Pith. sign in

REVIEW 3 major objections 6 minor 17 references

The perturbative computation of the gradient flow coupling for the twisted Eguchi-Kawai model with the numerical stochastic perturbation theory

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Stochastic lattice perturbation theory in the twisted Eguchi–Kawai model extracts the universal one-loop beta function of SU(∞) gauge theory from the flow-time dependence of the gradient flow coupling, with the fitted coefficient matching…

desk verdict The paper's own fits do not pin down the one-loop coefficient at the claimed <10% level; the NSPT computation is real but the analysis is not yet convincing. read the letter →

arxiv 2501.18175 v1 pith:J2JH2SRB submitted 2025-01-30 hep-lat

Hironori Takei , Ken-Ichi Ishikawa , Masanori Okawa This is my paper · ORCID
classification hep-lat
keywords gradientflowcouplingtwistedEguchi-Kawaimodelnumericalstochasticperturbationtheorybetafunctionlarge-NfactorizationSU(∞)gaugelatticetimedependence
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

Gradient flow is a renormalization scheme in which gauge fields are diffused in a fictitious flow time, making flowed observables finite and regularization-independent. This paper uses numerical stochastic perturbation theory (NSPT) on the twisted Eguchi–Kawai (TEK) model to compute the perturbative coefficients $r_1(t)$ and $r_2(t)$ of the gradient flow coupling for SU(∞) gauge theory, and tries to show that the flow-time dependence of these coefficients encodes the universal $\beta$-function running once lattice and finite-volume effects are controlled. With the current data, the one-loop $\beta$ function is determined to better than 10% accuracy, whereas the two-loop coefficient is reproduced only with large errors. The paper also verifies that large-N factorization holds for flowed operators at finite flow time, meaning the variance of perturbative coefficients shrinks as the matrix size $N$ grows. This matters because the gradient flow scheme is regularization-independent, so a controllable perturbative calculation in the small TEK matrix offers a cheap route to precision $\beta$-function coefficients and to connecting lattice results with the $\overline{\mathrm{MS}}$ scheme.

What carries the argument

The machinery is the gradient flow coupling $\lambda_\rho$ defined through the flowed energy density $E(t)$ in Eq. (5), normalized by $N(t)$, and its perturbative expansion in the lattice bare coupling. The coefficients $r_i(t)$ are obtained from NSPT by solving the hierarchical flow equation (2) order by order, with the TEK twist phases $z_{\mu\nu}$ implementing the SU(∞) theory on a finite matrix of size $N$. The argument is carried by fitting those coefficients to the analytic flow-time dependence $\log(\sqrt{2t})+\gamma_E/2$ (plus an optional $A_0/t$ lattice-artifact term), because the slope of that logarithm is the $\beta$-function coefficient; the extraction is valid only if finite-volume corrections $O(t^2/N^2)$ stay small inside the chosen window.

What would settle it

Repeat the NSPT computation at larger N (for example N=625 or 841) and extend the flow-time window to smaller t, then fit the coefficients with a model that includes explicit $O(t^2/N^2)$ and $O(t^4/N^4)$ terms; if the extracted $B_0$ moves away from 0.046439 by more than the quoted ~10% error, the $\beta$-function extraction is not clean.

Watch

Extended reading notes

Core claim

The central claim is that in the TEK model with twist phase $\theta \simeq 0.40$, the gradient flow coupling expanded in the lattice bare coupling as $\lambda_\rho = \lambda_0 + r_1(t)\lambda_0^2 + r_2(t)\lambda_0^3 + \cdots$ has coefficients whose scale dependence is governed by the universal $\beta$ function. Fitting $r_1(t)$ to $f(t)=B_0(\log(\sqrt{2t})+\gamma_E/2)+F_1$ with an optional $A_0/t$ term, the flow-time window $t \in [0.9,6.3]$ and the $A_0$ term give $B_0 = 0.04725(349)$, consistent with the analytic value $0.046439$; omitting the $A_0$ term gives $B_0 = 0.04315(222)$, showing the sensitivity to lattice artifacts. The analogous fit to $r_2(t)-r_1(t)^2$ yields $B_1 = 0.00157(491)$ against $0.00090897$, too noisy to determine the two-loop $\beta$ function. The variance of $r_1(t)$ and $r_2(t)$ at $t=6.0$, extrapolated to the large-$N$ limit with a global fit including $O(t^4/N^4)$ terms, supports large-N factorization at finite flow time.

