{"id":"8e344a30-a017-4b4d-88f7-694868fbdbc2","arxiv_id":"2501.04465","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"For N=2 and N=3, LPA rate functions match Monte Carlo shapes to within 1% after rescaling a global amplitude and zeta, but the uncorrected universal constants differ by 20-25%.","lead":"This paper computes the entire family of universal probability distributions of the order parameter for the 3D O(N) model at criticality, using a functional renormalization group method. It finds that the simplest local-potential approximation reproduces the shape of these distributions after rescaling a global amplitude and the scaling variable zeta.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The two fitted scales (r_I, r_zeta) are determined from the same Monte Carlo data used to claim the <1% agreement; without an independent next-order shape check, the assertion that LPA's functional form is exact is not established.","rationale":"The paper is transparent about its limitations: Table III reports the 20-25% discrepancies in Delta I_N and zeta_c, and Sec. IV B explicitly states the two-scale rescaling procedure. The large-N derivation in Sec. II and the numerical benchmark in Sec. III D are solid, and the regulator dependence at LPA is small, so I do not question the numerical implementation. My concern is with the inference from a two-parameter fit to the exactness of the functional form of the rate function. The reader's weakest-assumption analysis identifies the same load-bearing premise, and the CONDITIONAL verdict already captures the right level of confidence. The proposed DE2 computation is the natural decisive check: it directly tests whether LPA truncation errors are confined to the two global scales. If DE2 changes the shape of I(rho, zeta) beyond those scales, the paper's strongest claim should be downgraded to an interpolation statement; if DE2 confirms the two-scale pattern, the claim becomes well supported. Since no code or data are shipped, independent reimplementation would also help, but the scientific crux is the next-order shape comparison.","tokens_in":21354,"tokens_out":7939,"duration_ms":85112,"concrete_test":"Compute the O(2) rate function at the next order of the derivative expansion (DE2) in d=3, for zeta=0 and at least zeta=+/-2, using the same finite-size FRG setup and regulator optimization as in Sec. III C. After optimally rescaling the DE2 result to the published MC data with the same two-parameter procedure (r_I, r_zeta), compare the maximum relative error on the full (rho/rho_0, zeta) domain of Fig. 10 with the <1% LPA residual. If the DE2 residual is not substantially smaller, the LPA truncation error is not confined to the two global scales and the central claim is unsupported; if it drops substantially, the two-scale hypothesis is validated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on the ansatz that all LPA truncation error in the rate function is absorbed by two global multiplicative constants: I^dagger = r_I I and zeta^dagger = r_zeta zeta, introduced in Sec. IV B. These factors are obtained by best-collapsing the FRG and MC curves of Delta I_N(zeta) in Fig. 9, i.e. they are fitted to the same MC data that later certify the <1% agreement in Fig. 10. Table III shows that the two largest LPA errors are precisely the overall amplitude Delta I_N (about 20-25%) and the critical ratio zeta_c (about 25%), so the fit removes the two dominant discrepancies by construction. The residual 1% agreement on the full rate-function shape is genuine evidence, but it tests the two-scale hypothesis only under the assumption that no other LPA error exists. Nothing in the paper independently checks that assumption: the N=infinity limit cannot help because LPA is exact there, no next-order derivative-expansion result is provided, and the full zeta-family comparison shown in Fig. 10 is for N=2 only, not N=3. The conclusion that 'the error induced by the LPA is primarily concentrated in the calculation of the two universal constants' is therefore an inference from the two-parameter fit rather than a tested property of the approximation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper generalizes the FRG-based method for computing order-parameter rate functions, previously developed for the Ising model, to the O(N) model in d=3. It derives the exact large-N rate-function family (Sec. II), implements the LPA flow equations for the constraint effective action (Sec. III), and compares the LPA results with Monte Carlo simulations for N=2 and N=3 (Sec. IV). The central claim is that, after rescaling the rate-function amplitude by r_I(N) and the scaling variable zeta by r_zeta(N), the LPA reproduces the full functional form of the universal rate-function family to better than 1% relative error, and that the LPA error is concentrated in the two universal constants Delta I_N and zeta_c.","tokens_in":21752,"tokens_out":5779,"duration_ms":59365,"significance":"The exact large-N solution and its FRG derivation are clean, self-contained, and carefully benchmarked; the numerical integration of the LPA flow is validated against