(63b) Long Range Interactions in Machine Learning Models of Molecular Systems

Computer simulations can provide important insights into the thermodynamics of molecular systems. The accuracy of simulation predictions relies on how well interatomic interactions are modeled. For charged and polar systems relevant to electrochemical processes, for example, long range electrostatic interactions are particularly important. These interactions dictate screening and response to applied fields, yet still pose computational and theoretical difficulties. In this talk, I will first review the importance of long-range interactions in molecular systems. As part of this discussion, I will give a brief overview of machine learning-based neural network models. Then, I will discuss how a physical understanding of interatomic interactions can be used to include the effects of long-range interactions in neural network models. By focusing on the correct physics, the resulting models are partially transferable and can describe electronic and nuclear response to external fields. I will then demonstrate the accuracy and transferability of this neural network approach – the self-consistent field neural network (SCFNN) – on model systems before closing with a discussion of ongoing research relevant to materials for energy.