{"id":"a514c42a-9d9a-4675-8423-0e23ec28d4ba","arxiv_id":"1908.11394","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The CTEQ-TEA collaboration releases CT18, a new global fit of nucleon parton distributions at NNLO accuracy based on HERA and LHC data.","lead":"A major collaboration has produced new maps of the quarks and gluons inside the proton, called CT18 parton distribution functions, by combining the latest accelerator data from HERA and the LHC. The result gives researchers a standard reference for predicting what happens when protons collide at high energy.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The x-dependent factorization scale is tuned on the HERA data it is then claimed to improve, so the resummation-mimicry result is in-sample and unvalidated.","rationale":"The reader identified the x-dependent scale tuning as the weakest assumption; the paper's Section II explicitly confirms that the scale coefficients are fit to the HERA data. Because the same data are then used to demonstrate the chi2 improvement, the improvement is in-sample and cannot independently validate the scale as a resummation surrogate. This is the load-bearing issue for the paper's 'comparable improvement' claim and for the small-x behavior of CT18X/Z. However, the central artifact, the CT18 PDF release, is supported by the reported global fit and public availability, so the overall CONDITIONAL verdict stands. I recommend no change.","tokens_in":10680,"tokens_out":3110,"duration_ms":29710,"concrete_test":"Perform a cross-validation: split the HERA I+II inclusive DIS data into two halves (e.g., by random bins or by x/Q cells), re-optimize the three coefficients in mu^2_F,x on the training half, and evaluate the chi2 difference (x-dependent vs. nominal scale) on the held-out half. If the held-out improvement falls well short of the reported >50 units, the improvement is overfitting and the resummation-mimicry claim should be downgraded.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II states that the coefficients in mu^2_F,x = 0.82 (Q^2 + 0.3 GeV^2/x^0.3) are chosen to minimize chi2 for the HERA DIS data, and then uses the same HERA data to report a >50 unit chi2 improvement relative to the fixed-Q^2 scale. This is an in-sample, fitted improvement, not an out-of-sample validation; it therefore does not establish that the x-dependent scale is a legitimate proxy for low-x resummation as claimed. The small-x gluon and sea quark PDFs in the CT18X/Z variants are shifted by this ad hoc scale, so the part of the 'progress' claim that relies on those variants is not data-driven. The comparison with resummed fits of Refs. [15,16] is thus a comparison of two tuned descriptions, not a confirmation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents the CTEQ-TEA collaboration's new CT18 global QCD analysis at NNLO, which is offered as the successor to CT14. Four PDF families are released (CT18, CT18A, CT18X, CT18Z), based on a global fit to 3681 data points for the default set with chi^2/Npt = 1.17 at NNLO. The paper describes the inclusion of new LHC data on jets, W/Z, Drell-Yan, and top-quark pair production, along with methodological advances (PDFSense, ePump, parallelized fitting). A central feature is an x-dependent factorization scale mu_F,x^2 = 0.82(Q^2 + 0.3 GeV^2 / x^0.3) used for DIS, which is reported to reduce the HERA I+II chi^2 by more than 50 units relative to the conventional scale mu_F^2 = Q^2, an improvement described as comparable to the inclusion of low-x resummation. The paper also compares PDF luminosities at the 13 TeV LHC with other groups and presents Lagrange Multiplier scans for the gluon and strangeness PDFs.","tokens_in":10928,"tokens_out":3992,"duration_ms":42815,"significance":"If the CT18 analysis is sound, it represents a valuable update to a widely used PDF set, incorporating a large body of new LHC data and improved fitting methodology. The public availability of the PDF grids and the careful comparisons with other groups are clear strengths. The x-dependent scale observation is intriguing: if it were shown to be a theoretically legitimate proxy for low-x resummation, it would be an important result for small-x phenomenology. However, the evidence presented in the manuscript does not currently establish that claim, because the scale's coefficients are tuned on the same HERA data used to demonstrate the chi^2 improvement. The four-family spread is also presented as a way to explore PDF uncertainty, but the selection of the alternative variants is driven by post-hoc data choices, so this spread is not a calibrated uncertainty band. These issues do not undermine the basic existence of the CT18 fit, but they do affect the strength of the paper's central 'progress' claims.","major_comments":[{"comment":"The paper states that the numerical coefficients in mu_F,x^2 = 0.82(Q^2 + 0.3 GeV^2 / x^0.3) are 'chosen to minimize chi^2 for the HERA DIS data', and then uses the same HERA I+II data to report a chi^2 reduction of more than 50 units. This is an in-sample, fitted