betaFeedback

WorksPhysics and Astronomy

New measurement of double beta decays of 100^{100}Mo to excited states of 100^{100}Ru with the CUPID-Mo experiment

Abstract

The CUPID-Mo experiment, located at Laboratoire Souterrain de Modane (France), was a demonstrator experiment for CUPID. It consisted of an array of 20 Li2100_2^{100}MoO4_4 (LMO) calorimeters each equipped with a Ge light detector (LD) for particle identification. In this work, we present the result of a search for two-neutrino and neutrinoless double beta decays of 100^{100}Mo to the first 0+^+ and 2+2^+ excited states of 100^{100}Ru using the full CUPID-Mo exposure (2.71 kg×\timesyr of LMO). We measure the half-life of 2νββ2νββ decay to the 01+0^{+}_1 state as T1/22ν→01+=7.5±0.8 (stat.) −0.3+0.4 (syst.))×1020 yrT_{1/2}^{2ν\rightarrow 0_1^+}=7.5\pm 0.8 \ \text{(stat.)} \ ^{+ 0.4}_{-0.3} \ \text{(syst.)} )\times 10^{20} \ \mathrm{yr}. The bolometric technique enables measurement of the electron energies as well as the gamma rays from nuclear de-excitation and this allows us to set new limits on the two-neutrino decay to the 21+2_1^+ state of T^{2ν\rightarrow 2_1^+}_{1/2}>4.4\times 10^{21} \ \mathrm{yr} \ \text{(90 % c.i.)} and on the neutrinoless modes of T_{1/2}^{0ν\rightarrow 2_1^+}>2.1\times10^{23} \ \mathrm{yr}\ \text{(90 % c.i.)}, T_{1/2}^{0ν\rightarrow 0_1^+}>1.2\times10^{23} \ \mathrm{yr}\ \text{(90 % c.i.)}. Information on the electrons spectral shape is obtained which allows us to make the first comparison of the single state (SSD) and higher state (HSD) 2νββ2νββ decay models for the 01+0_1^+ excited state of 100^{100}Ru.

I Introduction

Two-neutrino double beta decay (2νββ2\nu\beta\beta) is an allowed Standard Model process which occurs in some even-even nuclei for which single beta decays are energetically forbidden or heavily disfavored due to large changes in angular momentum [1, 2]. These decays have been observed to the 0+0^+ ground states (0g.s.+0^+_{\mathrm{g.s.}}) in eleven nuclei and to the first zero-plus excited state (01+0^+_1) for two nuclei, 100Mo{}^{100}\mathrm{Mo} and 150Nd{}^{150}\mathrm{Nd} [1–3]. In addition, if neutrinos are Majorana particles then an additional decay mode becomes allowed: neutrinoless double beta decay (0νββ0\nu\beta\beta) [4–6]. Whereas 2νββ2\nu\beta\beta decay conserves lepton number, 0νββ0\nu\beta\beta decay would violate this symmetry [4, 7, 8] and provide a possible path to explain the predominance of matter over antimatter in the universe [4, 8–11].

One of the most promising experimental techniques to search for 0νββ0\nu\beta\beta decay are cryogenic calorimeters. These detectors provide an excellent energy resolution, high detection efficiency and are scalable to tonne scale arrays, such as the CUORE experiment (Cryogenic Underground Observatory for Rare Events) [12]. The CUPID experiment [13] (CUORE Upgrade with Particle IDentification) is an upgrade for CUORE which will use a scintillating bolometer technology to discriminate β/γ\beta/\gamma and α\alpha particles. CUPID-Mo [14] was a demonstrator experiment for CUPID aiming to validate the technology of lithium molybdate (LMO) scintillating calorimeters.

Multiple models exist to describe 0νββ0\nu\beta\beta decay [15, 16, 17, 18, 19, 20, 21] [7, 8], however the minimal extension to the Standard Model needed to explain 0νββ0\nu\beta\beta decay is the light Majorana neutrino exchange mechanism [22], in which 0νββ0\nu\beta\beta decay is mediated by a Majorana neutrino. In this framework, the decay rate of 0νββ0\nu\beta\beta decay is related to the effective Majorana neutrino mass ⟨mββ⟩\langle m_{\beta\beta}\rangle (a weighted sum of the three neutrino masses) by:

(T1/20ν)−1=G0νgA4∣M0ν∣2⟨mββ⟩2me2,(1)\left(T^{0\nu}_{1/2}\right)^{-1}=G_{0\nu}g_A^4\lvert M_{0\nu}\rvert^2\frac{\langle m_{\beta\beta}\rangle^2}{m_e^2}, \tag*{(1)}

where T1/20νT^{0\nu}_{1/2} is the 0νββ0\nu\beta\beta decay half-life, G0νG_{0\nu} is the phase space factor, M0νM_{0\nu} is the nuclear matrix element (NME), mem_e is the electron mass and gAg_A is the axial vector coupling constant. The phase space factor G0νG_{0\nu} can be computed accurately, but the NME is the result of complicated many-body nuclear physics calculations [23]. Several models with different approximations are used, but these only agree to within a factor of a few on the value of the NME (cf. [24, 25, 26, 27]). At present the upper limit on the effective Majorana mass, ⟨mββ⟩\langle m_{\beta\beta}\rangle, ranges from 60 to 600 meV [12, 28, 29, 30] [8]. Despite its low exposure, CUPID-Mo has set the most stringent limit on 0νββ0\nu\beta\beta decay in 100Mo{}^{100}\mathrm{Mo}, with T1/20ν>1.8×1024T^{0\nu}_{1/2}>1.8\times10^{24} yr or ⟨mββ⟩<(0.28−0.49)\langle m_{\beta\beta}\rangle<(0.28-0.49) eV [31].

For 2νββ2\nu\beta\beta decay, the half-life is related to the NME as:

(T1/22ν)−1=G2ν∣M2νeff∣2,(2)\left(T^{2\nu}_{1/2}\right)^{-1}=G_{2\nu}\lvert M^{\mathrm{eff}}_{2\nu}\rvert^2, \tag*{(2)}

where G2νG_{2\nu} is the phase space factor and and M2νeffM^{\mathrm{eff}}_{2\nu} is the dimensionless effective nuclear matrix element including the factor gA2g_A^2, M2νeff=gA2×M2νM^{\mathrm{eff}}_{2\nu}=g_A^2\times M_{2\nu} [2]. Measurements of T1/22νT^{2\nu}_{1/2} for both ground states (g.s.) and excited states (e.s.) of a daughter nucleus provide experimental data which can be used to validate the methods to calculate the NME M2νM_{2\nu}, and by extension, provide better confidence in the calculation of M0νM_{0\nu}.

Whilst the phase space, and therefore decay probability, is lower for decays to e.s. compared to transitions to the g.s., the presence of monochromatic γ\gamma’s can lead to a very clear detection signature. In the case of a positive observation of 0νββ0\nu\beta\beta decay, measurements of the 0νββ0\nu\beta\beta decay to e.s. will be a useful tool to understand the decay mechanism [3]. Additionally, whilst the 0νββ0\nu\beta\beta decay to 01+0^+_1 state is expected to be suppressed relative to the decay to 0g.s.+0^+_{\mathrm{g.s.}} state, it can still improve the overall sensitivity to ⟨mββ⟩\langle m_{\beta\beta}\rangle. Finally, a 2νββ2\nu\beta\beta decay to 21+2^+_1 e.s. is expected to be strongly suppressed by angular momentum conservation. However, this mode is expected to be much more likely in a framework with Bosonic neutrinos [32, 33] and can be used to test this model.

We can exploit the specific combination of energy depositions associated with these events to measure ββ\beta\beta decays to e.s. As shown in Figure 1, the two β\beta’s are accompanied by one or more γ\gamma particles. These have a much longer range (∼10\sim10 cm [34] for 1 MeV γ\gamma) compared to β\beta (∼2\sim2 mm [35]) in LMO. A ββ\beta\beta e.s. event therefore has a high probability to induce a multi-detector signature with the γ\gammas escaping the detector where the decay took place and being measured in neighbouring detectors. The additional information provided by these multi-detector events allows us to significantly reduce the background rate while the γ\gamma peaks provide a clean signature to robustly measure the decay rate [36].

Transition scheme for the decay of 100Mo to 100Ru

Figure 1. Transition scheme for the decay of 100Mo{}^{100}\mathrm{Mo} to 100Ru{}^{100}\mathrm{Ru} showing the double beta decays to the 01+0^+_1 (1130.3 keV) and 21+2^+_1 (539.5 keV) states. The decay to the 01+0^+_1 state proceeds via the 21+2^+_1 state with ∼100%\sim100\% probability, so both a 539.5 keV and 590.8 keV γ\gamma are emitted in every decay, while only the 539.5 keV γ\gamma is emitted in the double beta decay to the 21+2^+_1 state.

Since decays to higher e.s. are generally disfavored by the lower phase space due to lower Q-value (and also by angular momentum conservation for 2+2^{+} decays) we focus on the lowest energy e.s.: the first zero plus (01+0^{+}_{1}) and two plus (21+2^{+}_{1}) e.s. of 100^{100}Ru at 1130.3 keV and 539.5 keV (see Figure 1).

The 2νββ2\nu\beta\beta decay of 100^{100}Mo to 01+0^{+}_{1} e.s. has been measured in several experiments [37, 38, 39, 40, 41, 42], mostly using high purity Germanium (HPGe) counters (cf. [43] [2, 3] for a recent review) with an external Mo (enriched) source and also NEMO-3 using a tracking detector and external sources. For the HPGe measurements given that the source and detector did not coincide, only γ\gamma’s were detected. The 2νββ2\nu\beta\beta decay to 21+2^{+}_{1} e.s. has not been observed and only lower limits on the half-life are available [39, 42]. Placing constraints on this decay is complicated by the overlap with the decays to the 01+0^{+}_{1} state since both decays include a 539.5 keV γ\gamma emission. By using a setup where the source is embedded in the detector, such as cryogenic calorimeters, we are able to measure the electron energies as well as the γ\gamma’s. This provides additional information to separate the 2νββ2\nu\beta\beta decays to the 21+2^{+}_{1} and 01+0^{+}_{1} e.s., and to validate that the decay is indeed 2νββ→01+2\nu\beta\beta\to0^{+}_{1} and not a background process. This type of detector also allows us to effectively distinguish the neutrinoless and two-neutrino decay modes, which is not possible with HPGe detectors using an external ββ\beta\beta source.

In this work, we present an analysis of 2νββ/0νββ→01+/21+2\nu\beta\beta/0\nu\beta\beta\to0^{+}_{1}/2^{+}_{1} e.s. which exploits the measurement of both β\beta and γ\gamma energies. In Section II we describe the CUPID-Mo experiment; we introduce the search technique and Geant4 Monte Carlo simulations in Section III. In Section IV, an overview of the experimental data and treatment of multi-site events used for this analysis are given. We describe our Bayesian analysis framework in Section V, our treatment of systematic uncertainties in Section VI, and the obtained results in Section VII. We conclude in Section VIII by comparing our measurement of 2νββ→01+2\nu\beta\beta\to0^{+}_{1} state to theoretical calculations and we discuss the prospects to further investigate the decay mechanism as well as sensitivity to models beyond the light Majorana neutrino exchange.

Experiment

CUPID-Mo consisted of an array of twenty ∼200\sim200 g LMO detectors (Figure 2) enriched to ∼97\sim97 % in 100^{100}Mo, with a total mass of 4.16 kg of LMO [14] and 2.26 kg of 100^{100}Mo. These detectors were operated as cryogenic calorimeters at ∼20\sim20 mK, in the EDELWEISS [44] cryostat at the Laboratoire Souterrain de Modane (LSM), France. This technique allows for a very high detection efficiency (∼75\sim75 % for 0νββ0\nu\beta\beta decay to g.s.) due to the detector containing the source, and excellent energy resolution (σ∼0.1\sigma\sim0.1%) at 3 MeV [14]. In addition, CUPID-Mo employed a dual readout with twenty cryogenic light detectors (LDs) consisting of Ge wafers also operated as calorimeters. These allow for discrimination between α\alpha’s and β/γ\beta/\gamma’s since LMO is a scintillating crystal and the amount of light produced for β/γ\beta/\gamma is about five times higher compared to that produced by α\alpha particles of the same energy [14, 45].

