{"id":"062c3e7d-17e9-4733-9aaa-9850cfe8a04c","arxiv_id":"1908.07529","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Joint resummation of two angularities, rather than one or many, yields the largest gain in predicting other angularities in e+ e- dijet events.","lead":"This paper asks how many simultaneous QCD logarithms must be resummed to reliably predict jet shapes, using e+ e- angularities as a testbed. Reweighting a flat phase-space generator with two angularities already gives an order of magnitude better predictions for other angularities than one, while adding more helps little.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"n=2 improvement may be an artifact of the low-dimensional flat phase space used for reweighting; k=4-6 checks do not probe realistic jet multiplicities.","rationale":"The reader's weakest assumption identifies the same broad area: the flat phase-space sample may lack QCD correlations, so the improvement could be an artifact. My concern is a sharper mechanism: the finite and small number of particles k in the Rambo sample creates an artificially low-dimensional phase space, where two angularities over-constrain the kinematics and produce overly narrow conditional distributions. The authors' tests at k=4,5,6 are all in this same small-k regime and therefore cannot distinguish this artifact from a real physical effect. This is load-bearing because the abstract's quantitative headline, the order-of-magnitude improvement for n=2, is computed entirely through eq. (2.2). If the conditional correlations in flat phase space are not representative, the improvement could shrink or vanish when the sampler's dimensionality approaches that of a real parton shower. The reader's other points (in-sample optimization, missing code, analytic projection inconsistency) are valid but secondary: the in-sample optimization affects the size of the improvement but not its existence, the missing code is a reproducibility issue, and the analytic inconsistency is confined to the secondary NLL demonstration and is explicitly acknowledged in appendix A.2. The proposed test is decisive and practical: increasing k to 10-20 is straightforward with Rambo, and comparing conditional widths from Herwig directly isolates the reweighting's key assumption. Given that the reader already returned CONDITIONAL, this stress-test does not alter the verdict; it just sharpens the condition that should be satisfied before the central claim is accepted at face value.","tokens_in":24517,"tokens_out":7744,"duration_ms":526919,"concrete_test":"Repeat the full reweighting analysis with flat k-body phase space for k=10, 15, and 20 (or with k sampled from a Poisson distribution matched to the mean Herwig multiplicity), using the same Herwig target distributions. If the ratio chi^2_min(n=2)/chi^2_min(n=1) moves substantially toward 1 as k increases, the claimed improvement is a low-dimensional artifact; if it remains near 0.1 with acceptable statistical errors, the claim is robust. Additionally, directly compute from Herwig events the conditional distribution width of a held-out angularity (e.g., alpha_j=2.6) given the two optimal inputs (0.2 and 1.8) and compare with the corresponding flat phase-space conditional width; a large discrepancy would invalidate the reweighting premise.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, an order-of-magnitude improvement from n=1 to n=2, rests entirely on the reweighting procedure in eq. (2.2). That equation uses the flat k-body phase-space distribution to supply the conditional distribution of a held-out angularity given the input angularities; the resummed/Herwig/Pythia input only fixes the marginals of the inputs. If the flat phase-space conditional correlations are not QCD-like, the measured improvement and its saturation at n=2 are not physically meaningful. The authors' robustness checks vary k=4,5,6, but all are far below the typical number of partons in an e+e- dijet event at Q=1 TeV, where a parton shower generates many emissions. In a low-dimensional phase space, two angularities can nearly determine the remaining kinematic degrees of freedom, artificially sharpening conditional distributions and inflating the apparent predictive power of n=2 relative to n=1. The similar trend for k=4-6 does not exclude this artifact because all three values lie in the same small-k regime; the conditional width of an angularity given two others likely decreases as k is reduced, regardless of QCD. Thus the order-of-magnitude improvement may be an upper bound set by the sampler's dimensionality rather than a property of QCD final states.