pith. sign in

arxiv: 2605.31190 · v1 · pith:TORYKZZAnew · submitted 2026-05-29 · ⚛️ physics.ins-det · hep-ex

Exploring Low Energy Excess in MINER with sapphire detectors using Convolutional Variational Autoencoder (CVAE)

D. Mondal , W. Baker , M. Chaudhuri , J. B. Dent , B. Dutta , V. Iyer , A. Jastram , V. K. S. Kashyap
show 14 more authors
A. Kubik K. Lang R. Mahapatra S. Maludze N. Mirabolfathi M. Mirzakhani B. Mohanty H. Neog J. L. Newstead M. Platt S. Sahoo J. Sander L. E. Strigari J. Walker
This is my paper

Pith reviewed 2026-06-28 19:59 UTC · model grok-4.3

classification ⚛️ physics.ins-det hep-ex
keywords low-energy excesscryogenic detectorssapphire detectorconvolutional variational autoencoderrise-time discriminationbackground mitigationMINER experiment
0
0 comments X

The pith

A convolutional variational autoencoder identifies a rise-time deviation allowing rejection of 53 percent of low-energy excess events in sapphire detectors.

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

The paper establishes that low-energy excess background persists in cryogenic sapphire detectors at energies near 200 eV and reappears after each warm-up cycle. An unsupervised CVAE model flags anomalous pulse shapes and reveals that LEE events exhibit a measurable rise-time shift relative to Monte Carlo ideal signals. From this feature the authors construct a rise-time selection cut that removes up to 53 percent of the excess while projecting a nearly 10 percent sensitivity gain for the MINER experiment at HFIR. The result is framed as evidence that a substantial fraction of LEE arises from bulk crystal defects or microfractures, offering a data-driven mitigation path for future low-threshold searches.

Core claim

A convolutional variational autoencoder applied to pulse data from a sapphire detector in the MINER experiment identifies a significant rise-time deviation in low-energy excess events compared with Monte Carlo simulated ideal signals. A discrimination pipeline based on this rise-time feature achieves up to 53 percent rejection of LEE events, corresponding to an expected sensitivity improvement of nearly 10 percent for MINER at HFIR. The observations remain consistent with a scenario in which a substantial fraction of the LEE originates from bulk-related defects or microfractures within the detector crystal.

What carries the argument

Convolutional variational autoencoder that produces a reconstruction-based anomaly score to isolate events whose pulse shapes deviate from the training distribution, specifically surfacing the rise-time difference used for discrimination.

If this is right

  • The rise-time cut can be deployed in the MINER analysis at HFIR to reduce LEE background and improve reach for rare-event searches.
  • Similar CVAE-based anomaly detection may be applied to other cryogenic detector materials facing LEE.
  • Material studies targeting bulk defects or microfractures could further suppress the excess beyond the 53 percent level achieved by timing selection.
  • Reproducible LEE reappearance after warm-up cycles indicates the background component is intrinsic to the detector rather than transient external sources.

Where Pith is reading between the lines

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

  • The method could be tested on other detector substrates to determine whether the rise-time signature is material-specific or universal.
  • Combining the CVAE anomaly score with additional pulse-shape variables might yield higher rejection without further efficiency cost.
  • If the defect hypothesis is correct, controlled irradiation or annealing studies on sapphire crystals would provide a direct check on LEE production rates.

Load-bearing premise

The rise-time deviation found by the CVAE is a stable, reproducible feature that separates LEE from true signals without large acceptance losses or analysis bias.

What would settle it

An independent dataset containing both LEE events and injected true signals at approximately 200 eV, with measured signal efficiency after the rise-time cut applied, would directly test whether the claimed rejection and sensitivity gain hold.

read the original abstract

As cryogenic detectors push toward ever-lower energy thresholds, their sensitivity is increasingly constrained by a persistent low-energy background known as the low-energy excess (LEE). We report observation of LEE in the MINER experiment using a sapphire ($\mathrm{Al_2O_3}$) detector at energies around 200 eV, with the excess reproducibly reappearing after each non-operational warm-up period. To address this limiting background, we implement an unsupervised convolutional variational autoencoder (CVAE) framework that identifies anomalous events through a reconstruction-based anomaly score. For the first time in a pulse-shape driven analysis, we uncover a significant deviation in the rise-time of LEE events relative to Monte Carlo simulated ideal signals. Using this feature, we develop a discrimination pipeline based on rise-time selection. This method achieves up to 53\% rejection of LEE events, corresponding to an expected sensitivity improvement of nearly 10\% for MINER at HFIR. These findings are consistent with a scenario in which a substantial fraction of the LEE originates from bulk-related defects or microfractures within the detector crystal, while leaving room for additional detector-related contributions. Our result provides a powerful, data-driven pathway for mitigating LEE and enhancing the discovery potential of next-generation cryogenic experiments.

