Related Experiment Video
Updated: Jun 12, 2025
10:03
Characterization of Electrode Materials for Lithium Ion and Sodium Ion Batteries Using Synchrotron Radiation Techniques
Published on: November 11, 2013
25.4K
Tracking Li atoms in real-time with ultra-fast NMR simulations
Angela F Harper1, Tabea Huss1, Simone S Köcher1,2
1Fritz-Haber Institute of the Max Planck Society, Berlin, Germany. harper@fhi.mpg.de.
Faraday Discussions
|September 18, 2024
Summary
This study introduces a multiscale machine learning method to simulate atomic structures and dynamics for solid-state Nuclear Magnetic Resonance (ssNMR) in lithium-ion solid electrolytes. The approach accurately predicts lithium-ion diffusion rates, advancing NMR crystallography for materials discovery.
Area of Science:
- Computational Materials Science
- Solid-State Nuclear Magnetic Resonance (ssNMR)
- Machine Learning in Chemistry
Background:
- Accurate simulation of atomic structure and dynamics is crucial for understanding solid-state electrolyte materials like Li3PS4 (LPS).
- Solid-state Nuclear Magnetic Resonance (ssNMR) observables, particularly 7Li quadrupolar frequencies, provide insights into Li-ion diffusion.
- Existing machine learning force fields often struggle to capture the microsecond timescales required for ssNMR analysis of Li-ion dynamics.
Purpose of the Study:
- To develop and apply a multiscale machine learning approach for joint simulation of atomic structure, dynamics, and ssNMR observables.
- To investigate Li-ion diffusion in Li3PS4 (LPS) using spin-alignment echo (SAE) NMR and machine learning.
- To enable accurate predictions of ssNMR data on experimentally relevant timescales and temperatures.
Main Methods:
- Utilized ultra-fast potentials (UFPs), a novel class of machine learning interatomic potentials, to access long timescales (>1 microsecond).
- Developed a machine learning model to predict 7Li electric field gradient (EFG) tensors in LPS.
- Combined UFP trajectories with EFG tensor predictions to compute the autocorrelation function (ACF) of 7Li quadrupolar frequencies.
Main Results:
- Successfully simulated Li-ion diffusion dynamics in both crystalline β-LPS and amorphous LPS on microsecond to millisecond timescales.
- Extracted Li hopping rates from the ACF of quadrupolar frequencies, showing good agreement with Li dynamics.
- Demonstrated that the machine learning approach can predict Li hopping rates on experimentally relevant timescales.
Conclusions:
- The multiscale machine learning approach enables efficient and accurate simulation of ssNMR observables for solid-state materials.
- This method opens new avenues for NMR crystallography, particularly for polycrystalline and glass ceramic materials.
- The study highlights the potential of machine learning to bridge the gap between simulation and experimental observations in materials science.
