Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 9: Poster

MM 9.62: Poster

Montag, 17. März 2025, 18:30–20:30, P1

Integrating Long-Range Interactions into Machine Learning Interatomic Potentials — •Tulga-Erdene Sodjargal^1,2, Egor Rumiantsev¹, Philip Loche¹, and Michele Ceriotti¹ — ¹Laboratory of Computational Science and Modeling (COSMO), Institute of Materials, École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland — ²Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology (KAIST), 34141 Daejeon, Republic of Korea

Machine learning-based interatomic potentials (MLIPs) often rely on the locality ansatz, calculating atomic energies based on a fixed cutoff radius. While effective for many systems, this nearsightedness can lead to inaccuracies when long-range interactions, such as ionic interactions, dominate. To overcome this limitation, we integrate Particle Mesh Ewald (PME) techniques into existing MLIP frameworks. Our extension is modular and plug-and-play, requiring minimal modifications to incorporate into various models. We demonstrate significant improvements in both simple architectures, such as Behler-Parrinello Neural Networks, and advanced models, including state-of-the-art graph neural networks like the Point Edge Transformer (PET).

Keywords: Machine Learning; Long-Range; Particle Mesh Ewald (PME); Graph Neural Networks