Minutes by Claude+transcript
NIUWG Meeting, 9 November 2026
Chair: Laura Munteanu
MINERvA open data as a near detector sample (Rik Gran)
Rik thanked Facundo López and Thomas Lund, two undergraduates who did most of this work over the summer. All plots are work in progress.
Sample and setup
- MINERvA tracker-fiducial CC inclusive events with MINOS-matched muons, θμ < ~20° (sliding cut). About 5M events split between FHC ν and RHC ν̄.
- The hadronic energy estimator uses a subset of the downstream HCAL energy and is a good proxy for available energy.
- The aim is not a cross-section extraction. The sample is exported to GUNDAM and binned in 3D: pT, pz and reconstructed hadronic energy. That gives 1500 bins per mode (3000 total), including an overflow bin. A separate q3 vs. Eavail binning exists as a cross-check but was not shown.
- The model is GENIE v2 with a MINERvA tune close to the Ascencio inclusive paper. Coherent is weighted up to stand in for diffractive.
- The flux now includes the low-ν constraint, which only acts above ~7.5 GeV.
- The x-axis bins grow by 10% each and are not bin-width normalised, so the area under each curve is the true event count.
Fit configuration and status
- The fit uses full statistics, with ν and ν̄ fitted simultaneously.
- There are 54 parameters: the native GENIE v2 knobs, knobs built from MINERvA tunes, and a few ported to or from nusystematics.
- Flux has three simplified parameters (5% normalisation, a high-energy tail and a low-energy tail).
- No detector systematics yet. They are lateral shifts and need a dedicated GUNDAM interface. No uncertainties on the prefit prediction yet either.
- The fit converges, but about 10–12 parameters (roughly a third) sit at their ±3σ limits. Parameter values are therefore not interpretable and were not shown; only the post-fit predictions were.
- This was expected at these statistics because there isn’t enough freedom, and ~100 parameters are probably needed. Fitting 1/10 of the data gives more sensible results and could serve as a progress metric.
Physics observations (most have a counterpart in earlier MINERvA publications)
- Prefit, QE is under-predicted while 2p2h, Δ and higher resonances are over-predicted, by ~20%. The trend is the same in ν and ν̄.
- Prediction is low at high W (~3 GeV), low Q² and high energy transfer. This region is largely unstudied, even by MINERvA, and no knob covers it.
- ν̄ QE neutron final states populate the lowest hadronic energy bins, which is the behaviour Eavail was designed around.
- Post-fit residuals remain at 5–10%.
- Just above the flux peak there is wide-ranging tension in ν mode. This could be residual flux uncertainty that the three simple flux parameters can’t absorb.
- DIS is split into three Q² ranges (< 1, 1–4 and > 4 GeV²), following discussion with Jorge Morfín.
- The Q² ~ 1–2 transition region is front and centre in MINERvA data.
- Some isolated bins are ~5σ off with no knob to move them.
- The sample overlaps strongly with the ongoing MINERvA DIS analysis.
- In the highest panel (pT > 2 GeV, pz < 40 GeV), the low-x excess (x < ~0.15) looks like shadowing in ν but not in ν̄. Energy-scale systematics could also contribute.
- Any 3D inclusive basis spans the kinematic space. The fit can therefore in principle see:
- Q²-dependent form factors
- the energy dependence of W3
- individual resonances in W
- GENIE’s RES/DIS transition boundary
- A first look with six W-binned resonance knobs suggests some sensitivity in W.
- Resolutions:
- Eμ/pz: ~10%. Binning is coarser than this.
- pT: ~12%. Binning is also coarser.
- Hadronic energy: ~30%, non-Gaussian and model dependent. Binning is up to 5× finer than this at high Ehad and 1–2× at low Ehad.
- Advertisement: a new preprint applies PDG methods to inclusive σ/E from historical data (NOMAD, MINOS, MINERvA).
Discussion
Detector systematics (Stephen Dolan and Ciro Riccio, in chat and live)
- Option 1: build toys in the MINERvA framework and add detector freedom through a covariance matrix, as T2K did for years. The matrix could be reduced with PCA.
- Option 2: convert event-by-event lateral shifts into binned response functions. This avoids ~3000 parameters.
