{"id":"8bf4c244-1ae7-4462-911d-a60ab40ee19f","arxiv_id":"2411.18606","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A framework that turns missing higher-order perturbative coefficients into fit-able theory nuisance parameters, giving correlated and statistically meaningful theory uncertainties.","lead":"Physicists often estimate the uncertainty from missing higher-order calculations by varying unphysical renormalization scales. This paper replaces that guesswork with theory nuisance parameters that represent the unknown higher-order terms themselves, so the uncertainty can be correlated, propagated, and even reduced by data.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"In-sample calibration of TNP natural-size distribution leaves the claimed 68% theory CL unvalidated for new quantities.","rationale":"The reader's weakest_assumption correctly identifies the load-bearing point: the statistical interpretation of TNP uncertainties is only as good as the identification in eq (5.11), and the validation in Sec 5.3 is circular to a significant degree. The normalization N_n was chosen to make the known sample O(1), and the same sample is used to confirm the Gaussian shape and unit variance. This is an in-sample calibration, not an out-of-sample prediction. The paper is honest about this being an assumption, but the central claim of the abstract and Sec 3.1 that the approach provides a well-defined statistical interpretation depends on it. The concern is not about internal inconsistency or the general parameterization strategy; it is that the numerical confidence statements (68% theory CL) may not transfer to new predictions. A holdout experiment, using known higher-order coefficients that were not used in the calibration, would directly test this. The qT application is impressive and the correlations are structurally correct, but the statistical calibration is the weakest link. The paper remains CONDITIONAL: the method is promising, but the statistical claims require out-of-sample validation and, ideally, a public implementation of the calibration sample and procedure. I found no other concern that is more load-bearing than this one.","tokens_in":51130,"tokens_out":4406,"duration_ms":89439,"concrete_test":"Hold out all n=4 matrix-element constants and, separately, the qT hard/soft/beam boundary coefficients used in Sec 6.2 from the sample in Sec 5.3.2; recalibrate N_n and the Gaussian fit using only the remaining entries (n<=3 and other series). Then test whether the held-out theta-hat values are consistent with N(0,1): check that the mean is within 2 sigma, the variance is within ~20% of 1, and the empirical coverage of the 68% interval matches 68%. If the coverage is significantly off, the identification in eq (5.11) is not validated for new quantities.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that TNP uncertainties have a meaningful statistical interpretation rests on eq (5.11), which identifies the distribution of an unknown theta_n with the empirical distribution of known normalized coefficients. This identification is explicitly an assumption (Sec 5.3.1), and the calibration is in-sample: the normalization N_n = 4^n C_n (n-1)! in eq (5.4) was chosen precisely so that the known sample has O(1) variance. Sec 5.3.3 admits that the (n-1)! factor was discovered by adjusting the normalization to reduce the variance of the sample. The same sample is then used to validate the N(0,1) distribution. Selection effects in the known-series sample are acknowledged but not quantified. If a new quantity is not drawn from the same population, or if the qT-relevant ingredients in Sec 6.2 (with their different conventions, including the square root for soft functions and the factor 3/2 in eq 6.13) are not representative, the claimed 68% theory CL and Gaussianity do not follow. Without an out-of-sample check, the statistical interpretation is an assumption, not an empirically validated property.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a method for estimating perturbative theory uncertainties by promoting missing higher-order series coefficients to 'theory nuisance parameters' (TNPs) with true but unknown values, instead of varying unphysical renormalization scales. The method is presented in two parts: a general framework (Secs. 2–4) for constructing TNP parameterizations that preserve correlations among related predictions, and a specific application to transverse-momentum (qT) resummation in Drell-Yan production (Sec. 6). The theory constraints on scalar TNPs are derived in Sec. 5 by normalizing known perturbative coefficients and studying the empirical distribution of these normalized coefficients across a collection of known series. The paper argues that the resulting uncertainties have a