betaFeedback

WorksPhysics and Astronomy

An assay-based background projection for the MAJORANA DEMONSTRATOR using Monte Carlo Uncertainty Propagation

Abstract

The background index is an important quantity which is used in projecting and calculating the half-life sensitivity of neutrinoless double-beta decay (0νββ0νββ) experiments. A novel analysis framework is presented to calculate the background index using the specific activities, masses and simulated efficiencies of an experiment's components as distributions. This Bayesian framework includes a unified approach to combine specific activities from assay. Monte Carlo uncertainty propagation is used to build a background index distribution from the specific activity, mass and efficiency distributions. This analysis method is applied to the MAJORANA DEMONSTRATOR, which deployed arrays of high-purity Ge detectors enriched in 76^{76}Ge to search for 0νββ0νββ. The framework projects a mean background index of [8.95±0.36]×10−4\left[8.95 \pm 0.36\right] \times 10^{-4}cts/(keV kg yr) from 232^{232}Th and 238^{238}U in the DEMONSTRATOR's components.

Introduction

A variety of low background experiments form the rich experimental program searching for neutrinoless double-beta decay (0νββ0\nu\beta\beta). While the decay remains unobserved in all candidate isotopes, the half-life has been constrained to be above 1025−102610^{25} - 10^{26} years by recent experiments [1, 2, 3, 4, 5]. The next generation of proposed experiments target a half-life sensitivity, T1/2T_{1/2}, of 102710^{27}–102810^{28} years [6, 7, 8]. To achieve these sensitivities, 0νββ0\nu\beta\beta experiments require underground locations, large isotopic mass and low-background construction materials. Extensive assay screenings are conducted to determine if the experiments structural components meet the targeted background levels.

In terms of the total electron kinetic energy, the experimental signature of 0νββ0\nu\beta\beta is a peak at the Q-value of the decay, QββQ_{\beta\beta}. If no background is present in this region of interest (ROI), the sensitivity of a 0νββ0\nu\beta\beta experiment scales linearly with the product of its isotopic mass, MM, and exposure time, tt [9]. However, if the specific background rate, bb, is large enough (such that the uncertainty on the background level is proportional to bΔE\sqrt{b\Delta E} [10]) the half-life sensitivity (T1/2T_{1/2}) scales with:

T1/2∝MtbΔE.(1)T_{1/2} \propto\sqrt{\frac{Mt}{b\Delta E}}. \tag*{(1)}

The width of the ROI, ΔE\Delta E, is related to bb and the energy resolution at QββQ_{\beta\beta}. The specific background rate is measured in counts per keV in the ROI, per kg of detector mass, per year (cts/(keV kg yr)). This observable is also referred to as the background index (BI). Given the low-background nature of 0νββ0\nu\beta\beta experiments, the number of background counts in the ROI is typically too low to estimate the BI (note that no signal would have to be assumed). In such cases, a wider proxy region is needed to increase statistics. This background estimation window (BEW) is often asymmetric around QββQ_{\beta\beta} to avoid running into the 2νββ2\nu\beta\beta spectrum and known gamma lines.

Germanium detector technology has been developed for decades, finding applications in radiometric assays and γ\gamma-ray spectroscopy. Ge detectors offer superb energy resolution – resulting in a narrower ROI – and can be readily enriched to ∼90%\sim90\% 76Ge{}^{76}\mathrm{Ge} [11]. The Majorana Demonstrator and GERDA experiments exploited this technology to search for 0νββ0\nu\beta\beta in 76Ge{}^{76}\mathrm{Ge}-enriched high purity Ge (HPGe) detectors [12]. Both experiments have the lowest backgrounds of any experiment in the present 0νββ0\nu\beta\beta experimental landscape [1, 2].

At the 2039 keV Q-value of 76Ge{}^{76}\mathrm{Ge} [13, 14], the Demonstrator achieved a BI of 6.2−0.5+0.6×10−36.2^{+0.6}_{-0.5} \times10^{-3} cts/(keV kg yr) in its low background configuration [1]. This is not in agreement with the originally projected BI of <8.75×10−4< 8.75 \times10^{-4} cts/(keV kg yr) [15]. The projection was based on simulations of the design geometry and the results of an extensive radioassay program. To account for any discrepancies between the design and as-built geometries of the Demonstrator, a new set of simulations was performed with the updated geometries. This as-built model did not find a significant deviation from the originally predicted BI. Nevertheless, neither calculation captured the often significant uncertainties from assay in the contribution of natural radiation (assay BI) to the total BI. Additionally, a systematic review of assay results highlighted the prevalence of measured values that did not agree within error, motivating the development of a technique which properly accounts for this spread and propagates it into the BI.

The total BI includes subdominant contributions from, amongst others, external and cosmogenically induced backgrounds. For this work, only backgrounds quantified by assay – 232Th{}^{232}\mathrm{Th}, 238U{}^{238}\mathrm{U} – are considered. While the components in the Demonstrator are also screened for 40K{}^{40}\mathrm{K}, it is excluded from the analysis since the γ\gamma-rays produced in the 40K{}^{40}\mathrm{K} decay chain have energies well below the BEW.

The assay BI is calculated by weighting the activity of a component with the Demonstrator’s simulated array-detection efficiency of decays originating from it. An upper limit on the activity of a component translates into an upper limit on its BI. In the previous projections, such upper limits were directly summed to BIs calculated from measured activities when summing over all modeled components of the Demonstrator. Thus, the total projected BI was reported as an upper limit itself. Null efficiencies were computed for some components far away from the detector array and failed to contribute to the BI. However, given the high activity of some of these components, their true contribution could still be significant. Therefore, simulation statistics must be properly taken into account.

