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:2211.09584

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

Marcel Rodekamp, Christoph Gäntgen

Published Nov 17, 2022• 0 citations
arXivHTMLPDF

Abstract

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.

DOI
10.48550/arxiv.2211.09584

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•1 citation

Machine learning to alleviate Hubbard-model sign problems

Jan-Lukas Wynen, Evan Berkowitz, Stefan Krieg +2 more

Lattice Monte Carlo calculations of interacting systems on nonbipartite lattices exhibit an oscillatory imaginary phase known as the phase or sign problem, even at zero chemical potential. One method to alleviate the sign problem is to analytically continue the integration region of the state variables into the complex plane via holomorphic flow equations. For asymptotically large flow times, the state variables approach manifolds of constant imaginary phase known as Lefschetz thimbles. However, flowing such variables and calculating the ensuing Jacobian is a computationally demanding procedure. In this paper, we demonstrate that neural networks can be trained to parametrize suitable manifolds for this class of sign problem and drastically reduce the computational cost for different severely afflicted small volume systems. In particular, we apply our method to the Hubbard model on the triangle and tetrahedron, both of which are nonbipartite. At strong interaction strengths and modest temperatures, the tetrahedron suffers from a severe sign problem that cannot be overcome with standard reweighting techniques, while it quickly yields to our method. We benchmark our results with exact calculations and comment on future directions of this work.

arXiv:2006.11221•Physical review. B./Physical review. B•Mar 25, 2021•29 citations

Mitigating the Hubbard sign problem with complex-valued neural networks

Marcel Rodekamp, Evan Berkowitz, Christoph Gäntgen +3 more

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 an 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.

arXiv:2203.00390•Physical review. B./Physical review. B•Sep 22, 2022•18 citations

Reducing the Sign Problem in Auxiliary Field Monte Carlo Simulations

Gäntgen, Christoph

The numerical sign problem is a hindrance for various areas of research that rely on stochastic sampling methods. It occurs when the sign or phase of the weight function in a path integral is not positive-definite. This leads to cancellations in the Monte Carlo sampling, which can significantly slow down convergence and render the simulation infeasible. Calculations of the Hubbard model are affected by the sign problem when the chemical potential is non-zero or when the spatial lattice is not bipartite. The overall objective of this thesis is the reduction of the sign problem in the Hubbard model with methods that may be applied to other models as well. The approach is to deform the integration contour into complex space and find a transformation that efficiently improves the distribution of the complex phase. The first publication of this cumulative thesis explores the simplest case of such a transformation, which is a constant shift of the real plane in imaginary direction. This transformation showed surprising potential and had the numerical advantage of not requiring the calculation of a Jacobian determinant, as opposed to general transformations. However, it was not obvious which offset would yield the best results. We present an algorithm that numerically approaches the optimal offset iteratively. The drawback of this method is that it relies on Markov Chain Monte Carlo sampling, which comes at significant computational cost and is affected by the sign problem itself, leading to reduced reliability. Therefore, we also present two analytic approximations that can be used directly or as a starting point for the iterative algorithm. To test the method, we calculate single particle correlation functions and the charge expectation value of carbon nanostructures, such as graphene sheets and fullerenes. The second publication introduces a refinement of the iterative optimization algorithm, increasing the stability and convergence speed. This enhancement is achieved by fitting a Gaussian curve to the average phase and its first two derivatives. We again present the physical observables, now including a tetrahedron and a larger honeycomb lattice. Additionally we extract single particle energies from the correlation functions, achieving a good agreement with the exactly solvable tetrahedron. The third publication applies a group equivariant neural network as transformation. The goal was to surpass the limitations of the constant shift while keeping the model as efficient as possible. For this purpose, we encoded the physical symmetries into the model architecture. Results show the potential to reduce the sign problem further than the constant shift. A comparison with dense neural networks shows more efficient use of parameters and training data. Tests with transfer learning lead to mixed results, indicating situational applicability. Overall these developments allow for more efficient calculations of sign problem affected systems.

