Pith. sign in

REVIEW 3 major objections 6 minor 2 cited by

Scaling of Stochastic Normalizing Flows in $\mathrm{SU}(3)$ lattice gauge theory

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Stochastic normalizing flows for SU(3) lattice gauge theory inherit a linear-in-volume scaling from non-equilibrium Monte Carlo.

desk verdict First credible SNF implementation in 4D SU(3) with a scaling claim that mostly holds, but the unquantified transfer-learning step and missing error bars keep it conditional. read the letter →

arxiv 2412.00200 v3 pith:A7P5GJO7 submitted 2024-11-29 hep-lat cond-mat.stat-mechcs.LGstat.ML

Andrea Bulgarelli , Elia Cellini , Alessandro Nada This is my paper · ORCID
classification hep-latcond-mat.stat-mechcs.LGstat.ML PACS 11.15.Ha05.10.Ln12.38.Gc
keywords StochasticNormalizingFlowslatticegaugetheorySU(3)Yang-Millsnon-equilibriumMonteCarloJarzynskiequalitycriticalslowingdowngauge-equivariantcouplinglayersscalingwithvolume
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces Stochastic Normalizing Flows (SNFs) to SU(3) lattice gauge theory in four dimensions: an architecture that interleaves trainable gauge-equivariant maps with out-of-equilibrium Monte Carlo updates. It aims to show that the sampling quality of both plain non-equilibrium Monte Carlo and the trained SNFs is controlled by a single ratio, nstep/(L/a)^4, where nstep is the number of updates and (L/a)^4 is the number of lattice sites. If true, the cost of reaching a fixed sampling quality grows linearly with the number of degrees of freedom, and SNFs inherit this favorable scaling while adding a roughly factor-two efficiency gain over the stochastic baseline. The authors also show that the smearing parameters learned on short flows transfer to longer flows and other volumes, so the training cost stays a small fraction of the total sampling cost. This matters because a sampler with controlled scaling would give a practical route to fine lattice spacings, where equilibrium Markov chains suffer critical slowing down.

What carries the argument

The carrying mechanism is the Stochastic Normalizing Flow built from gauge-equivariant stout-smearing coupling layers. Each layer transforms a subset of links via U' = exp(iQ)U, where Q is built from staples of frozen links with one learned smearing parameter per layer, and is interleaved with one heatbath plus four over-relaxation updates. Crooks' theorem provides the precise bookkeeping: the work of a full evolution is W = S - S0 - Q - log J, with log J the sum of Jacobian logarithms of the layers, so training the parameters by minimizing the average dissipated work is equivalent to minimizing the KL divergence between forward and reverse evolutions. The layer-wise training objective makes memory use independent of nstep, and the observed collapse of the learned parameters when plotted against n/nstep justifies transferring them to other step counts and volumes.

What would settle it

A direct comparison of SNFs at L/a = 20 and nstep = 512 (or larger) using transferred parameters from nstep = 64 versus parameters trained at that volume and step count; if the KL divergence or effective sample size differ beyond statistical errors, the transfer assumption and the reported training-cost savings would fail. Equally, computing the two metrics at fixed nstep for a sequence of volumes and finding a deviation from the single-curve collapse would falsify the central scaling claim.

Watch

Extended reading notes

Core claim

The central claim is that the KL divergence between forward and reverse non-equilibrium evolutions, and the effective sample size, do not depend separately on the number of steps nstep and the volume, but only on the ratio nstep/(L/a)^4. This collapse onto a single curve is demonstrated for NE-MCMC over lattice sizes L/a = 10, 12, 16, 20, and the same collapse is inherited by the trained SNFs, even though deterministic gauge-equivariant layers sit between the Monte Carlo updates. The paper further claims that, at equal nstep, SNFs reach the same metric values at roughly half the number of steps of NE-MCMC, and that the layer parameters trained only for nstep = 16, 32, 64 (on L/a = 20 for the main ensembles) can be interpolated and transferred to larger nstep and smaller volumes with no retraining. This makes the SNF roughly twice as efficient at an overhead of about 25% per update, and it grounds the assertion that the architecture scales linearly with the degrees of freedom of the system.

Load-bearing premise

The paper's cost claims rely on the assumption that smearing parameters trained only on short flows (nstep = 16, 32, 64) and on one lattice size can be interpolated and reused for longer flows and other volumes without retraining, a compatibility the authors describe as observed but do not quantify.

