Accelerating NMR Shielding Calculations Through Machine Learning Methods: Application to Magnesium Sodium Silicate Glasses | JoVE Visualize

Area of Science:

Materials Science
Computational Chemistry
Solid-State NMR Spectroscopy

Background:

Nuclear Magnetic Resonance (NMR) spectroscopy is crucial for characterizing materials.
Predicting NMR parameters like isotropic magnetic shielding (σ_iso) for complex systems is computationally intensive.
Silicate glasses containing magnesium and sodium are important industrial materials.

Purpose of the Study:

To develop a machine learning (ML) model for predicting NMR isotropic magnetic shielding (σ_iso) in (Mg, Na)-silicate glasses.
To assess the impact of training dataset size and diversity on ML model performance and transferability.
To demonstrate the utility of the ML model in simulating ²⁹Si Magic Angle Spinning (MAS) NMR spectra for large systems.

Main Methods:

Kernel Ridge Regression (KRR) with a Least Square Support Vector Regression (LSSVR) approach was employed.
The ML model was trained using the Smooth Overlap of Atomic Position (SOAP) descriptor to represent atomic environments.
Isotropic chemical shielding values were computed using Density Functional Theory (DFT) with the Gauge-Included-Projector-Augmented-Wave (GIPAW) method.
Training datasets were generated using molecular dynamics (MD) simulations at various temperatures and inter-atomic potentials.

Main Results:

The ML model successfully predicted NMR isotropic magnetic shielding for ¹⁷O, ²³Na, ²⁵Mg, and ²⁹Si nuclei.
A wide exploration of the configurational space during training significantly improved the ML regressor's transferability.
Simulations of ²⁹Si MAS NMR spectra for systems up to 20,000 atoms were achieved by averaging hundreds of MD configurations.
The ML approach provided results at a fraction of the computational cost of traditional quantum mechanical calculations.

Conclusions:

Machine learning, specifically KRR with LSSVR, is a powerful tool for predicting NMR parameters in complex materials.
The developed ML model enables efficient simulation and interpretation of NMR spectra for large silicate glass systems.
This method significantly reduces computational time, making it valuable for materials characterization and discovery.