Pith. sign in

REVIEW 4 major objections 5 minor 4 cited by

Maximum entropy reconstructions are provably sound only with a near-true prior or near-zero noise — and better priors beat lower noise.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 23:10 UTC pith:TLZKFZLH

load-bearing objection The MSE analysis is genuinely useful, but the paper's headline claim overstates the regime: the 'improved-prior' limit is defined on the estimator, not on the prior, and the abstract's phrasing doesn't follow. the 4 major comments →

arxiv 2511.06915 v1 pith:TLZKFZLH submitted 2025-11-10 physics.comp-ph cond-mat.str-elhep-lat

The noiseless limit and improved-prior limit of the maximum entropy method and their implications for the analytic continuation problem

Thomas Chuna , Nicholas Barnfield , Paul Hamann , Sebastian Schwalbe , Michael P. Friedlander , Tobias Dornheim This is my paper
classification physics.comp-ph cond-mat.str-elhep-lat
keywords analytic continuationmaximum entropy methodBryan's algorithmBayesian priormean squared errornoiseless limitimproved prior limitquantum Monte Carlo
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper investigates when the maximum entropy method (MEM) — the standard tool for analytically continuing noisy imaginary-time quantum Monte Carlo data to real-frequency spectra — is actually justified. The authors show that stochastic sampling algorithms reduce to MEM when the Bayesian prior is close to the true spectral function, establishing the first concrete scenario in which a known mean-field approximation holds. They also show that in this 'improved-prior' limit, the MEM estimate becomes linear, which justifies the use of a widely used fast solver (Bryan's algorithm) that is otherwise controversial. Their mean-squared-error analysis indicates that improving the prior yields quadratically shrinking error, whereas reducing noise alone gives an error controlled by the numerically infinite sum of inverse squared singular values of the kernel. The paper concludes that practitioners should focus on building better data-driven priors rather than only gathering higher-precision data.

Core claim

The central claim is that the maximum entropy estimator (the solution of the entropy-regularized least squares problem) becomes a linear estimator in the improved-prior limit defined by |x̂_i − μ_i|/μ_i ≪ 1, and that this same limit makes a mean-field expansion of stochastic sampling exact, so that stochastic analytic continuation collapses to entropy maximization. In the noiseless limit σ² → 0 the estimator also becomes linear, which explains why Bryan's algorithm — a modified Levenberg-Marquardt method that neglects small singular-value components — is accurate there. The improved-prior limit is distinct: it supplies a finite-variance linearization with bias and covariance that decay quadr

What carries the argument

The key object is the linearized maximum entropy estimator, obtained by Taylor-expanding the logarithmic term ln(x̂/μ) under the assumption that the estimate stays near the prior: x̂ = μ + M A(x0 − μ) + M ε, with M = (AᵀC⁻¹A + α diag(1/μ))⁻¹ AᵀC⁻¹. This linear form yields closed-form bias and covariance formulas. A second mechanism is the mean-field (saddle-point) approximation linking stochastic sampling to MEM: the paper shows that in the improved-prior limit the scaled fluctuations of x_i/μ_i vanish, satisfying the Ginzburg criterion, so the entropy-regularized problem is the exact limit of stochastic sampling.

Load-bearing premise

The paper's central conclusion that 'improve the prior' is the best lever rests on equating a prior close to the true solution with an estimate close to the prior (Eq. 6); if the data noise is large and regularization weak, a good prior does not guarantee |x̂−μ|/μ ≪ 1, and the linearization and its MSE formulas break down.

What would settle it

Set up the double Gaussian problem with the prior exactly at the true solution (c=0) and data noise at the high end (σ0/√N_S = 0.1) with weak regularization (small α). If the resulting estimate violates |x̂−μ|/μ ≪ 1 and the observed MSE deviates from the linearized formula, then the improved-prior analysis does not extend to that regime, contradicting the paper's blanket conclusion (c).

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Bryan's algorithm, which neglects small singular-value basis vectors, is valid precisely in the noiseless and improved-prior limits; in between, other algorithms should outperform it.
  • Stochastic sampling methods need not be run whenever the prior is already good; solving the entropy-regularized problem gives the same answer at lower cost.
  • The MSE formulas give quantitative guidance: error decays quadratically with prior error, but only linearly (via σ²) with noise variance—and for the Laplace kernel the noise path has an essentially infinite prefactor.
  • Data-driven priors (e.g., informed by physical approximations) should be the primary focus for improving analytic continuation, rather than solely increasing QMC sampling accuracy.
  • The noiseless limit does not satisfy the mean-field condition, so stochastic methods and MEM may genuinely differ at finite noise even when data are very accurate.