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.

Referee Report

3 major / 2 minor

Summary. The manuscript reports the observation of low-energy excess (LEE) events in a sapphire detector in the MINER experiment at energies around 200 eV. It describes an unsupervised convolutional variational autoencoder (CVAE) used to identify anomalous pulse shapes, leading to the discovery of a rise-time deviation in LEE events relative to Monte Carlo simulations. A rise-time selection pipeline is then applied, claimed to reject up to 53% of LEE events while yielding an expected sensitivity improvement of nearly 10% for MINER at HFIR. The authors interpret this as evidence that a substantial fraction of LEE may originate from bulk-related defects or microfractures.

Significance. If the discrimination performance holds after proper validation, the result would offer a practical, data-driven method for reducing a key background in cryogenic detectors, potentially improving the reach of experiments searching for low-mass dark matter or coherent neutrino scattering. The unsupervised CVAE approach and the identification of a reproducible rise-time feature are positive elements that could be broadly useful if the efficiency and bias are quantified.

major comments (3)
  1. [Abstract] Abstract: The central claim of 53% LEE rejection and ~10% sensitivity gain is presented without any reported signal efficiency, acceptance curves, or ROC analysis. This makes it impossible to determine whether the rise-time cut preserves high acceptance for true nuclear recoils or calibration events across the energy range of interest, or whether the quoted figures are affected by post-hoc optimization.
  2. [Abstract] Abstract: No systematic uncertainty budget, cross-checks against injected signals, or comparison of the rise-time distribution before/after the CVAE anomaly score cut is provided. Without these, the discrimination pipeline cannot be evaluated for analysis bias or stability across warm-up cycles.
  3. [Abstract] Abstract: The sensitivity projection of nearly 10% improvement assumes that background reduction is essentially cost-free; the absence of any efficiency metric or live-time loss estimate means the net gain cannot be verified from the presented information.
minor comments (2)
  1. [Abstract] The abstract states that the excess 'reproducibly reappearing after each non-operational warm-up period' but provides no quantitative measure of reproducibility (e.g., rate stability or pulse-shape consistency metrics).
  2. [Abstract] Notation for the CVAE anomaly score and the precise definition of the rise-time feature used in the selection pipeline are not defined in the abstract; these should be introduced with equations or explicit formulas in the methods section.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their detailed review and constructive comments. We agree that the abstract would benefit from additional details on efficiencies and uncertainties to allow proper evaluation of the claims. We will revise the manuscript to address these points and provide the requested information.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of 53% LEE rejection and ~10% sensitivity gain is presented without any reported signal efficiency, acceptance curves, or ROC analysis. This makes it impossible to determine whether the rise-time cut preserves high acceptance for true nuclear recoils or calibration events across the energy range of interest, or whether the quoted figures are affected by post-hoc optimization.

    Authors: We acknowledge this limitation in the current abstract. The full manuscript includes a description of the rise-time selection pipeline and its application to LEE events, but does not explicitly present acceptance curves or ROC analysis for signal events. We will add these elements, including signal efficiency as a function of energy and an ROC curve, to a new section in the revised manuscript and update the abstract accordingly. This will allow readers to assess the trade-off between LEE rejection and signal acceptance. revision: yes

  2. Referee: [Abstract] Abstract: No systematic uncertainty budget, cross-checks against injected signals, or comparison of the rise-time distribution before/after the CVAE anomaly score cut is provided. Without these, the discrimination pipeline cannot be evaluated for analysis bias or stability across warm-up cycles.

    Authors: We agree that these elements are important for validating the pipeline. The manuscript discusses the reproducibility of LEE after warm-up cycles, but lacks the specific cross-checks mentioned. In the revision, we will include a systematic uncertainty budget, results from injected signal simulations or data, and before/after comparisons of rise-time distributions. This will demonstrate the stability and lack of bias in the CVAE-based selection. revision: yes

  3. Referee: [Abstract] Abstract: The sensitivity projection of nearly 10% improvement assumes that background reduction is essentially cost-free; the absence of any efficiency metric or live-time loss estimate means the net gain cannot be verified from the presented information.

