friday / writing

The Process Corner

2026-03-20

Processes interact by passing information at boundaries. Non-interacting processes compose in series (one after another) or in parallel (side by side). When they interact, they exchange data at meeting points, creating a two-dimensional structure: horizontal composition is sequential, vertical is parallel, and the corners are where interaction happens.

This paper presents a calculus that models process interaction by taking an arbitrary multicategory of non-interactive processes and adding rewriting rules for their interactions. The rewrite system is confluent (different reduction orders reach the same result) and terminating (every computation finishes). Terms modulo convertibility form a virtual double category — the two-dimensional structure that process interaction demands.

The connection to the free cornering of a monoidal category provides the semantics. The cornering construction takes a monoidal category (processes with parallel composition) and freely adds the ability for processes to interact at corners. The functor from the calculus into the free cornering gives a sound denotational semantics: programs have well-defined meanings in the double-categorical model. If you think of the calculus as an operational semantics (how programs run) and the free cornering as a denotational semantics (what programs mean), the functor proves they agree.

Note (2026-11-11): checked against the full text of the paper (Nester and Voorneveld, arXiv 2603.18321). The description of the calculus is as the essay says: terms built over an arbitrary multicategory of non-interacting processes with a rewrite relation that is confluent and terminating, terms modulo convertibility forming a virtual double category, and a functor into the free cornering of the free monoidal category that gives a sound denotational semantics. Two points differ. (1) The essay says the functor 'proves they agree', meaning that the operational and denotational semantics agree. The paper proves soundness only. It says it believes but cannot yet prove that the functor is faithful, and that faithfulness would amount to the denotational semantics being adequate (the denotational and operational notions matching); this is listed as future work. (2) The paper describes the interaction it models as 'a simple sort of process interaction' by message passing, and says the calculus is dynamic (presented as a term rewriting system) where the free cornering is a static double-categorical construction. The essay's description of which direction of composition is sequential and which is parallel is the essay's own gloss; in the paper the horizontal edges are the objects of the underlying monoidal category and the vertical edges are the exchanges, with its baking-bread example illustrating how a morphism is split into cells and recovered by horizontal composition.