Other 0νββ0\nu\beta\beta experiments use techniques which address these issues in part. By repeatedly drawing samples from an activity probability density function (PDF) and weighting them by the simulated efficiency, a BI distribution is generated. The CUORE Collaboration generates activity PDFs from fits to preliminary data [16]. On the other hand, the nEXO Collaboration promotes assay-based activities to a truncated-at-zero Gaussian PDF [17]. The latter technique is adopted by this work and expanded on by promoting the single-valued efficiency to a distribution as well. Ref. [18] provides a comprehensive summary of the methods used to estimate the BI of various 0νββ0\nu\beta\beta experiments.

In this article, an assay-based Bayesian framework to project the BI of low-background experiments is presented and applied to the Majorana Demonstrator. The framework takes as input all the assay and simulation efficiency data with the goal of:

  1. Combining multiple assay measurements, including upper limits, using a unified averaging method which properly accounts for the spread in data.

  1. Calculating uncertainties in non-Gaussian regimes such as those posed by null simulation efficiencies.

  1. Combining uncertainties in assay results, component masses and simulation statistics.

  1. Preserving the generality of this method, allowing for its adoption by the low-background community.

The Majorana Demonstrator

The Majorana Demonstrator consisted of two modules (M1, M2) [19] where a total of 40.4 kg of HPGe detectors (27.2 kg enriched to 88% in 76Ge{}^{76}\mathrm{Ge}) operated in vacuum at the 4850-foot-level (4300 meter water equivalent) of the Sanford Underground Research Facility (SURF) [20] in Lead, South Dakota. The ultra-low background and world-leading energy resolution achieved by the Demonstrator enabled a sensitive 0νββ0\nu\beta\beta decay search, as well as additional searches for physics beyond the Standard Model.

Both vacuum cryostats and all structural components of the detector arrays were machined from ultra-low-background underground electroformed copper (UGEFCu) [15, 21] and low-background plastics, such as DuPont™ Vespel® and polytetrafluoroethylene (PTFE). Low-radioactivity parylene was used to coat UGEFCu threads to prevent galling and for the cryostat seal [19]. A layered shield enclosed both modules. The innermost layer consisted of 5 cm of UGEFCu. Five cm of commercial oxygen-free high conductivity copper (OFHCCu) and 45 cm of high-purity lead followed. The shield and module volume were constantly purged with low-radon liquid nitrogen boil-off gas. The aluminum enclosure that isolated this Rn-excluded region was covered with a plastic active muon veto which provided near-4π4\pi coverage [22]. The near-detector readout system, which was designed for the Demonstrator, included low-mass front end (LMFE) electronics [23] and low-mass cables and connectors [24]. Cables were guided out of each module following a UGEFCu cross-arm which penetrated the layered shield. The cross-arm connected the cryostat with vacuum and cryogenic hardware. Control and readout electronics were just outside the Rn-excluded region. The entire assembly was surrounded by 5 cm of borated polyethylene and 25 cm of pure polyethylene to shield against neutrons.

All components inside the Rn-excluded region populate the as-built model. These were modeled in MaGe, the Geant4-based simulation software jointly developed by the GERDA and Majorana collaborations [25]. Due to possible radioactive shine through the cross-arm, the vacuum hardware was included as well. The as-built model is based on the Aug. 2016 to Nov. 2019 configuration of the Demonstrator, where up to 32.1 kg of detectors were operational. In Nov. 2019, M2 was upgraded with an improved set of cables and connectors and additional cross-arm shielding. The upgrade, in combination with a reconfiguration of M2 detectors, resulted in the final configuration of up to 40.4 kg of operational HPGe detectors. The final active enriched exposure of the Demonstrator was 64.5 kg yr. A low-background dataset — with an active exposure of 63.3 kg yr — was obtained by excluding data taken previous to the installation of the inner UGEFCu shield. From this dataset, a BI of 6.2−0.5+0.6×10−36.2^{+0.6}_{-0.5} \times10^{-3} cts/(keV kg yr) was calculated [1]. The previous data release of the Demonstrator does not include post-M2-upgrade data. The low-background dataset BI in this release is 4.7±0.8×10−34.7 \pm0.8 \times10^{-3} cts/(keV kg yr) with 21.3 kg yr of active exposure [26].

To simulate 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} decays originating from the hardware components of the Demonstrator, their respective decay chains were divided into 10 and 4 segments respectively, following the prescription in Ref. [27]. Within each segment secular equilibrium was assumed. The breakpoints in the chains — which generally correspond to isotopes with half-lives longer than 3 days — allow a break in secular equilibrium. However, given that the concentration of these isotopes is unknown, secular equilibrium of the chain as a whole was assumed as well. For a particular component and decay chain, the same number of decays were simulated for each segment and the energy depositions in the detectors were recorded. These were later combined with the appropriate branching ratios to produce a spectrum. Section IV describes how the component efficiency is extracted from this simulated spectrum.

A unified approach to combine assay results

The Demonstrator’s radioassay program delivered an extensive specific activity database (232Th{}^{232}\mathrm{Th}, 238U{}^{238}\mathrm{U}, 40K{}^{40}\mathrm{K}) of the materials used to build the experiment [15]. During the commissioning of the Demonstrator, additional samples were collected and assayed, thus continually growing this database. Amongst others, inductively coupled plasma mass spectrometry (ICPMS), γ\gamma-count and neutron activation analysis (NAA) measurements were performed. The most sensitive technique, ICPMS, measures specific elements within a decay chain and therefore secular equilibrium is assumed to project a specific activity. The concentration of 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} in HPGe detectors is far too low to be detected by ICPMS measurements. Thus, the 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} contaminations are deduced by searching for time-correlated α\alpha decays from their respective decay chains in the Demonstrator’s low-background dataset. These data-driven results (as opposed to the assay-driven projections published here) will be reported in a future publication. Despite the high detection efficiency of decays originating within Ge, the bulk 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} contamination is anticipated to be so low [28] that the BI contribution from these sources is expected to be sub-dominant.

