{"id":"9335f30f-86c0-4439-a81f-1f365481decf","arxiv_id":"2501.18175","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"NSPT on the twisted Eguchi-Kawai model yields gradient flow coupling coefficients whose flow-time dependence reproduces the universal one-loop beta function and, with large errors, a two-loop value.","lead":"Using numerical stochastic perturbation theory, this paper computes the perturbative coefficients of the gradient flow coupling in the twisted Eguchi-Kawai model and extracts the one- and two-loop beta function coefficients of large-N gauge theory. A generalist might read it to see how stochastic quantization on a lattice can connect a reduced large-N model to known analytic renormalization results.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The <10% one-loop accuracy claim is not robust: the two accepted fits in the extended window differ by 19%, and the fit that matches the analytic value has worse chi2 than the fit that misses it.","rationale":"The reader's CONDITIONAL verdict is appropriate. My pass agrees that the limiting assumption is the fit ansatz and window separation of the logarithmic running from lattice and finite-volume effects, but I sharpen the concern: the inconsistency is visible inside the paper's own tables. The f-fit in Table 3 has a lower chi2 than the preferred g-fit but misses the analytic value by 19%, while the g-fit reaches the analytic value only after introducing an A0 term that changes B0 by about 26%. This pattern suggests B0 and A0 are compensating and that the fit model is not uniquely selected by the data. A decisive check is a model-averaging or global-fit reanalysis; if the spread persists, the paper needs a quoted systematic uncertainty. The current 'less than 10%' claim is therefore not yet fully supported, but the paper is honest about the two-loop limitation and delegates details to Refs. [10,11]. I leave the verdict unchanged at CONDITIONAL.","tokens_in":6492,"tokens_out":8570,"duration_ms":90397,"concrete_test":"Reanalyze the r1(t) data from Refs. [10,11] with a global correlated fit in t and N that includes the lattice-artifact term A0/t and a finite-volume term C t^2/N^2, using all three lattice sizes before the large-N extrapolation; report B0 with a model average over f, g, and g+C t^2/N^2 and over both fit windows. If the model-averaged B0 plus the between-model spread brackets 0.046439 within 10%, the concern is resolved; if it does not, the '<10%' claim should be relaxed to a wider systematic range.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim is that the one-loop beta-function coefficient is recovered with <10% accuracy. The paper's own fits do not uniquely support this. In the extended window t in [0.9,6.3], the f-fit (Eq. 8 without A0) gives B0=0.03762(144), 19% below the analytic value 0.046439 with chi2/dof=3.2, while the g-fit (Eq. 8 with A0/t) gives B0=0.04725(349), inside 1.7%, but with worse chi2/dof=4.2. In the shorter window t in [2.1,6.3], both f and g give chi2/dof~12 with B0 about 7% low. Thus the match to the universal beta function is contingent on preferring g in the extended window, a choice not justified by fit quality, information criteria, or any stated selection rule. No systematic error from fit-model and window dependence is quoted, so 'less than 10%' is not an established uncertainty. Because the two-loop coefficient is already statistically inconclusive (B1=0.00157(491) vs 0.000909), the one-loop extraction is the main numerical result, and this uncontrolled systematic is load-bearing.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":6821,"tokens_out":4463,"duration_ms":43566,"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":[{"comment":"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.","section":"§4 and Tables 2–3"},{"comment":"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.","section":"§3.1, Figs. 1–2"},{"comment":"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=∞.","section":"§3.2"}],"minor_comments":[{"comment":"The column header 'Statics' appears to be a typo for 'Statistics'.","section":"Table 1"},{"comment":"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)'.","section":"Fig. 2 caption"},{"comment":"The phrase 'confirm the existence the large-N factorization' should be 'confirm the existence of large-N factorization'.","section":"§3.2"},{"comment":"The fit functions are written as f(x) and g(x), but the variable is called t everywhere else; please use a consistent notation.","section":"Eq. (8)"},{"comment":"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.","section":"Fig. 3"},{"comment":"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.","section":"Eq. (1)"}],"recommendation":"major_revision","confidential_remarks":"This is a proceedings contribution, so it is reasonable for some technical details to be deferred to companion papers. However, the central numerical claim of better-than-10% accuracy needs to be qualified with a systematic uncertainty estimate or a clear statement that the quoted accuracy applies only to the preferred fit. The current presentation conflates one good fit with an established result."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper reports an NSPT computation of the gradient flow coupling in the TEK model at N=289, 441, and 529. The underlying computation is nontrivial and the setup is described clearly. The paper is also honest that the two-loop coefficient is statistically inconclusive, and the variance check for large-N factorization is a useful sanity check. Credit where due: this is the first NSPT calculation of the gradient flow coupling in TEK, and the data appear real.