Tackling the Sign Problem in the Doped Hubbard Model with Normalizing Flows

summary

The gist

The paper presents a new annealing-based normalizing flow method to reliably sample the Hubbard model in its spin basis at finite chemical potential…

In short

This work introduces a new annealing-based normalizing flow method to sample the Hubbard model in its spin basis at finite chemical potential, overcoming severe sign problems that limit traditional simulations. The method significantly outperforms hybrid Monte Carlo by achieving an order of magnitude better accuracy and uncertainty for system sizes up to Nx=18.

Key concepts

Hubbard Model Sign Problem
The Hubbard model describes doped correlated systems but suffers from a 'sign problem' when simulating at finite chemical potential. This occurs because the probability weight, though real, is not positive definite, making standard sampling methods unreliable.
Sign Quenching
Instead of directly dealing with complex weights, the authors use sign quenching by replacing the Boltzmann weight with its absolute value. This allows them to simulate a theory where weights fluctuate in sign but are treated as real numbers, which is then corrected during reweighting.
Normalizing Flows (NF)
Normalizing flows are generative models that learn a mapping between a simple starting distribution and the complex target distribution of the physical system. They allow for efficient sampling from high-dimensional distributions by learning this bijective transformation.
Annealing Scheme
An annealing scheme gradually deforms the simulation from an easy Gaussian state (at zero chemical potential) to the full interacting Hubbard action. This gradual transition helps the normalizing flow learn signals across all possible states, including complex multimodal structures.

Terminology used across episodes

This episode discusses

The paper

Tackling the Sign Problem in the Doped Hubbard Model with Normalizing Flows · Read on arXiv

Dominic Schuh, Lena Funcke, Janik Kreit, Thomas Luu, Simran Singh

University of Bonn · Helmholtz Institute for Radiation and Nuclear Physics Helmholtz Institute for Radiation and Nuclear Physics Forschungszentrum J¨ulich Institute for Advanced Simulation 4

The Hubbard model at finite chemical potential is a cornerstone for understanding doped correlated systems, but simulations are severely limited by the sign problem. In the auxiliary-field formulation, the spin basis mitigates the sign problem, yet severe ergodicity issues have limited its use. We extend recent advances with normalizing flows at half-filling to finite chemical potential by introducing an annealing scheme enabling ergodic sampling. Compared to state-of-the-art hybrid Monte Carlo in the charge basis, our approach accurately reproduces exact diagonalization results while reducing statistical uncertainties by an order of magnitude, opening a new path for simulations of doped correlated systems.

DOI: 10.1103/93pv-5hlm

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: We're looking at the paper: "Tackling the Sign Problem in the Doped Hubbard Model with Normalizing Flows".

Mira: Here's the gist of the paper. The paper presents a new annealing-based normalizing flow method to reliably sample the Hubbard model in its spin basis at finite chemical potential, which outperforms state-of-the-art hybrid Monte Carlo simulations by up to an order of magnitude in both accuracy and uncertainty for system sizes up to Nx = 18 Model Formulation and Sign Problem Context The Hubbard model at finite density is a cornerstone for understanding doped correlated systems, but simulations are severely limited by the sign problem The model Hamiltonian in the spin basis is given by H = −κ X⟨x,y⟩σ c†x,σcy,σ − U2X x(nx,↑ − nx,↓)2 − µX qx This formulation offers a structural advantage because the fermion matrix is real for each spin species separately The introduction of a finite chemical potential explicitly breaks the symmetry that guarantees exact positivity of the measure at half-filling This leads to a situation where the probability weight is real but not positive definite, which is described as a sign problem Sign Quenching and Normalizing Flows When simulating theories with a complex action, one typically follows one of two strategies: (i) simulate the phase-quenched theory

38: , in which the complex phase of the Boltzmann weight is neglected and subsequently restored by reweighting; or (ii) simulate the sign-quenched theory

39, 40: , where the Boltzmann weight is replaced by the absolute value of its real part, and sign fluctuations are reintroduced through sign reweighting In this work, we employ sign quenching, defining wq = wt for real weights wt that fluctuate in sign Observables in the full theory are recovered via reweighting using the formula ⟨O⟩t = R Dϕ wt(ϕ)O(ϕ) / R Dϕ wt(ϕ) = ⟨wt wq · O⟩q ⟨wt wq⟩q Normalizing flows are a class of generative models that enable efficient sampling from high-dimensional distributions by learning a bijective map fθ between a simple prior distribution qz and a target distribution qy, fθ: z ∼ qz 7→ y ∼ qy Annealing Scheme for Ergodicity To describe systems across a range of couplings, the authors employ an annealing scheme

