Graph-EAM: An Interpretable and Efficient Graph Neural Network Potential Framework

The development of deep learning interatomic potentialshas enabledefficient and accurate computations in quantum chemistry and materialsscience, circumventing computationally expensive ab initio calculations. However, the huge number of learnable parameters indeep learning models and their complex architectures hinder physicalinterpretability and affect the robustness of the derived potential.In this work, we propose graph-EAM, a lightweight graph neural network(GNN) inspired by the empirical embedded atom method to model theinteratomic potential of single-element structures. Four materialsystems: platinum, niobium, silicon, and amorphous-carbon, for whichquantum simulation data sets are publicly available, are examinedto demonstrate that graph-EAM can achieve high energy and force predictionaccuracy comparable or better than existing state-of-the-artmachine learning models with much fewer parameters. It is alsoshown that the explicit inclusion of the angular information via three-bodyatomic density increases the prediction accuracy. The accuracy andefficiency of potentials obtained from graph-EAM can help acceleratethe molecular dynamics simulation.