Machine learning a time-local fluctuation theorem for nonequilibrium steady states

• DOI: 10.48550/arxiv.2305.19457
• Publication Date: 2023-01-01
• Authors: Stephen K. Sanderson; Charlotte F. Pete...; Debra J. Searles
• Keywords: Fluctuation theorem; Fluctuation-dissipation theorem

Fluctuation theorems (FTs) quantify the thermodynamic reversibility of a system, and for deterministic systems they are defined in terms dissipation function. However, nonequilibrium steady state dynamics, phase space distribution is unknown, making function difficult to evaluate without extra information. As such, FTs date have required either that trajectory segment interest relatively long, or information available about entire surrounding segment. In this work, it shown simple machine learning model trained predict whether given being played forward backward time calculates satisfies an FT relies solely on within interest. The satisfied even very short segments where approximate relation derived from theory breaks down, far equilibrium, various dynamics. It further demonstrated any well-calibrated predictor time's arrow must satisfy fluctuation theorem, local can be which depends only its correlations with non-local dissipation.