Physics-informed deep learning enables fast ultrahigh-resolution nuclear magnetic resonance spectroscopy

Area of Science:

• Nuclear Magnetic Resonance (NMR) Spectroscopy
• Artificial Intelligence (AI) in Chemistry
• Molecular Spectroscopy

Background:

• Pure shift NMR spectroscopy offers ultrahigh-resolution molecular insights but suffers from lengthy data acquisition times.
• The additional time dimension required for pure shift evolution limits its practical applications.

Purpose of the Study:

• To develop a fast and robust AI-assisted methodology for pure shift NMR spectroscopy.
• To overcome the limitations of slow data acquisition in pure shift NMR.

Main Methods:

• Implementation of a novel AI-assisted NMR protocol combining non-uniform chunk sampling with physics-informed deep learning (DL) reconstruction.
• Utilizing DL for artifact suppression, weak signal recovery, and peak intensity fidelity in sparse sampled spectra.

Main Results:

• The proposed DL protocol successfully accelerates pure shift NMR acquisition while maintaining spectral quality.
• Demonstrated broad applicability across 1D, 2D, and multi-dimensional pure shift NMR experiments.
• Successful application in in-situ monitoring of 1-butanol electrooxidation.

Conclusions:

• Established a robust AI-assisted NMR framework for rapid molecular structure and dynamics analysis in complex systems.
• The methodology enables high temporal and spectral resolution, with potential applications across diverse chemistry disciplines.