CUPID-Mo experimental geometry showing the CUPID-Mo towers and a single detector module

Figure 2. CUPID-Mo experimental geometry showing (Left) the CUPID-Mo towers mounted in the EDELWEISS cryostat and single detector module (Right) with LMO crystal, PTFE clamps, NTD thermistor, Si heater and Cu holder. The Ge LD is seen through the LMO crystal.

Each LMO crystal is assembled into an independent detector module with a Ge LD, NTD-Ge [46] thermistors on both LMO and LD, Si heater, copper holder, reflective foil (3M Vikuiti™, which guides the light to the LDs), and PTFE clamps (shown in Figure 2). While this design allows for a modular pre-assembly and is compatible with the operation of multiple payloads in the EDELWEISS cryogenic system, it results in a relatively high copper/LMO mass ratio of ∼1.5\sim1.5. The individual detector modules are arranged in an array of five towers (see Figure 2) so that each LMO detector faces two LDs (apart from those on the top floor which only have a lower light detector).

CUPID-Mo took data between 2019 and 2020, collecting a total exposure of 2.71 kg×\timesyr of LMO. The data-taking was organized into 9 datasets, periods of stable operations lasting around a month. As in [47], three short periods of data-taking which could not be calibrated to the same accuracy were discarded.

Search Technique and Simulations

In ββ\beta\beta decays to e.s. the electrons are accompanied by γ\gamma’s so they often reconstruct in multi-site or coincident events where multiple detectors are triggered simultaneously. The particular set of energies obtained allows us to make a pure selection of e.s. decay events with very low background. We define the multiplicity, M\mathcal{M}, as the number of crystals subject to simultaneous energy depositions above the analysis energy threshold, chosen as 40 keV, well above the detector trigger thresholds and within a time window of ±10\pm10 ms well above the detector time resolution [31].

For an event which is a time coincidence between several LMO detectors, our minimal experimental observable information is the energy deposited in each detector. We define energies E1,E2,E3,…,EME_1, E_2, E_3, \ldots, E_M as the energy deposited in each detector, sorted so that:

E1>E2>E3…>EM.(3)E_1 > E_2 > E_3 \ldots> E_M. \tag*{(3)}

In principle, the most sensitive approach exploiting all the information contained in the data, would be a fit directly to this multidimensional spectrum. However, this analysis has significant complexity due to the difficulty of quantifying the background shape in this high dimensional space.

An alternative approach that has been used by the CUORE-0, CUORE, and CUPID-0 experiments [48, 49, 50] is to select categories of signal signatures consistent with being from an e.s. decay. For each category only the energy variable containing the γ\gamma (or two electrons - ββ\beta\beta for 0νββ0\nu\beta\beta decay) peak is fitted (the peak energy EpE_{\mathrm{p}}). The other energy variables are referred to as “projected out” energies. This technique has the advantage that the decay rate can be extracted as a set of one dimensional fits to a peak over a background, which is a simple and robust technique. We use this approach in our analysis, however we modify this technique by dividing up categories based on their projected out energy variables. Since the decays to different e.s. significantly overlap this allows us to better discriminate between them. Furthermore, it avoids integrating together regions with different background levels.

MC simulations

We use dedicated Monte Carlo (MC) simulations to identify the most promising categories of events to include and optimize the associated energy cuts. The objective is to identify categories which possess a high signal to background ratio, to define the energy variable to project onto, to select the energy boundaries for each signal category, and to assess the containment efficiency therein. A Geant4 [51] model has been created which implements the CUPID-Mo detector into the existing EDELWEISS MC software package [52]. Special care is

taken to implement the detector structure as accurately as possible, in particular accounting for the individual dimensions of each LMO crystal, and the passive holder material close to the detectors. We simulate both 2νββ2\nu\beta\beta decays and 0νββ0\nu\beta\beta decays to the 01+0^{+}_{1} and 21+2^{+}_{1} e.s.. This accounts for the angular correlation between 540 keV and 591 keV γ\gamma’s from the 01+0^{+}_{1} (01+−21+−0g.s.+0^{+}_{1}-2^{+}_{1}-0^{+}_{\mathrm{g.s.}}) cascade as [53]:

ρ(θ)=1−3cos⁡2(θ)+4cos⁡4(θ).(4)\rho(\theta) = 1 - 3\cos^{2}(\theta) + 4\cos^{4}(\theta). \tag*{(4)}

Here ρ(θ)\rho(\theta) is the probability for γ\gamma’s to be emitted at angle θ\theta from each other. Particles are then propagated through the experimental geometry using the Livermore low energy physics list [54], applicable down to very low energy (250 eV). For the 2νββ2\nu\beta\beta decay to 01+0^{+}_{1} e.s. simulations are performed using both the single state dominance (SSD) and higher state dominance (HSD) models [55]. Currently, no experimental data favors one model over another for the 01+0^{+}_{1} state, however the SSD model is strongly favored for the ground state decay [56, 57] and so we use it as our default. The difference to the HSD model is then treated as a systematic uncertainty. We combine the MC simulations with inputs from experimental data to reproduce the detector response of each individual detector, accounting for:

  • The energy threshold (set at 40 keV) [31];

  • The multiplicity of events;

  • Energy resolution of each detector-dataset pair (energy dependent);

  • Rejected periods of detector operations due to periods of cryogenic instability (i.e. due to environmental disturbances);

  • Scintillation light, light detector energy resolution, and cuts based on the LDs (see Section IV);

In our MC simulations, we observe that most excited states events are in M1−2\mathcal{M}_{1-2} with a small contribution of M3\mathcal{M}_{3} (Figure 3). The large fraction of M1\mathcal{M}_{1} events is caused by the significant volume of passive material (Cu housing) and a relatively small modular detector array. We exclude M1\mathcal{M}_{1} 2νββ2\nu\beta\beta decay events as the continuous energy spectrum of the β\beta’s does not produce a peak to perform a fit. For 0νββ0\nu\beta\beta decays, there are mono-energetic peaks in M1\mathcal{M}_{1}, albeit with a high background. We therefore focus on M2\mathcal{M}_{2} events which have the highest sensitivity, and we also include some categories of M3\mathcal{M}_{3} events.

Figure 3. Distribution of multiplicities (MC simulation) for each e.s. decay. Due to the small crystal array and significant amount of non-sensitive holder material, both decay modes to the (01+0^{+}_{1}, 21+2^{+}_{1}) state are dominated by single crystals hits, (10–20%) of events cause energy deposits in two crystals, and O(1%)O(1\%) in three or more.

2νββ decay categories

We first describe categories for 2νββ2\nu\beta\beta decay analysis. For the analysis of M2\mathcal{M}_{2} events the energies can be visualized as two-dimensional histograms. We plot these two-dimensional energy spectra for both e.s. decays (2νββ2\nu\beta\beta and 0νββ0\nu\beta\beta decay) to 01+0^{+}_{1} and 21+2^{+}_{1} e.s. in Figure 4. When selecting categories of events to use in our analysis, we require at least one peak to perform a fit. This selection is equivalent to looking for lines or points on the two-dimensional plane.

Figure 4. Monte Carlo simulated two-dimensional M2\mathcal{M}_2 energy distributions for decays to e.s. Top left: 2νββ2\nu\beta\beta decay to 01+0^+_1. Top Right: 2νββ2\nu\beta\beta decay to 21+2^+_1. Bottom Left: 0νββ0\nu\beta\beta decay to 01+0^+_1. Bottom Right: 0νββ0\nu\beta\beta decay to 21+2^+_1. We show red boxes to highlight the M2\mathcal{M}_2 signal categories used in the analysis (see Tables I, II) and we label the associated category numbers.

For 2νββ2\nu\beta\beta decays in MC simulation we observe a very clear signature of events where either E1E_{1} or E2E_{2} has a γ\gamma energy (either 539.5 or 590.8 keV) as shown in Figure 4 (top). These events can be interpreted as the ββ\beta\beta being contained in one detector, giving rise to a continuous distribution, and a γ\gamma that is fully contained in a second LMO detector. In the case of the decay to the 01+0^{+}_{1} state, the second γ\gamma either escapes the detector or is contained in the crystal with the ββ\beta\beta’s. We notice that the 21+2^{+}_{1} and 01+0^{+}_{1} decays (partially) overlap and the vertical/horizontal lines cover a wide range of projected out (ββ\beta\beta) energies. This motivates us to divide the observed signal distribution into five slices, two for the horizontal and three for the vertical lines since the vertical lines cover a larger range in energy. The boundaries of this categories are defined to maximise the experimental sensitivity using MC simulations as explained in Appendix A. This leads to the energy categories highlighted in Figure 4 (top) and listed in Table I (index 0 to 4).

To identify M3\mathcal{M}_{3} categories we note that in MC simulations generally the electrons carry the largest kinetic energy and are thus likely to be contained in the E1E_{1} detector due to the relatively low γ\gamma energies. We hence focus our search on events with γ\gamma’s in the E2E_{2} and E3E_{3} variables. In particular, this leads to two categories, one where the γ\gamma energy is split between E2,E3E_{2}, E_{3} and another where one γ\gamma is fully contained in E2E_{2} with a Compton scatter event in E3E_{3}.

Combined with the M2\mathcal{M}_{2} categories this leads to a total of 7 independent categories used for the 2νββ2\nu\beta\beta decay analysis.

0νββ decay categories

Unlike for 2νββ2\nu\beta\beta decay, in 0νββ0\nu\beta\beta decay the electrons have a fixed total energy so we are also able to search for a peak from the summed electron energy. Therefore a different set of categories are necessary, however the same strategy can be used. For M1\mathcal{M}_{1} events we include three categories corresponding to the ββ\beta\beta and possibly γ\gamma’s being contained in the detector where the decay took place. For M2\mathcal{M}_{2} events, we show the two-dimensional distributions in Figure 4 (bottom two figures). We observe a variety of features, vertical, horizontal diagonal lines and points. Each of these require a different set of energy cuts to contain the signal, as shown in Table II. These cuts are carefully constructed to ensure that none of the categories overlap. Some categories feature a mono-energetic peak in both energy variables, for these we still choose one as the peak variable and use a range of ±10 keV for the other, projected energy, much larger than the energy resolution. We employ a similar strategy to the 2νββ2\nu\beta\beta decay categories, dividing the horizontal lines, events with a ββ\beta\beta peak (1904, 2444, 2494 keV) in E1E_{1} and a Compton scatter in E2E_{2}, into 3 categories each to account for the large changes in background level. This is not necessary for other categories due to the low background, or a small range in the projected out energy. We neglect some possible categories which have very small containmeasurement efficiency (events with E2>600E_2 > 600 keV), that overlap significantly with 2νββ2\nu\beta\beta decay (events with E2∼500E_2 \sim500–650 keV, E1<1904E_1 < 1904 keV) or with 214^{214}Bi background, diagonal lines with energy summing to ∼2400\sim2400 keV. This leads to 10 M2\mathcal{M}_2 0νββ0\nu\beta\beta decay categories which are most easily interpreted geometrically (see the boxes in Figure 4).

Cat.MultiplicityPeak VariableEnergy cutsInterpretation
02E1E_1E2∈[220,410]E_2 \in[220,410]γ\gamma in E1E_1, ββ\beta\beta in E2E_2
12E1E_1E2∈[410,500]E_2 \in[410,500]γ\gamma in E1E_1, ββ\beta\beta in E2E_2
22E2E_2E1∈[500,890]E_1 \in[500,890]γ\gamma in E2E_2, ββ\beta\beta in E1E_1
32E2E_2E1∈[890,1190]E_1 \in[890,1190]γ\gamma in E2E_2, ββ\beta\beta in E1E_1
42E2E_2E1∈[1190,1880]E_1 \in[1190,1880]γ\gamma in E2E_2, ββ\beta\beta in E1E_1
53E2E_2-γ\gamma in E2E_2, ββ\beta\beta in E1E_1, Compt. in E3E_3
63E2+E3E_2+E_3E2<500E_2 < 500γ\gamma in E2+E3E_2+E_3 (shared Compt.), ββ\beta\beta in E1E_1

Table 1. Categories for 2νββ2\nu\beta\beta decay to 01+0^+_1, 21+2^+_1 e.s., these are given a name describing the multiplicity, peak energy variable and then an index (if necessary to distinguish between categories). The table lists the energy cuts for the variable that is projected out, while the projected energy range for each category is 500–650 keV. For every category the γ\gamma’s are either 540 or 591 keV.

focusing on those where the electrons are contained in E1E_1. We choose four categories: For events with E1∼1904E_1 \sim1904 keV, we split the data into E2∈[500,650]E_2 \in[500,650] keV, E2+E3∈[500,650]E_2+E_3 \in[500,650] keV or neither of these two. For events with E1∼2400E_1 \sim2400 keV the background is expected to be low so we just include one category. This procedure leads to 17 experimental signatures as shown in Table II.