Load-bearing premise

The load-bearing premise is that the chosen fitting formula and flow-time window cleanly separate the logarithmic running from small lattice-spacing and finite-volume corrections, so no extra term is biasing the fitted slopes; the paper notes that one of its fits has a poor chi-squared per degree of freedom of 12.3, which suggests this separation might not be fully clean.

Editorial extensions

If this is right

  • The one-loop beta function of SU(∞) gauge theory can be reproduced from NSPT on the TEK model to better than 10% accuracy, without large-volume simulations.
  • Including the $A_0/t$ term and the flow-time window $t \in [0.9,6.3]$ suppresses the lattice-spacing error enough that the fitted coefficient stays close to the analytic value.
  • Large-N factorization holds for flowed operators at finite flow time, so increasing the matrix size $N$ reduces the variance of the perturbative coefficients.
  • The two-loop beta function remains out of reach with current statistics: the fitted $B_1$ is consistent with the analytic value only within a large error.
  • The same pipeline is expected to work at larger $N$ or with more statistics, and the variance analysis suggests that larger matrices will make higher-loop determinations cheaper.

Reading between the lines

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

  • Extension: pushing the same NSPT computation to larger matrix sizes (for example $N=625$ or larger) should bring the two-loop coefficient $B_1$ to a determinate value, because the demonstrated factorization implies variance drops like $1/N^2$; the paper does not itself run at these sizes.
  • Extension: the regularization independence of the gradient flow coupling means the extracted one-loop coefficient could serve as a numerical bridge between lattice-regularized SU(∞) observables and the $\overline{\mathrm{MS}}$ scheme, a consequence the paper leaves implicit.
  • Extension: applying a global fit in both flow time and matrix size, as the paper does for the variances, to the coefficients themselves would directly test whether the poor $\chi^2/\mathrm{dof}=12.3$ of the $f$-fit comes from an unmodeled finite-volume term.
Share X Bluesky LinkedIn Reddit HN

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 manuscript reports a numerical stochastic perturbation theory (NSPT) computation of the gradient flow coupling in the twisted Eguchi-Kawai (TEK) model. The authors generate perturbative configurations at N=289, 441, and 529, evolve the gradient flow equation perturbatively, extract the one-loop and two-loop coefficients r1(t) and r2(t) of the coupling, and fit the flow-time dependence to logarithmic forms motivated by the perturbative beta function. They compare the fitted coefficients with analytic values from the literature, claim a determination of the one-loop beta function to better than 10% accuracy, and present variance extrapolations purporting to confirm large-N factorization of flowed operators at finite flow time. The paper also explicitly acknowledges that the two-loop coefficient is not yet precisely determined and that more statistics are needed.

Significance. If the claims are correct, the work would be a useful demonstration that NSPT can be applied to the gradient flow coupling in a reduced large-N model, and that the universal one-loop coefficient can be recovered from lattice perturbation theory with controlled systematics. The paper has several strengths: it reports raw fit tables with analytic benchmarks, it uses multiple lattice sizes, and it correctly refrains from overclaiming the two-loop result. The extraction is not circular in the sense that the fitted coefficients are free parameters compared with external analytic values. However, the central quantitative claim of less-than-10% one-loop accuracy is not robust to the choice of fit function and fit window, as documented in Tables 2 and 3. Because the one-loop coefficient is the main numerical result, this issue is load-bearing.

