Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 17: Development of Calculation Methods
MM 17.2: Vortrag
Mittwoch, 19. März 2025, 10:30–10:45, H22
Charge Equilibration in Machine Learning Potentials — •Martin Vondrak1,2, Johannes Margraf1, and Karsten Reuter2 — 1Bayreuth University, Bayreuth, Germany — 2Fritz-Haber-Institut der Max-Planck-Gesellschaft, Berlin, Germany
Machine learning (ML) techniques have recently been shown to bridge the gap between accurate first-principles methods and computationally cheap empirical potentials. This is achieved by learning a mapping between a system's structure and its physical properties. To this end, state-of-the-art models typically represent chemical structures in terms of local atomic environments. This inevitably leads to the neglect of long-range interactions (most prominently electrostatics) and non-local phenomena (e.g. charge transfer), resulting in significant errors in the description of e.g. polar molecules or materials in non-isotropic environments. To overcome these issues, we are developing ML frameworks for predicting charge distributions in molecules based on Charge Equilibration (QEq). Here, atomic charges are derived from a physical model using environment-dependent atomic electronegativities. In this presentation, we will demonstrate strategies for creating long-range interatomic potentials on the example of Kernel Charge Equilibration (kQEq) models combined with local Gaussian Approximation Potentials (GAP). An alternative approach, incorporating QEq into the equivariant MACE neural network scheme will also be discussed.
Keywords: Machine Learning; Charge Transfer; Interatomic Potentials; Electrostatics