Neural networks as low-cost surrogates for impurity solvers in quantum embedding methods

Kipton Barros; Philip M. Dee; Rohan Nain; Steven Johnston; Thomas A. Maier

arxiv: 2603.25557 · v2 · submitted 2026-03-26 · ❄️ cond-mat.str-el

Neural networks as low-cost surrogates for impurity solvers in quantum embedding methods

Rohan Nain , Philip M. Dee , Kipton Barros , Steven Johnston , Thomas A. Maier This is my paper

Pith reviewed 2026-05-15 00:38 UTC · model grok-4.3

classification ❄️ cond-mat.str-el

keywords neural networkimpurity solverdynamical mean-field theoryquantum Monte Carlosurrogate modelcorrelated electronsmachine learningGreen's function

0 comments

The pith

Neural network serves as low-cost surrogate for impurity solvers in dynamical mean-field theory, matching CT-QMC accuracy on interpolation and accelerating it by up to five times as initial guess.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that a neural network can be trained with one to two orders of magnitude fewer examples than previous efforts to replace the impurity solver inside dynamical mean-field theory loops. Inside the range of training parameters the network reproduces continuous-time quantum Monte Carlo results to high accuracy. When the target temperature drops below the training window the network predictions lose fidelity, yet they still supply starting configurations that let the full Monte Carlo solver converge up to five times faster. The combination therefore supports rapid sweeps across interaction strengths and temperatures in models of strongly correlated electrons.

Core claim

A neural network trained in the low-data regime serves as an efficient substitute for the impurity solver in dynamical mean-field theory simulations of correlated electron models. The network achieves accuracy comparable to continuous-time quantum Monte Carlo solvers when interpolating between samples within the training set. Its output provides an excellent initial guess for CT-QMC solvers, accelerating the time to solution by up to a factor of five even when extrapolating to lower temperatures outside the training distribution.

What carries the argument

Neural network surrogate that maps DMFT input parameters to the impurity Green's function or self-energy and is used either in place of or as initializer for the CT-QMC solver.

If this is right

Self-consistent DMFT calculations become feasible on larger parameter grids with fixed computational budget.
Hybrid NN-plus-CT-QMC workflows reduce total runtime while retaining Monte Carlo statistical control.
Exploratory scans of interaction strength, doping, and temperature can be performed orders of magnitude faster inside the interpolation regime.
The same trained network can be reused across multiple material models that share similar local interaction ranges.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Retraining the network on progressively lower temperatures could eventually eliminate the need for the hybrid Monte Carlo step.
The surrogate approach may transfer directly to cluster DMFT or other quantum embedding schemes that also rely on expensive impurity solvers.
Feeding the network output straight into analytic continuations or response-function calculations could bypass Monte Carlo entirely for screening studies.

Load-bearing premise

The training distribution sufficiently covers the relevant parameter space so that the network remains useful when the target simulation moves to lower temperatures outside the training set.

What would settle it

A DMFT run at a temperature below the training range in which the NN-initialized CT-QMC shows no reduction in wall-clock time or iteration count relative to a standard random initialization.

Figures

Figures reproduced from arXiv: 2603.25557 by Kipton Barros, Philip M. Dee, Rohan Nain, Steven Johnston, Thomas A. Maier.

**Figure 3.** Figure 3: FIG. 3. A comparison of the convergence of the DMFT loop [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4. The evolution of (a) the low-frequency slope in the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 7.** Figure 7: FIG. 7. (a) Double occupancy [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

**Figure 9.** Figure 9: FIG. 9. Imaginary-time Green’s functions [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗

read the original abstract

A promising application of machine learning is the creation of low-cost surrogate models to mitigate computational bottlenecks in quantum many-body simulations. Here, we explore whether a neural network (NN) can be trained in the low-data regime, with one to two orders of magnitude fewer training examples than previous works, as an efficient substitute for the impurity solver in dynamical mean-field theory simulations of correlated electron models. We show that the NN solver achieves accuracy comparable to popular continuous-time quantum Monte Carlo (CT-QMC) impurity solvers when interpolating between samples within the training set. While the NN's performance decreases notably when extrapolating to lower temperatures outside the training distribution, its output still provides an excellent initial guess for input to more accurate CT-QMC impurity solvers, thus accelerating the time to solution up to a factor of five. We discuss our results in the context of rapid phase-space exploration.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