statistical interpretation (68% theory CL) and correct correlations, and it illustrates the approach with numerical results for Z, W, and their ratios, showing strong cancellation of uncertainties. The application has been used in a CMS W-mass measurement.","tokens_in":51359,"tokens_out":4765,"duration_ms":48374,"significance":"If the statistical interpretation of TNP uncertainties can be sustained, this is a significant methodological advance. The paper correctly identifies a key weakness of scale variations—their inability to provide meaningful correlations or a parametric statistical interpretation—and offers a principled alternative. The qT application is concrete and nontrivial: the correlations across qT, between processes, and between different scales are derived from the known RGE structure of the hard, soft, and beam functions, and the numerical demonstrations in Sec. 6.3 are carefully presented and reproducible in principle. The paper also transparently recognizes that the identity between the distribution of unknown coefficients and the empirical distribution of known coefficients is an assumption (Sec. 5.3.1). However, the load-bearing statistical claim—that the uncertainties have a well-defined 68% theory CL—rests on in-sample calibration: the normalization N_n^f in Eq. (5.4) was adjusted on the same sample used for validation, and the sample selection is acknowledged to be potentially unrepresentative.","major_comments":[{"comment":"The central statistical interpretation hinges on identifying the distribution of an unknown theta_n with the empirical distribution of known normalized coefficients, as in Eq. (5.11), which the paper explicitly states is an assumption. The subsequent validation in Figs. 1–3 is performed on the same sample that was used to determine the normalization N_n^f in Eq. (5.4): Sec. 5.3.3 reports that the factorial factor (n-1)! was discovered by adjusting the normalization to reduce the variance of the sample. Consequently, the fitted mean near zero and sigma near one are in-sample properties, and they do not by themselves establish that a new, not-yet-calculated coefficient will follow the same distribution. This is a load-bearing issue for the abstract's claim of a 'well-defined statistical interpretation' and the '68% theory CL' used throughout Sec. 6. An out-of-sample test (e.g., holding out one or more series from the calibration and testing the prediction on them, or a cross-validation across categories) or an explicit scoping of the claim as conditional on the population assumption should be provided.","section":"Sec. 5.3.1, Eq. (5.11)"},{"comment":"The sample of known perturbative series is acknowledged to be biased toward quantities that are 'naturally simpler to calculate' (Sec. 5.3.2). The paper asserts, based on experience, that this does not affect representativeness, but no quantitative evidence is offered. Since the entire empirical validation of the natural-size normalization and the Gaussian distribution rests on this sample, the representativeness claim needs support. For example, a stability study that removes subsets of closely related series (e.g., all form factors, or all nf-dependent quantities) and checks whether the fitted distribution parameters remain stable would quantify the sensitivity, or the paper could report the evolution of the fitted sigma and mean as new higher-order results have been added over time.","section":"Sec. 5.3.2, Appendix A"},{"comment":"The beam-function TNP parameterization in Eq. (6.13) introduces an ad hoc factor of 3/2 'to be conservative' without a derivation from the same statistical calibration that is used elsewhere in Sec. 5. This factor directly changes the size of the beam-function uncertainty, which is a dominant component in the qT application (especially in ratios, as shown in Figs. 7–8). If the intended meaning is that the uncertainty corresponds to a 68% theory CL, then this factor should be derived from the calibration procedure or its effect on the claimed coverage should be quantified; otherwise, the size of the resulting theory uncertainty is not statistically calibrated.","section":"Sec. 6.2.3, Eq. (6.13)"}],"minor_comments":[{"comment":"The notation N^{m+k}LO, where '1+1' does not equal '2', is introduced in the text but would benefit from a formal definition at first use in Section 2.2 or in a footnote, since it is central to the paper's language and appears throughout.","section":"Sec. 2.2, Eq. (2.5)"},{"comment":"The captions of Tables 2 and 3 clearly identify the bold entries as the true values of the normalized coefficients, but the rows could be made even more explicit by indicating that these bold values are the θ_n values used to build the distributions in Figs. 1–3.","section":"Tables 2 and 3"},{"comment":"The