Automated discovery of Hamiltonian models for quantum spin systems

With the rise of quantum technologies such as sensing and computation, the accurate characterization of quantum many-body systems is important for future developments. In this work, we introduce a machine learning framework that integrates discrete and continuous optimization to efficiently explore the space of model Hamiltonians that describe quantum many-body systems. Our discovery algorithm represents each multi-spin system as a graph, with nodes to describe the individual spins and edges for the available couplings. Our algorithm performs an efficient exhaustive search and identifies the minimal resources required to reproduce a given set of observations. This work highlights the promise of artificial scientific discovery in understanding quantum many-body systems.