Monoidal computer II: Normal complexity by string diagrams

Abstract

In Monoidal Computer I, we introduced a categorical model of computation where the formal reasoning about computability was supported by the simple and popular diagrammatic language of string diagrams. In the present paper, we refine and extend that model of computation to support a formal complexity theory as well. This formalization brings to the foreground the concept of normal complexity measures, which allow decompositions akin to Kleene's normal form. Such measures turn out to be just those where evaluating the complexity of a program does not require substantially more resources than evaluating the program itself. The usual time and space complexity are thus normal measures, whereas the average and the randomized complexity measures are not. While the measures that are not normal provide important design time information about algorithms, and for theoretical analyses, normal measures can also be used at run time, as practical tools of computation, e.g. to set the bounds for hypothesis testing, inductive inference and algorithmic learning.