arxiv: 2604.21869 · v1 · submitted 2026-04-23 · ✦ hep-ph

Recognition: unknown

Analytical and Machine Learning Methods for Model Discernment at CEνNS Experiments

Bhaskar Dutta, Doojin Kim, Iain A. Bisset, Joel W. Walker, Samiran Sinha

Authors on Pith no claims yet

Pith reviewed 2026-05-09 21:28 UTC · model grok-4.3

classification ✦ hep-ph

keywords CEνNSsterile neutrinosnon-standard interactionsmodel discriminationshape analysisconvolutional neural networkscoherent neutrino scatteringneutrino new physics

0 comments

The pith

Shape information from baseline, recoil energy and timing distinguishes sterile neutrinos from NSI at CEνNS even after removing total rate.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that in stopped-pion CEνNS experiments, retaining multidimensional correlations among baseline, recoil energy and timing separates a 3+1 sterile-neutrino scenario from neutral-current non-standard interactions in nontrivial regions of parameter space. This matters because total-rate comparisons alone are vulnerable to source-normalization uncertainties and can leave models degenerate. Likelihood analyses recover the separation power, while convolutional neural networks continue to discriminate after the overall event count is explicitly removed from the input, confirming that the distinguishing information resides in the shape of the distribution. The same observables also permit approximate localization of the underlying sterile benchmark point in favorable regions of parameter space.

Core claim

Using multidimensional shape information from baseline, recoil energy, and timing in CEνNS, the 3+1 sterile-neutrino framework and neutral-current NSI can be distinguished in nontrivial regions of parameter space. Likelihood-based analysis and convolutional neural networks demonstrate that substantial discrimination persists even after the total event rate is removed, showing that the relevant information is encoded in the shape of the CEνNS distribution. Multi-class classification further supports approximate localization of sterile-neutrino benchmark points.

What carries the argument

Multidimensional correlations in baseline, recoil energy, and timing, extracted through likelihood ratios and convolutional neural networks, to discriminate BSM models while remaining insensitive to overall normalization.

If this is right

Model discrimination power increases when shape information supplements rate-only analyses in low-statistics neutrino experiments.
Convolutional neural networks can extract model-discriminating features without access to the total event normalization.
The same observables support approximate localization of sterile-neutrino parameters in favorable regions.
Neutrino new-physics searches can transition from anomaly detection to physics interpretation by retaining multidimensional shape data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar shape-based methods could be tested in other low-statistics neutrino setups such as reactor or solar experiments where normalization uncertainties are also large.
Detector designs that improve timing and spatial resolution may gain extra reach for new-physics interpretation beyond what raw statistics alone provide.
Combining analytical likelihoods with machine-learning classifiers offers a general template for resolving rate-degenerate BSM scenarios in future neutrino data.
The approach suggests that experiments should archive and analyze full event-by-event information rather than collapsing to one-dimensional rate measurements.

Load-bearing premise

The simulated event distributions accurately capture the shape differences between the 3+1 sterile and NSI benchmarks under realistic detector conditions, and these shapes are less sensitive to source-normalization uncertainties than the total rate.

What would settle it

Collecting real CEνNS data in which the multidimensional likelihood or CNN classifier shows no separation power between the sterile and NSI benchmarks after standard systematics are applied would falsify the discrimination claim.

read the original abstract

Neutrino experiments are often limited by low statistics, sizable systematic uncertainties, and coarse observable binning, which can hinder discrimination among competing beyond-the-Standard-Model (BSM) explanations of anomalous signals. In particular, analyses based primarily on total event-rate differences are vulnerable to source-normalization uncertainties and to degeneracies among models that induce similar inclusive yields. Using stopped-pion coherent elastic neutrino-nucleus scattering (CE$\nu$NS) as a benchmark environment, we study how much model-discrimination power can be obtained from correlations in baseline, recoil energy, and timing that are less sensitive to the total rate. As benchmark BSM scenarios, we consider a $3+1$ sterile-neutrino framework and neutral-current non-standard neutrino interactions (NSI). We show with a likelihood-based analysis that these scenarios can be distinguished in nontrivial regions of parameter space once multidimensional shape information is retained. We further demonstrate with convolutional neural networks that substantial discrimination remains possible even after the total event rate is explicitly removed from the input, indicating that the relevant information is genuinely encoded in the shape of the CE$\nu$NS distribution. Finally, through multi-class classification within the sterile parameter space, we show that in favorable regions the same observables can support approximate localization of the underlying sterile-neutrino benchmark point. Our results highlight the complementary roles of conventional and machine-learning-based inference in moving neutrino new-physics searches from anomaly detection to physics interpretation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Shape correlations in baseline, recoil energy and timing let you separate 3+1 sterile neutrinos from NSI in stopped-pion CEνNS, and CNNs keep the separation even after the total rate is stripped out.

