Go with the Winners: a General Monte Carlo Strategy

Abstract

We describe a general strategy for sampling configurations from a given distribution, NOT based on the standard Metropolis (Markov chain) strategy. It uses the fact that nontrivial problems in statistical physics are high dimensional and often close to Markovian. Therefore, configurations are built up in many, usually biased, steps. Due to the bias, each configuration carries its weight which changes at every step. If the bias is close to optimal, all weights are similar and importance sampling is perfect. If not, ``population control" is applied by cloning/killing partial configurations with too high/low weight. This is done such that the final (weighted) distribution is unbiased. We apply this method (which is also closely related to diffusion type quantum Monte Carlo) to several problems of polymer statistics, reaction-diffusion models, sequence alignment, and percolation.