonnegative reals ordered in reverse, so joins are infima and composition is addition. Both are genuine quantaloids, not loose metaphors for combining information.

Structural Signature

  • Objects and typed arrows. For objects A,B, the arrows A → B form a hom-set Q(A,B); only arrows with matching endpoints compose. Without these types, the composition operation is not the category operation used here.[1]
  • Complete local order. Each Q(A,B) is an antisymmetric complete join-lattice. Every set-indexed family of parallel arrows has a join, including the empty family, whose join is the local bottom arrow ⊥₍A,B₎. A finite-join order alone is insufficient.[1]
  • Associative composition and identities. If f:A→B and g:B→C, then g∘f:A→C; composition is associative and each object has a two-sided identity 1ₐ:A→A. Join-rich typed data without these category laws are insufficient.[1]
  • Separate preservation of arbitrary joins. Holding one compatible arrow fixed, composition preserves every join in the other hom-set: g∘(⋁ᵢ fᵢ)=⋁ᵢ(g∘fᵢ) and (⋁ⱼ gⱼ)∘f=⋁ⱼ(gⱼ∘f). The two laws apply to families of parallel arrows with the required endpoints. Their empty-family instances imply g∘⊥=⊥ and ⊥∘f=⊥ in the appropriately typed homs. They do not assert that composition preserves joins of arbitrary pairs in a cartesian product.[1]

These are the constitutive roles. Typed residuals arise from them: for example, Stubbe's Lemma 2.2(1) expresses the greatest x satisfying g∘x≤h as the join of all such x.[1] This is a derived operation, not a fifth choice in the definition. At each object A, Q(A,A) is a unital quantale; with one object it is the whole quantaloid, whereas a many-object quantaloid has further typed homs.[1]

What It Is Not

A generic category has objects, arrows, identities and associative composition, but need not have complete ordered hom-sets or join-preserving composition. A bare complete lattice has joins but need not supply typed arrows or categorical composition. The category of sets and functions is not a substitute for Rel: an arbitrary union of functions need not be a function, so the relation example's joins do not stay inside function homs.[2]

The base quantaloid Q is also distinct from a category enriched in Q. Such a Q-category uses arrows of Q as typed hom-values subject to its own axioms; it is a further construction, not an additional condition that Q must have been put to use.[1] Stubbe proves that the category Dist(Q) of Q-categories and distributors is itself a quantaloid, but it is a constructed quantaloid distinct from its base Q.[1]

Scope of Application

The scope is formal mathematics where objects, parallel-arrow orders and composition can all be specified. Two source-attested whole structures illustrate the range:

  • Relations among sets: Rel takes sets as objects and all relations R⊆A×B as arrows A→B. Its hom-lattice is the powerset of A×B, ordered by inclusion; joins are unions, composition is existential relational composition, and identities are equality diagonals. It has all four defining roles.[2]
  • One-object cost algebra: Take one object ∗ and hom [0,∞] with the order opposite to ordinary numerical order. Joins are numerical infima, including ⋁∅=∞; composition is extended addition, with a+∞=∞+a=∞, and identity is 0. This unital quantale is a one-object quantaloid.[1][2]
  • Cost matrices as a supported variant: Schubert's Mat V construction applied to the preceding cost quantale has sets as objects and cost-valued matrices as arrows. Composition is the infimum over intermediate indices of sums; the identity has 0 on its diagonal and ∞ elsewhere. This is a mathematical specialization of his general matrix construction, not a third measured cost study.[2]

The definition does not require a metric, a measured travel cost, a software implementation or an empirical performance result. It admits proper-class object collections under the declared size convention so long as each hom is a set.[1]

Clarity

“Complete” is local: Q(A,B) has joins of arrows with the same endpoints. There is no operation joining an arrow A→B directly to an arrow C→D without further structure. In Rel, more ordered means a larger subset, and the empty join is the empty relation. Its composite satisfies (S∘R)(a,c) exactly when some intermediate b has both R(a,b) and S(b,c). That existential clause distributes over unions on either side, including an empty union. The diagonal relation is neutral.[2]

