Hydrodynamic Backflow for Easing the Fermion Sign in Finite-Temperature Electron Path Integral Simulations

Abstract: Some notable systems, such as room-temperature superconductors and materials for controlled nuclear fusion, require an accurate description of finite-temperature quantum matter. Stochastic path integral methods are finite-temperature and numerically exact, but scale poorly with system size due the notorious Fermion sign problem. To somewhat mitigate this, we use a hydrodynamical backflow coordinate transformation. Our first attempt was a continuous normalizing flow machine learning approach to determine the optimal parameters. We found this to reduce the error of the total energy, approximately, three times at medium sign severity. Numerical issues challenged training effectively. Thus, a semi-analytic approach was developed to estimate the optimal parameters. We do this by using a derived expression dependent on a Bosonic observable. Hence, the calculation of these values does not have a sign problem. The resulting backflow transformations reduce the problem by multiple orders of magnitude, specifically, in the case of a harmonically trapped, two-dimensional electron gas at finite-temperature. The total energy of the system agrees with previous, backflow untransformed, studies and we calculate energies for up to 32 electrons. The limiting factor is found to be, primarily, the $O(N^3)$ calculation of the Jacobian, stemming from the coordinate transformation of the backflow. A more thorough implementation may further improve this scaling. Otherwise, a pathway for simulating electron systems at currently unreachable regimes is obtained. Finally, as a specific practical use case in energy storage systems, the quantum capacitance for graphene quantum dot materials is calculated.

• The paper demonstrates that hydrodynamic backflow transformations substantially mitigate the Fermion sign problem in finite-temperature electron PIMD simulations.
• It employs neural ODE-based normalizing flows and semi-analytical optimization to align nodal surfaces closer to true Fermionic nodes.
• The methods achieve significant sign improvement (e.g., average sign rising from 0.2 to 0.5) and reliable energy calculations up to 32 electrons.

Hydrodynamic Backflow for Easing the Fermion Sign in Finite-Temperature Electron Path Integral Simulations

Overview

The study presents an investigation into the mitigation of the notorious Fermion sign problem in finite-temperature path-integral molecular dynamics (PIMD) simulations of electron systems. The approach leverages hydrodynamic backflow coordinate transformations, optimized both via ML and a robust semi-analytic framework, to deform the configuration space such that nodal surfaces more closely align with the true Fermionic nodes. The authors focus on the realistic context of strongly-correlated, finite-temperature electron gases in harmonic traps, delivering not only methodological advances, but also specific application to the quantum capacitance of graphene quantum dots.

Theoretical and Computational Methods

The Fermion sign problem arises from the antisymmetry of the many-fermion wavefunction, producing rapidly oscillating signs in path integral estimators and leading to exponential variance growth with system size, temperature, or interaction strength. Traditional approaches, such as extrapolation from bosonic behavior or restricted/path-averaged nodes, only partially ameliorate this scaling catastrophe.

The key innovation is the application of a hydrodynamic backflow transformation to the electron coordinates:

$$\tilde{\mathbf{r}}(\mathbf{r}) = \mathbf{r} + \sum_{\mathbf{r}_i \neq \mathbf{r}} A \frac{\mathbf{r} - \mathbf{r}_i}{1 + \left(|\mathbf{r} - \mathbf{r}_i|/l\right)^3}$$

Here, $A$ controls the backflow strength and $l$ its range, inducing collective coordinate displacements reflecting many-electron correlations. Two strategies are adopted for choosing the optimal parameters:

• Neural ODE-based Normalizing Flow Optimization: Extending continuous normalizing flow methods, the authors train neural networks to maximize the average sign $\langle \sigma \rangle$ in transformed configuration space, using stochastic samples from PIMD as input data. Although nontrivial to stabilize, this method demonstrates a strong sign improvement at moderate severity.
• Semi-Analytical Parameter Optimization: By deriving a closed-form, first-order perturbative estimate for the effect of the backflow transformation on the average sign (Eq. 24), a gradient-based grid search efficiently locates optimal backflow parameters. Importantly, this evaluation does not suffer from the sign problem itself, as it relies on bosonic (sign-free) observables.

