Regensburg 2022 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 29: Data Driven Materials Science: Design of Functional Materials
MM 29.5: Talk
Thursday, September 8, 2022, 11:15–11:30, H45
Predicting oxidation and spin states by high-dimensional neural networks — •Knut Nikolas Lausch1, Marco Eckhoff1, Peter Blöchl2, and Jörg Behler1 — 1Georg-August-Universität Göttingen, Institut für Physikalische Chemie, Theoretische Chemie, Göttingen, Germany — 2Technische Universität Clausthal, Institut für Theoretische Physik, Clausthal-Zellerfeld, Germany
Machine learning potentials (MLP) such as high-dimensional neural network potentials (HDNNP) provide first-principles quality energies and forces enabling large-scale molecular dynamics simulations at low computational costs. However, most current MLPs do not provide any information about the electronic structure of the system, which is often important for a detailed understanding of complex systems such as transition metal oxides. The lithium intercalation compound LixMn2O4 (0≤ x ≤ 2), a commercially used cathode material in lithium ion batteries, is such a system since the manganese ions adopt different oxidation states based on the lithium content and distribution. Here, we propose a high-dimensional neural network (HDNN) that can predict atomic oxidation and spin states as a function of the local atomic environments in LixMn2O4. The HDNN can complement HDNNP-driven MD simulations giving insights into the underlying electronic processes that give rise to complex phenomena such as a charge ordering transition, and electrical conductance.