Editorial extensions

If this is right

  • For a fixed target KL divergence or effective sample size, the required number of updates grows as (L/a)^4, so the total sampling cost grows linearly with the number of degrees of freedom.
  • SNFs keep the same scaling as NE-MCMC while needing about half the updates, so at fixed quality they are roughly a factor two cheaper, even after accounting for the smearing overhead.
  • The transfer of smearing parameters from cheap short flows means the training expenditure is a negligible part of the cost for large nstep, making the method practical without full retraining.
  • The same ratio collapse appears for the coarser-spacing ensemble, so the scaling is not an artifact of a single coupling range.
  • For linear protocols in the inverse coupling, keeping the KL divergence fixed requires nstep proportional to (β − β0)^2, quantifying how the cost grows when targeting finer lattice spacings.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: The transfer-learning result points to a universal protocol structure: the learned stout parameters collapse to a single curve when plotted against n/nstep, suggesting the optimal driving schedule may be a property of the coupling change alone, independent of volume.
  • Inference: A testable extension would be to train on a single small volume and reuse parameters on a larger volume for a boundary-condition protocol (switching from open to periodic boundary conditions); because the modified degrees of freedom form a three-dimensional surface, the expected scaling would be nstep ∝ (L/a)^3, potentially cheaper than the β-shift studied here.
  • Inference: Another unstated consequence is that the factor-two advantage and the ratio scaling may erode as the flow becomes more expressive (e.g., neural-network-parameterized smearing) and nstep becomes small; the authors flag this as future work, and it is a natural stress test of the linear-cost claim.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper reports the first implementation of Stochastic Normalizing Flows (SNFs) for four-dimensional SU(3) lattice gauge theory. The architecture interleaves gauge-equivariant stout-smearing coupling layers with non-equilibrium Monte Carlo updates, and the training minimizes the dissipated work, i.e. the KL divergence between forward and reverse path distributions. The central empirical claim is that both NE-MCMC and the trained SNFs have sampling-quality metrics that depend on the number of steps and the lattice volume only through the ratio nstep/(L/a)^4, and that SNFs reach the same KL divergence or ESS with roughly half the number of steps. The paper also reports a scaling of the KL divergence with the squared change in β for linear protocols and analyzes the work distributions underlying the Jarzynski estimator.

Significance. If the scaling claim holds, the computational cost of reaching a fixed sampling quality grows only linearly with the number of lattice degrees of freedom, making SNFs a potentially practical tool for large-volume SU(3) simulations. The paper's theoretical framework, which generalizes Crooks' theorem to include deterministic gauge-equivariant layers, is carefully laid out, and the empirical data cover two ensembles and four volumes. The transfer-learning strategy, if quantitatively validated, would be a notable practical advance because it avoids retraining at every nstep and volume. However, the two main empirical pillars, the scaling collapse and the factor-of-two improvement, are currently supported only by qualitative visual inspection without error bars or a quantitative comparison against direct training, so the strength of the claim is not yet commensurate with its stated significance.

major comments (3)
  1. [§IV.A and Figs. 3–4] The SNF scaling data in Figs. 3 and 4 are obtained almost entirely with parameters transferred from training on L/a = 20 lattices at nstep = 16, 32, 64. The manuscript states that 'we observed no differences in the relevant metrics' for nstep > 64 and that parameters transferred to smaller volumes are 'compatible' with direct training, but no quantitative comparison is shown. If the transferred parameters are noticeably suboptimal at larger nstep or at smaller volumes, the SNF points in Figs. 3 and 4 do not represent properly trained flows, and the observed collapse could be an artifact of imposing one parameter profile across all volumes. Because this transfer is load-bearing for the scaling-inheritance claim, the authors should provide a quantitative comparison, e.g. a plot or table of KL divergence and ESS for directly trained versus transferred parameters at least for one larger-nstep case and one smaller-volume case, with statistical uncertainties.
  2. [§IV.B, Figs. 2–4] No error bars are shown for the KL divergence or ESS in Figs. 2–4, and the claimed collapse onto a universal curve is assessed only visually. The central quantitative statements, including the factor-of-two improvement and the volume scaling, require a more rigorous treatment. For example, the authors could compute bootstrap or jackknife uncertainties for representative points, fit a common function of nstep/(L/a)^4, and report residuals or a goodness-of-fit statistic. Without such analysis, the reader cannot distinguish a genuine scaling collapse from a qualitative coincidence, especially given that the SNF points at several volumes share transferred parameters.
  3. [§IV.B, Fig. 2 and Conclusions] The claim that SNFs are 'roughly a factor 2 more efficient' is made in terms of nstep, while the actual computational cost per step includes the stout-smearing layers, which the text states are about 25% as expensive as a full MCMC update. If the comparison is meant to be wall-clock cost, the efficiency gain is closer to 1.6, not 2; if it is meant to be nstep only, this should be stated explicitly and the overhead discussion made consistent throughout. The authors should also report the statistical uncertainty on any factor-of-two estimate, since Fig. 2 contains overlapping curves at some nstep values.
minor comments (6)
  1. [Title page] The section heading 'INTRODUCTION AND MOTIV A TION' contains a typographical artifact ('V A TION' should be 'VATION').
  2. [Eq. (8)] In the denominator of Eq. (8), 'pc(n)(Un − 1)' is ambiguous; it should be 'pc(n)(Un−1)' to denote the configuration at step n−1 rather than the configuration Un minus 1.
  3. [Fig. 1] The learned parameters ρ(n) are shown without error bars or a description of run-to-run variability, which makes it difficult to assess the collapse for nstep = 64 where the text notes the parameters are noisier.
  4. [§IV.A] The phrase 'the training was performed uniquely on values of nstep which are much smaller than the ones showed in fig. 2' is awkward; 'showed' should be 'shown', and the sentence could be rephrased for clarity.
  5. [Conclusions] The sentence 'These results points to an underlying structure' contains a subject-verb agreement error; it should be 'These results point to an underlying structure'.
  6. [General] The manuscript does not state where the code and data are available; for a numerical study of this type, a reproducibility statement or repository link would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the scaling result is measured from new numerical data, not derived from fitted parameters or self-citations.