improvement, not an out-of-sample validation. The comparison with the resummed fits of Refs. [15,16] is therefore a comparison of two tuned descriptions of the same data, and it does not by itself establish that the x-dependent scale is a legitimate proxy for low-x resummation. I request either an out-of-sample test (for example, fitting the scale parameters to a subset of the data and evaluating on a held-out subset, or checking consistency against independent low-x data sets not used in the tuning) or an explicit reframing of this result as an exploratory observation that does not carry the weight of a validation of resummation mimicry.","section":"Sec. II, 'Combined HERA I+II DIS data and an x-dependent factorization scale'"},{"comment":"The decision to exclude the ATLAS 7 TeV W/Z data from the default CT18 fit and include it only in CT18Z, and to remove CDHSW in CT18Z, is made after examining the tension indicators (the SE distribution in Fig. 8). Such post-hoc choices mean that the spread among CT18, CT18A, CT18X, and CT18Z is not a systematically defined PDF uncertainty band. The quoted chi^2/Npt values are conditional on these decisions. If the four families are intended to 'explore the full range of PDF behavior consistent with the available hadronic data', the manuscript should either specify selection criteria that are defined before inspecting the data or clearly label the family spread as a sensitivity study rather than an uncertainty envelope. As written, the claim is too strong for the presented methodology.","section":"Sec. I and Sec. II, data set selection (ATLAS 7 TeV W/Z, CDHSW)"},{"comment":"Several details that are essential for independently assessing the CT18 determination are deferred to an upcoming publication: the complete list of fitted data sets with their kinematic cuts and treatment of correlated systematic uncertainties, the explicit functional forms of the PDF parametrizations (beyond the mention of Bernstein polynomials), the precise definition of chi^2 including normalization penalties, and the exact source of the '0.5% uncorrelated error'. For a paper that proposes CT18 as a replacement for CT14, these are load-bearing for reproducibility. If this is intended as a preliminary proceedings contribution, that status should be stated clearly and the full record should be made permanently accessible, for example in a public repository, so that the CT18 result can be independently validated rather than resting on the upcoming publication.","section":"General (Sections I and II)"}],"minor_comments":[{"comment":"The phrase 'multi-prone effort' should be 'multi-pronged effort'.","section":"Sec. II, 'Advancements in fitting methodology'"},{"comment":"The typesetting of chi^2 as 'χ2' is inconsistent; the exponential form 'χ²' would be clearer.","section":"Throughout"},{"comment":"The sentence 'The right Fig. 7 shows the χ2/Npt values' should be 'The right panel of Fig. 7 shows...' for clarity.","section":"Sec. II, Fig. 7 caption"},{"comment":"The formula for mu_F,x is given in the text without an equation number; adding a numbered equation would aid referencing in the discussion.","section":"Sec. II, x-dependent scale equation"},{"comment":"The phrase 'take charm pole mass to be 1.4 GeV, instead of the nominal value of 1.3 GeV' could be made more precise by stating whether this is the pole mass in the MS-bar scheme or a specific threshold parameter; a brief definition would avoid ambiguity.","section":"Sec. I, description of CT18Z variants"}],"recommendation":"major_revision","confidential_remarks":"This paper appears to be a conference proceedings contribution announcing the CT18 PDF release, and the CT18 grids are already public and widely used. The basic fit is credible, and the methodological improvements are useful. However, the manuscript goes beyond a simple report by making a physics claim about the x-dependent scale mimicking resummation, and that claim is currently supported only by an in-sample chi^2 improvement. The authors may be able to fix this with a moderate revision that reframes the claim or adds a cross-check. I do not see grounds for rejection, but the revision should be more than cosmetic."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"CT18 is a genuine update with real shelf life: a new global PDF fit with eleven new LHC data sets, a more flexible parametrization, and four released ensembles. The PDF sets are public. The work is careful about data tensions and parametrization bias—they tried 90+ functional forms—and the Lagrange-multiplier scans are informative. That part is credible.\n\nThe weak spot is exactly what the stress-test note flags. Section II says the coefficients in mu^2_F,x = 0.82(Q^2 + 0.3 GeV^2/x^0.3) are chosen to minimize chi^2 for the HERA DIS data, and then the same HERA data produce the reported >50-unit chi^2 improvement. So the improvement is in-sample. Comparing it with the low-x resummation fits of Refs. [15,16] is comparing two tuned descriptions, not an independent confirmation. To the paper's credit, it is explicit about the tuning—\"chosen to minimize chi^2\"—but the surrounding text invites a stronger reading. The small-x gluon and strangeness shifts in the CT18X/Z variants should be treated as exploratory, not as evidence that fixed-order NNLO plus a tuned scale reproduces resummation.