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:2605.26814

Neural Autoregressive Control Variates for the Quantum Monte Carlo Sign Problem

Bei Qiao, Lei Wang

Published May 26, 2026• 2 citations
arXivHTMLPDF

Abstract

We train a pair of autoregressive models to construct zero-mean control variates to mitigate the sign problem in quantum Monte Carlo simulations. The two autoregressive networks are confined to the positive- and negative-sign sectors with strictly disjoint support, and each is exactly normalized over its sector. Their difference is therefore structurally zero-mean, providing an unbiased auxiliary observable whose correlation with the sign estimator controls the variance reduction. We implement the method within the stochastic series expansion framework, which we extend to frustrated lattices by developing an incremental loop-topology update. Sign-ergodic sampling is achieved through a twist channel, which is the unique sign-changing mechanism on non-bipartite lattices. We implement the control variates as autoregressive transformers with an end-of-sequence parity mask that enforces exact sign-sector resolution, while the incremental loop-count change and cumulative frustration parity are incorporated as topological features. On the triangular-lattice Heisenberg antiferromagnet, we benchmark the method in the small-$N$ limit. The control variate reduces the standard error of the average sign by up to an order of magnitude and that of the energy estimator by a factor of three to five, remaining effective even when the average sign drops below $10^{-3}$. This work lays out the framework and provides a proof-of-principle demonstration that autoregressive control variates can effectively mitigate the sign problem. Scaling to larger systems with physics-informed architectures is the subject of future work.

DOI
10.48550/arxiv.2605.26814

Related papers

Reduction of the sign problem near T=0 in quantum Monte Carlo simulations

Jonathan D’Emidio, Stefan Weßel, Frédéric Mila

Building on a recent investigation of the Shastry-Sutherland model [S. Wessel et al., Phys. Rev. B 98, 174432 (2018)], we develop a general strategy to eliminate the Monte Carlo sign problem near the zero-temperature limit in frustrated quantum spin models. If the Hamiltonian of interest and the sign-problem-free Hamiltonian, obtained by making all off-diagonal elements negative in a given basis, have the same ground state and this state is a member of the computational basis, then the average sign returns to one as the temperature goes to zero. We illustrate this technique by studying the triangular and kagome lattice Heisenberg antiferrromagnet in a magnetic field above saturation, as well as the Heisenberg antiferromagnet on a modified Husimi cactus in the dimer basis. We also provide detailed Appendices on using linear programming techniques to automatically generate efficient directed loop updates in quantum Monte Carlo simulations.

arXiv:2003.13356•Physical review. B./Physical review. B•Aug 20, 2020•11 citations

Signal-to-noise problems in fault-tolerant quantum computation

Hyunwoo Oh

Quantum computation offers a promising alternative to classical Monte Carlo methods by naturally avoiding sign problems through unitary evolution. A common misconception follows from this: because quantum simulation sidesteps the sign problem, it is often assumed to be an essentially perfect substitute for classical sampling. This is false---quantum measurements are intrinsically stochastic, and estimating observables often requires a number of samples that grows exponentially with system size, time separation, or inverse temperature, exactly as in the classical signal-to-noise problem. In this work, we analyze the origin and structure of signal-to-noise problems in quantum computation, drawing parallels with classical Monte Carlo, and demonstrate that such noise persists even in sign-problem-free regimes. We discuss strategies to mitigate it, in particular a quantum generalization of Stein control variates construction, in which unbiased estimators are built from operator commutators $[X,A]$ and their discrete diagonal counterparts, and validate the approach with exactly solvable numerical examples that quantify the achievable variance reduction. Our results underscore that variance reduction, not merely sign-problem avoidance, is required to make quantum algorithms practically viable for many-body physics.

quant-pharXiv:2610.15210•Oct 13, 2026•0 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

Sign problem in Monte Carlo simulations of frustrated quantum spin systems

Patrik Henelius, Anders W. Sandvik

We discuss the sign problem arising in Monte Carlo simulations of frustrated quantum spin systems. We show that for a class of ``semifrustrated'' systems [Heisenberg models with ferromagnetic couplings ${J}_{z}(r)<0$ along the z axis and antiferromagnetic couplings ${J}_{\mathrm{xy}}(r)=\ensuremath{-}{J}_{z}(r)$ in the $\mathrm{xy}$ plane, for arbitrary distances $r],$ the sign problem present for algorithms operating in the z basis can be solved within a recent ``operator-loop'' formulation of the stochastic series expansion method [a cluster algorithm for sampling the diagonal matrix elements of the power series expansion of $\mathrm{exp}(\ensuremath{-}\ensuremath{\beta}H)$ to all orders]. The solution relies on the identification of operator loops which change the configuration sign when updated (``merons'') and is similar to the meron-cluster algorithm recently proposed by Chandrasekharan and Wiese for solving the sign problem for a class of fermion models [Phys. Rev. Lett. 83, 3116 (1999)]. Some important expectation values, e.g., the internal energy, can be evaluated in the subspace with no merons, where the weight function is positive definite. Calculations of other expectation values require sampling of configurations with only a small number of merons (typically zero or two), with an accompanying sign problem which is not serious. We also discuss problems which arise in applying the meron concept to more general quantum spin models with frustrated interactions.

