Search for Neutrinoless Double-Beta Decay with the Upgraded EXO-200 Detector
Results from a search for neutrinoless double-beta decay () of Xe are presented using the first year of data taken with the upgraded EXO-200 detector. Relative to previous searches by EXO-200, the energy resolution of the detector has been improved to =1.23%, the electric field in the drift region has been raised by 50%, and a system to suppress radon in the volume between the cryostat and lead shielding has been implemented. In addition, analysis techniques that improve topological discrimination between and background events have been developed. Incorporating these hardware and analysis improvements, the median 90% confidence level half-life sensitivity after combining with the full data set acquired before the upgrade has increased 2-fold to yr. No statistically significant evidence for is observed, leading to a lower limit on the half-life of yr at the 90% confidence level.
Neutrinoless double-beta decay (), in which a nucleus with mass number and charge undergoes the decay with the emission of no neutrinos Furry (1939), provides the most sensitive test of the Majorana nature of neutrinos Dell’Oro et al. (2016). While the corresponding two-neutrino double-beta decay () has been observed for several nuclides Patrignani et al. (2016), the observation of would provide direct evidence for a beyond-the-Standard-Model process that violates lepton number conservation and constrain the absolute neutrino mass scale Engel and Menéndez (2017). Motivated by these implications, a variety of experiments are searching for in a number of nuclides, reaching half-life sensitivities in excess of years (e.g. Albert et al. (2014); Gando et al. (2016); Agostini et al. (2017)), with the most stringent for Xe at yr Gando et al. (2016).
EXO-200 is searching for in Xe (see Albert et al. (2014, 2014); Auger et al. (2012) for a detailed description). The detector consists of a cylindrical time projection chamber (TPC) filled with liquid xenon (LXe) enriched to 80.6% Xe. The TPC is split into two drift regions by a common cathode, each with radius 18 cm and drift length 20 cm. Energy depositions in the LXe produce both scintillation light and ionization. The ionization charge is read out after being drifted to crossed-wire planes at each anode by an electric field, while the scintillation light is collected by arrays of avalanche photodiodes (APDs) Neilson et al. (2009) located behind the wire planes. For each interaction, the location of the deposited charge in the directions perpendicular to the drift field ( and ) is determined from the wire signals. The position is reconstructed from the time delay between the prompt light signal and the delayed charge signals, using the measured ionization drift velocity Albert et al. (2017). The total energy deposited is determined from the combination of the charge and light signals, optimally accounting for their anticorrelation Conti et al. (2003).
The LXe is housed in a thin-walled copper vessel, and surrounded by several layers of passive and active shielding, including 50 cm of HFE-7000 cryofluid 3M HFE-7000, http://www.3m.com () and 25 cm of lead in all directions Auger et al. (2012). A plastic scintillator muon veto surrounds the experiment on four sides Albert et al. (2014, 2016). The detector is located at the Waste Isolation Pilot Plant (WIPP) near Carlsbad New Mexico, which provides an overburden of 1624 meters water equivalent Albert et al. (2016).
To reconstruct events in the TPC, charge and light signals are first grouped into individual energy deposits within each event. Events with a single reconstructed deposit are identified as “single-site” (SS), while events with multiple, spatially-separated deposits are denoted as “multi-site” (MS). This topological SS/MS classification has been used in previous EXO-200 analyses and provides discrimination between backgrounds, which are primarily MS, and the signal, which is primarily SS. A detailed detector Monte Carlo (MC) simulation based on Geant4 Allison et al. (2006) is used to model the energy deposits produced in the LXe by various backgrounds and the signal. The MC simulation propagates the charge deposits through the detector and produces simulated waveforms for each readout channel and event. The MC waveforms are processed using the same analysis framework as the data waveforms. In order to calibrate the detector energy scale and validate the accuracy of the MC simulation, runs are taken using sources positioned within 10 cm of the LXe vessel at locations around the cathode plane and at both anode planes. Figure 1 shows the agreement between the MC simulation and data acquired with Co, Ra, and Th sources.
