Non-equilibrium MCMC in general spaces - Joris Bierkens

External forcing may increase mixing speed. This observation makes perfect sense when one thinks of stirring a cup of coffee with sugar inside.

In Markov Chain Monte Carlo, a limitation of fundamental algorithms (e.g. Metropolis-Hastings and Gibbs sampling) is that these provide time reversible Markov chains. In the analogy of the cup of coffee: no stirring takes place.

We discuss examples in the literature where mixing speed is increased by applying non-reversible methods. Also we discuss a framework which in principle should be able to provide non-reversible MCMC algorithms for general probability densities.