•Sep 29, 2025•0 citations

Path integral representation for the Hubbard model with reduced number of Lefschetz thimbles

Maksim Ulybyshev, S. N. Valgushev

The concept of Lefschetz thimble decomposition is one of the most promising possible modifications of Quantum Monte Carlo (QMC) algorithms aimed at alleviating the sign problem which appears in many interesting physical situations, e.g. in the Hubbard model away from half filling. In this approach one utilizes the fact that the integral over real variables with an integrand containing a complex fluctuating phase is equivalent to the sum of integrals over special manifolds in complex space ("Lefschetz thimbles"), each of them having a fixed complex phase factor. Thus, the sign problem can be reduced if the resulting sum contains terms with only a few different phases. We explore the complexity of the sign problem for the few-site Hubbard model on square lattice combining a semi-analytical study of saddle points and thimbles in small lattices with several steps in Euclidean time with results of test QMC calculations. We check different variants of conventional Hubbard-Stratonovich transformation based on Gaussian integrals. On the basis of our analysis we reveal a regime with minimal number of relevant thimbles in the vicinity of half filling. In this regime we found only two relevant thimbles for the few-site lattices studied in the paper. There is also indirect evidence of the existence of this regime in more realistic systems with large number of Euclidean time slices. In addition, we derive a new non-Gaussian representation of the interaction term, where the number of relevant Lefschetz is also reduced in comparison with conventional Gaussian Hubbard-Stratonovich transformation.

arXiv:1712.02188•Dec 6, 2017•11 citations

Lefschetz Thimbles and Wigner Functions

Vladimir S. Filinov, Alexander S. Larkin

The major difficulty for the Feynman Path Integral Monte Carlo (PIMC) simulations of the quantum systems of particles is the so called “sign problem”, arising due to the fast oscillations of the path integral integrand depending on the complex-valued action. Our aim is to find universal techniques being able to solve this problem. The new method combines the basic ideas of the Metropolis and Hasting algorithms and is based on the Picard-Lefschetz theory and complex-valued version of Morse theory. The basic idea is to choose the Lefschetz thimbles as manifolds approaching the saddle point of the integrand. On this thimble the imaginary part of the complex-valued action remains constant. As a result the integrand on each thimble does not oscillate, so the “sign problem” disappears and the integral can be calculated much more effectively. The developed approach allows also finding saddle points in the complexified space of path integral integration. Some simple test calculations and comparisons with available analytical results have been carried out.

