Artificial and synthetic nanoscale machines are engineered systems designed to mimic biological functionality by performing controlled tasks at the nanometer scale, often operating far from thermal equilibrium. These systems process information obtained from the environment and use it to perform tasks. We aim to study their thermodynamic properties using stochastic and information thermodynamics. This theoretical framework is particularly motivated by its ability to model simple decision-making and environmental sensing, as observed in biological systems. Such models therefore serve as ideal platforms for exploring how information processing, energy dissipation, and function are intertwined in active systems. The key challenges in these systems include their active nature, partial observability, and unknown dynamics.

The term coarse-graining is commonly used to accelerate calculations by reducing computational cost in molecular dynamics simulations. However, in our case, coarse-graining arises from the finite spatiotemporal resolution of the experimental setup, as well as from unobserved states or transitions between states. As a result, one may or may not recover the full dissipation information from such coarse-grained descriptions. Therefore, we seek a thermodynamically consistent coarse-graining method that can accurately reproduce the full dissipation even when only coarse-grained observations are available, making it directly relevant for practical applications.

Systems away from thermal equilibrium are inherently irreversible. Various measures and methods for identifying irreversibility have been proposed in the literature. A recent review articleon this topic from our group can be found here. We are interested in developing methods to identify and quantify irreversibility in partially observed systems.