In many cases, measurements of duplicate parts returned specific activities that did not agree within error. A method is thus needed to properly combine these results. The method should take into account the possible sample-to-sample variation of contaminants and the different detection limits of assay methods. It should also allow for the inclusion of assays which result in upper limits.

Following the methodology of the Particle Data Group (PDG) for unconstrained averaging, a standard weighted least-squares approach is employed [29]. The average specific activity and its uncertainty are calculated as

aˉ±δaˉ=∑kwkak∑kwk±(∑kwk)−1/2,(2)\bar{a} \pm\delta\bar{a} = \frac{\sum_k w_k a_k}{\sum_k w_k} \pm\left(\sum_k w_k\right)^{-1/2}, \tag*{(2)}

where the kk-th specific activity from assay, aka_k, is weighted by

wk=1/(δak)2.(3)w_k = 1/(\delta a_k)^2. \tag*{(3)}

NaN_a assays populate the sum, including all assays resulting in a measured value. The PDG states “We do not average or combine upper limits except in a very few cases where they may be re-expressed as measured numbers with Gaussian errors.” [29] Exactly the latter is used to treat specific activity upper limits. More concisely, only the most stringent 90% C.L. upper limit, ll, is re-expressed as a measured number, ak=0a_k = 0, with Gaussian error δak=l/[2erf⁡−1(0.9)]\delta a_k = l/[\sqrt{2}\operatorname{erf}^{-1}(0.9)]. Note that an issue would arise if multiple upper limits are combined. Combining two identical ak±δaka_k \pm\delta a_k results in a smaller combined uncertainty δaˉ=δak/2\delta\bar{a} = \delta a_k/\sqrt{2}. This is a desirable property for measured values, but not for upper limits. Repeated results are likely if the activity of a source is significantly below the detection limit of the assay apparatus. Combining such measurements would result in an unrealistically lowered upper limit. Therefore only the most stringent upper limit is chosen. A maximum of one upper limit is thus included in the sum of Eq. (2).

Once the average is computed, χ2=∑kwk(aˉ−ak)2\chi^2 = \sum_k w_k(\bar{a} - a_k)^2 is calculated. χ2/(Na−1)\chi^2/(N_a - 1) is used as a discriminant as follows:

  1. If χ2/(Na−1)≤1\chi^2/(N_a - 1) \le1 the average, aˉ\bar{a}, and uncertainty, δaˉ\delta\bar{a}, are accepted.

  1. If χ2/(Na−1)>1\chi^2/(N_a - 1) > 1 the uncertainty, δaˉ\delta\bar{a}, is scaled by a factor Σ=[χ2/(Na−1)]1/2\Sigma= [\chi^2/(N_a - 1)]^{1/2}.

Fig. 1 exemplifies the technique for combining assay results. Following the nEXO collaboration’s treatment of activities, these are visualized as truncated-at-zero Gaussian distributions. In the figure two example assay results, ak±δaka_k \pm\delta a_k, are plotted as such. These are averaged using the technique described above and the result, aˉ±δaˉ\bar{a} \pm\delta\bar{a}, is shown. An upper limit lower than a measured value lowers the measured value Figure 1(a), whereas an upper limit higher than a measured value has almost no effect on the same Figure 1(b). Note that in Figure 1(a) a scaling factor of Σ\Sigma is applied since χ2>1\chi^2 > 1. If two measured values agree within error, the combined uncertainty decreases Figure 1(c). If there is no agreement, the combined uncertainty increases since a scaling factor of Σ\Sigma is applied Figure 1(d).

Figure 1. The combination of an upper limit with a measured value (a,b) and the combination of two measured values (c,d) are shown as truncated-at-zero Gaussian distributions in solid red. Toy data is used for the raw upper limits and measured values, which are also depicted as truncated-at-zero Gaussian distributions but in dashed grey.

The levels of 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} in the Demonstrator’s components often lie below the detection limits of the most sensitive assays, leading to a high prevalence of upper limits. The motivation to include upper limits, not only in the sum of Eq. (2) but also in the calculation of the scaling factor Σ\Sigma, stems from this prevalence. To justify their use, the effect of combining an upper limit, ll, with a fixed measured value, aˉ±δaˉ\bar{a} \pm\delta\bar{a}, with small error (δa<aˉ\delta a < \bar{a}), was evaluated. This case is representative of many assay results. Figure 2 shows the χ2\chi^2, average and uncertainty with respect to a Gaussian upper limit, δal\delta a_l, between 0 and 1.1a1.1a. It demonstrates that averaging with an upper limit leads to the following desirable properties:

Figure 2. The χ2\chi^2, average (aˉ\bar{a}) and uncertainty (δaˉ\delta\bar{a}), of a measured value, a±δaa\pm\delta a, and the Gaussian upper limit, δal\delta a_l, defined in the text. δa\delta a is set to 0.2a0.2a. A dashed and dash-dotted vertical line is shown at δal=δa\delta a_l=\delta a and at the δal\delta a_l which results in χ2=1\chi^2=1.

  1. For δal>a+δa\delta a_l > a + \delta a, aˉ\bar{a} and δaˉ\delta\bar{a} rapidly converge to aa and δa\delta a respectively. Smaller δa\delta a lead to faster convergence. In other words, upper limits higher than measured values have little effect on the same. The lower the measured value’s error, the lower the impact. In this regime χ2<1\chi^2 < 1.

