{"id":"ca75fe21-d6c3-4817-bb8a-09422c370aaf","arxiv_id":"2501.13167","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Differential NNLO Drell-Yan interpolation grids are released, with closure tests showing sub-per-mille accuracy and a validation that PDF fits are unaffected by the K-factor approximation.","lead":"The authors built an interface between the NNLOJET event generator and the PINEAPPL grid library, releasing fast NNLO interpolation grids for Drell-Yan measurements. These grids let PDF fits re-evaluate theory predictions for different PDFs and scales almost instantly, and the paper validates the common K-factor shortcut.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Closure tests validate only two of the released grids; the universal sub-per-mille accuracy claim is extrapolated to high-mass and 13 TeV forward grids without direct evidence.","rationale":"The reader's weakest_assumption correctly identifies the closure-test generalization as the main soft spot. My read agrees: the paper's central utility rests on the accuracy of every released grid, but only two datasets are shown to close. The concern is genuine, but it is not fatal: the paper itself acknowledges few-per-mille deviations in extreme regions and provides a dilution argument for δNNLO; PINEAPPL is a mature library with well-understood interpolation; and the grids are released with metadata and runcards, allowing any user to validate. Therefore the appropriate verdict remains ACCEPT (UNCHANGED), with the concrete closure test above serving as a low-cost verification that would either confirm or refute the extrapolation. I do not see an internally inconsistent step or a more load-bearing assumption in the K-factor fit comparison, which is carefully controlled (same seed, same settings, worst-case DY-only fit).","tokens_in":12555,"tokens_out":8719,"duration_ms":90325,"concrete_test":"Run the same closure test shown in Fig. 1 for three additional released grids that cover the unvalidated extremes: ATLAS high-mass DY 2D at 8 TeV (Ref. [47]), LHCb Z→ee at 13 TeV (Ref. [58]), and CMS W rapidity at 8 TeV (Ref. [51]). For each grid, evaluate the PINEAPPL grid with the NNPDF4.0 PDF recorded in the metadata and compare bin-by-bin against the NNLOJET reference results stored under the 'result' metadata key, separately for LO, δNLO, and δNNLO. If the maximum per-bin deviation on the δNLO coefficient exceeds 0.5% or on δNNLO exceeds 1%, the universal 'completely negligible' claim requires qualification.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central accuracy claim — that interpolation errors are typically below per-mille and completely negligible for phenomenology — is supported by Fig. 1 closure tests for only two of the 22 released datasets: CMS 7 TeV DY 2D (Ref. [50]) and LHCb 8 TeV W→μ (Ref. [55]). No closure test is shown for the high-mass grids (ATLAS 7/8 TeV up to 1500 GeV, Refs. [42,47]) or the 13 TeV LHCb forward grids (Ref. [58]). These regions probe larger x and rapidities where interpolation is hardest; Fig. 1b already shows δNNLO deviations of a few per-mille in the forward LHCb region. The paper's '0.01‰ on the final result' dilution argument is applied to δNNLO, but interpolation errors on the LO and δNLO coefficients (which together dominate the cross section) are not separately quantified. If an untested grid had a 1% δNLO interpolation error in some bin, the absolute prediction would shift by roughly 0.3–0.5%, which is not negligible compared with sub-percent PDF uncertainties. The extrapolation from two datasets to all released grids is therefore the least secure step in the paper's central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports on a new interface between the NNLOJET parton-level Monte Carlo generator and the PINEAPPL interpolation-grid library, and releases grids for 22 Drell-Yan datasets that commonly enter global PDF determinations. After describing the grid metadata and presenting closure tests for two representative datasets (CMS 7 TeV DY and LHCb 8 TeV W) to quantify interpolation errors, the paper demonstrates the use of the grids for scale and PDF uncertainty studies, investigates accidental cancellations between partonic channels at NNLO (including a rotation to the evolution basis), and validates the widely used K-factor approximation for DY in the context of PDF fits. The K-factor validation includes dedicated PDF fits with identical settings that compare exact NNLO grids against the K-factor approximation, both in a global-fit setup and in a DY-only worst-case scenario.","tokens_in":12789,"tokens_out":5211,"duration_ms":52210,"significance":"If the grids are accurate as claimed, this work fills a genuine gap: no differential NNLO interpolation grids for Drell-Yan were previously available, despite the process contributing roughly 20% of the data in recent global PDF fits. The released grids would allow fast, exact NNLO re-evaluations for arbitrary PDFs and scale choices, which is directly useful for the PDF-fitting community. The K-factor study is well designed: the PDF fits differ only in the theory treatment while