NN surrogate matches CT-QMC on interpolation with far less training data and still speeds it up by up to 5x as an initial guess, even when extrapolating to lower T.

read the letter

The main point is that this neural-network surrogate for the impurity solver works with far less training data than previous attempts and still delivers a useful speed-up when its output is fed as an initial guess into CT-QMC, even for temperatures below the training set. The authors train the network on one to two orders of magnitude fewer examples and report accuracy comparable to CT-QMC inside the training distribution. The new practical piece is the explicit demonstration that the NN output accelerates the Monte Carlo run by up to a factor of five for lower-temperature DMFT calculations. That combination of low-data training plus runtime gain is what they add beyond earlier neural-network solver papers. The evidence is empirical and rests on direct comparison to CT-QMC reference data, so the claims are straightforward to check. No circular fitting is involved. The paper is open about the fact that extrapolation performance drops, which keeps the discussion grounded. The soft spot is the lack of a clear bound on how much the NN can deviate before the initial-guess advantage goes away. The abstract notes the degradation but does not show, for example, the distance to the true solution or the number of CT-QMC steps saved as a function of that distance. If the NN stays reasonably close even when accuracy falls, the speed-up holds; if the error grows too large, the benefit may be smaller than claimed. Readers will want to see those convergence curves in the figures. This paper is for condensed-matter groups that already run DMFT and want to explore larger parameter spaces without paying the full cost at every point. A reader who knows the CT-QMC literature will find the numbers easy to evaluate and decide whether to try the approach. I would send it out for peer review. The central result is practical and testable, and the stated limitation gives referees something concrete to examine.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes training neural networks on CT-QMC data to serve as low-cost surrogates for impurity solvers in DMFT/quantum embedding calculations. With one to two orders of magnitude fewer training examples than prior works, the NN matches CT-QMC accuracy on interpolation within the training distribution; in extrapolation to lower temperatures the accuracy degrades, yet the NN output remains an effective initial guess that accelerates CT-QMC convergence by up to a factor of five, enabling faster phase-space exploration.

Significance. If the central claims hold, the work provides a practical route to reducing the computational bottleneck of impurity solvers while remaining anchored to exact CT-QMC results. The low-data training regime and the demonstrated initial-guess acceleration constitute concrete strengths that could facilitate broader parameter scans in correlated-electron studies.

major comments (2)

[Abstract and §4.3] Abstract and §4.3: the headline claim that NN output accelerates CT-QMC by up to a factor of five in the extrapolation regime (lower T) is load-bearing for the practical utility argument, yet no quantitative bound is supplied on the maximum deviation between NN prediction and true solution beyond which the initial-guess benefit vanishes. This omission leaves the extrapolation speedup claim without a clear robustness criterion.
[§3.2 and Table 2] §3.2 and Table 2: the training-set coverage of the relevant parameter space (especially temperature) is asserted to be sufficient for useful extrapolation, but the manuscript provides no systematic test (e.g., distance-to-training-set metrics or ablation on temperature spacing) that would confirm the weakest assumption remains valid when the target lies outside the training distribution.

minor comments (2)

[Figure 4] Figure 4: convergence-time histograms would be clearer if the conventional starting-point baseline were shown on the same axes for direct visual comparison.
[Methods] Notation: the definition of the impurity-model parameters (U, T, doping) should be restated once in the methods section to avoid cross-referencing the abstract.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address each major point below and have revised the manuscript to incorporate additional analysis where appropriate.

read point-by-point responses

Referee: [Abstract and §4.3] Abstract and §4.3: the headline claim that NN output accelerates CT-QMC by up to a factor of five in the extrapolation regime (lower T) is load-bearing for the practical utility argument, yet no quantitative bound is supplied on the maximum deviation between NN prediction and true solution beyond which the initial-guess benefit vanishes. This omission leaves the extrapolation speedup claim without a clear robustness criterion.