As in 2νββ2\nu\beta\beta decay analysis we also consider M3\mathcal{M}_3 events

Cat.MultiplicityPeak VariableRange [keV]Energy cuts [keV]Interpretation
01E1E_{1}1864–1944ββ\beta\beta in E1E_{1}, γ\gamma escape
11E1E_{1}2384–2534ββ\beta\beta in E1E_{1}, γ\gamma escape
21E1E_{1}2984–3084ββ\beta\beta and γ\gamma in E1E_{1}
32E1E_{1}1864–1944E2<180E_{2}<180ββ\beta\beta in E1E_{1}, Compt. γ\gamma in E2E_{2}
42E1E_{1}1864–1944E2∈[180,310]E_{2}\in[180,310]ββ\beta\beta in E1E_{1}, Compt. γ\gamma in E2E_{2}
52E1E_{1}1864–1944E2∈[310,500]E_{2}\in[310,500]ββ\beta\beta in E1E_{1}, Compt. γ\gamma in E2E_{2}
62E1E_{1}2384–2534E2<180E_{2}<180ββ\beta\beta in E1E_{1}, Compt. γ\gamma in E2E_{2}
72E1E_{1}2384–2534E2∈[180,280]E_{2}\in[180,280]ββ\beta\beta in E1E_{1}, Compt. γ\gamma in E2E_{2}
82E1E_{1}2384–2534E2∈[280,500]E_{2}\in[280,500]ββ\beta\beta in E1E_{1}, Compt. γ\gamma in E2E_{2}
92E2E_{2}500–650E1∈[1894,2904]E_{1}\in[1894,2904]ββ\beta\beta in E1E_{1}, γ\gamma in E2E_{2}
102E2E_{2}500–650E1∈[1944,2384]E_{1}\in[1944,2384], E1+E2>2534E_{1}+E_{2}>2534ββ+\beta\beta+ Compt. E1E_{1}, γ\gamma in E2E_{2}
112E1+E2E_{1}+E_{2}2984–3084E1>2534E_{1}>2534, E2<500E_{2}<500ββ+\beta\beta+ Compt. E1E_{1}, Compt. E2E_{2}
122E1−E2E_{1}-E_{2}1734–2034E2<650E_{2}<650, E1>2384E_{1}>2384, E1+E2∈[3024,3044]E_{1}+E_{2}\in[3024,3044]ββ\beta\beta in E1E_{1}, γ\gamma in E2E_{2}
133E2+E3E_{2}+E_{3}500–650E1∈[1894,1914]E_{1}\in[1894,1914], E2<500E_{2}<500ββ\beta\beta E1E_{1}, Compt. E2/E3E_{2}/E_{3}
143E2E_{2}500–650E1∈[1894,1914]E_{1}\in[1894,1914]ββ\beta\beta in E1E_{1}, γ\gamma in E2E_{2}, Compt. E3E_{3}
153E1E_{1}1854–1954E2∉[500,650]E_{2}\notin[500,650], E2+E3∉[500,650]E_{2}+E_{3}\notin[500,650]ββ\beta\beta in E1E_{1} and Compt. E2/E3E_{2}/E_{3}
163E1E_{1}2384–2534ββ\beta\beta in E1E_{1}

Table 2. Categories for 0νββ0\nu\beta\beta to excited states, for each category we give an index, a name, the projected energy variable and its energy window, the cuts used to define the category and a physical interpretation. Most of these cuts are more easily visualized in Figure 4.

Determination of signal shape and efficiency

We use MC simulations to determine the signal shapes and efficiencies for our Bayesian analysis. For each category we make the energy cuts as described in Tables I, II. We also apply a set of selection cuts to the data and MC simulations to remove events that likely arise from a known γ\gamma-ray background peak. The cuts are chosen to correspond to roughly ±ΔE(FWHM)\pm\Delta E(\mathrm{FWHM}) with the values shown in Table III. After the projection onto the peak energy variable we extract one-dimensional histograms for each decay mode (01+0^{+}_{1} or 21+2^{+}_{1}) in each category from MC simulations.

VariableCut [keV]Decay
E1E_{1}1333±51333\pm560Co{}^{60}\mathrm{Co}
E1E_{1}1173±51173\pm560Co{}^{60}\mathrm{Co}
E1+E2E_{1}+E_{2}1333±51333\pm560Co{}^{60}\mathrm{Co}
E1+E2E_{1}+E_{2}1173±51173\pm560Co{}^{60}\mathrm{Co}
E1+E2E_{1}+E_{2}1461±51461\pm5
E1E_{1}2615±72615\pm7208Tl{}^{208}\mathrm{Tl}
E1+E2E_{1}+E_{2}2615±72615\pm7208Tl{}^{208}\mathrm{Tl}

Table 3. Table showing the γ\gamma exclusion cuts used to remove events likely originating from a background γ\gamma ray.

From these histograms we obtain the containment efficiency εcont,r0+/2+\varepsilon^{0^{+}/2^{+}}_{\mathrm{cont},r}, or the probability for a decay to reconstruct in category rr. We also compute the total containment efficiency (summed over all categories) as 4.4/2.0 % for 2νββ2\nu\beta\beta to 01+/21+0^{+}_{1}/2^{+}_{1} e.s. and 46/57 % respectively for 0νββ0\nu\beta\beta decay.

We fit these histograms to phenomenological functions to extract analytic signal shapes to use in our analysis. These functions are described in more detail in Appendix B, the photo-peak is approximated via three Gaussian’s to account for the slight non-Gaussianity of the line-shape (due to each detector-dataset pair having a different resolution), this is discussed more in Section IV A. Each function is normalized to unity and then we model the MC simulated spectrum as:

f(E)=Ntot∑i=1Npi⋅fi(E),(5)f(E)=N_{\mathrm{tot}}\sum_{i=1}^{N}p_{i}\cdot f_{i}(E), \tag*{(5)}

where the sum runs over the normalized functions fif_{i}, pip_{i} is the probability for an event to be in a given spectral feature ii of the signal distribution for the considered category, for example in one of the gamma peaks or in the continuum, and NtotN_{\mathrm{tot}} is the total number of events. We show two example fits (both 01+0^{+}_{1} signal) in Figure 5, 2νββ2\nu\beta\beta category 2 and Figure 6, 0νββ0\nu\beta\beta category 2. We see that these models describe the MC simulations very well, in particular the slightly non-Gaussian photo-peak shape. These functions encode most of our systematic uncertainties, in particular the resolution and peak position of Gaussian and normalization of the peaks which can then be treated as nuisance parameters of the final fit.

Figure 5. Parameterization of 2νββ2\nu\beta\beta to 01+0^{+}_{1} category 2 signal. The model consists of two photo-peaks, with their respective 100Mo{}^{100}\mathrm{Mo} X-ray escape peaks, two smeared step functions modeling Compton interactions and a linear background. The residual defined as (value − model)/error is shown in the upper figure.

Parameterization of $0\nu\beta\beta$ decay to $0^{+}_{1}$ category 2 signal

Figure 6. Parameterization of 0νββ0\nu\beta\beta decay to 01+0^{+}_{1} category 2 signal. The model contains a photo-peak, X-ray escape peak, and linear smeared step functions. The residual defined as (value − model)/error is shown in the upper figure.

Data analysis

For this search we use the full CUPID-Mo data corresponding to 2.71 kg×yr exposure of LMO. We process our data using a C++ software, Diana [58, 59], developed by the Cuoricino, CUORE, and CUPID-0 collaborations and further developed by CUPID-Mo. Most of the data processing steps are the same as described in [31, 47]. An optimal trigger is used to identify physics events and an optimal filter which maximizes signal-to-noise ratio is used to compute amplitudes (both LD and LMO detectors) [60]. Thermal gain stabilization and calibration are performed using data collected with a Th/U calibration source. For each LMO detector we associate ”side-channels”, as the LDs which face this crystal. For M1\mathcal{M}_{1} data (3 0νββ0\nu\beta\beta decay categories) we use the same selection cuts as in [31]: a principal component analysis (PCA) based pulse shape discrimination (PSD) cut [61], a normalized light distance cut, delayed coincidence (DC) cuts to remove 222Rn{}^{222}\mathrm{Rn}, 228Th{}^{228}\mathrm{Th} decay chain events, and muon veto anti-coincidence.

However, for coincidence events (M>1\mathcal{M}>1) it is beneficial to modify several of these steps. Random coincidences between non-causally related events (hereafter accidentals) could provide a large possible background. Similar to CUORE [58] and as described in [31] we adjust

just the time difference of events to account for the characteristic detector response based time offsets between detector pairs. This allows us to reduce the time window used to define coincidences to 10 ms (from 100 ms in previous analyses [47]) and thus reduce the accidental coincidence background.

In order to select a clean sample of higher multiplicity e.s. candidate events, we make use of the dual read-out and implement a light yield based cut for M>1\mathcal{M}>1 events. Unlike the M1\mathcal{M}_{1} analysis where this cut primarily rejects α\alpha backgrounds, it is designed here to tag and reject events where an energetic electron escapes an LMO crystal, punches through an LD, and is stopped in the adjacent crystal within a tower. In addition, this cut very efficiently tags events from a 60Co{}^{60}\mathrm{Co} contamination identified on one of the LDs where the β\beta particle typically leaves significant excess energy in the LD. Developing a light cut to remove these events is less straightforward than for M1\mathcal{M}_{1} data, since a coincidence between two vertically adjacent crystals is accompanied by scintillation light signal that can be absorbed in the common, intermediate LD. For each event we compute the expected scintillation light deposited on the LD based on the values observed in M1\mathcal{M}_{1} events (mainly 2νββ2\nu\beta\beta decays to g.s.). This accounts for these multiple contributions to light yield. We then normalize the light energies using a conservative estimate of the LD resolution (σi,sLD\sigma_{i,s}^{\mathrm{LD}}) measured in high energy background data (M2\mathcal{M}_{2} total energy >1500>1500 keV) as:

ni,s=Ei,sLD−Ei,sLD,expσi,sLD,(6)n_{i,s}=\frac{E_{i,s}^{\mathrm{LD}}-E_{i,s}^{\mathrm{LD,exp}}}{\sigma_{i,s}^{\mathrm{LD}}}, \tag*{(6)}

where ii refers to the pulse index (one for E1E_{1} etc.), ss is the side channel (either 0 or 1), Ei,sLDE_{i,s}^{\mathrm{LD}} is the measured LD energy, while Ei,sLD,expE_{i,s}^{\mathrm{LD,exp}} is the expected LD energy for γ/β\gamma/\beta like event and σi,sLD\sigma_{i,s}^{\mathrm{LD}} is the predicted energy resolution of the LD. We then place a cut of ∣ni,s∣<10|n_{i,s}|<10 for each LD and side channel in an M>1\mathcal{M}>1 event. We place a cut of −10<ni,s<3-10<n_{i,s}<3 for an LD with the 60Co{}^{60}\mathrm{Co} contamination. This contamination can also lead to background events in other LMO detectors; to remove this we make a global LD anti-coincidence cut. We remove any LMO event (excluding those who directly face this LD) with a trigger on this LD with LD energy >5>5 keV within a ±5\pm5 ms window.

