View article

Stochastic thermodynamics is a theoretical framework that extends the laws of classical thermodynamics to small system at the molecular and cellular scale. In particular processing information at theses scales is continuously corrupted by thermal fluctuations. Examples involve translating information from DNA to proteins, bacteria that sense their environment or neurons that fire action potentials. In all of these examples, energy is consumed to process information or to shield the process against thermal fluctuations. This thesis investigates the relation between information and thermodynamics in physical systems.

We develop a framework for two continuously coupled systems, which is called stochastic thermodynamics of bipartite systems. This framework includes information and refines the standard second law of thermodynamics. In the first part we consider feedback-driven engines, where one subsystem is controlled by a second subsystem that constitutes the feedback controller. The feedback controller continuously acquires information about the controlled subsystem and uses it to rectify thermal fluctuations, ie, to “convert information into energy”. We compare two information theoretic quantities that characterize the performance of the feedback controller the transfer entropy rate and the learning rate. We find that only the latter both (i) bounds the rate of energy extraction from the medium due to the controlled subsystem and (ii) is itself bounded by the thermodynamic cost to maintain the dynamics of the feedback controller. This insight is one of the main results and provides a modern view on classical thought experiments first proposed by …