Weak-coupling tensor cross interpolation impurity solver for nonequilibrium dynamical mean-field theory

arXiv:2607.00702 · cond-mat.str-el · Submitted 2026-07-01 · Read on arXiv

Listen

Radio episode about this paper

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: "Weak-coupling tensor cross interpolation impurity solver for nonequilibrium dynamical mean-field theory".

Mira: Here's the gist of the paper. Simulating nonequilibrium quantum many-body systems remains a major challenge due to the exponential growth of the computational complexity with real time Page 1.

Kai: Here we implement a nonequilibrium impurity solver based on the weak-coupling expansion and the tensor cross interpolation (TCI), and apply it to nonequilibrium dynamical mean-field theory (DMFT) Page 1.

Mira: The method approximates the integrands of the high-dimensional integrals arising in the weak-coupling expansion in a tensor-train form, enabling efficient evaluations without stochastic sampling and thereby mitigating the sign problem affecting continuous-time quantum Monte Carlo (CT-QMC) methods Page 1.

Kai: The core of the approach involves developing a nonequilibrium impurity solver that combines the weak-coupling expansion with the TCI algorithm to compute the two-time Green’s functions required for nonequilibrium DMFT Page 1.

Paper summary: Kai: So to wrap up on this paper, they developed a nonequilibrium TCI impurity solver based on the weak-coupling expansion and applied it to nonequilibrium DMFT for both transient and steady-state calculations. They showed that this method systematically converges with increasing expansion order and for moderate bond dimensions, indicating that the nonequilibrium integrand admits an efficient low-rank tensor train representation.

Mira: The paper confirms that they reproduced the known half-filled quench dynamics and the rapid thermalization near the dynamical transition point around U equals three times v. Crucially, it showed that away from half filling, where sign problems are more severe for continuous time quantum Monte Carlo methods, this approach remains well controlled.

Lev: The authors conclude that this nonequilibrium TCI impurity solver is a promising route to nonequilibrium DMFT in parameter regimes that are difficult to access with existing approaches. They also point out the method is systematically improvable by increasing the maximum order and allows for reliable results even in the intermediate-to-strong coupling regime, at least for short times.

Kai: It means they've given us a tool that works well where other methods struggle, especially when we look at how dynamics change with doping. We see that the dynamical transition at half filling evolves into a smooth crossover upon doping, which is a qualitative change in the relaxation behavior.

Mira: The main point is that this solver gives us numerically exact evidence of this crossover, and it marks a qualitative shift from smooth relaxation on the weak U side to an atomic-like oscillatory post-quench dynamics on the strong U side.

Lev: The real future direction they are pushing for is compressing those Keldysh indices so you can reach polynomial scaling in time. That’s what makes it useful for actual computation, not just theoretical exploration.

Conclusion: Kai: So, we're finishing up on this paper about nonequilibrium DMFT that uses a weak-coupling expansion and tensor cross interpolation.

Mira: Yeah, Kai, looking at the title again, it really points to a specific way they tackle those hard problems in quantum many-body systems.

Lev: From where I sit with building hardware, I'm interested in how much overhead this adds to running things on real machines.

Kai: Well, the authors are showing how they can solve those complex impurity problems—the little bits of the system we have to look at—when things aren't in perfect equilibrium.

Mira: They’re essentially building a tool that lets us calculate how a system reacts when you suddenly change an interaction strength.

Lev: And what I see is that this method works up to twentieth order in the expansion, which is pretty substantial for handling complexity.

Kai: That’s the low-rank structure they found in the high-dimensional integrals, which means they don't have to do all that massive computation every time.

Mira: It’s about finding a shortcut—a specific mathematical shape—that lets them approximate those huge integrals very efficiently without needing messy stochastic sampling.

Lev: So, it avoids the sign problem issues that plague other methods when you look at systems away from perfect half-filling.

Kai: Exactly, and the results on how doping changes that fast thermalization behavior are what really stick out to me.

Mira: They’re showing that even when things get more complicated with doping, like at three quarters filling, the method stays numerically stable.

Lev: That stability is key for any real physics application because you need reliable numbers, not just a theoretical curve.

Kai: So it seems this solver opens up access to studying these nonequilibrium dynamics in parameter regimes that were previously too hard to reach with existing techniques.

