In meteorology and climate science, fluid models are simulated on time scales very long compared to the characteristic Lyapunov time of chaotic growth, with the goal of generating a data set suitable for statistical analysis. The choice of a numerical discretization scheme for a problem carries with it a certain bias in the statistical data generated in long simulations. In this talk I will discuss research on the use of thermostat techniques, commonly used in molecular dynamics, to control the invariant measure of a discretized model, with the goals of correcting bias or effecting a statistically consistent model reduction. These will be illustrated for a point-vortex gas and a Burgers/KdV equation. Finally I will discuss new extensions of the approach to cases where observations are available.
view more