Categories in control: applied PROPs

Abstract

Control theory uses `signal-flow diagrams' to describe processes where real-valued functions of time are added, multiplied by scalars, differentiated and integrated, duplicated and deleted. These diagrams can be seen as string diagrams for the PROP FinRel_k, the strict version of the category of finite-dimensional vector spaces over the field of rational functions k = R(s) and linear relations, where the variable s acts as differentiation and the monoidal structure is direct sum rather than the usual tensor product of vector spaces. Control processes are also described by controllability and observability---whether the input can drive the process to any state, and whether any state can be determined from later outputs. For any field k we give a presentation of FinRel_k in terms of generators of the free PROP of signal-flow diagrams together with the equations that give FinRel_k its structure. The `cap' and `cup' generators, missing when the morphisms are linear maps, make it possible to model feedback. The relations say, among other things, that the 1-dimensional vector space k has two special commutative dagger-Frobenius structures, such that the multiplication and unit of either one and the comultiplication and counit of the other fit together to form a bimonoid. This sort of structure, but with tensor product replacing direct sum, is familiar from the `ZX-calculus' obeyed by a finite-dimensional Hilbert space with two mutually unbiased bases. In order to address controllability and observability, we construct the PROP Stateful_k and relate it back to the PROP of signal-flow diagrams. This provides a way to graphically express controllability and observability for linear time-invariant processes.