First results from the DEAP-3600 dark matter search with argon at SNOLAB

This paper reports the first results of a direct dark matter search with the DEAP-3600 single-phase liquid argon (LAr) detector. The experiment was performed 2 km underground at SNOLAB (Sudbury, Canada) utilizing a large target mass, with the LAr target contained in a spherical acrylic vessel of 3600 kg capacity. The LAr is viewed by an array of PMTs, which would register scintillation light produced by rare nuclear recoil signals induced by dark matter particle scattering. An analysis of 4.44 live days (fiducial exposure of 9.87 tonne-days) of data taken with the nearly full detector during the initial filling phase demonstrates the detector performance and the best electronic recoil rejection using pulse-shape discrimination in argon, with leakage $<1.2\times 10^{-7}$ (90% C.L.) between 16 and 33 keV$_{ee}$. No candidate signal events are observed, which results in the leading limit on WIMP-nucleon spin-independent cross section on argon, $<1.2\times 10^{-44}$ cm$^2$ for a 100 GeV/c$^2$ WIMP mass (90% C.L.).

It is well established from astronomical observations that dark matter (DM) constitutes most of the matter in the Universe [1], accounting for 26.8% of the energy density, compared to 4.9% for ordinary matter.Weakly Interacting Massive Particles (WIMPs) are one of the leading dark matter candidates, predicted by a number of theoretical extensions of the Standard Model.Direct detection of WIMPs from the galactic halo is possible via elastic scattering interactions, which produce nuclear recoils of a few tens of keV.Such detection requires large target mass in ultralow background detectors located deep underground to suppress background associated with cosmic rays.This paper reports on the analysis of commissioning data from DEAP-3600, the first liquid argon (LAr) based dark matter detector exceeding a 1 tonne target mass.DEAP-3600 uses single-phase technology, which registers only the primary scintillation light from the target medium.We determine the best limit on WIMP-nucleon cross section measured with argon, in the high WIMP mass regime second only to xenon based searches, and demonstrate the best low-threshold electronic recoil re-jection using pulse-shape discrimination (PSD) in argon.The substantial difference in LAr scintillation timing between nuclear recoils (NR) and electronic recoils (ER) allows sufficient rejection of the dominant β/γ backgrounds [2,3] using only the primary scintillation light.The DEAP-3600 single phase LAr detector design offers excellent scalability to ktonne-scale target masses [4,5].
The detector is comprised of an atmospheric LAr target contained in a transparent acrylic vessel (AV) cryostat capable of storing 3600 kg of argon.The AV is viewed by 255 Hamamatsu R5912-HQE photomultiplier tubes (PMTs) operated near room temperature to detect scintillation light generated in the target medium.The PMTs are coupled to the AV by 50 cm-long acrylic light guides (LGs) that also provide neutron shielding and reduce the heat load on the AV.The inner AV surface was coated in-situ with a thin layer of wavelength shifter, 1,1,4,4-tetraphenyl-1,3-butadiene (TPB) to convert 128 nm Ar scintillation light into visible blue light, which is efficiently transmitted through acrylic.The AV neck is wrapped with optical fibers read out by 4 PMTs from both ends, to veto light emission in the AV neck region.The detector is housed in a stainless steel spherical shell, which is immersed in an 8 m diameter ultrapure water tank instrumented with 48 PMTs.This tank serves as a radiation shield and Cherenkov veto for cosmogenic muons.All detector materials were selected to achieve the background target of <0.6 events in a 3 tonne-year exposure [5].To avoid 222 Rn/ 210 Pb contamination of the bulk acrylic and TPB, the inner 0.5 mm surface layer of the inner AV was removed in-situ after construction.The Rn exposure was then strictly limited, with the AV and the access glovebox purged with Rn-scrubbed N 2 , evacuated and baked before filling.Argon was delivered as cryogenic liquid, stored underground, purified as gas with SAES Megatorr PS5 to sub-ppb impurity levels (for O 2 , H 2 O, CO, CO 2 , H 2 , N 2 and hydrocarbons), scrubbed of Rn [6] and liquified in the AV.
