Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 27: Transport in Materials: Diffusion, Charge or Heat Conduction
MM 27.3: Vortrag
Donnerstag, 20. März 2025, 10:45–11:00, H22
Modelling complex proton transport phenomena - Exploring the limits of fine-tuning and transferability of foundational machine-learned force fields — •Christian Dreßler1, Malte Grunert2, Jonas Hänseroth1, Max Großmann2, and Erich Runge2 — 1TU Ilmenau, Institute of Physics, Theoretical Solid State Physics — 2TU Ilmenau, Institute of Physics, Group of Theoretical Physics 1
The solid acids CsH2PO4 and Cs7(H4PO4)(H2PO4)8 pose significant challenges for the simulation of proton transport phenomena. In this talk, we present the use of the recently developed machine-learned force field MACE to model proton dynamics on nanosecond timescales for these systems and compare its performance with long-term ab initio molecular dynamics (AIMD) simulations. The MACE-MP-0 foundation model shows remarkable performance for all observables derived from molecular dynamics simulations, but minor quantitative discrepancies remain compared to the AIMD reference data. However, we show that minimal fine-tuning - fitting to as little as 1 ps of AIMD data - leads to full quantitative agreement between the radial distribution and autocorrelation functions of MACE force field and AIMD simulations. Long-time AIMD simulations are unable to capture the correct qualitative trends in diffusion coefficients due to their inherent time scale limitations. In contrast, we demonstrate that accurate and convergent diffusion coefficients, consistent with experimental data, can only be reliably achieved through multi-nanosecond molecular dynamics simulations utilizing machine-learned force fields.
Keywords: machine-learned force field; ab initio molecular dynamics; solid acids; proton diffusion; machine-learned interatomic potential