Mira: It’s a systematic way to push the limits of what we can calculate in these complex quantum models.

Lev: That leads us into the next part, how they actually did it, and if that complexity is something we can tame for actual hardware simulations.

Shuta Matsuura, Hiroshi Shinaoka, Philipp Werner, Naoto Tsuji

Department of Physics, University of Tokyo · Department of Physics, Saitama University · Department of Physics, University of Fribourg · RIKEN Center for Emergent Matter Science (CEMS) · Trans-scale Quantum Science Institute, University of Tokyo

cond-mat.str-el

Submitted: 2026-07-01

Updated: 2026-11-09

Comments: 30 pages, 19 figures

Journal ref: Phys. Rev. B 114, 295108 (2026)

DOI: 10.1103/bvds-1gxf

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 91/100

The gist: Simulating nonequilibrium quantum many-body systems remains a major challenge due to the exponential growth of the computational complexity with real time Page 1.

Key concepts

Nonequilibrium Impurity Solver
This is a specialized computational tool designed to solve the difficult quantum mechanics problem of an impurity (a small part of the system) when the entire system is evolving over time. It combines a weak-coupling expansion with TCI to efficiently calculate necessary Green's functions for DMFT.
Weak-Coupling Expansion
This is a mathematical technique used to approximate complex interactions in quantum systems by expanding them in terms of small interaction strengths. The paper uses this to break down high-dimensional integrals that arise during the calculation, making them manageable.
Tensor Cross Interpolation (TCI)
TCI is an algorithm used to efficiently approximate high-dimensional integrals by exploiting low-rank structures. The paper shows that the integrands from the weak-coupling expansion have this low-rank structure, allowing TCI to evaluate these integrals without needing slow stochastic sampling.
Dynamical Mean Field Theory (DMFT)
DMFT is a theoretical framework used to study strongly correlated electron systems. It simplifies a complex lattice problem by mapping it onto an effective single-site impurity problem that evolves in time, which this solver is designed to address.

Terminology

Summary

Simulating nonequilibrium quantum many-body systems remains a major challenge due to the exponential growth of the computational complexity with real time Page 1. p. 0 Here we implement a nonequilibrium impurity solver based on the weak-coupling expansion and the tensor cross interpolation (TCI), and apply it to nonequilibrium dynamical mean-field theory (DMFT) Page 1. p. 0 The method approximates the integrands of the high-dimensional integrals arising in the weak-coupling expansion in a tensor-train form, enabling efficient evaluations without stochastic sampling and thereby mitigating the sign problem affecting continuous-time quantum Monte Carlo (CT-QMC) methods Page 1. p. 0

Methodology for Solving Impurity Problems

The core of the approach involves developing a nonequilibrium impurity solver that combines the weak-coupling expansion with the TCI algorithm to compute the two-time Green’s functions required for nonequilibrium DMFT Page 1. The method benchmarks against an exactly solvable nonequilibrium impurity model and shows that the integrands appearing in the weak-coupling expansion indeed exhibit low-rank structures that can be efficiently captured by TCI Page 1. p. 1 This allows for evaluation of the perturbation series up to twentieth order with moderate computational cost Page 1. p. 1

Application to Nonequilibrium DMFT

The solver is applied to interaction quenches in the Hubbard model, reproducing fast thermalization at a critical interaction strength with accuracy comparable to CT-QMC at half filling Page 1. Crucially, away from half filling where the sign problem is more severe for CT-QMC, the present approach remains well controlled, revealing a crossover instead of a sharply defined fast thermalization point in the 3/4-filled case Page 1. p. 0 The solver can also be applied to steady-state DMFT problems, yielding accurate spectral functions in the metallic regime without analytic continuation Page 1. p. 0

Benchmark Results and Low-Rank Structure

In the particle-hole symmetric case, the TCI results converge systematically to the exact solution as nmax is increased, suggesting that the 20-dimensional integrals can be accurately evaluated by the tensor-train method with a relatively low bond dimension χ = 40 Page 11. The analysis shows that the integrand of the weak-coupling expansion has a low-rank structure, which is related to the lesser and greater components of the Weiss Green’s function being constant for a specific choice of α Page 14. p. 9