this exact result to high precision, and the regulator dependence is small. If the two-scale hypothesis for finite N were independently established, the paper would provide a strong nonperturbative tool: the entire universal family of critical rate functions would be obtained from LPA, leaving only two global constants to be determined by other means. However, the finite-N evidence for this hypothesis is currently weakened by the fact that the two rescaling factors are fitted to the very Monte Carlo data that later certify the shape agreement, and the full rate-function family comparison is shown for N=2 only. The paper is honest about the 20-25% discrepancies in Delta I_N and zeta_c, and the residual shape agreement is genuine evidence, but the stronger conclusion about where the LPA error lives is an inference from a two-parameter fit rather than a tested property of the approximation.","major_comments":[{"comment":"The central claim that 'the error induced by the LPA is primarily concentrated in the calculation of the two universal constants Delta I_N and zeta_c' is not supported by an independent test. The factors r_I(N) and r_zeta(N) are obtained by best-collapsing the FRG and MC curves of Delta I_N(zeta) in Fig. 9, i.e. they are fitted to the same Monte Carlo data that later certify the <1% agreement in Fig. 10. Table III shows that the two largest LPA errors are precisely the overall amplitude Delta I_N (about 20-25%) and the critical ratio zeta_c (about 25%), so the fit removes the two dominant discrepancies by construction. The residual 1% agreement in the rate-function shape is genuine evidence for the two-scale ansatz, but it tests that ansatz only under the assumption that no other LPA truncation error exists. The paper currently provides no out-of-sample check of this assumption: a next-order derivative-expansion result for Delta I_N or zeta_c is not given, the full zeta-family comparison in Fig. 10 is for N=2 only, and no train/test split of the MC data is made. I recommend either providing such an independent check (e.g., fixing r_I and r_zeta on a subset of zeta values or on N=3 and predicting the rest, or computing a next-order derivative-expansion estimate) or tempering the conclusion to state that the two-scale ansatz is consistent with, rather than established by, the present data.","section":"Sec. IV B, Fig. 10"},{"comment":"The paper claims to compute 'the entire family of universal scaling functions' for finite N, but the full rate-function family comparison between FRG and MC is shown only for N=2 (Fig. 10). For N=3, only the zeta=0 rate function (Fig. 7) and the integrated quantity Delta I_N(zeta) (Fig. 9) are compared. Since the central claim is about the functional form of the whole family in zeta and rho, the N=3 case should either be shown with the same family plot or the claim should be restricted to N=2. This is particularly important because the fitted rescalings differ between N=2 and N=3, so the universality of the two-scale hypothesis across N is not yet demonstrated.","section":"Sec. IV B, Fig. 10"},{"comment":"The assessment of the 1% agreement in Sec. IV B is made after also normalizing the field variable by rho_0, the position of the minimum of the rate function at zeta=0. This normalization removes a third non-universal scale in addition to the two fitted factors r_I and r_zeta. The paper acknowledges that the field scale is non-universal, but the role of the rho_0 normalization should be made explicit in the error budget: the <1% relative error in Fig. 10 is a statement about the shape in the normalized variable rho/rho_0, not about the absolute field scale. I ask the authors to state clearly that three rescalings (r_I, r_zeta, rho_0) are used in the comparison, and to discuss how the conclusion would be affected if rho_0 were treated as an additional fitted parameter.","section":"Sec. IV B, Eq. (49)"}],"minor_comments":[{"comment":"Equation (58) appears to contain a typesetting error: the term should read (d-2) * check_rho * tilde_I'_k, but the tilde on I is missing on the right-hand side, making the equation dimensionally inconsistent as typeset.","section":"Eq. (35)"},{"comment":"In Eq. (35), the coefficient '2d' is ambiguous; it should be typeset as 2^d (or, for d=3, 8) divided by L^d (rho_0 - rho). As printed, '2d Ld(rho0 - rho)' could be misread as 2d times a logarithm or as a product, and the LaTeX is missing a fraction or parentheses.","section":"Fig. 5"},{"comment":"The caption of Fig. 5 refers to 'Probability distributions P(check_rho)' but the text immediately below explains that the plotted quantity is actually <delta(s^2/2 - check_rho)>, which differs from P(s) by a Jacobian factor rho^{(N-2)/2}. The caption should be corrected to avoid confusion between the PDF and the histogram of the squared magnetization.","section":"Sec. III C"},{"comment":"The statement that using the fixed-point potential as the initial condition 'ensures there will not be corrections to scaling' is only strictly true if the fixed point is exact. At LPA for finite N, U* is an approximate fixed point, and setting all irrelevant