Monday 21 Oct 2013: Speeding up convergence to equilibrium for diffusion processes
Greg Pavliotis - Imperial College, London
Harrison 101 15:00-16:00
In many applications it is necessary to sample from a probability distribution that is known up to a constant. A standard technique for doing this is by simulating a stochastic differential equation whose invariant measure is the probability measure from which we want to sample. There are (infinitely) many different diffusion processes that have the same invariant distribution. It is natural to choose a diffusion process that converges as quickly as possible to equilibrium, since this reduces the computational cost. In this talk I will present some recent results on optimizing the rate of convergence to equilibrium by adding an appropriate non-reversible perturbation to the standard reversible overdamped Langevin dynamics. This is joint work with T. Leliever and F. Nier.