full rationale

The paper's central claim—that both NE-MCMC and trained SNFs in SU(3) show KL divergence and ESS collapsing as functions of nstep/(L/a)^4—is an empirical collapse of measured quantities (Figs. 3 and 4), not an identity derived from the fitted smearing parameters. The smearing parameters rho(n) are trained by minimizing DKL (eq. 27) for nstep = 16, 32, 64 at L/a = 20 and then interpolated/transferred; however, the reported KL and ESS values at larger nstep and other volumes are computed from simulations, not obtained from the interpolation formula, so the transfer is an extrapolation assumption rather than a circular reduction. The use of DKL as both training loss and evaluation metric is standard variational practice and does not force the factor-of-2 SNF improvement, which is a measured gap over untrained NE-MCMC at matched nstep. Self-citations to refs. [24, 44, 49, 50, 83, 84] provide background and previously observed NE-MCMC scaling, but the present paper independently reproduces the NE-MCMC scaling in Figs. 3–4 and the SNF scaling is new data, so no load-bearing step reduces to a self-citation. No uniqueness theorem is imported, and no known result is merely renamed. The transfer-learning validation gap flagged in the skeptic summary is a robustness or evidence concern, not circularity.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim is an empirical scaling law, so the ledger is light. The only fitted quantities are the smearing parameters of the equivariant layers, which are standard trainable components of the architecture and do not enter the scaling argument directly. The main non-standard assumption is the transferability of these parameters across volumes and nstep, which is essential for the reported training-cost savings and for the SNF scaling data at larger nstep.

free parameters (2)
  • Stout smearing coefficient ρ(n) per mask per layer = Trained via Adam; values scale as ~O(10^-4) to O(10^-3) times nstep (see Fig. 1)
    These 8 parameters per layer (one per mask) are optimized to minimize the dissipated work (KL divergence). The trained values are used for SNF sampling, but the scaling claim is an observed property of the resulting sampler, not a direct function of the fitted values.
  • Global interpolation of ρ(n) vs n/nstep for transfer learning = Not given numerically; interpolated from nstep = 16, 32, 64 trainings
    Used to set smearing parameters for nstep > 64 without retraining. This is a fitted curve, and the authors report that direct training at larger nstep gives compatible metrics, but no quantitative comparison is shown.
assumptions (5)
  • standard math Jarzynski equality and Crooks theorem hold for the non-equilibrium evolutions
    Core of NE-MCMC; used in Eqs. (4), (10)-(13). Assumes detailed balance of the Markov chain updates (Eq. 8). These are established results from non-equilibrium statistical mechanics.
  • domain assumption Heatbath and over-relaxation updates satisfy detailed balance with respect to the Wilson action at each intermediate β(n)
    Required for Crooks theorem to hold; standard for these local updates in lattice gauge theory, stated in Section II and used throughout the numerical simulations.
  • domain assumption The prior distribution q0 is sampled at equilibrium
    Stated in Section II: 'the only exception is that if the prior distribution q0 is sampled with a MCMC, it must be at equilibrium.' The simulations thermalize at β0 before the non-equilibrium evolution begins.
  • domain assumption The stout-smearing coupling layers are invertible and their Jacobian is computed exactly
    Needed for Eqs. (22)-(26). Invertibility is argued by reference to the link-level flow in ref. [29]; the paper does not prove invertibility itself but relies on the cited result.
  • ad hoc to paper Learned ρ(n) transfer across volumes and larger nstep without retraining
    The transfer learning procedure (training on nstep = 16, 32, 64 and L/a = 20, then interpolating ρ for larger nstep and smaller volumes) is an empirical assumption. The authors validate it only by comparing final metrics, not by showing full training curves for all cases, so it is a load-bearing premise for the reported SNF scalability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scaling of Stochastic Normalizing Flows in $\mathrm{SU}(3)$ lattice gauge theory." pith.science (2026). https://pith.science/paper/A7P5GJO7

@misc{pith2026241200200,
  author       = {Pith},
  title        = {Pith review of: Scaling of Stochastic Normalizing Flows in $\mathrmSU(3)$ lattice gauge theory},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A7P5GJO7}},
  note         = {Machine review of arXiv:2412.00200}
}
abstract

Non-equilibrium Markov Chain Monte Carlo (NE-MCMC) simulations provide a well-understood framework based on Jarzynski's equality to sample from a target probability distribution. By driving a base probability distribution out of equilibrium, observables are computed without the need to thermalize. If the base distribution is characterized by mild autocorrelations, this approach provides a way to mitigate critical slowing down. Out-of-equilibrium evolutions share the same framework of flow-based approaches and they can be naturally combined into a novel architecture called Stochastic Normalizing Flows (SNFs). In this work we present the first implementation of SNFs for $\mathrm{SU}(3)$ lattice gauge theory in 4 dimensions, defined by introducing gauge-equivariant layers between out-of-equilibrium Monte Carlo updates. The core of our analysis is focused on the promising scaling properties of this architecture with the degrees of freedom of the system, which are directly inherited from NE-MCMC. Finally, we discuss how systematic improvements of this approach can realistically lead to a general and yet efficient sampling strategy at fine lattice spacings for observables affected by long autocorrelation times.

