(746c) Improved Coarse-Grained Models through Local Density Potentials Optimized with the Relative Entropy

Bottom-up multiscale techniques are often used to develop coarse grained (CG) models of atomistic molecular simulations to probe large length and time scales. Conventionally, CG models use pair potentials for nonbonded forces to capture the renormalization of detailed interactions of a reference all-atom system. Such pair potentials, while computationally attractive, neglect the inherently multibody contributions of the local environment of a CG site to its energy, which is inevitably embedded in the CG model as a result of integrating over removed degrees of freedom. This makes CG models highly sensitive to the thermodynamic state point (temperature, density, concentration, etc.) of the all-atom reference from which they are parameterized, and as such, limits their transferability to other state points.

 Here, we propose to improve CG models through computationally fast many body potentials, in which multibody energies depend in a mean-field manner on local densities of different atomic species and serve as additive corrections to conventional CG pair potentials. These local density potentials have the same mathematical structure as embedded-atom or bond-order models designed to capture multibody electronic effects in metallic systems. We show that the relative entropy coarse-graining framework offers a systematic route to parameterizing local density potentials.

 We explore this approach in the development of implicit solvation strategies for aqueous solvation of hydrophobic solutes. CG models of solutes and macromolecules that represent solvent implicitly through effective intra-solute interactions offer tremendous potential speed-ups through the elimination of explicit solvent molecules. Our results reveal that the local density approach enables the models to capture higher order cooperativity inherent to water-mediated hydrophobic interactions, which results in better transferability of local density augmented implicit solvent CG forcefields. More generally, this approach can be easily extended to arbitrary CG models of systems beyond solvation, e.g. phase behavior, where density and concentration transferability play an important role.