Distorted Universality: Universal Properties Modulo Controlled Distortion

Distorted Universality: Universal Properties Modulo Controlled Distortion Joaquim Reizi Higuchi December 12, 2025 Abstract Universal properties can be read as the transport of equality-based uniqueness ∀x∃ ! y P ( x, y ) to a categorical regime where object-level uniqueness is relaxed to “up to isomorphism.” In many semantic models motivated by computation, physics, and resource-sensitive reasoning, even isomorphism is too rigid: exchange can be lossy or irreversible, and unavoidable overhead breaks strict commutativity conditions. This paper develops distorted universality, where universal constructions are required to exist and be unique only modulo a monoidal congruence generated by a pair of distortion operators (Λ , σ ): a defect/overhead operator Λ : id ⇒id and a constrained exchange σ : ⊗ ⇒ ⊗ ◦ τ . We emphasize necessity and usefulness: distorted universality restores canonical constructions in over-refined categories (e.g. cost-annotated morphisms), supports non-invertible exchange while preserving a universal calculus via quotient categories, and clarifies the two-layer methodology in which canonicality is proved in a quotient while optimization and budgeting live in the original category. We also isolate optional coherence regimes for σ (braided/symmetric up to the congruence) to ensure that exchange behaves as intended in the quotient. 1 Introduction Universal properties encode canonical constructions (limits, colimits, adjoints, Kan extensions, free objects) via a disciplined uniqueness pattern. Logically, one may view universality as a categorical refinement of the equality-based schema ∀x∃!y P(x, y), where the uniqueness component is expressed with respect to equality. Category theory replaces object-level equality by isomorphism and retains equality of arrows inside hom-sets. However, many models refined to reflect implementation (cost, latency, noise, loss, scheduling friction) do not validate strict universal properties: •swapping resources can be lossy or irreversible (non-invertible exchange), •units can be coherent only after inserting normalization overhead, •commutativity constraints often hold only modulo “acceptable” distortion. These phenomena are familiar in resource-sensitive and process semantics, and resonate with the ethos of linear logic and resource theories, where structural rules (notably exchange) are carefully controlled. 1 The aim of this paper is to formalize a minimal remedy: keep the refined model, but reinterpret the equalities inside universal properties modulo a structured congruence. We introduce a pair (Λ , σ ) and use it to generate a monoidal congruence ≈ on morphisms. Distorted universality is then defined as ordinary universality in the quotient category D/≈. Two design imperatives guide the development: 1. Necessity: without quotienting, refined categories frequently lose existence/uniqueness of universal mediators. 2. Usefulness: quotienting restores canonicality while the original category still retains finegrained information for optimization and budgeting. 2 Uniqueness modulo an equivalence 2.1 Uniqueness modulo a relation Definition 2.1 (Uniqueness modulo ∼ ).Let A be a set and ∼ an equivalence relation on A . We write ∃!∼a∈A . Q(a)to mean: ∃a Q(a)and ∀a′(Q(a′)⇒a′∼a). Equivalently: there is at least one solution and all solutions are equivalent under ∼. Remark 2.2. Category theory often uses ∃ ! ∼ with ∼ being isomorphism of objects (for object-level uniqueness) or equality of arrows (for arrow-level uniqueness). In this paper, ∼ will be a distorted equivalence ≈generated by designated distortions. 2.2 Representability as a template for universal properties Many universal properties can be phrased as representability. For example, A×B represents X7→ Hom(X, A)×Hom(X, B)via a natural bijection Hom(X, A ×B)∼ =Hom(X, A)×Hom(X, B). Distorted universality will demand representability after quotienting hom-sets by ≈. 