Computing the ground state of interacting quantum matter, especially for complex two-dimensional systems, is challenging. Neural quantum states (NQSs), encoding many-body wavefunctions in artificial neural networks, offer potential but are limited by existing optimization algorithms unsuitable for large-scale deep networks. This work introduces a minimum-step stochastic-reconfiguration (MinSR) optimization algorithm, enabling the training of deep NQSs with up to 10⁶ parameters. Applied to frustrated spin-1/2 models, MinSR achieves machine precision and improved variational energies, providing numerical evidence for gapless quantum-spin-liquid phases.

Ao Chen, Markus Heyl