read the letter

The main takeaway is that in stopped-pion CEνNS, keeping the joint distribution across baseline, recoil energy and timing gives you nontrivial separation between a 3+1 sterile neutrino scenario and neutral-current NSI. The paper shows this with a standard binned likelihood and then repeats the exercise with convolutional networks after explicitly removing the overall normalization from the input. They also run a multi-class version that can localize the sterile parameters in favorable regions of space. This is the concrete advance: moving from rate-only anomaly hunting to shape-driven model discrimination in a low-statistics neutrino setting where normalization uncertainties are large. The CNN check is useful because it confirms the information really lives in the shape rather than just the count. The work is straightforward and directly tied to an existing experimental setup, so the idea is easy to test or adapt. The soft spot is the Monte Carlo foundation. The separation power rests on how faithfully the simulated event distributions capture the differences between the two benchmarks once realistic detector response, energy resolution, timing jitter and acceptance variations are included. If those effects introduce correlated shape distortions that the two models feel differently, the reported discrimination will shrink. The abstract claims the results hold, but without seeing the exact treatment of shape systematics in both the likelihood and the network training, it is difficult to judge how conservative the numbers are. This is the kind of paper experimentalists and phenomenologists working on CEνNS or similar neutrino scattering experiments would want to see. A reader who needs practical tools for breaking BSM degeneracies in low-rate data will get something usable. It deserves a serious referee because the central claim is testable and the methods are not exotic. I would send it to review and ask the referees to focus on the simulation validation and any cross-checks against real detector performance.

Referee Report

2 major / 2 minor

Summary. The paper claims that in stopped-pion CEνNS experiments, 3+1 sterile-neutrino and NSI BSM scenarios can be distinguished in nontrivial regions of parameter space by retaining multidimensional shape information in baseline, recoil energy, and timing. Likelihood analysis supports this distinction, while CNNs demonstrate that substantial discrimination power remains even after the total event rate is explicitly removed from the input. Multi-class classification is further shown to enable approximate localization of sterile-neutrino benchmark points in favorable regions.

Significance. If the central results hold under realistic conditions, the work demonstrates the complementary value of shape-based observables for moving neutrino new-physics searches from anomaly detection to model interpretation. The explicit demonstration that shape correlations suffice even without rate information, together with the use of both likelihood and CNN methods, is a clear strength and addresses a common limitation in low-statistics experiments.

major comments (2)

[Event generation and detector modeling] Event generation and detector modeling: The central claim that shape correlations in baseline-recoil-timing are less sensitive to source-normalization uncertainties and allow nontrivial discrimination rests on the fidelity of the Monte Carlo event distributions. The manuscript must explicitly show how correlated detector effects (energy resolution, timing jitter, baseline-dependent acceptance, nuclear form-factor uncertainties) are modeled and propagated through both the binned likelihood and CNN training, since these effects could impact the two BSM benchmarks differently.
[Likelihood analysis] Likelihood analysis: The abstract asserts that likelihood results support distinction once multidimensional shape information is retained, yet quantitative performance metrics, background modeling details, and systematic treatment are not provided at a level that allows verification of the reported separation power.

minor comments (2)

[Abstract] The abstract could usefully indicate the specific ranges of sterile mass/mixing or NSI parameters where the reported discrimination holds.
[CNN analysis] In the CNN section, explicitly state the procedure used to remove the total event rate from the input features to confirm that no implicit rate information remains.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation of the work's significance and for the constructive major comments, which identify areas where additional detail will improve verifiability. We address each point below and will revise the manuscript to incorporate the requested clarifications and quantitative information.

read point-by-point responses

Referee: Event generation and detector modeling: The central claim that shape correlations in baseline-recoil-timing are less sensitive to source-normalization uncertainties and allow nontrivial discrimination rests on the fidelity of the Monte Carlo event distributions. The manuscript must explicitly show how correlated detector effects (energy resolution, timing jitter, baseline-dependent acceptance, nuclear form-factor uncertainties) are modeled and propagated through both the binned likelihood and CNN training, since these effects could impact the two BSM benchmarks differently.