  1. For δal<a+δa\delta a_l < a + \delta a, aˉ\bar{a} decreases monotonically with δal\delta a_l.

  1. The uncertainty, δaˉ\delta\bar{a} is maximized when δal=δa\delta a_l = \delta a. For this value, aˉ=a/2\bar{a} = a/2. Note the importance of scaling the uncertainty and thus the need to include the upper limit in the calculation of Σ\Sigma.

If assays have to be averaged more than once, multiple upper limits may contribute to the result. For example, 40 of the Demonstrator’s lead bricks were assayed with the following procedure. One or more samples were taken from each brick, and each sample was used to prepare one or more dilutions to conduct ICPMS measurements. In some cases each dilution was separated into multiple vials, which were separately measured. Thus averaging is performed at the dilution, sample, brick and global levels, with upper limits contributing at each stage.

The specific activity averages and uncertainties of the materials used in the Majorana Demonstrator are presented in Table I. These values are used to calculate the BI contributions of all 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} sources used to model the experiment. In this table, NaN_a represents the number of assays contributing at the global averaging level. While the set of assayed components is limited, it is assumed to be representative of all those used in the Demonstrator.

MaterialGroup232Th^{232}\mathrm{Th} [µBq/kg]238U^{238}\mathrm{U} [µBq/kg]NaN_aMethod
LMFEFront Ends6950 ± 83010600 ± 3002ICPMS
HV CableCables87.9 ± 52.4231 ± 343ICPMS
Signal CableCables546 ± 112530 ± 583ICPMS
Stock UGEFCuElectroformed Cu0.188 ± 0.0290.137 ± 0.0445ICPMS
Machined UGEFCuElectroformed Cu0.575 ± 0.0880.752 ± 0.08312ICPMS
OFHCCuOFHC Cu Shielding1.10 ± 0.141.37 ± 0.182ICPMS
Pb BricksPb Shielding9.53 ± 1.0125.6 ± 1.529, 19ICPMS
Female ConnectorConnectors390 ± 7540 ± 91ICPMS
Male ConnectorConnectors28.8 ± 2.0130 ± 111ICPMS
PTFE O-ringOther Plastics39.7 ± 31.9< 1052, 1NAA
PTFEDetector Unit PTFE0.101 ± 0.008<4.971NAA
DuPont™ Vespel®Other Plastics360 ± 234403 ± 1793ICPMS
PTFE GasketOther Plastics<20.7<94.51NAA
Stainless SteelVacuum Hardware13000 ± 4000<50001γ-count
Glass BreakVacuum Hardware49000 ± 8000160000 ± 100001γ-count
PTFE TubingOther Plastics6.09 ± 7.30<38.61ICPMS
ParyleneParylene2150 ± 1203110 ± 7501ICPMS

Table I. Measured (Na=1N_a = 1) or combined (Na>1N_a > 1, using the technique described in Section III) bulk 232Th^{232}\mathrm{Th} and 238U^{238}\mathrm{U} specific activities of materials used in the Majorana Demonstrator. Where applicable, the most stringent 90% C.L. upper limit is used. The assay methodology is the same for all NaN_a assays in a given row. If NaN_a is different for 232Th^{232}\mathrm{Th} and 238U^{238}\mathrm{U}, then comma separated values are given. Machined UGEFCu samples were fabricated from stock UGEFCu. The latter is assumed for the specific activity of the inner copper shield of the Demonstrator. Signal cables were separated into two sections. Custom connectors were designed and built to connect these sections, with the female end wired to an LMFE and the male end wired to pre-amplification and readout electronics. When calculating the BI, all components made from a given material are collected in the group shown in the second column. A total of NaN_a assays – sourced from Ref. [15] and additional measurement campaigns – are combined.

Background index

Simulations predict an approximately flat background in the 370 keV split BEW covering 1950-2350 keV. The BEW has three 10 keV regions removed, centered at 2103 keV, the 208Tl{}^{208}\mathrm{Tl} (232Th{}^{232}\mathrm{Th}) single escape peak, and 2118 keV and 2204 keV, the 214Bi{}^{214}\mathrm{Bi} (238U{}^{238}\mathrm{U}) peaks. This BEW is used to estimate the BI at QββQ_{\beta\beta}, for both simulations and data. When estimating the BI from data an additional 10 keV region, centered at QββQ_{\beta\beta}, is removed [1].

The contribution to the Demonstrator’s BI from a natural radiation source ii – the 232Th{}^{232}\mathrm{Th} or 238U{}^{238}\mathrm{U} contamination in a component of mass mim_i – is calculated as follows. For each segment, jj, of the decay chain of the source’s contaminant, MaGe simulates NijN_{ij} decays. For a given source ii, all NijN_{ij} are equal up to computational errors. Simulated events leading to energy depositions in an operational detector are subject to the same anti-coincidence and pulse shape analysis cuts applied to data (designed to select 0νββ0\nu\beta\beta-like single-site events). The number of counts passing all cuts, nijn_{ij}, is calculated by integrating the resulting combined detector spectra over the BEW. The corresponding segment efficiency is thus:

ϵijMJD=nijNij(4)\epsilon_{ij}^{\mathrm{MJD}} = \frac{n_{ij}}{N_{ij}} \tag*{(4)}

The segment efficiencies are weighted by the branching ratio of each segment, BjB_j, and summed over the total number of segments, SiS_i, to produce the source efficiency, ϵiMJD\epsilon_i^{\mathrm{MJD}}:

ϵiMJD=∑jSiBjϵijMJD(5)\epsilon_i^{\mathrm{MJD}} = \sum_{j}^{S_i} B_j\epsilon_{ij}^{\mathrm{MJD}} \tag*{(5)}

The source efficiency, mass and averaged specific activity, aˉi\bar{a}_i, are used to calculate the BI of the source:

