Large Deviations in Thermodynamic Computation*

We propose a framework for the study of large deviations in thermodynamic computation with a long memory. The thermodynamic computation is done by a machine in contact with thermal reservoir known as information engine which manipulates information on the input to tape to information on the output tape by some given stochastic law. Information on the input tape can be highly correlated with a really long memory and as a result, Non-Markovian which makes the problem non-trivial. Here we study large deviations of computation accuracy and also cost of computation.