Fluctuations of entropy production rate in turbulent thermal convection

We use Direct Numerical Simulation (DNS) to compute the entropy production rate ($\sigma$) in turbulent thermal convection. The overall entropy production rate is measured injecting a large number of pointwise lagrangian tracers within the flow and simultaneously sampling velocity and temperature fields along the tracers trajectory. For an isolated system, classical thermodynamics prescribes that $\sigma$ always increases until equilibrium. However, we show here that the entropy production rate is characterized by large fluctuations and becomes often negative. We also discuss the fluctuations of $\sigma$ averaged over a time lag $\tau$ in the framework of the fluctuation theorem.