Stochastic thermodynamics for classical non-Markov jump processes

Abstract

Stochastic thermodynamics investigates energetic and entropic bounds in small systems, but its foundational results rely on the Markov (memoryless) assumption. Although the Markov assumption is questionable in real experimental setups, extending stochastic thermodynamics to general non-Markov systems has proven challenging. Fundamentally, it has been elusive how to model memory-dependent non-Gaussian fluctuations consistently with thermodynamic laws. Here we establish stochastic thermodynamics for classical non-Markov jump processes with equilibrium bath degrees of freedom. Building on Markov embedding, we introduce a key technique, called the Fourier embedding, which converts non-Markov jump processes into Markovian field dynamics of auxiliary Fourier modes. This yields necessary and sufficient conditions for time-reversal symmetry and enables the derivation of the second law for a broad class of Fourier-embeddable dynamics with strong memory. Remarkably, our cumulative entropy production is determined by observations of the target system alone, independently of the embedding. We illustrate our framework with two non-Markov models: (i) a history-dependent two-state model and (ii) a history-dependent random walk. Interestingly, these models can be interpreted as jump-process counterparts of Zwanzig's model and thus admit a transparent microscopic interpretation. Our work offers a guiding principle for thermodynamically consistent, physics-informed modelling of history-dependent fluctuations.

6 pages, 2 figures + Methods and Supplementary Information