Where Pith is reading between the lines

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

  • The linearization condition |x̂−μ|/μ ≪ 1 is an assumption about the estimate, while the paper's conclusions are phrased about the prior being near the true solution; these coincide only when the data are sufficiently informative or regularization strong. A testable extension would be to bound the regime (in terms of noise and α) where prior improvements actually produce the promised quadratic MSE
  • The paper's MSE analysis suggests a general principle for inverse problems with entropic regularization: prior engineering is a more powerful lever than noise reduction, which may transfer to other ill-posed problems (e.g., deconvolution, tomography) using Bayesian priors.
  • Iterating MEM, using each solution as the next prior, failed historically because bias accumulates; the improved-prior analysis clarifies this: only a prior closer to the truth than the current estimate helps, so iterative schemes need an external source of prior improvement.
  • The dual formulation used in the numerics may allow full-basis solutions, making it a natural tool to test the linearity regime and to detect when Bryan's algorithm is unreliable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper studies the maximum entropy method (MEM) for analytic continuation, focusing on two regimes: the noiseless limit and the 'improved-prior' limit. It derives linearized mean-squared-error formulas, argues that Bryan's algorithm is valid when the MEM estimator becomes linear, presents numerical experiments on the double-Gaussian benchmark, and claims that the improved-prior limit satisfies Beach's mean-field condition, thereby reducing stochastic sampling to MEM. The headline conclusions are that stochastic sampling reduces to MEM when the prior is near the truth, Bryan's algorithm is then valid, and improving the prior is a better return on investment than reducing noise.

Significance. If the claims were established as stated, the paper would give valuable practical guidance: MEM and Bryan's algorithm are trustworthy precisely in the noiseless and improved-prior limits, and prior engineering is the highest-leverage improvement. The explicit MSE formulas in Eqs. (13)-(18) and the numerical double-Gaussian study are useful contributions, and the connection to Bryan's linearity assumption is a helpful clarification. However, the main analytic results are proven for a condition on the estimator, Eq. (6), not for a prior close to the true solution; this gap affects all three headline conclusions. The multinomial covariance assumption in Sec. III also limits the reach of conclusion (a). With a clear restatement of the hypotheses and a controlled comparison of the two limits, the results would be a credible and useful contribution.

major comments (4)
  1. The improved-prior limit is defined by |x_hat_i - mu_i|/mu_i << 1, i.e., the estimator is close to the prior, not the prior close to the true solution. Every subsequent formula, Eqs. (10)-(18), and the Section III argument require this condition. From Eq. (11), if mu = x0 then x_hat - mu = M epsilon, so Eq. (6) becomes a constraint on the noise level, alpha, and the conditioning of A, not a consequence of ||x0 - mu|| small. The numerical study in Sec. II.D.1 restricts c to [0.05, 1.0] because for very small c 'the ITCF data may actually make the result worse'; this is exactly the regime where the prior is close to x0 but the estimator does not track the prior. Thus conclusions (a)-(c) as phrased in the abstract and Sec. IV are not established. The authors should either prove a regularity condition under which prior closeness implies Eq. (6), or rephrase the claims as 'when the MEM estima
  2. The 'better scaling' of the improved-prior MSE is not a controlled asymptotic comparison. The noiseless MSE (9) is evaluated as sigma^2 -> 0 with alpha fixed; the improved-prior formulas (13)-(18) are evaluated as |x_hat - mu|/mu -> 0. When sigma^2 is also small in the improved-prior regime, Eq. (18) reduces to Eq. (9), as the paper notes, so the numerically large sigma^2 Tr Sigma^{-2} term reappears. The statement that M is windowed by alpha diag(1/mu) and hence avoids small singular values holds only away from the sigma -> 0 limit; no quantitative regime supporting 'better return on investment' is given. Provide explicit asymptotic dependence on c (or ||x0 - mu||) and sigma.
  3. The reduction of stochastic sampling to MEM rests on the multinomial covariance model (25) and on identifying x_hat with the MEM solution. This is a model-level argument, not a derivation for a concrete stochastic continuation algorithm; the 'intuitive' statement around Eq. (26) that a sampler initialized near the prior makes only small perturbations is not a proof. Since conclusion (a) is a headline claim, the paper should either provide a more direct derivation for an actual stochastic algorithm or state explicitly that the claim is conditional on the multinomial/saddle-point assumptions.
  4. The noiseless-limit derivation drops the entropy term from Eq. (7). The correct limiting condition is sigma^2 alpha -> 0, not simply sigma -> 0; the authors note that chi^2-kink might choose alpha -> infinity but only give a numerical observation for one problem. Also, Eq. (9) requires A^T A to be invertible, while the AC kernel is severely ill-conditioned and may be rank-deficient on the chosen grid; calling Tr Sigma^{-2} 'numerically infinite' is not the same as an MSE statement. This caveat should be stated because it underlies conclusion (c).
minor comments (5)
  1. The notation ||x0 - mu||_H^2 is used without defining the H-weighted norm. Please define H = M A and the associated inner product.
  2. 'Ginzberg criterion' should be 'Ginzburg criterion.'
  3. The axis label '||xdual N. xBryan||' appears to be missing an operator; it should read ||x_dual - x_Bryan||.
  4. 'Gunnarson' should be 'Gunnarsson.'
  5. The analytic formulas (8) and (18) assume C = sigma^2 I, while the numerical noise model (21) is heteroscedastic. A sentence reconciling this mismatch would improve the connection between the theory and the numerical claims.