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates how many jointly resummed angularities are needed to predict other angularities in e+e- dijet events. The authors derive a next-to-leading-logarithmic (NLL) joint resummation for an arbitrary number of angularities using SCET, obtaining factorization formulas, power corrections, and a matching procedure over phase-space regions. They then combine this analytic input with Herwig and Pythia distributions to reweight flat k-body phase space, using n angularities as input (Eq. (2.2)) and measuring the quality of predictions for held-out angularities through a global chi-squared (Eq. (2.4)). The central claim is that reweighting with n=2 angularities improves the held-out predictions by an order of magnitude over n=1, with diminishing returns for larger n. Robustness checks are presented for Herwig versus Pythia, k=4,5,6, restricted angularity sets, and center-of-mass energies Q=0.2,1,4 TeV.","tokens_in":24660,"tokens_out":6380,"duration_ms":69242,"significance":"If the central claim is correct, the paper provides a useful quantitative answer to a practical question in jet substructure and machine-learning applications: two jointly resummed observables capture most of the predictive information in the angularity family. The derivation of the NLL joint resummation for n angularities, especially the general factorization formula in Eq. (3.16) and the power-correction hierarchy in Eqs. (3.27)-(3.30), is a nontrivial technical contribution. The paper is also careful in several respects: it uses external Monte Carlo generators as independent benchmarks, provides a replica-based estimate of statistical uncertainties, and includes multiple variations of the setup in Appendix A. These strengths make the qualitative trend credible, but they do not by themselves resolve two load-bearing methodological concerns about the optimal-set selection and the role of flat phase-space correlations.","major_comments":[{"comment":"The optimal input set I is selected by minimizing the same global chi-squared (2.4) that is later reported as chi^2_min. This is in-sample selection: the performance of the chosen set is evaluated on the very angularities used to choose it, so the reported value measures training performance rather than predictive performance. The order-of-magnitude drop from n=1 to n=2 in the right panel of Fig. 7 could therefore be inflated by overfitting to the 15-member angularity family, especially because the replica median does not remove selection bias within each replica. The qualitative trend may survive, but the quantitative 'order of magnitude' claim should be verified with a proper cross-validation scheme, for example by optimizing I on one half of the angularities and evaluating chi^2 on the other half.","section":"Sec. 2.2, Eqs. (2.3)-(2.4), Fig. 7"},{"comment":"The reweighting procedure constrains only the marginal distributions of the input angularities; the conditional distribution of a held-out angularity given the inputs is inherited entirely from flat massless k-body phase space. If the flat-sampler correlations are not QCD-like, the measured improvement and saturation at n=2 could be artifacts of the low-dimensional sampler rather than properties of QCD final states. The k=4,5,6 checks in Figs. 9-10 do not exclude this, since all three values are in the same small-k regime and all share the same Rambo-based conditional structure. A concrete test would be to repeat the procedure with a base sample that has QCD-like correlations, for example parton-shower events reweighted to the same target marginals, or to increase k to values comparable to typical parton multiplicities in e+e- dijets and check whether the improvement from n=1 to n=2 persists.","section":"Sec. 2.2, Eq. (2.2), Figs. 9-10"},{"comment":"The authors find that projections from higher-dimensional analytic NLL distributions do not reproduce the direct one-dimensional distributions, with chi^2 values up to 0.0109 for projections from three angularities, and attribute this to binning. Because the analytic reweighting results in Figs. 6 and 13 are constructed from such projected distributions, the analytic leg of the robustness argument inherits this inconsistency. The claim that the analytic input reproduces the same qualitative trend is therefore contingent on the binning resolution of the n=3 spectra; a check with a larger number of bins, or at least a quantitative estimate of the resulting uncertainty on chi^2_min, would be needed to support that claim.","section":"Sec. A.2, Fig. 12, Figs. 6 and 13"}],"minor_comments":[{"comment":"The label 'HEWRIG' in the left panel of Fig. 7 is