Dresden 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 19: Poster Session
MM 19.37: Poster
Dienstag, 10. März 2026, 18:00–20:00, P5
Neural Network Potentials for Molecular Dynamics Simulations of NASICON Solid Electrolytes — •Insa F. de Vries and Nikos L. Doltsinis — Institute of Solid State Theory, University of Münster, Wilhelm-Klemm-Straße 10,48149 Münster
Solid-state electrolytes are by now well-established components for the development of safe and efficient batteries. Among them, the sodium superionic conductor (NASICON) family offers an appealing degree of tunability. Their open framework allows for the substitution of lattice sites and the adjustment of diffusing ion concentration, making them an attractive choice for material design.
Accurately modeling and predicting ion and thermal transport properties in these systems by ab initio molecular dynamics (AIMD) simulations poses a challenge due to the large supercells and time scales required. In this study, we therefore train neural network (NN) potentials for various members of the Na1+xZr2SixP3−xO12 (x=0, 1, 2, 3) family using training data obtained by accelerated AIMD [1]. The NN potentials are then employed to calculate the sodium ion diffusion coefficient – a key transport property. The performance and results obtained with the NN potential are validated against those obtained from ab initio trajectories. In addition, both the NN potential and the ab initio results are then compared to corresponding data generated with a previously developed force field [2].
[1] D.Hamelberg et. al, J. Chem. Phys. 2004, 120, 11919
[2] P. Kumar & S. Yashonath, J. Am. Chem. Soc. 2002, 124, 3828
Keywords: AIMD; NASICON; diffusion; neural network