Simulating quadrupolar NMR dynamics in solid electrolyte Li10GeP2S12.

Area of Science:

• Solid-state chemistry
• Materials science
• Computational chemistry

Background:

• Quadrupolar solid-state nuclear magnetic resonance (NMR) spectroscopy is sensitive to lithium ion diffusion dynamics in solid electrolytes.
• Interpreting NMR data is challenging due to material complexity and motional narrowing effects.
• Atomic simulations are computationally expensive and often use idealized models.

Purpose of the Study:

• To develop a machine learning (ML)-assisted workflow to address experimental complexity in solid-state NMR studies of battery materials.
• To enable microsecond-scale molecular dynamics (MD) simulations and predict electric field gradient (EFG) tensors for lithium ion conductors.
• To accurately predict NMR observables and differentiate ion dynamics in lithium-ion battery materials.

Main Methods:

• Utilized ML acceleration for microsecond-scale MD simulations of Li10GeP2S12 (LGPS).
• Employed a tensorial model for efficient EFG tensor predictions from MD trajectories.
• Computed temperature-dependent 7Li NMR quadrupolar observables and emulated spin-alignment echo (SAE) experiments in silico.

Main Results:

• Achieved excellent agreement between predicted (24 kHz) and experimental (23 kHz) quadrupolar coupling for tetragonal LGPS.
• Successfully predicted temperature dependence of 7Li NMR quadrupolar observables, accounting for motional narrowing.
• Extracted correlation times for Li ion motion in different LGPS crystal structures using inverse Laplace transform on simulated SAE data.

Conclusions:

• The ML-assisted workflow effectively overcomes limitations in interpreting complex solid-state NMR data for battery materials.
• Accurate prediction of NMR observables and ion dynamics is achievable, enhancing understanding of lithium-ion diffusion.
• The developed methodology shows promise for differentiating inter-grain vs. intra-grain ion dynamics in solid electrolytes.