A novel algorithm for creating density dependent, coarse-grained models for the simulation of surfactant systems

Abstract

Large-scale simulations of solvated molecules that treat the solvent explicitly are very computationally expensive, and as a result work has been done on modifying the potentials to treat solvent implicitly. Implicit solvation is well-known in Brownian Dynamics of dilute solutions, but offers promise to speed up many other types of molecular simulations as well, including studies of proteins and colloids where the local density can vary considerably. This work examines implicit solvent potentials within a more general coarse-graining framework. While a pairwise potential between solute sites is relatively simple and ubiquitous, an additional parameterization based on the local solute concentration has the possibility to increase the accuracy of the simulations with only a marginal increase in computational cost. In this thesis we describe a method in which the radial distribution function (RDF) and excess chemical potential of solute insertion ([mu]ex) for a system of Lennard-Jones particles are first measured in a fully explicit, all-particle simulation, and then reproduced across a range of solute particle densities in an implicit solvent simulation. The resulting potentials are densitydependent, implicit solvent (DDIS) potentials. We then test the transferability of DDIS potentials to mixtures and systems of chains without additional optimization. We find that RDF transferability to mixtures is very good and RDF errors in systems of chains increase linearly with chain length. Excess chemical potential transferability is good for mixtures at low solute concentration, chains, and chains of mixed composition; at higher solute concentrations in mixtures, chemical potential transferability fails due to the unique property of DDIS potentials that inserting a single particle changes the densities of all neighboring particles. Based these results, we demonstrate that DDIS potentials derived for pure solutes can be used effectively in the study of many important systems including those involving mixtures, chains and chains of mixed composition. Finally, the DDIS potentials are used to examine the self-assembly of a model surfactant system.

(cont.) We demonstrate that the coarse-grained DDIS potentials generated by this method accurately predict the trends in critical micelle concentration (CMC) for two surfactant types, but that the absolute values of the predicted CMC are an order of magnitude higher than previously established estimates for the same surfactants using atomistic simulations. Additionally, the micelles formed are less densely packed than the corresponding all-atom micelles, leading to a larger average aggregation number. By examining a series of simulations of increasing molecular complexity, we identify the source of this error with the transferability of the DDIS potentials. The results suggest that deriving the DDIS potentials directly from simulations of chain molecules in solvent could improve the ability of such potentials to reproduce surfactant properties accurately. The method for deriving DDIS potentials is extremely general and can be applied to study a variety of solvated systems of any chemical complexity. For example, the results of this work can be extended to study problems in protein folding, drug uptake in micellar systems, and biological membranes.