Leibniz MMS Days 2024 - Abstract

Werner, Marco

Machine learning of an implicit solvent for dynamic Monte Carlo simulations

We discuss an implicit solvent model based on an artificial neural network (NN) for dynamic Monte Carlo simulations, where the dynamics is implemented via local particle displacements (elementary moves). The training data was obtained from coarse grained simulations using the bond fluctuation model with explicit solvent for single homopolymers under variation of solvent quality. The NN based implicit solvent model takes into account only the information of the local environment of monomers in order to predict a distribution of possible acceptance rates of an attempted elementary monomer move in a given configuration. We show that NN based implicit solvent simulations reproduce coil-globule transition as seen in the explicit solvent model ? including static and dynamic properties of the polymer in a liquid globule state.