We also adjust the PCA based PSD cuts [31, 47] to place a cut on the shape of each pulse in a higher multiplicity event. We require that each pulse has a normalized PCA reconstruction error of less than 23 median absolute deviations. This was optimized by comparing the efficiency of M2\mathcal{M}_{2} events εs\varepsilon_{\mathrm{s}}, obtained from events summing to the 40K{}^{40}\mathrm{K} 1461 keV peak and estimating the background efficiency, εb\varepsilon_{\mathrm{b}}, in a side-band E2∈(450−520)E_{2}\in(450-520) keV. This sideband, which is only used in optimizing the PCA cut, approximates the background from the dominant 2νββ2\nu\beta\beta decay categories (2, 3 and 4) whilst not including any signal peaks. We then maximize a figure of merit εs/εb\varepsilon_{\mathrm{s}}/\sqrt{\varepsilon_{\mathrm{b}}} which is proportional to the experimental sensitivity. For M1\mathcal{M}_1 data a cut of <9<9 on the PCA normalized reconstruction error is used (as in [31]).

Finally, we employ a data blinding by adding simulated MC events directly to the data files. The rate is chosen by sampling randomly from a uniform distribution with range, Γ∈Γmeasured⋅(2,10)\Gamma\in\Gamma_{\mathrm{measured}}\cdot(2,10), where Γmeasured\Gamma_{\mathrm{measured}} is the current leading limit or measurement [2] and (2,10)(2,10), is a uniform distribution between 2 and 10 chosen to ensure the injected signal is significantly larger than any possible signal in the data. The injected rate is hidden during the analysis of the blinded data. The blinded data are used to optimize and test the Bayesian fitting routine and prevent biasing our results.

Energy resolution and linearity

We determine the response of our detector to a monochromatic energy deposit using both calibration sources and γ\gamma peaks from natural radioactivity using the same procedure described in detail in [31]. In particular, a simultaneous fit, over each detector in each dataset, of the 2615 keV peak in calibration data is used to extract the resolution of each detector-dataset pair at 2615 keV σchan,dataset2615\sigma_{\mathrm{chan,dataset}}^{2615}.

We model the line-shape of our peaks in M1\mathcal{M}_1 physics data as an exposure weighted sum of Gaussians for each channel-dataset pair. The individual Gaussians have a common mean μ\mu but a resolution which is a product of σchan,dataset2615\sigma_{\mathrm{chan,dataset}}^{2615} and a global scaling R(E)R(E). We fit this line-shape model to our peaks to extract R(E)R(E) and μ(E)\mu(E) for each peak in physics data and parameterize the energy dependence as R(E)=r02+r1ER(E)=\sqrt{r_0^2+r_1E} and μ(E)=b0+b1E+b2E2\mu(E)=b_0+b_1E+b_2E^2. The resulting parametrization is used in the MC simulations to model the detector response. In particular, we account for the uncertainty on the parameters of these functions as systematic uncertainties (Section VI).

Cut efficiencies

To measure 2νββ/0νββ2\nu\beta\beta/0\nu\beta\beta decay rates it is necessary to determine the analysis efficiency, or probability that a signal event will pass all selection cuts. We employ the same strategy that was utilised in the M1\mathcal{M}_1 0νββ0\nu\beta\beta analysis [31] in order to compute these. We evaluate the pileup efficiency, or probability that a signal event will not have another trigger in the same waveform using random triggers. Several other cuts (multiplicity selection, muon veto cut, and LD anti-coincidence) induce dead times in one or more detectors. For these we evaluate the efficiencies using 210Po{}^{210}\mathrm{Po} Q-value peak events by counting the number of events passing the cuts with energy in (5407±50)(5407\pm50) keV. These are a proxy for physical M1\mathcal{M}_1 events due to the high energy and modular structure meaning α\alpha particles can only deposit energy in one crystal.

Finally, for our pulse shape and light yield cuts the efficiency is evaluated using γ\gamma peaks in M1\mathcal{M}_1 and summed M2\mathcal{M}_2 energy which are a clean sample of signal like events. For each prominent γ\gamma peak, we fit the energy distribution of events passing and failing each cut to a Gaussian plus linear background. From the measured number of events NpassN_{\mathrm{pass}}, NfailN_{\mathrm{fail}} (events in the Gaussian peak only for the two fits) we compute the efficiency ϵ=Npass/(Npass+Nfail)\epsilon=N_{\mathrm{pass}}/(N_{\mathrm{pass}}+N_{\mathrm{fail}}). We estimate numerically the uncertainty on ϵ\epsilon by sampling from the measured uncertainties on NpassN_{\mathrm{pass}}, NfailN_{\mathrm{fail}}. We observe an energy independent efficiency for all of our cuts (shown in col. 2 of Table IV). The energy range for the M1\mathcal{M}_1 3034 keV category of 0νββ0\nu\beta\beta decay lies outside the range of our γ\gamma peaks. Similar to [31] we use a 1st order polynomial to extrapolate the PCA and light yield cut efficiency to 3034 keV, accounting for the possibility of a slight energy dependence of the cut (shown in col. 3 of Table IV). We summarise the cut efficiencies in Table IV, and use them as nuisance parameters with Gaussian priors in our Bayesian analysis.

Bayesian analysis

We perform a Bayesian analysis to extract decay rates. In particular, we use an extended unbinned maximum likelihood fit implemented using the Markov Chain Monte Carlo (MCMC) of the Bayesian Analysis Toolkit (BAT) [62] for all categories except for two 0νββ0\nu\beta\beta decay categories, for which we use a binned fit due to their exceptionally high statistics. We use a fit of the summed data of all 19 detectors and 9 datasets.

Two separate fits are created, one to extract the decay rates of 2νββ2\nu\beta\beta decay and another for the 0νββ0\nu\beta\beta decay. Both fits are implemented in the same framework. In each category (index rr) we model our experimental data as:

fr(E)=ϵa,rn2ν+f2+,r(E)+ϵa,rn0ν+f0+,r(E)f_r(E)=\epsilon_{a,r}n_{2\nu+}f_{2+,r}(E)+\epsilon_{a,r}n_{0\nu+}f_{0+,r}(E)
+nr,bkg⋅fr,bkg(E)+∑p=15nr,pfp(E)(7){}+n_{r,\mathrm{bkg}}\cdot f_{r,\mathrm{bkg}}(E)+\sum_{p=1}^{5}n_{r,p}f_p(E) \tag*{(7)}

where

  • n0+/2+n_{0+/2+} are the observed number of counts to 01+/21+0_1^+/2_1^+ excited states.

  • f0+/2+,r(E)f_{0+/2+,r}(E) are the phenomenological functions describing the signal shape of category rr, normalized to unity over the sum of all categories.

  • ϵa,r\epsilon_{a,r} is the analysis efficiency for category rr (described in Section VI).

  • fr,bkg(E)f_{r,\mathrm{bkg}}(E) is a function describing the background in category rr, either exponential or flat for categories with low statistics.

  • nr,pn_{r,p} are the number of background counts from known γ\gamma lines (a subset of 511, 583, 609, 2448 and 2505 keV) with index pp and fp(E)f_p(E) is the model of this spectral shape, a single Gaussian.

We show in Figure 7 the fit of the 2νββ2\nu\beta\beta decay category 3 and the contribution from the exponential background, signal and background peaks. Our likelihood function is:

Figure 7. Fit of the 2νββ2\nu\beta\beta decay category 3. The experimental data are binned for visualization and we show the best fit reproduction of our model and the contributions from the signal peaks, background peaks and the exponential background.

log⁡(L)=∑n=1Nlog⁡(fr(n)(En))+∑r=1C−λr+nrlog⁡(λr)−log⁡(nr!)+∑b=1B∑i=1Nbin−λi+ni,b⋅log⁡(λi,b)−log⁡(ni,b!)(8)\log(\mathcal{L}) = \sum_{n=1}^{N} \log\left(f_{r(n)}(E_n)\right) + \sum_{r=1}^{C} -\lambda_r + n_r\log(\lambda_r) - \log(n_r!) + \sum_{b=1}^{B}\sum_{i=1}^{N_{\mathrm{bin}}} -\lambda_i + n_{i,b}\cdot\log(\lambda_{i,b}) - \log(n_{i,b}!) \tag*{(8)}

where the first sum is over the events in the experimental data (unbinned categories), r(n)r(n) is the category of event nn, CC is the number of categories, λr\lambda_r is the total predicted number of events in category rr, while nrn_r is the observed number of events. The last sums are a binned likelihood where the sum bb runs over the BB categories which use a binned likelihood. These are the categories 0 and 1 for the 0νββ0\nu\beta\beta decay where a large number of events make an unbinned fit computationally unfeasible. λi\lambda_i is the expectation value for bin ii and ni,bn_{i,b} is the number of events in experimental data. We use 0.5 keV bins for these fits, much smaller than the energy resolution over the full spectral range.

We use BAT to sample the full posterior distribution of the parameters of interest θ⃗=(n01+,n21+)\vec{\theta} = (n_{0_1^+}, n_{2_1^+}) and nuisance parameters ν⃗\vec{\nu}, which include all of the background components (both the exponential/flat background and the γ\gamma peaks) and also parameters of systematic uncertainties (see § 1)

p(θ⃗,ν⃗∣D)=P(D∣θ⃗,ν⃗)P(θ⃗,ν⃗)P(D)(9)p(\vec{\theta},\vec{\nu}\mid\mathcal{D}) = \frac{P(\mathcal{D}\mid\vec{\theta},\vec{\nu})P(\vec{\theta},\vec{\nu})}{P(\mathcal{D})} \tag*{(9)}

where D\mathcal{D} represents the data. We use uninformative (flat) priors on all background model components and parameters of interest n2+/0+n_{2^+/0^+}. We define observables for the decay rates of 01+/21+0_1^+/2_1^+ decay (Γ0+/2+\Gamma_{0^+/2^+})

Γ0+/2+=n0+/2+⋅Wϵcont,0+/2+⋅η⋅Mt⋅Na⋅1000(10)\Gamma_{0^+/2^+} = \frac{n_{0^+/2^+}\cdot W}{\epsilon_{\mathrm{cont},0^+/2^+}\cdot\eta\cdot Mt\cdot N_a\cdot1000} \tag*{(10)}

where WW is the molecular mass for enriched Li2100MoO4\mathrm{Li}_2{}^{100}\mathrm{MoO}_4, NaN_a is the Avogadro constant, MtMt is the exposure (in kg ×\times yr), η\eta is the isotopic abundance of 100Mo{}^{100}\mathrm{Mo} and ϵcont,0+/2+\epsilon_{\mathrm{cont},0^+/2^+} is the containment efficiency for 01+/21+0_1^+/2_1^+ signal, i.e. the total probability a simulated event is in one of the categories. The half-life is then given by ln⁡(2)/Γ\ln(2)/\Gamma. For each step of the Markov chains the values of Γ0+/2+\Gamma_{0^+/2^+} are stored and used to compute the marginalized posterior distributions. We include all of our systematic uncertainties as nuisance parameters in our analysis as described in § 1.

Systematic uncertainties

Energy resolution and bias

We account for the uncertainty in the energy resolution of our peaks. Both the terms r1r_1 and r0r_0 in the resolution scaling function R(E)R(E), see Section IV A, are given Gaussian priors based on the measured uncertainty from the fit to γ\gamma peak resolution. A multivariate prior is not necessary since the correlation is very low (−0.2%-0.2\%) because the r0r_0 term is fixed very well by noise events (random triggers used to estimate energy resolution at 0 keV). At each stage in the Markov chain this resolution is applied to all the signal components and background peaks, by adjusting the resolution of the functions f2+/0+,r(E)f_{2^+/0^+,r}(E), fr,bkg(E)f_{r,\mathrm{bkg}}(E).

We also account for the uncertainty in the energy scale. Our energy bias is parameterized as a second order polynomial as described earlier in Section IV A. We adjust the mean position of all peaks in our signal and background template functions (f0+/2+f_{0^+/2^+}, fbkgf_{\mathrm{bkg}}) by this bias. We add a nuisance parameter to the model accounting for the uncertainty on the bias, which is given a Gaussian prior with 0.07 keV σ\sigma, and vary the position of each peak (in a correlated way) by this amount. For the 0νββ0\nu\beta\beta decay analysis the peaks cover a wide range in energy and we take a conservative approach varying the position of all the peaks by the largest uncertainty of 0.3 keV (from 3034 keV).

