{"id":"50fe543f-363f-44d8-91a8-9dc619bf5cf1","arxiv_id":"2412.06473","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Recalculating 98 tenth-order electron g-2 integrals removes the 5-sigma gap between AHKN2020 and Volkov, giving Set V = 6.800(128).","lead":"A long-standing 5 sigma gap between two calculations of the tenth-order QED correction to the electron's magnetic moment is traced to 98 Feynman integrals of one diagram class; after recalculating them with more Monte Carlo statistics, the authors obtain a new value consistent with the independent calculation. The result matters because the electron g-2 is one of the most precise tests of the Standard Model and is used to extract the fine-structure constant.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table I's gap-equation check shows extreme outliers (X350 ~11σ, X352 ~17σ) outside the re-evaluated 98, so the 'no individual discrepancy' premise for the revised 6.800(128) is not supported.","rationale":"The reader's conditional verdict is appropriate, but the most load-bearing concern is not only the reuse of VEGAS or the post hoc selection; it is the internal contradiction in Table I. The paper's verification procedure produces residuals as large as 17σ for diagrams outside the 98 that were re-evaluated. This undermines the claim that the gap-equation check validates all individual diagrams and that the total discrepancy is fully explained by the 98-diagram group. Even if the re-evaluated integrals are unbiased, the overall verification is incomplete because other diagrams show significant disagreements that are neither explained nor addressed. The proposed test would distinguish between two possible failure modes: underestimated Monte Carlo errors or incorrect gap equations. Until this is resolved, the revised central value 6.800(128) remains conditional. I do not change the reader's verdict because the paper already received a conditional recommendation; I strengthen the basis for it by pointing to a specific, internally checkable failure.","tokens_in":24286,"tokens_out":6157,"duration_ms":62628,"concrete_test":"Recompute the AHKN ΔM_G and the Volkov vertex-sum integrals for diagrams X350 and X352 using an independent numerical method (e.g., deterministic adaptive cubature or a non-VEGAS Monte Carlo with a different variable mapping). If the residuals from Table I persist at >5σ, the gap-equation formalism or the quoted error bars are wrong, invalidating the verification. If the residuals vanish, the original VEGAS estimates carry underestimated errors and the 0.128 total uncertainty needs revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on the gap-equation consistency check in Table I, which is asserted to show no apparent discrepancy in any of the 389 diagrams. That assertion is internally contradicted by the table itself: X350 has a residual of 0.0090±0.0008 (~11σ) and X352 has 0.0103±0.0006 (~17σ). Further examples include X382 (~5σ), X329 (~4.3σ), and X227 (~3.6σ). These diagrams are not in the 98 re-evaluated integrals listed in Table II, so the proposed replacement cannot remediate them. If the residuals are genuine, either the quoted uncertainties are badly underestimated (making the final error 0.128 unreliable) or the gap equations constructed in Section V are incorrect for these diagrams (invalidating the comparison method that supports the whole verification). The paper's statement in Section VI that 'there is no apparent inconsistency in any of the 389 diagrams' is therefore not supported by the presented data. The post hoc selection of the 98-diagram subset also lacks a statistical correction for the look-elsewhere effect: with 389 residuals, some large deviations are expected by chance, and the paper does not quantify how unusual the observed sum for the chosen 98 is.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper addresses a known ~5σ discrepancy between two independent numerical evaluations of the tenth-order QED contribution to the electron anomalous magnetic moment from diagrams without fermion loops (Set V): the authors' earlier AHKN2020 value and Volkov's results. The authors decompose the 389 independent tenth-order self-energy diagrams and construct gap equations that relate AHKN's finite magnetic moment amplitudes to Volkov's sums of vertex-diagram amplitudes through finite renormalization-constant differences. A diagram-by-diagram comparison is presented in Table I. The authors find that the numerical differences between the two calculations accumulate in 98 diagrams that contain exactly one second-order self-energy subdiagram and no other self-energy