3 Distortion data (Λ, σ)and distorted equivalence 3.1 Distortion profile Let (D,⊗, I, α, λ, ρ) be a (not necessarily strict) monoidal category. Let τ : D × D → D × D be the swap functor τ(X, Y )=(Y, X). Definition 3.1 (Distortion operators).Adistortion profile on Dis a pair (Λ, σ)consisting of (i) a natural endomorphism Λ : idD⇒idD, i.e. morphisms ΛX:X→Xnatural in X, (ii) a natural transformation σ : ⊗ ⇒ ⊗ ◦ τ , i.e. morphisms σX,Y : X⊗Y→Y⊗X natural in X, Y . 2 Discussion 3.2 (Interpretation).Λcaptures intrinsic overhead/defect (delay, normalization, noise injection, budget consumption). σ captures constrained exchange (swap with friction), not assumed invertible. The point of the theory is not to pretend distortion is absent, but to control how it is ignored when establishing canonicality. 3.2 Monoidal congruences Definition 3.3 (Monoidal congruence).Amonoidal congruence ≈ on D is a family of equivalence relations ≈X,Y on each hom-set HomD(X, Y )such that: (i) (composition stability) if f≈g : X→Y , then for all h : Y→Z and k : W→X , h◦f≈h◦g and f◦k≈g◦k, (ii) (tensor stability) if f≈g : X→X′ and f′≈g′ : Y→Y′ , then f⊗f′≈g⊗g′ : X⊗Y→X′⊗Y′ . 3.3 The distorted equivalence generated by (Λ, σ) Definition 3.4 (Compatibility and generated congruence).Fix a distortion profile (Λ , σ ). A monoidal congruence ≈ is compatible with (Λ , σ )if it contains the following generating identifications: (i) (defect insertion) for all f:X→Y, f≈f◦ΛXand f≈ΛY◦f, (ii) (exchange naturality up to ≈)for all f:X→X′,g:Y→Y′, σX′,Y ′◦(f⊗g)≈(g⊗f)◦σX,Y . We write ≈(Λ,σ)for the smallest monoidal congruence compatible with (Λ, σ). Remark 3.5 (On “collapse” and what is (not) lost).From defect insertion with f = idX we obtain idX≈ Λ X , hence [Λ X ]=[ idX ]in the quotient. This is intentional: the quotient is the idealized layer in which overhead is ignored for the purpose of canonicality. Information is not lost from D itself; it is merely not visible to universal properties interpreted in D/≈ . Optimization questions (e.g. minimum cost, least distortion) are handled in D by additional structure (orders, weights, or budgets); see Section 7. 3.4 Coherence regimes for σ A reviewer concern is whether σ behaves as an “exchange” in the quotient. The minimal definition above only makes σ a natural transformation, hence it may remain a “generic arrow” unless one enforces further relations. We therefore separate regimes. Definition 3.6 (Braided/symmetric exchange up to ≈ ).Let ( D,⊗, I, α, λ, ρ )be monoidal with congruence ≈and a natural σ. (i) σ is braided up to ≈ if the hexagon identities hold modulo ≈ , i.e. the usual braided hexagon diagrams commute in D/≈. (ii) σis symmetric up to ≈if it is braided up to ≈and additionally σY,X ◦σX,Y ≈idX⊗Yfor all X, Y. 3 (iii) σis unit-normalized up to ≈if σX,I ≈ρ−1 X◦λX, σI,X ≈λ−1 X◦ρX. Proposition 3.7 (Exchange becomes braiding/symmetry in the quotient).If σ is braided (resp. symmetric) up to ≈, then its image [σ]in D/≈is a braiding (resp. symmetry). In particular, if σ is symmetric up to ≈, then [σX,Y ]is an isomorphism in D/≈with inverse [σY,X ]. Discussion 3.8 (Why we allow the minimal regime).Some applications require partial or unreliable exchange: not every swap is meaningful, and coherence laws may fail. The minimal regime treats σ as “available but not ideal,” while the braided/symmetric-up-to- ≈ regimes recover the familiar behavior in the quotient when appropriate. This makes the framework adaptable rather than over-committing to braidings. 4 Quotient categories and reduction of universal reasoning Definition 4.1 (Quotient category).Given a monoidal congruence ≈on D, define D/≈by: •Ob(D/≈) = Ob(D), •HomD/≈(X, Y ) = HomD(X, Y )/≈, •composition induced from D, well-defined by composition stability. If ≈is monoidal, then D/≈is monoidal by tensor stability. Proposition 4.2 (Monoidal structure descends).If D is monoidal and ≈ is a monoidal congruence, then D/≈ inherits a monoidal structure, and the quotient functor Q : D → D/≈ is strong monoidal. Proposition 4.3 (Reduction principle).Let ≈ be a congruence on D and Q : D → D/≈ the quotient functor. A cone (resp. cocone) in D is a distorted limit (resp. colimit) cone w.r.t. ≈ if and only if its image under Qis an ordinary limit (resp. colimit) cone in D/≈. Proof. Immediate