\n\nOther soft spots are proportionate. The ATLAS 7 TeV W/Z tension is handled by defining a separate family (CT18Z) rather than resolved; CDHSW is dropped in that family. That is defensible for a PDF release, but it means the paper deliberately offers a family of fits rather than one definitive answer. A lot of the methodology is deferred to a promised comprehensive paper, so a referee cannot fully verify the downstream claims from this text alone.\n\nBottom line: this deserves a serious referee. The central artifact—the CT18 PDF ensemble—is plausible, public, and useful for LHC phenomenology. The resummation-mimicry claim is the part to read with skepticism. If this goes to peer review, the main thing I would ask for is an out-of-sample check: fix the scale on a subset of the HERA data, test on held-out data, and show that the improvement is not just a flexible parametrization absorbing noise.","headline":"CT18 is a solid public PDF update, but the x-dependent scale improvement is fitted on the same HERA data it is then used to describe.","tokens_in":11408,"tokens_out":2386,"would_cite":true,"duration_ms":25711,"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 CT18 global analysis claims that an x-dependent factorization scale reproduces the HERA-data improvement attributed to low-x resummation.","keywords":["Quantum chromodynamics","parton distribution functions","global QCD analysis","next-to-next-to-leading order","Hessian method","deep-inelastic scattering","factorization scale","LHC data"],"falsifier":"Take the tuned scale $\\mu^2_{F,x}$ with its published coefficients and apply it, without refitting, to independent low-x DIS data not included in the CT18 fit, such as future EIC structure-function measurements; if the out-of-sample improvement is not comparable to the reported >50-unit gain on HERA, the scale is absorbing fit noise rather than the low-x logarithms it is claimed to reproduce.","tokens_in":10402,"feed_emoji":"📐","tokens_out":12288,"duration_ms":106981,"temperature":0.7,"pith_summary":"This paper introduces the CT18 family of parton distribution functions, the new NNLO global QCD analysis based on 3681 data points with a default chi-squared per point of 1.17. The new PDFs include a wide range of LHC measurements—W and Z production, Drell-Yan, jets, and top-quark pairs—and replace the earlier CT14 set. The central physics claim is that an x-dependent factorization scale, $\\mu^2_{F,x}=0.82(Q^2+0.3\\,\\mathrm{GeV}^2/x^{0.3})$, improves the fixed-order NNLO description of the combined HERA deep-inelastic data by more than 50 units of chi-squared, comparable to what low-x resummation was previously reported to achieve. If this is correct, standard NNLO tools can describe small-x DIS without explicit resummation, and the small-x gluon and sea-quark shapes in CT18 shift accordingly. Because the ATLAS 7 and 8 TeV W/Z data cannot be fitted together with the rest of the world data, four PDF families—CT18, CT18A, CT18X, CT18Z—are provided to bracket the resulting ambiguity.","feed_headline":"A tuned scale mimics low-x resummation, cutting HERA chi-squared by 50","feed_subtitle":"CT18 fit reports chi-squared per point 1.17 on 3681 points; the x-dependent scale does resummation's work.","key_machinery":"The central mechanism is the $x$-dependent factorization scale $\\mu^2_{F,x}=0.82\\,(Q^2+0.3\\,\\mathrm{GeV}^2/x^{0.3})$, used in place of the conventional $\\mu_F^2=Q^2$ when evaluating NNLO DIS cross sections. It is intended as a fixed-order proxy for enhanced low-$x$ logarithms; the paper reports that it lowers $\\chi^2$ for the HERA I+II inclusive data by more than 50 units in the stated region, matching the improvement attributed to resummation. The supporting apparatus is the Hessian global-fit machinery: fast grid interfaces (APPLgrid and fastNLO) for cross sections, the PDFSense and ePump programs for deciding which data sets constrain the fit, Lagrange Multiplier scans for PDF constraints, and flexible Bernstein-polynomial parametrizations tested over more than 90 functional forms.","core_discovery":"The paper claims that the CT18 family of NNLO parton distribution functions, fitted to 3681 data points with $\\chi^2/N_{\\mathrm{pt}}=1.17$ in the default set, supersede CT14 and CT14HERA2 for LHC phenomenology. Its central physics result is that an $x$-dependent factorization scale in fixed-order NNLO deep-inelastic scattering reproduces the HERA-data improvement previously attributed to low-$x$ resummation, a reduction of more than 50 units of $\\chi^2$ in the kinematic region $Q>2$ GeV and $x>10^{-5}$, while also moving the small-$x$ gluon and sea quarks. Because the high-precision ATLAS $W/Z$ measurements cannot be accommodated together with the rest of the world data, four PDF families (CT18, CT18A, CT18X, CT18Z) are presented; the spread among them is the paper's estimate of the present ambiguity in NNLO PDFs.","pith_inferences":["Beyond the paper, the central claim is directly testable out of sample: take the fixed coefficients of $\\mu^2_{F,x}$ and apply them to low-x DIS data not used in the fit; the paper reports no such held-out test.","The claimed equivalence with resummation is demonstrated only for inclusive HERA data; a natural extension would be to check charm and bottom structure functions or transverse-momentum distributions, where resummation and a scale change can diverge.","If the tuned scale is instead a phenomenological knob, then the spread across CT18, A, X, and Z would be the honest statement of current small-x uncertainty, and future precision data would be needed to break the degeneracy between scale choice and resummation."],"forward_implications":["The default CT18 set becomes the reference NNLO PDF ensemble for LHC calculations, replacing CT14 and CT14HERA2, with Hessian error sets for uncertainty propagation.","If the tuned scale is equivalent to low-x resummation for inclusive DIS, fixed-order NNLO calculations can be used in kinematic regions where explicit resummation was previously thought necessary.","The four CT18 families quantify the range of NNLO PDF behavior consistent with current data; using only the default set would understate the ambiguity revealed by the ATLAS W/Z tension.","CT18Z, the variant that includes the ATLAS W/Z data, removes CDHSW, adopts a saturation-inspired DIS scale, and reduces the NNLO gluon-fusion Higgs cross section by about 1% relative to CT14 and CT18.","The new parametrization, with an SU(3)-symmetric strange sea at x->0 and a d/u ratio free at x->1, changes strange-quark and high-x d-quark predictions relevant to W-mass and new-physics analyses."],"supporting_citations":[{"why":"Defines the CT14 baseline PDFs that CT18 replaces and is compared against.","marker":"[2]"},{"why":"Provides the transitional CT14HERA2 PDFs used as the reference for ratio plots and the prior fit to the final HERA data.","marker":"[3]"},{"why":"Supplies the ATLAS 7 TeV W/Z rapidity data whose tension with other experiments motivates the separate CT18Z and CT18A fits.","marker":"[6]"},{"why":"Provides the APPLgrid fast interpolation tables used to compute NLO cross sections for new LHC processes, with NNLO K-factors.","marker":"[7]"},{"why":"Is the combined HERA I+II inclusive DIS data set, the dominant constraint on the fit and the dataset showing the >50-unit chi-squared improvement.","marker":"[13]"},{"why":"Reports the resummed fit whose HERA-data improvement the x-dependent scale is claimed to match.","marker":"[15]"},{"why":"Gives a second resummation study cited as a benchmark for the same HERA-data improvement.","marker":"[16]"},{"why":"Describes the PDFSense program used to predict which candidate data sets will constrain the PDFs before fitting.","marker":"[22]"},{"why":"Describes the ePump reweighting method used to estimate the impact of candidate data on the published Hessian PDFs.","marker":"[23]"},{"why":"Supplies the fastNLO tables used for fast NLO predictions with NNLO corrections in the global fit.","marker":"[24]"}],"fun_headline_variants":["CT18 NNLO PDFs: x-dependent scale mimics low-x resummation","Four CT18 families capture ATLAS W/Z tension in NNLO fits","Scale trick in CT18 cuts HERA chi2 by 50, no resummation needed","CTEQ-TEA CT18: x-scale reproduces resummation gain at HERA","New CT18 NNLO PDFs: ATLAS data spread into four variants"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The x-dependent scale's coefficients are chosen to minimize chi-squared on the HERA data that are then quoted as the improvement, so the small-x physics claim rests on that scale being a genuine theoretical proxy for resummation rather than a flexible fit.","fun_headline_variants_meta":{"raw":{"variants":["CT18 NNLO PDFs: x-dependent scale mimics low-x resummation","Four CT18 families capture ATLAS W/Z tension in NNLO fits","Scale trick in CT18 cuts HERA chi2 by 50, no resummation needed","CTEQ-TEA CT18: x-scale reproduces resummation gain at HERA","New CT18 NNLO PDFs: ATLAS data spread into four variants"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000364,"raw_usage":{"total_tokens":1941,"prompt_tokens":907,"completion_tokens":1034,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":926}},"tokens_in":523,"tokens_out":1034,"duration_ms":8708,"temperature":1.0,"reasoning_tokens":926,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:15:39.974684+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the tuned scale $\\mu^2_{F,x}$ with its published coefficients and apply it, without refitting, to independent low-x DIS data not included in the CT18 fit, such as future EIC structure-function measurements; if the out-of-sample improvement is not comparable to the reported >50-unit gain on HERA, the scale is absorbing fit noise rather than the low-x logarithms it is claimed to reproduce.","supporting_citations":[],"review_version":1}