subdiagrams. These 98 integrals are re-evaluated with increased VEGAS statistics, and replacing the old values yields a revised Set V result of 6.800 ± 0.128, consistent with Volkov's combined estimate 6.828(60). The paper concludes that the discrepancy is resolved.","tokens_in":24563,"tokens_out":3070,"duration_ms":35093,"significance":"If the revised value and its uncertainty are correct, this is an important result: it resolves a long-standing tension between two major numerical efforts in tenth-order QED and restores consistency with the independent Volkov calculations. The gap-equation framework is a useful new tool for diagram-by-diagram comparison and is presented with enough detail to be checked. The paper also provides a large computational campaign (about 3.2×10^7 core-hours) and publishes detailed tables of old and new integrals, which is valuable for future work. However, the central claim rests on a post hoc selection of 98 diagrams and on the assumption that the numerical bias in the old AHKN2020 integrals was confined to that class. That assumption is not yet validated with an independent integration method or a control sample, and the paper's own Table I contains residuals far outside the quoted errors, which undermines the assertion that no individual diagram shows a discrepancy.","major_comments":[{"comment":"The statement in Section VI that 'there is no apparent inconsistency in any of the 389 diagrams' is not supported by the numbers in Table I. For example, X350 has a residual of 0.0090 ± 0.0008 (about 11σ), X352 has 0.0103 ± 0.0006 (about 17σ), X382 has 0.0382 ± 0.0076 (about 5σ), X329 has 0.0275 ± 0.0064 (about 4.3σ), and X227 has 0.0341 ± 0.0094 (about 3.6σ). X350, X352, and X382 are not in the 98 re-evaluated integrals listed in Table II, so the proposed replacement cannot remediate those residuals. If these residuals are genuine, either the quoted uncertainties are underestimated or the gap equations are not correct for these diagrams; either way, the claim of diagram-by-diagram consistency needs to be replaced with a statistical treatment of the full set of 389 residuals, including a look-elsewhere correction or a global goodness-of-fit test.","section":"Section VI, Table I"},{"comment":"The identification of the 98-diagram class is made after inspecting the data and is not pre-registered, and the conclusion that the bias is confined to this class is not verified with an independent integration method. The new evaluation uses the same VEGAS algorithm as the old one, with larger statistics and only slightly modified stretching; a correlated bias in the VEGAS integration would not necessarily be detected by this procedure. Furthermore, although the individual old–new differences in Table II are mostly within 1–2σ, 85 of the 98 integrals decrease, and the total shift is −0.8032 ± 0.0717, an approximately 11σ effect. This strongly indicates a systematic effect in the old integrals, but it also raises the question of whether the new integrals carry a residual correlated bias. A control check is needed: for instance, re-evaluating several non-98 diagrams (such as X350, X352, and X382) with the same new setup, or computing a subset with an independent integration method, would test whether the bias is truly confined to the 98-diagram class.","section":"Section VII, Table II"},{"comment":"The final uncertainty of 0.128 in Eq. (9) is the combined statistical error of the new VEGAS evaluations. If the numerical bias in the old integrals is correlated across diagrams that share the same self-energy substructure, the new integrals could carry a common systematic error that is not reflected in this uncertainty. The paper does not provide any estimate of such a systematic uncertainty. To make the central claim load-bearing, the authors should either demonstrate with a control sample that the new evaluations are free of the bias or add a systematic error term to the final result. Without this, the claim that the discrepancy is resolved is conditional on an unverified assumption about the source of the bias.","section":"Eq. (9) and Section VII"}],"minor_comments":[{"comment":"The paper refers to the unpublished AHKN2020 result as a reference value throughout; it would be helpful to state explicitly that this value has not been published in a refereed journal and to describe the exact contents of the dataset that defines AHKN2020.","section":"Section I, Eq. (4)"},{"comment":"The statement that the 98 integrals 'had not been reevaluated with increased sampling points exceeding 1×10^10 using double-double precision' could be clarified: does this mean that