Re: Interesting, but has one major flaw (Score 5, Informative)

Author of the paper here. The method actually is variational, after all itâ(TM)s just a variant of variational quantum Monte Carlo with neural networks. And it is perhaps not systematically improvable in the way, say, the coupled cluster theory is, but it is certainly improvable. See our preprint for a taste of that, https://arxiv.org/abs/2010.053..., or have a look at the FermiNet ansatz by Deepmind, which achieves even higher accuracy at significantly larger computational cost.