Circularity Check

1 steps flagged

Central claims inherit a definitional conflation: the 'improved-prior limit' is defined by |x̂−μ|/μ small, not by prior closeness to the true solution.

specific steps
  1. self definitional [Section II.A, Eq. (6), and Section IV Conclusions]
    "and the limit that the Bayesian prior is near the true solution, which we refer to as the “improved-prior” limit |x̂_i−µ_i|/µ_i ≪ 1."

    The limit is defined on the estimator-vs-prior deviation |x̂_i−μ_i|/μ_i, not on the prior-to-truth distance ‖μ−x0‖. All improved-prior results—linearized estimator (10)–(11), bias/variance formulas (13)–(18), and Section III’s reduction of (25) to M^{−1}—use only Eq. (6). Yet the abstract and conclusions restate the condition as 'the Bayesian prior is near the true solution.' The paper itself concedes the gap: for very small c (prior closest to x0) it says 'the ITCF data may actually make the result worse' and therefore restricts c∈[0.05,1.0]. Thus the claimed conclusions (a)–(c) follow from the estimator being close to the prior, not from the prior being close to the truth, so the stated premise is the definition of the limit rather than an independently established condition.

full rationale

The analytic derivations are not circular in the fitted-parameter sense: Eq. (10) is a Taylor expansion about μ, Eq. (11) solves the linearized optimality condition, and the MSE formulas follow algebraically. No fitted value is renamed as a prediction, and the self-citations (dual MEM [48], data-driven priors [62,63]) are supporting rather than load-bearing for the main limits. The genuine issue is a definitional conflation at the center of the paper’s framing. The 'improved-prior limit' is introduced with language that identifies it with 'the Bayesian prior is near the true solution,' but the operational condition (6) is about the estimator being close to the prior. Since the derivations rely only on (6), the conclusions as stated about prior closeness to x0 do not follow; the paper’s own exclusion of very small c acknowledges that prior closeness to truth alone does not ensure (6). This is a significant load-bearing flaw, but the underlying calculus of the expansions is self-contained, so the overall circularity score is moderate rather than extreme.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

No new physical entities are introduced. The analysis relies on standard linear algebra and cited methodological premises; the two load-bearing assumptions are Rothkopf's linearity criterion for Bryan's algorithm and the multinomial covariance model for stochastic sampling in Section III.

free parameters (1)
  • regularization weight α = selected by χ²-kink algorithm (not fixed analytically)
    The MSE formulas and the improved-prior/noiseless limits depend on α; the paper never derives α from first principles, so all quantitative MSE results are conditional on the data-driven choice of α. In the noiseless limit the derivation requires σ²α→0, which is verified numerically only for the double-Gaussian problem.
axioms (4)
  • domain assumption Rothkopf's criterion: Bryan's algorithm is valid iff the MEM estimator is linear.
    Invoked in Section IV(b) and the introduction to transfer linearity of the estimator into validity of Bryan's null-space trick. The paper does not prove or derive this criterion; it cites Rothkopf [44].
  • ad hoc to paper Multinomial covariance model for stochastic sampling bin counts: Cov(x_i,x_j) = −x̂_i x̂_j / M.
    Section III, after Eq. (24), 'Following Asakawa, Hatsuda, Nakahara's monkey argument, we assume the covariance between x_i and x_j is described by a multinomial distribution.' This assumption is necessary to conclude that the improved-prior limit satisfies Beach's mean-field approximation; no derivation from a concrete stochastic AC algorithm is given.
  • domain assumption Asymptotic σ²α→0 in the noiseless limit.
    Section II.B drops the α ln(x̂/μ) term from Eq. (7) to obtain the OLS estimate. The paper checks numerically for one problem that the χ²-kink choice of α does not violate this, but there is no general proof.
  • standard math Gauss-Markov theorem: the OLS estimate is BLUE.
    Used in Section II.B to identify the covariance of the noiseless-limit estimator.