BIi=1M×ΔϵiMJD×aˉi×mi(6)\mathrm{BI}_i = \frac{1}{M\times\Delta}\epsilon_i^{\mathrm{MJD}}\times\bar{a}_i\times m_i \tag*{(6)}

Where Δ=370 keV\Delta=370\ \mathrm{keV} is the width of the BEW and MM is the mass of operational detectors in the array. Note that the total number of sources is equal to twice the number of components used to model the Demonstrator, since 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} are accounted for in all components. The source contributions to the total assay BI =∑iBIi=\sum_i\mathrm{BI}_i, can be summed as desired, either by source material, contaminant, or component group. If only an upper limit, lil_i, is available, it can be used in place of aˉi\bar{a}_i. This calculates a BI which is the direct sum of upper limits and central values and is referred to as the direct BI in the text. The component-group-combined 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} direct BIs are calculated by summing over the appropriate sources and are presented in Table II. Note that the design geometry projection of Ref. [15] used this method. It is not possible to assign an uncertainty to the direct BI because of the inclusion of upper limits. However, a proper treatment of uncertainties can be obtained by promoting the single-valued aˉi\bar{a}_i (or lil_i), mim_i and ϵiMJD\epsilon_i^{\mathrm{MJD}} to distributions and using Monte Carlo uncertainty propagation to combine these in a final BI distribution.

Promoting single values to distributions

As described in Section III the specific activity can be expressed as a truncated-at-zero Gaussian. The mass, mim_i, is also promoted to a distribution of this form. Depending on the method used to measure or estimate the mass, a Gaussian uncertainty is assigned. This uncertainty ranges from 1%, to account for sample to sample variation in direct mass measurements, to 10% for masses that were estimated from geometry. In practice the mass PDFs are indistinguishable from a true Gaussian given the assigned level of uncertainty.

The derivation of the efficiency, ϵiMJD\epsilon_i^{\mathrm{MJD}}, PDF follows. Starting from Eq. (4), the probability of finding nn counts in the BEW – with expectation value λ\lambda – is given by the Poisson distribution, P(n∣λ)=λne−λ/n!P(n|\lambda) = \lambda^n e^{-\lambda}/n!. Conversely, P(λ∣n)P(\lambda|n) is deduced via Bayes’ theorem, P(λ∣n)∝P(λ)P(n∣λ)P(\lambda|n) \propto P(\lambda)P(n|\lambda), using the prior,

P(λ)=1λ1−1/S.(7)P(\lambda) = \frac{1}{\lambda^{1-1/S}}. \tag*{(7)}

The choice of uninformative prior corresponds to a flat prior on the sum of the SS decay chain segments. The functional form of the prior is derived in Appendix § 0.9.

P(λ∣n)∝λn−1+1/Se−λn!P(\lambda|n) \propto\frac{\lambda^{n-1+1/S}e^{-\lambda}}{n!}
∝Γ(λ∣α=n+1/S, β=1)(8)\propto\Gamma(\lambda|\alpha= n + 1/S,\ \beta= 1) \tag*{(8)}

The resulting posterior is a Gamma distribution with parameters α=n+1/S\alpha= n + 1/S and β=1\beta= 1. The probability of obtaining an efficiency ϵiMJD\epsilon_i^{\mathrm{MJD}}, given an ensemble of counts {nij}\{n_{ij}\}, is thus,

P(ϵiMJD∣{nij})=∑jSiBjP(ϵiMJD∣nij)=∑jSiBjP(λij∣nij)Nij.(9)\begin{aligned} P(\epsilon_i^{\mathrm{MJD}}|\{n_{ij}\}) = \sum_{j}^{S_i} B_jP(\epsilon_i^{\mathrm{MJD}}|n_{ij}) \\ &= \sum_{j}^{S_i} B_j\frac{P(\lambda_{ij}|n_{ij})}{N_{ij}}. \tag*{(9)} \end{aligned}

This result is obtained by replacing nijn_{ij} by P(λij∣nij)P(\lambda_{ij}|n_{ij}) in Eq. (4) and carrying it into Eq. (5). The PDF of ϵiMJD\epsilon_i^{\mathrm{MJD}} is computed numerically. Taking a random draw from the PDF in Eq. (8) for each segment, weighing by the appropriate factors and summing them as in Eq. (9). This process is repeated >105>10^5 times to generate the PDF or only once if one efficiency sample is needed.

Not all segments are included in the sum for far away sources, only segments which produce γ\gamma’s with sufficiently high energies to lead to energy depositions in the BEW. The excluded segments have a true zero efficiency.

Monte Carlo uncertainty propagation

The assay-based background index can be promoted to a distribution by sampling the PDFs described in § 0.5. The BIi_i distributions are thus generated in a similar manner as that of the efficiency. A random draw is taken from the specific activity, mass, and efficiency PDFs. These are multiplied and scaled as in Eq. (6). The process is repeated 10610^6 times to produce a PDF for the BI of the source. Each set of draws constitutes a toy experiment resulting in a different BI. The uncertainty of the efficiency, specific activity and mass – embedded in their corresponding PDFs – is propagated into the BIi_i distribution through this process, referred to as Monte Carlo uncertainty propagation.

The BIi_i PDFs are combined by taking the direct sum of the samples that were used to generate them. Figure 3 shows the result of summing all the BIi_i PDFs belonging to a component group in the Demonstrator. The mean and uncertainties that are extracted from these distributions are reported in Table II and plotted as error bars in Figure 3. The total BI distribution of 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} sources in the Demonstrator is computed in a similar manner and displayed at the bottom of the figure.

Figure 3. The mean BI and 1σ1\sigma uncertainties of the Demonstrator’s component groups and their sum are plotted as vertical black lines through dark grey error bars. Statistical uncertainty contributions are shown in red. These are overlaid on their corresponding PDFs in light grey (with the central 2σ2\sigma region shaded). All PDFs contain 10610^{6} samples and are scaled vertically such that they have the same height.