discussion of subleading effects in Sec. 6.4 lists power corrections, quark masses, QED/EW effects, and nonperturbative corrections, but does not mention PDF uncertainties. Since the beam function involves PDFs and their evolution, a brief comment on how PDF uncertainties are separated from the TNP uncertainties of the beam function would be useful.","section":"Sec. 6.4"},{"comment":"The sentence 'the pull t_n is invariant under a linear transformation' is correct, but the distribution of t_n is also affected by the choice of the estimator u_n; the paper could state more precisely that Eq. (5.9) defines the structure of the estimator, not an empirical claim.","section":"Sec. 5.3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a single-author methodological contribution from a leading practitioner, and the TNP approach is already having tangible impact (CMS W-mass measurement). The technical derivation of the qT correlations is solid and well grounded in the known RGE structure. The main weakness is the in-sample nature of the statistical validation: the normalization and the Gaussianity claim are both calibrated and tested on the same set of known series. This is not hidden by the author, but the abstract and conclusions overstate the degree of validation. I recommend major revision rather than rejection because the issue is fixable: either provide an out-of-sample validation (which the author, with access to the sample, can likely do) or explicitly weaken the statistical claims to 'under the assumption that the new coefficient is drawn from the same population as the calibrated sample.' The paper would also be strengthened by quantifying the representativeness of the sample and by deriving the beam-function factor 3/2 from the calibration procedure rather than leaving it as an ad hoc choice."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe short version: this is the most serious attempt so far to turn missing higher-order QCD corrections into genuine nuisance parameters, and the application to qT resummation is already doing real work in the CMS W-mass analysis. Read it if you care about theory uncertainties in fits. The caveat to keep in mind is that the statistical interpretation — the 68% theory CL — is calibrated in-sample.\n\nWhat is actually new: the systematic TNP framework, the parameterization guide in Sec 4, and the qT implementation with correct correlations across qT and between W and Z. The derivation in Secs 2.2 and 3.1 is clean and internally consistent. The paper is also honest: it explicitly states that eq (5.11), identifying the distribution of unknown coefficients with the empirical distribution of known ones, is an assumption. The numerical results show how different TNP components produce different shapes in qT, and the cancellation in W/Z ratios is striking.\n\nThe soft spots, in proportion. The main one is the calibration. The normalization N_n = 4^n C_n (n-1)! in eq (5.4) was chosen to make the variance of the known sample O(1); the same sample is then used to validate the N(0,1) distribution. That is an in-sample fit, and the paper admits it. As a result, the claimed 68% confidence level is best treated as a plausible, empirically motivated prior rather than a rigorously validated coverage property for new quantities. The author acknowledges selection effects but does not quantify them. This matters, but it is not a fatal flaw: the framework does not collapse if the distribution is not exactly Gaussian, it just means the uncertainty band has a weaker meaning than advertised. A public implementation of the TNP variations would help; SCETlib is cited but the modified version is not shipped.\n\nWho is this for: anyone doing precision QCD phenomenology, especially fits to differential spectra. It deserves a serious referee; the methodology is important and the application is nontrivial. I would engage with it.\n\nRecommendation: send to peer review. The referee should push on out-of-sample validation and clarify the statistical interpretation, but this is a real contribution.","headline":"A serious, well-written methodology paper that turns missing higher-order QCD terms into genuine nuisance parameters; the main caveat is that the 68% confidence claim rests on an in-sample calibration, so treat it as a motivated prior rather than a fully validated coverage property.","tokens_in":51918,"tokens_out":2038,"would_cite":true,"duration_ms":21036,"reading_group":"yes","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 argues that the uncertainty from missing higher-order perturbative terms should be parameterized by genuine theory nuisance parameters (TNPs) — parameters with true but unknown values — rather than by varying unphysical…","keywords":["perturbative theory