keeping all other settings fixed, and the DY-only fit provides a meaningful worst-case test. The evolution-basis analysis offers a novel and plausible explanation for the accidental NNLO cancellations. The paper also ships reproducible artifacts: the grids, the NNLOJET runcards, and the PINEAPPL-based analysis tools are publicly available, which strengthens its practical value.","major_comments":[{"comment":"The closure tests are presented for only two of the 22 released datasets (CMS 7 TeV DY, Ref. [50], and LHCb 8 TeV W, Ref. [55]). The paper then states that interpolation errors are 'completely negligible' for any phenomenological application of the provided grids. This universal accuracy claim is extrapolated to untested datasets, including the high-mass ATLAS grids reaching m_ll = 1500 GeV (Refs. [42,47]) and the 13 TeV LHCb forward grids (Ref. [58]), which probe larger x and rapidity ranges where interpolation is hardest. The authors should either provide closure tests for at least one high-mass and one 13 TeV forward dataset, or explicitly restrict the accuracy claim to the tested kinematic regions.","section":"Sec. 2.2, Figs. 1a/1b"},{"comment":"The argument that 'an interpolation error of 1 permille on the coefficient translates to a 0.01 permille level of exactitude on the final results' applies only to the delta_NNLO coefficient, which is a few percent of the full cross section. The LO and delta_NLO coefficients contribute directly to the cross section, and Fig. 1b shows deviations of a few permille for these coefficients in the forward LHCb region; a 1% interpolation error on the delta_NLO coefficient in an untested bin would shift the absolute prediction by roughly 0.3-0.5%, which is not negligible compared with sub-percent PDF uncertainties. The closure test should separately quantify and report the interpolation error on the LO and delta_NLO coefficients, and the 'completely negligible' claim should be reassessed in light of those numbers.","section":"Sec. 2.2, paragraph on dilution argument"}],"minor_comments":[{"comment":"The sentence 'no differential predictions are available' is imprecise: Drell-Yan has been computed differentially at NNLO (Refs. [18-21]) and N3LO (Refs. [26-30]); the intended meaning is that no differential interpolation grids are available. Please rephrase.","section":"Sec. 1"},{"comment":"Typo: 'an approximation basted on NLO grids' should read 'based on NLO grids'.","section":"Sec. 1"},{"comment":"The sentence 'conversion from and to (pineappl [import|export] --help) this format facilitate its use as an universal converter' is garbled; please rephrase to clearly explain the import/export syntax.","section":"Sec. 2.1, footnote 1"},{"comment":"The phrase 'combined with the flexibly to evolve' should read 'combined with the flexibility to evolve'.","section":"Sec. 3.1"},{"comment":"The horizontal-axis label 'F (GeV)' should be 'mu_F (GeV)' in both panels for clarity.","section":"Fig. 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is a resource/release note for a well-maintained toolchain. The central technical content is sound, but the universal accuracy claim for all released grids rests on a narrow set of closure tests; I would like to see those tests extended before publication, since the grids are meant for public use in PDF fits where untested interpolation errors could matter."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a genuinely useful release. Drell-Yan is a big chunk of the data in global PDF fits, and until now there were no differential NNLO grids for it. The paper closes that gap: a new NNLOJET-PINEAPPL interface, a set of grids covering the NNPDF4.0 DY datasets, and two small phenomenological studies on top.\n\nWhat is actually new is the grids and the interface. The K-factor stability study is the most carefully done piece: same fit settings, same seeds, only the theory treatment changed, plus a DY-only worst-case fit. That is the right way to isolate the effect, and the conclusion that the K-factor approximation is safe at the current precision level is credible. The evolution-basis decomposition of the NNLO channel cancellation is also a nice observation, even if it is more of a curiosity than a result with direct implications.\n\nThe closure tests are the appropriate check for interpolation error, and the two datasets shown bracket the central and forward regions reasonably well. The paper is careful in its language: it says \"typically\" below per-mille, and it explicitly acknowledges few-per-mille deviations in the LHCb forward region. The dilution argument about the delta_NNLO coefficient being a small part of the total is legitimate, and the figure does plot separate ratios for LO, delta_NLO, and delta_NNLO, so the worry that LO/NLO errors are unquantified is partly answered, at least for those two datasets.\n\nThe real soft spot is the extrapolation from two closure tests to all 22 released grids. That is a fair concern, but it is minor because the grids are public and anyone can rerun the test. It would be a stronger paper with one more closure test on a high-mass grid, but I would not treat this as a blocking flaw. The reliance on the non-public NNLOJET code for the exact reference numbers is a practical limitation, not an error.