Journal of Applied Mathematics and Physics•Jan 1, 2020•1 citation
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.
Lattice Monte Carlo calculations of interacting systems on nonbipartite lattices exhibit an oscillatory imaginary phase known as the phase or sign problem, even at zero chemical potential. One method to alleviate the sign problem is to analytically continue the integration region of the state variables into the complex plane via holomorphic flow equations. For asymptotically large flow times, the state variables approach manifolds of constant imaginary phase known as Lefschetz thimbles. However, flowing such variables and calculating the ensuing Jacobian is a computationally demanding procedure. In this paper, we demonstrate that neural networks can be trained to parametrize suitable manifolds for this class of sign problem and drastically reduce the computational cost for different severely afflicted small volume systems. In particular, we apply our method to the Hubbard model on the triangle and tetrahedron, both of which are nonbipartite. At strong interaction strengths and modest temperatures, the tetrahedron suffers from a severe sign problem that cannot be overcome with standard reweighting techniques, while it quickly yields to our method. We benchmark our results with exact calculations and comment on future directions of this work.
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 an 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.
The numerical sign problem is a hindrance for various areas of research that rely on stochastic sampling methods. It occurs when the sign or phase of the weight function in a path integral is not positive-definite. This leads to cancellations in the Monte Carlo sampling, which can significantly slow down convergence and render the simulation infeasible. Calculations of the Hubbard model are affected by the sign problem when the chemical potential is non-zero or when the spatial lattice is not bipartite. The overall objective of this thesis is the reduction of the sign problem in the Hubbard model with methods that may be applied to other models as well. The approach is to deform the integration contour into complex space and find a transformation that efficiently improves the distribution of the complex phase. The first publication of this cumulative thesis explores the simplest case of such a transformation, which is a constant shift of the real plane in imaginary direction. This transformation showed surprising potential and had the numerical advantage of not requiring the calculation of a Jacobian determinant, as opposed to general transformations. However, it was not obvious which offset would yield the best results. We present an algorithm that numerically approaches the optimal offset iteratively. The drawback of this method is that it relies on Markov Chain Monte Carlo sampling, which comes at significant computational cost and is affected by the sign problem itself, leading to reduced reliability. Therefore, we also present two analytic approximations that can be used directly or as a starting point for the iterative algorithm. To test the method, we calculate single particle correlation functions and the charge expectation value of carbon nanostructures, such as graphene sheets and fullerenes. The second publication introduces a refinement of the iterative optimization algorithm, increasing the stability and convergence speed. This enhancement is achieved by fitting a Gaussian curve to the average phase and its first two derivatives. We again present the physical observables, now including a tetrahedron and a larger honeycomb lattice. Additionally we extract single particle energies from the correlation functions, achieving a good agreement with the exactly solvable tetrahedron. The third publication applies a group equivariant neural network as transformation. The goal was to surpass the limitations of the constant shift while keeping the model as efficient as possible. For this purpose, we encoded the physical symmetries into the model architecture. Results show the potential to reduce the sign problem further than the constant shift. A comparison with dense neural networks shows more efficient use of parameters and training data. Tests with transfer learning lead to mixed results, indicating situational applicability. Overall these developments allow for more efficient calculations of sign problem affected systems.
The concept of Lefschetz thimble decomposition is one of the most promising possible modifications of Quantum Monte Carlo (QMC) algorithms aimed at alleviating the sign problem which appears in many interesting physical situations, e.g. in the Hubbard model away from half filling. In this approach one utilizes the fact that the integral over real variables with an integrand containing a complex fluctuating phase is equivalent to the sum of integrals over special manifolds in complex space ("Lefschetz thimbles"), each of them having a fixed complex phase factor. Thus, the sign problem can be reduced if the resulting sum contains terms with only a few different phases. We explore the complexity of the sign problem for the few-site Hubbard model on square lattice combining a semi-analytical study of saddle points and thimbles in small lattices with several steps in Euclidean time with results of test QMC calculations. We check different variants of conventional Hubbard-Stratonovich transformation based on Gaussian integrals. On the basis of our analysis we reveal a regime with minimal number of relevant thimbles in the vicinity of half filling. In this regime we found only two relevant thimbles for the few-site lattices studied in the paper. There is also indirect evidence of the existence of this regime in more realistic systems with large number of Euclidean time slices. In addition, we derive a new non-Gaussian representation of the interaction term, where the number of relevant Lefschetz is also reduced in comparison with conventional Gaussian Hubbard-Stratonovich transformation.
The major difficulty for the Feynman Path Integral Monte Carlo (PIMC) simulations of the quantum systems of particles is the so called “sign problem”, arising due to the fast oscillations of the path integral integrand depending on the complex-valued action. Our aim is to find universal techniques being able to solve this problem. The new method combines the basic ideas of the Metropolis and Hasting algorithms and is based on the Picard-Lefschetz theory and complex-valued version of Morse theory. The basic idea is to choose the Lefschetz thimbles as manifolds approaching the saddle point of the integrand. On this thimble the imaginary part of the complex-valued action remains constant. As a result the integrand on each thimble does not oscillate, so the “sign problem” disappears and the integral can be calculated much more effectively. The developed approach allows also finding saddle points in the complexified space of path integral integration. Some simple test calculations and comparisons with available analytical results have been carried out.