uncertainty","missing higher orders","nuisance parameters","scale variations","theory correlations","QCD resummation","transverse momentum spectrum","statistical interpretation"],"falsifier":"Compute the genuinely next-order coefficient for a set of quantities outside the paper's calibration sample, normalize each with the paper's formulas, and compare the pull distribution to a standard Gaussian; if substantially more than 32% of the pulls have $|\\hat\\theta_n|>1$, or the distribution is strongly non-Gaussian, the 68% 'theory CL' interpretation would be invalid for the general population.","tokens_in":50841,"feed_emoji":"⚛️","tokens_out":7496,"duration_ms":67385,"temperature":0.7,"pith_summary":"The paper argues that the standard way of estimating perturbative theory uncertainties — varying unphysical renormalization scales — is fundamentally unable to provide meaningful uncertainties, because the varied object has no true value and cannot be interpreted as a parameter. The author proposes instead to treat the missing higher-order coefficients of a perturbative series as true theory nuisance parameters (TNPs): well-defined parameters whose true values are unknown, normalized to have natural size O(1), and constrained by theory or data. This makes the theory uncertainty a genuine parametric uncertainty, so it can be propagated, combined, and profiled in fits like any experimental systematic. The paper validates the normalization on a large sample of known QCD series, shows the resulting distribution is approximately Gaussian, and applies the method to the transverse-momentum spectrum of W and Z bosons, where it yields explicit bin-by-bin correlations and strong cancellation in ratios. If correct, this would replace the current ad hoc correlation models for scale variations with a systematically improvable, statistically meaningful treatment.","feed_headline":"Missing QCD higher orders become true nuisance parameters","feed_subtitle":"Theory uncertainties gain correct correlations, profile in fits, and can shrink with data — unlike scale variations.","key_machinery":"The carrying object is the theory nuisance parameter $\\theta_n$, defined through $f_n(\\theta_n)=N_n\\,\\theta_n$, where $N_n$ is a normalization factor chosen so that $|\\hat\\theta_n|\\lesssim 1$ generically. For matrix-element constants the paper takes $N_n = 4^n C_n (n-1)!$ with leading color factor $C_n=C_r C_A^{n-1}$, and for anomalous dimensions $N_n = 4^{n+1} C_{n+1}$; these choices strip off conventional loop factors and factorial growth so the remaining unknown coefficients have natural size $O(1)$. This turns a missing term from an unknown number into a parameter with a true value, enabling error propagation, profiling in fits, and the sharing of 100% correlations between predictions that depend on the same ingredient. In the $q_T$ application, the renormalization-group equations predict the functional dependence on $q_T$, $Q$, and the process, so the remaining unknowns reduce to a small set of scalar anomalous dimensions and boundary conditions, which are exactly the TNPs.","core_discovery":"The central discovery is that the unknown higher-order series coefficients $f_n$ themselves — not the renormalization scale — are the true sources of perturbative theory uncertainty, and that they can be included in the prediction as theory nuisance parameters $\\theta_n$ with normalization $f_n(\\theta_n)=N_n\\,\\theta_n$. The parameterization is constructed so that $\\theta_n$ has a true value $\\hat\\theta_n$, so the prediction $f(\\alpha,\\theta_n)$ is a genuine parametric function; constraining $\\theta_n$ to $0\\pm 1$ then yields a 68% theory-$\\sigma$ uncertainty, and because different predictions share the same TNPs, their uncertainties are 100% correlated where they share the same perturbative ingredient. The paper provides a general parameterization guide for when the coefficient is a function of kinematic or internal variables, a statistical validation of the $0\\pm1$ default constraint based on the empirical distribution of known QCD series (Gaussian with $\\sigma\\simeq0.9$–$1.0$), and a full application to $q_T$ resummation for $Z$ and $W$ production, where seven TNPs produce the correlations across the spectrum and between processes.","pith_inferences":["Editorially, the same logic applies to any truncated expansion whose coefficients have a known natural size, not just QCD: the TNP construction could be carried over to electroweak higher-order corrections or SMEFT power counting, where the paper notes similar strategies already exist.","A concrete test of the Gaussian calibration would be to reserve a hold-out set of series not used in the calibration sample, then check the pull distribution of their next-order coefficients as they become