PMT signals are decoupled from the high voltage by a set of custom analog signal-conditioning boards and split into three outputs: high-and low-gain, and a twelve channel analog sum used to generate trigger decisions.Outputs are digitized with CAEN V1720 digitizers and handled by the MIDAS DAQ system, described in [7].
The PMT charge response functions are calibrated daily with a system of optical fibres injecting 435 nm light from a pulsed LED source onto 20 PMTs uniformly spaced around the detector (and 2 additional fibres in the detector neck), allowing study of a range of occupancies in all PMTs with a combination of reflected and direct illumination.A detailed model of the charge response function for each PMT gives the mean single photoelectron (SPE) charges with uncertainty less than 3% [5,8].A full PMT signal simulation is implemented in a detailed Monte Carlo model of the detector and electronics, using the Geant4-based RAT [9].For accurate simulation of PSD, the PMT simulation uses in-situ measured time vs. charge distributions from calibration data for noise sources, including late, double, and after-pulsing (AP) for each PMT [5,8,10].
The charge of each identified pulse is divided by the PMT-specific mean SPE charge to extract the number of photoelectrons (PEs).F prompt is then defined for each event as the ratio of prompt to total charge, where Q is the pulse charge in PE and t is the pulse time with respect to the event time.The relative timing of each PMT channel is calibrated with a fast laser source; the resulting overall time resolution is 1.0 ns.F prompt is a powerful PSD variable because it is sensitive to the ratio of excited singlet to triplet states in LAr, with lifetimes of 6 and 1300 ns [11], respectively.The detector trigger was designed to accept all low energy events above threshold, all high-F prompt NR events and to cope with approx. 1 Bq/kg 39 Ar activity of LAr [12], by prescaling the resulting low-F prompt ER events.The signal from the inner PMT analog sum is continuously integrated in windows 177 ns and 3100 ns wide, from which the prompt energy (E trigger ) and ratio of prompt and wide energies (F trigger ) are calculated.Triggers with NR-like E trigger >40 PE are digitized for all events, while only 1% of 39 Ar-decay-like events are digitized; summary information is recorded for all events.While for NR-like events in this analysis the trigger is highly efficient, (100 +0.0 −0.1 )%, as determined by running in a very low threshold mode, for ER-like events below 120 PE the efficiency decreases because of their lower prompt charge.This result was validated with a dedicated random trigger run, processed offline with a simulated physics trigger algorithm.
Stability of the LAr triplet lifetime, τ 3 , was verified with a fit accounting for PMT AP, dark noise, and TPB fluorescence [13].From this fit τ 3 =1399±20 (PMT syst.)±8(fit syst.)±6(TPB syst.)±7(APsyst.)ns, where systematic uncertainties are evaluated by performing the fit separately on individual PMTs, varying the fit range, and varying the TPB fluorescence decay time and times of the AP distributions within uncertainties.This result is consistent with the literature value of 1300±60 ns [11] and is stable throughout the analyzed dataset.Stability over a longer period is shown in Fig. S1.
