Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2311.18312 · International Symposium on Lattice Field Theory (LATTICE)

From Theory to Practice: Applying Neural Networks to Simulate Real Systems with Sign Problems

Marcel Rodekamp, Evan Berkowitz, Maria Dincă, Christoph Gäntgen, Stefan Krieg, Thomas Luu

Published Dec 27, 2023• 3 citations
arXivHTMLPDF

Abstract

The numerical sign problem poses a seemingly insurmountable barrier to the simulation of many fascinating systems. We apply neural networks to deform the region of integration, mitigating the sign problem of systems with strongly correlated electrons. In this talk we present our latest architectural developments as applied to contour deformation. We also demonstrate its applicability to real systems, namely perylene.

DOI
10.22323/1.453.0031

Related papers

Mitigating the Hubbard Sign Problem. A Novel Application of Machine Learning

Marcel Rodekamp, Christoph Gäntgen

Many fascinating systems suffer from a severe (complex action) sign problem preventing us from calculating them with Markov Chain Monte Carlo simulations. One promising method to alleviate the sign problem is the transformation of the integration domain towards Lefschetz Thimbles. Unfortunately, this suffers from poor scaling originating in numerically integrating of flow equations and evaluation of an induced Jacobian. In this proceedings we present a new preliminary Neural Network architecture based on complex-valued affine coupling layers. This network performs such a transformation efficiently, ultimately allowing simulation of systems with a severe sign problem. We test this method within the Hubbard Model at finite chemical potential, modelling strongly correlated electrons on a spatial lattice of ions.

International Symposium on Lattice Field Theory (LATTICE)•Nov 7, 2022•2 citations

Mitigating the Hubbard Sign Problem. A Novel Application of Machine Learning

Marcel Rodekamp, Christoph Gäntgen

Many fascinating systems suffer from a severe (complex action) sign problem preventing us from calculating them with Markov Chain Monte Carlo simulations. One promising method to alleviate the sign problem is the transformation of the integration domain towards Lefschetz Thimbles. Unfortunately, this suffers from poor scaling originating in numerically integrating of flow equations and evaluation of an induced Jacobian. In this proceedings we present a new preliminary Neural Network architecture based on complex-valued affine coupling layers. This network performs such a transformation efficiently, ultimately allowing simulation of systems with a severe sign problem. We test this method within the Hubbard Model at finite chemical potential, modelling strongly correlated electrons on a spatial lattice of ions.

arXiv:2211.09584•Nov 17, 2022•1 citation

Exploring Group Convolutional Networks for Sign Problem Mitigation via Contour Deformation

Christoph Gäntgen, Thomas Luu, Marcel Rodekamp

The sign problem that arises in Hybrid Monte Carlo calculations can be mitigated by deforming the integration manifold. While simple transformations are highly efficient for simulation, their efficacy systematically decreases with decreasing temperature and increasing interaction. Machine learning models have demonstrated the ability to push further, but require additional computational effort and upfront training. While neural networks possess the capacity to learn physical symmetries through proper training, there are anticipated advantages associated with encoding them into the network’s structure. These include enhanced accuracy, accelerated training, and improved stability. The objective of the present study is twofold. First, we investigate the benefits of group convolutional models in comparison to fully connected networks, with a specific focus on the effects on the sign problem and on computational aspects. Second, we examine their capabilities for transfer learning, demonstrating the ability to further reduce training cost. We perform our investigations on the Hubbard model on select low-dimensional systems.

arXiv:2502.04104•International Symposium on Lattice Field Theory (LATTICE)•Feb 5, 2025•0 citations

Neural network wave functions and the sign problem

Attila Szabó, Claudio Castelnovo

This paper proposes a neural network architecture and learning protocol to solve the problem of convergence to ground states with a nontrivial sign structure. The authors apply their scheme to conventional antiferromagnets and show that, while it works well on those systems, it is unable to find ground states on frustrated magnets, finding instead low-energy states that exhibit the unfrustrated Marshall sign rule

arXiv:2002.04613•Physical Review Research•Jul 15, 2020•129 citations

Fermionic sign problem minimization by constant path integral contour shifts

Christoph Gäntgen, Evan Berkowitz, Thomas Luu +2 more

The path integral formulation of quantum mechanical problems including fermions is often affected by a severe numerical sign problem. We show how such a sign problem can be alleviated by a judiciously chosen constant imaginary offset to the path integral. Such integration contour deformations introduce no additional computational cost to the Hamiltonian Monte Carlo algorithm, while its effective sample size is greatly increased. This makes otherwise unviable simulations efficient for a wide range of parameters. Applying our method to the Hubbard model, we find that the sign problem is significantly reduced. Furthermore, we prove that it vanishes completely for large chemical potentials, a regime where the sign problem is expected to be particularly severe without imaginary offsets. In addition to a numerical analysis of such optimized contour shifts, we analytically compute the shifts corresponding to the leading and next-to-leading order corrections to the action. We find that such simple approximations, free of significant computational cost, suffice in many cases. We present a simulation of ${\mathrm{C}}_{60}$ fullerenes (buckyballs) that are successful over a wide parameter range.

