Transferable local density-dependent friction in tert-butanol/water mixtures

Abstract

Coarse-grained (CG) models informed from molecular dynamics simulations provide a way to represent the structure of an underlying all-atom (AA) model by deriving an effective interaction potential. However, this leads to a speed-up in dynamics due to the lost friction, which is especially pronounced in CG implicit solvent models. Applying a thermostat based on the Langevin equation (LE) provides a way to represent the long-time dynamics of CG particles by reintroducing friction to the system. To improve the representability of CG models of heterogeneous molecular mixtures and their transferability over the mixture compositions, we parameterize an LE thermostat in which the friction coefficient depends on the local particle density (LD). The thermostat friction was iteratively optimized with a Markovian variant of the recently introduced Iterative Optimization of Memory Kernels (IOMK) method. We simulated tert-butanol/water mixtures over a range of compositions, which show a distinct clustering behavior. Our model with LD-dependent friction reproduces the AA diffusion coefficients well over the full range of mixtures and is, therefore, transferable with respect to dynamics.