Parallel Monte Carlo simulations

Abstract

The Monte Carlo (MC) method is an important tool in sampling the state space of a chosen statistical ensemble. It allows the study of thermodynamic averages of configurational properties by generating ‘‘moves’’ in a system and accepting or rejecting the thus generated new state depending on the energy of the new system and/or a random choice. These moves are intrinsically sequential and complicate parallel implementation. We propose a method which allows the parallel generation of MC moves, and which is especially useful for simulations with unavoidably low acceptance rates, such as for long chain molecules.