major comments (3)
  1. [§4 and Tables 2–3] The claim in §4 that 'the one-loop beta function was determined with an accuracy of less than 10%' is not supported by the full set of fits. The only fit consistent with the analytic value B0=0.046439 is the g(x) fit in the extended window t∈[0.9,6.3], giving B0=0.04725(349) with χ²/dof=4.2. The f(x) fit in the same window gives B0=0.03762(144), 19% below the analytic value, with a better χ²/dof=3.2. In the shorter window t∈[2.1,6.3], both fits give B0≈0.043 with χ²/dof≈12. Without a stated model-selection criterion or a quoted systematic error from fit-model and window dependence, the <10% accuracy claim is not established.
  2. [§3.1, Figs. 1–2] The text presents r1(t) and r2(t) 'in the large-N limit' but does not describe how the N→∞ extrapolation is performed. The reader needs to know the fit form in 1/N², the data included (all three N values or a subset), whether the extrapolation is correlated, and the resulting uncertainties at each flow time. Since the beta-function extraction in §3.1 uses these extrapolated values, an unexplained extrapolation is a load-bearing missing detail. If the procedure is fully described in Refs. [10,11], the relevant fit form and uncertainties should at least be summarized here.
  3. [§3.2] The claim that 'these results confirm the existence of the large-N factorization at finite flow times' is stronger than the evidence shown. The simple linear fit in 1/N² yields a finite large-N variance, and the global fit including an O(t^4/N^4) term gives a smaller extrapolated value, but no fit parameters, errors, or goodness-of-fit are reported. With only three lattice sizes, the statement should be qualified as suggestive or preliminary, unless the quantitative extrapolation is shown to be consistent with zero variance at N=∞.
minor comments (6)
  1. [Table 1] The column header 'Statics' appears to be a typo for 'Statistics'.
  2. [Fig. 2 caption] The phrase 'the results is subtracted the two-loop coefficient' is ungrammatical; it should read 'the results after subtracting r1(t)^2 from r2(t)'.
  3. [§3.2] The phrase 'confirm the existence the large-N factorization' should be 'confirm the existence of large-N factorization'.
  4. [Eq. (8)] The fit functions are written as f(x) and g(x), but the variable is called t everywhere else; please use a consistent notation.
  5. [Fig. 3] The black dash-dotted lines from the global fit are mentioned in the text but not labeled in the figure caption; please add a legend or clarify in the caption.
  6. [Eq. (1)] The relation between ρ and μ is given as μ²t=ρ immediately after Eq. (1), but the parameter ρ in λ_ρ is introduced in Eq. (5) without an explicit definition; please define ρ when the coupling λ_ρ is first used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the beta-function coefficients are fitted from NSPT data and benchmarked against independent analytic values.

full rationale

The paper's central derivation is self-contained and non-circular. The perturbative coefficients r1(t) and r2(t) are generated by NSPT integration of the lattice flow equation (Eq. 2), and the one- and two-loop beta-function coefficients are obtained as free parameters B0 and B1 in the fits (Eqs. 8 and 9). The analytic values 0.046439 and 0.00090897 quoted in Tables 2-4 are external benchmarks from Refs. [15,16]; they are not substituted into the fit or used to define r1 or r2. Self-citations [10,11] point to companion papers for technical details and a tree-level finite-volume analysis, but those references do not enter the fits or fix B0 and B1. The large-N factorization check is an independent variance analysis, not a re-statement of the beta-function result. The reviewer's concern that the two accepted fits in the extended window differ by 19% from each other, and that the best-match fit has worse chi2, is a legitimate systematic-uncertainty criticism of the '<10% accuracy' claim, but it concerns the statistical model and fit selection rather than a reduction of the prediction to its own inputs. No equation or parameter is defined in terms of the target result, and no load-bearing premise relies on an unverified self-citation.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The paper introduces no invented entities. The central results depend on five fit parameters (B0, F1, A0, B1, F2) and variance extrapolation coefficients, and on standard domain assumptions about NSPT convergence, TEK volume reduction, the continuum flow-time form, and the size of finite-volume corrections. These are all reasonable within lattice field theory, but the fit regions and artifact terms are choices that affect the extracted values.