    Authors: The sensitivity improvement estimate in the manuscript is based on the observed LEE rejection rate. However, we recognize the need to account for any efficiency losses. We will revise the sensitivity section to include explicit signal efficiency metrics and any potential live-time impacts, providing a more complete assessment of the net sensitivity gain. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's central result is an empirical measurement: CVAE anomaly scores reveal a rise-time deviation in LEE events, after which a rise-time selection cut is applied and its rejection fraction (53%) is reported directly from the data. No equations, fitted parameters, or self-citations are invoked that would make this rejection rate or the ~10% sensitivity projection reduce to a definition or input by construction. The discrimination pipeline rests on an observed data feature rather than a tautological renaming or load-bearing self-reference, rendering the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; full methods section would be required to audit any implicit thresholds or simulation assumptions.

pith-pipeline@v0.9.1-grok · 5869 in / 1071 out tokens · 25296 ms · 2026-06-28T19:59:54.886185+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

27 extracted references · 11 canonical work pages · 2 internal anchors

  1. [1]

    Angloher et al.,Latest observations on the low energy excess in CRESST-III,SciPost Phys

    G. Angloher et al.,Latest observations on the low energy excess in CRESST-III,SciPost Phys. Proc.(2023) 013

    2023

  2. [2]

    Abeleet al.(NUCLEUS), Characterization of the Low Energy Excess using a NUCLEUS Al 2O3 detector, (2026), arXiv:2603.07687

    H. Abele et al.,Characterization of the Low Energy Excess using a NUCLEUSAl 2O3 detector,arXiv preprint(2026) [2603.07687]. [7]Ricochetcollaboration,First demonstration of 30 eVee ionization energy resolution with Ricochet germanium cryogenic bolometers,Eur. Phys. J. C84(2024) 186 [2306.00166]. [8]SuperCDMScollaboration,Investigating the sources of low-en...

    work page arXiv 2026

  3. [3]

    Adari et al.,EXCESS workshop: Descriptions of rising low-energy spectra,SciPost Phys

    P. Adari et al.,EXCESS workshop: Descriptions of rising low-energy spectra,SciPost Phys. Proc.9(2022) 001 [2202.05097]

    work page arXiv 2022

  4. [4]

    Baxter, R

    D. Baxter, R. Essig, Y. Hochberg, M. Kaznacheeva, B. von Krosigk, F. Reindl et al., Low-Energy Backgrounds in Solid-State Phonon and Charge Detectors,Ann. Rev. Nucl. Part. Sci.75(2025) 301 [2503.08859]

    work page arXiv 2025

  5. [5]

    Mondal et al.,Ceνns search with cryogenic sapphire detectors at miner: Results from the triga reactor data and future sensitivity at hfir,arXiv preprint(2025) [2510.15999]

    D. Mondal et al.,Ceνns search with cryogenic sapphire detectors at miner: Results from the triga reactor data and future sensitivity at hfir,arXiv preprint(2025) [2510.15999]

    work page arXiv 2025

  6. [6]

    Anthony-Petersen et al.,A stress-induced source of phonon bursts and quasiparticle poisoning,Nature Communications15(2024)

    2024

  7. [7]

    Romani,Aluminum relaxation as the source of excess low energy events in low threshold calorimeters,J

    R.K. Romani,Aluminum relaxation as the source of excess low energy events in low threshold calorimeters,J. Appl. Phys.136(2024) 124502 [2406.15425]

    work page arXiv 2024

  8. [8]

    Nordlund, F

    K. Nordlund, F. Kong, F. Djurabekova, M. Heikinheimo, K. Tuominen, A. Kuronen et al., Spontaneous damage annealing reactions as a possible source of low energy excess in semiconductor detectors,Phys. Rev. Mater.9(2025) 113603

    2025

  9. [9]

    P. Du, D. Egana-Ugrinovic, R. Essig and M. Sholapurkar,Sources of low-energy events in low-threshold dark-matter and neutrino detectors,Phys. Rev. X12(2022) 011009. – 19 –

    2022

  10. [10]

    Khosa, L

    C.K. Khosa, L. Mars, J. Richards and V. Sanz,Convolutional Neural Networks for Direct Detection of Dark Matter,J. Phys. G47(2020) 095201 [1911.09210]. [25]NEXTcollaboration,Demonstration of background rejection using deep convolutional neural networks in the NEXT experiment,JHEP01(2021) 189 [2009.10783]. [26]DAR WINcollaboration,Model-independent searches...

    work page arXiv 2020

  11. [11]