EXO-200 has previously reported results on a search for Albert et al. (2014) using 80% of the data from its first run (“Phase I”), which spans from Sept. 2011 to Feb. 2014. In Feb. 2014, EXO-200 was forced to suspend operations, because of accidents at the WIPP facility and recover the Xe from the detector. After access to the experiment was regained in early 2015, the detector was recommissioned and refilled with LXe in Jan. 2016. Between Jan. and May 2016 the detector was upgraded with new electronics primarily aimed at improving the APD read-out noise. In addition, a system was installed to suppress radon in the air gap between the copper cryostat and the lead shield. After installation of this system, direct sampling of the air indicated that the average radon level was reduced by more than a factor of 10. Fits to the energy and location of backgrounds in physics data also indicated a lower best-fit value for this background component, although more data are required to demonstrate a statistically significant reduction. Finally, the electric field in the drift region of the detector was raised from 380 V/cm (cathode voltage, kV) to 567 V/cm ( kV). The data taking run with the upgraded detector began in May 2016 (“Phase II”).
The primary goal of the electronics upgrade was to minimize the APD read-out noise observed in Phase I. While this noise was accounted for in previous analyses and partially suppressed using a software “de-noising” algorithm Davis et al. (2016), the hardware upgrade provides substantially improved performance. The effect on the energy resolution is shown in Fig. 2. In Phase I, the SS resolution at the decay energy of keV Redshaw et al. (2007) after applying the software de-noising algorithm is %, averaged over live time and position. In Phase II, this figure is 1.23% and its time variation is greatly reduced. These values account for the spatial variation of the resolution, including events taken with the calibration source behind the anodes. Because of the source’s proximity to the readout when at the anode, these events present better energy resolution than those in Fig. 2.
The selection cuts for this analysis closely follow those used in previous EXO-200 analyses Albert et al. (2014). Both Phase II data and the previously examined Phase I data were “blinded” to remove candidate events in the energy region between 2345 keV and 2570 keV. After data quality cuts Albert et al. (2014), the total exposure considered here is 596.7 d and 271.8 d for Phase I and Phase II, respectively.
Only a fiducial volume (FV) within the detector is considered. The FV selection requires the position of all charge deposits in an event to be reconstructed within a hexagon with apothem of 162 mm and more than mm away from the anode and cathode wire planes, as well as from the cylindrical PTFE reflector inside the field-shaping rings. This FV corresponds to 74.7 kg of Xe, i.e. atoms, resulting in a total exposure of 177.6 kgyr or 1307 molyr. The individual exposure in Phase I and Phase II are 122 kgyr and 55.6 kgyr, respectively, or 898 molyr and 409 molyr.
To suppress backgrounds correlated in time, events are required to have only a single reconstructed scintillation signal and to occur s from all other reconstructed events. The corresponding signal reconstruction efficiency is found to be consistent between phases within errors, % (%) for Phase I (Phase II). The inefficiency is dominated by the 1 s anticoincidence cut and by incomplete reconstruction of events with small, separated energy deposits from bremsstrahlung. Its errors are determined from the difference in the observed absolute rate for calibration source data and MC using the known source activity, and measurements of the individual cut efficiencies for low-background events. It includes the estimation of the uncertainty in the FV, 2.8% in Phase I (2.6% in Phase II), the dominant term in this error.
This analysis introduces a cut to reduce the rate of background events arising from cosmogenically produced Xe Albert et al. (2016), which decays via emission with a total energy of 4173 keV Browne and Tuli (2007). Events in coincidence with the muon veto detector, and depositing energy consistent with the cascade s emitted after the neutron capture on Xe, are used to veto subsequent events in the same TPC half within 19.1 min, corresponding to . This cut was estimated to reduce the number of Xe events by %, with a loss in exposure of 3.5% (2.8%) in Phase I (Phase II). This reduction is consistent with the Xe rate entirely attributed to cosmogenic sources Albert et al. (2016).