Figures

Figures reproduced from arXiv: 2412.00200 by the authors.

Figure 1
Figure 1. FIG. 1: Value of the learned parameter [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Results for the KL divergence (top panel) and [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4: Results for the KL divergence (top panel) and [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: FIG. 6: Results for the dimensionless free energy ∆ [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: Distribution of the work [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Stochastic Quantization as Optimal Control

    hep-lat 2026-07 conditional novelty 6.0 of 10

    Stochastic quantization is re-expressed as finite-time optimal control, in which a learned Doob force plus exact path weights reach the Gibbs measure without waiting for equilibrium.

  2. Studying Effective String Theory using deep generative models

    hep-lat 2025-08 conditional novelty 4.0 of 10

    Flow-based samplers numerically confirm the next-to-leading-order width and the resummed string-tension conjecture for the Nambu-Goto effective string in 2+1 dimensions.

Reference graph

Works this paper leans on

101 extracted references · 5 canonical work pages · cited by 2 Pith papers

  1. [1]

    quasi-static

    More generally, this quantity gives us a quantita- tive description of the distribution of the weights of exp(−W (U )) across different evolutions: indeed, inves- tigating this distribution represents an effective way to monitor the overlap with the target distribution. One simple and effective approach to either reduce the KL divergence of eq. (14) or to...

    2000

  2. [2]

    Aoki et al

    Y. Aoki et al. (Flavour Lattice Averaging Group (FLAG)), Eur. Phys. J. C 82, 869 (2022), arXiv:2111.09849 [hep-lat]

    arXiv 2022

  3. [3]

    Aoki et al

    Y. Aoki et al. (Flavour Lattice Averaging Group (FLAG)), (2024), arXiv:2411.04268 [hep-lat]

    arXiv 2024

  4. [4]

    Wolff, Nucl

    U. Wolff, Nucl. Phys. B Proc. Suppl. 17, 93 (1990)

    1990

  5. [5]

    Alles, G

    B. Alles, G. Boyd, M. D’Elia, A. Di Giacomo, and E. Vicari, Phys. Lett. B 389, 107 (1996), arXiv:hep- lat/9607049

    arXiv 1996

  6. [6]

    Del Debbio, G

    L. Del Debbio, G. M. Manca, and E. Vicari, Phys. Lett. B 594, 315 (2004), arXiv:hep-lat/0403001

    arXiv 2004

  7. [7]

    Schaefer, R

    S. Schaefer, R. Sommer, and F. Virotta (ALPHA), Nucl. Phys. B 845, 93 (2011), arXiv:1009.5228 [hep-lat]

    arXiv 2011

  8. [8]

    Luscher and S

    M. Luscher and S. Schaefer, JHEP 07, 036 (2011), arXiv:1105.4749 [hep-lat]

    arXiv 2011

Show all 101 references
  1. [9]

    Luscher and S

    M. Luscher and S. Schaefer, Comput. Phys. Commun. 184, 519 (2013), arXiv:1206.2809 [hep-lat]

    2013 arXiv

  2. [10]

    A. Laio, G. Martinelli, and F. Sanfilippo, JHEP 07, 089 (2016), arXiv:1508.07270 [hep-lat]

    2016 arXiv

  3. [11]

    Eichhorn, G

    T. Eichhorn, G. Fuwa, C. Hoelbling, and L. Varnhorst, Phys. Rev. D 109, 114504 (2024), arXiv:2307.04742 [hep-lat]

    2024 arXiv

  4. [12]

    Albandea, P

    D. Albandea, P. Hern´ andez, A. Ramos, and F. Romero- L´ opez, Eur. Phys. J. C 81, 873 (2021), [Erratum: Eur.Phys.J.C 83, 508 (2023)], arXiv:2106.14234 [hep- lat]

    2021 arXiv

  5. [13]

    Cranmer, G

    K. Cranmer, G. Kanwar, S. Racani` ere, D. J. Rezende, and P. E. Shanahan, Nature Rev. Phys. 5, 526 (2023), arXiv:2309.01156 [hep-lat]

    2023 arXiv

  6. [14]

    Luscher, Commun

    M. Luscher, Commun. Math. Phys. 293, 899 (2010), arXiv:0907.5491 [hep-lat]

    2010 arXiv

  7. [15]

    Rezende and S

    D. Rezende and S. Mohamed, in Proceedings of the 32nd International Conference on Machine Learning , Vol. 37 (PMLR, 2015) pp. 1530–1538, arXiv:1505.05770 [stat.ML]

    2015 arXiv

  8. [16]

    M. S. Albergo, G. Kanwar, and P. E. Shanahan, Phys. Rev. D 100, 034515 (2019), arXiv:1904.12072 [hep-lat]

    2019 arXiv

  9. [17]