Authors: We agree that a more explicit description of detector modeling is required to support the central claims. The present manuscript outlines the overall event-generation approach but does not provide the requested level of detail on how the listed correlated effects are implemented and propagated. In the revised version we will add a dedicated subsection (new Section 3.2) that specifies: the functional form and numerical values used for energy-resolution and timing-jitter smearing; the baseline-dependent acceptance parametrization; the nuclear form-factor model together with its uncertainty treatment; and the precise manner in which these effects are folded into the binned histograms for the likelihood analysis and into the two-dimensional image inputs for the CNN. We will also include a short discussion of any differential impact on the sterile-neutrino versus NSI benchmarks. These additions will allow readers to assess the robustness of the reported shape-based discrimination. revision: yes
Referee: Likelihood analysis: The abstract asserts that likelihood results support distinction once multidimensional shape information is retained, yet quantitative performance metrics, background modeling details, and systematic treatment are not provided at a level that allows verification of the reported separation power.

Authors: We acknowledge that the likelihood section would benefit from additional quantitative metrics and explicit background and systematic information to permit independent verification. While the manuscript presents the overall methodology and qualitative separation results, it does not tabulate specific test-statistic values, background rates and shapes, or the full treatment of nuisance parameters. In the revision we will expand the likelihood analysis (Section 4) to include: (i) quantitative performance metrics such as expected separation significances or profiled likelihood ratios for the benchmark points; (ii) a detailed description of the background model, including its normalization, spectral shape, and associated uncertainties; and (iii) the systematic covariance matrix together with the profiling procedure used for source-normalization and detector-related uncertainties. These elements will be presented in the main text or a supplementary appendix so that the claimed discrimination power can be reproduced. revision: yes

Circularity Check

0 steps flagged

No circularity; discrimination claims rest on forward simulation of distinct BSM models

full rationale

The paper derives its model-discrimination results from explicit Monte Carlo generation of event distributions for two independent BSM scenarios (3+1 sterile neutrinos and NSI) in a stopped-pion CEνNS setup, followed by binned likelihood analysis and CNN training on multidimensional observables (baseline, recoil energy, timing). No equations or procedures reduce a claimed prediction to a fitted input by construction, no load-bearing self-citations justify uniqueness theorems, and no ansatz is smuggled via prior work. The abstract and description frame the retained shape information as an output of the simulation pipeline, not a tautological re-expression of normalization or rate. This is the standard non-circular case for simulation-based inference papers.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the assumption that simulated CEνNS event distributions for the two benchmark models faithfully represent the shape differences observable in a real detector.

axioms (1)

domain assumption The CEνNS cross sections, nuclear form factors, and detector response functions are accurately modeled in the Monte Carlo simulation used for both likelihood and CNN studies.
Required for the reported shape-based discrimination to translate to real data.

pith-pipeline@v0.9.0 · 5574 in / 1241 out tokens · 28658 ms · 2026-05-09T21:28:00.295521+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

26 extracted references · 20 canonical work pages

[1]

Acero, C.A

M.A. Acero, C.A. Arg¨ uelles, M. Hostert, D. Kalra, G. Karagiorgi, K.J. Kelly et al.,White paper on light sterile neutrino searches and related phenomenology,Journal of Physics G: Nuclear and Particle Physics51(2024) 120501

2024
[2]

Bisset, B

I.A. Bisset, B. Dutta, W.-C. Huang and L.E. Strigari,Short baseline neutrino anomalies at stopped pion experiments,Journal of High Energy Physics2024(2024) 3. [3]Neutrino Non-Standard Interactions: A Status Report, vol. 2, 2019. 10.21468/SciPostPhysProc.2.001. [4]COHERENTcollaboration,Observation of Coherent Elastic Neutrino-Nucleus Scattering, Science357...

