Petri nets are monoids

AbstractPetri nets are widely used to model concurrent systems. However, their composition and abstraction mechanisms are inadequate: we solve this problem in a satisfactory way. We start by remarking that place/transition Petri nets can be viewed as ordinary, directed graphs equipped with two algebraic operations corresponding to parallell and sequential composition of transitions. A distributive law between the two operations captures a basic fact about concurrency. New morphisms are defined, mapping single, atomic transitions into whole computations, thus relating system descriptions at different levels of abstraction. Categories equipped with products and coproducts (corresponding to parallel and nondeterministic compositions) are introduced for Petri nets with and without initial markings. Petri net duality is expressed as a duality functor, and several new invariants are introduced. A tensor product is defined on nets, and their category is proved to be symmetric monoidal closed. This construction is generalized to a large class of algebraic theories on graphs. These results provide a formal basis for expressing the semantics of concurrent languages in terms of Petri nets. They also provide a new understanding of concurrency in terms of algebraic structures over graphs and categories that should apply to other models besides Petri nets and thus contribute to the conceptual unification of concurrency