    K. A. Nicoli, C. J. Anders, L. Funcke, T. Hartung, K. Jansen, P. Kessel, S. Nakajima, and P. Stornati, Phys. Rev. Lett. 126, 032001 (2021), arXiv:2007.07115 [hep-lat]

    2021 arXiv

  10. [18]

    K. A. Nicoli, C. J. Anders, T. Hartung, K. Jansen, P. Kessel, and S. Nakajima, Phys. Rev. D 108, 114501 (2023), arXiv:2302.14082 [hep-lat]

    2023 arXiv

  11. [19]

    Del Debbio, J

    L. Del Debbio, J. M. Rossney, and M. Wilson, Phys. Rev. D 104, 094507 (2021), arXiv:2105.12481 [hep-lat]

    2021 arXiv

  12. [20]

    Gerdes, P

    M. Gerdes, P. de Haan, C. Rainone, R. Bondesan, and M. C. N. Cheng, SciPost Phys. 15, 238 (2023), arXiv:2207.00283 [hep-lat]

    2023 arXiv

  13. [21]

    Singha, D

    A. Singha, D. Chakrabarti, and V. Arora, Phys. Rev. D 107, 014512 (2023), arXiv:2207.00980 [hep-lat]

    2023 arXiv

  14. [22]

    S. Chen, O. Savchuk, S. Zheng, B. Chen, H. Stoecker, L. Wang, and K. Zhou, Phys. Rev. D 107, 056001 (2023), arXiv:2211.03470 [hep-lat]

    2023 arXiv

  15. [23]

    Caselle, E

    M. Caselle, E. Cellini, and A. Nada, JHEP 02, 048 (2024), arXiv:2307.01107 [hep-lat]

    2024 arXiv

  16. [24]

    Albandea, L

    D. Albandea, L. Del Debbio, P. Hern´ andez, R. Kenway, J. Marsh Rossney, and A. Ramos, Eur. Phys. J. C 83, 676 (2023), arXiv:2302.08408 [hep-lat]

    2023 arXiv

  17. [25]

    Bulgarelli, E

    A. Bulgarelli, E. Cellini, K. Jansen, S. K¨ uhn, A. Nada, S. Nakajima, K. A. Nicoli, and M. Panero, Phys. Rev. Lett. 134, 151601 (2025), arXiv:2410.14466 [quant-ph]

    2025 arXiv

  18. [26]

    Kanwar, M

    G. Kanwar, M. S. Albergo, D. Boyda, K. Cranmer, D. C. Hackett, S. Racani` ere, D. J. Rezende, and P. E. Shanahan, Phys. Rev. Lett. 125, 121601 (2020), arXiv:2003.06413 [hep-lat]

    2020 arXiv

  19. [27]

    Boyda, G

    D. Boyda, G. Kanwar, S. Racani` ere, D. J. Rezende, M. S. Albergo, K. Cranmer, D. C. Hackett, and P. E. Shanahan, Phys. Rev. D 103, 074504 (2021), arXiv:2008.05456 [hep-lat]

    2021 arXiv

  20. [28]

    Bacchio, P

    S. Bacchio, P. Kessel, S. Schaefer, and L. Vaitl, Phys. Rev. D 107, L051504 (2023), arXiv:2212.08469 [hep- lat]

    2023 arXiv

  21. [29]

    Singha, D

    A. Singha, D. Chakrabarti, and V. Arora, Phys. Rev. D 108, 074518 (2023), arXiv:2306.00581 [hep-lat]

    2023 arXiv

  22. [30]

    Abbott et al

    R. Abbott et al. , (2023), arXiv:2305.02402 [hep-lat]. 14

    2023 arXiv

  23. [31]

    Gerdes, P

    M. Gerdes, P. de Haan, R. Bondesan, and M. C. N. Cheng, (2024), arXiv:2410.13161 [hep-lat]

    2024

  24. [32]

    M. S. Albergo, G. Kanwar, S. Racani` ere, D. J. Rezende, J. M. Urban, D. Boyda, K. Cranmer, D. C. Hackett, and P. E. Shanahan, Phys. Rev. D 104, 114507 (2021), arXiv:2106.05934 [hep-lat]

    2021 arXiv

  25. [33]

    Finkenrath, (2022), arXiv:2201.02216 [hep-lat]

    J. Finkenrath, (2022), arXiv:2201.02216 [hep-lat]

    2022 arXiv

  26. [34]

    M. S. Albergo, D. Boyda, K. Cranmer, D. C. Hackett, G. Kanwar, S. Racani` ere, D. J. Rezende, F. Romero- L´ opez, P. E. Shanahan, and J. M. Urban, Phys. Rev. D 106, 014514 (2022), arXiv:2202.11712 [hep-lat]

    2022 arXiv

  27. [35]

    Abbott et al

    R. Abbott et al. , Phys. Rev. D 106, 074506 (2022), arXiv:2207.08945 [hep-lat]

    2022 arXiv

  28. [36]