free parameters (6)
  • B0 (one-loop coefficient in r1(t) fit) = 0.04725(349) (g-fit, t in [0.9,6.3]); 0.04315(222) (f-fit, t in [2.1,6.3]); analytic 0.046439
    Fitted to the NSPT data for r1(t) using Eq. (8); central to the claim that the one-loop beta function is reproduced.
  • F1 (constant in r1(t) fit) = 0.21778(337) (g-fit, t in [0.9,6.3]); 0.22466(139) (f-fit, t in [2.1,6.3]); analytic 0.217862
    Fitted constant in Eq. (8); part of the same one-loop extraction.
  • A0 (lattice artifact coefficient) = 0.00577(139) (g-fit, t in [0.9,6.3]); 0.00030(634) (g-fit, t in [2.1,6.3])
    Coefficient of the ad hoc 1/t term added in g(x) to absorb lattice-spacing errors O(a^2/t).
  • B1 (two-loop coefficient in r2(t)-r1(t)^2 fit) = 0.00157(491); analytic 0.00090897
    Fitted from the two-loop combination using Eq. (9); error is too large for a precise determination.
  • F2 (constant in two-loop fit) = 0.00617(293); analytic 0.00673711
    Fitted constant in Eq. (9).
  • Variance extrapolation coefficients A0, A1, A2 = not quoted in the text
    Coefficients in f(t,N)=A0+A1*t^2/N^2+A2*t^4/N^4 used to extrapolate Var(r1) and Var(r2) to large N in Section 3.2.
assumptions (5)
  • domain assumption NSPT gives the correct perturbative expansion of lattice gauge theory to each order in the bare coupling.
    The entire computation of r1 and r2 rests on the stochastic quantization equivalence; invoked in Section 2, Eqs. (2)-(4).
  • domain assumption The TEK model with the twist phase theta ~ 0.40 and the chosen matrix sizes N=289, 441, 529 is equivalent to SU(N) gauge theory in the large-N limit with controllable finite-volume corrections.
    Used to extrapolate to large-N in Section 3.1 and 3.2; only three N values are available.
  • domain assumption The continuum analytic flow-time dependence of the gradient flow coupling (log plus constant, Eqs. (8)-(9)) from Refs. [15,16] applies to this large-N system.
    The fit functions are taken from this known result; if the TEK observable has different functional form, the extracted B0 and B1 would be biased.
  • domain assumption Finite-volume corrections to flowed observables enter at O(t^2/N^2), with next corrections O(t^4/N^4), and are negligible for t <= 6.3.
    Stated in Section 3.1; used to justify the fit window and the large-N variance extrapolation in Figure 3.
  • domain assumption The numerical integration of the flow equation with epsilon=0.01 introduces negligible systematic error.
    Stated in Section 3 with reference to Ref. [10]; not independently verified here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The perturbative computation of the gradient flow coupling for the twisted Eguchi-Kawai model with the numerical stochastic perturbation theory." pith.science (2026). https://pith.science/paper/J2JH2SRB

@misc{pith2026250118175,
  author       = {Pith},
  title        = {Pith review of: The perturbative computation of the gradient flow coupling for the twisted Eguchi-Kawai model with the numerical stochastic perturbation theory},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J2JH2SRB}},
  note         = {Machine review of arXiv:2501.18175}
}
abstract

The gradient flow method is a renormalization scheme in which the gauge field is flowed by the diffusion equation. The gradient flow scheme has benefits that the observables composed of flowed gauge fields do not require further renormalization and do not depend on the regularization. From the independence of the regularization, this scheme allows us to relate the lattice regularization and the dimensional regularization such as the $\overline{\mathrm{MS}}$ scheme. We compute the gradient flow coupling for the twisted Eguchi--Kawai model using the numerical stochastic perturbation theory. In this presentation we show the results of the perturbative coefficients of the gradient flow coupling and its flow time dependence. We investigate the beta function from the flow time dependence and discuss the lattice artifacts in the large flow time in taking the large-$N$ limit.

Figures

Figures reproduced from arXiv: 2501.18175 by the authors.

Figure 1
Figure 1. Flow time dependence of the one-loop coefficient 𝑟1 (𝑡ˆ) in the large-𝑁 limit. Top (bottom) panels present the fitting results in the flow time region 𝑡ˆ ∈ [2.1, 6.3] (𝑡ˆ ∈ [0.9, 6.3]). Red crosses represent simulation results, while black dashed lines correspond to continuum analytic expressions. Purple solid lines represent the fitting results with 𝑓 (𝑥) (left panels) and 𝑔(𝑥) (right panels) [PITH_FULL_IMAGE:figu… view at source ↗
Figure 2
Figure 2. Same as [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Variances of the coefficients 𝑟1 (𝑡ˆ) (left panel) and 𝑟2 (𝑡ˆ) (right panel) at 𝑡ˆ = 6.0. The red diamonds indicate the numerical results of the variances for the finite 𝑁 = 289, 441, and 529. The red down-triangles denote the results for large-𝑁 obtained using simple linear fitting (red solid line). 6 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 6 canonical work pages

  1. [1]

    Narayanan and H

    R. Narayanan and H. Neuberger,JHEP 03, 064 (2006),arXiv:hep-th/0601210, doi:10. 1088/1126-6708/2006/03/064

    arXiv 2006

  2. [2]

    Lüscher, JHEP 08, 071 (2010), arXiv:1006.4518 [hep-lat] , doi:10.1007/ JHEP08(2010)071, [Erratum: JHEP 03, 092 (2014)]

