Quantum states from normalizing flows

We introduce an architecture for neural quantum states for many-body quantum-mechanical systems, based on normalizing flows. The use of normalizing flows enables efficient uncorrelated sampling of configurations from the probability distribution defined by the wavefunction, mitigating a major cost of using neural states in simulation. We demonstrate the use of this architecture for both ground-state preparation (for self-interacting particles in a harmonic trap) and real-time evolution (for one-dimensional tunneling). Finally, we detail a procedure for obtaining rigorous estimates of the systematic error when using neural states to approximate quantum evolution.