Neural Network Ansätze for Infinite Matter

Artificial neural networks have shown tremendous promise as a flexible ansatz for quantum many-body problems. In this work, we approximately solve the Schrödinger equation by performing variational Monte Carlo calculations with a deep, permutation-invariant neural network as a Jastrow correlator. We discuss the reinforcement learning scheme and the stochastic reconfiguration algorithm which helps stabilize the optimization of the wave function parameters. Ground state energies for the three-dimensional electron gas and infinite neutron matter will be compared to standard variational and diffusion Monte Carlo results.