The dominant source of scintillation events is 39 Ar β decay, resulting in low-F prompt ER events.In order to define an F prompt cut constraining the leakage of 39 Ar events into the NR band, the F prompt distribution of ER and its energy dependence were fitted with an 11parameter empirical model of F prompt vs. PE, based on a smeared Gamma distribution, The 2dimensional fit of the model to the data from 80 to 260 PE has a χ 2 ndf of 5581/(5236-11).As an example, a 1-dimensional slice from the model and the data at 80 PE is shown in Fig. 1(a).The PSD leakage measured in the 120 to 240 PE window with a 90% NR acceptance is shown in Fig. 1(b).The extrapolated leakage is approximately 10 times lower than projected in the DEAP-3600 design [3].As further reduction in the PSD leakage is expected from an analysis relying on SPE counting [14], the original goal of a 120 PE analysis threshold in 3 years livetime from PSD will likely be surpassed.The energy calibration uses internal detector backgrounds and external radioactive sources.The internal calibration uses β's from 39 Ar decay, with an endpoint of 565 keV.These are uniformly distributed in the detector, as WIMP-induced NR's would be.The external calibration uses a 22 Na source, which produces 1.27 MeV γ's and a 30-50 keV photo-absorption feature, both near the AV surface, similar to surface backgrounds.The simulated spectra of 39 Ar and 22 Na are fit to the data to find the energy response function relating T eff [keV ee ] (electron-equivalent energy) to detected PE, The internal and external sources are fit separately, be-cause of their different spatial distributions.The error bars on the fit parameters are scaled by χ 2 ndf to account for the systematic uncertainty, and the best fit values are combined in a weighted average to produce the final response function, which is shown in Fig. 2 together with the 39 Ar data, which spans from below to above the energy window for this analysis (see Fig. S2 for the 22 Na fit).The energy response function fits for 39 Ar and 22 Na agree within errors, however an additional systematic uncertainty, 0.5 keV ee is assessed to account for the non-zero offset term.As a cross-check, the response function is extrapolated to compare with high energy γ lines, particularly 1461 keV from 40 K and 2614 keV from 208 Tl in the detector materials, showing good agreement until 40 K and then diverging in the regime where PMT saturation and non-linearities in the DAQ become more significant.The light yield (LY) at 80 PE is 7.36 +0.61 −0.52 (fit syst.)±0.22(SPEsyst.)PE/keV ee , where systematic uncertainties from the fitting procedure and the SPE charge calibration are included.A Gaussian energy resolution function is used to smear the spectra in the fit, with variance σ 2 = p 1 • PE.Extrapolated resolution at 80 PE from best fit values for 39 Ar and 22 Na is 13±1% and 16±1%, respectively.The difference is attributed to a larger spread in light collection efficiency for events near the AV surface as measured with the 22 Na source.A lower bound on the energy resolution at 80 PE is 12% (p 1 = 1.185), determined from Poisson counting statistics widened by the measured insitu SPE charge resolution.Since at low WIMP masses the broader the resolution the stronger the limit, because of the steeply falling WIMP-induced NR spectrum, using this lower bound is conservative.
The nuclear recoil acceptance of the F prompt cut is determined from a simulation of 40 Ar recoils distributed uniformly in LAr.The simulation assumes the quenching factor and triplet/singlet ratio energy dependence as measured by SCENE [15] at zero electric field and applies the full response of the detection and analysis chain.PMT AP dominates the effect of the detector response on the mean F prompt with the average AP probability of (7.6±1.9)%[5].Comparison of external neutron AmBe source data with a simplified detector simulation in Fig. 3(b) shows qualitative agreement and serves as a validation of the model.The simulation includes neutrons and 4.4 MeV γ's from the AmBe source and considers scattering-or capture induced γ's only for neutrons that entered the LAr.AmBe data is not used directly to model the WIMP-induced NR acceptance as a significant fraction (59% in the 120-240 PE window) of AmBe events contain multiple elastic neutron scatters.
The region-of-interest (ROI) in this analysis, as shown in Fig. 3(a), was defined by allowing for an expectation of 0.2 leakage events from the 39 Ar band, determined with the PSD model, while maintaining the NR acceptance of >5% at the lowest energies.The smaller number of 39 Ar events in the short exposure and the low F prompt leakage allowed us to set the energy threshold at 80 PE (11 keV ee ), lower than the nominal 120 PE threshold originally projected [3].Above 150 PE the lower limit on F prompt is chosen to remove 5% of NR events in each bin.The ROI also has a maximum F prompt chosen to remove 1% of NR events in each 1 PE bin.The maximum energy was set to 240 PE to reduce possible backgrounds from α activity of the AV surface [16].
The first LAr fill of the detector took approx.100 days between May and mid-August 2016.For the majority of this time, Ar gas was introduced into the detector from the purification system for cooling.In the final phase of the fill, shortly after the dataset discussed in this work was taken, a leak in the detector neck contaminated LAr with clean Rn-scrubbed N 2 .The detector was subsequently emptied and refilled and has been taking data since Nov. 1, 2016, with a slightly lower liquid level.
In this work, we focus on the period Aug. 5 to Aug. 15 (9.09 days), when no Ar had been introduced into the detector.Because of the much higher scintillation rate, photon yield and refractive index in liquid than in gas, there is a very sharp drop in rate between PMTs facing the liquid and PMTs facing the vapour space, which allows determination of the fill level, 590±50 mm above the AV centre, and the full LAr mass, 3322±110 kg.