a typo and should read 'HERWIG'.","section":"Fig. 7, left panel"},{"comment":"The quantity chi^2 in Eq. (2.3) is not a standard chi-square statistic: it is an unnormalized L2 distance between binned distributions. The 'order of magnitude' improvement is therefore metric-dependent, and this should be stated explicitly when the abstract refers to an order-of-magnitude improvement.","section":"Sec. 2.2, Eq. (2.3)"},{"comment":"The transition thresholds x_i and x_f in Eq. (3.33) are chosen by fixing the power corrections to 10% and 50%, respectively, but no variation of these thresholds is reported. Since they are free parameters in the matching procedure, a short scan or an estimate of the induced uncertainty would help establish that the main conclusions are insensitive to them.","section":"Sec. 3.5, Eq. (3.33)"},{"comment":"For the analytic reweighting, the optimal input set is taken from the analogous Herwig procedure rather than re-optimized for the analytic distributions. The text explains this choice, but a brief discussion of whether the Herwig-optimal set is also optimal, or nearly optimal, for the analytic input would strengthen the comparison.","section":"Fig. 6 and Sec. 4"},{"comment":"The conclusion states that augmenting Monte Carlo predictions with NLL analytic resummation is 'probably not that useful' due to perturbative uncertainty at this order, but no perturbative scale uncertainty band is shown for the analytic predictions. Adding such a band, or at least a qualitative estimate of its size, would substantiate this statement.","section":"Sec. 5"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid phenomenological study with a plausible qualitative conclusion, and the technical derivation of the n-angularity NLL resummation is a useful contribution. However, the central quantitative claim is currently supported by a procedure that selects the input set using the same figure of merit used to evaluate performance, and the held-out predictions rely on flat phase-space correlations that are not validated at realistic multiplicities. Both concerns are testable, so I recommend major revision rather than rejection. If the authors cannot fully address the selection-bias point, they should at least soften the abstract's order-of-magnitude statement and present the chi^2_min curve as an upper bound on achievable improvement rather than as a measurement of predictive power."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear X,\n\nWhat you should know: this is a careful case study answering a practical question—how many jointly resummed angularities do you need to predict other angularities in e+e- dijets? Their answer is two. The genuinely new piece is the NLL joint resummation of an arbitrary number n of angularities: Lund-plane regime classification, the factorization formula in eq. (3.16), the resummed cumulative expression (3.20), and the power-correction matching. Prior work had two angularities; this extends the framework to n. The saturation study, reweighting flat phase space with n-angularity distributions from Herwig, Pythia, or the analytic calculation and measuring held-out chi^2, is also new and motivated by ML jet substructure.\n\nThe central claim—an order-of-magnitude improvement from n=1 to n=2 with diminishing returns beyond—is supported by several variations: Herwig versus Pythia, k=4,5,6, different center-of-mass energies, and restricted alpha sets. The authors are honest about limitations. Appendix A.2 admits that projections from higher-dimensional analytic distributions do not match direct lower-dimensional ones; they attribute it to binning and route around it. That is a real soft spot: before the n-angularity factorization is used as a precision tool, that inconsistency should be understood, not just suspected.\n\nOther soft spots are minor. The reported chi^2_min is an in-sample optimum, so the magnitude is an upper bound for any fixed input set; the paper partially mitigates this by showing per-angularity results for the global best set. No code or data are shipped, so the numbers are not directly auditable. The stress-test concern—that the n=2 gain may reflect the low dimensionality of the flat k-body phase space rather than QCD correlations—is the one I would press on in review. k=4,5,6 are all small compared with realistic e+e- dijet parton multiplicities. The fact that the trend persists at k=5 and k=6 is reassuring but does not fully eliminate the worry, since all three are in the same small-k regime. I would not call the central argument broken; it is a statement about this reweighting procedure, and the paper says so.