Learning Long-Range Interactions in Equivariant Machine Learning Interatomic Potentials via Electronic Degrees of...

Area of Science:

• Computational Materials Science
• Machine Learning
• Quantum Mechanics

Background:

• Machine learning interatomic potentials (MLIPs) offer efficient alternatives to quantum mechanical simulations.
• Current MLIPs using graph neural networks struggle with long-range interactions and charge transfer due to local descriptors.

Purpose of the Study:

• Develop a novel equivariant MLIP that explicitly incorporates long-range Coulomb interactions.
• Improve the accuracy and efficiency of MLIPs for systems with charge heterogeneity.

Main Methods:

• Incorporated long-range Coulomb interactions by treating electronic degrees of freedom and global charge distribution.
• Utilized a charge equilibration scheme based on predicted atomic electronegativities.
• Developed a new equivariant MLIP model.

Main Results:

• The new MLIP outperforms existing short-range and long-range MLIPs in energy and force predictions.
• Achieved higher accuracy with a smaller cutoff radius (4 Å) compared to short-range models (6 Å).
• Demonstrated superior performance on both periodic and nonperiodic benchmark datasets.

Conclusions:

• The developed MLIP accurately simulates systems with long-range interactions and charge heterogeneity.
• This advancement expands the applicability of MLIPs in computational materials science.
• Offers a more efficient and accurate approach for materials simulations.