Convergence of variational Monte Carlo simulation and scale-invariant pre-training

Nilin Abrahamsen, Zhiyan Ding, Gil Goldshlager, Lin Lin

Abstract: Highlights•The variational Monte Carlo (VMC) algorithm is an important and established method in ab-initio quantum chemistry.•We provide the first proof (concurrently with https://arxiv.org/pdf/2303.10599.pdf) of convergence for the VMC algorithm.•We show empirically that pre-training for the VMC algorithm can be accelerated using a scale-invariant loss function.•We base our analysis on the observation that the finite-sample gradient estimator in VMC is directionally unbiased.