{"id":"d46136ec-0708-4d5c-9227-65ba05b0d04b","arxiv_id":"2411.10379","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Pade and D-Log approximants recover the hadronic vacuum polarization contribution to the muon anomalous magnetic moment from simulated MUonE data to within 0.06%.","lead":"This paper tests whether Pade and D-Log Pade approximants can turn future MUonE scattering data into a precise value for the hadronic contribution to the muon's magnetic moment. In simulated MUonE data, the approximants recover the input value within 0.06%, suggesting a model-independent route to one of the key uncertainties in the muon g-2 puzzle.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The load-bearing assumption is that Stieltjes convergence theorems, proven for Taylor-coefficient PAs, remain valid for least-squares fits to noisy finite-range data; Section 5 concedes this is not strictly valid and only 'apparently satisfied'.","rationale":"The reader's weakest assumption correctly identifies the central soft spot: the transition from Taylor-coefficient Padé approximants, where Stieltjes convergence theorems are rigorous, to least-squares fits of noisy finite-range data. The paper itself flags this limitation in Section 5, so the concern is not manufactured. My stress test would settle it by checking the method against multiple Stieltjes models and by testing coverage and selection bias; this directly probes the load-bearing claim. The single-model GdR closure test is a secondary weakness, but the primary hinge is the empirical convergence of fitted approximants and the reliability of the quoted uncertainties. I do not see a reason to reject the paper: it is an honest proceedings contribution that demonstrates an internally consistent method on toy data, provides a canonical Taylor-coefficient construction that does satisfy the Stieltjes bounds, and explicitly acknowledges the theoretical gap. The appropriate verdict remains CONDITIONAL: the method is promising, but the fitted-approximant convergence and the resulting uncertainty coverage must be validated before claiming a model-independent extraction from real MUonE data. My read agrees with the reader's verdict, so no adjustment is needed.","tokens_in":7058,"tokens_out":4777,"duration_ms":46977,"concrete_test":"Generate 1000 MUonE-like pseudo-data sets from at least three independent Stieltjes models with different singularity structures (e.g., one single vector-meson pole, one with a broad continuum threshold, and one with two separated thresholds), then apply the exact fit and defect-selection pipeline of Sections 5-6. For each model, check (i) whether the bounding order of Eqs. (6)-(7) holds for the fitted approximants, (ii) whether the accepted-fits median deviates from the model's aHVP_LO by more than the quoted 68% CL, and (iii) whether the defect-discard rate is correlated with the fitted aHVP_LO value. If any model shows a bias exceeding the quoted uncertainty or a bounding violation, the empirical extension of the convergence theorems fails and the model-independence claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that PA and D-Log approximants provide a model-independent, controlled extrapolation of MUonE data to obtain aHVP_LO. The extrapolation guarantee rests on Stieltjes convergence and bounding theorems (Eqs. 6-7), which are proven when the approximant is built from the Taylor coefficients of a Stieltjes function. In the MUonE application, however, the approximants are obtained by least-squares fits to 30 noisy data points over a finite interval. Section 5 explicitly concedes: 'the convergence theorems, which are not strictly valid when the approximants are not built from the Taylor series, are apparently satisfied in all these cases.' This is an empirical assertion, not a proven property, and it is the hinge of the extrapolation beyond x = 0.93. The problem is amplified by the fit-selection procedure: 30% of P2_2 fits and 56% of P3_2 fits are discarded as defects, and the final value is the median of accepted fits. If defect occurrence correlates with the true underlying value, the accepted-fits median is biased and the quoted 68% CL (e.g., 6987+46-34) would not have the claimed coverage. The method's model-independence and uncertainty calibration therefore rest on an unverified extension of the Stieltjes theory.