Biased Metropolis Sampling for Rugged Free Energy Landscapes

Abstract

Metropolis simulations of all-atom models of peptides (i.e. small proteins) are considered. Inspired by the funnel picture of Bryngelson and Wolyness, a transformation of the updating probabilities of the dihedral angles is defined, which uses probability densities from a higher temperature to improve the algorithmic performance at a lower temperature. The method is suitable for canonical as well as for generalized ensemble simulations. A simple approximation to the full transformation is tested at room temperature for Met-Enkephalin in vacuum. Integrated autocorrelation times are found to be reduced by factors close to two and a similar improvement due to generalized ensemble methods enters multiplicatively.

Plenary talk at the Los Alamos conference, The Monte Carlo Method in Physical Sciences: Celebrating the 50th Anniversary of the Metropolis Algorithm, to appear in the proceedings, 11 pages, 4 figures, one table. Inconsistencies corrected and references added