Abstract

Whether externally manipulated or autonomously controlled, systems driven out of thermodynamic equilibrium often display features that are most naturally described in the language of information theory. Such a connection is immediately established at the level of the entropy production, which is linked via the fluctuation theorem of stochastic thermodynamics to the informatic distinguishability of the ensembles of forward paths and their time reverse, i.e., to the (resolution-dependent) arrow of time. In this talk I will discuss two recent works at the intersection of nonequilibrium information and control theory: in the first, I will demonstrate how mutual information provides a natural cost function to describe mixing at low Reynolds number and show that a variational approach may be employed to derive analytical thermodynamic bounds reminiscent of the Landauer bound for irreversible computation; in the second, I will discuss the energetics of a generalization of the Active Brownian Particle, a canonical model in motile active matter, showing how the multi-objective optimization problem of balancing the cost of information-processing and actuation leads to the emergence of nontrivial Pareto fronts, i.e. of precision-dissipation tradeoffs.