pith-pipeline@v1.3.0-alltime-deepseek · 12031 in / 18158 out tokens · 181182 ms · 2026-08-03T23:10:05.598791+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of The noiseless limit and improved-prior limit of the maximum entropy method and their implications for the analytic continuation problem." pith.science (2026). https://pith.science/paper/TLZKFZLH

@misc{pith2026251106915,
  author       = {Pith},
  title        = {Pith review of: The noiseless limit and improved-prior limit of the maximum entropy method and their implications for the analytic continuation problem},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TLZKFZLH}},
  note         = {Machine review of arXiv:2511.06915}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Quantum Monte Carlo (QMC) methods are uniquely capable of providing exact simulations of quantum many-body systems. Unfortunately, the applications of a QMC simulation are limited because extracting dynamic properties requires solving the analytic continuation (AC) problem. Across the many fields that use QMC methods, there is no universally accepted analytic continuation algorithm for extracting dynamic properties, but many publications compare to the maximum entropy method. We investigate when entropy maximization is an acceptable approach. We show that stochastic sampling algorithms reduce to entropy maximization when the Bayesian prior is near to the true solution. We investigate when is Bryan's controversial optimization algorithm [Bryan, Eur. Biophys. J. 18, 165-174 (1990)] for entropy maximization (sometimes known as the maximum entropy method) appropriate to use. We show that Bryan's algorithm is appropriate when the noise is near zero or when the Bayesian prior is near to the true solution. We also investigate the mean squared error, finding a better scaling when the Bayesian prior is near the true solution than when the noise is near zero. We point to examples of improved data-driven Bayesian priors that have already leveraged this advantage. We support these results by solving the double Gaussian problem using both Bryan's algorithm and the newly formulated dual approach to entropy maximization [Chuna et al., J. Phys. A: Math. Theor. 58, 335203 (2025)].

Figures

Figures reproduced from arXiv: 2511.06915 by Michael P. Friedlander, Nicholas Barnfield, Paul Hamann, Sebastian Schwalbe, Thomas Chuna, Tobias Dornheim.

Figure 1
Figure 1. Figure 1: FIG. 1. Plot of the double Gaussian test problem from [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Heatmap of the mean squared error (MSE) over var [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Heatmap of the average relative distance between [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Monte-Carlo Event Generation for X-Ray Thomson Scattering Analysis

    hep-ph 2026-04 unverdicted novelty 7.0

    A proof-of-principle event-driven Monte-Carlo framework samples XRTS events from differential cross sections and passes them through detector simulations to preserve kinematic information and support model-agnostic analysis.

  2. Model-free interpretation of X-ray Thomson scattering measurements

    physics.plasm-ph 2026-04 unverdicted novelty 2.0

    The paper reviews the use of the imaginary-time correlation function to extract temperature, normalization, and Rayleigh weight from XRTS spectra without model dependence.

  3. Overview of X-ray Thomson scattering measurements of extreme states of matter

    physics.plasm-ph 2026-04 unverdicted novelty 2.0

    XRTS has become a leading diagnostic for extreme states of matter, and this review compiles prior experiments, analysis methods, and future directions.

  4. Quantum effects in plasmas

    physics.plasm-ph 2026-04 unverdicted novelty 2.0

    Quantum effects govern behavior in warm dense matter and inertial fusion plasmas and are best modeled by combining quantum methods through downfolding from first-principles simulations.

Reference graph

Works this paper leans on

72 extracted references · 6 linked inside Pith · cited by 4 Pith papers

  1. [1]

    In particular, we study synthetic data produced from the double Gaussian problem presented in Goulkoet al

    Description of problem We present numeric tests of the MSE. In particular, we study synthetic data produced from the double Gaussian problem presented in Goulkoet al. [51]: x0 = 2 ∑ i=1 ci σi e − (ω−zi )2 2σ2 i ,(19) where the moments are defined,c 1 =0.62;σ 1 =0.12;z 1 = 0.74 andc 2 =0.41;σ 2 =0.064;z 2 =2.93, the transforma- tion kernel is defined as A=...

  2. [2]

    Each plot in Figure 1, shows the estimate obtained for different quality priorsc=1.0,0.7,0.5

    Numeric Results We investigate the double Gaussian problem using the dual formulation of the entropic regularization [48]. Each plot in Figure 1, shows the estimate obtained for different quality priorsc=1.0,0.7,0.5. For each prior, we show the estimate converging towards the true solution in the noiseless limit in Figure 1. We see that a flat priorc=1.0 ...

