Battery charging in collision models with Bayesian risk strategies

Abstract

We construct a collision model where measurements in the system, together with a Bayesian decision rule, are used to classify the incoming ancillas as having either high or low ergotropy (maximum extractable work). The former are allowed to leave, while the latter are redirected for further processing, aimed at increasing their ergotropy further. The ancillas play the role of a quantum battery, and the collision model therefore implements a Maxwell demon. To make the process autonomous, and with a well defined limit cycle, the information collected by the demon is reset after each collision by means of a cold heat bath.