First-principles NMR of oxide glasses boosted by machine learning

Area of Science:

• Materials Science
• Spectroscopy
• Computational Chemistry

Background:

• Solid-state Nuclear Magnetic Resonance (NMR) spectroscopy is crucial for oxide glass structure elucidation.
• First-principles calculations and molecular-dynamics (MD) simulations aid NMR data interpretation but face computational challenges (size, time, resources).
• Accurate simulation of NMR spectra requires efficient methods to handle large systems and finite-temperature effects.

Purpose of the Study:

• To develop a machine-learning (ML) framework to enhance the predictive modeling of NMR spectra.
• To overcome the limitations of computational cost and system size in traditional NMR data interpretation.
• To enable efficient simulation of NMR spectra for large models and incorporate vibrational effects.

Main Methods:

• Utilized kernel ridge regression (least-squares support vector regression and linear ridge regression).
• Employed smooth overlap of atomic position (SOAP) atom-centered descriptors to predict NMR interactions (isotropic magnetic shielding and electric field gradient tensor).
• Applied the ML framework to simulate magic-angle spinning (MAS) and multiple-quantum magic-angle spinning (MQMAS) NMR spectra for large models (>10,000 atoms) and averaged NMR properties over MD trajectories.

Main Results:

• Achieved accurate prediction of NMR parameters (isotropic chemical shift and electric field gradient) with 1-2% accuracy.
• Enabled simulation of NMR spectra for very large models, significantly expanding the scope of analysis.
• Incorporated finite-temperature effects by efficiently averaging NMR properties over nanosecond MD trajectories.
• Proposed a method to scale the electric field gradient tensor using time auto-correlation functions for vibrational effects.

Conclusions:

• The developed ML framework significantly boosts the predictive modeling of NMR spectra for oxide glasses.
• This approach overcomes computational limitations, allowing for the analysis of larger systems and inclusion of temperature effects.
• The method provides a powerful and efficient tool for interpreting solid-state NMR data in materials science.