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Padé and D-Log Padé approximants as a model-independent framework for extracting the hadronic vacuum polarization contribution to the muon anomalous magnetic moment from simulated MUonE data. The authors use the Greynat–de Rafael model as a benchmark truth, generate 1000 toy data sets over the expected MUonE kinematic range x in [0.2, 0.93], and fit low-order approximants with Stieltjes-inspired constraints. The median of the accepted fits for the highest-order approximants yields aHVP_LO = (6987 +46 -34) x 10^-11 for PAs and (6988 +48 -39) x 10^-11 for D-Logs, both within ~0.1% of the model benchmark 6992.4 x 10^-11. The central claim is that Stieltjes convergence and bounding theorems (Eqs. 6–7) carry over to least-squares fits from noisy finite-range data, making the extrapolation beyond x = 0.93 controlled.","tokens_in":7359,"tokens_out":2721,"duration_ms":21193,"significance":"If the claim holds, the method is directly relevant to the MUonE physics program and offers an alternative to fixed-shape fitting functions for an experiment whose statistical reach demands controlled extrapolation. The paper's strengths are that the toy-data pipeline is internally consistent, the benchmark comparison is explicit, and the analysis is an honest exploration of a real methodological question: whether Stieltjes convergence theory, proven for Taylor-coefficient approximants, survives the transition to least-squares fits to finite-range noisy data. This is an important question for the community and the paper is a clear, readable contribution to that discussion.","major_comments":[{"comment":"The load-bearing premise of the paper is that the Stieltjes convergence theorems, stated in Equations (6) and (7), remain valid when PAs and D-Logs are fitted to 30 noisy data points over a finite interval rather than built from the Taylor coefficients. Section 5 itself concedes that these theorems \"are not strictly valid when the approximants are not built from the Taylor series,\" and cites only empirical evidence from other analyses. This is a correctness-risk concern that the manuscript acknowledges but does not quantify. I would like to see either (a) a direct numerical test of the monotone bounding structure in the actual fitting setup, e.g., checking the inequalities of Eq. (6) hit-by-hit on the toy data and reporting the rate and size of violations, or (b) a statement that the claimed coverage of the quoted 68% CL intervals is not guaranteed by the present evidence, and therefore that the final uncertainties should be treated as indicative rather than as rigorous confidence intervals.","section":"Section 5, Eq. (6)–(7)"},{"comment":"The procedure discards 30% of P2_2 fits and 56% of P3_2 fits as defects before quoting the median as the final result. The paper does not address whether the defect rate is correlated with the underlying value of the extrapolated integral. If the defect probability depends on the value being fitted, the median of accepted fits is a biased estimator and the quoted 68% CL (e.g., 6987+46-34) would not have nominal coverage. I would ask the authors to report, at minimum, the means of the accepted and rejected distributions, and to demonstrate on the toy data that the accepted-fits median retains the claimed coverage (for example, by reporting the empirical coverage of the 68% intervals across the 1000 pseudo-experiments).","section":"Section 6, defect-discard rates"},{"comment":"The benchmark truth is the Greynat–de Rafael model, and the pseudo-data are generated from that same model. The test is therefore a closure test within one function class: it demonstrates that the method recovers a Stieltjes function when the true function is Stieltjes, which is a necessary but not sufficient condition for the method to perform well on real hadronic vacuum polarization. I do not regard this as circular in a damaging sense, but the paper should state more explicitly that the claimed model independence is only tested within the Stieltjes class and that no non-Stieltjes or near-Stieltjes stress test (e.g., adding a small non-Stieltjes component or a further singularity beyond the data region) is performed in this work.","section":"Section 4, GdR model reliance"}],"minor_comments":[{"comment":"The term \"modified chi-squared function\" is introduced without specifying what is modified. Please state the exact definition of the objective function used in the fits.","section":"Section 5, paragraph 1"},{"comment":"The table caption lists chi2/n.d.o.f. distributions, but the quoted values are given as medians with 68% CL intervals. Please explicitly state that these are medians of the distribution over the 1000 pseudo-experiments, to avoid confusion with the per-fit chi-squared value of a single fit.","section":"Section 6, Table 1"},{"comment":"The description of the pseudo-data generation says errors range from 0.7% to 6.7% but does not state the bin widths or the exact x-binning beyond \"equally distributed.