from the definition of limits/colimits in the quotient: equalities of morphisms are interpreted as ≈-equivalences in D. 5 Distorted universality 5.1 General definition Definition 5.1 (Distorted universal property).Fix (Λ , σ )and let ≈ = ≈(Λ,σ) . A diagram F : J → D is said to have a distorted limit (w.r.t. (Λ , σ )) if Q◦F : J → D/≈ has a (strict) limit in D/≈ . Dually for distorted colimits. Remark 5.2 (On the phrase “strict limit”).We say “strict” only in the sense of ordinary 1categorical limits in D/≈ , i.e. with literal equality of morphism classes. Pulling back along Q , this means “commutes up to ≈” in D, and mediators are unique up to ≈. 4 5.2 Distorted products as a concrete illustration Definition 5.3 (Distorted product).Fix ≈ on D . A distorted product of A, B is an object P with morphisms π1 : P→A , π2 : P→B such that for every X and f : X→A , g : X→B , there exists u:X→Pwith π1◦u≈fand π2◦u≈g, and such uis ≈-unique: if u′also satisfies these conditions, then u′≈u. Discussion 5.4 (A diagrammatic view).The distorted commutativity conditions can be displayed as 2-cells witnessing ≈-commutativity. For instance, π1◦u≈fcan be depicted as: X A f π1◦u ≈ and similarly for π2◦u≈g. 5.3 Uniqueness up to ≈-isomorphism Definition 5.5 ( ≈ -isomorphism).A morphism u : X→Y is an ≈ -isomorphism if there exists v:Y→Xsuch that v◦u≈idXand u◦v≈idY. Theorem 5.6 (Uniqueness of distorted universals).If distorted limits exist (w.r.t. ≈ ), then they are unique up to ≈ -isomorphism. In particular, distorted products (when they exist) are unique up to ≈-isomorphism. Proof sketch with explicit correspondence. In D/≈ , distorted limits are ordinary limits and hence unique up to unique isomorphism. Let L, L′ be two distorted limit objects in D . Their images Q ( L ) , Q ( L′ )are isomorphic in D/≈ . Choosing an isomorphism class representative yields morphisms u : L→L′ and v : L′→L in D whose classes satisfy [ v ] ◦ [ u ] = [ idL ]and [ u ] ◦ [ v ] = [ idL′ ], i.e. v◦u≈idLand u◦v≈idL′. Thus uis an ≈-isomorphism. 6 Why distorted universality is necessary This section isolates failure modes of classical universality in refined semantics and explains how (Λ, σ)repairs them. 6.1 Failure mode I: over-refinement destroys existence/uniqueness Example 6.1 (Cost-annotated functions: strict products fail).Fix the commutative monoid ( N, + , 0). Define a category SetN whose objects are sets and morphisms are pairs ( f, c )where f:X→Yis a function and c∈N. Composition is (g, d)◦(f, c):= (g◦f, c +d),idX= (idX,0). (For a noncommutative cost monoid ( M, ·, 1), one would use ( g, d ) ◦ ( f, c ) = ( g◦f, c·d )or ( g◦f, d·c ); the present example is commutative, so order issues do not arise.) Let A, B be sets and consider A×Bwith projections π1:= (pr1,0) : A×B→A, π2:= (pr2,0) : A×B→B. 5 Given ( f, c ) : X→A and ( g, d ) : X→B , any candidate mediator must be ( ⟨f, g⟩, e ) : X→A×B . Then π1◦(⟨f, g⟩, e)=(f, e), π2◦(⟨f, g⟩, e)=(g, e), so strict commutativity forces e = c = d . If c = d , no mediator exists, hence the strict product universal property fails. Discussion 6.2 (Diagrammatic location of the failure).The failure is visible as an incompatibility of costs on two triangles: A×B A B π1π2vs. X A B (f,c)(g,d) A single mediator (⟨f, g⟩, e)must synchronize the two costs, which is generally impossible. 6.2 Repair: quotienting out controlled distortion Example 6.3 (A distortion profile on SetN ).Define Λ X: = ( idX, 1) : X→X , representing unit overhead. Define σX,Y : = ( swapX,Y , s )where swapX,Y ( x, y )=( y, x )and s∈N is a fixed swap-cost. If s > 0, then σX,Y is not invertible in SetN(an inverse would require negative cost). Lemma 6.4 (In SetN , ≈(Λ,σ) forgets cost).Let (Λ , σ )be as above and let ≈ = ≈(Λ,σ) . Then ( f, c ) ≈ ( f, c′ )for all c, c′∈N , i.e. two arrows are ≈ -equivalent iff they have the same underlying function. Proof sketch. Defect insertion yields ( f, c ) ≈ ( f, c + 1) by composing with Λon source or target. By symmetry and transitivity of ≈ , all costs on the same underlying function become equivalent. Conversely, the generating rules do not change the underlying function, so ≈ cannot identify two different functions. Discussion 6.5 (Necessity: universality returns, but at the right abstraction level).By Lemma 6.4, the quotient SetN/≈ is equivalent to Set , so ordinary products exist in the quotient. Hence SetN admits distorted products even though strict products fail in Example 6.1. This illustrates the core necessity claim: in refined semantics, strict universality is often the wrong notion; distorted universality restores canonical structure by quotienting exactly the distortion one intends to ignore for canonical reasoning. 