Learning Hamiltonians for solid-state quantum simulators
A preprint introduces an unsupervised autoencoder method for extracting effective Hamiltonians directly from experimental data taken on solid-state quantum simulators. The decoder is constrained by scattering-matrix physics, so the inferred model parameters are meant to represent physically meaningful Hamiltonian terms rather than arbitrary features. The authors present the approach as a generalizable framework for Hamiltonian identification in these systems.
If the physics-constrained autoencoder performs well on noisy transport or spectroscopy data, spin-qubit groups could replace manual Hamiltonian characterization with a single learned calibration pass within two years.