Authors: We agree that a quantitative bound on the deviation would strengthen the robustness of the speedup claim. In the revised manuscript we have added a new paragraph and figure in §4.3 that plots the observed CT-QMC convergence speedup versus the initial L2 deviation (Frobenius norm between the NN-predicted Green's function and the fully converged solution). The data show that the factor-of-five acceleration persists for initial deviations up to ~0.12; beyond this threshold the benefit drops below 2×. We have updated the abstract to reference this practical bound. revision: yes
Referee: [§3.2 and Table 2] §3.2 and Table 2: the training-set coverage of the relevant parameter space (especially temperature) is asserted to be sufficient for useful extrapolation, but the manuscript provides no systematic test (e.g., distance-to-training-set metrics or ablation on temperature spacing) that would confirm the weakest assumption remains valid when the target lies outside the training distribution.

Authors: We acknowledge that systematic validation of the extrapolation regime was missing. In the revised version we have added to §3.2 (i) the Euclidean distance in normalized (U,T,μ) space from each extrapolation target to its nearest training point and (ii) an ablation study that successively coarsens the temperature grid while keeping the total number of training examples fixed. Both the distance metric and the ablation results are now reported in Table 2 and a new supplementary figure; they confirm that useful initial guesses are obtained even when the target temperature lies 30–50 % below the lowest training temperature. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical NN training against independent CT-QMC benchmarks

full rationale

The paper trains a neural network surrogate on CT-QMC-generated data for impurity solvers in DMFT and validates interpolation accuracy plus initial-guess acceleration via direct comparison to fresh CT-QMC runs. All performance metrics are measured against external reference data rather than derived from the network itself. No equation reduces to a fitted parameter by construction, no self-citation supplies a load-bearing uniqueness theorem, and the central claims remain falsifiable against the independent Monte Carlo solver. The work is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper rests on standard DMFT formalism and standard supervised learning assumptions; no new axioms or invented entities are introduced.

pith-pipeline@v0.9.0 · 5459 in / 1026 out tokens · 27366 ms · 2026-05-15T00:38:51.121718+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

neural network (NN) can be trained ... to predict the impurity site’s self-energy Σ(τ) ... from U, β, and G0(τ) Legendre coefficients
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Legendre polynomial basis ... lmax=60 ... even coefficients only

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

46 extracted references · 46 canonical work pages

[1]

Carleo, I

G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zdeborov´ a, Machine learning and the physical sciences, Rev. Mod. Phys.91, 045002 (2019)

work page 2019
[2]

Carrasquilla, Machine learning for quantum matter, Advances in Physics: X5, 1797528 (2020)

J. Carrasquilla, Machine learning for quantum matter, Advances in Physics: X5, 1797528 (2020)

work page 2020
[3]

Bedolla, L

E. Bedolla, L. C. Padierna, and R. Casta˜ neda-Priego, Machine learning for condensed matter physics, Journal of Physics: Condensed Matter33, 053001 (2020)

work page 2020
[4]

Johnston, E

S. Johnston, E. Khatami, and R. Scalettar, A perspec- tive on machine learning and data science for strongly correlated electron problems, Carbon Trends9, 100231 (2022)

work page 2022
[5]

Jinnouchi, F

R. Jinnouchi, F. Karsai, and G. Kresse, On-the-fly ma- chine learning force field generation: Application to melt- ing points, Phys. Rev. B100, 014105 (2019)

work page 2019
[6]

No´ e, A

F. No´ e, A. Tkatchenko, K.-R. M¨ uller, and C. Clementi, Machine learning for molecular simulation, Annual Re- view of Physical Chemistry71, 361 (2020)

work page 2020
[7]

Praˇ snikar, M

E. Praˇ snikar, M. Ljubiˇ c, A. Perdih, and J. Boriˇ sek, Ma- chine learning heralding a new development phase in molecular dynamics simulations, Artificial Intelligence Review57, 102 (2024)

work page 2024
[8]

J. Liu, Y. Qi, Z. Y. Meng, and L. Fu, Self-learning Monte Carlo method, Phys. Rev. B95, 041101 (2017)

work page 2017
[9]