    Abbott, A

    R. Abbott, A. Botev, D. Boyda, D. C. Hackett, G. Kan- war, S. Racani` ere, D. J. Rezende, F. Romero-L´ opez, P. E. Shanahan, and J. M. Urban, Phys. Rev. D 109, 094514 (2024), arXiv:2401.10874 [hep-lat]

    2024 arXiv

  29. [37]

    No´ e, S

    F. No´ e, S. Olsson, J. K¨ ohler, and H. Wu, Science 365, eaaw1147 (2019), https://www.science.org/doi/pdf/10.1126/science.aaw1147

    2019 doi

  30. [38]

    Invernizzi, A

    M. Invernizzi, A. Kr¨ amer, C. Clementi, and F. No´ e, The Journal of Physical Chemistry Letters 13, 11643–11649 (2022)

    2022

  31. [39]

    Wirnsberger, A

    P. Wirnsberger, A. J. Ballard, G. Papamakar- ios, S. Abercrombie, S. Racani` ere, A. Pritzel, D. Jimenez Rezende, and C. Blundell, The Journal of Chemical Physics 153 (2020), 10.1063/5.0018903

    2020 doi

  32. [40]

    Abbott et al

    R. Abbott et al. , Eur. Phys. J. A 59, 257 (2023), arXiv:2211.07541 [hep-lat]

    2023 arXiv

  33. [41]

    Komijani and M

    J. Komijani and M. K. Marinkovic, PoS LA T- TICE2022, 019 (2023), arXiv:2301.01504 [hep-lat]

    2023 arXiv

  34. [42]

    Jarzynski, Phys

    C. Jarzynski, Phys. Rev. Lett. 78, 2690 (1997), arXiv:cond-mat/9610209 [cond-mat]

    1997 arXiv

  35. [43]

    Jarzynski, Phys

    C. Jarzynski, Phys. Rev. E 56, 5018–5035 (1997), arXiv:cond-mat/9707325 [cond-mat]

    1997 arXiv

  36. [44]

    Caselle, G

    M. Caselle, G. Costagliola, A. Nada, M. Panero, and A. Toniato, Phys. Rev. D 94, 034503 (2016), arXiv:1604.05544 [hep-lat]

    2016 arXiv

  37. [45]

    Caselle, A

    M. Caselle, A. Nada, and M. Panero, Phys. Rev. D 98, 054513 (2018), arXiv:1801.03110 [hep-lat]

    2018 arXiv

  38. [46]

    Francesconi, M

    O. Francesconi, M. Panero, and D. Preti, JHEP 07, 233 (2020), arXiv:2003.13734 [hep-lat]

    2020 arXiv

  39. [47]

    Bulgarelli and M

    A. Bulgarelli and M. Panero, JHEP 06, 030 (2023), arXiv:2304.03311 [quant-ph]

    2023 arXiv

  40. [48]

    Bulgarelli and M

    A. Bulgarelli and M. Panero, JHEP 06, 041 (2024), arXiv:2404.01987 [quant-ph]

    2024 arXiv

  41. [49]

    G. E. Crooks, Phys. Rev. E 60, 2721 (1999)

    1999

  42. [50]

    Bonanno, A

    C. Bonanno, A. Nada, and D. Vadacchino, JHEP 04, 126 (2024), arXiv:2402.06561 [hep-lat]

    2024 arXiv

  43. [51]

    Bonanno, A

    C. Bonanno, A. Nada, and D. Vadacchino, in 41st In- ternational Symposium on Lattice Field Theory (2024) arXiv:2411.00620 [hep-lat]

    2024 arXiv

  44. [52]

    Schmiedl and U

    T. Schmiedl and U. Seifert, Phys. Rev. Lett. 98 (2007), 10.1103/physrevlett.98.108301, arXiv:cond- mat/0701554 [cond-mat.stat-mech]

    2007

  45. [53]

    Gomez-Marin, T

    A. Gomez-Marin, T. Schmiedl, and U. Seifert, The Journal of Chemical Physics 129 (2008), 10.1063/1.2948948, arXiv:0803.0269 [cond-mat.stat- mech]

    2008 arXiv

  46. [54]

    Aurell, C

    E. Aurell, C. Mej ´ ıa-Monasterio, and P. Muratore- Ginanneschi, Phys. Rev. Lett. 106 (2011), 10.1103/physrevlett.106.250601, arXiv:1012.2037 [cond-mat.stat-mech]

    2011 arXiv

  47. [55]

    Aurell, C

    E. Aurell, C. Mej ´ ıa-Monasterio, and P. Muratore- Ginanneschi, Phys. Rev. E 85 (2012), 10.1103/phys- reve.85.020103, arXiv:1111.2876 [cond-mat.stat-mech]

    2012 arXiv

  48. [56]

    M. V. S. Bonan¸ ca and S. Deffner, Phys. Rev. E 98 (2018), 10.1103/physreve.98.042103, arXiv:1803.07050 [cond-mat.stat-mech]

    2018 arXiv

  49. [57]

    L. P. Kamizaki, M. V. S. Bonan¸ ca, and S. R. Muniz, Phys. Rev. E 106 (2022), 10.1103/physreve.106.064123, arXiv:2204.07145 [cond-mat.stat-mech]

    2022 arXiv

  50. [58]