Analysis efficiency

We account for the uncertainty in the analysis efficiency, or the probability that a signal like event would pass all cuts. These cut efficiencies are included as nuisance parameters in our model which are given Gaussian priors based on the estimated uncertainty (see Section IV B). A nuisance parameter is included for each cut, including two separate parameters for the constant and extrapolated light cut, and PCA <9< 9 cut. We then predict the efficiency in category rr as:

εr=∏c=111εcp(M(r),c),(11)\varepsilon_r = \prod_{c=1}^{11} \varepsilon_c^{p(M(r),c)}, \tag*{(11)}

where the product cc runs over all cuts (both energy independent and extrapolated, see Table IV) and p(M(r),c)p(M(r),c) is a power which represents how many times a cut was applied to an event. For example, for PSD the cut is applied for both pulses so the efficiency is raised to the power of the multiplicity.

CutEfficiency Energy Independent [%]Efficiency Extrapolated [%]Method of Evaluation
Pileup95.7±1.095.7 \pm1.0-Noise events
Multiplicity99.55±0.0799.55 \pm0.07-210Po{}^{210}\mathrm{Po}
LD Anti-coincidence99.976±0.01799.976 \pm0.017-210Po{}^{210}\mathrm{Po}
Delayed coincidence99.16±0.0199.16 \pm0.01-210Po{}^{210}\mathrm{Po}
Muon Veto99.62±0.0799.62 \pm0.07-210Po{}^{210}\mathrm{Po}
Light Cut (M1\mathcal{M}_1)99.4±0.499.4 \pm0.499.7±0.899.7 \pm0.8Gamma Peaks (M1\mathcal{M}_1)
Light Cut M>1\mathcal{M} > 197.7±1.897.7 \pm1.8-Gamma Peaks (M2\mathcal{M}_2)
PCA <23< 23 (M1\mathcal{M}_1)99.2±0.399.2 \pm0.3-Gamma Peaks (M1\mathcal{M}_1)
PCA <9< 9 (M1\mathcal{M}_1)95.2±0.595.2 \pm0.594.3±1.594.3 \pm1.5Gamma Peaks (M1\mathcal{M}_1)

Table IV. Efficiency for analysis cuts, the energy independent efficiencies are computed assuming the cut efficiency does not depend on energy. For the extrapolated efficiency in M1\mathcal{M}_1 data at 3034 keV, we fit the efficiency as a function of energy to a first order polynomial and extrapolate this to 3034 keV, this is only relevant for the M1\mathcal{M}_1 signatures with 3034 keV energy.

Containment Efficiency

The final systematic uncertainty we account for is the containment efficiency from MC simulations. In particular, this is related to the accuracy of our Geant4 MC simulations which can broadly be divided into two parts: the accuracy of the simulated geometry and the implemented physics process. For the simulated geometry we vary the amounts of the three main materials in the experimental setup (see Figure 2):

  • The dimensions of the LMO crystals (and therefore density since the mass is very well known);

  • The thickness of the copper holders;

  • The thickness of the Ge LDs.

We run a set of simulations varying the LMO dimensions (by ±100/200\pm100/200 μ\mum or around ±0.2/0.4%\pm0.2/0.4\%), the Cu holder thickness (by ±100/200\pm100/200 μ\mum or ±2.5/5%\pm2.5/5\%) and the Ge LD thickness (by ±10/20\pm10/20 μ\mum or around ±6/12%\pm6/12\%). We also account for the possible inaccuracy of Geant4 physics cross sections by running MC simulations varying both the Compton effect and photo-effect cross sections by ±5%\pm5\% which is a conservative estimate of Geant4 cross section accuracy [63, 64]. For each systematic effect (Cu thickness, LMO dimension, Ge thickness, Compton or photo-effect cross section scale) we introduce a parameter νi\nu_i. For every effect, decay, category (index rr) and peak (index pp) we parameterize the variation in peak containment efficiency. This is the product of the overall containment efficiency and the Gaussian peak probability (see § 0.3 D). We use first order polynomial fits to obtain dεcont,r,p/dνid\varepsilon_{\mathrm{cont},r,p}/d\nu_i. We show the results of all these tests in Appendix C. We observe that the most significant effects are the photo-effect cross section (higher scale tends to increase containment) and the Compton cross section (higher tends to decrease containment). The geometrical uncertainties provide a much smaller effect. Effects where the variation in containment efficiency is consistent with 0 (within 2σ2\sigma) are not included. Based on the parameters ν⃗\vec{\nu} we predict the efficiency for category rr and peak pp as:

εcont,r,p=εcont,r,p0+∑c=15νc⋅dεcont,r,pd(νc),(12)\varepsilon_{\mathrm{cont},r,p} = \varepsilon^0_{\mathrm{cont},r,p} + \sum_{c=1}^{5} \nu_c \cdot\frac{d\varepsilon_{\mathrm{cont},r,p}}{d(\nu_c)}, \tag*{(12)}

where the sum cc runs over systematic effects, and εcont,r,p0\varepsilon^0_{\mathrm{cont},r,p} is the default efficiency (for the MC simulation without varying any parameters). In our fit, the nuisance parameters accounting for uncertainties of copper thickness and LMO dimension are given Gaussian priors with a standard deviation of 100 μ\mum; for the Ge thickness we use 20 μ\mum, and we use 5% for the Compton and photo-effect physics cross sections. At each step of the Markov chain the amplitude of each peak is adjusted according to these parameters. The normalization of the fit functions are adjusted accordingly and we recompute the containment efficiency for both 01+/21+0^+_1/2^+_1 signals.

Choice of 2νββ2\nu\beta\beta decay to 01+0^+_1 state model

The final systematic uncertainty we consider is the 2νββ2\nu\beta\beta to 01+0^+_1 state decay model. By default we use the SSD model, however we also run MC simulations and a fit using the HSD model. We then consider the difference between these two results as a systematic on the 2νββ2\nu\beta\beta decay to 01+0_1^+ state decay rate.

Results

Toy Monte-Carlo sensitivity and bias

To validate our fitting routine and predict the median exclusion sensitivity we use an ensemble of pseudoexperiments (toy MC). We generate 2000 toy datasets for both 0νββ0\nu\beta\beta and 2νββ2\nu\beta\beta decay analysis based on the background parameters from the fit to data (see § 1.5 B 1,VII B 2). We assume the 2νββ→01+2\nu\beta\beta\rightarrow0_1^+ signal from the best fit to data and zero signal of 2νββ→21+2\nu\beta\beta\rightarrow2_1^+ or 0νββ0\nu\beta\beta decay modes. We sample from a Poisson distribution for the number of counts of each component in each toy experiment. We fit each dataset to determine the distribution of possible limits and therefore the median sensitivity for each decay. This is shown in Figure 8 and Figure 9. We extract median exclusion sensitivities, T^1/2\hat{T}_{1/2}, at 90% credibility interval (c.i.) for the three decays which have not yet been measured of:

Figure 8. Sensitivity for 0νββ0\nu\beta\beta decay to both 01+0_1^+ and 21+2_1^+ excited states obtained from an ensemble of 2000 toy experiments. We show the median sensitivity (dashed line) and the limit obtained on real data (solid lines). In both cases the limits on real data are consistent with those expected from toy experiments.

Figure 9. Sensitivity for 2νββ2\nu\beta\beta decay to 21+2_1^+ excited state obtained from an ensemble of 2000 toy experiments. As with 0νββ0\nu\beta\beta decay we show the median sensitivity (dashed line) and the limit obtained on real data (solid line). Again the limit on real data is consistent with the toy experiments.

T^1/22ν→21+=3.8×1021 yr,(13)\hat{T}_{1/2}^{2\nu\rightarrow2^+_1}=3.8\times10^{21}\ \mathrm{yr}, \tag*{(13)}
T^1/20ν→21+=3.1×1023 yr,(14)\hat{T}_{1/2}^{0\nu\rightarrow2^+_1}=3.1\times10^{23}\ \mathrm{yr}, \tag*{(14)}
T^1/20ν→01+=9.8×1022 yr.(15)\hat{T}_{1/2}^{0\nu\rightarrow0^+_1}=9.8\times10^{22}\ \mathrm{yr}. \tag*{(15)}

We also use these toys to verify the fitting procedure is not biased. We fit our toy datasets for 2νββ2\nu\beta\beta decay analysis with both a model including the 21+2_1^+ signal called H0=H(01++21++B)H_0=H(0_1^+ + 2_1^+ + B) and one including only 01+0_1^+ signal, H1=H(01++B)H_1=H(0_1^+ + B). We compute

ri=n0+i−nˉ0+σ(n0+),(16)r_i=\frac{n_{0^+}^i-\bar{n}_{0^+}}{\sigma(n_{0^+})}, \tag*{(16)}

where nˉ0+\bar{n}_{0^+} is the marginalized mode from the fit to real data used to generate the toys, n0+in_{0^+}^i is the marginalized mode in toy ii, and σ(n0+)\sigma(n_{0^+}) is the estimated error (central 68% c.i.). The distribution of rir_i for both models is shown in Figure 10. We observe a clear bias in rir_i for the H0H_0 model with a mean of −0.58σ-0.58\sigma, while rir_i is unbiased for the H1H_1 model (mean of −0.005σ-0.005\sigma). Hence, we decide to use the H1H_1 model for the case that no evidence of a decay to 21+2_1^+ state is found. This bias is due to non-negative Bayesian priors on the rate of 21+2_1^+ decay which allow the rate of 21+2_1^+ to fluctuate only up. Since the 01+0_1^+ and 21+2_1^+ rates are anti-correlated this causes the 01+0_1^+ rate to be biased.

Figure 10. The distribution of rir_i (see Eq. (16)) for both H0H_0 and H1H_1 models for our fits to toy data. We observe a clear bias for the H0H_0 model while the H1H_1 model is unbiased.

Fit of 2νββ2\nu\beta\beta with H0=H(01+21++B)H_0=H(0_1^+2_1^+ + B) model

We run the 2νββ2\nu\beta\beta decay fit with both 2νββ2\nu\beta\beta decay to 21+2_1^+ and 01+0_1^+ e.s. signals and background components, i.e. the

H0H_0 model. We can use this fit to determine if we observe evidence for the decay to 21+2_1^+ state and to set a limit if no evidence is found. We consider first a fit including all possible background peaks, the maximal model. We then repeat the fit removing any background peak where the central 68% confidence interval contains 0 counts and replacing the exponential background with a constant if the slope is compatible with 0 (within 1σ1\sigma). We call this the minimal model and it is this model we use for statistical inference. We show the marginalized posterior distribution of the decay rate Γ\Gamma for 2νββ2\nu\beta\beta decay to 21+2_1^+ state (including all systematics) in Figure 11 (left). We see that the mode is at 0 rate and therefore find no evidence for 2νββ2\nu\beta\beta to 21+2^+_1 state. We correspondingly set a limit (including all systematics) of:

Figure 11. Posterior probability distributions for the decay rates, Γ2ν\Gamma^{2\nu} to 21+2^{+}_{1} state (top left) and 01+0^{+}_{1} state (top right). The mode for 21+2^{+}_{1} decay is at zero rate and so we find no evidence for this decay and we extract the measurement of 01+0^{+}_{1} mode using the H(01++B)H(0^{+}_{1}+B) model. Bottom left/right are the posteriors of Γ0ν\Gamma^{0\nu} to 21+2^{+}_{1} and 01+0^{+}_{1} state respectively, again showing no evidence of the decay.

T1/22ν→21+>4.4×1021 yr (90% c.i.)(17)T^{2\nu\rightarrow2^+_1}_{1/2} > 4.4 \times10^{21}\ \mathrm{yr}\ (90\%\ \mathrm{c.i.}) \tag*{(17)}

This is the most stringent constraint on this process in 100Mo{}^{100}\mathrm{Mo}, an improvement of ∼80%\sim80\% over the previous constraint, 2.5×1021 yr2.5 \times10^{21}\ \mathrm{yr} [42].

