Improving Monte Carlo atomic simulations with coordinate transformation

PHYS 358

Ivan Vinogradov, vinogri@mcmaster.ca and Paul W. Ayers, ayers@mcmaster.ca. Department of Chemistry, McMaster University, Hamilton, ON L8S 4M1, Canada
Traditional Monte Carlo simulations of molecular systems perform trial moves one atom at a time. When the motions of the atoms are highly correlated, however, a "smarter" algorithm would move a group of closely interacting atoms instead. This would result at a higher rate of acceptance and fewer simulation steps. The potential energy surface of a system should provide enough information about connectedness of atoms. By transforming atomic coordinates into collective coordinates that take this connectedness into account, we can perform Monte Carlo step in collective coordinates and then invert the transformation to recover new atomic coordinates. Results of application of this approach to Lennard-Jones clusters will be presented. Our ultimate goal is to apply this method to proteins.