    2022

  3. [3]

    W. M. C. Foulkes, L. Mitas, R. J. Needs, and G. Ra- jagopal, Quantum monte carlo simulations of solids, Rev. Mod. Phys.73, 33 (2001)

    2001

  4. [4]

    Anderson,Quantum Monte Carlo: Origins, Devel- opment, Applications(Oxford University Press, USA, 2007)

    J. Anderson,Quantum Monte Carlo: Origins, Devel- opment, Applications(Oxford University Press, USA, 2007)

    2007

  5. [5]

    D. M. Ceperley, Path integrals in the theory of condensed helium, Rev. Mod. Phys67, 279 (1995)

    1995

  6. [6]

    Boninsegni, N

    M. Boninsegni, N. V. Prokofev, and B. V. Svistunov, Worm algorithm and diagrammatic Monte Carlo: A new approach to continuous-space path integral Monte Carlo simulations, Phys. Rev. E74, 036701 (2006)

    2006

  7. [7]

    Rabani, D

    E. Rabani, D. R. Reichman, G. Krilov, and B. J. Berne, The calculation of transport properties in quantum liquids using the maximum entropy nu- merical analytic continuation method: Applica- tion to liquid para-hydrogen, Proceedings of the National Academy of Sciences99, 1129 (2002), 7 https://www.pnas.org/doi/pdf/10.1073/pnas.261540698

    doi:10.1073/pnas.261540698 2002

  8. [8]

    Dornheim, Z

    T. Dornheim, Z. Moldabekov, P. Tolias, M. B¨ ohme, and J. Vorberger, Physical insights from imaginary-time density–density correlation functions, Matter and Radi- ation at Extremes8, 056601 (2023)

    2023

  9. [9]

    Dornheim, Z

    T. Dornheim, Z. A. Moldabekov, and J. Vorberger, Non- linear density response from imaginary-time correlation functions: Ab initio path integral monte carlo simulations of the warm dense electron gas, The Journal of Chemical Physics155, 054110 (2021)

    2021

  10. [10]

    Jarrell and J

    M. Jarrell and J. Gubernatis, Bayesian inference and the analytic continuation of imaginary-time quantum monte carlo data, Physics Reports269, 133 (1996)

    1996

  11. [11]

    S. H. Glenzer and R. Redmer, X-ray thomson scattering in high energy density plasmas, Rev. Mod. Phys81, 1625 (2009)

    2009

  12. [12]

    Dornheim, Z

    T. Dornheim, Z. A. Moldabekov, K. Ramakrishna, P. To- lias, A. D. Baczewski, D. Kraus, T. R. Preston, D. A. Chapman, M. P. B¨ ohme, T. D¨ oppner, F. Graziani, M. Bonitz, A. Cangi, and J. Vorberger, Electronic den- sity response of warm dense matter, Physics of Plasmas 30, 032705 (2023)

    2023

  13. [13]

    Sch¨ orner, M

    M. Sch¨ orner, M. Bethkenhagen, T. D¨ oppner, D. Kraus, L. B. Fletcher, S. H. Glenzer, and R. Redmer, X-ray thomson scattering spectra from density functional the- ory molecular dynamics simulations based on a modified chihara formula, Phys. Rev. E107, 065207 (2023)

    2023

  14. [14]

    D. S. Bespalov, U. Zastrau, Z. A. Moldabekov, T. Gawne, T. Dornheim, M. Meshhal, A. Amouretti, M. Andrze- jewski, K. Appel, C. Baehtz, E. Brambrink, K. Buakor, C. Camarda, D. Chin, G. Collins, C. Crepisson, A. Descamps, J. Eggert, L. Fletcher, A. Forte, G. Gre- gori, M. Harmand, O. S. Humphries, H. Hoeppner, J. Kuhlke, W. Lynn, J. Luetgert, M. Masruri, E....

