Multi-particle sampling in Monte Carlo simulations on fluids: Efficiency and extended implementations
Résumé
Various technical aspects affecting efficiency of a recently proposed novel Monte Carlo simulation scheme based on biased simultaneous displacements/rotations of {\it all} particles of the system are investigated using two polarizable models of water, the Chialvo-Cummings and Brodholt-Sampoli-Vallauri models, as a test case. Necessary expressions for polarizable site-site interaction models are derived along with a novel smoothing of the potential at the cutoff distance. In addition to the common thermodynamic and structural properties, the mean squared displacements, rotation relaxation, speed of equilibration (translational order parameter), and autocorrelation coefficients have been computed as well in order to assess efficiency of the method. Gain in speed by parallelization has also been examined. Performance of the method is compared with both the standard one-particle move method and available approximate methods. It is shown that the multi-particle move method performs about by a factor of 10 faster for the systems considered when compared with the common Monte Carlo scheme, and several times faster when compared with approximate methods. Parallelized codes of the multi-particle move method may then perform about seventy times faster than the conventional Monte Carlo. These conclusions hold true for the system size simulated (N=256) because the efficiency of the multi-particle method depends on the size of the system: its efficiency even increases with increasing number of particles.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...