When the simulation gives up, this spectrometer simply learns from the sample itself

An NMR spectrometer can reveal a molecule's structure atom by atom, but its signal is faint — and for the complex systems researchers most want to study, no computer can calculate in advance the perfect sequence of pulses to coax out the strongest reading. So Aarhus University's team stopped calculating. They wired their spectrometer to a machine-learning algorithm that tries a pulse sequence on the real sample, measures what comes back, learns, and proposes the next try — the instrument and the sample themselves becoming part of the experiment. In the harder of two test systems, the self-taught sequence produced about 70% more signal than the best reference method, in work published in Science Advances. 'What is new is that we let the spectrometer learn for itself how to perform the best experiments,' says first author Filip V. Jensen. Better quantum sensors, food analysis and environmental measurement are among the doors it could open.

SDG 9 · Industry, innovation and infrastructure — Mechanism: tech innovation