Atomistic Line Graph Neural Network for improved materials property predictions

Graph neural networks (GNNs) have shown significant improvements in atomistic material representation and modeling compared to descriptor-based machine learning models. However, most GNNs for atomistic predictions rely on atomic distance information without explicitly incorporating bond angles, which are crucial for differentiating many atomic structures and influence material properties. This paper introduces the Atomistic Line Graph Neural Network (ALIGNN), a GNN architecture that performs message passing on both the interatomic bond graph and its line graph (representing bond angles). ALIGNN demonstrates improved performance on multiple atomistic prediction tasks by efficiently incorporating angle information. The model's effectiveness is showcased by predicting 52 solid-state and molecular properties from JARVIS-DFT, Materials Project, and QM9 databases, outperforming some existing GNN models in accuracy and/or training speed.

Kamal Choudhary, Brian DeCost