Scanning tunnelling microscopes can probe the spin and orbital excitations of individual magnetic molecules on surfaces, but working out the underlying quantum Hamiltonian from the measured spectra is difficult. Researchers at Aalto University have introduced 'molecular Hamiltonian learning', a machine-learning method that reconstructs these parameters from how the spectra change as the tip–sample distance is varied.

Trained solely on theoretical many-body simulations, the algorithm extracts Hamiltonian parameters directly from experimental spectra of iron phthalocyanine molecules on ferroelectric SnTe, turning routine setpoint-dependent spectroscopy into a quantitative tool. The authors are Greta Lupi, Adolfo Fumega and Jose Lado of QMAT's Correlated Quantum Materials group, together with Mohammad Amini, Robert Drost and Peter Liljeroth of Aalto's Atomic Scale Physics group.

The preprint has not yet been peer reviewed.

Image: QMAT