Generalized Langevin equation thermostats for ab initio molecular dynamics.

Molecular dynamics can be used as a very effective, general-purpose approach to gen-erate atomic configurations for a material or chemical compound that are consistent with prescribed experimental conditions. It is an ideal companion to ab initio electronic structure methods, that allow one to evaluate the inter-atomic forces from first principles, avoiding the time-consuming and at times inaccurate process of designing an empirical model. Un-fortunately, ab initio methods are still computationally demanding, and so it is desirable to modify the properties of the dynamics to get a more effective exploration of phase space than allowed by the intrinsic system’s dynamics. A very effective approach to manipulate the sampling behaviour of a molecular dynamics simulation exploits non-Markovian, Generalized Langevin Equations (GLEs) that have been used in the past to model the coupling of the system with a physical bath. Within this framework, the impact of the correlated noise on static and dynamic properties of the trajectory can be predicted analytically in the harmonic limit. One can therefore optimize the performances of the GLE a priori, and obtain effects that range from effective sampling of all the relevant time scales, to fine-tuned control o