Liquid xenon scintillation measurements and pulse shape discrimination in the LUX dark matter detector

Weakly interacting massive particles (WIMPs) are a leading candidate for dark matter and are expected to produce nuclear recoil (NR) events within liquid xenon time-projection chambers. We present a measurement of the scintillation timing characteristics of liquid xenon in the LUX dark matter detector and develop a pulse shape discriminant to be used for particle identification. To accurately measure the timing characteristics, we develop a template-fitting method to reconstruct the detection times of photons. Analyzing calibration data collected during the 2013–2016 LUX WIMP search, we provide a new measurement of the singlet-to-triplet scintillation ratio for electron recoils (ER) below 46 keV, and we make, to our knowledge, a first-ever measurement of the NR singlet-to-triplet ratio at recoil energies below 74 keV. We exploit the difference of the photon time spectra for NR and ER events by using a prompt fraction discrimination parameter, which is optimized using calibration data to have the least number of ER events that occur in a 50% NR acceptance region. We then demonstrate how this discriminant can be used in conjunction with the charge-to-light discrimination to possibly improve the signal-to-noise ratio for nuclear recoils.

Figure 1

Ratio of corrected to raw S1 pulse area in detected photons (phd) for both tritium (blue) and DD neutron (red) calibration data. The corrected photon count is computed using Eq. (3.2) with $C=1.04$. The black dashed line has a slope $m=1$ and is shown for comparison.

Figure 2

Example simulated waveform (black curve). Three photons arriving at $t=-0.75$, 1.60, and 6.50 samples ($1\mathrm{sample}=10\mathrm{ns}$) are used to generate the simulated signal. The photon timing algorithm described in Sec. 3 reconstructs three photons arriving at $t=-0.76$, 1.51, and 6.45 samples (grey).

Figure 3

(a) A scintillation pulse from a ${\mathrm{CH}}_3\mathrm{T}$ calibration event, separated by the PMT channel. (b) The reconstructed peak times for the photons in the pulse are shown. The times are weighted by the fitted area of the template to correct for unresolved pileup of multiple photons arriving in a single channel.

Figure 4

An example of a photon detection time distribution for a PMT located on the bottom array, facing the strobed LED on the top array. The photon detection times shown are measured relative to the LED trigger and are histogrammed in 2 ns bins. The points between the data are interpolated with a linear spline, which is shown with the dotted red trace. The dashed blue line shows 10% of the peak height and is compared to the spline fit to obtain the offset for that channel, shown by the solid green trace.

Figure 5

Simulated time distribution of photons detected in the LUX detector, calculated using the ray-tracing capabilities in the LUXSim Geant4 simulation package. The photon sources are 10 cm thick, 20 cm diameter cylinders, centered at different depths below the liquid surface: 42 cm (grey curve), 32 cm (green curve), 22 cm (orange curve), and 12 cm (blue curve. These slices in depth correspond to the drift time bins used in the WS2014-16 dark matter search [1]. Solid lines show the second term in Eq. (4.2), fitted to these simulations. The excess in the first bin from direct-path photon arrivals is parametrized by a constant $A$ and is not reflected in the curves shown. Values of the parameters for each position are given in Table 1.

Figure 6

Average photon detection time spectra for events with pulse area between 40 and 50 phd ($\sim 11–13\mathrm{ke}{\mathrm{V}}_{ee}$). The probability distribution of NR events, from DD neutron data, is shown by the red diamonds, and the MC simulated NR time spectrum is shown in red. The measured distributions for ER events from tritium and ${}^{14}\mathrm{C}$ are shown in green and blue, respectively. The ER time spectrum generated from MC simulations is shown in blue. The time at which the pulse area reaches 5% of the total, denoted T05 in the text, is used as $t=0$. The vertical uncertainties in each bin are calculated from Poisson statistics, while the horizontal error bars represent bin width.

Figure 7

Singlet/triplet ratio ($C_1{\tau }_1/C_3{\tau }_3$) measured for nuclear recoils (top) and electron recoils (bottom) using LUX calibration data. Only statistical uncertainties of the data are shown. Calibration sources are DD neutrons (red), tritium (blue), and ${}^{14}\mathrm{C}$ (green). Measurements in different energy bins are shown by the square points, and the best fit constant model by the solid line. The shaded region indicates the statistical uncertainty on the constant model. The shaded gray bar indicated the systematic uncertainty of the constant model. A power law is also fit to the data and is presented by the dashed line. We also show measurements of the ER singlet/triplet ratio at zero field from Ref. [11] (cyan diamonds), and a measurement using a ${}^{207}\mathrm{Bi}$ internal conversion source at $4\mathrm{kV}/\mathrm{cm}$ from Ref. [8] (purple diamond). In Ref. [11], the singlet fraction (denoted $F_1$) is given rather than the singlet/triplet ratio. For direct comparison to this work we make the conversion $(C_1{\tau }_1)/(C_3{\tau }_3)=F_1/(1+F_1)$.

Figure 8

Normalized prompt fraction distributions for NR (red) and ER (blue) events with raw S1 pulse areas from 40 to 50 phd. Solid red and blue traces are the simulated NR and ER prompt fraction distributions for this pulse area. The vertical error bars indicate the Poisson uncertainty, while the horizontal error bars represent bin width. The solid black line indicates the median of the NR distribution, and the prompt fraction region above the DD median is considered to be the NR acceptance region. The $26\pm 2\%$ of the ER distribution was found to lie within this region.

Figure 9

PF distribution for NR events (left), and ER events (right). The red and blue dots indicate the NR and ER bands, respectively (median and median $\pm 34$ percentile), calculated from data. The traces indicate the bands calculated from simulated data using the constant value model fitted to the singlet/triplet ratio (Fig. 7). The power law model produces bands similar to the constant value model in this energy region and is omitted for clarity. The solid traces indicate the medians, and the dashed lines indicate the median $\pm 34$ percentile. We define the region above the NR median as the NR acceptance region.

Figure 10

Fraction of ER events that leak into the NR region. The points show the leakage fraction obtained from the data while the solid trace indicates the leakage fraction calculated from simulation. The NR median is used to define the $\sim 50\%$ NR acceptance region. The horizontal error bars indicate the bin widths. The counting errors from Poisson statistics and the errors in the NR median are added in quadrature to calculate the error in the leakage fraction. For both data and simulations, the same PFD windows are used.

Figure 11

(Top) Two parameter event discrimination using PFD and ${\log }_{10}(S2/S1)$. The distribution contains NR (red) and ER (blue) events with S1 pulse areas between 40 and 50 phd. The yellow ellipse indicates the 90% NR region, and the solid black line indicates the discriminator. These events are projected onto the dashed black axis perpendicular to the NR median. (Bottom) The distance of these events from the NR median is presented. The region to the left of the solid black line is defined as the NR acceptance region and is chosen to preserve 50% of the ellipse’s NR region while minimizing the number of ER events in this region.