\n\nOverall: serious work, real technical extension, limitations disclosed. It deserves a proper referee and likely publication after a moderate revision. I would bring it to a reading group and would cite it for the n-angularity factorization.","headline":"A genuine technical extension—NLL joint resummation of n angularities—plus a saturation study whose central n=2 claim is plausible but partly hostage to the flat-phase-space prior.","tokens_in":25322,"tokens_out":4608,"would_cite":true,"duration_ms":197899,"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":"Two jointly resummed angularities cut event-shape prediction error tenfold; adding more scarcely helps.","keywords":["joint resummation","angularities","event shapes","e+e- annihilation","soft-collinear effective theory","Lund plane","reweighting","next-to-leading logarithmic accuracy"],"falsifier":"Repeat the reweighting with a phase-space sample whose angularity marginals are correct but whose conditional correlations are deliberately scrambled, for instance by permuting the angularity values between particles in each event; if the $\\chi^2_{\\min}$ drop from $n=1$ to $n=2$ disappears, the benefit is genuine QCD correlation, while if it persists, the conclusion would not come from the correlations in the phase-space sample and would need to be re-evaluated.","tokens_in":24202,"feed_emoji":"⚛️","tokens_out":6605,"duration_ms":66162,"temperature":0.7,"pith_summary":"This paper asks how many QCD observables must be resummed together before the full radiation pattern of a jet is captured. Using the family of $e^+e^-$ event shapes called angularities, the authors compute the cross section differential in $n$ angularities at next-to-leading logarithmic accuracy and use it to reweigh a flat multiparticle phase space, then test how well the reweighted events predict angularities that were not part of the input. The central finding is that reweighing with $n=2$ angularities improves predictions by an order of magnitude over $n=1$, while $n=3,4,5$ give only diminishing returns. The same trend appears whether the input distributions come from analytic resummation or from parton-shower generators, and across phase-space multiplicities and collision energies. If this result holds, the practical payoff of joint resummation is already achieved with two observables.","feed_headline":"Two angularities capture nearly the whole resummation payoff","feed_subtitle":"Reweighting e+e− dijet phase space with n=2 outperforms n=1 tenfold; n=3 to 5 add little.","key_machinery":"The machinery has two parts. First, the Lund-plane description of $n$ angularity measurements divides the $n$-dimensional phase space into regions characterized by soft, collinear, and collinear-soft modes; each region gets a factorization formula in Soft-Collinear Effective Theory, with the cross section written as a product of convolutions of hard, jet, soft, and collinear-soft functions and resummed through renormalization-group evolution in Laplace space. Second, the optimal-reweighing map of eq. (2.2), which multiplies flat phase space by the ratio of the resummed $n$-angularity distribution to the flat one, so that the held-out angularity is predicted from the correlations already present in flat phase space after the marginal distributions are corrected.","core_discovery":"The discovery is that the information needed to predict any angularity from a set of jointly resummed angularities saturates at $n=2$. Concretely, reweighing flat $k$-body phase space by the $n$-angularity distribution and then projecting onto a held-out angularity $\\ell_{\\alpha_j}$ yields a global goodness-of-fit $\\chi^2_{\\min}$ that drops roughly tenfold from $n=1$ to $n=2$, and then flattens. The authors identify the optimal input sets through global minimization, verify the trend with $k=4,5,6$ phase-space bodies, with two different parton-shower generators as references, and at three different center-of-mass energies, and conclude that the benefit of joint resummation is already realized at two angularities.","pith_inferences":["A natural testable extension would be to measure the mutual information between angularities in a high-statistics $e^+e^-$ dataset and compare where it saturates against the $\\chi^2$ saturation found here.","The saturation at $n=2$ may reflect an effective low dimensionality of the resummation-region phase space; if so, other observables such as the pair $(q_T,\\text{threshold})$ in hadronic collisions should show a similar two-variable sweet spot.","The reweighting method itself could become a lightweight way to inject analytic resummation into existing Monte Carlo samples: training