Mutual information driven colloidal heat engine*

We report on the direct measurement of the mutual information as a function of error size for a Brownian information engine operating in non-equilibrium steady state. Each engine cycle consists of the measurement of the particle position, feedback control, resetting of the particle position and relaxation. The measurement involves a Gaussian noise of controlled width. The performance of the information engine depends on the cycle period τ and the width of the noise N. The mutual information decreases with decrease in τ and increase in N, thereby reducing the amount of work extraction. Our system operates as a cooling or a heating device depending on the noise width. The efficiency of information-to-work conversion increases as the system is allowed to relax more at the end of each cycle. The maximum efficiency is obtained at the finite value of N. The generalized Jarzynski equality was found to be valid either when the initial state of the system is in thermal equilibrium, or when noise and signal width are equal.