Dynamics Away from Half-Filling

Away from half filling, where the sign problem becomes even more severe for CT-QMC, the TCI approach remains numerically well controlled, and it is shown that the 3/4-filled case can be treated with nearly the same numerical effort as the particle-hole-symmetric case Page 1. The results indicate that the sharp fast-thermalization behavior observed at half filling is absent within the accessible interaction range Page 1. p. 1

Steady-State Calculations and Limitations

The method can be used for steady-state DMFT calculations, where the distribution function fst(ω) is specified as an input to calculate the spectral function Aσ(ω) directly Page 1. However, the computational cost grows exponentially with the perturbation order due to explicit summation over Keldysh indices Page 3. p. 7 In transient calculations, the maximum reachable time is limited to approximately Utmax ≲ 14. p. 67, and in steady-state formalism, accurate calculations are restricted to U ≲ 2v Page 21. p. 0

Future Directions

The most important next step is to compress the information carried by the Keldysh indices within the tensor-train representation so that the explicit summation over them no longer is necessary, which could reduce computational cost to a polynomial scaling Page 23. Another important direction is the extension to cluster impurity solvers, as the present weak-coupling solver is naturally suited for cluster impurity problems Page 23. p. 1 The TCI approach establishes itself as a promising route to nonequilibrium DMFT in parameter regimes that are difficult to access with existing approaches Page 23.

Summary and Outlook

In this work, we developed a nonequilibrium TCI impurity solver based on the weak-coupling expansion on the Kadanoff–Baym contour and applied it to nonequilibrium DMFT for both transient and steady-state calculations Page 1. The benchmarks against an exactly solvable impurity problem showed systematic convergence with increasing expansion order and for moderate bond dimensions, indicating that the nonequilibrium integrand admits an efficient low-rank tensor-train representation Page 23. p. 0

The method reproduced the known half-filled quench dynamics and the rapid thermalization near the dynamical transition point around U = 3v Page 15. p. 1 Moreover, it remained well controlled in the 3/4-filled case, where the rapid-thermalization regime observed at half filling evolves into a much broader crossover and no comparably fast thermalization behavior is found Page 19. p. 0

The present calculations thus provide numerically exact evidence that the dynamical transition at half filling evolves into a smooth crossover upon doping Page 19. p. 1 The crossover marks a qualitative change from a relatively smooth relaxation on the weak-U side to an atomic-like oscillatory post-quench dynamics on the strong-U side Page 19.

The TCI solver establishes itself as a useful approach for nonequilibrium DMFT Page 23. p. 1 The method is systematically improvable by increasing the maximum order and allows us to obtain reliable results even in the intermediate-to-strong-coupling regime, at least for short times Page 23. p. 8 p. 8

The present work establishes the nonequilibrium TCI impurity solver as a promising route to nonequilibrium DMFT in parameter regimes that are difficult to access with existing approaches Page 23. The method is systematically improvable by increasing the maximum order and allows us to obtain reliable results even in the intermediate-to-strong-coupling regime, at least for short times Page 23.

--- Page 1 ---

Page 1</ref:2607.018

Page 1</ref:2653

Page 1</ref:2653

Page 3</ref:49

Page 4</ref:50

Page 18</ref:77

Page 22</ref:81

Pruning redundant reasoning tokens

The method approximates the integrands of the high-dimensional integrals arising in the weak-coupling expansion in a tensor-train form, enabling efficient evaluations without stochastic sampling and thereby mitigating the sign problem affecting continuous-time quantum Monte Carlo (CT-QMC) methods Page 1. Benchmark calculations for an exactly solvable nonequilibrium impurity model agree well with the exact results and reveal a low-rank structure of the integrands Page 1. p. 0

How the training signal is built

To calculate the interacting Green’s function Gσ(z, z′) using Eqs. (26) and (27), we need to evaluate high-dimensional integrals, whose computational cost grows exponentially with the maximum order nmax when a standard quadrature method is used Page 6. p. 5 The TCI algorithm is applied to approximate the function Q. p. 0

Where the gains come from

The low-rank structure of the integrand is understood because the lesser and greater components of the Weiss Green’s function for the spindown electrons are constant for this choice of α Page 14. This reduces the number of terms contributing to the determinant from (n+1)! p. 4 to n! Page 14. The computational cost is estimated as O(2Utmax)(U tmax)2χ 2d Page 14. p. 8

