NewtonNet: A Newtonian message passing network for deep learning of interatomic potentials and forces

TL;DR

NewtonNet is a physics-inspired deep learning model that accurately predicts interatomic potentials and forces, leveraging Newtonian principles to improve interpretability, efficiency, and performance across diverse chemical datasets.

Contribution

The paper introduces NewtonNet, a novel message passing network that incorporates Newtonian physics, achieving state-of-the-art results in predicting energies and forces with enhanced interpretability and efficiency.

Findings

Achieves state-of-the-art accuracy on multiple datasets.

Demonstrates greater data and computational efficiency.

Maintains rotational equivariance and interpretable physical features.

Abstract

We report a new deep learning message passing network that takes inspiration from Newton's equations of motion to learn interatomic potentials and forces. With the advantage of directional information from trainable latent force vectors, and physics-infused operators that are inspired by the Newtonian physics, the entire model remains rotationally equivariant, and many-body interactions are inferred by more interpretable physical features. We test NewtonNet on the prediction of several reactive and non-reactive high quality ab initio data sets including single small molecule dynamics, a large set of chemically diverse molecules, and methane and hydrogen combustion reactions, achieving state-of-the-art test performance on energies and forces with far greater data and computational efficiency than other deep learning models.