Long-range interactions in machine learning interatomic potentials

Develop methods to accurately incorporate long-range electrostatic and dispersion interactions into machine learning interatomic potentials for atomistic simulations while preserving physical consistency (e.g., energy conservation) and enabling stable, efficient simulations.

Background

Throughout the tutorial, the authors emphasize that most widely used machine-learning potentials rely on local, short-range descriptors and models, which can lead to significant errors in systems where long-range electrostatic or dispersion interactions are important. They review several approaches (e.g., third- and fourth-generation neural network potentials, charge equilibration schemes, and long-distance equivariant representations) but conclude that fully addressing non-local effects remains an open challenge.

In the Summary and Outlook, the authors explicitly identify long-range interactions as one of the open questions for the field, underscoring the need for new methods that integrate long-range physics into machine learning potentials without sacrificing accuracy or computational tractability.

References

There are still many open questions and challenges to be addressed, such as the long-range interactions, generalisation and interpretability.

— Introduction to machine learning potentials for atomistic simulations  (2410.00626 - Thiemann et al., 2024) in Summary and Outlook (Section 8)

To the best of our knowledge, such a treatment for dielectric systems has not yet been combined with MLIPs.

— Electrostatic Phenomenology Benchmarks for Machine-Learned Interatomic Potentials in Electrochemistry: Beyond the Energy-Force Metric  (2608.14153 - Sumić et al., 14 Aug 2026) in Section 3, subsection “2) Ionic and electronic screening”