no other Set V integrals have ever been evaluated with such statistics, or only that these particular ones had not been? The sentence as written is ambiguous.","section":"Section VII"},{"comment":"The caption says 'Old Value' and 'New Value' but does not state the precision or the exact statistics used for each; adding a note on the number of sampling points per diagram, or at least the range, would improve reproducibility.","section":"Table II"},{"comment":"There are minor typographical issues, such as 'difficulty' and other non-ASCII ligatures in the text, and the reference formatting is inconsistent (e.g., Refs. [20] and [36] have slightly different journal-name styles). These do not affect the content.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses an important problem and provides a plausible resolution, but the central claim requires stronger validation. The extreme residuals in Table I outside the re-evaluated 98 diagrams are a serious concern that the authors need to address, perhaps by re-evaluating those diagrams as a control or by providing a full statistical analysis of the 389 residuals. The post hoc selection and the lack of an independent integration check are the main reasons I cannot recommend acceptance in the current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe new thing here is real: the authors build symbolic gap equations for all 389 Set V self-energy diagrams, introducing finite constants δL_G(i) to convert AHKN's and Volkov's different renormalization schemes into a common language. That is a clever, useful technique, and the fact that the sum of the 389 gap-equation differences reproduces the total 0.780(165) discrepancy is a good bookkeeping check. The revised Set V value, 6.800(128), is also credible in the loose sense that it agrees with Volkov's independent 6.828(60). If this holds up, it settles a long-running 4.5–5σ headache.\n\nThat said, the paper overreaches in its own table. Table I reports residuals between the two calculations and the gap equations. The text says 'no apparent inconsistency in any of the 389 diagrams,' but the table shows X352 at 0.0103(6), roughly 17σ; X350 at 0.0090(8), roughly 11σ; X382 at 0.0382(76), about 5σ; X329 about 4.3σ; X227 about 3.6σ. These are not tiny effects relative to the quoted errors. Either the VEGAS error estimates are badly underestimated, or the gap equations are missing terms for these diagrams. Since those diagrams are not among the 98 re-evaluated in Table II, the replacement cannot fix them. The paper never addresses this. The claim that both calculations are valid for every individual diagram is simply not supported by the data as presented.\n\nThe post hoc selection of the 98-diagram class adds to the concern. The class is identified after examining the data, and there is no look-elsewhere correction for having scanned 389 residuals. The fact that 85 of 98 new integrals shift in the same direction is suggestive, but the statistical significance of that clustering is not quantified against the full set of residuals. And the new integrals use the same VEGAS algorithm, just more samples and slightly modified stretching; an independent integration method would be a much stronger check.\n\nStill, the central numerical conclusion may survive. The shift in the 98 integrals is large enough to move the total into agreement with Volkov, and the final error of 0.128 is not tight enough to be sensitive to the individual outliers. The paper deserves a serious referee, but it needs a careful statistical treatment of Table I and a response to the outliers before the 'verified' language is warranted.\n\nBring it to review, and push the authors to either fix the gap equations or properly account for the tail in their residuals.\n\nBest,\n[You]","headline":"A genuinely new diagram-by-diagram gap-equation audit of the two tenth-order QED calculations, with a credible revised value, but Table I itself contains multi-sigma outliers that contradict the paper's central 'no discrepancy' claim.","tokens_in":25107,"tokens_out":2106,"would_cite":false,"duration_ms":22384,"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":"The paper attributes a 5-sigma disagreement in the electron's magnetic moment to biased Monte Carlo integrals in 98 diagrams, and a higher-statistics recomputation yields a revised value that resolves it.","keywords":["anomalous magnetic moment","electron g-2","QED perturbation theory","tenth order","no-fermion-loop diagrams","self-energy diagram decomposition","gap equation","Monte Carlo integration"],"falsifier":"Recomputing the same 98 integrals with an integration