Gallavotti–Cohen fluctuation theorem: Universal law of non-equilibrium statistical mechanics.

The irreversible behaviour of macroscopic processes governed by reversible laws of classical or quantum physics has been a captivating subject in statistical mechanics going back at least to the pioneering work of Boltzmann. The general consensus reached by the middle of the last century was that the second law of thermodynamics, which states that entropy increases with time, is empirical in nature, and that the probability of a negative fluctuation of entropy is so small that it cannot be observed in practice. A natural question is the quantitative description of that claim. A breakthrough on this subject came in the middle of nineties due to the work of Evans-Searles and Gallavotti-Cohen.

In this talk, I shall illustrate their discovery on the simplest example of finite state Markov chains. It will be shown that, under some natural hypotheses, one can define an entropy production observable whose time averages satisfy the large deviation principle. The resulting rate function possesses a symmetry property which implies that the probability of a negative value for the mean entropy production is exponentially suppressed by that of the opposite positive value. I shall also discuss the realisation of this programme in the context of fluid flows.