perturbations to zero at k_* does not by itself eliminate corrections to scaling from the truncation. The paper's regulator-dependence tests mitigate this concern, but the phrase is too strong as written.","section":"Sec. IV, Table III"},{"comment":"The MC values of zeta_c are quoted as 3.2 +/- 0.2 (N=2) and 2.7 +/- 0.2 (N=3). Because the fitted rescalings r_zeta(2) = 1.33 and r_zeta(3) = 1.12 are derived from the same data, the reader should be told explicitly in the table caption that the MC zeta_c values are not independent of the rescaling factors used in the shape comparison; otherwise the 25% discrepancy and the near-perfect collapse in Fig. 10 may appear contradictory.","section":"Sec. IV B"},{"comment":"The sentence 'This has also been observed in the calculation of the rate function of the 3D Ising model using FRG [49] and perturbative RG [45]' is an important supporting reference, but since the present N=2 result is the main finite-N evidence in this paper, the phrasing could be sharpened to distinguish the N=1 and N=2 cases and to emphasize that the two-scale ansatz is an empirical observation at LPA for finite N, not a proven property of the exact theory.","section":"Sec. IV B"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest and the large-N part is solid; the numerical implementation is carefully benchmarked. The main risk is overinterpreting the two-parameter fit to the same Monte Carlo data as evidence that LPA's functional form is exact. I would be comfortable with a revised version that either provides an independent out-of-sample test (e.g., fixing r_I and r_zeta on one N or on a subset of zeta and predicting the rest), adds a next-order derivative-expansion estimate, or clearly downgrades the conclusion to 'consistent with a two-scale description.' The N=3 family comparison should be added or the claim restricted. This is a fixable issue rather than a fundamental flaw, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the large-N rate function computation is clean and exact: a full family of universal order-parameter distributions parametrized by zeta, derived from an explicit regulator-independent gap equation. That is new and solid, and the FRG numerics reproduce it to 1e-4 for analytic regulators. Second, the finite-N claim is not as strong as the abstract suggests. The 'excellent' agreement is obtained after fitting two global rescaling factors r_I and r_zeta to the same Monte Carlo data that later certify the <1% agreement. Those fits remove the two largest LPA errors: Delta I_N and zeta_c are both off by about 25%, so the residual shape agreement is real but tests the two-scale ansatz only under the assumption that no other LPA shape error exists. The paper says this plainly, so it is not a hidden flaw, but it is a load-bearing gap. The conclusion that LPA error 'is primarily concentrated' in those two constants is an inference from the two-parameter fit, not a tested property of the approximation.\n\nWhat is genuinely good: the large-N limit is self-contained, the derivation of the rate-function gap equation is explicit enough to reimplement, and the observation that the N=2 and N=3 rate functions nearly collapse is an honest, unexplained result left for future work. The MC comparison is careful, with histogram reweighting and L=96/128 convergence checks, and the paper corrects a statement from their earlier Ising paper about a non-universal amplitude, which is good scholarly hygiene. The LPA equations and numerical scheme are described well enough for independent reimplementation, though no code or data is shipped.\n\nSoft spots beyond the calibration issue: the N=2 vs N=3 near-collapse is not explained, and N=4 being closer to N=infinity muddies the ordering. The <1% relative error in Sec. IV B is demonstrated for N=2 only; N=3 is not shown as a full-family comparison. The stress-test note's concern is real but not disqualifying, because the paper is transparent about the two rescaling factors and the large-N benchmark gives confidence the numerics are sound.\n\nBottom line: this is a solid methods-and-benchmark paper. The exact large-N result deserves citation; the finite-N shape claim should be read with the calibration caveat. No one should cite this as 'LPA predicts the full PDF shape without input from MC.' A serious referee should engage, mainly to push for a next-order check or an independent determination of r_I and r_zeta (say, from a single zeta value) if that is feasible.","headline":"Solid exact large-N benchmark and an honest but conditional finite-N shape claim; the two fitted rescaling factors make the <1% agreement weaker than it looks.","tokens_in":22231,"tokens_out":1753,"would_cite":true,"duration_ms":17116,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82B27","82B28","60F10"],"pacs":[],"model":"deepseek-v4-flash","headline":"The lowest-order functional renormalization group captures the full functional shape of critical order-parameter probability distributions for O(N) models, up to two global rescaling factors.","keywords":["O(N) model","order-parameter distribution","rate function","functional