Fit of 2 with H1=H(01++B) model

Since we find no evidence of the 2νββ2\nu\beta\beta decay to 21+2^+_1 state and toy experiments show that including this parameter in our fit would bias our measurement, we run the 2νββ2\nu\beta\beta decay H1=H(01++B)H_1 = H(0^+_1+B) fit without the 21+2^+_1 decay contribution. We find that this model is able to describe our data in all seven categories very well (see Figure 7 for an example and all fits in Appendix D). We verify the goodness of fit using our ensemble of toy experiments. For each fit we extract the global mode or the set of parameters θ~\tilde{\theta} with the maximum probability. We use the posterior probability of these parameters:

k=p(θ~∣D),(18)k = p(\tilde{\theta}\mid D), \tag*{(18)}

as a test statistic to quantify the quality of the fit by comparing the probability distribution obtained in toy experiments p(k′)p(k') to the fit on real data. We extract the pp-value:

p=∫0kp(k′) dk′=0.49(19)p = \int_{0}^{k} p(k')\,\mathrm{d}k' = 0.49 \tag*{(19)}

indicating that the model describes the data well. We extract the posterior distribution on Γ2ν→01+\Gamma^{2\nu\rightarrow0^+_1} from the fit on the data and we extract the central 68.3% c.i. on the decay rate as:

Γ2ν→01+=(9.2±1.0 (stat.+syst.))×10−22 yr−1.(20)\Gamma_{2\nu\rightarrow0^+_1} = (9.2 \pm1.0\ (\mathrm{stat.}+\mathrm{syst.})) \times10^{-22}\ \mathrm{yr}^{-1}. \tag*{(20)}

As the systematics are allowed to float freely in our fit this error is the sum of the statistical and systematic components. To evaluate the statistical error, we repeat the fit fixing all nuisance parameters connected to systematics, obtaining:

Γ2ν→01+=(9.1±1.0 (stat.))×10−22 yr−1.(21)\Gamma_{2\nu\rightarrow0^+_1} = (9.1 \pm1.0\ (\mathrm{stat.})) \times10^{-22}\ \mathrm{yr}^{-1}. \tag*{(21)}

A difference in the error compared to the statistical plus systematic fit is only observed after the first digit. We also run a fit using the HSD model, which leads to:

Γ2ν→01+HSD=(8.9±1.0 (stat.+syst.))×10−22 yr−1,(22)\Gamma^{\mathrm{HSD}}_{2\nu\rightarrow0^+_1} = (8.9 \pm1.0\ (\mathrm{stat.}+\mathrm{syst.})) \times10^{-22}\ \mathrm{yr}^{-1}, \tag*{(22)}

a 4% decrease in the decay rate. Under the assumption that statistical and systematic errors add in quadrature we obtain:

Γ2ν→01+=(9.2±1.0 (stat.) −0.5+0.3 (syst.))×10−22 yr−1,(23)\Gamma_{2\nu\rightarrow0^+_1} = (9.2 \pm1.0\ (\mathrm{stat.})\,{}^{+0.3}_{-0.5}\ (\mathrm{syst.})) \times10^{-22}\ \mathrm{yr}^{-1}, \tag*{(23)}

or converting to half life:

T1/22ν→01+=(7.5±0.8 (stat.) −0.3+0.4 (syst.))×1020 yr.(24)T^{2\nu\rightarrow0^+_1}_{1/2} = (7.5 \pm0.8\ (\mathrm{stat.})\,{}^{+0.4}_{-0.3}\ (\mathrm{syst.})) \times10^{20}\ \mathrm{yr}. \tag*{(24)}

This is a new independent measurement of this decay, with total uncertainty consistent with the previous leading measurement [42].

2 0 decay fit

From the fit for 0νββ0\nu\beta\beta decay we find no evidence of either the decay to 01+/21+0^+_1/2^+_1 excited states. The best fit reproductions of the experimental data are shown in Figure 11 (bottom left and right), this leads to limits (both at 90% c.i. including all systematic uncertainties) of:

T1/20ν→21+>2.1×1023 yr (90% c.i.),(25)T^{0\nu\rightarrow2^+_1}_{1/2} > 2.1 \times10^{23}\ \mathrm{yr}\ (90\%\ \mathrm{c.i.}), \tag*{(25)}
T1/20ν→01+>1.2×1023 yr (90% c.i.).(26)T^{0\nu\rightarrow0^+_1}_{1/2} > 1.2 \times10^{23}\ \mathrm{yr}\ (90\%\ \mathrm{c.i.}). \tag*{(26)}

These are new leading limits on these processes, a factor of 1.3 stronger than previous limits from [39] in both cases, despite a factor of ∼6\sim6 lower exposure of 100Mo{}^{100}\mathrm{Mo}.

3 2 decay spectral shape

The CUPID-Mo source equals detector geometry also allows us to investigate the 2νββ→01+2\nu\beta\beta\rightarrow0^+_1 spectral shape. This provides a concrete demonstration of a method for this analysis which can be applied to future experiments with large statistics. Our analysis described in § 0.3 does not reconstruct the ββ\beta\beta spectral shape directly, so we develop a procedure to extract this shape. For simplicity this analysis is only applied to M2M_{2} data which is the main contribution to the sensitivity.

We select all events that have either E1E_{1} or E2E_{2} in the range [539.5 − 1.75, 539.5 + 1.75] keV and [590.8 − 1.75, 590.8 + 1.75] keV. This choice of the interval is the FWHM energy resolution, roughly optimal for maximising the sensitivity S/BS/\sqrt{B}. We note there is some ambiguity in this selection when E1∼E2E_{1} \sim E_{2} (when the ββ\beta\beta has the same energy as the γ\gamma) however the excellent CUPID-Mo energy resolution makes this negligible. We then define Eββ,540E_{\beta\beta,540}, Eββ,591E_{\beta\beta,591} as the energy which does not contain the peak for both 540 and 591 keV peaks respectively and we construct histograms of these energies. This procedure is applied to both the data, signal MC simulations, and to the preliminary background model reconstruction of the data.

To compare the SSD and HSD models we use a simultaneous binned Bayesian likelihood fit of the two ββ\beta\beta

spectra. We use three components; background, 01+0^{+}_{1} signal and 21+2^{+}_{1} signal and 50 keV bins. Using the SSD model, this fit reconstructs the half-life as:

T1/22ν→01+=(7.3±1.0 (stat.))×1020 yr,(27)T^{2\nu\to0^{+}_{1}}_{1/2} = (7.3 \pm1.0\ \text{(stat.)}) \times10^{20}\ \text{yr}, \tag*{(27)}
T1/22ν→21+>3.6×1021 yr.(28)T^{2\nu\to2^{+}_{1}}_{1/2} > 3.6 \times10^{21}\ \text{yr}. \tag*{(28)}

These are consistent with the values from § 1.5 B,VII B 1 and should be considered a cross check of this more robust analysis which does not depend on the quality of the background model. We show this fit reproduction in Figure 12. Repeating the fit using MC simulations obtained with the HSD model, we extract the evidence for both models p(D∣SSD/HSD)p(\mathrm{D}|\mathrm{SSD}/\mathrm{HSD}) and therefore the probabilities for the HSD or SSD models of:

Figure 12. Plots showing the 2νββ2\nu\beta\beta decay to e.s. ββ\beta\beta spectral shape for 540 keV peak (left) and 591 keV peak (right). We overlay the best fit prediction. We find that in the region >600> 600 keV, the spectrum is dominated by 2νββ→01+2\nu\beta\beta\to0^{+}_{1}.

p(SSD∣D)=0.51,p(HSD∣D)=0.49.(29)p(\mathrm{SSD}|\mathrm{D}) = 0.51,\quad p(\mathrm{HSD}|\mathrm{D}) = 0.49. \tag*{(29)}

This indicates that the CUPID-Mo data is not able to differentiate between the two models. However, this analysis provides a method which could be used to differentiate these two models in a future experiment with larger statistics.

Discussion

Matrix element for 2νββ2\nu\beta\beta decay to 01+0^{+}_{1}

From our measurement of 2νββ→01+2\nu\beta\beta\to0^{+}_{1} excited state we can measure experimentally the nuclear matrix element for this process based on:

Γ2ν/ln⁡(2)=G2ν⋅∣M2νeff∣2.(30)\Gamma^{2\nu}/\ln(2) = G_{2\nu}\cdot\lvert M^{\mathrm{eff}}_{2\nu}\rvert^{2}. \tag*{(30)}

The phase space factor is given by (60.55−65.18)×10−21(60.55 - 65.18) \times10^{-21} 1/yr depending on whether the HSD or SSD model is used [65]. Therefore the matrix element (assuming the SSD model) is given by:

M2ν→01+eff=Γln⁡(2)⋅G±12ln⁡(2)GΔΓΓ=0.143±0.008 (stat.+syst.).(31)\begin{aligned} M^{\mathrm{eff}}_{2\nu\to0^{+}_{1}} = \sqrt{\frac{\Gamma}{\ln(2)\cdot G}} \pm\frac{1}{2\sqrt{\ln(2)G}}\frac{\Delta\Gamma}{\sqrt{\Gamma}} \\ &= 0.143 \pm0.008\ \text{(stat.+syst.)}. \tag*{(31)} \end{aligned}

We compare the theoretical predictions of dimensionless M2νeff=gA2M2νM^{\mathrm{eff}}_{2\nu}=g_{A}^{2}M_{2\nu} calculated assuming and unquenched value of gA=1.27g_{A}=1.27. These are 0.395, 0.595, 0.185 for the shell model [66], microscopic interacting Boson model [67], and quasi-particle random phase approximation [68], respectively. This shows the decay rate is quenched relative to theoretical predictions.

Effective Majorana neutrino mass, mββm_{\beta\beta}

The limits on 0νββ→01+0\nu\beta\beta\to0^{+}_{1}, despite the lower phase space, can be used to set a limit on mββm_{\beta\beta}. We use the phase space factor (3.162×10−15 yr−1(3.162 \times10^{-15}\ \mathrm{yr}^{-1} from [65]) and the NMEs from [66, 67] to obtain:

mββ<(11−15) eV,(32)m_{\beta\beta} < (11 - 15)\ \mathrm{eV}, \tag*{(32)}

depending on the NMEs used. Whilst this is still several orders of magnitude above the constrains from the g.s. decay this can be improved significantly in future for a large experiment with minimized dead material.

Bosonic neutrinos

Double beta decays to J≠0J \ne0 excited states can be sensitive to a Bosonic contribution to the neutrino wavefunction [33]. In particular, under the SSD hypothesis, the predicted half-lives for 2νββ2\nu\beta\beta to the 21+2^{+}_{1} e.s. are 2.4×10222.4 \times10^{22}, 1.7×10231.7 \times10^{23} yr for Bosonic and Fermionic neutrinos respectively. The current limit of 4.4×10214.4 \times10^{21} yr is still below these predictions.

SSD vs HSD models

Both NEMO-3 and CUPID-Mo have demonstrated that SSD hypothesis describes the experimental data very well for the 2νββ2\nu\beta\beta to 0g.s.+0^{+}_{\mathrm{g.s.}} decay in 100Mo{}^{100}\mathrm{Mo} [56, 57]. However, for the decay to 01+0^{+}_{1} state the mechanism (SSD or HSD/closure approximation) is not known. CUPID-Mo is only the second experiment to reconstruct the 2νββ2\nu\beta\beta to 01+0^{+}_{1} ββ\beta\beta spectral shape. However, the present level of statistics is insufficient to distinguish the two modes. The excellent energy resolution and low background rates mean this will be possible in a future experiment. In this paper, we demonstrate a method to quantify numerically this and obtained the first result on the compatibility of these two models with data.