G. Pan, C. Chen, and Z. Y. Meng, Self-learning Monte Carlo method: A review, arXiv:2507.12554 (2025)

work page arXiv 2025
[10]

Huang and L

L. Huang and L. Wang, Accelerated Monte Carlo simu- lations with restricted Boltzmann machines, Phys. Rev. B95, 035105 (2017)

work page 2017
[11]

Nagai, H

Y. Nagai, H. Shen, Y. Qi, J. Liu, and L. Fu, Self-learning Monte Carlo method: Continuous-time algorithm, Phys. Rev. B96, 161102 (2017). 10

work page 2017
[12]

H. Shen, J. Liu, and L. Fu, Self-learning Monte Carlo with deep neural networks, Phys. Rev. B97, 205140 (2018)

work page 2018
[13]

C. Chen, X. Y. Xu, J. Liu, G. Batrouni, R. Scalettar, and Z. Y. Meng, Symmetry-enforced self-learning Monte Carlo method applied to the Holstein model, Phys. Rev. B98, 041102 (2018)

work page 2018
[14]

S. Li, P. M. Dee, E. Khatami, and S. Johnston, Accel- erating lattice quantum Monte Carlo simulations using artificial neural networks: Application to the Holstein model, Phys. Rev. B100, 020302 (2019)

work page 2019
[15]

S. S. Bakshi, Y. Fan, and G.-W. Chern, Machine learn- ing modeling of charge-density-wave recovery after laser melting, arXiv:2602.03761 (2026)

work page arXiv 2026
[16]

Georges, G

A. Georges, G. Kotliar, W. Krauth, and M. J. Rozen- berg, Dynamical mean-field theory of strongly correlated fermion systems and the limit of infinite dimensions, Rev. Mod. Phys.68, 13 (1996)

work page 1996
[17]

, author Jarrell, M

T. Maier, M. Jarrell, T. Pruschke, and M. H. Hettler, Quantum cluster theories, Rev. Mod. Phys.77, 1027 (2005)

work page 2005
[18]

M. J. Rozenberg, G. Kotliar, and X. Y. Zhang, Mott- Hubbard transition in infinite dimensions. II, Phys. Rev. B49, 10181 (1994)

work page 1994
[19]

Bulla, T

R. Bulla, T. A. Costi, and D. Vollhardt, Finite- temperature numerical renormalization group study of the Mott transition, Phys. Rev. B64, 045103 (2001)

work page 2001
[20]

Joo and V

J. Joo and V. Oudovenko, Quantum Monte Carlo calcu- lation of the finite temperature Mott-Hubbard transition, Phys. Rev. B64, 193102 (2001)

work page 2001
[21]

T. A. Maier, M. S. Jarrell, and D. J. Scalapino, Structure of the pairing interaction in the two-dimensional Hub- bard model, Phys. Rev. Lett.96, 047005 (2006)

work page 2006
[22]

E. Gull, O. Parcollet, and A. J. Millis, Superconductiv- ity and the pseudogap in the two-dimensional Hubbard model, Phys. Rev. Lett.110, 216405 (2013)

work page 2013
[23]

X. Dong, E. Gull, and A. J. Millis, Quantifying the role of antiferromagnetic fluctuations in the superconductivity of the doped Hubbard model, Nature Physics18, 1293 (2022)

work page 2022
[24]

P. Mai, S. Karakuzu, G. Balduzzi, S. Johnston, and T. A. Maier, Intertwined spin, charge, and pair correlations in the two-dimensional Hubbard model in the thermody- namic limit, Proceedings of the National Academy of Sci- ences119, e2112806119 (2022)

work page 2022
[25]

Kumar, D

A. Kumar, D. S´ en´ echal, and A.-M. S. Tremblay, Cluster dynamical mean-field study of intra-unit-cell charge ne- maticity in hole-doped cuprates, Proceedings of the Na- tional Academy of Sciences122, e2419534122 (2025)

work page 2025
[26]

Kotliar, S

G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko, O. Parcollet, and C. A. Marianetti, Electronic struc- ture calculations with dynamical mean-field theory, Rev. Mod. Phys.78, 865 (2006)

work page 2006
[27]

