Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 17: Data-driven Materials Science: Big Data and Workflows II
MM 17.3: Talk
Tuesday, March 10, 2026, 14:30–14:45, SCH/A251
Simultaneous Learning of Static and Dynamic Charges — Philipp Stärk1, •Philip Loche2, Marcel Langer1, Henrik Stooß1,3, Michele Ceriotti2, and Alexander Schlaich1,3 — 1Stuttgart Center for Simulation Science, University of Stuttgart, Germany — 2Laboratory of Computational Science and Modeling, École Polytechnique Fédérale de Lausanne, Switzerland — 3Institute for Physics of Functional Materials, Hamburg University of Technology, Germany
Long-range interactions and electric response are essential for accurate modeling of condensed-phase systems, yet remain challenging for atomistic machine learning. Static charges modulate Coulomb interactions, while dynamic charges such as atomic polar tensors describe the response to external electric fields. We compare strategies for learning both types of charges: independent models; coupled learning with or without an isotropic dielectric correction; and coupled learning with an environment-dependent screening. While screening corrections are crucial in the coupled case, assuming homogeneous, isotropic screening fails in heterogeneous systems such as water clusters. Learning a local screening restores accuracy for dynamic charges but offers negligible improvement over independent models while increasing computational cost.
Keywords: Atomistic machine learning; Long-range interactions; Charge predictions; Born effective charges; Atomic polar tensors