    Pith/arXiv arXiv 2025

  15. [15]

    Filinov and M

    A. Filinov and M. Bonitz, Collective and single-particle excitations in two-dimensional dipolar bose gases, Phys. Rev. A86, 043628 (2012)

    2012

  16. [16]

    Dornheim, Z

    T. Dornheim, Z. A. Moldabekov, J. Vorberger, and B. Militzer, Path integral monte carlo approach to the structural properties and collective excitations of liq- uid 3he without fixed nodes, Scientific Reports12, 708 (2022)

    2022

  17. [17]

    Boninsegni and D

    M. Boninsegni and D. M. Ceperley, Density fluctuations in liquid 4He. Path integrals and maximum entropy, Jour- nal of Low Temperature Physics104, 339 (1996)

    1996

  18. [18]

    Ferr´ e and J

    G. Ferr´ e and J. Boronat, Dynamic structure factor of liquid 4He across the normal-superfluid transition, Phys. Rev. B93, 104510 (2016)

    2016

  19. [19]

    Asakawa, Y

    M. Asakawa, Y. Nakahara, and T. Hatsuda, Maximum entropy analysis of the spectral functions in lattice qcd, Progress in Particle and Nuclear Physics46, 459 (2001)

    2001

  20. [20]

    Rothkopf, Heavy quarkonium in extreme conditions, Physics Reports858, 1 (2020)

    A. Rothkopf, Heavy quarkonium in extreme conditions, Physics Reports858, 1 (2020)

    2020

  21. [21]

    Yoon, J.-H

    H. Yoon, J.-H. Sim, and M. J. Han, Analytic continuation via domain knowledge free machine learning, Phys. Rev. B98, 245101 (2018)

    2018

  22. [22]

    Fournier, L

    R. Fournier, L. Wang, O. V. Yazyev, and Q. Wu, Artifi- cial neural network approach to the analytic continuation problem, Phys. Rev. Lett.124, 056401 (2020)

    2020

  23. [23]

    X. Xie, F. Bao, T. Maier, and C. Webster, Analytic continuation of noisy data using adams bashforth resid- ual neural network, Discrete and Continuous Dynamical Systems-Series S15(2021)

    2021

  24. [24]

    L. Wang, S. Shi, and K. Zhou, Reconstructing spectral functions via automatic differentiation, Physical Review D106, L051502 (2022)

    2022

  25. [25]

    L. Wang, S. Shi, and K. Zhou, Unsupervised learning spectral functions with neural networks, inJournal of Physics: Conference Series, Vol. 2586 (IOP Publishing,

  26. [26]

    Aarts, K

    G. Aarts, K. Fukushima, T. Hatsuda, A. Ipp, S. Shi, L. Wang, and K. Zhou, Physics-driven learning for in- verse problems in quantum chromodynamics, Nature Re- views Physics7, 154 (2025)

    2025

  27. [27]

    Groth, T

    S. Groth, T. Dornheim, and J. Vorberger, Ab initio path integral monte carlo approach to the static and dynamic density response of the uniform electron gas, Physical Review B99, 235122 (2019)

    2019

  28. [28]

    S. Shi, L. Wang, and K. Zhou, Rethinking the ill- posedness of the spectral function reconstruction—why is it fundamentally hard and how artificial neural net- works can help, Computer Physics Communications282, 108547 (2023)

    2023

  29. [29]

    Dornheim, S

    T. Dornheim, S. Groth, J. Vorberger, and M. Bonitz, Ab initio path integral monte carlo results for the dynamic structure factor of correlated electrons: From the electron liquid to warm dense matter, Physical Review Letters 121, 255001 (2018)

    2018

  30. [30]

    Vitali, M

    E. Vitali, M. Rossi, L. Reatto, and D. E. Galli, Ab initio low-energy dynamics of superfluid and solid 4He, Phys. Rev. B82, 174510 (2010)

    2010

  31. [31]

    N. V. Prokof’ev and B. V. Svistunov, Spectral analysis by the method of consistent constraints, JETP letters97, 649 (2013)

    2013

  32. [32]

    Saccani, S

    S. Saccani, S. Moroni, and M. Boninsegni, Excitation spectrum of a supersolid, Phys. Rev. Lett.108, 175301 (2012)

    2012

  33. [33]

    N. S. Nichols, P. Sokol, and A. Del Maestro, Parameter- free differential evolution algorithm for the analytic con- tinuation of imaginary time correlation functions, Phys. Rev. E106, 025312 (2022)

    2022

  34. [34]

    F. Bao, Y. Tang, M. Summers, G. Zhang, C. Webster, V. Scarola, and T. A. Maier, Fast and efficient stochastic optimization for analytic continuation, Physical Review B94, 125149 (2016)

    2016

  35. [35]

    Shao and A

    H. Shao and A. W. Sandvik, Progress on stochastic ana- lytic continuation of quantum monte carlo data, Physics Reports1003, 1 (2023)

    2023

  36. [36]

    Shu, H.-T

    H.-T. Shu, H.-T. Ding, O. Kaczmarek, S. Mukherjee, and H. Ohno, A stochastic approach to the reconstruc- tion of spectral functions in lattice qcd, arXiv preprint arXiv:1510.02901 (2015)