H. Aoki, N. Tsuji, M. Eckstein, M. Kollar, T. Oka, and P. Werner, Nonequilibrium dynamical mean-field theory and its applications, Rev. Mod. Phys.86, 779 (2014)

work page 2014
[28]

L.-F. m. c. Arsenault, A. Lopez-Bezanilla, O. A. von Lilienfeld, and A. J. Millis, Machine learning for many- body physics: The case of the Anderson impurity model, Phys. Rev. B90, 155136 (2014)

work page 2014
[29]

Rogers, T.-H

J. Rogers, T.-H. Lee, S. Pakdel, W. Xu, V. Do- brosavljevi´ c, Y.-X. Yao, O. Christiansen, and N. Lanat` a, Bypassing the computational bottleneck of quantum- embedding theories for strong electron correlations with machine learning, Physical Review Research3, 013101 (2021)

work page 2021
[30]

Sheridan, C

E. Sheridan, C. Rhodes, F. Jamet, I. Rungger, and C. Weber, Data-driven dynamical mean-field theory: An error-correction approach to solve the quantum many- body problem using machine learning, Phys. Rev. B104, 205120 (2021)

work page 2021
[31]

Agapov, O

E. Agapov, O. Bertomeu, A. Carballo, C. B. Mendl, and A. Sander, Predicting interacting Green’s functions with neural networks, arXiv:2411.13644 (2024)

work page arXiv 2024
[32]

H. Lee, Z. Zhao, G. H. Booth, W. Ge, and C. Weber, Language-inspired machine learning approach for solving strongly correlated problems with dynamical mean-field theory, Phys. Rev. B112, 035165 (2025)

work page 2025
[33]

E. Gull, A. J. Millis, A. I. Lichtenstein, A. N. Rubtsov, M. Troyer, and P. Werner, Continuous-time Monte Carlo methods for quantum impurity models, Rev. Mod. Phys. 83, 349 (2011)

work page 2011
[34]

Liebsch and H

A. Liebsch and H. Ishida, Temperature and bath size in exact diagonalization dynamical mean field theory, Jour- nal of Physics: Condensed Matter24, 053201 (2011)

work page 2011
[35]

Bulla, T

R. Bulla, T. A. Costi, and T. Pruschke, Numerical renor- malization group method for quantum impurity systems, Rev. Mod. Phys.80, 395 (2008)

work page 2008
[36]

Parcollet, M

O. Parcollet, M. Ferrero, T. Ayral, H. Hafermann, I. Krivenko, L. Messio, and P. Seth, TRIQS: A toolbox for research on interacting quantum systems, Computer Physics Communications196, 398 (2015)

work page 2015
[37]

Boehnke, H

L. Boehnke, H. Hafermann, M. Ferrero, F. Lechermann, and O. Parcollet, Orthogonal polynomial representation of imaginary-time Green’s functions, Phys. Rev. B84, 075145 (2011)

work page 2011
[38]

Loshchilov and F

I. Loshchilov and F. Hutter, Decoupled weight decay regularization, inInternational Conference on Learning Representations (ICLR)(2019)

work page 2019
[39]

ISAAC-NG cluster: System overview,https: //oit.utk.edu/hpsc/isaac-open-enclave-new-kpb/ system-overview-cluster-at-kpb/, accessed: 25-03- 2026

work page 2026
[40]

Kotliar, S

G. Kotliar, S. Murthy, and M. J. Rozenberg, Compress- ibility divergence and the finite temperature Mott tran- sition, Phys. Rev. Lett.89, 046401 (2002)

work page 2002
[41]

Bl¨ umer,Mott-Hubbard Metal-Insulator Transition and Optical Conductivity in High Dimensions, Ph.D

N. Bl¨ umer,Mott-Hubbard Metal-Insulator Transition and Optical Conductivity in High Dimensions, Ph.D. thesis, Universit¨ at Augsburg (2003)

work page 2003
[42]

