Representation Independent Decompositions of Computation (2504.04660v2)

Abstract: Constructing complex computation from simpler building blocks is a defining problem of computer science. In algebraic automata theory, we represent computing devices as semigroups. Accordingly, we use mathematical tools like products and homomorphisms to understand computation through hierarchical decompositions. To address the shortcomings of some of the existing decomposition methods, we generalize semigroup representations to semigroupoids by introducing types. On the abstraction level of category theory, we describe a flexible, iterative and representation independent algorithm. Moving from the specific state transition model to the abstract composition of arrows unifies seemingly different decomposition methods and clarifies the three algorithmic stages: collapse, copy and compress. We collapse some dynamics through a morphism to the top level; copy the forgotten details into the bottom level; and finally we apply compression there. The hierarchical connections are solely for locating the repeating patterns in the compression. These theoretical findings pave the way for more precise computer algebra tools and allow for understanding computation with other algebraic structures.