\" Please provide the explicit x-values of the 30 data points, as the extrapolation dependence on the last bin is relevant to the result.","section":"Section 6, paragraph 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a proceedings-style contribution that transparently reports the method and its limitations, and it may be of interest to the MUonE community. My main concern is that the central claim of a controlled extrapolation rests on an unproven and under-tested extension of the Stieltjes convergence theorems to the least-squares setting, combined with a defect-discard procedure whose bias properties are not examined. These issues are fixable within the scope of the paper, but the current version does not yet make the case convincingly. I would also note that the reference to the full paper [13] is appropriate and may relieve some of the pressure on the present text, but the proceedings should still stand on its own for the claims it makes."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a look if you're tracking MUonE analysis tools, but this is a proceedings digest of the authors' PRD arXiv:2405.13638. The toy-data demonstration is internally consistent: they generate 1000 pseudo-data sets from the Greynat-de Rafael model, fit Pade and D-Log approximants, and recover the benchmark within 0.06%. The bounding sequences in Eqs. (6)-(7) follow from Stieltjes theory and are laid out clearly. Credit is also due for the transparency about the extrapolation region x > 0.93 and for reporting defect-discard rates.\n\nThe soft spots are the load-bearing ones. First, the convergence theorems that justify the extrapolation are proven for Pade approximants built from Taylor coefficients of a Stieltjes function. Here the approximants are least-squares fits to 30 noisy points, and Section 5 concedes the theorems are 'not strictly valid' and only 'apparently satisfied.' That is an empirical hope, not a proven property. If it fails on real MUonE data, the extrapolation beyond x = 0.93 is uncontrolled. Second, the fit selection: 30% of P2_2 fits and 56% of P3_2 fits are discarded as defects, and the final value is the median of accepted fits. The paper does not examine whether defect occurrence correlates with the underlying value; if it does, the quoted 68% CL has no guaranteed coverage. Third, the test uses the same model that generated the pseudo-data, so it is a closure test within one function class, not a demonstration of model independence.\n\nNone of this is fatal to the method, since it is compatible with the broader literature on Pade approximants used as fitting functions, but the proceedings paper is not the place to resolve it; the full analysis is in Ref. [13]. The paper is honest about its own limitations, and the writing is clear.\n\nThis is for the g-2/MUonE hadronic-vacuum-polarization community. It is a useful overview and a fair advertisement for the PRD, but as a stand-alone proceedings it is an extended abstract. As a journal submission it would be desk-rejected for redundancy; as a proceedings contribution it is within scope. If the authors ever make a full paper out of this, the fit-selection bias is what I would push on first.","headline":"A competent proceedings summary of a solid PRD paper; the MUonE context is new but the method's extrapolation guarantee rests on an empirical extension of Stieltjes convergence theorems and unexamined fit-selection bias.","tokens_in":703,"tokens_out":883,"would_cite":false,"duration_ms":35563,"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":"Padé and D-Log Padé approximants recover the muon's hadronic contribution from MUonE data to better than 0.06%.","keywords":["Padé approximants","D-Log Padé approximants","MUonE experiment","hadronic vacuum polarization","muon g-2","Stieltjes functions","space-like running of alpha","model-independent extrapolation"],"falsifier":"If a reanalysis of the actual MUonE data, or a stress test with an independent model of the hadronic vacuum polarization, finds that the bracketing sequence of the highest-order approximants does not contain the value obtained by lattice QCD or by the dispersive e+e- method, or that the median deviates by more than the quoted 68% intervals, then the empirical extension of the convergence theorems is falsified and the extrapolation beyond x=0.93 is uncontrolled.","tokens_in":6853,"feed_emoji":"🧲","tokens_out":11833,"duration_ms":101423,"temperature":0.7,"pith_summary":"The paper aims to establish that