New techniques have been developed to further improve -background rejection among events classified as SS by using the detailed topological information available for each interaction in the TPC. By implementing transverse electron diffusion (coefficient /s Albert et al. (2017)) and the three-dimensional geometry of the wire planes in the detector model, the number of channels that collect charge signals (denoted as “number of channels”) is now accurately simulated. Figure 3 (a) shows that SS backgrounds are more likely to deposit energy on more than one neighboring channel than the signal. In addition, extending this concept to the -direction, the distribution of the rise time of the charge pulse (defined as the time between collection of 5% to 95% of the total signal) is more likely to extend to large values for backgrounds relative to events (Fig. 3 (b)). Finally, the “standoff distance,” denoting the minimum distance between a cluster and the closest TPC surface, excluding the cathode, is used to constrain backgrounds originating from sources external to the LXe (Fig. 3 (c)).
A multivariate discriminator was developed by combining these topological variables in a boosted decision tree (BDT) using the TMVA software package Hoecker et al. (2007). The separation between SS and the most prominent backgrounds (U, Th, and Co) was maximized using a subset of the MC. Its performance was then tested on a statistically independent MC data set. Agreement between data and MC for calibration sources for both the BDT and its constituent variables was used to validate its performance for the main backgrounds with high statistics, while the corresponding distributions for signal-like events were investigated using a pure sample of SS events with energy near the . The ranked importance of the individual discriminator variables—defined as the weighted fraction of decision tree cuts for which each variable was used—was found to be 42%, 39%, and 19% for the rise time, standoff distance, and number of channels, respectively.
Figure 3 shows a comparison between the simulated and observed data distributions for calibration sources, and for the measured background-subtracted distribution. Overall, the data and MC distributions for the input variables and the overall discriminator agree to better than 10% at every bin. The detailed binning and range used for each variable was optimized to minimize systematic errors arising from imperfections in the MC simulation, while maintaining as much discriminating power as possible. As described below, the systematic errors resulting from the differences between the data and MC distributions are evaluated using toy MC studies. These residual differences contribute a sub-dominant uncertainty to the backgrounds and signal efficiency.
To search for a signal, the Phase I and Phase II data are separately fit to models using a binned maximum-likelihood (ML) fit. These models consist of the signal and backgrounds originating from the detector and surrounding materials. The background model closely follows that used in previous EXO-200 analyses Albert et al. (2014). The shape of the spectrum for each of the fit observables is determined from the MC simulation for each background and signal component. The energy spectra for the SS and MS data are fit simultaneously, and unlike the previous analysis of Phase I data Albert et al. (2014), the BDT variable (including the standoff variable) is added as a fit dimension for the SS data. Toy studies indicated that the addition of the BDT or standoff to the MS fit did not enhance sensitivity for this search. Systematic errors are included in the ML fit as nuisance parameters, constrained by normal distributions. An overall normalization constrained to unity is included to account for the error on the detection efficiency.
The balance between SS and MS events, parameterized by the “SS fraction,” is allowed to vary around the expected value from MC for each component within a systematic error. This error was determined by comparing the SS fraction for source calibration data and MC, as shown in the inset to Fig. 1. Averaging over all calibration positions acquired throughout Phase I (Phase II) gives a relative error on the SS fraction of 5.0% (8.8%). An 85% correlation between the SS fractions of the -like components is included in the constraint, justified by similar levels observed in calibration source data.
Since the ML fit relies on accurately modeling the shapes of the various background components, the impact of shape differences between data and MC was investigated for each fit observable (see Figs. 1 and 3). In these studies, the shapes of the -originated background components are corrected by using the residual differences between calibration source data and simulation, while the shapes of the SS components are corrected by using the residual differences of the measured background-subtracted spectrum. A large number of simulated data sets were drawn from the best-fit background model using the corrected PDFs, and were fit with the original simulated shapes. The resulting bias between the fitted and true value of backgrounds near is included as an additional systematic error on the normalization of the background components. Toy studies indicate that these shape errors are 2.1% (1.7%) for Phase I (Phase II). The contribution to this error caused by spatial and temporal energy resolution variations that are not fully accounted for by the MC simulation was determined to be 1.5% (1.2%) in Phase I (Phase II).