Data were analyzed from runs where (1) the difference between the maximum and minimum AV pressures in a run corresponded to <1 cm change in the liquid level and (2) no more than one PMT read <50% of its average charge, determined from approx.5 min samples.During this dataset one PMT was turned off (and has since returned to operation).In all cases, pressure excursions were correlated with periods of the cryocoolers operating at reduced power.34% of the data are removed by failing both criteria and additional 11% by failing criterion 2 alone.The total run time after run selection corresponds to 4.72 d, out of which 0.28 days (5.9%) is deadtime from 17.5 µs following each trigger.
Acceptance for NR events, shown in Fig. 4, is determined using a combination of 39 Ar events, which are uniformly distributed in the LAr volume, and simulation of F prompt for NR events.The sample of 39 Ar single-recoils is obtained first by applying low level cuts to remove events (1) from DAQ calibration, (2) highly asymmetric (with more than 40% of charge in a single PMT) e.g.Cherenkov events in LGs and PMTs or (3) from pile-up.The approach of measuring acceptance for NR events using ER events is used since none of the acceptance cut variables depend on the pulse time information, only F prompt does, which is handled separately.The F prompt simulation for NR's is validated by comparison with the AmBe calibration data.
Quality cuts are applied to 39 Ar events within the energy window in order to determine the ER acceptance, as shown in Table I.The fiducial acceptance is determined with respect to the events remaining after the quality cuts in order to factor out light coming from outside the LAr volume.The event time cut requires that the scintillation peak is positioned early in the waveform, which ensures reliable evaluation of F prompt .Cuts on the fraction of charge in the brightest PMT and on the neck veto remove high charge afterpulses triggering the detector as well as events caused by light emission (e.g.Cherenkov) in the AV neck acrylic.We have identified a class of background events originating in the neck region and are characterizing it for future larger-exposure searches.
Fiducialization in this analysis employs low-level PE ratio variables.These are the fraction of scintillationinduced PE [14] with AP correction [10] in the PMT in a given event detecting the most light, and the fraction of charge in the top 2 rows of PMTs in the detector.These are strongly correlated with the radial and vertical event positions, respectively.The fiducial mass, 2223±74 kg, is determined from the full LAr mass and acceptance of the fiducial cuts.The expected activity of 39 Ar contained in this mass is 2245±198 Bq [12], consistent with the fiducial rate observed in DEAP-3600, 2239±8 Hz.
Position reconstruction algorithms have been developed and tested on the detector data.However, in the analysis presented here, they are used only as a crosscheck.A maximum likelihood fitter relies on the full Monte Carlo of the detector including its optical properties, and minimizes the difference between the observed  pattern of PMT charges and the one expected based on a PDF constructed from simulation, under the assumption that the illumination of the detector is symmetric around the axis of the event position vector.Residual position bias is corrected for using the uniformly distributed population of 39 Ar β's.To study reconstruc-tion of events from the inner AV surface, as expected for α backgrounds, we apply the 39 Ar-derived calibration to 22 Na events, which are strongly peaked near the surface.Fig. S3 shows that qualitative agreement results from this procedure.We plan to use reconstructed positions to further reduce backgrounds in longer exposure runs.
The main background sources are α activity, neutrons, leakage from 39 Ar and other ER interactions.
As shown in Table II, 222 Rn, 218 Po and 214 Po α decays can be identified in the LAr bulk in well-defined high energy peaks with an activity of 1.8×10 −1 µBq/kg based on time delayed coincidence with α-α ( 222 Rn-218 Po and 220 Rn-216 Po) or β-α ( 214 Bi-214 Po) tags.The activity of 214 Po in the bulk is consistent with the earlier part of the chain, which indicates that it is mostly mixed within the LAr volume, see Table II and Fig. S4.Out-of-equilibrium 210 Po α decays can be identified with degraded energies characteristic of α's coming from below the 3 µm thick TPB layer on the surface of the acrylic.Activity of 210 Po is determined with a fit of simulated spectra to the data (see Fig. S5), assuming contamination either on the acrylic surface or distributed uniformly in an 80 µm deep acrylic surface layer.The result for bulk contamination assumes no additional backgrounds in the fit range and is considered an upper limit.
