Functorial Statistical Physics: Feynman--Kac Formulae and Information Geometries

The main results of this paper comprise proofs of the following two related facts: (i) the Feynman–Kac formula is a functor $F_{\ast }$, namely, between a stochastic differential equation and a dynamical system on a statistical manifold, and (ii) a statistical manifold is a sheaf generated by this functor with a canonical gluing condition. Using a particular locality property for $F_{\ast }$, recognised from functorial quantum field theory as a `sewing law,’ we then extend our results to the Chapman–Kolmogorov equation {\it via} a time-dependent generalisation of the principle of maximum entropy. This yields a partial formalisation of a variational principle which takes us beyond Feynman–Kac measures driven by Wiener laws. Our construction offers a robust glimpse at a deeper theory which we argue re-imagines time-dependent statistical physics and information geometry alike.