57: , in which an auxiliary parameter λ scales the fermionic contribution to the action, Sq,λ

ϕ: = X x,t∈Λ ϕ2x,t2U˜ − λ · log det (M

ϕκ, ˜ µ˜: M

-ϕκ, ˜ µ˜: ) For λ = 0, the fermion determinant is absent and the target distribution is Gaussian, while λ = 1 recovers the full interacting Hubbard action with its multimodal structure This gradual deformation mitigates mode-seeking behavior of reverse KL divergence by allowing the NF to receive training signals from all modes Performance and Validation The results show that simulations performed in the spin basis using the NF approach maintain a substantially larger average sign across the region µ ∈

1.0, 2.0: compared to optimized HMC simulations in the charge basis The NF approach reproduces exact results with high precision while reducing uncertainties by an order of magnitude compared to HMC for system sizes up to Nx = 18 In the absence of exact results at this system size, the NF method finds good agreement with HMC and produces smaller uncertainties for noise-sensitive observables The proposed annealing scheme enables controlled sampling of multimodal target distributions without introducing significant training overhead or strong dependence on the system size Future Directions The overall scalability of the method is primarily determined by the properties of the underlying generative model, with training cost expected to increase for larger volumes Building on recent work on improving the numerical stability of action-based Hubbard model simulations, the authors plan to extend their annealing scheme to more expressive and scalable generative models in future work Taken together, these results establish a broadly applicable framework for efficient and accurate sampling of the Hubbard model in regimes where conventional Monte Carlo methods are limited by ergodicity loss or severe sign problems Appendix Details At vanishing chemical potential, the distribution takes the form of a correlated Gaussian, but introducing a chemical potential splits the distribution into two fairly disconnected modes indicating that practical ergodicity issues still persist In Fig.

Paper summary: Kai: So we're looking at how this paper tackles that sign problem in the doped Hubbard model using normalizing flows. The central thesis is that by introducing an annealing scheme to scale the fermionic action, they can enable ergodic sampling of the system even when a finite chemical potential breaks symmetry.

Mira: They are claiming this method reliably samples the Hubbard model in its spin basis at finite chemical potential, and they state it outperforms current state-of-the-art hybrid Monte Carlo simulations by up to an order of magnitude in both accuracy and uncertainty for system sizes up to Nx equals eighteen <ref:2603.18205#pg2>.

Lev: The context is that the Hubbard model at finite density has been limited by the sign problem, even though using the spin basis helps mitigate this somewhat, but severe ergodicity issues still limit its use. This paper extends recent advances with normalizing flows from half-filling to finite chemical potential specifically by introducing an annealing scheme enabling ergodic sampling.

Kai: The key mechanism they use is sign quenching, where they define the weight as wq equals the absolute value of its real part, and then recover physical observables by reweighting them using that quenched weight. They use normalizing flows to learn a map between a simple prior distribution and the complex target distribution.

Mira: What this means for us is that we have a new framework for efficiently sampling these highly correlated systems in regimes where conventional Monte Carlo methods are severely hindered by sign problems or ergodicity loss, opening up a path for studying doped correlated systems.

Lev: It’s interesting from the perspective of error correction researcher because it suggests a way to sample distributions that are inherently multimodal without relying on the simulation getting stuck in one mode. The annealing scheme is specifically designed to handle that multimodal structure by providing signals across all modes during training.

Conclusion: Kai: So we’ve looked at how this paper, "Tackling the Sign Problem in the Doped Hubbard Model with Normalizing Flows," proposes using annealing and normalizing flows to handle sign problems in doped correlated systems. It's a method developed by Schuh, Funcke, Kreit, Luu and Singh.

Mira: The core implication is that this new technique offers a path for accurate sampling of the Hubbard model at finite chemical potential where existing methods are failing due to sign problems or ergodicity loss. It shows that learning complex mappings between simple distributions can overcome some of those fundamental simulation hurdles.

Lev: From an engineering standpoint, it means we have a method that can produce results with much lower statistical uncertainty than hybrid Monte Carlo for specific system sizes, which is valuable for setting benchmarks and understanding what's physically achievable on current computational resources.

Kai: It’s not just about getting one better number; it’s about creating a framework that lets us explore these complex physics regimes more effectively. The authors are pointing toward extending the annealing scheme to more expressive generative models for larger volumes, which is the next logical step in this research.

Mira: Exactly. The paper suggests that instead of just quenching the sign and hoping for the best, we can learn a transformation that guides us through complex probability landscapes where traditional sampling gets stuck. It’s about moving beyond just mitigating problems to actually learning to navigate them.

More episodes

← Home