    D. A. Sivak and G. E. Crooks, Phys. Rev. Lett. 108 (2012), 10.1103/physrevlett.108.190602, arXiv:1201.4166 [cond-mat.stat-mech]

    2012 arXiv

  51. [59]

    P. R. Zulkowski, D. A. Sivak, G. E. Crooks, and M. R. DeWeese, Phys. Rev. E 86 (2012), 10.1103/phys- reve.86.041148, arXiv:1208.4553 [cond-mat.stat-mech]

    2012 arXiv

  52. [60]

    G. M. Rotskoff and G. E. Crooks, Phys. Rev. E 92 (2015), 10.1103/physreve.92.060102, arXiv:1510.06734 [cond-mat.stat-mech]

    2015 arXiv

  53. [61]

    Brandner and K

    K. Brandner and K. Saito, Phys. Rev. Lett. 124 (2020), 10.1103/physrevlett.124.040602, arXiv:1907.06780 [cond-mat.stat-mech]

    2020 arXiv

  54. [62]

    Blaber and D

    S. Blaber and D. A. Sivak, The Journal of Chemical Physics 153 (2020), 10.1063/5.0033405, arXiv:2009.14354 [cond-mat.stat-mech]

    2020 arXiv

  55. [63]

    Blaber and D

    S. Blaber and D. A. Sivak, J. Phys. Commun. 7, 033001 (2023), arXiv:2212.00706 [cond-mat.stat-mech]

    2023 arXiv

  56. [64]

    R. M. Neal, Statistics and Computing 11, 125 (2001), arXiv:physics/9803008 [physics.comp-ph]

    2001 arXiv

  57. [65]

    C. Dai, J. Heng, P. Jacob, and N. Whiteley, Journal of the American Statistical Association 117, 1587 (2022), arXiv:2007.11936 [stat.CO]

    2022 arXiv

  58. [66]

    Sohl-Dickstein, E

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, in Proceedings of the 32nd International Conference on Machine Learning , Proceedings of Ma- chine Learning Research, Vol. 37 (PMLR, 2015) pp. 2256–2265, arXiv:1503.03585 [cs.LG]

    2015 arXiv

  59. [67]

    J. Ho, A. Jain, and P. Abbeel, in Advances in Neu- ral Information Processing Systems , Vol. 33 (2020) pp. 6840–6851, arXiv:2006.11239 [cs.LG]

    2020 arXiv

  60. [68]

    L. Wang, G. Aarts, and K. Zhou, JHEP 05, 060 (2024), arXiv:2309.17082 [hep-lat]

    2024 arXiv

  61. [69]

    Q. Zhu, G. Aarts, W. Wang, K. Zhou, and L. Wang (2024) arXiv:2410.19602 [hep-lat]

    2024 arXiv

  62. [70]

    Aarts, D

    G. Aarts, D. E. Habibi, L. Wang, and K. Zhou, in 38th conference on Neural Information Processing Systems (2024) arXiv:2410.21212 [hep-lat]

    2024 arXiv

  63. [71]

    Arbel, A

    M. Arbel, A. Matthews, and A. Doucet, in Interna- tional Conference on Machine Learning (PMLR, 2021) pp. 318–330, arXiv:2102.07501 [stat.ML]

    2021 arXiv

  64. [72]

    A. G. D. G. Matthews, M. Arbel, D. J. Rezende, and A. Doucet, in International Conference on Ma- chine Learning (PMLR, 2022) pp. 15196–15219, arXiv:2201.13117 [stat.ML]

    2022 arXiv

  65. [73]

    Nets: A non-equilibrium transport sampler,

    M. S. Albergo and E. Vanden-Eijnden, “Nets: A non-equilibrium transport sampler,” (2024), arXiv:2410.02711 [cs.LG]

    2024 arXiv

  66. [74]

    Hasenbusch, Phys

    M. Hasenbusch, Phys. Rev. D 96, 054504 (2017), arXiv:1706.04443 [hep-lat]

    2017 arXiv

  67. [75]

    Bonanno, C

    C. Bonanno, C. Bonati, and M. D’Elia, JHEP 03, 111 (2021), arXiv:2012.14000 [hep-lat]

    2021 arXiv

  68. [76]

    Bonanno, M

    C. Bonanno, M. D’Elia, B. Lucini, and D. Vadacchino, Phys. Lett. B 833, 137281 (2022), arXiv:2205.06190 [hep-lat]

    2022 arXiv

  69. [77]

    Bonanno, M

    C. Bonanno, M. D’Elia, and L. Verzichelli, JHEP 02, 156 (2024), arXiv:2312.12202 [hep-lat]. 15

    2024 arXiv

  70. [78]

    Bonanno, J

    C. Bonanno, J. L. Dasilva Gol´ an, M. D’Elia, M. Garc ´ ıa P´ erez, and A. Giorgieri, Eur. Phys. J. C 84, 916 (2024), arXiv:2403.13607 [hep-lat]

    2024 arXiv

  71. [79]

    Bonanno, G

    C. Bonanno, G. Clemente, M. D’Elia, L. Maio, and L. Parente, JHEP 08, 236 (2024), arXiv:2404.14151 [hep-lat]

    2024 arXiv

  72. [80]