renormalization group","local potential approximation","finite-size scaling","large deviations","critical phenomena"],"falsifier":"Compute the rate function at the next order of the derivative expansion and compare it, after the same two-parameter rescaling, with the LPA curve and with the Monte Carlo data. If the higher-order shape differs from the LPA shape by more than the Monte Carlo statistical error, the claim that LPA captures the exact functional form is false; a direct alternative is to take Monte Carlo data at several zeta values not used to fix r_I and r_zeta, such as zeta = 1 and zeta = 3 for N=2, and check whether the asserted sub-1% agreement persists.","tokens_in":21160,"feed_emoji":"📊","tokens_out":7324,"duration_ms":69699,"temperature":0.7,"pith_summary":"This paper extends a functional renormalization group (FRG) computation of the probability distribution function (PDF) of the order parameter at criticality from the Ising case to the O(N) model. Its central claim is that the Local Potential Approximation (LPA), the lowest order of the derivative expansion, reproduces the whole family of universal scaling functions once the overall amplitude of the logarithm of the PDF and the ratio zeta = L/xi_infinity are each corrected by one N-dependent factor. The family is parametrized by how the thermodynamic limit and the critical point are approached, and it includes a second family reached from the ordered phase. If this claim is right, the nontrivial functional form of these large-deviation distributions is already encoded at the simplest level of the FRG, with all approximation error concentrated in two universal constants.","feed_headline":"Lowest-order RG predicts critical spin PDF shapes to 1 percent","feed_subtitle":"One simple RG approximation gives every critical order-parameter distribution, up to two fitted scales.","key_machinery":"The central object is the scale-dependent rate function I_k(s), obtained as the M -> infinity limit of a modified effective action Gamma_{M,k} that enforces the fixed order-parameter constraint through a large mass term ($M^{2}$/2)(integral_x (phi - s))^2. Within the LPA ansatz, I_k obeys a flow equation that is identical to the effective potential flow except that the zero-momentum mode is removed from the regulator trace, which is what allows the rate function to remain non-convex at finite size. The flow is initialized from the dimensionless fixed-point solution of the effective potential, and a relevant perturbation controls zeta = L/xi_infinity. In the large-N limit the exact solution is encoded by a universal function check F_d(z) with a pole at z = -pi, replacing the function F_d(z) that appears in the effective potential.","core_discovery":"For the three-dimensional O(N) model with N=2 and N=3, the rate function I(s) obtained from the LPA matches Monte Carlo simulations across the entire zeta-family to better than 1% relative error after the rescaling I -> r_I I and zeta -> r_zeta zeta. Equivalently, the LPA error is concentrated in the universal amplitude $\\Delta$ I_N = L^d (I_min - I_0)/N and in the critical ratio zeta_c, while the shape of I(s) as a function of the scaled field s $L^{{(d-2+eta)/2}}$ and of zeta is essentially exact. In the large-N limit the paper derives an exact rate function from a saddle-point gap equation and verifies that its LPA flow reproduces it to about $10^{{-4}}$ relative error, independent of the regulator. The paper also computes the separate family of universal PDFs reached by approaching criticality from the low-temperature phase, and reports a surprising near-coincidence of the N=2 and N=3 rate functions once normalized by N.","pith_inferences":["If the two-parameter absorption of LPA error is generic, the same procedure could convert low-order FRG results for other large-deviation functions, such as work fluctuations or entanglement measures, into quantitatively predictive shapes with only fitted amplitudes.","The near-identity of the normalized N=2 and N=3 rate functions suggests that the N-dependence enters mostly through the two universal constants; a direct test would be a high-precision Monte Carlo study at N=5 and N=6 to see whether the normalized curves continue to cluster.","A stronger test of the central claim would be to repeat the computation at second order in the derivative expansion and check whether the resulting rate-function shape differs from the LPA shape only by the same two rescaling factors; if it does not, the exactness of the LPA shape is an artifact of the lowest order.","The ratios r_I(N) and r_zeta(N) should themselves be universal functions of N and d, so computing them at higher orders could turn the LPA-plus-rescaling scheme into a fully predictive method that needs no Monte Carlo input."],"forward_implications":["For a given N, computing the absolute critical PDF reduces to determining two universal constants, the amplitude Delta I_N and the critical ratio zeta_c; the LPA supplies the full functional form of the family.","The functional shape of the entire zeta-family, including the non-convex regime and the ordered-phase family, is