Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
  • Open Access

Improving sensitivity to low-mass dark matter in LUX using a novel electrode background mitigation technique

D. S. Akerib et al. (LUX Collaboration)
Phys. Rev. D 104, 012011 – Published 19 July 2021
PDFHTMLExport Citation
Abstract
Authors
Article Text
  • INTRODUCTION
  • DATA SELECTION AND ANALYSIS
  • MACHINE LEARNING TECHNIQUE FOR REMOVING…
  • RESULTS AND SENSITIVITY
  • ACKNOWLEDGMENTS
    • References

    Abstract

    This paper presents a novel technique for mitigating electrode backgrounds that limit the sensitivity of searches for low-mass dark matter (DM) using xenon time projection chambers. In the Large Underground Xenon (LUX) detector, signatures of low-mass DM interactions would be very low-energy (∼keV) scatters in the active target that ionize only a few xenon atoms and seldom produce detectable scintillation signals. In this regime, extra precaution is required to reject a complex set of low-energy electron backgrounds that have long been observed in this class of detector. Noticing backgrounds from the wire grid electrodes near the top and bottom of the active target are particularly pernicious, we develop a machine learning technique based on ionization pulse shape to identify and reject these events. We demonstrate the technique can improve Poisson limits on low-mass DM interactions by a factor of 1.7–3 with improvement depending heavily on the size of ionization signals. We use the technique on events in an effective 5 tonne·day exposure from LUX’s 2013 science operation to place strong limits on low-mass DM particles with masses in the range mχ∈0.15–10  GeV. This machine learning technique is expected to be useful for near-future experiments, such as LUX-ZEPLIN and XENONnT, which hope to perform low-mass DM searches with the stringent background control necessary to make a discovery.

    • Figure
    • Figure
    • Figure
    • Figure
    • Figure
    • Figure
    • Figure
    6 More
    • Received 29 November 2020
    • Accepted 6 May 2021

    DOI:https://doi.org/10.1103/PhysRevD.104.012011

    Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. Funded by SCOAP3.

    Published by the American Physical Society

    Physics Subject Headings (PhySH)

    1. Research Areas
    Dark matterParticle dark matter
    1. Physical Systems
    Weakly interacting massive particles
    1. Techniques
    Dark matter detectors
    Gravitation, Cosmology & AstrophysicsParticles & Fields

    Authors & Affiliations

    Click to Expand

    Article Text

    Click to Expand

    References

    Click to Expand
    Issue

    Vol. 104, Iss. 1 — 1 July 2021

    Reuse & Permissions
    Author publication services for translation and copyediting assistance advertisement

    Authorization Required

    Log In
    Other Options
    ×

    Images

    • Figure 1
      Figure 1

      A schematic illustration of the LUX detector.

      Reuse & Permissions
    • Figure 2
      Figure 2

      Trigger [27], single scatter, and S2-quality cut efficiencies, as well as their combined efficiency (including the 1.6% acceptance loss from the single photon S1 cut). The curves labeled “S2 detection” and “S1+S2 detection” encompass the liquid xenon microphysics of signal production and detector physics of signal collection as modeled with NEST v2.0.1 [26]. The latter tapers off more quickly at low energy due to LUXs requirement that S1s be composed of photon signals in two or more PMT channels. It is not applied in this analysis, but is shown to illustrate the extra low-mass dark matter sensitivity gained in this search.

      Reuse & Permissions
    • Figure 3
      Figure 3

      All WS2013 events in range 3.5<ne<50.5 containing both an S1 (passing the two PMT coincidence requirement) and S2 pulse, and passing all vetoes/quality cuts applied to S2-only events. Gate, bulk, and cathode events are defined by drift time cuts: td<7  μs, 7  μs<td<321  μs, and 321  μs<td. The event rate is vastly higher at the gate and cathode drift times suggesting electrode events are the dominant source of backgrounds.

      Reuse & Permissions
    • Figure 4
      Figure 4

      Panels (a)–(e) show typical S2 pulse shapes obtained from skew-Gaussian fits to LUX data. The top panels have symmetric shapes that are characteristic of bulk events near the top, middle, and bottom of the TPC (drift times ∼10, ∼150, and ∼300  μs). Panels (d) and (e) show asymmetry characteristic of gate S2s whose electrons travel through a range of electric fields. Panel (f) shows a typical electric field profile around a single gate wire (reproduced from [33]). Electric field profiles of cathode wires have a similarly wide range of magnitudes although most field lines, except those at the very top of the wire, point downward. The five red circles in panels (a)–(e) are parameters (t10, t25, t50, t75, and t90) that mark the times at which the pulse attains 10%, 25%, 50%, 75%, and 90% of its total area. These parameters were used as input to the machine learning algorithm along with maximum pulse height, the time at which the pulse attains its maximum height, and the times at which the rising and falling edges of the pulse reach 0  phd/sample. Note the bulk event profiles on the top panels can also occur for gate and cathode S2s originating on the top of a wire where field curvature is less dramatic.