work page doi:10.21468/scipostphysproc.2.001 2024
[3]

Dark matter signals from timing spectra at neutrino experiments,

B. Dutta, D. Kim, S. Liao, J.-C. Park, S. Shin and L.E. Strigari,Dark matter signals from timing spectra at neutrino experiments,Phys. Rev. Lett.124(2020) 121802 [1906.10745]

work page arXiv 2020
[4]

Dutta, D

B. Dutta, D. Kim, S. Liao, J.-C. Park, S. Shin, L.E. Strigari et al.,Searching for dark matter signals in timing spectra at neutrino experiments,JHEP01(2022) 144 [2006.09386]. [10]COHERENTcollaboration,First Probe of Sub-GeV Dark Matter beyond the Cosmological Expectation with the COHERENT CsI Detector at the SNS,Phys. Rev. Lett.130(2023) 051803 [2110.114...

work page arXiv 2022
[5]

Barinov et al.,Search for electron-neutrino transitions to sterile states in the BEST experiment,Phys

V.V. Barinov et al.,Search for electron-neutrino transitions to sterile states in the BEST experiment,Phys. Rev. C105(2022) 065502 [2201.07364]. [17]SAGEcollaboration,Measurement of the response of the Russian-American gallium experiment to neutrinos from a Cr-51 source,Phys. Rev. C59(1999) 2246 [hep-ph/9803418]. [18]GALLEXcollaboration,Final results of t...

work page arXiv 2022
[6]

Backhouse and R.B

C. Backhouse and R.B. Patterson,Library Event Matching event classification algorithm for electron neutrino interactions in the NOνA detectors,Nucl. Instrum. Meth. A778(2015) 31 [1501.00968]

work page arXiv 2015
[7]

Aurisano, A

A. Aurisano, A. Radovic, D. Rocco, A. Himmel, M.D. Messier, E. Niner et al.,A Convolutional Neural Network Neutrino Event Classifier,JINST11(2016) P09001 [1604.01444]. [22]MicroBooNEcollaboration,Convolutional Neural Networks Applied to Neutrino Events in a Liquid Argon Time Projection Chamber,JINST12(2017) P03011 [1611.05531]. [23]MicroBooNEcollaboration...

work page arXiv 2016
[8]

Baldi, J

P. Baldi, J. Bian, L. Hertel and L. Li,Improved Energy Reconstruction in NOvA with Regression Convolutional Neural Networks,Phys. Rev. D99(2019) 012011 [1811.04557]. [25]KM3NeTcollaboration,Event reconstruction for KM3NeT/ORCA using convolutional neural networks,JINST15(2020) P10005 [2004.08254]. [26]DUNEcollaboration,Neutrino interaction classification w...

work page arXiv 2019
[9]

Villarreal, J

J. Villarreal, J. Woodward, J. Hardin and J. Conrad,Machine learning-informed 3+1 sterile neutrino global fits using posterior density estimation of electron disappearance data,The European Physical Journal C86(2026) 326

2026
[10]

Psihas, M

F. Psihas, M. Groh, C. Tunnell and K. Warburton,A Review on Machine Learning for Neutrino Experiments,Int. J. Mod. Phys. A35(2020) 2043005 [2008.01242]

work page arXiv 2020
[11]

D. Kim, K. Kong, K.T. Matchev, M. Park and P. Shyamsundar,Deep-learned event variables for collider phenomenology,Phys. Rev. D107(2023) L031904 [2105.10126]

work page arXiv 2023
[12]

Franceschini, D

R. Franceschini, D. Kim, K. Kong, K.T. Matchev, M. Park and P. Shyamsundar,Kinematic variables and feature engineering for particle phenomenology,Rev. Mod. Phys.95(2023) 045004 [2206.13431]

work page arXiv 2023
[13]

Anderson, J

P. Agrawal, N. Craig, A. Madden and I.V. Lombera,The FERMIACC: Agents for Particle Theory,2603.22538

work page arXiv
[14]

Ai agents can already autonomously perform experimental high energy physics,

E.A. Moreno, S. Bright-Thonney, A. Novak, D. Garcia and P. Harris,AI Agents Can Already Autonomously Perform Experimental High Energy Physics,2603.20179. [36]SHiNESScollaboration,Search for hidden neutrinos at the European Spallation Source: the SHiNESS experiment,JHEP03(2024) 148 [2311.18509]. [37]COHERENTcollaboration,The COHERENT Experiment: 2026 Updat...

work page arXiv 2024
[15]

Probing the dark sector with accelerators: New opportunities!