known; a systematic overdispersion would falsify the $0\\pm1$ unit-variance claim.","If TNPs are profiled in fits to precise data, the resulting post-fit values of $\\theta_n$ provide a direct empirical estimate of missing higher-order coefficients; comparing these to the $0\\pm1$ prior would test whether the prior is biased in practice.","The ratio cancellation seen in the W/Z example suggests that any observable that is a ratio of closely related processes will inherit dramatically smaller theory uncertainties; this could be exploited in global SM fits, but the correlation structure depends on correctly assigning shared TNPs, which the paper argues scales cannot do."],"forward_implications":["Theory uncertainties become genuine nuisance parameters that can be profiled in fits, so data can reduce the theory error instead of merely inflating the total uncertainty.","Correlations between bins, processes, and energies are fixed by the sharing of perturbative ingredients; for W and Z production the common TNPs make the leading uncertainties cancel by roughly a factor of ten in the ratio.","Partial or approximate higher-order information can be incorporated as soon as it is known, shrinking the error without waiting for a complete formal next order.","The default $0\\pm1$ theory constraint carries a quantitative statistical meaning (68% theory CL under a Gaussian), matching how experimental systematics are handled.","A systematic parameterization guide is provided, so the method can be extended to other observables and to subleading power corrections as they become relevant."],"supporting_citations":[{"why":"Original paper advocating the TNP approach; the method developed here is presented as its systematic elaboration.","marker":"[12]"},{"why":"High-precision W-boson mass measurement that the TNP model for the $q_T$ spectrum enabled, providing real-world evidence that the method works.","marker":"[1]"},{"why":"Earlier use of TNP parameterizations for QCD splitting functions and parton distributions, supplying a working example and the PDF set used in the numerical application.","marker":"[13]"},{"why":"One of the earlier applications of TNPs to estimate missing perturbative ingredients, supporting the claim that the method has already been used.","marker":"[14]"},{"why":"Another earlier application of TNPs, further supporting the assertion of prior successful use.","marker":"[15]"},{"why":"Forthcoming companion paper using the TNP approach to study theory uncertainties in $\\alpha_s$ extraction from the $Z$ $q_T$ spectrum.","marker":"[16]"},{"why":"First application of the TNP approach to fixed-order single-differential distributions, demonstrating the method's generality beyond resummation.","marker":"[17]"},{"why":"The $q_T$ resummation framework (with its factorization and renormalization-group structure) on which the numerical application is built, providing the functional forms that reduce the unknowns to scalar TNPs.","marker":"[42, 44]"}],"fun_headline_variants":["Missing higher orders become true nuisance parameters","Theory errors as fit parameters, not scale variations","Beyond scale variation: missing terms as nuisance parameters","Profile theory uncertainty with data via nuisance parameters"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire 68% confidence statement rests on the assumption that the true values of the still-unknown coefficients behave statistically like the known coefficients used to calibrate the normalization; if a new coefficient belongs to a different population, the stated confidence no longer holds.","fun_headline_variants_meta":{"raw":{"variants":["Missing higher orders become true nuisance parameters","Theory errors as fit parameters, not scale variations","Beyond scale variation: missing terms as nuisance parameters","Profile theory uncertainty with data via nuisance parameters"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000612,"raw_usage":{"total_tokens":2924,"prompt_tokens":1102,"completion_tokens":1822,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":718,"completion_tokens_details":{"reasoning_tokens":1766}},"tokens_in":718,"tokens_out":1822,"duration_ms":14581,"temperature":1.0,"reasoning_tokens":1766,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:01:23.536274+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the genuinely next-order coefficient for a set of quantities outside the paper's calibration sample, normalize each with the paper's formulas, and compare the pull distribution to a standard Gaussian; if substantially more than 32% of the pulls have $|\\hat\\theta_n|>1$, or the distribution is strongly non-Gaussian, the 68% 'theory CL' interpretation would be invalid for the general population.","supporting_citations":[],"review_version":1}