Padé and D-Log Padé approximants, built from the known analytic properties of the hadronic contribution to the running electromagnetic coupling $\\Delta\\alpha_{\\rm had}(t)$, can turn the narrow kinematic window of the MUonE experiment into a full, model-independent determination of the leading-order hadronic vacuum polarization $a_\\mu^{\\rm HVP,LO}$. Because $\\Delta\\alpha_{\\rm had}(t)$ is a Stieltjes function, sequences of these approximants are claimed to bracket the true function, and the paper shows on 1000 toy data sets with realistic MUonE bin errors that the highest-order approximants reproduce the reference model with deviations below 0.06% for Padé and 0.05% for D-Log approximants. The practical payoff is a way to use MUonE's $e\\mu$ scattering data to test the current tension between lattice and dispersive determinations of the muon's anomalous magnetic moment.","feed_headline":"Padé fits hit the muon's hadronic term to 0.06%","feed_subtitle":"By exploiting the analytic structure of the vacuum polarization, it reaches beyond MUonE's narrow window without a fixed model.","key_machinery":"The mechanism is the Stieltjes integral representation $\\Delta\\alpha_{\\rm had}(t)=\\int_0^\\infty d\\phi(u)/(1+t\\,u)$, in which the hadronic vacuum polarization is an analytic function generated by a positive spectral measure. A Padé approximant $P_N^M(t)=Q_N(t)/R_M(t)$ is a ratio of polynomials matched to the Taylor expansion, and for Stieltjes functions its poles are real and positive; the sequences $P_{M+k}^M$ with $k\\ge -1$ bound the function from above and below. A D-Log Padé approximant $D_N^M(t)=f(0)\\exp(\\int \\bar P_N^M(t)\\,dt)$ instead approximates the logarithmic derivative first, turning branch cuts into simple poles and yielding an unbiased estimate of the position and multiplicity of the hadronic cut; it reproduces the first $M+N+2$ Taylor coefficients of $f$. These two objects, together with the Stieltjes property of $\\Delta\\alpha_{\\rm had}$, carry the extrapolation strategy and supply the systematic interpretation of the spread among approximants.","core_discovery":"The central claim is that $\\Delta\\alpha_{\\rm had}(t)$ is a Stieltjes function of $t$, so rational approximants inherit guaranteed bounding and convergence properties: the sequences $P_1^1\\le P_2^2\\le\\cdots\\le\\Delta\\alpha_{\\rm had}\\le\\cdots\\le P_3^2\\le P_2^1$, and analogous D-Log sequences, hold. Fitting these approximants to simulated MUonE data in $x\\in[0.2,0.93]$ and extrapolating to $x=1$ gives $a_\\mu^{\\rm HVP,LO}$ medians of $6987^{+46}_{-34}\\times10^{-11}$ (Padé) and $6988^{+48}_{-39}\\times10^{-11}$ (D-Log), inside the pseudo-data distribution $6991^{+22}_{-20}\\times10^{-11}$ and within 0.06% and 0.05% of the reference model. The paper presents this as evidence that the approximants, applied as least-squares fitting functions rather than as Taylor-coefficient constructions, still satisfy the Stieltjes convergence pattern and provide competitive, conservative uncertainties whose dominant part comes from the extrapolation beyond the data region.","pith_inferences":["Editorial inference: if the convergence pattern holds on real data, the difference between the Padé final value and the D-Log final value could be treated as a direct, statistic-free estimate of extrapolation systematics, a quantity dispersive and lattice determinations do not currently quote.","Editorial inference: the same D-Log machinery should transfer to any space-like observable that is a Stieltjes function with branch points, such as semileptonic form factors or hadronic tau spectral functions; one could test this by deliberately fitting with a wrong model and checking whether the bracketing sequence still brackets the input.","Editorial inference: the high defect rates quoted for low-order Padé approximants (roughly 30–56% discarded for $P_2^2$ and $P_3^2$) while D-Logs defected far less suggest that D-Logs may be the safer default for a real MUonE analysis; this is an inference from the paper's reported discard rates, not a stated conclusion.","Editorial inference: a decisive test would use a second independent model, not the reference model that generated the toy data; if the 0.05–0.06% agreement persists, the method is less dependent on the benchmark model's particular analytic structure than a skeptical reader might fear."],"forward_implications":["If the bracketing pattern survives real MUonE data, the spread between the highest-order Padé and D-Log approximants gives a data-driven systematic error for $a_\\mu^{\\rm HVP,LO}$ that requires no external hadronic model.","The