a generator on a two-angularity NLL-corrected weight and then validating on data would test whether the order-of-magnitude gain persists at the level of actual measurements."],"forward_implications":["Monte Carlo samples used for jet-substructure studies need only be constrained by two jointly resummed angularities to reproduce the full family of angularity predictions at the level probed here.","Adding a third, fourth, or fifth angularity to the reweighting buys little, so the practical cost of higher-dimensional joint resummation is not justified by the predictive gain for angularities.","The same qualitative conclusion holds for both parton-shower-based and analytic NLL inputs, indicating that the saturation at $n=2$ is not specific to one generator.","The factorization framework for $n$ angularities requires no new ingredients beyond the two-angularity case at NNLL, so the saturation found at NLL is expected to persist at higher logarithmic order."],"supporting_citations":[{"why":"Defines angularities, the observable family at the center of the study.","marker":"[1]"},{"why":"Provides the factorization of double-differential cross sections that the generalization to $n$ angularities extends.","marker":"[40]"},{"why":"Supplies the joint resummation of two angularities, including the consistency relations and natural scales that the $n$-angularity formula builds on.","marker":"[45]"},{"why":"Motivates the question by showing that jet discrimination saturates at a small number of observables.","marker":"[57]"},{"why":"Introduces the Rambo flat phase-space sampler used to generate the reweighting sample.","marker":"[58]"},{"why":"Supplies the improved flat phase-space sampling that enables reliable angularity predictions.","marker":"[59]"},{"why":"Gives the modified angularity definition used for large angles in the implementation.","marker":"[63]"},{"why":"Supplies the inverse-error weighting procedure adapted for matching factorization regions.","marker":"[70]"}],"fun_headline_variants":["Saturation at two angularities in joint QCD resummation","Two angularities reap the resummation benefit","Tenfold gain from adding second angularity, then plateau","Resummation payoff saturates at n=2 angularities","Two angularities suffice for joint resummation gains"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The procedure assumes that flat, massless $k$-body phase space, after being reweighted to match the $n$-angularity distribution, already contains the correct QCD correlations needed to predict any other angularity; if flat phase space lacks those correlations, the improvement and its saturation at $n=2$ could be an artifact of the sampler.","fun_headline_variants_meta":{"raw":{"variants":["Saturation at two angularities in joint QCD resummation","Two angularities reap the resummation benefit","Tenfold gain from adding second angularity, then plateau","Resummation payoff saturates at n=2 angularities","Two angularities suffice for joint resummation gains"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000594,"raw_usage":{"total_tokens":2763,"prompt_tokens":910,"completion_tokens":1853,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":526,"completion_tokens_details":{"reasoning_tokens":1771}},"tokens_in":526,"tokens_out":1853,"duration_ms":13971,"temperature":1.0,"reasoning_tokens":1771,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:05:01.423278+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the reweighting with a phase-space sample whose angularity marginals are correct but whose conditional correlations are deliberately scrambled, for instance by permuting the angularity values between particles in each event; if the $\\chi^2_{\\min}$ drop from $n=1$ to $n=2$ disappears, the benefit is genuine QCD correlation, while if it persists, the conclusion would not come from the correlations in the phase-space sample and would need to be re-evaluated.","supporting_citations":[{"cited_title":"Joint resummation of two angularities at next-to-next-to-leading logarithmic order","cited_arxiv_id":"1806.10622","evidence_quote":"Supplies the joint resummation of two angularities, including the consistency relations and natural scales that the $n$-angularity formula builds on."},{"cited_title":"Kleiss, W","cited_arxiv_id":null,"evidence_quote":"Introduces the Rambo flat phase-space sampler used to generate the reweighting sample."},{"cited_title":"Matching factorization theorems with an inverse-error weighting","cited_arxiv_id":"1801.01480","evidence_quote":"Supplies the inverse-error weighting procedure adapted for matching factorization regions."}],"review_version":1}