Search for dark matter with a 231-day exposure of liquid argon using DEAP-3600 at SNOLAB

DEAP-3600 is a single-phase liquid argon (LAr) direct-detection dark matter experiment, operating 2 km underground at SNOLAB (Sudbury, Canada). The detector consists of 3279 kg of LAr contained in a spherical acrylic vessel. This paper reports on the analysis of a $758\mathrm{tonne}·\mathrm{day}$ exposure taken over a period of 231 live-days during the first year of operation. No candidate signal events are observed in the WIMP-search region of interest, which results in the leading limit on the WIMP-nucleon spin-independent cross section on a LAr target of $3.9\times {10}^{-45}{\mathrm{cm}}^2$ ($1.5\times {10}^{-44}{\mathrm{cm}}^2$) for a $100\mathrm{GeV}/{\mathrm{c}}^2$ ($1\mathrm{TeV}/{\mathrm{c}}^2$) WIMP mass at 90% C.L. In addition to a detailed background model, this analysis demonstrates the best pulse-shape discrimination in LAr at threshold, employs a Bayesian photoelectron-counting technique to improve the energy resolution and discrimination efficiency, and utilizes two position reconstruction algorithms based on the charge and photon detection time distributions observed in each photomultiplier tube.

Figure 1

Cross section of the DEAP-3600 detector components located inside the water tank (not shown). Inside the steel shell are inward-looking PMTs, light guides, filler blocks, and the AV, which holds the liquid argon target and the gaseous argon layer. Located on the outer surface of the steel shell are muon veto PMTs. Above this, a steel neck contains the neck of the AV, acrylic flow guides and the cooling coil. The neck is coupled to a central support assembly on which the glovebox is located. Shown also is the neck veto fiber system (green).

Figure 2

A block diagram of the DEAP-3600 data acquisition system, adapted from [10]. Shown are the PMTs, the digitizer and trigger module (DTM), the signal conditioning boards (SCBs), the event builder, the light injection system, the test pulser systems, the fast high-gain channel digitizers (V1720s), and the slow low-gain channel digitizers (V1740s).

Figure 3

Average ${}^{39}\mathrm{Ar}$ pulse shape before correction of instrumental effects (black) shown together with a model fit (red). The fit accounts for the following effects, which are shown individually: LAr singlet, triplet, and intermediate [23] light emission (green dashed), TPB prompt and delayed light emission [24] (blue dash-dotted), afterpulsing following all the previous components (pink dotted), and stray light (grey filled), which accounts for dark noise and the delayed TPB emission from previous events. The pulse shape made from pulses that use the pulse-by-pulse AP removal algorithm (see text) is also shown (grey solid).

Figure 4

The ${}^{39}\mathrm{Ar}$ model (blue line) fit to data (black). Included in the fit is the expected background contribution from $\gamma$-rays and ${}^{39}\mathrm{Ar}$ pileup events (green).

Figure 5

The energy response function (red), showing the number of detected PE for an event depositing energy $E$ in the LAr. The uncertainties of the response function are also shown (yellow band). The response function agrees with the number of PE detected from known monoenergetic sources of $\gamma$-rays from the detector materials.

Figure 6

Comparison of the detected PE distribution for high-${\mathrm{F}}_{\text{prompt}}$ NR-like events from an AmBe neutron source simulation (pink) and data (black). The peak at low PE is due to the ${\mathrm{F}}_{\text{prompt}}$ cut that removed ERs; due to the high rate of ERs in coincidence with neutron-induced NRs in the calibration data, particularly strong cuts are needed to obtain a clean NR spectrum. Uncertainties shown are all statistical, using nominal values for the quenching and detector optics models.

Figure 7

${\mathrm{F}}_{\text{prompt}}$ distribution for ERs from standard physics data in the lowest $1{\mathrm{keV}}_{\mathrm{ee}}$ energy bin in the WIMP-search region of interest. The PSD model is fit to the data to the right of the vertical dashed line, where the trigger efficiency is approximately unity. Agreement between the best fit model and the data can be seen; below the ${\mathrm{F}}_{\text{prompt}}$ fit region, trigger efficiency corrections to the data show that the model agrees when extrapolated to lower values. Solid vertical lines show the ${\mathrm{F}}_{\text{prompt}}$ value in the corresponding PE bins above which 90% or 50% of NRs are expected to be found. In the bottom plot, black points show residuals over the range where the model is fit; blue points compare the extrapolated model directly to the observed data prior to correcting for the decreasing trigger efficiency, while pink points compare the extrapolated model to the data after making these corrections.

Figure 8

${\mathrm{F}}_{\text{prompt}}$ vs PE distribution for data taken with an AmBe neutron source near the equator of the stainless steel shell. The WIMP-search ROI is shown in red. Separation can be seen between the NR band (${\mathrm{F}}_{\text{prompt}}\sim 0.7$) and ER band (${\mathrm{F}}_{\text{prompt}}\sim 0.3$). Events in the prescaled region (ER band with $\mathrm{PE}\gtrsim 300$) are weighted with a correction factor of 100, corresponding to the prescale factor. Note that multiple neutron scatters and pileup with Cherenkov and ER signals bias the NR spectrum away from the expected distribution for pure single-scatter NRs and populate the region between both bands.

Figure 9

${\mathrm{F}}_{\text{prompt}}$ distribution in the 120–200 PE range of events from AmBe data (black) and simulations of single-scatter neutrons (red dashed). Also shown are simulated events from an AmBe source (pink), the ER PSD model (green) and their sum (blue).