      Reuse & Permissions
    • Figure 5
      Figure 5

      WS2013 gate, cathode, and tritium calibration data used to train a boosted decision tree to recognize electrode backgrounds. All events pass the S2-only quality cuts, except some gate events that fail the radial cut. This cut was removed to increase the number of gate training events. Before training, these data were reweighted to give identical spectra for the three sources, as well as a 1∶1 ratio of gate/cathode events.

      Reuse & Permissions
    • Figure 6
      Figure 6

      ROC curves for (a) bulk tritium vs cathode and (b) bulk tritium vs gate test data. The curves can be used to estimate a BDT discriminator threshold that maximizes sensitivity to dark matter signals in an extreme scenario where only gate or cathode backgrounds are present. Two additional curves are plotted to guide the eye: εs=εb and εs=εb, the thresholds that must be exceeded by the ROC curves for a BDT discriminator threshold to improve sensitivity in a Poisson or background subtracted DM analysis. Stars indicate the points of optimal εs/εb.

      Reuse & Permissions
    • Figure 7
      Figure 7

      Signal acceptance and background rejection capability of the boosted decision tree used to tag and remove gate- and cathodelike events. The cut is not applied below the software threshold.

      Reuse & Permissions
    • Figure 8
      Figure 8

      Half-width distributions of training data before and after applying a discriminator cut tuned to 60% signal efficiency.

      Reuse & Permissions
    • Figure 9
      Figure 9

      ROC curves of training+testing data for a BDT using all shape quantifying parameters compared to a BDT using only half-width. Points of maximum limit improvement are shown by stars and circles for background subtracted and Poisson cases, respectively.

      Reuse & Permissions
    • Figure 10
      Figure 10

      Predicted maximum improvement in exclusion limits for background subtracted (εs/εb) and Poisson (εs/εb) scenarios, calculated separately for extreme cases of only gate backgrounds or only cathode backgrounds. Points of maximum improvement correspond to the stars and circles from Fig. 9 for the BDT using all parameters. Since we do not know the energy spectrum of the background, distinct values are given for each S2 size bin; the final improvement would be a weighted average of the values shown.

      Reuse & Permissions
    • Figure 11
      Figure 11

      Sensitivity calculations employed NEST v2.0.1 NR and ER charge yield models (solid black curves), with a hard cutoff in yield below 0.3  keVnr and 0.186  keVee, respectively. Charge yield measurements for NRs are from [37, 38], while those for ERs are from [30, 39, 40, 41].

      Reuse & Permissions
    • Figure 12
      Figure 12

      DM search data from May through September 2013 before and after applying the BDT cut and hand scanning to remove events that originated on the electrodes. The BDT cut reduces the observed event rate by a factor of ∼4 while retaining approximately 60% signal efficiency independent of S2 size. DM spectra at the 90% confidence interval cross-section limit are overlaid for comparison.

      Reuse & Permissions
    • Figure 13
      Figure 13

      Upper limits on the spin-independent DM-nucleon cross section at 90% C.L. The result of the S2-only analysis with an NR signal model is shown in black, and the result of the S2-only analysis with a signal model based on the Migdal effect is shown in gray. Also shown are limits from DarkSide-50 [18] (S2-only, binomial fluctuation assumption), CDMSlite [48], CRESST-III [49], XENON100 S2-only [16], XENON1T S1+S2 [6], XENON1T S2-only [17] (NEST 2.0.1 yields), XENON1T S2-only with Migdal effect [47], and past LUX searches using S1+S2 events [5], including S1s with single photons using double photoelectron emission (DPE) [13], and the Migdal effect [11].

      Reuse & Permissions
    ×

    Sign up to receive regular email alerts from Physical Review D

    Reuse & Permissions

    It is not necessary to obtain permission to reuse this article or its components as it is available under the terms of the Creative Commons Attribution 4.0 International license. This license permits unrestricted use, distribution, and reproduction in any medium, provided attribution to the author(s) and the published article's title, journal citation, and DOI are maintained. Please note that some figures may have been included with permission from other third parties. It is your responsibility to obtain the proper permission from the rights holder directly for these figures.

    ×

    Log In

    Cancel
    ×

    Search

    Article Lookup

    Paste a citation or DOI
    Enter a citation
    ×