Simulation statistical uncertainties can be isolated by collapsing the specific activity and mass distributions to their respective means and using them to weigh the samples that were drawn from the efficiency distribution. This information can be used by the analyst as a guide to set the number of decays in future simulations.

Monte Carlo uncertainty propagation allows for all potential sources of natural radiation to be incorporated in the model, not just those which produced counts in the BEW. With no counts in the BEW, the single-valued efficiency is zero. However, the efficiency distribution will take an exponential form, effectively setting an upper limit on the BI (once weighted properly) for the source in question. This upper limit is fully dependent on the number of simulated decays and given enough computational resources should be driven down to the point where it is negligible compared to other contributions to the BI. This is the case for the Vacuum Hardware and 238U{}^{238}\mathrm{U} contamination.

Projected background and conclusions

As Section VI describes, the mean BI and the corresponding uncertainties of the Majorana Demonstrator’s component groups are extracted from their distributions in Figure 3 and reported in Table II. The mean total natural radiation BI, determined from its distribution at the bottom of the figure, is

assayBI=[8.95±0.16(stat.)±0.20(act.)]×10−4{}^{\mathrm{assay}}\mathrm{BI} = [8.95 \pm0.16\mathrm{(stat.)} \pm0.20\mathrm{(act.)}] \times10^{-4}

in units of cts/(keV kg yr). The uncertainty from activity (act.) dominates assayBI{}^{\mathrm{assay}}\mathrm{BI} and most component group BIs. The statistical uncertainty from simulation (stat.) has been calculated with the method described in Section VI. Note that the symmetric uncertainty around the mean does not capture the asymmetry of the distribution. The contributions from the 232Th{}^{232}\mathrm{Th} and 238U{}^{238}\mathrm{U} decay chains, given in units of cts/(keV kg yr), follow:

232BI=[7.37±0.12(stat.)±0.22(act.)]×10−4,{}^{232}\mathrm{BI} = [7.37 \pm0.12\mathrm{(stat.)} \pm0.22\mathrm{(act.)}] \times10^{-4},
238BI=[1.58±0.11(stat.)±0.01(act.)]×10−4.{}^{238}\mathrm{BI} = [1.58 \pm0.11\mathrm{(stat.)} \pm0.01\mathrm{(act.)}] \times10^{-4}.

By comparing the direct BI of Table II with the design geometry projection of Ref. [15] for the same components, a 44% increase is found. Uncertainties were not calculated for the design geometry projection. On the other hand, Monte Carlo uncertainty propagation captured the uncertainties from specific activity, component mass and simulation efficiencies in the BI. Multiple contaminants, which have null single-valued efficiencies, contribute to the BI under this new framework.

The projected assayBI{}^{\mathrm{assay}}\mathrm{BI} does not capture the uncertainty associated with the assumption of secular equilibrium or account for the possible introduction of backgrounds during the construction of the Demonstrator. Additionally, given the limited number of assayed components, the variation in contaminants may be larger than captured by the averaged assay uncertainty. Furthermore, possible systematic uncertainties in simulated component geometry were not taken into account. Nevertheless, the Monte Carlo uncertainty propagation framework that has been developed can be extended to account for these effects.

GroupMean 232^{232}Th BI [cts/(keV kg yr)]Mean 238^{238}U BI [cts/(keV kg yr)]Mean BI [cts/(keV kg yr)]Direct BI [cts/(keV kg yr)]
Electroformed Cu2.44−0.19+0.19×10−42.44^{+0.19}_{-0.19} \times10^{-4}3.41−0.23+0.23×10−53.41^{+0.23}_{-0.23} \times10^{-5}2.78−0.19+0.19×10−42.78^{+0.19}_{-0.19} \times10^{-4}2.75×10−42.75 \times10^{-4}
OFHC Cu Shielding5.49−0.77+0.77×10−55.49^{+0.77}_{-0.77} \times10^{-5}5.79−1.21+1.21×10−65.79^{+1.21}_{-1.21} \times10^{-6}6.07−0.78+0.78×10−56.07^{+0.78}_{-0.78} \times10^{-5}6.05×10−56.05 \times10^{-5}
Pb Shielding7.75−1.38+1.38×10−57.75^{+1.38}_{-1.38} \times10^{-5}2.52−1.06+1.06×10−52.52^{+1.06}_{-1.06} \times10^{-5}1.03−0.18+0.18×10−41.03^{+0.18}_{-0.18} \times10^{-4}9.70×10−59.70 \times10^{-5}
Cables1.41−0.17+0.17×10−41.41^{+0.17}_{-0.17} \times10^{-4}3.29−0.20+0.20×10−53.29^{+0.20}_{-0.20} \times10^{-5}1.74−0.17+0.17×10−41.74^{+0.17}_{-0.17} \times10^{-4}1.93×10−41.93 \times10^{-4}
Connectors2.09−0.03+0.03×10−52.09^{+0.03}_{-0.03} \times10^{-5}3.46−0.07+0.07×10−63.46^{+0.07}_{-0.07} \times10^{-6}2.44−0.03+0.03×10−52.44^{+0.03}_{-0.03} \times10^{-5}2.44×10−52.44 \times10^{-5}
Front Ends1.54−0.13+0.13×10−41.54^{+0.13}_{-0.13} \times10^{-4}4.67−0.10+0.10×10−54.67^{+0.10}_{-0.10} \times10^{-5}2.01−0.13+0.13×10−42.01^{+0.13}_{-0.13} \times10^{-4}2.15×10−42.15 \times10^{-4}
Detector Unit PTFE1.96−0.18+0.18×10−71.96^{+0.18}_{-0.18} \times10^{-7}1.05−0.57+0.57×10−61.05^{+0.57}_{-0.57} \times10^{-6}1.24−0.57+0.57×10−61.24^{+0.57}_{-0.57} \times10^{-6}2.36×10−62.36 \times10^{-6}
Other Plastics3.18−0.97+0.97×10−53.18^{+0.97}_{-0.97} \times10^{-5}5.99−2.52+2.53×10−65.99^{+2.53}_{-2.52} \times10^{-6}3.78−1.00+1.00×10−53.78^{+1.00}_{-1.00} \times10^{-5}4.10×10−54.10 \times10^{-5}
Vacuum Hardware1.06−0.25+0.25×10−61.06^{+0.25}_{-0.25} \times10^{-6}<7.80×10−8< 7.80 \times10^{-8}1.10−0.25+0.25×10−61.10^{+0.25}_{-0.25} \times10^{-6}9.41×10−79.41 \times10^{-7}
Parylene1.10−0.08+0.08×10−51.10^{+0.08}_{-0.08} \times10^{-5}2.55−0.48+0.48×10−62.55^{+0.48}_{-0.48} \times10^{-6}1.35−0.09+0.09×10−51.35^{+0.09}_{-0.09} \times10^{-5}1.35×10−51.35 \times10^{-5}
SUM7.37−0.34+0.34×10−47.37^{+0.34}_{-0.34} \times10^{-4}1.58−0.11+0.11×10−41.58^{+0.11}_{-0.11} \times10^{-4}8.95−0.36+0.36×10−48.95^{+0.36}_{-0.36} \times10^{-4}9.23×10−49.23 \times10^{-4}