method that does not share VEGAS's variable-stretching heuristics, such as deterministic quasi-Monte Carlo, and comparing the summed value with $57.0023 \\pm 0.0327$ would settle the claim: a significantly different total would show the bias was not removed.","tokens_in":24091,"feed_emoji":"🧲","tokens_out":14347,"duration_ms":129359,"temperature":0.7,"pith_summary":"Two independent numerical calculations of the tenth-order QED contribution to the electron's anomalous magnetic moment from diagrams without electron loops disagreed by about $5\\sigma$, a serious obstacle to using the electron's magnetic moment as a precision test of the Standard Model. This paper attempts to locate the source of that disagreement by decomposing the contribution into 389 self-energy-diagram integrals and checking each one against a 'gap equation' that relates the two calculations' renormalization conventions. The check finds no individual diagram out of line, but the differences of 98 diagrams with a common structure—one second-order self-energy subdiagram and no others—accumulate with a clear sign. Recomputing those 98 integrals with the same Monte Carlo algorithm at higher statistics shifts the total by $-0.8032 \\pm 0.0717$, giving a revised value $6.800 \\pm 0.128$ that agrees with the independent result. If correct, the paper removes a well-known tension in the theoretical prediction of the electron's magnetic moment.","feed_headline":"New electron g-2 calculation ends 5-sigma clash","feed_subtitle":"Tenth-order QED correction revised to 6.800(128), matching independent results and sharpening the Standard Model test.","key_machinery":"The load-bearing object is the gap equation. For each of the 389 independent tenth-order self-energy diagrams $G$, the difference between the finite magnetic moment $\\Delta M_G$ from one calculation and the sum of the vertex-diagram amplitudes $\\sum_i \\Delta M_{G(i)}$ from the other is expressed, via the Ward-Takahashi identity, in terms of lower-order renormalization constants and magnetic moments. Because those lower-order quantities are known to high accuracy, the equation predicts the difference; comparing that prediction with the two numerical integrals checks both calculations diagram by diagram. The classification by self-energy subdiagram structure then isolates the 98 integrals that carry the accumulated bias.","core_discovery":"The paper claims that the $5\\sigma$ disagreement between the two known tenth-order QED contributions to the electron's anomalous magnetic moment from diagrams without fermion loops originates in a systematic bias in the Monte Carlo integration of 98 particular Feynman integrals. These are the diagrams that contain one second-order self-energy subdiagram and no other self-energy subdiagrams. Recomputing those 98 integrals with the same VEGAS algorithm at higher statistics and with slightly modified variable stretching produced a summed shift of $-0.8032 \\pm 0.0717$, changing the total from $7.604 \\pm 0.140$ to $6.800 \\pm 0.128$. The new value is consistent with the independent result $6.824 \\pm 0.089$ and its 2024 update $6.857 \\pm 0.081$, so the authors conclude the discrepancy is resolved.","pith_inferences":["An implication the authors leave implicit is that the earlier gap was a numerical artifact of under-sampled VEGAS integration rather than a sign of new physics or a flaw in renormalization.","Because 85 of the 98 new integrals moved downward, the bias looks directional and tied to the VEGAS variable stretching; this suggests the same auditing procedure could be applied to other slow-converging diagram classes, a step the paper does not take.","A natural testable extension is to recompute the 98 integrals with an independent numerical method; the paper's resolution currently rests on one algorithm run at higher statistics, so such a check would settle whether the bias is truly gone."],"forward_implications":["The revised no-fermion-loop contribution is $6.800 \\pm 0.128$, replacing $7.604 \\pm 0.140$ and agreeing with the independent estimates $6.824 \\pm 0.089$ and $6.857 \\pm 0.081$.","The gap-equation consistency check across all 389 self-energy diagrams validates that both constructions of the Feynman integrals are correct; the discrepancy was in numerical integration, not in the formulation.","The accumulated bias is localized to the 98 integrals containing a single second-order self-energy subdiagram, whose new sum differs from the old by $-0.8032 \\pm 0.0717$.","The electron's tenth-order QED prediction is now compatible with the independent value, so the comparison between the measured