method converts the experiment's blind spot beyond $x=0.93$ into a controlled extrapolation: truncating at $x_{\\max}=0.990$ covers 99.1% of the integral and lowers the uncertainty by about 25%, a concrete trade-off for the experiment's design.","A single fixed fitting function is not needed; the framework is claimed to be superior because its model dependence is explicit and its uncertainty estimate conservative.","With only 30 bins over $x\\in[0.2,0.93]$, the reported precision indicates that analytic constraints, not data density, control the final answer, so the method could serve as a cross-check of lattice HVP results."],"supporting_citations":[{"why":"defines the MUonE proposal and the accessible kinematic range x in [0.2,0.93] that forces the extrapolation.","marker":"[7–9]"},{"why":"supplies the reference model for the hadronic vacuum polarization used to generate the toy data and the benchmark value 6992.4 x 10^-11.","marker":"[12]"},{"why":"is the companion full analysis containing the complete methodology and the detailed convergence-pattern claims summarized here.","marker":"[13]"},{"why":"states the Padé approximant definition and the convergence theorems for Stieltjes functions that anchor the bounding inequalities.","marker":"[16]"},{"why":"provides the classical results on pole structure for Padé approximants to Stieltjes functions.","marker":"[17]"},{"why":"applies Padé theory to the vacuum polarization, establishing the Stieltjes framework used here.","marker":"[20]"},{"why":"derives the model-independent parametrization of the hadronic vacuum polarization and the Stieltjes property in the t variable.","marker":"[21]"},{"why":"is the g_mu minus 2 application of rational approximants as fitting functions that motivates the empirical extension to least-squares fits.","marker":"[27]"}],"fun_headline_variants":["Padé approximants tame MUonE extrapolation to 0.06%","D-Log Padé matches muon hadronic to 0.05%","Model-free Padé fits hit muon hadronic to <0.1%","Padé and D-Log approximants probe muon hadronic within 0.06%","Analyticity-based Padé fits reach 0.06% on muon hadronic"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Stieltjes convergence and bounding theorems, proven only for approximants built from Taylor coefficients, still govern least-squares fits to noisy, finite-range data; a secondary premise is that the reference model used to generate the toy data is a realistic stand-in for the true hadronic vacuum polarization.","fun_headline_variants_meta":{"raw":{"variants":["Padé approximants tame MUonE extrapolation to 0.06%","D-Log Padé matches muon hadronic to 0.05%","Model-free Padé fits hit muon hadronic to <0.1%","Padé and D-Log approximants probe muon hadronic within 0.06%","Analyticity-based Padé fits reach 0.06% on muon hadronic"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000972,"raw_usage":{"total_tokens":4152,"prompt_tokens":986,"completion_tokens":3166,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":602,"completion_tokens_details":{"reasoning_tokens":3055}},"tokens_in":602,"tokens_out":3166,"duration_ms":24453,"temperature":1.0,"reasoning_tokens":3055,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:41:25.702773+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"If a reanalysis of the actual MUonE data, or a stress test with an independent model of the hadronic vacuum polarization, finds that the bracketing sequence of the highest-order approximants does not contain the value obtained by lattice QCD or by the dispersive e+e- method, or that the median deviates by more than the quoted 68% intervals, then the empirical extension of the convergence theorems is falsified and the extrapolation beyond x=0.93 is uncontrolled.","supporting_citations":[{"cited_title":"Hadronic Vacuum Polarization and the MUonE proposal","cited_arxiv_id":"2202.10810","evidence_quote":"supplies the reference model for the hadronic vacuum polarization used to generate the toy data and the benchmark value 6992.4 x 10^-11."},{"cited_title":"Model-independent extrapolation of MUonE data with Pad\\'e and D-Log approximants","cited_arxiv_id":"2405.13638","evidence_quote":"is the companion full analysis containing the complete methodology and the detailed convergence-pattern claims summarized here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"states the Padé approximant definition and the convergence theorems for Stieltjes functions that anchor the bounding inequalities."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the classical results on pole structure for Padé approximants to Stieltjes functions."}],"review_version":1}