For costs, the order direction reverses: a numerically higher cost is lower in the lattice order. Thus the lattice join is a numerical infimum, and the empty join is ∞, not 0. Addition by a fixed extended nonnegative value preserves those infima; the empty-family and fixed-∞ cases give ∞ on both sides. The identity is the numeric 0, while ∞ is absorbing. This order convention is why a familiar minimum-of-sums formula is an instance of join-preserving composition.[1][2]

Manages Complexity

The four laws offer a short check for otherwise quite different constructions. First fix source and target, then inspect the complete local order; next check category composition and identities; finally check join preservation in each composition variable, including the empty family. The empty check detects structures whose composition is monotone yet fails to absorb local bottom arrows.

The same laws give additional calculations without adding axioms. Because each fixed-arrow composition map preserves joins, it has a typed right adjoint; Stubbe states the resulting residual formulas explicitly.[1] This lets a composition inequality be solved inside a hom-lattice while keeping the source and target types visible. It does not make an arbitrary individual arrow invertible or give that arrow its own right adjoint.

Abstract Reasoning

The definition supports reasoning at two levels. One may work within a single hom-lattice, using its order and arbitrary joins, then transport those joins through a compatible composition. Or one may specialize to Q(A,A), where composition and 1ₐ make a quantale. Passing to one endohom does not erase the distinct homs and object types of a many-object quantaloid.[1]

Candidate tests have edge cases. Empty joins are part of “arbitrary joins”; an identity need not be the greatest endomorphism. In Rel on a set with at least two elements, the equality diagonal is below the total relation, so a top identity is not universal. In Rel on the empty set, the sole endomorphism is both its identity and its bottom arrow, so nontriviality is not universal either. Commutativity of composition also fails in general for relations. These follow directly from the relation construction and the stated laws.[2]

Knowledge Transfer

The reusable test travels from Boolean relations to numerical costs without treating their carriers as interchangeable. For Rel, union is the hom-join and existential composition preserves it. For the one-object cost quantale, numerical infimum is the hom-join and addition preserves it. Schubert's Boolean matrices recover Rel as Mat 2, while cost matrices use his same typed matrix scheme with the cost quantale substituted.[2] The matrix equation changes from an existential Boolean combination to an infimum of sums; no symmetry or triangle inequality follows merely from being a cost matrix.

Quantaloids can serve as bases for enrichment. Stubbe's Q-category definition is a use of the base laws, and his distributor construction produces a further quantaloid Dist(Q).[1] Recognizing these levels prevents a result about an enriched category or a cost-valued matrix from being mistaken for a new axiom of the base quantaloid.

Examples

In Rel, take A={a₁,a₂}, B={b} and C={c}. The relations R₁={(a₁,b)} and R₂={(a₂,b)} join to R₁∪R₂. If S={(b,c)}, then S∘(R₁∪R₂)=(S∘R₁)∪(S∘R₂). The same argument works for any family, because existential choice of an intermediate element distributes over a union; for an empty family both sides are the empty relation. Equality diagonals are identities. This small calculation instantiates Schubert's general Rel structure rather than reporting a separate empirical case.[2]

In the one-object cost quantaloid, consider the arrows 2, 5 and ∞. Their join in the reversed order is min{2,5,∞}=2. Composing with arrow 3 gives 3+2=5, matching inf{3+2,3+5,3+∞}=5. The identity arrow is 0; the empty join is ∞, and 3+∞=∞. These values merely illustrate the original extended-nonnegative quantale, whose preservation law holds for all families and every fixed arrow, including fixed ∞.[1][2]

The first case has many object types and inclusion-ordered sets of relations; the second has one object and reversed-order numerical costs. Both satisfy the same typed complete-hom and composition laws. No actual metric space is needed for the second case.

Structural Tensions

The inspected definitions establish no opposing requirements that must be traded against each other. Object size requires a declared foundation, and the order convention changes what a join looks like, but neither is a design tension internal to the quantaloid axioms.[1][2] In particular, the cost example's reversed order is a coherent choice that makes infimum the join; it is not a failure of completeness.

