Regensburg 2019 – wissenschaftliches Programm
CPP 25.29: Poster
Dienstag, 2. April 2019, 14:00–16:00, Poster B1
Machine learning to predict three-body contributions in coarse-graining — •René Scheid, Christoph Scherer, Denis Andrienko, and Tristan Bereau — Max Planck Institute for Polymer Research, Mainz, Germany
Coarse-graining (CG) is a method to systematically reduce the degrees of freedom of a given system. Projecting the fine-grained system into a space with lower degrees of freedom leads to higher order interaction terms which are not considered in most CG models. We employ Kernel-based Machine learning (ML) to predict CG contributions based on decomposed CG energies and forces of atomistic simulations. The ML scheme is implemented in the VOTCA-CSG toolkit. First, we show that the ML scheme can recover two-body CG force fields generated by standard force-matching. This demonstrates that the approach is suitable to complement and expand standard force-matched models. Furthermore, we examine the expansion to three-body contributions. Decomposing the CG atomistic interactions into two- and three-body terms, the residual three-body potentials could be used to improve existing two-body models. The approach is illustrated on a Lennard-Jones liquid as test system, liquid water, and liquid methanol.