\n\nThe central claim that the one-loop beta function is determined to better than 10% is not supported by the fits shown. In the extended flow-time window t in [0.9,6.3], the f-fit (no 1/t term) gives B0=0.03762(144), 19% below the analytic value 0.046439, with chi2/dof=3.2. The g-fit (with 1/t term) gives B0=0.04725(349), within 1.7% of the analytic value, but with chi2/dof=4.2. In the shorter window t in [2.1,6.3], both fits give B0 about 7% low with chi2/dof around 12. The paper prefers the g-fit in the extended window because it matches the analytic value, but no selection rule is given and the chi2 difference is not meaningful. The 'less than 10%' statement is therefore a statistical error on one preferred fit, not a systematic error budget. The stress-test concern is valid.\n\nA related issue is that the manuscript repeatedly defers to Refs. [10,11] for details, including the error analysis of the numerical integration and the large-N factorization discussion. As a standalone arXiv paper, it is a summary of work documented elsewhere. The factorization section itself is suggestive rather than conclusive: a simple 1/N^2 extrapolation leaves a finite variance, and the conclusion rests on a global fit with O(t^4/N^4) corrections using only three N values.\n\nThe paper is written for lattice large-N and NSPT specialists. It is a proceedings contribution, so the bar is lower, but even by that standard the one-loop claim needs rewording. As a journal submission it would need a proper treatment of fit-model and window systematics, or an explicit note that the <10% figure is the statistical error from one fit.\n\nFor peer review: this deserves referee time rather than a desk reject. The computation is plausible and the community may want the numbers. But the referee should insist on a systematic error analysis for the fit extraction, and the paper should be conditional on either adding that analysis or softening the conclusion.\n\nI would not cite this paper itself; I would wait for the companion papers with the full results.","headline":"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.","tokens_in":7367,"tokens_out":2767,"would_cite":false,"duration_ms":28984,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"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…","keywords":["gradient flow coupling","twisted Eguchi-Kawai model","numerical stochastic perturbation theory","beta function","large-N factorization","SU(∞) gauge theory","lattice perturbation theory","flow time dependence"],"falsifier":"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.","tokens_in":6279,"feed_emoji":"🎯","tokens_out":12888,"duration_ms":111340,"temperature":0.7,"pith_summary":"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.","feed_headline":"One-loop beta function measured to under 10% in large-N gauge theory","feed_subtitle":"A stochastic lattice method confirms the coupling's universal running, with finite-volume effects under control.","key_machinery":"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.","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"Introduces numerical stochastic perturbation theory, the method used to generate order-by-order perturbative gauge configurations.","marker":"[4–6]"},{"why":"Defines the twisted Eguchi–Kawai model, which provides the SU(∞) large-N gauge theory used here.","marker":"[7–9]"},{"why":"Supplies the NSPT implementation for the TEK model, including the hierarchical flow equation and simulation setup.","marker":"[12]"},{"why":"Defines the gradient flow coupling used as the observable.","marker":"[13]"},{"why":"Gives the analytic one- and two-loop coefficients against which the NSPT fits are compared.","marker":"[15,16]"},{"why":"Justifies the twist phase used for a smooth approach to the large-N limit.","marker":"[14]"},{"why":"Reports the detailed analysis of finite-volume corrections and the tree-level energy-density behavior that underlies the large-N factorization check.","marker":"[10, 11]"},{"why":"Provides the gradient-flow integration scheme used to evolve the flowed fields.","marker":"[2]"}],"fun_headline_variants":["Stochastic lattice verifies one-loop beta in large-N gauge theory","Gradient flow coupling's one-loop coefficient matches universal theory","Large-N gauge theory: stochastic method confirms one-loop coupling","Twisted Eguchi-Kawai: one-loop beta matches prediction, two-loop noisy","Stochastic perturbation theory: one-loop beta agrees with universal value"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Stochastic lattice verifies one-loop beta in large-N gauge theory","Gradient flow coupling's one-loop coefficient matches universal theory","Large-N gauge theory: stochastic method confirms one-loop coupling","Twisted Eguchi-Kawai: one-loop beta matches prediction, two-loop noisy","Stochastic perturbation theory: one-loop beta agrees with universal value"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001399,"raw_usage":{"total_tokens":5679,"prompt_tokens":992,"completion_tokens":4687,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":608,"completion_tokens_details":{"reasoning_tokens":4598}},"tokens_in":608,"tokens_out":4687,"duration_ms":34986,"temperature":1.0,"reasoning_tokens":4598,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T00:25:16.478609+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":"Numerical stochastic perturbation theory applied to the twisted Eguchi-Kawai model","cited_arxiv_id":"1902.09847","evidence_quote":"Supplies the NSPT implementation for the TEK model, including the hierarchical flow equation and simulation setup."}],"review_version":1}