There is a distinction between primitive laws and their consequences: residuals follow from join-preserving composition, and enriched categories are a possible use of the resulting base.[1] Calling either an extra required role would obscure the definition rather than expose a genuine source-established conflict.

Structural–Framed Character

Quantaloid is a formal mathematical structure. Its constitutive test is literal: typed category laws, complete antisymmetric local orders and separate preservation of arbitrary joins.[1]

  • Evaluative weight: No good/bad ranking is built into membership; satisfying the axioms determines it.
  • Human practice: Mathematicians choose notation, examples and uses, but those choices do not add a law to the object.
  • Institutional origin: A publication may establish a definition, yet no institution confers quantaloid status on a structure that fails it.
  • Vocabulary travel: Relations, costs and matrices meet one formal test through different carriers and operations; a metaphorical use of “quantaloid” would still need all four roles.
  • Import versus recognition: The task is to recognize complete typed homs and join-stable composition, not to import an observed application or a favored quantale convention into every instance.

Its character: structural. The category and arbitrary-join laws determine membership across relational and cost carriers; empirical interpretation, if any, belongs to a separate application.

Structural Core vs. Domain Accent

The portable core is a category with typed arrows, complete local hom orders, identities and associative composition that preserves arbitrary joins separately in both variables. Relational inclusion, union and existential composition are the relation-specific realization; reversed numerical order, infimum and addition are the cost-specific realization.[1][2] Neither accent is a compulsory material of every quantaloid.

A wider Prime would require genuinely nonmathematical, role-complete cases in which typed composition and both arbitrary-join laws survive the change of carrier. These sources establish the mathematical structure and its mathematical examples; they do not establish that broader transfer. Dropping either local completeness or empty-join preservation to make a loose analogy would remove part of the present identity.

Quantaloid shares the objects, arrows and composition of a mathematical category, while adding complete ordered homs and the separate join laws. A complete lattice appears at each hom Q(A,B); a single lattice is not the whole category when several object types and homs are present. Sets and membership underlie the familiar Rel example, whose arrows are relations between sets.[1][2]

These nearby concepts clarify the ingredients without asserting that one ordinary set of objects, one complete lattice or every category is a quantaloid. In particular, large quantaloids may have a proper class of objects under the stated foundation even though their homs are sets.[1]

Neighborhood in Abstraction Space

Quantaloid sits in a sparse region of the domain-specific corpus (85th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2835 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-31

Not to Be Confused With

  • An arbitrary category: it may lack complete hom-joins or their preservation by composition.
  • A complete lattice alone: it supplies local order but not necessarily typed categorical composition.
  • Rel alone: it is an example, not the entire class of quantaloids.[2]
  • The category Set of functions: its arrows are functions, and unions of them need not remain functions.
  • A Q-enriched category: it uses a base quantaloid Q; it is not identical to that base. Its distributor category Dist(Q) is itself a quantaloid constructed at another level.[1]
  • A metric or a measured cost: the extended nonnegative cost quantale is an algebraic source example; metric conditions belong to a further enrichment or application.[1][2]

No universal commutativity, integral identity, nontrivial hom or small object set is implied by the quantaloid definition.[1][2]

References

[1] Isar Stubbe. (2004). “Categorical structures enriched in a quantaloid: categories, distributors and functors,” arXiv:math/0409473v1, posted 24 September 2004, manuscript dated 10 May 2004, original author preprint PDF. Relevant preprint Definition 2.1 p. 4, Lemma 2.2(1) p. 5, Example 2.4 p. 5, Definition 3.1 pp. 6–7, and Proposition 3.4 p. 8. Page numbers refer to the inspected 2004 preprint, not the later journal layout. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27

[2] Christoph Schubert. (2006). Lax Algebras — A Scenic Approach, doctoral dissertation, Universität Bremen, Lax Algebras — A Scenic Approach. Relevant pp. 13, 31, 38–39: relations, Definition 1.4.1, Example 1.5.8(3), and matrix Equations (1.10)–(1.11). Its additional commutative and nontrivial quantale conventions concern that author's chosen examples and do not define all quantaloids. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q