    Pith/arXiv arXiv 2015

  37. [37]

    A. S. Mishchenko, N. V. Prokof’ev, A. Sakamoto, and B. V. Svistunov, Diagrammatic quantum monte carlo study of the fr¨ ohlich polaron, Phys. Rev. B62, 6317 (2000)

    2000

  38. [38]

    Otsuki, M

    J. Otsuki, M. Ohzeki, H. Shinaoka, and K. Yoshimi, 8 Sparse modeling approach to analytical continuation of imaginary-time quantum monte carlo data, Phys. Rev. E 95, 061302 (2017)

    2017

  39. [39]

    Otsuki, M

    J. Otsuki, M. Ohzeki, H. Shinaoka, and K. Yoshimi, Sparse modeling in quantum many-body problems, Jour- nal of the Physical Society of Japan89, 012001 (2020), https://doi.org/10.7566/JPSJ.89.012001

    doi:10.7566/jpsj.89.012001 2020

  40. [40]

    A. B. Robles, P.-A. Hofmann, T. Chuna, T. Dornheim, and M. Hecht, Pylit: Reformulation and implementa- tion of the analytic continuation problem using kernel representation methods, arXiv preprint arXiv:2505.10211 (2025)

    arXiv 2025

  41. [41]

    Boninsegni and D

    M. Boninsegni and D. M. Ceperley, Density fluctuations in liquid4he. path integrals and maximum entropy, Jour- nal of Low Temperature Physics104, 339 (1996)

    1996

  42. [42]

    Kora and M

    Y. Kora and M. Boninsegni, Dynamic structure factor of superfluid 4He from quantum monte carlo: Maximum entropy revisited, Phys. Rev. B98, 134509 (2018)

    2018

  43. [43]

    Beach, Identifying the maximum entropy method as a special limit of stochastic analytic continuation, arXiv preprint cond-mat/0403055 (2004)

    K. Beach, Identifying the maximum entropy method as a special limit of stochastic analytic continuation, arXiv preprint cond-mat/0403055 (2004)

    Pith/arXiv arXiv 2004

  44. [44]

    Bryan, Maximum entropy analysis of oversampled data problems, European Biophysics Journal18, 165 (1990)

    R. Bryan, Maximum entropy analysis of oversampled data problems, European Biophysics Journal18, 165 (1990)

    1990

  45. [45]

    S. F. Gull, Developments in maximum entropy data analysis, inMaximum Entropy and Bayesian Methods (Springer, 1989) pp. 53–71

    1989

  46. [46]

    Rothkopf, Bryan’s maximum entropy method—diagnosis of a flawed argument and its remedy, Data5, 85 (2020)

    A. Rothkopf, Bryan’s maximum entropy method—diagnosis of a flawed argument and its remedy, Data5, 85 (2020)

    2020

  47. [47]

    heavy quarkonium in ex- treme conditions

    M. Asakawa, Comment on “heavy quarkonium in ex- treme conditions”’, arXiv preprint arXiv:2001.10205 (2020)

    Pith/arXiv arXiv 2001

  48. [48]

    Rothkopf, Improved maximum entropy method with an extended search space, arXiv preprint arXiv:1208.5162 (2012)

    A. Rothkopf, Improved maximum entropy method with an extended search space, arXiv preprint arXiv:1208.5162 (2012)

    Pith/arXiv arXiv 2012

  49. [49]

    Kelly, A

    A. Kelly, A. Rothkopf, and J.-I. Skullerud, Bayesian study of relativistic open and hidden charm in anisotropic lattice qcd, Physical Review D97, 114509 (2018)

    2018

  50. [50]

    Chuna, N

    T. Chuna, N. Barnfield, T. Dornheim, M. P. Friedlan- der, and T. Hoheisel, Dual formulation of the maximum entropy method applied to analytic continuation of quan- tum monte carlo data, Journal of Physics A: Mathemat- ical and Theoretical58, 335203 (2025)

    2025

  51. [51]

    Fuchs, T

    S. Fuchs, T. Pruschke, and M. Jarrell, Analytic continu- ation of quantum monte carlo data by stochastic analyti- cal inference, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics81, 056701 (2010)

    2010

  52. [52]

    Huang, Acflow: An open source toolkit for analytic continuation of quantum monte carlo data, Computer Physics Communications292, 108863 (2023)

    L. Huang, Acflow: An open source toolkit for analytic continuation of quantum monte carlo data, Computer Physics Communications292, 108863 (2023)

    2023

  53. [53]

    Goulko, A

    O. Goulko, A. S. Mishchenko, L. Pollet, N. Prokof’ev, and B. Svistunov, Numerical analytic continuation: An- swers to well-posed questions, Phys. Rev. B95, 014102 (2017)