\n\nWho should read this: PDF fitters and anyone doing fast re-evaluations of DY theory predictions. The paper deserves a serious referee, mostly to verify that the released grids match the paper's claims and to ask for that additional closure test. I would accept it with minor revision.","headline":"Useful, well-scoped technical release that fills the missing NNLO Drell-Yan grid gap; the K-factor fit study is the strongest part.","tokens_in":13307,"tokens_out":1845,"would_cite":true,"duration_ms":21531,"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 establishes that NNLO Drell-Yan theory can be stored in interpolation grids and reused for any PDF with negligible error, retiring the need for K-factor approximations.","keywords":["Drell-Yan process","NNLO QCD","interpolation grids","PDF fits","K-factor approximation","accidental cancellations","DGLAP evolution","Fast-Kernel tables"],"falsifier":"Take one released grid outside the two closure-tested datasets, such as an ATLAS 8 TeV high-mass grid, evaluate it at the same NNLOJET reference settings, and compare bin-by-bin; if any bin's interpolation error on the NNLO coefficient exceeds the claimed few-per-mille level, or shifts the final cross section by more than one part in $10^{-5}$, the universal accuracy statement is wrong.","tokens_in":12365,"feed_emoji":"⚛️","tokens_out":14231,"duration_ms":127767,"temperature":0.7,"pith_summary":"The paper reports a new interface between an NNLO parton-level Monte Carlo generator and an interpolation-grid library, and releases the first NNLO differential grids for the Drell-Yan process across the Tevatron and LHC measurements that commonly enter global PDF fits. The grids reproduce the exact NNLO calculation to better than one per-mille over most of phase space, with errors rising to a few per-mille only at the forward LHCb edge; once the small size of the NNLO coefficient is taken into account, those errors are negligible. Using the grids, the paper diagnoses the NNLO accidental cancellations as a basis effect and shows that the widely used K-factor approximation is stable at the per-mille level, with PDF fits using exact NNLO grids instead of K-factors agreeing within quoted uncertainties. This makes exact NNLO Drell-Yan theory a practical option for PDF determinations and provides a stepping stone toward approximate N3LO grids.","feed_headline":"Exact NNLO Drell-Yan grids retire the K-factor workaround","feed_subtitle":"Sub-per-mille accurate grids let PDF fits drop the NNLO K-factor approximation.","key_machinery":"The central object is the interpolation grid produced by the NNLOJET-PINEAPPL interface: a discretized store of the partonic cross section convolved with basis eigenfunctions, with separate entries for the renormalisation scale $\\mu_R$, the factorisation scale $\\mu_F$, the momentum fractions $x_1$ and $x_2$, each partonic channel, and each perturbative order. Evaluating a hadronic cross section from the grid reduces to a weighted sum over nodes, which is what makes re-evaluation for arbitrary PDFs almost instantaneous. A second mechanism, the rotation of the DGLAP evolution operator into the singlet and non-singlet evolution basis, does the work of isolating the accidental cancellations: it decouples the evolution of independent components and makes the scale dependence nearly flat, showing the large flavour-basis cancellations are driven by DGLAP-induced correlations inside the singlet sector.","core_discovery":"By linking the NNLO parton-level generator NNLOJET to the interpolation library PINEAPPL, the paper supplies the first differential NNLO interpolation grids for Drell-Yan, covering the Tevatron and LHC measurements that typically enter a global PDF determination. Closure tests against the exact calculation show that interpolation errors stay below one per-mille over most of phase space, with a few per-mille in the forward LHCb region; because the NNLO coefficient itself is only a few percent of the full prediction, those grid errors shift the final cross section by roughly one part in $10^{-5}$, well below the Monte Carlo and experimental uncertainties. The grids reveal that the striking NNLO cancellation between $\\mathrm{q}\\bar{\\mathrm{q}}$ and $\\mathrm{qg}+\\bar{\\mathrm{q}}\\mathrm{g}$ channels is largely an artefact of the flavour basis: in the evolution basis the cancellation is an order of magnitude smaller and the scale dependence nearly flat. Finally, the paper shows that the widely used K-factor approximation for Drell-Yan, NLO grids multiplied by an NNLO K-factor computed with one PDF set, is stable at the per-mille level across PDF sets, and that PDF fits using the exact NNLO grids instead of K-factors agree within quoted uncertainties even when only Drell-Yan data are fitted.","pith_inferences":["The interface is process-agnostic, so the same recipe should produce NNLO grids for other NNLOJET processes such as jets, top-quark pairs, and DIS; a direct test would