V. M. Galitskii and A. B. Migdal, Application of quan- tum field theory methods to the many-body problem, Zh. Eksp. Teor. Fiz.34, 139 (1958)

work page 1958
[43]

V. M. Galitskii and A. B. Migdal, Application of quan- tum field theory methods to the many-body problem, Sov. Phys. JETP7, 96 (1958)

work page 1958
[44]

Gull and A

E. Gull and A. J. Millis, Energetics of superconductivity in the two-dimensional Hubbard model, Phys. Rev. B86, 241106 (2012)

work page 2012
[45]

E. G. C. P. van Loon, F. Krien, H. Hafermann, E. A. Stepanov, A. I. Lichtenstein, and M. I. Katsnelson, Dou- ble occupancy in dynamical mean-field theory and the dual boson approach, Phys. Rev. B93, 155162 (2016)

work page 2016
[46]

Valenti, I

A. Valenti, I. Park, A. Georges, A. J. Millis, and O. Par- collet, Neural-network quantum embedding solvers for correlated materials, arXiv:2603.15741 (2026)

work page arXiv 2026

[1] [1]

Carleo, I

G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zdeborov´ a, Machine learning and the physical sciences, Rev. Mod. Phys.91, 045002 (2019)

work page 2019

[2] [2]

Carrasquilla, Machine learning for quantum matter, Advances in Physics: X5, 1797528 (2020)

J. Carrasquilla, Machine learning for quantum matter, Advances in Physics: X5, 1797528 (2020)

work page 2020

[3] [3]

Bedolla, L

E. Bedolla, L. C. Padierna, and R. Casta˜ neda-Priego, Machine learning for condensed matter physics, Journal of Physics: Condensed Matter33, 053001 (2020)

work page 2020

[4] [4]

Johnston, E

S. Johnston, E. Khatami, and R. Scalettar, A perspec- tive on machine learning and data science for strongly correlated electron problems, Carbon Trends9, 100231 (2022)

work page 2022

[5] [5]

Jinnouchi, F

R. Jinnouchi, F. Karsai, and G. Kresse, On-the-fly ma- chine learning force field generation: Application to melt- ing points, Phys. Rev. B100, 014105 (2019)

work page 2019

[6] [6]

No´ e, A

F. No´ e, A. Tkatchenko, K.-R. M¨ uller, and C. Clementi, Machine learning for molecular simulation, Annual Re- view of Physical Chemistry71, 361 (2020)

work page 2020

[7] [7]

Praˇ snikar, M

E. Praˇ snikar, M. Ljubiˇ c, A. Perdih, and J. Boriˇ sek, Ma- chine learning heralding a new development phase in molecular dynamics simulations, Artificial Intelligence Review57, 102 (2024)

work page 2024

[8] [8]

J. Liu, Y. Qi, Z. Y. Meng, and L. Fu, Self-learning Monte Carlo method, Phys. Rev. B95, 041101 (2017)

work page 2017

[9] [9]

G. Pan, C. Chen, and Z. Y. Meng, Self-learning Monte Carlo method: A review, arXiv:2507.12554 (2025)

work page arXiv 2025

[10] [10]

Huang and L

L. Huang and L. Wang, Accelerated Monte Carlo simu- lations with restricted Boltzmann machines, Phys. Rev. B95, 035105 (2017)

work page 2017

[11] [11]

Nagai, H

Y. Nagai, H. Shen, Y. Qi, J. Liu, and L. Fu, Self-learning Monte Carlo method: Continuous-time algorithm, Phys. Rev. B96, 161102 (2017). 10

work page 2017

[12] [12]

H. Shen, J. Liu, and L. Fu, Self-learning Monte Carlo with deep neural networks, Phys. Rev. B97, 205140 (2018)

work page 2018

[13] [13]

C. Chen, X. Y. Xu, J. Liu, G. Batrouni, R. Scalettar, and Z. Y. Meng, Symmetry-enforced self-learning Monte Carlo method applied to the Holstein model, Phys. Rev. B98, 041102 (2018)

work page 2018

[14] [14]

S. Li, P. M. Dee, E. Khatami, and S. Johnston, Accel- erating lattice quantum Monte Carlo simulations using artificial neural networks: Application to the Holstein model, Phys. Rev. B100, 020302 (2019)

work page 2019

[15] [15]