    2017

  54. [54]

    Bergeron and A.-M

    D. Bergeron and A.-M. Tremblay, Algorithms for opti- mized maximum entropy and diagnostic tools for analytic continuation, Physical Review E94, 023303 (2016)

    2016

  55. [55]

    Gunnarsson, M

    O. Gunnarsson, M. Haverkort, and G. Sangiovanni, An- alytical continuation of imaginary axis data using max- imum entropy, Physical Review B—Condensed Matter and Materials Physics81, 155107 (2010)

    2010

  56. [56]

    Fei, C.-N

    J. Fei, C.-N. Yeh, and E. Gull, Nevanlinna analytical con- tinuation, Physical Review Letters126, 056402 (2021)

    2021

  57. [57]

    S. Kim, P. Petreczky, and A. Rothkopf, Quarkonium in- medium properties from realistic lattice nrqcd, Journal of High Energy Physics2018, 1 (2018)

    2018

  58. [58]

    James, D

    G. James, D. Witten, T. Hastie, R. Tibshirani,et al., An introduction to statistical learning, Vol. 112 (Springer, 2013)

    2013

  59. [59]

    Yeh, Bias-variance decomposition of mean squared error,https://chrisyeh96.github.io/2019/07/18/ bias-variance-decomposition.html(2019), last updated December 5, 2022

    C. Yeh, Bias-variance decomposition of mean squared error,https://chrisyeh96.github.io/2019/07/18/ bias-variance-decomposition.html(2019), last updated December 5, 2022. Accessed: 2025-10-29

    2019

  60. [60]

    J. J. Meyer, The multivariate bias-variance decom- position,https://johannesjakobmeyer.com/blog/ 005-multivariate-bias-variance-decomposition/ (2020), accessed: 2025-10-29

    2020

  61. [61]

    Kaufmann and K

    J. Kaufmann and K. Held, ana cont: Python package for analytic continuation, Computer Physics Communi- cations282, 108519 (2023)

    2023

  62. [62]

    Landau and K

    D. Landau and K. Binder,A guide to Monte Carlo simu- lations in statistical physics(Cambridge university press, 2021)

    2021

  63. [63]

    M. E. Tuckerman,Statistical mechanics: theory and molecular simulation(Oxford university press, 2023)

    2023

  64. [64]

    Chuna, N

    T. Chuna, N. Barnfield, J. Vorberger, M. P. Friedlander, T. Hoheisel, and T. Dornheim, Estimates of the dynamic structure factor for the finite temperature electron liquid via analytic continuation of path integral monte carlo data, Physical Review B112, 125112 (2025)

    2025

  65. [65]

    T. M. Chuna, J. Vorberger, P. Tolias, A. Benedix Rob- les, M. Hecht, P.-A. Hofmann, Z. A. Moldabekov, and T. Dornheim, Second roton feature in the strongly cou- pled electron liquid, The Journal of Chemical Physics 163(2025)

    2025

  66. [66]

    Dornheim, J

    T. Dornheim, J. Vorberger, S. Groth, N. Hoffmann, Z. Moldabekov, and M. Bonitz, The static local field cor- rection of the warm dense electron gas: An ab initio path integral Monte Carlo study and machine learning repre- sentation, J. Chem. Phys151, 194104 (2019)

    2019

  67. [67]

    Dornheim, Z

    T. Dornheim, Z. A. Moldabekov, and P. Tolias, Analyti- cal representation of the local field correction of the uni- form electron gas within the effective static approxima- tion, Physical Review B103, 165102 (2021)

    2021

  68. [68]

    Giuliani and G

    G. Giuliani and G. Vignale,Quantum Theory of the Elec- tron Liquid(Cambridge University Press, Cambridge, 2008)

    2008

  69. [69]

    Z. Wu, N. E. Huang, S. R. Long, and C.-K. Peng, On the trend, detrending, and variability of nonlinear and nonstationary time series, Proceedings of the National Academy of Sciences104, 14889 (2007)

    2007

  70. [70]

    P. J. Brockwell and R. A. Davis,Introduction to time series and forecasting(Springer, 2002)

    2002

  71. [71]

    Adhikari and R

    R. Adhikari and R. K. Agrawal, An introductory study on time series modeling and forecasting, arXiv preprint arXiv:1302.6613 (2013)

    Pith/arXiv arXiv 2013

  72. [72]

    B. A. Berg,Markov chain Monte Carlo simulations and their statistical analysis: with web-based Fortran code (World Scientific Publishing Company, 2004)

    2004