Models of low-dimensional magnets often assume that the coupling between neighbouring spins is the same everywhere, although structural and environmental imperfections can change it locally. An international team has used a local, length-independent machine-learning method to reconstruct spatially varying exchange interactions directly from inelastic scanning tunnelling spectroscopy maps.

Applied to nanographene spin chains, the method identified both nearly uniform and inhomogeneous chains. The reconstructed models reproduce the measured spectra and the correct scaling of the excitation gap with chain length. From QMAT, Greta Lupi and Jose Lado of the Correlated Quantum Materials group at Aalto University took part, with Empa, the University of Bern, TU Dresden and the Max Planck Institute of Microstructure Physics.

The preprint has not yet been peer reviewed.

Image: QMAT