Regensburg 2019 – wissenschaftliches Programm

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

MM: Fachverband Metall- und Materialphysik

MM 10: Methods in Computational Materials Modelling (methodological aspects, numerics)

MM 10.6: Vortrag

Montag, 1. April 2019, 17:00–17:15, H45

A Neural Network Potential for Lithium Manganese Oxides — •Marco Eckhoff¹, Peter Blöchl², and Jörg Behler¹ — ¹Universität Göttingen, Institut für Physikalische Chemie, Theoretische Chemie, Tammannstraße 6, 37077 Göttingen, Germany — ²Technische Universität Clausthal, Institut für Theoretische Physik, Leibnizstraße 10, 38678 Clausthal-Zellerfeld, Germany

The lithium manganese oxide spinel Li_xMn₂O₄, with 0<x<2, is an important cathode material in lithium ion batteries. Its accurate description by density functional theory (DFT) is far from trivial due to several energetically close electronic and magnetic states. In extensive benchmark studies we find that the hybrid functionals PBE0, HSE06, and PBE0r yield energetic, structural, electronic, and magnetic properties in good agreement with experiment. Building on such hybrid DFT data, we are able to extend the time and length scales of molecular dynamics simulations of Li_xMn₂O₄ using a high-dimensional neural network potential, which provides a first-principles quality description of the potential energy surface at a fraction of the computational costs.