Aarhus University uses AI to optimize NMR spectrometer pulse sequences

Aarhus University researchers have used AI to allow an NMR spectrometer to learn the best pulse sequence directly from a sample, overcoming the limitations of weak signals in revealing atomic-level molecular structure. The team connected a home-built dynamic nuclear polarization (DNP) spectrometer to a machine-learning algorithm based on Bayesian optimization. This system tests a pulse sequence on the sample, measures the result, and proposes the next experiment, rather than relying on pre-calculated models that may not capture real-world imperfections. Professor Niels Chr. Nielsen noted that NMR is relatively insensitive, requiring high sample concentrations. Ph.D. student Filip V. Jensen, the study’s first author, explained that the innovation lies in letting the spectrometer learn how to perform the best experiments itself.

Editor's note: Clear, technically accurate summary of the research and its methodology.