U, Th, and Co background components simulated at locations different from the default ones were individually inserted into the fit, and the resulting variation in the number of expected events near was determined. These studies estimate the error due to uncertainty in the location of the background model components to be 5.6% (5.9%) in Phase I (Phase II). All sources of systematic uncertainty on the background model near are treated as uncorrelated and result in a total error of 6.2% for both Phase I and Phase II, as summarized in Tab. 1.
|Source||Phase I||Phase II|
|Signal detection efficiency||3.0%||2.9%|
|Spectral shape agreement||2.1%||1.7%|
|Energy scale and resolution||1.5%||1.2%|
Two final constraints on the measured radon concentration in the LXe and relative rate of cosmogenically produced backgrounds were included in the fit, but verified to be unchanged from previous analyses Albert et al. (2014) for both Phase I and Phase II.
The analysis further accounts for a possible difference in the reconstructed energy for -like events, , relative to the energy scale determined from the calibration sources, . This difference is expressed through a multiplicative constant, , that scales the energy for all -like components, , which is allowed to float freely in the fit. is highly constrained by the spectrum, and consistent with unity to the sub-percent level.
After “unblinding” the combined data set, no statistically significant evidence for was observed. A lower limit on the half-life of yr at the 90% confidence level (CL) was derived from the ML fits after profiling over nuisance parameters. The data from each phase is fit separately and the profiles added together considering the difference in live time and signal detection efficiency. No correlation was considered between these two profiles. This conservative assumption was estimated to negligibly change the expected sensitivity. The profile-likelihood distribution was determined from toy MC simulations, following the same procedure to combine phases, and found to be in good agreement with Wilks’s theorem Wilks (1938); Cowan (1998). Under the assumption that neutrinos are Majorana particles, this corresponds to an upper limit on the Majorana neutrino mass, meV Dell’Oro et al. (2016), using the nuclear matrix elements of Barea et al. (2015); Rodriguez and Martínez-Pinedo (2010); Engel et al. (2014); Menéndez et al. (2009); Mustonen and Engel (2013) and phase space factor from Kotila and Iachello (2012). The best-fit value for the component is consistent with the null hypothesis at 1.5, corresponding to a -value of 0.12.
The results of the ML fits are presented in Fig 4. The measured rates were found to be consistent with Albert et al. (2014). The best-fit contributions from the primary background components within are summarized in the inset table in Fig 4 (top right). The best-fit total event rate is kgyrkeV [ kgyrkeV] when normalized to the full mass including all Xe isotopes for Phase I [Phase II].
The median 90% CL sensitivity was estimated from toy MC studies to be yr. This represents a factor of 2 improvement over the previous EXO-200 search Albert et al. (2014). In comparison to fits using the energy spectra and SS/MS classification alone, or with the addition of only the standoff distance, the use of the BDT discriminator provides a 15% increase in sensitivity.
The individual Phase I and Phase II data set lower limits of yr and yr at the 90% CL, respectively, with corresponding median sensitivity of yr and yr. Because of the detector upgrades and improved topological discrimination described here, the Phase II sensitivity from this analysis is already comparable to that of the previous EXO-200 search Albert et al. (2014) with an exposure that is half the size. The combined analysis of the Phase I and Phase II data provides one of the most sensitive searches for for any isotope Gando et al. (2016); Agostini et al. (2017) to date. Further operation of the upgraded detector is expected to continue improving sensitivity to , and holds promise for nEXO nEXO Collaboration (2017), a tonne-scale LXe TPC being designed to reach half-life sensitivity of yr.
Acknowledgements.EXO-200 is supported by DOE and NSF in the United States, NSERC in Canada, SNF in Switzerland, IBS in Korea, RFBR in Russia, DFG in Germany, and CAS and ISTCP in China. EXO-200 data analysis and simulation uses resources of the National Energy Research Scientific Computing Center (NERSC). We gratefully acknowledge the KARMEN collaboration for supplying the cosmic-ray veto detectors, and the WIPP for their hospitality.
- Furry (1939) W. H. Furry, Phys. Rev. 56, 1184 (1939).
