Grid computing multiple shooting algorithms for extended phase space sampling and long time propagation in molecular dynamics

Abstract

Grid computing refers to a well established computational platform of geographically distributed computers that offer a seamless, integrated, computational and collaborative environment. It provides the means for solving highly demanded, in computer time and storage media problems of molecular dynamics. However, because of the rather high latency network, to exploit the unprecedented amount of computational resources of the Grid, it is necessary to develop new or to adapt old algorithms for investigating dynamical and statistical molecular behaviors at the desired temporal and spatial resolution. In this chapter we review methods that assist one to harness the current computational Grid infrastructure for carrying out extended samplings of phase space and integrating the classical mechanical equations of motion for long times. Packages that allow to automatically submit and propagate trajectories in the Grid and to check and store large amounts of intermediate data are described. We report our experience in employing the European production Grid infrastructure for investigating the dynamics and free energy hypersurfaces of enzymes such as Cytochrome c Oxidases. Time autocorrelation functions of dynamical variables yield vibrational spectra of the molecule and reveal the localization of energy in specific bonds in the active site of the enzyme. Dynamical calculations and free energy landscapes of the Cytochrome c Oxidase protein interacting with gases like O2, CO and NO reveal the pathways for the molecules to penetrate in the cavities of the enzyme and how they reach the active site where the reactions take place. The discussed methods can be adopted in any intensive computational campaign, which involves the scheduling of a large number of long term running jobs.