Mitigating the Hubbard Sign Problem with Complex-Valued Neural Networks
Computational Physics
2022-09-30 v2 Strongly Correlated Electrons
High Energy Physics - Lattice
Abstract
Monte Carlo simulations away from half-filling suffer from a sign problem that can be reduced by deforming the contour of integration. Such a transformation, which induces a Jacobian determinant in the Boltzmann weight, can be implemented using neural networks. This additional determinant cost for a generic neural network scales cubically with the volume, preventing large-scale simulations. We implement a new architecture, based on complex-valued affine coupling layers, which reduces this to linear scaling. We demonstrate the efficacy of this method by successfully applying it to systems of different size, the largest of which is intractable by other Monte Carlo methods due to its severe sign problem.
Cite
@article{arxiv.2203.00390,
title = {Mitigating the Hubbard Sign Problem with Complex-Valued Neural Networks},
author = {Marcel Rodekamp and Evan Berkowitz and Christoph Gäntgen and Stefan Krieg and Thomas Luu and Johann Ostmeyer},
journal= {arXiv preprint arXiv:2203.00390},
year = {2022}
}