Table 2. Background indices and 1σ1\sigma uncertainties of the component groups of the Demonstrator as derived from their distributions. A 90% C.L. upper limit is given if the width of the central 2σ2\sigma region is higher than the width of the lower 2σ2\sigma region. The direct BIs are included for reference.

The techniques outlined in this article can be exploited to project the BI of future 0νββ0\nu\beta\beta experiments. Particularly, the design of such experiments can benefit from the ability to include component mass uncertainties, since the designed and as-built geometries often differ. Additionally, the framework informs the analyst on the number of decays to simulate, thus optimizing computational resources. The unified approach to average assay results can facilitate the standardization of assay reporting. This is of importance in the field, given that the design of new experiments often draws from assay data collected by others. The uncertainty extracted from the BI can be propagated into the projected sensitivity, thus further illuminating the physics reach of the next generation of 0νββ0\nu\beta\beta experiments.

Acknowledgments

This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics under contract / award numbers DE-AC02-05CH11231, DE-AC05-00OR22725, DE-AC05-76RL0130, DE-FG02-97ER41020, DE-FG02-97ER41033, DE-FG02-97ER41041, DE-SC0012612, DE-SC0014445, DE-SC0017594, DE-SC0080160, DE-SC0022339, and LANL/ LEML77/LANLLEM78. We acknowledge support from the Particle Astrophysics Program and Nuclear Physics Program of the National Science Foundation through grant numbers MR-0923142, PHY-1003399, PHY-1102292, PHY-1206314, PHY-1614611, PHY-1812409, PHY-1812356, PHY-2111140, and PHY-209530.

We gratefully acknowledge the support of the Laboratory Directed Research & Development (LDRD) program at Lawrence Berkeley National Laboratory for this work. We gratefully acknowledge the support of the U.S. Department of Energy through the Los Alamos National Laboratory LDRD Program, the Oak Ridge National Laboratory LDRD Program, and the Pacific Northwest National Laboratory LDRD Program for this work. We gratefully acknowledge the support of the South Dakota Board of Regents Competitive Research Grant. We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada, funding reference number SAPIN-2017-00023, and from the Canada Foundation for Innovation John R. Evans Leaders Fund. We acknowledge support from the 2020/2021 L’Oréal-UNESCO for Women in Science Programme. This research used resources provided by the Oak Ridge Leadership Computing Facility at Oak Ridge National Laboratory and by the National Energy Research Scientific Computing Center, a U.S. Department of Energy Office of Science User Facility. We thank our hosts and colleagues at the Sanford Underground Research Facility for their support.

Efficiency distribution prior derivation

The choice of prior, P(λij)=1/λij1−1/SiP(\lambda_{ij}) = 1/\lambda_{ij}^{1-1/S_i}, becomes apparent here when all nijn_{ij}, BjB_j, NijN_{ij} are equal for a given ii in (5). Setting nij=njn_{ij}=n_j and dropping the ii indices, the sum to be evaluated is ∑jSP(λj∣n)\sum_j^S P(\lambda_j|n). The sum of PDFs is given by their convolution. Defining the convolution ff with itself S−1S-1 times as

f{S}=f∗(f∗(f∗⋯∗(f∗f)))⏟S−1 convolutions,(10)f^{\{S\}} = \underbrace{f * (f * (f * \cdots* (f * f)))}_{S-1\ \text{convolutions}}, \tag*{(10)}

and rewriting P(λj∣n)P(\lambda_j|n) as f(λj)f(\lambda_j), an unknown function to be solved for, the sum becomes the following convolution:

∑jSf(λj)=f{S}(λj)(11)\sum_{j}^{S} f(\lambda_j)=f^{\{S\}}(\lambda_j) \tag*{(11)}

If the decay chain was simulated as a whole and not by segment, P(λ∣n)P(\lambda|n) would be given by Eq. (8) but with S=1S=1 and n=Snjn=Sn_j. The functional form of f(λj)f(\lambda_j) must be such that the sum of the PDFs of the segments equals the PDF of the full decay chain:

f{S}(λj)=λSnje−λ(12)f^{\{S\}}(\lambda_j)=\lambda^{Sn_j}e^{-\lambda} \tag*{(12)}

Taking the Laplace transform and working simultaneously on both sides yields:

