First double-differential cross-section measurements in proton multiplicity and kinematics for mesonless ν_μ charged-current interactions on argon using the MicroBooNE detector

This is the first multiplicity-resolved CC0piNp double-differential measurement on argon, using the full BNB dataset and stepping beyond the prior [9] result, which didn't resolve proton multiplicity and relied on tighter muon cuts. It does what a careful measurement paper should: FSI-sensitive binning, an honest treatment of the 38 constrained degrees of freedom coming from event sharing between blocks via the range-projected chi-squared (a useful methodological step in its own right), the AC-matrix correction applied to model comparisons to account for regularization bias, and a full data release with per-source covariances, response matrices, and the projection matrix. The FSI sensitivity is the real result - AR23 variants spanning global chi2/ndof from 488/145 (hA) to 1201/145 (hN) isolates the cascade treatment as a leading driver of the discrepancy - and the backward-angle Np=1 excess that no generator reproduces is a clean, reproducible claim. The stress-test concern holds up on reading, and it's the right caveat. The resimulation validation (Appendix B, Eq. B1) reuses the central-value efficiency, so it constrains migration but not selection efficiency; the authors acknowledge this at b456. The efficiency is lowest - 32.6% and 36.5% in the two most backward Np=1 bins, and a few percent in the high-momentum subleading bins - in exactly the regions where the tension lives. If the MC efficiency is mismodeled there, the unfolded excess moves. The NuWro fake-data closure is also partly outside the true data regime, since NuWro itself fails globally (574.8/145). These points are real, though not load-bearing: the same backward-angle pattern shows up in [9] with a different selection and in the accompanying nu_e analyses, and a common efficiency conspiracy across three independent selections is a stretch. The missing piece is explicit - the paper never isolates what a plausible efficiency error in the backward Np=1 region would do to the reported excess. Agree with ACCEPT, significance 7, soundness 8. Send it to referee and push the efficiency question in the backward-angle bins, or ask for the statement of how much efficiency error the excess tolerates. This becomes the reference for CC0piNp on argon.

The claim that no generator adequately describes the data, with the most pronounced tension at backward opening angles and for subleading protons (abstract; b5, b6), requires the unfolded cross sections to be unbiased in exactly the bins where the tension lives. The reader's flagged assumption — response-matrix model dependence — is real, and I can sharpen it. The resimulation study (Appendix B) builds alternative response matrices from resimulated data and MC, but Eq. (B1) explicitly reuses the central-value efficiency epsilon_CV from the GENIE MicroBooNE tune; the authors acknowledge in b456 that resimulated samples cannot estimate the selection efficiency. Consequently, the data-versus-MC resimulation comparison (Table III) is a pure migration-consistency test: any efficiency error cancels because both arms share the same epsilon_CV. The efficiency is steep and low in precisely the decisive bins: 32.6% for the most backward Np=1 bin (block 2), ~19%, ~10%, and ~4-19% for subleading protons in blocks 5-7. The sideband study (p = 0.98, 0.77) bounds the background, and the NuWro fake-data closure recovers the NuWro truth shape (chi2 ~ 1), but the paper's own results show NuWro itself disagrees with the data (global 574.8/145). The closure regime therefore lies partly outside the true data shape; the backward Np=1 excess is beyond every tested prediction, so no internal test exercises the detector response where the data most departs from simulation. If the simulation's backward near-threshold proton efficiency is mismodeled in either direction, the magnitude of the headline backward-angle discrepancy changes, and the finding that generators underpredict backward single-proton events is at least partially instrumental. This concern is acknowledged as a general limitation (b452, b456) but the specific unvalidated component — efficiency, as opposed to smearing — is not isolated anywhere in Sec. IV C.

The reader's weakest_assumption identifies the response matrix built from the GENIE MicroBooNE tune as the main residual risk, which matches my general concern. I sharpen it to a specific gap: of the two components of the response (efficiency and migration), only the migration is exercised by the data-driven resimulation study, because Eq. (B1) reuses the central-value efficiency for both data- and MC-derived matrices. The most decisive bins — backward Np=1 and subleading-proton bins — have the lowest and steepest efficiencies, so they are the least protected by the internal checks. The NuWro fake-data closure is genuine independent support but is weakened by the paper's own demonstration that NuWro is not an adequate description of the data, meaning the closure tests a regime that partially lies outside the data. On balance, the concern is real but does not overturn the verdict. The full-dataset failure of all generators is robust to the efficiency question: the best model has chi2/ndof of 428.8/145 and even the strongest per-block chi2 values (e.g., 42.7/32 for block 8) are far too large to be absorbed by any plausible efficiency bias, and an efficiency error large enough to remove the backward-angle tension would likely have shown up in the purity/efficiency comparisons of Table I. I therefore keep ACCEPT (verdict_should_be = UNCHANGED). The concrete test is a data-driven proton-efficiency check that both the collaboration and an external group using the released selection could run; it is the single check that would settle whether the backward-angle and subleading-proton findings are physics or partly detector response. My disagreement with the reader is only in emphasis: I attribute less evidential weight to the resimulated-data/MC comparison for the efficiency component, and I make the acknowledgment already present in b456 into the focal point rather than treating it as resolved.

