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

Area of Science:

• Quantum Chemistry
• Materials Science
• Computational Materials Science

Background:

• Deep learning interatomic potentials offer efficient alternatives to ab initio calculations.
• Complex deep learning models often lack physical interpretability and robustness due to numerous parameters.

Purpose of the Study:

• To introduce graph-EAM, a lightweight graph neural network (GNN) for modeling interatomic potentials in single-element structures.
• To enhance the interpretability and robustness of machine learning potentials.

Main Methods:

• Developed graph-EAM, a GNN inspired by the empirical embedded atom method.
• Incorporated angular information via three-body atomic density.
• Trained and validated on platinum, niobium, silicon, and amorphous carbon systems.

Main Results:

• Graph-EAM achieved high energy and force prediction accuracy comparable to or better than state-of-the-art models.
• The model demonstrated superior performance with significantly fewer parameters.
• Inclusion of angular information improved prediction accuracy.

Conclusions:

• Graph-EAM provides an accurate and efficient method for interatomic potential modeling.
• The lightweight architecture enhances interpretability and robustness.
• This approach can accelerate molecular dynamics simulations in materials science.