the same for FRG at LPA and Monte Carlo after the two rescaling factors are fixed.","The large-N limit provides a closed-form benchmark in which the LPA flow is exact, validating the numerical scheme and giving a template for benchmarking higher-order approximations.","The near-collapse of the normalized N=2 and N=3 rate functions, with N=4 closer to the large-N curve, shows that the shape depends only weakly on N and can serve as a target for other approximation schemes.","Improving beyond LPA should mainly correct the two universal constants rather than the shape, so next-order derivative-expansion calculations should be testable against Monte Carlo through Delta I_N and zeta_c alone."],"supporting_citations":[{"why":"Introduces the FRG method for rate functions that this paper generalizes from the Ising model to O(N), including the idea of correcting global amplitudes and zeta.","marker":"[49]"},{"why":"Supplies the standard large-N saddle-point derivation of the effective potential that the paper extends to the rate function.","marker":"[48]"},{"why":"Gives the large-N flow equation reasoning used to derive the exact rate-function gap equation in the N to infinity limit.","marker":"[60]"},{"why":"Provides the Wolff cluster algorithm used to generate the Monte Carlo data that serve as the exact proxy for the O(N) models.","marker":"[62]"},{"why":"Supplies the histogram reweighting technique used to extend Monte Carlo rate functions to a range of zeta values.","marker":"[63]"},{"why":"Provide the critical temperatures and correlation-length amplitudes for the O(2) and O(3) models used to set zeta in the Monte Carlo simulations.","marker":"[64–66]"},{"why":"Provides the fourth-order derivative-expansion critical exponents used as reference values showing that the LPA estimates of universal scales are approximate.","marker":"[58]"}],"fun_headline_variants":["LPA rate function matches Monte Carlo to under 1% after two rescalings","Two fit scales make LPA spin PDFs match MC to <1%","LPA matches MC for O(N) spin PDFs to <1%","Large-N O(N) rate function exactly from RG","LPA predicts entire universal PDF family with two fits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that every error made by the Local Potential Approximation appears only as one overall multiplication factor for the logarithm of the PDF and one rescaling of zeta, so that the functional shape itself is exact; if shape errors exist, the excellent collapse in Fig. 10 is produced by the two fitted factors rather than by physics.","fun_headline_variants_meta":{"raw":{"variants":["LPA rate function matches Monte Carlo to under 1% after two rescalings","Two fit scales make LPA spin PDFs match MC to <1%","LPA matches MC for O(N) spin PDFs to <1%","Large-N O(N) rate function exactly from RG","LPA predicts entire universal PDF family with two fits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.002006,"raw_usage":{"total_tokens":7838,"prompt_tokens":971,"completion_tokens":6867,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":587,"completion_tokens_details":{"reasoning_tokens":6774}},"tokens_in":587,"tokens_out":6867,"duration_ms":44088,"temperature":1.0,"reasoning_tokens":6774,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:31:55.525416+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the rate function at the next order of the derivative expansion and compare it, after the same two-parameter rescaling, with the LPA curve and with the Monte Carlo data. If the higher-order shape differs from the LPA shape by more than the Monte Carlo statistical error, the claim that LPA captures the exact functional form is false; a direct alternative is to take Monte Carlo data at several zeta values not used to fix r_I and r_zeta, such as zeta = 1 and zeta = 3 for N=2, and check whether the asserted sub-1% agreement persists.","supporting_citations":[{"cited_title":"Balog, A","cited_arxiv_id":null,"evidence_quote":"Introduces the FRG method for rate functions that this paper generalizes from the Ising model to O(N), including the idea of correcting global amplitudes and zeta."},{"cited_title":"Moshe and J","cited_arxiv_id":null,"evidence_quote":"Supplies the standard large-N saddle-point derivation of the effective potential that the paper extends to the rate function."},{"cited_title":"Tetradis and D","cited_arxiv_id":null,"evidence_quote":"Gives the large-N flow equation reasoning used to derive the exact rate-function gap equation in the N to infinity limit."},{"cited_title":"Wolff, Phys","cited_arxiv_id":null,"evidence_quote":"Provides the Wolff cluster algorithm used to generate the Monte Carlo data that serve as the exact proxy for the O(N) models."},{"cited_title":"Newman and G","cited_arxiv_id":null,"evidence_quote":"Supplies the histogram reweighting technique used to extend Monte Carlo rate functions to a range of zeta values."},{"cited_title":"De Polsi, I","cited_arxiv_id":null,"evidence_quote":"Provides the fourth-order derivative-expansion critical exponents used as reference values showing that the LPA estimates of universal scales are approximate."}],"review_version":1}