    Abbott, D

    R. Abbott, D. Boyda, D. C. Hackett, G. Kanwar, F. Romero-L´ opez, P. E. Shanahan, J. M. Urban, and M. S. Albergo, PoS LA TTICE2023, 011 (2024), arXiv:2404.11674 [hep-lat]

    2024 arXiv

  73. [81]

    H. Wu, J. K¨ ohler, and F. Noe, in Advances in Neu- ral Information Processing Systems , Vol. 33 (2020) pp. 5933–5944, arXiv:2002.06707 [stat.ML]

    2020 arXiv

  74. [82]

    Vaikuntanathan and C

    S. Vaikuntanathan and C. Jarzynski, The Journal of Chemical Physics 134 (2011), 10.1063/1.3544679, arXiv:1101.2612 [cond-mat.stat-mech]

    2011 arXiv

  75. [83]

    J. P. Nilmeier, G. E. Crooks, D. D. L. Minh, and J. D. Chodera, Proceedings of the National Academy of Sciences 108 (2011), 10.1073/pnas.1106094108, arXiv:1105.2278 [cond-mat.stat-mech]

    2011 arXiv

  76. [84]

    Caselle, E

    M. Caselle, E. Cellini, A. Nada, and M. Panero, JHEP 07, 015 (2022), arXiv:2201.08862 [hep-lat]

    2022 arXiv

  77. [85]

    Caselle, E

    M. Caselle, E. Cellini, and A. Nada, JHEP 02, 090 (2025), arXiv:2409.15937 [hep-lat]

    2025 arXiv

  78. [86]

    G. E. Crooks, Journal of Statistical Physics 90, 1481 (1998)

    1998

  79. [87]

    We refer to ref.[44] for an in-depth description on how to compute with NE-MCMC the pressure and the equation of state in the SU(3) pure gauge theory

    Across the manuscript ∆ F is more precisely the free energy in units of the temperature: we remark that in a non-zero temperature setup this quantity is indeed connected to the actual free energy of the theory. We refer to ref.[44] for an in-depth description on how to compute...

  80. [88]

    Nice: Non- linear independent components estimation,

    L. Dinh, D. Krueger, and Y. Bengio, “Nice: Non- linear independent components estimation,” (2015), arXiv:1410.8516 [cs.LG]

    2015 arXiv

  81. [89]

    L. Dinh, J. Sohl-Dickstein, and S. Bengio, in Interna- tional Conference on Learning Representations (2017) arXiv:1605.08803 [cs.LG]

    2017 arXiv

  82. [90]

    Cohen and M

    T. Cohen and M. Welling, inProceedings of The 33rd In- ternational Conference on Machine Learning , Proceed- ings of Machine Learning Research, Vol. 48, edited by M. F. Balcan and K. Q. Weinberger (PMLR, New York, New York, USA, 2016) pp. 2990–2999

    2016

  83. [91]

    K¨ ohler, L

    J. K¨ ohler, L. Klein, and F. No´ e, (2019), arXiv:1910.00753 [stat.ML]

    2019 arXiv

  84. [92]

    Morningstar and M

    C. Morningstar and M. J. Peardon, Phys. Rev. D 69, 054501 (2004), arXiv:hep-lat/0311018

    2004 arXiv

  85. [93]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” (2015), arXiv:1512.03385 [cs.CV]

    2015 arXiv

  86. [94]

    Tomiya and Y

    A. Tomiya and Y. Nagai, (2021), arXiv:2103.11965 [hep-lat]

    2021 arXiv

  87. [95]

    Necco and R

    S. Necco and R. Sommer, Nucl. Phys. B622, 328 (2002), arXiv:hep-lat/0108008

    2002 arXiv

  88. [96]

    D. P. Kingma and J. Ba, (2014), arXiv:1412.6980 [cs.LG]

    2014 arXiv

  89. [97]

    L¨ uscher, EPJ Web Conf

    M. L¨ uscher, EPJ Web Conf. 175, 01002 (2018), arXiv:1707.09758 [hep-lat]

    2018 arXiv

  90. [98]

    Giusti and M

    L. Giusti and M. L¨ uscher, Eur. Phys. J. C 79, 207 (2019), arXiv:1812.02062 [hep-lat]

    2019 arXiv

  91. [99]

    Francis, P

    A. Francis, P. Fritzsch, M. L¨ uscher, and A. Rago, Comput. Phys. Commun. 255, 107355 (2020), arXiv:1911.04533 [hep-lat]

    2020 arXiv

  92. [100]

    Bruno, M

    M. Bruno, M. C` e, A. Francis, P. Fritzsch, J. R. Green, M. T. Hansen, and A. Rago, JHEP 11, 167 (2023), arXiv:2307.15674 [hep-lat]

    2023 arXiv

  93. [101]

    Fritzsch, J

    P. Fritzsch, J. Bulava, M. C` e, A. Francis, M. L¨ uscher, and A. Rago, PoS LA TTICE2021, 465 (2022), arXiv:2111.11544 [hep-lat]

    2022 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.