Both approaches are benchmarked within a two-dimensional, harmonically trapped electron gas, simulated at finite temperature, with close comparison to established path-integral results.

Numerical Results

Sign Problem Suppression

Application of the optimal backflow transformation results in a substantial enhancement of the average sign. In the reference four-particle case, transformation increases the average sign from 0.2 (untransformed PIMD) to approximately 0.5 with a single learned backflow. Optimization at more severe sign-problem regimes (higher electron number, lower temperature) confirms that the exponential decay in $\langle \sigma \rangle$ is dramatically reduced, achieving average sign values as high as $0.07 \pm 0.01$ even at 16 electrons—a many-fold improvement over the untransformed baseline, where simulations become intractable beyond $\sim$10 electrons.

Physical Observables and Phase Structure

Crucially, the total energy computed with the optimal backflow matches previous high-accuracy results. Energies are extracted reliably up to 32 electrons, facilitating access to regimes corresponding to Wigner crystallization. The Wigner transition is evidenced by a discernible 'bump' in $\langle \sigma \rangle$ and energy curves at $N \approx 16$, in quantitative correspondence with the Lindemann melting criterion for two-dimensional trapped Coulomb systems.

Further exploration varying the interaction range ($|\mathbf{r}_i - \mathbf{r}_j|^\alpha$) elucidates that backflow transformations yield substantial improvements only when the sign problem is significant; for nearly bosonic limits (weak sign problem), backflow introduces unnecessary complexity. Similarly, temperature dependence studies reveal that maximal efficacy is achieved in intermediate to low temperature regimes.

Computational Considerations

The computational bottleneck is identified as the $A$0 cost associated with the evaluation of the Jacobian determinant for the transformed coordinates—a requirement for proper reweighting in the Metropolis-Hastings accept/reject step. While polynomial and not exponential, this scaling limits practical system sizes. The authors note avenues for algorithmic improvement based on advanced techniques in automatic differentiation and Jacobian estimation.

Application to Quantum Capacitance of Graphene Quantum Dots

As a demonstration of practical relevance, the methods are applied to calculate the quantum capacitance of graphene quantum dots, motivated by their promise as high-capacitance electrodes in energy storage devices. The quantum dot system is accurately modeled via the two-dimensional harmonic oscillator with realistic parameters, and the capacitance is obtained by differentiating the total energy with respect to electron number. The computed quantum capacitance, $A$1F/cm$A$2, exceeds the experimentally-measured total capacitance values for graphene-based devices. This suggests that the quantum contribution is not a limiting factor, and strategies for optimizing total device capacitance should address electrolyte screening and materials chemistry.

Implications and Future Directions

This work underlines the practical effectiveness of hydrodynamic backflow transformations in alleviating the Fermion sign problem for finite-temperature electron simulations—substantially extending the tractable system size for ab initio calculations. By connecting modern ML-based normalizing flows with analytic perturbative insights, the authors pave the way for more stable, interpretable coordinate transformations in stochastic quantum simulations.

On a theoretical level, the results reinforce the perspective that optimal nodal deformations—here, via physically motivated backflow—can convert exponential sign decay into a regime where polynomial scaling dominates, provided the transformation is sufficiently flexible and efficiently optimizable.

Future developments are likely to focus on improving the computational scaling of Jacobian evaluation, potentially through sparse-matrix methods or further ML-driven approximations. Extending the approach to higher dimensions, multiband systems, and real materials beyond idealized traps is a natural direction of inquiry. In parallel, deeper understanding of the physical structure of optimal backflow transformations—particularly with respect to phase transitions or emergent correlations—may elucidate the topology of Fermionic nodal surfaces in complex quantum systems.

Conclusion

By integrating hydrodynamic backflow transformations with data-driven and analytic optimization techniques, this work demonstrates effective mitigation of the Fermion sign problem in finite-temperature path integral electron simulations. The approach enables quantitative access to larger system sizes and lower temperature regimes than previously feasible with stochastic Monte Carlo methods, and is applicable to technologically relevant observables in energy materials. The central limitation remains computational, rather than fundamental, and ongoing research in both algorithmic efficiency and the identification of optimal coordinate transformations continues to push the boundaries of ab initio quantum simulation.