L[f{S}(λj)](t)=L[λSnje−λ](t){L[f(λj)](t)}S=(Snj)!(t+1)Snj+1f(λj)=(Snj)!S L−1[1(t+1)nj+1/S](λj)=(Snj)!Sλjnj−1+1/Se−λjΓ(nj+1/S)(13)\begin{aligned} \mathcal{L}\left[f^{\{S\}}(\lambda_j)\right](t)&=\mathcal{L}\left[\lambda^{Sn_j}e^{-\lambda}\right](t)\\ \left\{\mathcal{L}\left[f(\lambda_j)\right](t)\right\}^{S}&=\frac{(Sn_j)!}{(t+1)^{Sn_j+1}}\\ f(\lambda_j)&=\sqrt[S]{(Sn_j)!}\,\mathcal{L}^{-1}\left[\frac{1}{(t+1)^{n_j+1/S}}\right](\lambda_j)\\ &=\sqrt[S]{(Sn_j)!}\frac{\lambda_j^{n_j-1+1/S}e^{-\lambda_j}}{\Gamma(n_j+1/S)} \tag*{(13)} \end{aligned}

Eq. A4 has the same functional form as Eq. (8) up to a constant. Therefore when nijn_{ij}, BjB_j, NijN_{ij} are equal for a given ii, the prior must take the form P(λj)=1/λj1−1/SP(\lambda_j)=1/\lambda_j^{1-1/S}. The use of this prior is extended to all cases.

References

References

  1. [1]I. J. Arnquist et al. (MAJORANA Collaboration), Phys. Rev. Lett. 130, 062501 (2023).
  2. [2]M. Agostini et al. (GERDA Collaboration), Phys. Rev. Lett. 125, 252502 (2020).
  3. [3]S. Abe et al. (KamLAND-Zen Collaboration), Phys. Rev. Lett. 130, 051801 (2023).
  4. [4]G. Anton et al. (EXO-200 Collaboration), Phys. Rev. Lett. 123, 161802 (2019).
  5. [5]D. Q. Adams et al. (CUORE Collaboration), Nature 604, 53 (2022).
  6. [6]N. Abgrall et al. (LEGEND collaboration), AIP Conf. Proc. 1894, 020027 (2017), arXiv:1709.01980 [physics.ins-det].DOI
  7. [7]G. Adhikari et al. (nEXO Collaboration), Journal of Physics G: Nuclear and Particle Physics 49, 015104 (2021).
  8. [8]K. Alfonso et al. (CUPID Collaboration), Journal of Low Temperature Physics 211, 10.1007/s10909-022-02909-3 (2022).
  9. [9]F. T. Avignone, G. S. King, and Y. G. Zdesenko, New Journal of Physics 7, 6 (2005).
  10. [10]F. T. Avignone, S. R. Elliott, and J. Engel, Rev. Mod. Phys. 80, 481 (2008).
  11. [11]N. Abgrall et al. (MAJORANA), Nucl. Instrum. Meth. A 877, 314 (2018), arXiv:1707.06255 [physics.ins-det].DOI
  12. [12]F. T. Avignone III and S. R. Elliott, Frontiers in Physics 7, 6 (2019).
  13. [13]S. Rahaman, V.-V. Elomaa, T. Eronen, J. Hakala, A. Jokinen, J. Julin, A. Kankainen, A. Saastamoinen, J. Suhonen, C. Weber, and J. Äystö, Physics Letters B 662, 111 (2008).
  14. [14]G. Douysset, T. Frioff, C. Carlberg, I. Bergström, and M. Björkåge, Phys. Rev. Lett. 86, 4259 (2001).
  15. [15]N. Abgrall et al. (MAJORANA Collaboration), Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 828, 22 (2016).
  16. [16]C. Alduino et al. (Cuore Collaboration), The European Physical Journal C 77, 543 (2017).
  17. [17]J. B. Albert et al. (nEXO Collaboration), Phys. Rev. C 97, 065503 (2018).
  18. [18]R. H. M. Tsang, I. J. Arnquist, E. W. Hoppe, J. L. Orrell, and R. Saldanha, (2018), arXiv:1808.05307 [physics.data-an].arxiv.org/abs/1808.05307
  19. [19]N. Abgrall et al. (MAJORANA Collaboration), Advances in High Energy Physics 2014, 365432 (2014).
  20. [20]J. Heise, Journal of Physics: Conference Series 606, 012015 (2015).
  21. [21]E. Hoppe, C. Aalseth, O. Farmer, T. Hossbach, M. Liebers, H. Miley, N. Overman, and J. Reeves, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 764, 116 (2014).
  22. [22]W. Bugy, V. Efremenkо, and S. Vasilyev, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 758, 91 (2014).
  23. [23]N. Abgrall et al. (MAJORANA Collaboration), Journal of Instrumentation 17 (05), T05003.
  24. [24]N. Abgrall et al. (MAJORANA Collaboration), Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 823, 83 (2016).
  25. [25]M. Boswell et al., IEEE Transactions on Nuclear Science 58, 1212 (2011).
  26. [26]S. I. Alvis et al. (MAJORANA Collaboration), Phys. Rev. C 100, 025501 (2019).
  27. [27]T. F. Gillis, ProQuest Dissertations and Theses , 165 (2019).
  28. [28]M. Agostini et al., Limits on uranium and thorium bulk content in gerda phase i detectors, Astroparticle Physics 91, 15 (2017).
  29. [29]R. L. Workman et al. (Particle Data Group), PTEP 2022, 083C01 (2022).

Paper details

Contents