arXiv:cond-mat/0001351•Physical review. B, Condensed matter•Jul 1, 2000•154 citations

Approaching the Thermodynamic Limit with Neural-Network Quantum States

Luciano Loris Viteritti, Riccardo Rende, Subir Sachdev +1 more

Accessing the thermodynamic-limit properties of strongly correlated quantum matter requires simulations on very large lattices, a regime that remains challenging for numerical methods, especially in frustrated two-dimensional systems. We introduce the Spatial Attention mechanism, a minimal and physically interpretable inductive bias for Neural-Network Quantum States, implemented as a single learned length scale within the Transformer architecture. This bias stabilizes large-scale optimization and enables access to thermodynamic-limit physics through highly accurate simulations on unprecedented system sizes within the Variational Monte Carlo framework. Applied to the spin-$\tfrac12$ triangular-lattice Heisenberg antiferromagnet, our approach achieves state-of-the-art results on clusters of up to $42\times42$ sites. The ability to simulate such large systems allows controlled finite-size scaling of energies and order parameters, enabling the extraction of experimentally relevant quantities such as spin-wave velocities and uniform susceptibilities. In turn, we find extrapolated thermodynamic limit energies systematically better than those obtained with tensor-network approaches such as iPEPS. The resulting magnetization is strongly renormalized, $M_0=0.148(1)$ (about $30\%$ of the classical value), revealing that less accurate variational states systematically overestimate magnetic order. Analysis of the optimized wave function further suggests an intrinsically non-local sign structure, indicating that the sign problem cannot be removed by local basis transformations. We finally demonstrate the generality of the method by obtaining state-of-the-art energies for a $J_1$-$J_2$ Heisenberg model on a $20\times20$ square lattice, outperforming Residual Convolutional Neural Networks.

cond-mat.str-elCross-listedarXiv:2602.02665•Feb 4, 2026•28 citations

Autoregressive neural Slater-Jastrow ansatz for variational Monte Carlo simulation

Stephan Humeniuk, Yuan Wan, Lei Wang

Direct sampling from a Slater determinant is combined with an autoregressive deep neural network as a Jastrow factor into a fully autoregressive Slater-Jastrow ansatz for variational quantum Monte Carlo, which allows for uncorrelated sampling. The elimination of the autocorrelation time leads to a stochastic algorithm with provable cubic scaling (with a potentially large prefactor), i.e. the number of operations for producing an uncorrelated sample and for calculating the local energy scales like \mathcal{O}(N_s^3) 𝒪 ( N s 3 ) with the number of orbitals N_s N s . The implementation is benchmarked on the two-dimensional t-V t − V model of spinless fermions on the square lattice.