Conclusion

In this paper we have presented a new analysis of 0νββ0\nu\beta\beta and 2νββ2\nu\beta\beta transitions of 100Mo{}^{100}\mathrm{Mo} to the first two (21+/01+2^{+}_{1}/0^{+}_{1}) excited states of 100Ru{}^{100}\mathrm{Ru} using the full exposure of CUPID-Mo. This analysis exploits the information available for a source equals detector experiment, where both β\beta and γ\gamma’s can be measured. A measurement of 100Mo T1/22ν→01+{}^{100}\mathrm{Mo}\ T_{1/2}^{2\nu\to0^{+}_{1}} was obtained:

T1/22ν→01+=(7.5±0.8 (stat.) −0.3+0.4 (syst.))×1020 yr.(33)T_{1/2}^{2\nu\to0_1^+} = (7.5 \pm0.8\ \text{(stat.)}\,{}_{-0.3}^{+0.4}\ \text{(syst.)}) \times10^{20}\ \text{yr.} \tag*{(33)}

For the other three decay modes, no evidence was found and we extract the limits:

T1/22ν→21+>4.4×1021 yr (90% c.i.),(34)T_{1/2}^{2\nu\to2_1^+} > 4.4 \times10^{21}\ \text{yr}\ \text{(90\% c.i.)}, \tag*{(34)}
T1/20ν→01+>1.2×1023 yr (90% c.i.),(35)T_{1/2}^{0\nu\to0_1^+} > 1.2 \times10^{23}\ \text{yr}\ \text{(90\% c.i.)}, \tag*{(35)}
T1/20ν→21+>2.1×1023 yr (90% c.i.).(36)T_{1/2}^{0\nu\to2_1^+} > 2.1 \times10^{23}\ \text{yr}\ \text{(90\% c.i.)}. \tag*{(36)}

These are the leading limits on these processes. The sensitivity was limited by the small size of the array and large amount of dead material which results in a low containment efficiency. Future experiments such as CUPID [13] or CROSS [69] will feature a more tightly packed array, much lower amounts of dead material, and a much larger exposure. This will lead to a significantly improved sensitivity with a precision measurement of 2νββ2\nu\beta\beta decay to 01+0_1^+ state, distinction between the SSD and HSD mechanisms of the decay and the possibility to measure the 2νββ2\nu\beta\beta decay to 21+2_1^+ state.

Acknowledgements

This work has been partially performed in the framework of the LUMINEU program, a project funded by the Agence Nationale de la Recherche (ANR, France). The help of the technical staff of the Laboratoire Souterrain de Modane and of the other participant laboratories is gratefully acknowledged. We thank the mechanical workshops of CEA/SPEC for their valuable contribution in the detector conception and of LAL (now IJCLab) for the detector holders fabrication. F.A. Danevich, V.V. Kobychev, V.I. Tretyak and M.M. Zarytskyy were supported in part by the National Research Foundation of Ukraine Grant No. 2020.02/0011. O.G. Polischuk was supported in part by the project Investigations of rare nuclear processes of the program of the National Academy of Sciences of Ukraine Laboratory of young scientists. A.S. Barabash, S.I. Konovalov, I.M. Makarov, V.N. Shlegel and V.I. Umatov were supported by the Russian Science Foundation under grant No. 18-12-00003. We acknowledge the support of the P2IO LabEx (ANR-10-LABX0038) in the framework Investissements dAvenir (ANR-11-IDEX-0003-01 Project BSM-nu) managed by the Agence Nationale de la Recherche (ANR), France.

Additionally the work is supported by the Istituto Nazionale di Fisica Nucleare (INFN) and by the EU Horizon2020 research and innovation program under the Marie Sklodowska-Curie Grant Agreement No. 754496. This work is also based on support by the US Department of Energy (DOE) Office of Science under Contract Nos. DE-AC02-05CH11231, and by the DOE Office of Science, Office of Nuclear Physics under Contract Nos. DE-FG02-08ER41551, DE-SC0011091; by the France-Berkeley Fund, the MISTI-France fund and by the Chateau-briand Fellowship of the Office for Science & Technology of the Embassy of France in the United States. J. Kotila is supported by Academy of Finland (Grant Nos. 3314733, 320062, 345869) This research used resources of the National Energy Research Scientific Computing Center (NERSC). This work makes use of the Diana data analysis software which has been developed by the Cuoricino, CUORE, LUCIFER, and CUPID-0 Collaborations.

Russian and Ukrainian scientists have given and give crucial contributions to CUPID-Mo. For this reason, the CUPID-Mo collaboration is particularly sensitive to the current situation in Ukraine. The position of the collaboration leadership on this matter, approved by majority, is expressed at https://cupid-mo.mi.institute.edu/collaboration#statement . Majority of the work described here was completed before February 24, 2022.

References

Optimization of categories

As explained in § 0.3, some categories are divided up by their projected out energies. In particular, for the 2νββ2\nu\beta\beta decay analysis the difference in the QQ-values, Q2β,01+=1904.1Q_{2\beta,0^{+}_{1}}=1904.1 keV and Q2β,21+=2494.9Q_{2\beta,2^{+}_{1}}=2494.9 keV, leads to an expected difference in the EββE_{\beta\beta} spectral shape. This can be used to help reduce correlation between the decays to these two states.

We use a preliminary background model fit to optimize these choices. We emphasize that this fit is only used for the optimization of the choice of categories and not in the Bayesian analysis. We first optimize a choice of ββ\beta\beta energies to minimize the expected measurement error (S+B/S\sqrt{S+B}/S) for 2νββ2\nu\beta\beta to 01+0^{+}_{1} e.s. decay. This optimization is performed on the 540 keV peak, however a consistent result is also obtained with 591 keV. A separate optimization is performed for the vertical line where the peak is in E2E_{2} and a horizontal where the peak is in E1E_{1} (see Figure 4). This leads to accepting only events with E2>220E_{2}>220 keV in the first case, and E1<1900E_{1}<1900 keV in the second.

Next we optimize the choice of energies, EaE_{\rm a}, such that the beta energy with a γ\gamma in E1E_{1} (horizontal line) is divided into two slices separated by EaE_{\rm a}:

E2∈[220 keV−Ea], [Ea−500 keV].(37)E_{2}\in[220\ \mathrm{keV}-E_{\rm a}],\ [E_{\rm a}-500\ \mathrm{keV}]. \tag*{(37)}

Similarly, for the events with the γ\gamma in E2E_{2} (vertical line) we divide into slices separated by the energies Eb,EcE_{\rm b},E_{\rm c}:

E1∈[500 keV−Eb], [Eb−Ec], [Ec−1900 keV].(38)E_{1}\in[500\ \mathrm{keV}-E_{\rm b}],\ [E_{\rm b}-E_{\rm c}],\ [E_{\rm c}-1900\ \mathrm{keV}]. \tag*{(38)}

This optimization maximizes the limit setting sensitivity for 2νββ2\nu\beta\beta to 21+2^{+}_{1} e.s. We estimate this as S/BS/\sqrt{B} using the 540 keV peak, this is a proxy for the correlation between the two decays. These optimizations lead to Ea,Eb,Ec=410,890,1190E_{\rm a},E_{\rm b},E_{\rm c}=410,890,1190 keV respectively.

We use a similar optimization scheme for the 0νββ0\nu\beta\beta decay analysis. In this case the relevant categories are horizontal lines with peaks in E1E_{1} (see Figure 4). We therefore make cuts only on the E2E_{2} variable. This is performed separately for the peaks at 1900 keV and ∼2400\sim2400 keV, and results in the categories shown in Table II. Here we optimize by maximizing the approximate sensitivity to 01+0^{+}_{1} decay for the 1904 keV peak, and the 21+2^{+}_{1} decay for the 2400 keV peaks.

Signal model functions

To model the signal shape as described in § 0.3 we use phenomenological functions which are often used for modeling the shape of signals in cryogenic calorimeters (for example see [57, 58]). In particular we use a linear combination of:

fpeak(E;p⃗,μ,R⃗,σ)=∑i=13pi⋅N(E,μ,Ri⋅σ(E)),(39)f_{\mathrm{peak}}(E;\vec{p},\mu,\vec{R},\sigma)=\sum_{i=1}^{3}p_i\cdot\mathcal{N}(E,\mu,R_i\cdot\sigma(E)), \tag*{(39)}
fxray(E;μ,EX,σ)=N(E,μ−EX,σ(E)),(40)f_{\mathrm{xray}}(E;\mu,E_{\mathrm{X}},\sigma)=\mathcal{N}(E,\mu-E_{\mathrm{X}},\sigma(E)), \tag*{(40)}
fstep(E;Emin⁡,μ,s,σ)=N⋅Erfc⁡(μ−E2σ),(41)f_{\mathrm{step}}(E;E_{\min},\mu,s,\sigma)=N\cdot\operatorname{Erfc}\left(\frac{\mu-E}{\sqrt{2\sigma}}\right), \tag*{(41)}
fstep−neg=N⋅Erfc⁡(E−μ2σ),(42)f_{\mathrm{step-neg}}=N\cdot\operatorname{Erfc}\left(\frac{E-\mu}{\sqrt{2\sigma}}\right), \tag*{(42)}
fbkg(E;s;Emin⁡,Emax⁡)=N⋅(1+s⋅E),(43)f_{\mathrm{bkg}}(E;s;E_{\min},E_{\max})=N\cdot(1+s\cdot E), \tag*{(43)}
fbump=N(E,μ,σ).(44)f_{\mathrm{bump}}=\mathcal{N}(E,\mu,\sigma). \tag*{(44)}

Here N(E,μ,σ)\mathcal{N}(E,\mu,\sigma) is a Gaussian with mean μ\mu and standard deviation σ\sigma, NN is a normalization coefficient and Erfc is the complementary error function. The first function models a peak, either γ\gamma or ββ\beta\beta, the second models events where a Mo X-ray escapes the crystal. The third and fourth functions (step/step-neg.) are step functions to account for Compton scattering events. The step function accounts for the Compton scatter of a single γ\gamma leading to a partial energy deposition and E<EphotopeakE < E_{\mathrm{photopeak}} and the step negative accounts for a combination of photoabsorption of one γ\gamma and Compton scatter of the other and E>EphotopeakE > E_{\mathrm{photopeak}}. We include a linear background and a single Gaussian to model some features in the data where a diagonal line crosses the projected out energy leading to a very small but broader peak (bump). For example see the category of 0νββ0\nu\beta\beta decay to 21+2^{+}_{1} e.s. with E1=1904E_1=1904 keV and E2∈[500,650]E_2\in[500,650] keV, here a diagonal line crosses the box at around 591 keV. This line is caused by a 2494 keV ββ\beta\beta in one crystal and a 540 keV γ\gamma shared between this crystal and another.

Containment efficiency uncertainty

As mentioned in § 1 C, several sets of MC simulations were used to estimate the uncertainty in the containment efficiency. For each peak, category and systematic test the percentage change in the containment efficiency was extracted. These values are shown in Figure 13.

Figure 13. Plots showing the systematic uncertainty on containment efficiency for each peak, category, decay mode, and systematic effect. The z-axis scale is the percentage effect. Top left this is 2νββ2\nu\beta\beta decay to 01+0^{+}_{1} state and on the right for decay to 21+2^{+}_{1} state. Bottom left is 0νββ0\nu\beta\beta decay to 01+0^{+}_{1} state, while bottom right is 0νββ0\nu\beta\beta decay to 21+2^{+}_{1} state.

Fits to each category

We show the best fit reproduction of each category for both 2νββ2\nu\beta\beta and 0νββ0\nu\beta\beta decay analysis in Figure 14 and Figure 15. Note that the spectra are binned for visualization purposes and we exclude plots with 0 counts in the experimental data. For the 2νββ2\nu\beta\beta decay fits we use the H1=(01++B)H_1=(0^{+}_{1}+B) model.

Figure 14. Fits of the seven 2νββ2\nu\beta\beta decay regions of interest, the experimental data is shown in grey while the best fit (global mode) is shown in blue. We see clearly a signal in each of the regions of interest.

Figure 15. Fits of the 0νββ0\nu\beta\beta decay regions of interest, the experimental data is shown in black while the best fit (global mode) is shown in blue. We only show categories with non-zero counts in the fit region.