The dominant source of neutron events is expected to be from (α, n) reactions and spontaneous fission in the PMTs.The PMT borosilicate glass contribution can be constrained with in-situ measurements of the 2614 keV and 1764 keV γ-rays from the 232 Th and 238 U decay chains, respectively.The 238 U and 232 Th decay chain activities seen in-situ agree within a factor of two with a simulation based on the screening results.Events from neutron backgrounds in LAr can be measured in-situ by FIG.4: The acceptance with systematic error bands in 80-240 PE window for the trigger, event quality cuts, Fprompt cut, fiducial cuts and all cuts combined.For uncertainties on the acceptance of quality and fiducial cuts see also Table I.
Uncertainties on trigger acceptance measurement and Fprompt cut acceptance are discussed in the text.
The systematic uncertainties considered in the WIMP cross-section limit calculation include uncertainty on the NR energy response, exposure (from livetime and total LAr mass), and quality and fiducial cut acceptance (see Fig. 4).The uncertainty on the NR acceptance of the F prompt cut is determined by varying the simulation inputs: triplet/singlet ratio (within errors propagated from the SCENE [15] measurement of f 90 ), the triplet lifetime uncertainty (the difference between literature value [11] and this work), and the AP probability.The dominant uncertainty in the final exclusion curve comes from the uncertainty on the NR energy response.This effect is dominated by uncertainties in Eq. (3); however, there is also some uncertainty on the NR quenching factor, i.e. the reduction in NR scintillation yield relative to ER.
([keV r ]= L eff •[keV ee ], when referring to energies of NR, keV r , the unit of the full energy of the recoil, can be used.)We used measurements from SCENE, which reports two different quenching factors: L eff, 83m Kr , which is the ratio of LY measurement at various NR energies to the LY measured by a 83m Kr ER calibration, and L, which is the Lindhard-Birks quenching factor describing the suppression of quanta (scintillation photons or extracted electrons) at different NR energies.The difference between these two values at a given NR energy is due to non-unitary recombination at null field.We adjusted the Lindhard-Birks quenching factors fit to L to account for the relative recombination rates of NR and 83m Kr ER at null field, according to the NEST model [20], fitting Thomas-Imel and Doke-Birks recombination parameters to SCENE's L eff, 83m Kr values.Uncertainties on this fit were inflated to account for differences between the SCENE and DEAP-3600 detectors and the different recombination rates of the 83m Kr ER and the 22 Na low energy feature that we used for our energy calibration.These factors, along with uncertainty in SCENE's reported value of Birks' constant and the difference between L and L eff, 83m Kr were factored into the uncertainty of our quenching factor, to account for uncertainties in the recombination probabilities.
No events are observed in the ROI, see Fig. 5. Figure 6 shows the resulting upper limit on the spin-independent WIMP-nucleon scattering cross-section as a function of WIMP mass, based on the standard DM halo model [21].A 90% C.L. upper limit is derived after factoring in the Poisson fluctuation in the number of expected signal events, with the Highland-Cousins [22] method employed to account for the systematic uncertainties.For a more conservative limit, the backgrounds from 39 Ar leakage were not taken into account.DEAP-3600 has achieved stable operation at FIG. 6: Spin-independent WIMP-nucleon cross-section 90% C.L. exclusion from 4.44 live days of DEAP-3600 data.Also shown: the current results from XENON1T [23], LUX [24], PandaX-II [25], DarkSide-50 [26], CDMS-II [28], PICO-60 [27], and the full sensitivity for XENON1T and DEAP-3600, assuming no observed events in a fiducial volume of 1000 kg in three years of running with a threshold of 15 keVee.
