Efficient equivariant model for machine learning interatomic potentials

Abstract

In modern computational materials, machine learning has shown the capability to predict interatomic potentials, thereby supporting and accelerating conventional molecular dynamics (MD) simulations. However, existing models typically sacrifice either accuracy or efficiency. Moreover, efficient models are highly demanded for offering simulating systems on a considerably larger scale at reduced computational costs. Here, we introduce an efficient equivariant graph neural network (E²GNN) that can enable accurate and efficient interatomic potential and force predictions for molecules and crystals. Rather than relying on higher-order representations, E²GNN employs a scalar-vector dual representation to encode equivariant features. By learning geometric symmetry information, our model remains efficient while ensuring prediction accuracy and robustness through the equivariance. Our results show that E²GNN consistently outperforms the prediction performance of the representative baselines and achieves significant efficiency across diverse datasets, which include catalysts, molecules, and organic isomers. Furthermore, we conduct MD simulations using the E²GNN force field across solid, liquid, and gas systems. It is found that E²GNN can achieve the accuracy of ab initio MD across all examined systems.