Parts | Days | Selection | Search | Updates | Downloads | Help

KFM: Fachverband Kristalline Festkörper und deren Mikrostruktur

KFM 7: Materials for the Storage and Conversion of Energy (joint session MM/KFM)

KFM 7.1: Talk

Monday, March 18, 2024, 11:45–12:00, C 264

Symmetry Matters: Machine Learning EFG Tensors for NMR Parameter Simulations — •Angela F. Harper¹, Simone Köcher^1,2, Karsten Reuter¹, and Christoph Scheurer¹ — ¹Fritz-Haber-Institut der MPG, Berlin — ²IEK-9 Forschungszentrum Jülich

Electric field gradient (EFG) tensors are directly probed by experimental solid-state Nuclear Magnetic Resonance (NMR), and are crucial for deciphering the atomic-scale structure and dynamics of Li-ion battery materials. By employing a machine learning approach we devise a model capable of learning complete EFG tensors, using equivariant descriptors. We further show that it is not sufficient to simply learn scalar quantities derived from a tensor such as quadrupolar shift or asymmetry. To assess the model’s performance, we curate an extensive dataset comprising over 60,000 EFG tensors calculated for a diverse set of equilibrium and non-equilibrium crystal structures of Li₄Ti₅O₁₂ (LTO), a well-studied zero-strain insertion anode material in Li-ion batteries. We finally show that we predict the quadrupolar frequency to within a few kHz for the ⁷Li nucleus, which is well within the level of error required to make meaningful predictions for ⁷Li NMR. This work represents a significant step towards realizing in silico spectroscopy: the ability to calculate spectroscopic signals such as EFG tensors with the same accuracy as experimental spectroscopy, using machine learning.

Keywords: machine learning; NMR; lithium ion battery; spectroscopy; database