- Dell’Oro et al. (2016) S. Dell’Oro, S. Marcocci, M. Viel, and F. Vissani, Adv. High Energy Phys., 2162659 (2016), arXiv:1601.07512 [hep-ph] .
- Patrignani et al. (2016) C. Patrignani et al. (Particle Data Group), Chin. Phys. C 40, 100001 (2016).
- Engel and Menéndez (2017) J. Engel and J. Menéndez, Rept. Prog. Phys. 80, 046301 (2017), arXiv:1610.06548 [nucl-th] .
- Albert et al. (2014) J. B. Albert et al. (EXO-200), Nature 510, 229 (2014), arXiv:1402.6956 [nucl-ex] .
- Gando et al. (2016) A. Gando et al. (KamLAND-Zen), Phys. Rev. Lett. 117, 082503 (2016), [Addendum: Phys. Rev. Lett. 117, 109903(2016)], arXiv:1605.02889 [hep-ex] .
- Agostini et al. (2017) M. Agostini et al. (GERDA), Nature 544, 47 (2017), arXiv:1703.00570 [nucl-ex] .
- Albert et al. (2014) J. B. Albert et al. (EXO-200), Phys. Rev. C 89, 015502 (2014).
- Auger et al. (2012) M. Auger et al. (EXO-200), JINST 7, P05010 (2012), arXiv:1202.2192 [physics.ins-det] .
- Neilson et al. (2009) R. Neilson et al. (EXO-200), Nucl. Inst. & Meth. A 608, 68 (2009), arXiv:0906.2499 [physics.ins-det] .
- Albert et al. (2017) J. B. Albert et al. (EXO-200), Phys. Rev. C95, 025502 (2017), arXiv:1609.04467 [physics.ins-det] .
- Conti et al. (2003) E. Conti et al., Phys. Rev. B 68, 054201 (2003).
- (13) 3M HFE-7000, http://www.3m.com.
- Albert et al. (2016) J. B. Albert et al. (EXO-200), JCAP 1604, 029 (2016), arXiv:1512.06835 [nucl-ex] .
- Allison et al. (2006) J. Allison et al., IEEE Trans. Nucl. Sci. 53, 270 (2006).
- Davis et al. (2016) C. G. Davis et al. (EXO-200), JINST 11, P07015 (2016), arXiv:1605.06552 [physics.ins-det] .
- Redshaw et al. (2007) M. Redshaw, E. Wingfield, J. McDaniel, and E. G. Myers, Phys. Rev. Lett. 98, 053003 (2007).
- Browne and Tuli (2007) E. Browne and J. Tuli, Nucl. Data Sheets 108, 2173 (2007).
- Hoecker et al. (2007) A. Hoecker et al., PoS(ACAT), 040 (2007), arXiv:physics/0703039 .
- Wilks (1938) S. S. Wilks, Ann. Math. Statist. 9, 60 (1938).
- Cowan (1998) G. Cowan, Statistical Data Analysis, Oxford science publications (Clarendon Press, 1998).
- Barea et al. (2015) J. Barea, J. Kotila, and F. Iachello, Phys. Rev. C 91, 034304 (2015).
- Rodriguez and Martínez-Pinedo (2010) T. R. Rodriguez and G. Martínez-Pinedo, Phys. Rev. Lett. 105, 252503 (2010).
- Engel et al. (2014) J. Engel, F. Šimkovic, and P. Vogel, Phys. Rev. C 89, 064308 (2014).
- Menéndez et al. (2009) J. Menéndez, A. Poves, E. Caurier, and F. Nowacki, Nucl. Phys. A 818, 139 (2009).
- Mustonen and Engel (2013) M. T. Mustonen and J. Engel, Phys. Rev. C 87, 064302 (2013).
- Kotila and Iachello (2012) J. Kotila and F. Iachello, Phys. Rev. C 85, 034316 (2012), arXiv:1209.5722 [nucl-th] .
- nEXO Collaboration (2017) J. B. Albert et al. (nEXO), arXiv:1710.05075 [nucl-ex] (2017).