If the result stands, it demonstrates that current final-state interaction and multi-nucleon models are inadequate for the dominant mesonless channel at few-GeV energies, with a specific, localization of the failure in backward-angle single-proton events. The paper's strengths are substantial: the four GENIE CCQE/CCMEC tune parameters are taken from Ref. [18] and are not fitted here, so the tested predictions are independent of the measured data; the per-block chi-square values are reported honestly with no post-hoc exclusions; the NuWro fake-data study recovers an input shape different from the response-matrix prior; regularization bias is handled via the AC smearing matrix; and the data release includes covariance matrices separated by source, the unfolding and AC matrices, and the range-projected chi2 machinery, enabling independent model comparison. If adopted, these measurements and the released covariance structure constitute a benchmark for the SBN and DUNE generator programs.

This paper reports the first double-differential cross-section measurements of charged-current muon-neutrino interactions on argon that produce no mesons and one or more protons (the CC0πNp topology), binned jointly in proton multiplicity and the momentum and muon-proton opening angle of each of the four leading protons, plus the total hadronic available energy. The analysis uses the full MicroBooNE Booster Neutrino Beam dataset of $1.30 \times 10^{21}$ protons on target, unfolding eleven correlated distributions with a blockwise scheme that keeps cross-block correlations in the covariance. The central finding is that no tested generator reproduces the dataset: agreement is worst in proton multiplicity at backward muon-proton opening angles and for subleading proton kinematics, and the spread among model predictions is driven by the treatment of final-state interactions. The authors conclude that existing hadronic transport models are inadequate for this dominant mesonless channel at few-GeV neutrino energies, which bears directly on the systematic uncertainties of the DUNE and SBN oscillation programs.

What carries the argument. The measurement rests on the detector response matrix, factorized as selection efficiency $\epsilon_\mu$ times a migration matrix $M$, built from the GENIE MicroBooNE tune and applied in a blockwise unfolding scheme in which eleven double-differential distributions are unfolded independently while their cross-block correlations are carried in the covariance. Regularization bias is accounted for by the additional smearing matrix $A_C = U\Delta$, and global goodness-of-fit is evaluated with a range-projected $\chi^2$ statistic that removes the 38 constraints induced by event sharing between blocks. Because protons are ranked by momentum, the migration matrices must also absorb the roughly 10% rate at which reconstruction swaps the leading and subleading proton ranks.

Core claim. The paper establishes that the number and kinematics of final-state protons in CC0πNp events on argon are far more sensitive to how a generator models hadron propagation through the nucleus than current models handle. Against the full MicroBooNE dataset, the best global description comes from the GiBUU transport framework with in-medium corrections to nucleon-nucleon cross sections, yet even that achieves only $\chi^2/\mathrm{ndof} \approx 429/145$; the worst performers, NEUT and the GENIE AR23 hN cascade, exceed 1100 and 1200. A signature tension appears in single-proton events at backward opening angles, where data exceed every prediction, and in the second-leading proton, where all generators give $\chi^2/\mathrm{ndof}$ well above one. Through sideband and resimulation studies the paper argues that this disagreement is signal mismodeling rather than background or detector-response error.

The premise it rests on. The unfolded cross sections are correct only if the GENIE MicroBooNE tune simulation adequately captures selection efficiency and reconstruction smearing in every reported bin, especially for low-momentum subleading protons near the 0.25 GeV/c threshold.

No generator passing a global fit means that final-state interaction cascade models must be retuned before the CC0πNp channel can safely anchor neutrino-energy reconstruction in DUNE or SBN.

The spread in global $\chi^2$ across GENIE AR23 FSI variants alone (488 to 1200) shows that proton-multiplicity observables are a discriminating test bed for final-state interaction modeling.

The backward-angle, single-proton excess is echoed by MicroBooNE electron-neutrino measurements on both beamlines, pointing to a flavor-independent feature of the hadronic final state that any generator must capture.

The released data, source-by-source covariance matrices, and unfolding matrices allow any generator to be compared on all 183 bins without redoing the analysis.

If the backward-angle excess is genuine physics rather than a reconstruction artifact, then generators that suppress wide-angle emission, for example by hardening the nucleon-nucleon cross section in the medium, should move toward the data; this can be tested by refitting within the released covariance matrix.

All generators reproduce the near-isotropy of the subleading proton angular distributions, suggesting that rescattering washes out the primary-emission geometry; the discriminating information for FSI therefore sits in the leading proton, not the subleading ones.

The claim of bias-free unfolding is the softest load-bearing point: a future response matrix built entirely from a non-GENIE simulation could shift unfolded cross sections by more than the quoted systematic envelope, which would move the central measurement itself.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review