An Explicit Framework for Interaction Nets

Interaction nets are a graphical formalism inspired by Linear Logic proof-nets often used for studying higher order rewriting e.g. \Beta-reduction. Traditional presentations of interaction nets are based on graph theory and rely on elementary properties of graph theory. We give here a more explicit presentation based on notions borrowed from Girard’s Geometry of Interaction: interaction nets are presented as partial permutations and a composition of nets, the gluing, is derived from the execution formula. We then define contexts and reduction as the context closure of rules. We prove strong confluence of the reduction within our framework and show how interaction nets can be viewed as the quotient of some generalized proof-nets.

💡 Research Summary

The paper presents a novel, mathematically explicit framework for interaction nets, moving away from the traditional graph‑theoretic descriptions that have dominated the field. The authors begin by re‑encoding each interaction net as a partial permutation: ports become elements of a permutation, and agents correspond to specific groupings of these elements. This representation captures the connectivity of a net in a purely algebraic form, allowing for clean manipulation of structure without resorting to adjacency matrices or edge lists.

Building on Girard’s Geometry of Interaction, the authors introduce a composition operation they call “gluing”. Gluing takes two partial permutations (i.e., two nets) and applies the execution formula: overlapping ports are eliminated, and the remaining ports are re‑wired according to the composition of the underlying permutations. This operation is associative and respects the logical meaning of the nets, effectively providing a categorical‑like tensor product that preserves computational semantics.

The next major contribution is the formal definition of contexts and reduction rules within this algebraic setting. A context is a net with a distinguished “hole” where a rule can be inserted. Rules themselves are tiny pattern nets together with their replacement nets. Reduction is defined as the context closure of these rules: any net that can be expressed as a context glued to a left‑hand side of a rule may be rewritten by gluing the same context to the right‑hand side. This definition abstracts away from low‑level graph rewriting details and yields a uniform, compositional notion of reduction.

A central theorem of the paper is strong confluence (strong Church‑Rosser property) for the reduction relation defined above. The proof leverages the algebraic properties of partial permutations: because gluing is associative and the execution formula is confluent at the level of permutations, any two distinct reduction steps can be rearranged to reach a common successor net. The authors present the proof in a concise, diagram‑free style that highlights the underlying algebra rather than combinatorial case analysis.

Finally, the authors relate interaction nets to generalized proof‑nets. They show that interaction nets can be viewed as the quotient of a class of proof‑nets under an equivalence relation induced by the permutation representation. In other words, many syntactically different proof‑nets correspond to the same interaction net when viewed through the lens of partial permutations. This establishes a precise bridge between the two formalisms, clarifying how the logical content of proof‑nets is preserved in the more operational interaction‑net setting.

Beyond the theoretical contributions, the paper discusses practical implications. Representing nets as permutations is memory‑efficient and naturally suited to parallel execution, since permutation composition can be performed concurrently on disjoint subsets of ports. Moreover, the algebraic framework simplifies the implementation of reduction engines: rule application reduces to permutation composition and hole filling, operations that are straightforward to code and verify. The authors suggest that their approach could improve compiler back‑ends for functional languages, enable more robust verification of graph‑rewriting systems, and provide a solid foundation for extending interaction nets to richer logical systems.

In summary, the work delivers a rigorous, permutation‑based semantics for interaction nets, defines gluing and reduction via the execution formula, proves strong confluence in this setting, and situates interaction nets as a quotient of generalized proof‑nets. This unifies several strands of research—linear logic, Geometry of Interaction, and graph rewriting—into a coherent algebraic theory with clear advantages for both analysis and implementation.