Tightest bound on hidden entropy production from partially observed dynamics

TL;DR

This paper introduces a method to derive the tightest lower bound on entropy production in systems with hidden degrees of freedom by using observable data to infer an underlying hidden Markov model, improving estimates in non-Markovian dynamics.

Contribution

It presents a novel formalism that uses observable data to find the most accurate lower bound on entropy production for partially observed systems with hidden Markovian dynamics.

Findings

Provides a method to obtain the tightest lower bound on entropy production.

Applicable to systems with non-Markovian observable dynamics.

Illustrated with a simple example system.

Abstract

Stochastic thermodynamics allows us to define heat and work for microscopic systems far from thermodynamic equilibrium, based on observations of their stochastic dynamics. However, a complete account of the energetics necessitates that all relevant nonequilibrium degrees of freedom are resolved, which is not feasible in many experimental situations. A simple approach is to map the visible dynamics onto a Markov model, which produces a lower-bound estimate of the entropy production. The bound, however, can be quite loose, especially when the visible dynamics only have small or vanishing observable currents. An alternative approach is presented that uses all observable data to find an underlying hidden Markov model responsible for generating the observed non-Markovian dynamics. For masked Markovian kinetic networks, one obtains the tightest possible lower bound on entropy production of the full dynamics that is compatible with the observable data. The formalism is illustrated with a simple example system.