Entropy production and information processing in stochastic thermodynamics: Optimization, measurement, and erasure

Abstract

In this thesis we explore the relationship between information processing and physics. We use various techniques, including deep learning, to optimize the information-to-energy conversion in microscopic machines. The second law of thermodynamics implies an arrow of time in physics, but how can the reversible microscopic dynamics lead to irreversible macroscopic phenomena? One particular paradox which this thesis focuses on is Maxwell's demon; a machine that uses information to extract work directly from a heat bath, in violation of the second law. The resolution of this paradox revealed a deep connection between information and the laws of physics. Information processing is also ubiquitous in molecular biology systems, for example during the transcription of DNA, and therefore ideas from physics and information theory can be used to understand the operation of these molecules. This thesis focuses on the relationship between information and thermodynamics, and how to optimize the information to energy conversion in these microscopic machines. We will discuss ideas from information theory, such as logical reversibility, measurement and erasure, what the equivalent physical processes of these somewhat abstract concepts are, and how they relate to the macroscopic laws of thermodynamics.