R. Van de Water, “Probing the dark sector with accelerators: New opportunities!.” https://www.int.washington.edu/program/schedule/1205, April 17-21, 2023

2023
[16]

Freedman,Coherent Neutrino Nucleus Scattering as a Probe of the Weak Neutral Current,Phys

D.Z. Freedman,Coherent Neutrino Nucleus Scattering as a Probe of the Weak Neutral Current,Phys. Rev. D9(1974) 1389

1974
[17]

Scholberg,Prospects for measuring coherent neutrino-nucleus elastic scattering at a stopped-pion neutrino source,Phys

K. Scholberg,Prospects for measuring coherent neutrino-nucleus elastic scattering at a stopped-pion neutrino source,Phys. Rev. D73(2006) 033005 [hep-ex/0511042]

work page arXiv 2006
[18]

Formaggio and G.P

J.A. Formaggio and G.P. Zeller,From eV to EeV: Neutrino Cross Sections Across Energy Scales,Rev. Mod. Phys.84(2012) 1307 [1305.7513]

work page arXiv 2012
[19]

Helm,Inelastic and Elastic Scattering of 187-Mev Electrons from Selected Even-Even Nuclei,Phys

R.H. Helm,Inelastic and Elastic Scattering of 187-Mev Electrons from Selected Even-Even Nuclei,Phys. Rev.104(1956) 1466

1956
[20]

Mirzakhani et al.,MINER reactor based search for axionlike particles using sapphire (Al2O3) detectors,Phys

M. Mirzakhani et al.,MINER reactor based search for axionlike particles using sapphire (Al2O3) detectors,Phys. Rev. D112(2025) 032013 [2504.20960]

work page arXiv 2025
[21]

Iyer et al.,Large mass single electron resolution detector for dark matter and neutrino elastic interaction searches,Nucl

V. Iyer et al.,Large mass single electron resolution detector for dark matter and neutrino elastic interaction searches,Nucl. Instrum. Meth. A1010(2021) 165489 [2011.02234]

work page arXiv 2021
[22]

Dutta, Y

B. Dutta, Y. Gao, R. Mahapatra, N. Mirabolfathi, L.E. Strigari and J.W. Walker,Sensitivity to oscillation with a sterile fourth generation neutrino from ultra-low threshold neutrino-nucleus coherent scattering,Phys. Rev. D94(2016) 093002 [1511.02834]

work page arXiv 2016
[23]

Raftery,Bayesian model selection in social research,Sociological Methodology25(1995) 111

A.E. Raftery,Bayesian model selection in social research,Sociological Methodology25(1995) 111

1995
[24]

Multimodal nested sampling: an efficient and robust alternative to MCMC methods for astronomical data analysis

F. Feroz and M.P. Hobson,Multimodal nested sampling: an efficient and robust alternative to MCMC methods for astronomical data analysis,Mon. Not. Roy. Astron. Soc.384(2008) 449 [0704.3704]

work page Pith review arXiv 2008
[25]

MultiNest: an efficient and robust Bayesian inference tool for cosmology and particle physics

F. Feroz, M.P. Hobson and M. Bridges,MultiNest: an efficient and robust Bayesian inference tool for cosmology and particle physics,Mon. Not. Roy. Astron. Soc.398(2009) 1601 [0809.3437]. – 25 –

work page Pith review arXiv 2009
[26]

Feroz, M

F. Feroz, M.P. Hobson, E. Cameron and A.N. Pettitt,Importance Nested Sampling and the MultiNest Algorithm,Open J. Astrophys.2(2019) 10 [1306.2144]. – 26 –

work page arXiv 2019