- Rik said that for this calorimetric sample, MINERvA’s detector systematics are mostly 4–5 energy scales. The Geant-based tracking systematics may be skippable because no hadron tracking is used.
- Ciro offered help with the GUNDAM implementation. Rik thanked the GUNDAM team for their support over the summer.
- Rik has limited effort for the next 6–8 months and will focus on DIS parameters, so detector systematics are unlikely to go in before next summer.
High-pT tail (Jeremy Wolcott)
- Jeremy asked whether the high-pT tail could be a flux effect. Rik thinks this is unlikely to be the main cause and noted that the panel shares its events with the DIS analysis, which could be used to isolate the question.
- Jeremy noted the sample is on CH and that GENIE’s shadowing is scaled from iron, so a discrepancy this large is a surprise, although it sits in a corner of phase space. Rik agreed.
Off-scale points and missing freedom (Ciro Riccio)
- In the highest panel, some ratio points go off scale (~1.5–1.6). This is the only place in the sample where that happens.
- Ciro observed that the model does better at high energy than in the QE/Δ/RES region, where more freedom is needed.
- Rik pointed to the statistically powerful gaps: isolated bins, the high-W low-Q² region, and groups of bins that matter in aggregate.
- The large post-fit RES reduction (to ~1/3) is likely a symptom of a parameter being pulled to its limit.
Precision needed for DUNE (Stephen Dolan)
- DUNE needs cross-section uncertainties below 2% and energy-scale control at the level of tens of MeV. The bins here are hundreds of MeV wide.
- Small effects can bias a Δm² measurement: more energy given to neutrons in the cascade, or a different fraction of primary neutron final states.
- Stephen suggested projecting the post-fit into finer hadronic energy bins in the peak regions (e.g. slide 11), and comparing the energy-scale shift between prefit and post-fit.
- Rik agreed that finer Ehad binning has merit where statistics allow (2k–20k events per bin near the flux peak), but not finer pT/pz binning.
- Two lateral shifts are already in the base model, both motivated by GENIE’s removal energy being unrealistic:
- −25 MeV for resonances. A hand scan by Thomas confirmed the data prefers ~25 MeV, in line with what Ascencio found.
- +5 MeV for half of the QE events.
- Neither shift is exposed to GUNDAM as a free parameter yet.
FSI dials (Laura Munteanu)
- Laura asked about studies of individual dial impacts. She noted that GENIE FSI dials often appear to have an unrealistically small effect.
- Rik said earlier undergraduate projects did knob-by-knob studies for QE and RES, but not FSI.
- For a calorimetric observable, FSI changes the fraction of energy going to neutrons rather than the total hadronic energy, which explains the small effect. TKI variables would be more sensitive.
- The low-Ehad neutron final-state region and some off-peak regions may still show FSI effects.
- The group has Oscar Moreno’s INCL fate-by-fate weights and plans to try them.
Fake-data studies (Jeremy Wolcott, Deborah Harris)
- Jeremy said DUNE also needs to estimate how mismodelling propagates to oscillation results. He proposed building fake-data studies from this experience and running them through the DUNE oscillation fit.
- Rik said this would be easy, using physically motivated distortions GENIE knows nothing about, and volunteered to help. Combining with exclusive samples (e.g. TKI) would be harder.
- Jeremy said the first-order goal is a serious fake-data set that breaks the current ND fit.
- Deborah suggested the interesting question is distinguishing cross-section effects from detector effects that produce the same discrepancy.
- Jeremy replied that exactly degenerate knobs are hard to construct in DUNE. The more valuable exercise is a sequence of fake-data sets with effects outside the model, to catalogue which ones bias oscillation results (high-Q² DIS probably won’t, others will).
Action items
- Rik and team: expand the parameter set toward ~100, with a focus on DIS. Propagate prefit uncertainties. Implement detector energy-scale systematics, likely next summer.
- Ciro and the GUNDAM team: support the detector systematics implementation (covariance/PCA or response functions).
- Rik: try the INCL weights. Look at finer Ehad binning of the post-fit in the peak regions.
- Jeremy and Rik: discuss a fake-data study programme for the DUNE oscillation fit over the coming month.
Next meeting in two weeks (23 November).
There are minutes attached to this event.
Show them.