6.3 Failure mode II: exchange is present but not isomorphic In symmetric or braided monoidal categories, exchange is invertible. In resource/process semantics, exchange can be lossy or irreversible. The role of σ is to model exchange without forcing invertibility, while allowing one to recover ideal exchange laws in the quotient if one adopts the braided/symmetricup-to-≈regime of Section 3.4. 7 Why distorted universality is useful Distorted universality is not merely a repair mechanism; it is a methodology for building models that are both canonical and implementation-aware. 6 7.1 Two-layer semantics: canonicality vs optimization The congruence ≈ can (and often should) identify Λ X with idX in the quotient. This does not claim overhead is unreal; it asserts that universal properties are statements of ideal canonical structure. Practical questions—minimum cost, best schedule, least noise—require extra structure on the original category D, for example: •a preorder ⪯on morphisms encoding “no worse than” distortion (e.g. cost ordering), •a weight function w: Hom(X, Y )→R≥0∪ {∞}, •or a notion of admissible budgets (“acceptable distortions”). Then one can refine distorted universality to budgeted or optimal universality: existence and ≈ - uniqueness are proved in the quotient, while a chosen representative is selected in D by minimizing w or respecting a budget. This resolves the apparent tension that the quotient “forgets” cost: it forgets it only for canonicality, not for engineering. 7.2 Modularity: distortion profiles as design parameters Treating (Λ , σ )as explicit parameters yields modularity: different applications choose different generators for ≈ , hence different notions of “the same up to distortion.” This supports principled comparison (coarser vs finer congruence) and incremental refinement of semantics while preserving a universal calculus. 7.3 Diagrammatic clarity Because distorted commutativity is a central idea, diagrams are not cosmetic; they are semantic documentation. For example, the mediating arrow uin a distorted product satisfies: X A f π1◦u ≈X B g π2◦u ≈ These 2-cells are the operational meaning of “commutes up to controlled distortion.” 8 Computability, rewriting, and future work The definition of ≈(Λ,σ) as the smallest compatible congruence is mathematically clean but may be computationally opaque. A practical research direction is to present ≈ via a rewriting system and prove confluence/normalization. At minimum, concrete examples (such as SetN ) should identify ≈ explicitly (Lemma 6.4), so readers can see that the abstract construction reduces to a familiar quotient. Further directions: 1. A systematic comparison theory of distortion profiles via inclusion/equality of their induced congruences. 2. Distorted adjunctions and distorted Kan extensions: universal arrows modulo ≈. 3. Proof assistant formalization: quotients by monoidal congruences and automated transfer of universal constructions. 7 References [1] S. Mac Lane, Categories for the Working Mathematician, 2nd ed., Springer, 1998. [2] G. M. Kelly, Basic Concepts of Enriched Category Theory, Cambridge University Press, 1982. [3] A. Joyal and R. Street, Braided tensor categories, Advances in Mathematics 102 (1993), 20–78. [4] J.-Y. Girard, Linear Logic, Theoretical Computer Science 50 (1987), 1–101. [5] B. Coecke, T. Fritz, and R. W. Spekkens, A mathematical theory of resources, Information and Computation 250 (2016), 59–86. [6] T. Leinster, Higher Operads, Higher Categories, Cambridge University Press, 2004. [7] T. Jech, Set Theory, 3rd millennium ed., Springer, 2003. 8