    Bernardini, L

    M. Bernardini, L. Mayer, D. Reed and R. Feldmann,Predicting dark matter halo formation in N-body simulations with deep regression networks,Mon. Not. Roy. Astron. Soc.496 (2020) 5116 [1912.04299]

    work page arXiv 2020

  12. [12]

    Variational Autoencoders for New Physics Mining at the Large Hadron Collider

    O. Cerri, T.Q. Nguyen, M. Pierini, M. Spiropulu and J.-R. Vlimant,Variational Autoencoders for New Physics Mining at the Large Hadron Collider,JHEP05(2019) 036 [1811.10276]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  13. [13]

    van Beekveld, S

    M. van Beekveld, S. Caron, L. Hendriks, P. Jackson, A. Leinweber, S. Otten et al., Combining outlier analysis algorithms to identify new physics at the LHC,JHEP09(2021) 024 [2010.07940]

    work page arXiv 2021

  14. [14]

    Ackermann et al.,First observation of reactor antineutrinos by coherent scattering,2501.05206

    N. Ackermann et al.,Direct observation of coherent elastic antineutrino-nucleus scattering, Nature643(2025) 1229 [2501.05206]. [31]TEXONO Collaborationcollaboration,New limits on the coherent neutrino-nucleus elastic scattering cross section at the kuo-sheng reactor-neutrino laboratory,Phys. Rev. Lett. 134(2025) 121802. [32]CONNIEcollaboration,Search for ...

    work page arXiv 2025

  15. [15]

    D. Bank, N. Koenigstein and R. Giryes,Autoencoders, 2021

    2021

  16. [16]

    Sakurada and T

    M. Sakurada and T. Yairi,Anomaly detection using autoencoders with nonlinear dimensionality reduction, inAnomaly Detection Using Autoencoders with Nonlinear Dimensionality Reduction, MLSDA’14, (New York, NY, USA), p. 4–11, Association for Computing Machinery, 2014, DOI

    2014

  17. [17]

    Kingma and M

    D.P. Kingma and M. Welling,Auto-encoding variational bayes, 2022

    2022

  18. [18]

    I.H. others,beta-vae: Learning basic visual concepts with a constrained variational framework, in5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings, OpenReview.net, 2017, https://openreview.net/forum?id=Sy2fzU9gl

    2017

  19. [19]

    Dumoulin and F

    V. Dumoulin and F. Visin,A guide to convolution arithmetic for deep learning, 2018. – 20 –

    2018

  20. [20]

    Abadi et al.,Tensorflow: Large-scale machine learning on heterogeneous distributed systems, 2016

    M. Abadi et al.,Tensorflow: Large-scale machine learning on heterogeneous distributed systems, 2016

    2016

  21. [21]

    Maas, A.Y

    A.L. Maas, A.Y. Hannun and A.Y. Ng,Rectifier nonlinearities improve neural network acoustic models, inProc. ICML Workshop on Deep Learning for Audio, Speech and Language Processing, 2013

    2013

  22. [22]

    Lindeberg,Scale-space for discrete signals,IEEE Transactions on Pattern Analysis and Machine Intelligence12(1990) 234

    T.P. Lindeberg,Scale-space for discrete signals,IEEE Transactions on Pattern Analysis and Machine Intelligence12(1990) 234

    1990

  23. [23]

    Savitzky and M.J.E

    A. Savitzky and M.J.E. Golay,Smoothing and Differentiation of Data by Simplified Least Squares Procedures.,Anal. Chem.36(2002) 1627

    2002

  24. [24]

    Schmid, D

    M. Schmid, D. Rath and U. Diebold,Why and how savitzky–golay filters should be replaced, ACS Measurement Science Au2(2022) 185

    2022

  25. [25]

    Butterworth,On the Theory of Filter Amplifiers,Experimental Wireless & the Wireless Engineer7(1930) 536

    S. Butterworth,On the Theory of Filter Amplifiers,Experimental Wireless & the Wireless Engineer7(1930) 536

    1930

  26. [26]

    Mittermayr, S

    C. Mittermayr, S. Nikolov, H. Hutter and M. Grasserbauer,Wavelet denoising of gaussian peaks: A comparative study,Chemometrics and Intelligent Laboratory Systems34(1996) 187

    1996

  27. [27]

    Fracture Processes Observed with A Cryogenic Detector

    J. ˚Astr¨ om et al.,Fracture Processes Observed with A Cryogenic Detector,Phys. Lett. A356 (2006) 262 [physics/0504151]. – 21 –

    work page internal anchor Pith review Pith/arXiv arXiv 2006