Applied Math Seminar

Spring Quarter 2007

Friday October 05, 3:15p.m.
Sloan Mathematics Corner, Building 380, Room 380-C

Alexandre Chorin

Monte Carlo without chains.

Monte Carlo methods in statistical physics are synonymous with Markov chain Monte Carlo methods, for good reasons which I will briefly recapitulate. Yet there are problems, for example spin glass problems, where MCMC methods perform so poorly that it is time to consider alternatives.

I will present one such alternative, where spin arrays are sampled from the top down through nested sequences of marginals. This can work if one can estimate marginals rapidly. Applications will be discussed.

Seminar website: http://math.stanford.edu/~applmath/