arXiv:2307.06785•Physical review. B./Physical review. B•May 20, 2024•12 citations

Contour deformations for nonholomorphic actions

Scott Lawrence, S. N. Valgushev, Jianan Xiao +1 more

We show how contour deformations may be used to control the sign problem of lattice Monte Carlo calculations with nonholomorphic Boltzmann factors. Such actions arise naturally in quantum mechanical scattering problems. The approach is demonstrated in conjunction with the holomorphic gradient flow. As our central example we compute the real-time evolution of a particle in a one-dimensional analog of the Yukawa potential. Published by the American Physical Society 2024

arXiv:2401.16733•Physical review. D/Physical review. D.•Oct 18, 2024•3 citations
The numerical sign problem poses a seemingly insurmountable barrier to the simulation of many fascinating systems. We apply neural networks to deform the region of integration, mitigating the sign problem of systems with strongly correlated electrons. In this talk we present our latest architectural developments as applied to contour deformation. We also demonstrate its applicability to real systems, namely perylene.
Many fascinating systems suffer from a severe (complex action) sign problem preventing us from calculating them with Markov Chain Monte Carlo simulations. One promising method to alleviate the sign problem is the transformation of the integration domain towards Lefschetz Thimbles. Unfortunately, this suffers from poor scaling originating in numerically integrating of flow equations and evaluation of an induced Jacobian. In this proceedings we present a new preliminary Neural Network architecture based on complex-valued affine coupling layers. This network performs such a transformation efficiently, ultimately allowing simulation of systems with a severe sign problem. We test this method within the Hubbard Model at finite chemical potential, modelling strongly correlated electrons on a spatial lattice of ions.
Many fascinating systems suffer from a severe (complex action) sign problem preventing us from calculating them with Markov Chain Monte Carlo simulations. One promising method to alleviate the sign problem is the transformation of the integration domain towards Lefschetz Thimbles. Unfortunately, this suffers from poor scaling originating in numerically integrating of flow equations and evaluation of an induced Jacobian. In this proceedings we present a new preliminary Neural Network architecture based on complex-valued affine coupling layers. This network performs such a transformation efficiently, ultimately allowing simulation of systems with a severe sign problem. We test this method within the Hubbard Model at finite chemical potential, modelling strongly correlated electrons on a spatial lattice of ions.
The sign problem that arises in Hybrid Monte Carlo calculations can be mitigated by deforming the integration manifold. While simple transformations are highly efficient for simulation, their efficacy systematically decreases with decreasing temperature and increasing interaction. Machine learning models have demonstrated the ability to push further, but require additional computational effort and upfront training. While neural networks possess the capacity to learn physical symmetries through proper training, there are anticipated advantages associated with encoding them into the network’s structure. These include enhanced accuracy, accelerated training, and improved stability. The objective of the present study is twofold. First, we investigate the benefits of group convolutional models in comparison to fully connected networks, with a specific focus on the effects on the sign problem and on computational aspects. Second, we examine their capabilities for transfer learning, demonstrating the ability to further reduce training cost. We perform our investigations on the Hubbard model on select low-dimensional systems.
The path integral formulation of quantum mechanical problems including fermions is often affected by a severe numerical sign problem. We show how such a sign problem can be alleviated by a judiciously chosen constant imaginary offset to the path integral. Such integration contour deformations introduce no additional computational cost to the Hamiltonian Monte Carlo algorithm, while its effective sample size is greatly increased. This makes otherwise unviable simulations efficient for a wide range of parameters. Applying our method to the Hubbard model, we find that the sign problem is significantly reduced. Furthermore, we prove that it vanishes completely for large chemical potentials, a regime where the sign problem is expected to be particularly severe without imaginary offsets. In addition to a numerical analysis of such optimized contour shifts, we analytically compute the shifts corresponding to the leading and next-to-leading order corrections to the action. We find that such simple approximations, free of significant computational cost, suffice in many cases. We present a simulation of C60{\mathrm{C}}_{60}C60​ fullerenes (buckyballs) that are successful over a wide parameter range.