Diagrammatic Inference

Diagrammatic logics were introduced in 2002, with emphasis on the notions of specifications and models. In this paper we improve the description of the inference process, which is seen as a Yoneda functor on a bicategory of fractions. A diagrammatic logic is defined from a morphism of limit sketches (called a propagator) which gives rise to an adjunction, which in turn determines a bicategory of fractions. The propagator, the adjunction and the bicategory provide respectively the syntax, the models and the inference process for the logic. Then diagrammatic logics and their morphisms are applied to the semantics of side effects in computer languages.

💡 Research Summary

The paper revisits the foundations of diagrammatic logics, originally introduced in 2002, and provides a rigorous categorical account of the inference process. The authors observe that earlier formulations emphasized the relationship between specifications and models but left the mechanics of inference under‑specified. To remedy this, they introduce the notion of a propagator—a morphism between limit sketches. A limit sketch encodes both algebraic operations and equations in a graphical form, making it suitable for representing complex syntactic structures. A propagator preserves the sketch’s objects and arrows while transporting additional structure from a source sketch to a target sketch.

From any propagator one automatically obtains an adjunction between the associated categories of models: a free functor (building syntactic objects from the source sketch) left‑adjoint to a forgetful functor (interpreting those syntactic objects in the target sketch). This adjunction captures the dual nature of syntax (as freely generated) and semantics (as a constrained interpretation).

The adjunction, in turn, induces a bicategory of fractions—a categorical localization that formally inverts a chosen class of morphisms. In this setting the fractions serve as 2‑cells representing inference steps. By applying the Yoneda functor to this bicategory, each inference step is identified with a representable functor, turning the traditional proof tree into a composition of representable morphisms. This Yoneda‑based view makes inference object‑centric: every deduction is a universal property of the object being proved about, rather than a mere syntactic manipulation.

The authors argue that this three‑layered architecture—propagator (syntax), adjunction (models), bicategory of fractions with Yoneda (inference)—offers several advantages. First, it unifies the three core aspects of a logic within a single categorical framework, facilitating systematic morphisms between logics and modular extensions. Second, because the syntax is encoded as a limit sketch, one can directly model sophisticated type constructors, effectful operations, or domain‑specific constructs without leaving the categorical setting. Third, the Yoneda‑based inference is inherently constructive and compositional, which is beneficial for mechanised reasoning and proof assistants.

To demonstrate practical relevance, the paper applies the framework to the semantics of side‑effects in programming languages. A special sketch is introduced to represent mutable state, and a propagator maps this stateful sketch into a more conventional computational sketch. The resulting adjunction interprets state‑changing commands as morphisms in the target model, while the bicategory of fractions captures the permissible state transitions as invertible 2‑cells. The Yoneda inference then yields a structured proof system for reasoning about programs with side‑effects, offering a more disciplined alternative to ad‑hoc operational semantics.

In conclusion, the work provides a mathematically elegant and operationally useful reformulation of diagrammatic logics. By grounding syntax, semantics, and inference in the same categorical machinery, it opens pathways for robust logic morphisms, automated proof generation, and precise modeling of effectful computation. The authors suggest that future research could explore further applications in type theory, formal verification, and the design of domain‑specific languages, where the interplay of complex syntactic features and semantic effects demands a unified, diagrammatic treatment.