arXiv:2210.05871•SciPost Physics•Jun 27, 2023•14 citations
We train a pair of autoregressive models to construct zero-mean control variates to mitigate the sign problem in quantum Monte Carlo simulations. The two autoregressive networks are confined to the positive- and negative-sign sectors with strictly disjoint support, and each is exactly normalized over its sector. Their difference is therefore structurally zero-mean, providing an unbiased auxiliary observable whose correlation with the sign estimator controls the variance reduction. We implement the method within the stochastic series expansion framework, which we extend to frustrated lattices by developing an incremental loop-topology update. Sign-ergodic sampling is achieved through a twist channel, which is the unique sign-changing mechanism on non-bipartite lattices. We implement the control variates as autoregressive transformers with an end-of-sequence parity mask that enforces exact sign-sector resolution, while the incremental loop-count change and cumulative frustration parity are incorporated as topological features. On the triangular-lattice Heisenberg antiferromagnet, we benchmark the method in the small-NNN limit. The control variate reduces the standard error of the average sign by up to an order of magnitude and that of the energy estimator by a factor of three to five, remaining effective even when the average sign drops below 10−310^{-3}10−3. This work lays out the framework and provides a proof-of-principle demonstration that autoregressive control variates can effectively mitigate the sign problem. Scaling to larger systems with physics-informed architectures is the subject of future work.
Building on a recent investigation of the Shastry-Sutherland model [S. Wessel et al., Phys. Rev. B 98, 174432 (2018)], we develop a general strategy to eliminate the Monte Carlo sign problem near the zero-temperature limit in frustrated quantum spin models. If the Hamiltonian of interest and the sign-problem-free Hamiltonian, obtained by making all off-diagonal elements negative in a given basis, have the same ground state and this state is a member of the computational basis, then the average sign returns to one as the temperature goes to zero. We illustrate this technique by studying the triangular and kagome lattice Heisenberg antiferrromagnet in a magnetic field above saturation, as well as the Heisenberg antiferromagnet on a modified Husimi cactus in the dimer basis. We also provide detailed Appendices on using linear programming techniques to automatically generate efficient directed loop updates in quantum Monte Carlo simulations.
Quantum computation offers a promising alternative to classical Monte Carlo methods by naturally avoiding sign problems through unitary evolution. A common misconception follows from this: because quantum simulation sidesteps the sign problem, it is often assumed to be an essentially perfect substitute for classical sampling. This is false---quantum measurements are intrinsically stochastic, and estimating observables often requires a number of samples that grows exponentially with system size, time separation, or inverse temperature, exactly as in the classical signal-to-noise problem. In this work, we analyze the origin and structure of signal-to-noise problems in quantum computation, drawing parallels with classical Monte Carlo, and demonstrate that such noise persists even in sign-problem-free regimes. We discuss strategies to mitigate it, in particular a quantum generalization of Stein control variates construction, in which unbiased estimators are built from operator commutators [X,A][X,A][X,A] and their discrete diagonal counterparts, and validate the approach with exactly solvable numerical examples that quantify the achievable variance reduction. Our results underscore that variance reduction, not merely sign-problem avoidance, is required to make quantum algorithms practically viable for many-body physics.
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.
We discuss the sign problem arising in Monte Carlo simulations of frustrated quantum spin systems. We show that for a class of ``semifrustrated'' systems [Heisenberg models with ferromagnetic couplings Jz(r)<0{J}_{z}(r)<0Jz​(r)<0 along the z axis and antiferromagnetic couplings ${J}_{\mathrm{xy}}(r)=\ensuremath{-}{J}_{z}(r)$ in the xy\mathrm{xy}xy plane, for arbitrary distances r],r],r], the sign problem present for algorithms operating in the z basis can be solved within a recent ``operator-loop'' formulation of the stochastic series expansion method [a cluster algorithm for sampling the diagonal matrix elements of the power series expansion of $\mathrm{exp}(\ensuremath{-}\ensuremath{\beta}H)$ to all orders]. The solution relies on the identification of operator loops which change the configuration sign when updated (``merons'') and is similar to the meron-cluster algorithm recently proposed by Chandrasekharan and Wiese for solving the sign problem for a class of fermion models [Phys. Rev. Lett. 83, 3116 (1999)]. Some important expectation values, e.g., the internal energy, can be evaluated in the subspace with no merons, where the weight function is positive definite. Calculations of other expectation values require sampling of configurations with only a small number of merons (typically zero or two), with an accompanying sign problem which is not serious. We also discuss problems which arise in applying the meron concept to more general quantum spin models with frustrated interactions.
Accessing the thermodynamic-limit properties of strongly correlated quantum matter requires simulations on very large lattices, a regime that remains challenging for numerical methods, especially in frustrated two-dimensional systems. We introduce the Spatial Attention mechanism, a minimal and physically interpretable inductive bias for Neural-Network Quantum States, implemented as a single learned length scale within the Transformer architecture. This bias stabilizes large-scale optimization and enables access to thermodynamic-limit physics through highly accurate simulations on unprecedented system sizes within the Variational Monte Carlo framework. Applied to the spin-12\tfrac1221​ triangular-lattice Heisenberg antiferromagnet, our approach achieves state-of-the-art results on clusters of up to 42×4242\times4242×42 sites. The ability to simulate such large systems allows controlled finite-size scaling of energies and order parameters, enabling the extraction of experimentally relevant quantities such as spin-wave velocities and uniform susceptibilities. In turn, we find extrapolated thermodynamic limit energies systematically better than those obtained with tensor-network approaches such as iPEPS. The resulting magnetization is strongly renormalized, M0=0.148(1)M_0=0.148(1)M0​=0.148(1) (about 30%30\%30% of the classical value), revealing that less accurate variational states systematically overestimate magnetic order. Analysis of the optimized wave function further suggests an intrinsically non-local sign structure, indicating that the sign problem cannot be removed by local basis transformations. We finally demonstrate the generality of the method by obtaining state-of-the-art energies for a J1J_1J1​-J2J_2J2​ Heisenberg model on a 20×2020\times2020×20 square lattice, outperforming Residual Convolutional Neural Networks.
Direct sampling from a Slater determinant is combined with an autoregressive deep neural network as a Jastrow factor into a fully autoregressive Slater-Jastrow ansatz for variational quantum Monte Carlo, which allows for uncorrelated sampling. The elimination of the autocorrelation time leads to a stochastic algorithm with provable cubic scaling (with a potentially large prefactor), i.e. the number of operations for producing an uncorrelated sample and for calculating the local energy scales like \mathcal{O}(N_s^3) 𝒪 ( N s 3 ) with the number of orbitals N_s N s . The implementation is benchmarked on the two-dimensional t-V t − V model of spinless fermions on the square lattice.