References

  1. [12]ture 604, 53 (2022), arXiv:2104.06906 [nucl-ex].
  2. [13]W. R. Armstrong et al. (CUPID Interest Group), CUPID pre-CDR, arXiv:1907.09376 (2019).
  3. [14]E. Armengaud et al. (CUPID-Mo), The CUPID-Mo experiment for neutrinoless double-beta decay: performance and prospects, Eur. Phys. J. C 80, 44 (2020).DOI
  4. [15]F. F. Deppisch, M. Hirsch, and H. Päs, Neutrinoless double-beta decay and physics beyond the standard model, J. Phys G: Nucl. Part. Phys. 39, 124007 (2012).
  5. [16]W. Rodejohann, Neutrinoless double-beta decay and neutrino physics, J. Phys G: Nucl. Part. Phys. 39, 124008 (2012).DOI
  6. [17]G. Prézeau, M. Ramsey-Musolf, and P. Vogel, Neutrinoless double β decay and effective field theory, Phys. Rev. D 68, 034016 (2003).DOI
  7. [18]A. Atre, T. Han, S. Pascoli, and B. Zhang, The search for heavy Majorana neutrinos, Journal of High Energy Physics 2009, 030 (2009).DOI
  8. [19]M. Blennow, E. Fernandez-Martinez, J. Lopez-Pavon, and J. Menéndez, Neutrinoless double beta decay in seesaw models, Journal of High Energy Physics 2010, 96 (2010).
  9. [20]M. Mitra, G. Senjanović, and F. Vissani, Neutrinoless double beta decay and heavy sterile neutrinos, Nuclear Physics B 856, 26 (2012).
  10. [21]V. Cirigliano, W. Dekens, J. de Vries, M. L. Graesser, and E. Mereghetti, A neutrinoless double beta decay master formula from effective field theory, Journal of High Energy Physics 2018, 97 (2018).DOI
  11. [22]G. Benato, Effective Majorana Mass and Neutrinoless Double Beta Decay, Eur. Phys. J. C 75, 563 (2015), arXiv:1510.01089 [hep-ph].DOI
  12. [23]J. Engel and J. Menéndez, Status and Future of Nuclear Matrix Elements for Neutrinoless Double-Beta Decay: A Review, Rept. Prog. Phys. 80, 046301 (2017), arXiv:1610.06548 [nucl-th].DOI
  13. [24]P. K. Chandra, K. R. Chaturvedi, P. Lohani, P. K. Raina, and J. G. Hirsch, Neutrinoless ββ decay transition matrix elements within mechanisms involving light Majorana neutrinos, classical Majorons, and sterile neutrinos, Phys. Rev. C 88, 064322 (2013), arXiv:1308.0460 [nucl-th].
  14. [25]F. Šimkovic, V. Rodin, A. Faessler, and P. Vogel, 0νββ and 2νββ nuclear matrix elements, quasiparticle random-phase approximation, and isospin symmetry restoration, Phys. Rev. C 87, 045501 (2013), arXiv:1302.1509 [nucl-th].
  15. [26]N. L. Vaquero, T. R. Rodríguez, and J. L. Egido, Shape and pairing fluctuation effects on neutrinoless double beta decay nuclear matrix elements, Phys. Rev. Lett. 111, 142501 (2013).
  16. [27]J. Barea, J. Kotila, and F. Iachello, 0νββ and 2νββ Nuclear Matrix Elements in the Interacting Boson Model With Isospin Restoration, Phys. Rev. C 91, 034304 (2015).
  17. [28]M. Agostini et al. (GERDA), Final Results of GERDA on the Search for Neutrinoless Double-β Decay, Phys. Rev. Lett. 125, 252502 (2020).
  18. [29]A. Gando et al. (KamLAND-Zen), Search for Majorana Neutrinos Near the Inverted Mass Hierarchy Region with KamLAND-Zen, Phys. Rev. Lett. 117, 082503 (2016).
  19. [30]O. Azzolini et al. (CUPID-0), First Result on the Neutrinoless Double-β Decay of ⁸²Se with CUPID-0, Phys. Rev. Lett. 120, 232502 (2018).
  20. [31]C. Augier et al., Final results on the 0νββ decay half-life limit of ¹⁰⁰Mo from the CUPID-Mo experiment, (2022), arXiv:2202.08716 [nucl-ex].
  21. [32]A. Dolgov and A. Smirnov, Possible violation of the spin-statistics relation for neutrinos: Cosmological and astrophysical consequences, Physics Letters B 621, 1 (2005).
  22. [33]A. S. Barabash, A. D. Dolgov, R. Dvornicky, F. Simkovic, and A. Y. Smirnov, Statistics of neutrinos and the double beta decay, Nucl. Phys. B 783, 90 (2007), arXiv:0704.2944 [hep-ph].
  23. [34]J. H. Hubbell and S. M. Seltzer, X-Ray mass attenuation coefficients, NIST Standard Reference Database 126 (2004).
  24. [35]M. Berger et al., Stopping-Power & Range Tables for Electrons, Protons, and Helium Ions, NIST Standard Reference Database 124 https://doi.org/10.18434/T4NC7P (2017).
  25. [36]A. S. Barabash, A Possibility for experimentally observing two neutrino double beta decay, JETP Lett. 51, 207 (1990).
  26. [37]A. S. Barabash et al., Two neutrino double beta decay of ¹⁰⁰Mo to the first excited 0⁺ state in ¹⁰⁰Ru, Phys. Lett. B 345, 408 (1995).
  27. [38]A. S. Barabash, V. I. Umatov, R. Gurrarian, F. Hubert, and P. Hubert, 2νββ decay of ¹⁰⁰Mo to the first 0⁺ excited state in ¹⁰⁰Ru, Phys. Atom. Nucl. 62, 2039 (1999).
  28. [39]R. Arnold et al. (NEMO), Measurement of double beta decay of ¹⁰⁰Mo to excited states in the NEMO 3 experiment, Nucl. Phys. A 781, 209 (2007), arXiv:hep-ex/0609058.
  29. [40]M. F. Kidd, J. H. Esterline, W. Tornow, A. S. Barabash, and V. I. Umatov, New Results for Double-Beta Decay of ¹⁰⁰Mo to Excited Final States of ¹⁰⁰Ru Using the TUNL-ITEP Apparatus, Nucl. Phys. A 821, 251 (2009), arXiv:0902.4418 [nucl-ex].
  30. [41]P. Belli et al., New observation of 2βν decay of ¹⁰⁰Mo to the 0⁺₁ level of ¹⁰⁰Ru in the ARMONIA experiment, Nuclear Physics A 846, 143 (2010).
  31. [42]R. Arnold et al., Investigation of double beta decay of ¹⁰⁰Mo to excited states of ¹⁰⁰Ru, Nuclear Physics A 925, 25 (2014).
  32. [43]P. Belli et al., Double Beta Decay to Excited States of Daughter Nuclei, Universe 6, 239 (2020).
  33. [44]E. Armengaud et al. (EDELWEISS), Performance of the EDELWEISS-III experiment for direct dark matter searches, JINST 12 (08), P08010 (2017), arXiv:1706.01070 [physics.ins-det].
  34. [45]E. Armengaud et al. (LUMINEU), Development of ¹⁰⁰Mo-containing scintillating bolometers for a high-sensitivity neutrinoless double-beta decay search, Eur. Phys. J. C 77, 785 (2017).
  35. [46]E. E. Haller, N. P. Palaio, M. Rodder, W. L. Hansen, and E. Kreyza, NTD Germanium: A Novel Material for Low Temperature Bolometers, in Neutron Transmutation Doping of Semiconductor Materials, edited by R. D. Larrabee (Springer US, Boston, MA, 1984) pp. 21–36.
  36. [47]E. Armengaud et al. (CUPID), New Limit for Neutrinoless Double-Beta Decay of ¹⁰⁰Mo from the CUPID-Mo Experiment, Phys. Rev. Lett. 126, 181802 (2021), arXiv:2011.13243 [nucl-ex].
  37. [48]D. Q. Adams et al. (CUORE), Search for double-beta decay of ¹³⁰Te to the 0⁺ states of ¹³⁰Xe with CUORE, Eur. Phys. J. C 81, 567 (2021), arXiv:2101.10702 [nucl-ex].
  38. [49]C. Alduino et al. (CUORE), Double-beta decay of ¹³⁰Te to the first 0⁺ excited state of ¹³⁰Xe with CUORE-0, Eur. Phys. J. C 79, 795 (2019), arXiv:1811.10363 [nucl-ex].
  39. [50]O. Azzolini et al. (CUPID-0), Search of the neutrino-less double beta decay of ⁸²Se into the excited states of ⁸²Kr with CUPID-0, Eur. Phys. J. C 78, 888 (2018).
  40. [51]J. Allison et al., Recent developments in Geant4, Nucl. Instrum. Meth. A 835, 186 (2016).
  41. [52]E. Armengaud et al. (EDELWEISS), Background studies for the EDELWEISS dark matter experiment, Astropart. Phys. 47, 1 (2013), arXiv:1305.3628 [physics.ins-det].
  42. [53]D. R. Hamilton, On Directional Correlation of Successive Quanta, Phys. Rev. 58, 122 (1940).DOI
  43. [54]S. Chauvie et al., Geant4 low energy electromagnetic physics, in IEEE Symposium Conference Record Nuclear Science 2004., Vol. 3 (2004) pp. 1881–1885.
  44. [55]F. Simkovic, P. Domin, and S. V. Semenov, The Single state dominance hypothesis and the two neutrino double beta decay of ¹⁰⁰Mo, J. Phys. G 27, 2233 (2001), arXiv:nucl-th/0006084.
  45. [56]R. Arnaud et al. (NEMO-3), Detailed studies of ¹⁰⁰Mo two-neutrino double beta decay in NEMO-3, Eur. Phys. J. C 79, 440 (2019).
  46. [57]E. Armengaud et al. (CUPID-Mo), Precise measurement of 2νββ decay of ¹⁰⁰Mo with the CUPID-Mo detection technology, Eur. Phys. J. C 80, 674 (2020).
  47. [58]C. Alduino et al. (CUORE), Analysis Techniques for the Evaluation of the Neutrinoless Double-Beta Decay Lifetime in ¹³⁰Te with CUORE-0, Phys. Rev. C 93, 045503 (2016).
  48. [59]O. Azzolini et al. (CUPID-0), Analysis of cryogenic calorimeters with light and heat read-out for double beta decay searches, Eur. Phys. J. C 78, 734 (2018).
  49. [60]E. Gatti and P. Manfredi, Processing the signals from solid-state detectors in elementary-particle physics, Riv. Nuovo Cim. 9, 1 (1986).
  50. [61]R. Huang et al. (CUPID), Pulse Shape Discrimination in CUPID-Mo using Principal Component Analysis, JINST 16 (03), P03032 (2021), arXiv:2010.04033 [physics.data-an].
  51. [62]A. Caldwell, D. Kollár, and K. Kröninger, BAT The Bayesian analysis toolkit, Comput. Phys. Commun. 180, 2197 (2009).
  52. [63]G. Weidenspointner et al., Validation of Compton scattering Monte Carlo simulation models, in 2013 IEEE Nuclear Science Symposium and Medical Imaging Conference and Workshop on Room-Temperature Semiconductor Detectors (2013).
  53. [64]G. A. P. Cirrone, G. Cuttone, F. Di Rosa, L. Pandola, F. Romano, and Q. Zhang, Validation of the Geant4 electromagnetic photon cross-sections for elements and compounds, Nucl. Instrum. Meth. A 618, 315 (2010).
  54. [65]J. Kotila and F. Iachello, Phase-space factors for double-β decay, Phys. Rev. C 85, 034316 (2012).DOI
  55. [66]L. Coraggio et al., Shell-model calculation of ¹⁰⁰Mo double-β decay, Phys. Rev. C 105, 034312 (2022), arXiv:2203.01013 [nucl-th].
  56. [67]J. Barea, J. Kotila, and F. Iachello, 0νββ and 2νββ nuclear matrix elements in the interacting boson model with isospin restoration, Phys. Rev. C 91, 034304 (2015).
  57. [68]P. Pirinen and J. Suhonen, Systematic approach to β and 2νββ decays of mass A = 100–136 nuclei, Phys. Rev. C 91, 054309 (2015).
  58. [69]I. C. Bandac et al. (CROSS), The 0ν2β-decay CROSS experiment: preliminary results and prospects, JHEP 01, 018 (2020), arXiv:1906.10233 [nucl-ex].

Paper details

Contents