7.36 PE/keV ee light yield without recirculation, and demonstrated better-than-expected PSD (permitting a 39 keV r energy threshold), with promising α and neutron background levels.Analysis of the first 4.44 d of data reported here results in the best limit at low energies on discrimination of β-decay backgrounds using PSD in LAr at 90% NR acceptance, with measured leakage probability of <1.2×10 −7 (90% C.L.) in the energy window 16-33 keV ee (55-111 keV r ).This measurement has lower threshold than DEAP-1 [3] and higher statistics than DarkSide-50 [26].After NR selection cuts no events are observed, resulting in the best spin-independent WIMP-nucleon cross section limit measured in LAr [26] of <1.2×10 −44 cm 2 for a 100 GeV/c 2 WIMP (90% C.L.).Data collection has been ongoing since Nov. 2016 and forms the basis for a more sensitive DM search currently in progress.FIG.S3: Reconstructed radii of (a) 39 Ar uniformly distributed in the detector and (b) tagged events from an external 22 Na calibration source after correcting for radial bias in data (black), 22 Na Monte Carlo in (red), with added distribution of random coincidences of the source tag with 39 Ar decays (blue) and the sum of both 39 Ar and 22 Na distributions (magenta).Residuals are displayed in the bottom row.

FIG. 1 :
FIG.1:(a) Projection of the Fprompt distribution at 80 PE (full 2-dimensional distribution is shown in Fig.5) is shown together with the effective model.Fit is performed above the red dashed line, indicating the Fprompt value below which the trigger efficiency is <100%.The brown and orange lines correspond to 90% and 50% NR acceptance (n.r.a.).Each PE bin in the fit range contributes approximately equally to the overall χ 2 value.(b) Data and model for the 120-240 PE range with 1.87972×10 7 events, represented as leakage probability above a given Fprompt value.A conservative projection from DEAP-1[3] is also shown with its own NR acceptance lines (all three dashed).

FIG. 3 :
FIG.3:(a) AmBe source data after cuts, with the region-of-interest for WIMP search shown with a black box.(b) The Fprompt distribution for 140<PE<240 in AmBe calibration data (black), compared to summed simulated contributions for AmBe neutrons, and 4.4 MeV γ's and the 39 Ar Fprompt model normalized to the peak of the distribution.Also plotted is the simulation of single scatter nuclear recoils with flat energy spectrum (see legend).Error bars shown on the simulated distributions are statistical, not systematic.

FIG. 5 :
FIG.5: Fprompt vs the number of photoelectrons and energy in keVee for events passing cuts, with the WIMP search ROI shown in red.
FIG. S4: Peaks from tagged alphas in 222 Rn and 220 Rn chains in the detector data overlayed with Monte Carlo PDFs.

TABLE I :
Run selection criteria and cuts with their effects on livetime, integrated acceptance, the fiducial fraction, and the number of events left in the ROI.The acceptance is calculated individually for each run and then weighted by livetime to provide an overall acceptance with the uncertainties taken as maximum and minimum variations about this weighted mean from each run.See text for details about the fiducial fraction determination.The total number of triggers before any cuts was 1.38×10 9 , out of which 6.47×10 7 in 80-240 PE window.

TABLE II :
[19]ary of α activities, see text.These can be compared with results reported by other experiments, approx.: 66 µHz/kg of 222 Rn and 10 µHz/kg of 220 Rn in LUX[17], 6.57 µBq/kg of 222 Rn and 0.41 µBq/kg of 220 Rn in PandaX-II[18], and 10 µBq/kg of 222 Rn in XENON1T[19].searchingfor NR's followed by capture γ's.The efficiency of this technique was calibrated using neutrons from an AmBe source deployed near the PMTs.No neutron candidates were seen in 4.44 d (80-10000 PE window, no fiducial cuts), which is consistent with the expectation based on assays.
Supplemental Materials: First results from the DEAP-3600 dark matter search with argon at SNOLAB The long time constant from argon scintillation measured during the detector fill, determined with a simple 'exponential + linear' fit to the summed waveforms from 500 ns to 3000 ns.Such a fit is in general sensitive not only to the LAr triplet lifetime (1.3 µs) but also to other effects including PMT AP and TPB fluorescence; hence, it overestimates the triplet lifetime.Uncertainties from the fit are within the marker size.LAr was not recirculated/repurified throughout the entire period.The grey shaded area represents the dataset used for dark matter search presented here.The fit time constant is stable within that period to <1%.