Symmetry-adapted graph neural networks for constructing molecular dynamics force fields

TL;DR

This paper introduces MDGNN, a symmetry-adapted graph neural network framework that automatically constructs accurate, transferable force fields for molecular dynamics simulations of molecules and crystals, preserving key invariances.

Contribution

The paper presents a novel symmetry-adapted GNN architecture, MDGNN, that accurately and efficiently constructs force fields for large-scale molecular dynamics, including higher order features and transferability.

Findings

MDGNN accurately reproduces classical and first-principles MD results.

The model preserves translation, rotation, and permutation invariance.

Force fields from MDGNN show good transferability across systems.

Abstract

Molecular dynamics is a powerful simulation tool to explore material properties. Most of the realistic material systems are too large to be simulated with first-principles molecular dynamics. Classical molecular dynamics has lower computational cost but requires accurate force fields to achieve chemical accuracy. In this work, we develop a symmetry-adapted graph neural networks framework, named molecular dynamics graph neural networks (MDGNN), to construct force fields automatically for molecular dynamics simulations for both molecules and crystals. This architecture consistently preserves the translation, rotation and permutation invariance in the simulations. We propose a new feature engineering method including higher order contributions and show that MDGNN accurately reproduces the results of both classical and first-principles molecular dynamics. We also demonstrate that force fields constructed by the model has good transferability. Therefore, MDGNN provides an efficient and promising option for molecular dynamics simulations of large scale systems with high accuracy.