    M. Lüscher, JHEP 08, 071 (2010), arXiv:1006.4518 [hep-lat] , doi:10.1007/ JHEP08(2010)071, [Erratum: JHEP 03, 092 (2014)]

    arXiv 2010

  3. [3]

    M.LuscherandP.Weisz, JHEP02,051(2011), arXiv:1101.0963 [hep-th] ,doi:10.1007/ JHEP02(2011)051

    arXiv 2011

  4. [4]

    Parisi and Y.-s

    G. Parisi and Y.-s. Wu,Sci. Sin.24, 483 (1981)

    work page 1981

  5. [5]

    Di Renzo, G

    F. Di Renzo, G. Marchesini, P. Marenzoni and E. Onofri,Nucl. Phys. B Proc. Suppl.34, 795 (1994), doi:10.1016/0920-5632(94)90517-7

    work page doi:10.1016/0920-5632(94)90517-7 1994

  6. [6]

    Di Renzo, E

    F. Di Renzo, E. Onofri, G. Marchesini and P. Marenzoni,Nucl. Phys. B426, 675 (1994), arXiv:hep-lat/9405019, doi:10.1016/0550-3213(94)90026-4

    arXiv 1994

  7. [7]

    48,1063(1982),doi:10.1103/PhysRevLett.48.1063

    T.EguchiandH.Kawai, Phys.Rev.Lett. 48,1063(1982),doi:10.1103/PhysRevLett.48.1063

    doi:10.1103/physrevlett.48.1063 1982

  8. [8]

    Gonzalez-Arroyo and M

    A. Gonzalez-Arroyo and M. Okawa,Phys. Rev. D27, 2397 (1983), doi:10.1103/PhysRevD. 27.2397. 7 The perturbative computation of the gradient flow coupling for the twisted Eguchi–Kawai model with the numerical stochastic perturbation theory Hironori Takei

    doi:10.1103/physrevd 1983

Show all 17 references
  1. [9]

    A.Gonzalez-ArroyoandM.Okawa, Phys.Lett.B 120,174(1983),doi:10.1016/0370-2693(83) 90647-0

    1983 doi

  2. [10]

    Ishikawa, M

    K.-I. Ishikawa, M. Okawa and H. Takei (10 2024),arXiv:2410.15668 [hep-lat]

    2024 arXiv

  3. [11]

    Takei, Study of reduced matrix models in perturbation theory with numerical stochastic perturbation theory, PhD thesis, Hiroshima University (2025)

    H. Takei, Study of reduced matrix models in perturbation theory with numerical stochastic perturbation theory, PhD thesis, Hiroshima University (2025)

    2025

  4. [12]

    González-Arroyo, I

    A. González-Arroyo, I. Kanamori, K.-I. Ishikawa, K. Miyahana, M. Okawa and R. Ueno, JHEP06, 127 (2019),arXiv:1902.09847 [hep-lat] , doi:10.1007/JHEP06(2019)127

    2019 arXiv

  5. [13]

    Ramos, JHEP 11, 101 (2014), arXiv:1409.1445 [hep-lat] , doi:10.1007/ JHEP11(2014)101

    A. Ramos, JHEP 11, 101 (2014), arXiv:1409.1445 [hep-lat] , doi:10.1007/ JHEP11(2014)101

    2014 arXiv

  6. [14]

    Gonzalez-Arroyo and M

    A. Gonzalez-Arroyo and M. Okawa,JHEP 07, 043 (2010),arXiv:1005.1981 [hep-th] , doi:10.1007/JHEP07(2010)043

    2010 arXiv

  7. [15]

    Luscher and P

    M. Luscher and P. Weisz,Phys. Lett. B349, 165 (1995), arXiv:hep-lat/9502001, doi: 10.1016/0370-2693(95)00250-O

    1995 arXiv

  8. [16]

    R. V. Harlander and T. Neumann,JHEP 06, 161 (2016), arXiv:1606.03756 [hep-ph] , doi:10.1007/JHEP06(2016)161

    2016 arXiv

  9. [17]

    González-Arroyo, K.-I

    A. González-Arroyo, K.-I. Ishikawa, Y. Ji and M. Okawa,Int. J. Mod. Phys. A37, 2250210 (2022), arXiv:2208.01867 [hep-lat] , doi:10.1142/S0217751X22502104. 8

    2022 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.