Hamiltonian learning of triplon excitations in an artificial nanoscale molecular quantum magnet

Abstract: Extracting the Hamiltonian of nanoscale quantum magnets from experimental measurements is a significant challenge in correlated quantum matter. Here we put forward a machine learning strategy to extract the spin Hamiltonian from inelastic spectroscopy with scanning tunneling microscopy, and we demonstrate this methodology experimentally with an artificial nanoscale molecular magnet based on cobalt phthalocyanine (CoPC) molecules on NbSe$_2$. We show that this technique allows to directly extract the Hamiltonian parameters of a quantum magnet from the differential conductance, including the substrate-induced spatial variation of the different exchange couplings. Our methodology leverages a machine learning algorithm trained on exact quantum many-body simulations with tensor networks of finite quantum magnets, leading to a methodology that can be used to predict the Hamiltonian of CoPC quantum magnets of arbitrary size. Our results demonstrate how quantum many-body methods and machine learning enable learning a microscopic description of nanoscale quantum many-body systems with scanning tunneling spectroscopy.