S. S. Bakshi, Y. Fan, and G.-W. Chern, Machine learn- ing modeling of charge-density-wave recovery after laser melting, arXiv:2602.03761 (2026)

work page arXiv 2026

[16] [16]

Georges, G

A. Georges, G. Kotliar, W. Krauth, and M. J. Rozen- berg, Dynamical mean-field theory of strongly correlated fermion systems and the limit of infinite dimensions, Rev. Mod. Phys.68, 13 (1996)

work page 1996

[17] [17]

, author Jarrell, M

T. Maier, M. Jarrell, T. Pruschke, and M. H. Hettler, Quantum cluster theories, Rev. Mod. Phys.77, 1027 (2005)

work page 2005

[18] [18]

M. J. Rozenberg, G. Kotliar, and X. Y. Zhang, Mott- Hubbard transition in infinite dimensions. II, Phys. Rev. B49, 10181 (1994)

work page 1994

[19] [19]

Bulla, T

R. Bulla, T. A. Costi, and D. Vollhardt, Finite- temperature numerical renormalization group study of the Mott transition, Phys. Rev. B64, 045103 (2001)

work page 2001

[20] [20]

Joo and V

J. Joo and V. Oudovenko, Quantum Monte Carlo calcu- lation of the finite temperature Mott-Hubbard transition, Phys. Rev. B64, 193102 (2001)

work page 2001

[21] [21]

T. A. Maier, M. S. Jarrell, and D. J. Scalapino, Structure of the pairing interaction in the two-dimensional Hub- bard model, Phys. Rev. Lett.96, 047005 (2006)

work page 2006

[22] [22]

E. Gull, O. Parcollet, and A. J. Millis, Superconductiv- ity and the pseudogap in the two-dimensional Hubbard model, Phys. Rev. Lett.110, 216405 (2013)

work page 2013

[23] [23]

X. Dong, E. Gull, and A. J. Millis, Quantifying the role of antiferromagnetic fluctuations in the superconductivity of the doped Hubbard model, Nature Physics18, 1293 (2022)

work page 2022

[24] [24]

P. Mai, S. Karakuzu, G. Balduzzi, S. Johnston, and T. A. Maier, Intertwined spin, charge, and pair correlations in the two-dimensional Hubbard model in the thermody- namic limit, Proceedings of the National Academy of Sci- ences119, e2112806119 (2022)

work page 2022

[25] [25]

Kumar, D

A. Kumar, D. S´ en´ echal, and A.-M. S. Tremblay, Cluster dynamical mean-field study of intra-unit-cell charge ne- maticity in hole-doped cuprates, Proceedings of the Na- tional Academy of Sciences122, e2419534122 (2025)

work page 2025

[26] [26]

Kotliar, S

G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko, O. Parcollet, and C. A. Marianetti, Electronic struc- ture calculations with dynamical mean-field theory, Rev. Mod. Phys.78, 865 (2006)

work page 2006

[27] [27]

H. Aoki, N. Tsuji, M. Eckstein, M. Kollar, T. Oka, and P. Werner, Nonequilibrium dynamical mean-field theory and its applications, Rev. Mod. Phys.86, 779 (2014)

work page 2014

[28] [28]

L.-F. m. c. Arsenault, A. Lopez-Bezanilla, O. A. von Lilienfeld, and A. J. Millis, Machine learning for many- body physics: The case of the Anderson impurity model, Phys. Rev. B90, 155136 (2014)

work page 2014

[29] [29]

Rogers, T.-H

J. Rogers, T.-H. Lee, S. Pakdel, W. Xu, V. Do- brosavljevi´ c, Y.-X. Yao, O. Christiansen, and N. Lanat` a, Bypassing the computational bottleneck of quantum- embedding theories for strong electron correlations with machine learning, Physical Review Research3, 013101 (2021)

work page 2021

[30] [30]