be a grid release for one of those processes.","Because closure tests are shown for only two datasets, a prudent user would run the same grid-versus-exact comparison on the high-mass ATLAS and 13 TeV LHCb grids before relying on the sub-per-mille statement at the kinematic edges; this is a cheap verification the paper does not itself include.","The evolution-basis flatness suggests that DGLAP evolution, not the hard matrix element, is the main driver of the NNLO scale sensitivity; a testable extension is whether approximate N3LO predictions in the same basis close the known gap between NNLO uncertainty bands and N3LO central values."],"forward_implications":["All released Drell-Yan datasets can be re-evaluated at NNLO for arbitrary PDF sets, scale choices, and $\\alpha_s$ values in seconds, so global PDF fits can use exact NNLO theory instead of K-factor-approximated theory.","The evolution-basis analysis implies that comparisons of NNLO and N3LO Drell-Yan predictions should be made in the evolution basis, where the scale dependence is flat, rather than in the flavour basis.","Existing PDF fits that used NLO grids plus NNLO K-factors are not materially biased by that approximation, because the K-factor is stable at the per-mille level and the fit impact is within quoted uncertainties.","The released grids supply the NNLO ingredient needed to construct approximate N3LO grids from N3LO K-factors, the route the paper identifies toward N3LO PDF fits.","Scale-variation bands, cross-PDF pulls, and PDF uncertainties for Drell-Yan become effectively zero-cost analyses, making routine theory-uncertainty audits feasible for every dataset in the release."],"supporting_citations":[{"why":"The interpolation library whose grid format, convolution algorithms, and CLI the released grids use.","marker":"[4]"},{"why":"The global PDF fit that defines the dataset list for the released grids and supplies the reference PDF used in closure and fit comparisons.","marker":"[16]"},{"why":"The NNLO parton-level generator whose Drell-Yan implementation provides the exact cross sections that the grids interpolate.","marker":"[31]"},{"why":"The CMS 7 TeV Drell-Yan measurement used for one of the two closure tests shown.","marker":"[50]"},{"why":"The LHCb 8 TeV W measurement used for the second closure test, covering the forward region where interpolation errors reach a few per-mille.","marker":"[55]"},{"why":"Documents the NNLOJET Drell-Yan process implementation used to produce all released grids.","marker":"[59]"},{"why":"Provides the tools that convert grids into Fast-Kernel tables and generate the evolution kernel operators used in the basis analysis.","marker":"[13]"},{"why":"The PDF-fitting framework used to run the exact-versus-K-factor fits.","marker":"[67]"}],"fun_headline_variants":["Exact NNLO Drell-Yan grids make K-factor approximation obsolete","First NNLO interpolation grids for Drell-Yan released","Drell-Yan K-factor workaround retired by exact NNLO grids","NNLO Drell-Yan grids replace K-factor shortcut in PDF fits","Sub-per-mille NNLO Drell-Yan grids end K-factor era"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The universal sub-per-mille accuracy claim for all released grids rests on closure tests for only two of the many datasets, and even those show deviations of a few per-mille in the forward and high-mass corners.","fun_headline_variants_meta":{"raw":{"variants":["Exact NNLO Drell-Yan grids make K-factor approximation obsolete","First NNLO interpolation grids for Drell-Yan released","Drell-Yan K-factor workaround retired by exact NNLO grids","NNLO Drell-Yan grids replace K-factor shortcut in PDF fits","Sub-per-mille NNLO Drell-Yan grids end K-factor era"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000779,"raw_usage":{"total_tokens":3475,"prompt_tokens":1011,"completion_tokens":2464,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":627,"completion_tokens_details":{"reasoning_tokens":2370}},"tokens_in":627,"tokens_out":2464,"duration_ms":19375,"temperature":1.0,"reasoning_tokens":2370,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T16:24:01.571127+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one released grid outside the two closure-tested datasets, such as an ATLAS 8 TeV high-mass grid, evaluate it at the same NNLOJET reference settings, and compare bin-by-bin; if any bin's interpolation error on the NNLO coefficient exceeds the claimed few-per-mille level, or shifts the final cross section by more than one part in $10^{-5}$, the universal accuracy statement is wrong.","supporting_citations":[{"cited_title":"NNLOJET: a parton-level event generator for jet cross sections at NNLO QCD accuracy","cited_arxiv_id":null,"evidence_quote":"The NNLO parton-level generator whose Drell-Yan implementation provides the exact cross sections that the grids interpolate."},{"cited_title":"Precision phenomenology with fiducial cross sections in the triple-differential Drell-Yan process","cited_arxiv_id":"2301.11827","evidence_quote":"Documents the NNLOJET Drell-Yan process implementation used to produce all released grids."}],"review_version":1}