Sign Cancellation in the TCI Integration

The ratio of the original Green’s function to this integral of the absolute value, S Xσ(t, t′), is a measure of the severity of the sign cancellation during the integration Page 13. The TCI approach allows us to quantify these phase cancellations in the continuous-time integrals that are actually interpolated by the TCI algorithm Page 13. p. 0

Low-rank structure of the integrand

The origin of the lowrank structure of the integrand for α = 0.5 + δµ/U can be understood because the second determinant in Eq. p. 9 (30) becomes constant and independent of all indices Page 14. p. 6 This implies that the integrand has a low-rank tensor-train representation with respect to the variables u i Page 14. p. 9

Interaction Quenches in the Hubbard Model

The method reproduced the known half-filled quench dynamics and the rapid thermalization near the dynamical transition point around U = 3v Page 15. Away from half filling, where the sign problem becomes even more severe for CT-QMC, the present approach remains well controlled Page 1. p. 0

Relaxation toward thermal equilibrium

In the particle-hole symmetric case, r(t) can reach zero, within numerical accuracy, at certain interaction strengths and times Page 13. Away from half filling, however, this zero-touching behavior is lost Page 19. p. 1

Steady-State DMFT for the Hubbard Model

The steady-state DMFT calculation for the half-filled Hubbard model with T = 0, and U = 2v exhibits a quasiparticle peak at the Fermi level ω = 0 Page 16. The results for U = 3v and U = 8v show that the total energy is conserved in both the weak-coupling and strong-coupling regimes Page 15. p. 1

Conclusion

The present work establishes the nonequilibrium TCI impurity solver as a promising route to nonequilibrium DMFT in parameter regimes that are difficult to access with existing approaches Page 23. The method is systematically improvable by increasing the maximum order and allows us to obtain reliable results even in the intermediate-to-strong-coupling regime, at least for short times Page 21. p. 8

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Develop an AI system capable of efficiently simulating nonequilibrium dynamical mean-field theory (DMFT) for strongly correlated systems by implementing a weak-coupling expansion impurity solver based on tensor cross interpolation (TCI). This system can evaluate high-dimensional integrals arising in the expansion using a tensor-train form, enabling efficient evaluations without stochastic sampling.

  2. Enable the AI system to accurately predict the two-time Green's functions required for nonequilibrium DMFT by benchmarking against exactly solvable impurity models and showing that integrands exhibit low-rank structures suitable for TCI approximation. This allows for evaluation of the perturbation series up to twentieth order with moderate computational cost.

  3. Allow the AI system to handle particle-hole asymmetric impurity problems, such as those at 3/4-filling, with only a constant-factor overhead compared to particle-hole symmetric cases using the present TCI approach. This makes nonequilibrium DMFT feasible away from half filling where sign problems severely limit CT-QMC solvers.

  4. Provide the capability to calculate steady-state spectral functions of the Hubbard model directly from an input distribution function fst(omega) without analytic continuation, consistent with nonequilibrium DMFT for steady states in the metallic regime. This eliminates a major numerical hurdle in equilibrium imaginary-time solvers.

  5. Implement a mechanism to quantify sign or phase cancellations during TCI integration by calculating the ratio of original Green's functions to the integral of their absolute value, measuring the severity of the sign cancellation during the integration. This allows for assessment of numerical stability when average signs are small, such as 10-2.

  6. Integrate a method for smoothly extrapolating Weiss Green's functions from a finite time window [0, tmax] to an infinite time range [−∞, tmax] using linear combinations of h1(t) and h2(t). This extrapolation ensures the extrapolated function is differentiable at t = 0 and decays exponentially as t approaches negative infinity.

  7. Allow the AI system to estimate the required maximum expansion order nmax based on model parameters, such as interaction strength U and simulation time tmax, using a scaling estimate of nmax approximately proportional to e(2Utmax). This guides practical computational choices for balancing accuracy and cost.

  8. Enable the AI system to reconstruct the full Green's function by performing recursive convolutions of matrix-valued functions derived from the tensor-train decomposition, utilizing Chebyshev polynomial expansions for efficient convolution evaluation.

Sources

Related papers