Sheridan, C

E. Sheridan, C. Rhodes, F. Jamet, I. Rungger, and C. Weber, Data-driven dynamical mean-field theory: An error-correction approach to solve the quantum many- body problem using machine learning, Phys. Rev. B104, 205120 (2021)

work page 2021

[31] [31]

Agapov, O

E. Agapov, O. Bertomeu, A. Carballo, C. B. Mendl, and A. Sander, Predicting interacting Green’s functions with neural networks, arXiv:2411.13644 (2024)

work page arXiv 2024

[32] [32]

H. Lee, Z. Zhao, G. H. Booth, W. Ge, and C. Weber, Language-inspired machine learning approach for solving strongly correlated problems with dynamical mean-field theory, Phys. Rev. B112, 035165 (2025)

work page 2025

[33] [33]

E. Gull, A. J. Millis, A. I. Lichtenstein, A. N. Rubtsov, M. Troyer, and P. Werner, Continuous-time Monte Carlo methods for quantum impurity models, Rev. Mod. Phys. 83, 349 (2011)

work page 2011

[34] [34]

Liebsch and H

A. Liebsch and H. Ishida, Temperature and bath size in exact diagonalization dynamical mean field theory, Jour- nal of Physics: Condensed Matter24, 053201 (2011)

work page 2011

[35] [35]

Bulla, T

R. Bulla, T. A. Costi, and T. Pruschke, Numerical renor- malization group method for quantum impurity systems, Rev. Mod. Phys.80, 395 (2008)

work page 2008

[36] [36]

Parcollet, M

O. Parcollet, M. Ferrero, T. Ayral, H. Hafermann, I. Krivenko, L. Messio, and P. Seth, TRIQS: A toolbox for research on interacting quantum systems, Computer Physics Communications196, 398 (2015)

work page 2015

[37] [37]

Boehnke, H

L. Boehnke, H. Hafermann, M. Ferrero, F. Lechermann, and O. Parcollet, Orthogonal polynomial representation of imaginary-time Green’s functions, Phys. Rev. B84, 075145 (2011)

work page 2011

[38] [38]

Loshchilov and F

I. Loshchilov and F. Hutter, Decoupled weight decay regularization, inInternational Conference on Learning Representations (ICLR)(2019)

work page 2019

[39] [39]

ISAAC-NG cluster: System overview,https: //oit.utk.edu/hpsc/isaac-open-enclave-new-kpb/ system-overview-cluster-at-kpb/, accessed: 25-03- 2026

work page 2026

[40] [40]

Kotliar, S

G. Kotliar, S. Murthy, and M. J. Rozenberg, Compress- ibility divergence and the finite temperature Mott tran- sition, Phys. Rev. Lett.89, 046401 (2002)

work page 2002

[41] [41]

Bl¨ umer,Mott-Hubbard Metal-Insulator Transition and Optical Conductivity in High Dimensions, Ph.D

N. Bl¨ umer,Mott-Hubbard Metal-Insulator Transition and Optical Conductivity in High Dimensions, Ph.D. thesis, Universit¨ at Augsburg (2003)

work page 2003

[42] [42]

V. M. Galitskii and A. B. Migdal, Application of quan- tum field theory methods to the many-body problem, Zh. Eksp. Teor. Fiz.34, 139 (1958)

work page 1958

[43] [43]

V. M. Galitskii and A. B. Migdal, Application of quan- tum field theory methods to the many-body problem, Sov. Phys. JETP7, 96 (1958)

work page 1958

[44] [44]

Gull and A

E. Gull and A. J. Millis, Energetics of superconductivity in the two-dimensional Hubbard model, Phys. Rev. B86, 241106 (2012)

work page 2012

[45] [45]

E. G. C. P. van Loon, F. Krien, H. Hafermann, E. A. Stepanov, A. I. Lichtenstein, and M. I. Katsnelson, Dou- ble occupancy in dynamical mean-field theory and the dual boson approach, Phys. Rev. B93, 155162 (2016)

work page 2016

[46] [46]

Valenti, I

A. Valenti, I. Park, A. Georges, A. J. Millis, and O. Par- collet, Neural-network quantum embedding solvers for correlated materials, arXiv:2603.15741 (2026)

work page arXiv 2026