Figure 10

Estimates from the PE-based (red) and time residual-based (blue) algorithms of the contained mass of LAr within a radius of the reconstructed position. The estimate is based on the total fraction of ${}^{39}\mathrm{Ar}$ ER signals in the 95–200 PE range reconstructing within a given radius. Values from both data (solid) or simulations (dashed) are shown. It is assumed that the true positions of the ${}^{39}\mathrm{Ar}$ nuclei are uniformly distributed throughout the LAr target. Shown also is the estimated contained mass calculated by considering only the geometric volume (black). The bottom inset shows the difference between data and simulation for each algorithm.

Figure 11

Position resolution evaluated using the data-driven pseudoevent method, as a function of the average number of PE in both pseudoevents and the reconstructed radius drawn from the same original event, as returned by the PE-based algorithm. The $z$-axis scale denotes the resolution, defined as the characteristic width of the distribution of distances between reconstructed pseudoevents drawn from the same original event.

Figure 12

The $z$-coordinate of the reconstructed position (along the axis parallel to the AV neck) as estimated by the time residual-based algorithm minus the $z$-coordinate estimated by the PE-based algorithm. Note that ${}^{39}\mathrm{Ar}$ data (solid blue) and a simulation of WIMP NRs (dashed green) show similar distributions, centered around the origin. A significant offset is seen for simulated $\alpha$-decays in the neck (dotted magenta).

Figure 13

Shown over the time period spanned by this data set is the long lifetime component of LAr scintillation (top), the detector light yield (middle), and mean ${\mathrm{F}}_{\text{prompt}}$ of ER signals generated by 2.61 MeV $\gamma$-rays in the LAr target throughout the run. For the latter two plots, the predicted value based on the long lifetime value at that time is also shown (dashed).

Figure 14

Probability of an ER being detected above a given ${\mathrm{F}}_{\text{prompt}}$ value in the lowest $1{\mathrm{keV}}_{\mathrm{ee}}$ bin in the WIMP-search region of interest. For comparison, vertical lines show the values above which 90% or 50% of nuclear recoils are expected to be found.

Figure 15

Distribution of mapped $\alpha$-particle energies from ${\mathrm{F}}_{\text{prompt}}$ and PE for short-lived $\alpha$-decays in LAr. The event selection shown has no cut on position reconstruction and so includes events from ${}^{214}\mathrm{Po}$ $\alpha$-decays at the TPB surface (purple).

Figure 16

PE distribution from data of ${}^{210}\mathrm{Po}$ $\alpha$-decays candidates (black) alongside different simulated ${}^{210}\mathrm{Po}$ samples located on the AV surface at the TPB/AV interface (purple), at the LAr/TPB interface and within the TPB layer (yellow), and from beneath the inner surface of the AV (blue). Shown also is the neighboring PE distribution from ${}^{222}\mathrm{Rn}$ $\alpha$-decays in LAr (green). Overlaid is the combined fit (red) of each of these spectra to the observed PE spectrum from data. The event selections shown reconstruct with a $z$-position below the LAr fill level. No cut on the reconstructed radius is applied.

Figure 17

Leakage probability for simulated $\alpha$-decays in the WIMP PE range vs the contained LAr mass as determined by events within a given reconstructed radius. The contained LAr mass corresponding to the fiducial cut at 630 mm is shown. The systematic uncertainty on the probability is shown also (yellow band).

Figure 18

Top and middle panels: Cross-sectional illustration of the FG components in the AV neck. Shown are the three FG surfaces and the piston ring (not coated in LAr for purposes of illustration). Bottom panel: Simulated relationship in reconstructed $z$ vs PE for $\alpha$-decays on the IFG-IS (green), IFG-OS (pink), and OFG-IS (purple).

Figure 19

Top: Distribution of reconstructed $z$ vs PE for events at high ${\mathrm{F}}_{\text{prompt}}$ in data, using the PE-based algorithm. Shown are trends from $\alpha$-decays in the neck between 300–5000 PE and a background sideband at higher PE values. The blue boxes denote the sample regions used by the template fit to compare with spectra from simulations. The red boxes are used to sample a background component. Bottom: 1D-projection of the observed PE spectra (black points) and expected distributions from the background model components, normalized to the best-fit rates.

Figure 20

Illustration of the WIMP ROI (black) along with the ER (blue), NR (green), and neck $\alpha$-decay (pink) bands that define the boundaries. Each band is drawn about the median of each class of events, with 25% of such signals above and 25% below included in the shaded regions.

Figure 21

WIMP acceptance as a function of PE, broken down by cut type.

Figure 22

Observed ${\mathrm{F}}_{\text{prompt}}$ vs PE distribution after all cuts. The region of interest is shown in red.

Figure 23

Observed spatial distribution for all events surviving all cuts other than the cut on the reconstructed radius. The color scale in the background shows the acceptance for ${}^{39}\mathrm{Ar}$ events measured as a function of position after all but the radial cut; green points represent events in the ROI after all background rejection cuts. The fill level and radial fiducial cuts are drawn as well.

Figure 24

The 90% confidence upper limit on the spin-independent WIMP-nucleon cross sections based on the analysis presented in this paper (blue), compared to other published limits, including our previous limit [6], SuperCDMS [46], DarkSide-50 [7], LUX [47], PANDAX-II [48], and XENON1T [5].