MEAM Seminar: “Neural Network Assisted Open Boundaries for Molecular Dynamics Simulations”

Molecular dynamics simulations allow for the visualization and analysis of atoms and molecules. The choice of boundary conditions in these simulations can impact the overall system, and is an important design decision. In particular, open boundary molecular dynamics simulations is one type of methodology that allows for the treatment of atomistic models with non equilibrium conditions. In this seminar, I introduce an approach to the treatment of open boundaries using machine learning. A neural network-assisted design will be presented that can emulate the physics and reduce the computational cost of open boundary simulations. Particle influxes and neural network-derived forces are applied at the boundaries of an open domain consisting of explicitly modeled Lennard-Jones atoms in order to represent the effects of an unmodeled surrounding fluid. Canonical ensemble simulations with periodic boundaries are used to train the neural network and to sample boundary fluxes. The method, as implemented in the LAMMPS molecular simulation package, yields results comparable to those calculated using periodic molecular dynamics and runs two orders of magnitude faster than a comparable grand canonical molecular dynamics system.