magnetic moment and theory no longer contains this $5\\sigma$ tension."],"supporting_citations":[{"why":"Supplies the residual renormalization formula and the lower-order finite quantities that enter the gap equations.","marker":"[20]"},{"why":"The previously published Set V result whose dataset is extended by the later evaluation and whose old integrals are replaced.","marker":"[22]"},{"why":"The independent calculation that produced the 4.7-sigma disagreement used as the comparison target.","marker":"[23]"},{"why":"The later recalculation using a different renormalization scheme; its consistent value shows the discrepancy persists beyond one scheme.","marker":"[24]"},{"why":"Defines the VEGAS Monte Carlo algorithm used both for the old and new numerical integrations.","marker":"[26]"},{"why":"Describes the density and variable-stretching choices for Feynman-parameter integrals, which were slightly modified in the new runs.","marker":"[30]"},{"why":"Defines the K operation used to build the renormalization constants appearing in the gap equations.","marker":"[31]"},{"why":"The forest formula used to remove ultraviolet subdivergences when computing the finite constants that appear in the gap equations.","marker":"[35]"},{"why":"The R subtraction for linear infrared divergences in self-energy subdiagrams, which is the divergence structure of the 98 diagrams.","marker":"[36]"}],"fun_headline_variants":["Electron g-2: 5-sigma clash traced to MC bias","Tenth-order QED revised: 98 diagrams fixed the 5-sigma gap","5-sigma discrepancy in electron anomaly resolved","MC bias in 98 diagrams explains electron g-2 clash","Electron g-2 revised to 6.800(128), resolving 5-sigma"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The resolution rests on the premise that the systematic integration error was confined to the 98 diagrams identified after inspecting the data, and that re-running the same integration routine with more samples and slightly altered stretching removed that error.","fun_headline_variants_meta":{"raw":{"variants":["Electron g-2: 5-sigma clash traced to MC bias","Tenth-order QED revised: 98 diagrams fixed the 5-sigma gap","5-sigma discrepancy in electron anomaly resolved","MC bias in 98 diagrams explains electron g-2 clash","Electron g-2 revised to 6.800(128), resolving 5-sigma"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000202,"raw_usage":{"total_tokens":1356,"prompt_tokens":890,"completion_tokens":466,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":506,"completion_tokens_details":{"reasoning_tokens":369}},"tokens_in":506,"tokens_out":466,"duration_ms":4576,"temperature":1.0,"reasoning_tokens":369,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T19:38:41.442926+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recomputing the same 98 integrals with an integration method that does not share VEGAS's variable-stretching heuristics, such as deterministic quasi-Monte Carlo, and comparing the summed value with $57.0023 \\pm 0.0327$ would settle the claim: a significantly different total would show the bias was not removed.","supporting_citations":[{"cited_title":"Aoyama, M","cited_arxiv_id":null,"evidence_quote":"Supplies the residual renormalization formula and the lower-order finite quantities that enter the gap equations."},{"cited_title":"Volkov, Phys","cited_arxiv_id":null,"evidence_quote":"The independent calculation that produced the 4.7-sigma disagreement used as the comparison target."},{"cited_title":"Volkov, Phys","cited_arxiv_id":null,"evidence_quote":"The later recalculation using a different renormalization scheme; its consistent value shows the discrepancy persists beyond one scheme."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the VEGAS Monte Carlo algorithm used both for the old and new numerical integrations."},{"cited_title":"Kinoshita, Adv","cited_arxiv_id":null,"evidence_quote":"Describes the density and variable-stretching choices for Feynman-parameter integrals, which were slightly modified in the new runs."},{"cited_title":"Cvitanovi´ c and T","cited_arxiv_id":null,"evidence_quote":"Defines the K operation used to build the renormalization constants appearing in the gap equations."},{"cited_title":"Zimmermann, Commun","cited_arxiv_id":null,"evidence_quote":"The forest formula used to remove ultraviolet subdivergences when computing the finite constants that appear in the gap equations."},{"cited_title":"Aoyama, M","cited_arxiv_id":null,"evidence_quote":"The R subtraction for linear infrared divergences in self-energy subdiagrams, which is the divergence structure of the 98 diagrams."}],"review_version":1}