Dissipative Particle Dynamics and Coarse−Graining: Review of Existing Techniques‚ Trials with Evolutionary Computation

Abstract

This thesis provides a review of the dissipative particle dynamics (DPD) technique, a commonly used mesoscopic simulation tool in computational physics; and an investigation of the feasibility of using evolutionary optimization techniques for the determination of interactions in the DPD model from measurements in atomistic simulations. The text starts with a brief overview of the historical development of particle models to provide a foundation for the discussion of coarse-graining, i.e. the description of a system at a less detailed level by smoothing out fine details that are not relevant for a particular study. Detailed introductions of fundamental computational physics methods are presented, such as molecular dynamics and Monte Carlo simulations, together with their application areas. The DPD technique is introduced, with detailed information about its historical development, interpretation as a mesoscopic model, and application areas. The two parts of the DPD coarse-graining process, i.e. the determination of conservative and dissipative interactions, are discussed. Major existing techniques for DPD coarse-graining are presented, such as the inverse Monte Carlo (IMC) procedure specialized for the determination of conservative interactions from structural observables. The thesis continues with an investigation of the feasibility of using evolutionary computation, a generic optimization approach with its roots in the biological process of evolution, for the determination of interactions in the DPD model, based on fitness measures comparing equilibrium and transport properties of the system with those measured in atomistic simulations. Taking the simple point charge water model as a case study, the technique is first used for the determination of conservative interactions from the radial distribution function (with the aim of validating the approach by results from the IMC technique) and after that, for the determination of dissipative interactions based on escape time distributions. The practicality of having relatively long DPD simulations within fitness evaluations of such a procedure is confirmed, also establishing a general framework for applying evolutionary optimization techniques for the determination of functional forms in possibly other models within the field of computational physics.