KamLAND-Zen Collaboration

Search for Majorana Neutrinos with the Complete KamLAND-Zen Dataset

Abstract

We present a search for neutrinoless double-beta ($0\nu\beta\beta$) decay of ¹³⁶Xe using the full KamLAND-Zen 800 dataset with 745 kg of enriched xenon, corresponding to an exposure of $2.097$ ton yr of ¹³⁶Xe. This updated search benefits from a more than twofold increase in exposure, recovery of photo-sensor gain, and reduced background from muon-induced spallation of xenon. Combining with the search in the previous KamLAND-Zen phase, we obtain a lower limit for the $0\nu\beta\beta$ decay half-life of $T_{1/2}^{0\nu}>3.8\times 10^{26}$ yr at 90% C.L., a factor of 1.7 improvement over the previous limit. The corresponding upper limits on the effective Majorana neutrino mass are in the range 28–122 meV using phenomenological nuclear matrix element calculations.

The search for neutrinoless double-beta ($0\nu\beta\beta$) decay is an active field of research in nuclear and particle physics Agostini2023 . Observation of this decay would prove the Majorana nature of neutrinos and demonstrate the non-conservation of lepton number, a key feature of Leptogenesis, which can explain the imbalance of the matter over antimatter in the Universe Fukugita1986 . In the model of light Majorana neutrino exchange, the $0\nu\beta\beta$ decay rate is proportional to the square of the effective Majorana neutrino mass $\left<m_{\beta\beta}\right>\equiv\left|\Sigma_{i}U_{ei}^{2}m_{\nu_{i}}\right|$. The most stringent limit to date has been provided by KamLAND-Zen, reaching the inverted mass ordering (IO) region below 50 meV for the first time Abe2023a . An improved search covering the unexplored IO region would stringently test theoretical models predicting $\left<m_{\beta\beta}\right>$ in this range Harigaya2012 ; Asaka2020 ; Asai2020 , with a significantly increased chance to observe $0\nu\beta\beta$ decay Agostini2017 .

The KamLAND-Zen (KamLAND Zero-Neutrino Double-Beta Decay) Gando2012a ; Gando2012b ; Gando2013a ; Asakura2016 ; Gando2016 ; Gando2019 ; Abe2023a experiment utilizes the low-background KamLAND detector to search for $0\nu\beta\beta$ decay. Enriched xenon is dissolved in liquid scintillator (Xe-LS) contained in a teardrop-shaped inner balloon (IB). The IB is constructed from 25-$\mu$m-thick transparent nylon film and is suspended at the center of the KamLAND detector. The IB is surrounded by 1 kton of LS (Outer LS) contained in a 13-m-diameter spherical outer balloon. Scintillation light is viewed by 1,325 17-inch and 554 20-inch photomultiplier tubes (PMTs) mounted on an 18-m-diameter spherical stainless-steel tank. In the previous phase of the experiment (KamLAND-Zen 400) with 381 kg of enriched xenon, we probed the quasidegenerate neutrino mass region Gando2016 . To further improve the search sensitivity, the KamLAND-Zen collaboration upgraded the experiment (KamLAND-Zen 800) to 745 kg of enriched xenon, nearly twice the target mass of the previous phase. To hold the increased xenon, we constructed a larger and cleaner 3.80-m-diameter IB Gando2021 . The Xe-LS consists of 82% decane and 18% pseudocumene (1,2,4-trimethylbenzene) by volume, 2.4 g/liter of the fluor PPO (2,5-diphenyloxazole), and $(3.13\pm 0.01)$% by weight of isotopically enriched xenon gas. The isotopic abundances in the enriched xenon are $(90.85\pm 0.13)\%$ ¹³⁶Xe and $(8.82\pm 0.01)\%$ ¹³⁴Xe.

KamLAND-Zen 800 completed data-taking in January 2024 to start dismantling the current KamLAND detector in preparation for a detector upgrade, KamLAND2-Zen. This Letter reports on the $0\nu\beta\beta$ decay analysis with the full KamLAND-Zen 800 dataset collected between February 5, 2019, and January 12, 2024, and its combination with the earlier KamLAND-Zen 400 data.

The experimental signal is a characteristic peak in the energy distribution, produced by the two $e^{-}$ emitted from ¹³⁶Xe $\beta\beta$ decay. For hypothetical $0\nu\beta\beta$ decays, their sum energy is always 2.458 MeV (Q-value of ¹³⁶Xe $\beta\beta$ decay) Redshaw2007 , while for $2\nu\beta\beta$ decays the sum has a continuous spectrum up to the Q-value. Event vertex and energy are reconstructed based on the timing and charge distributions of photoelectrons (p.e.) recorded by the PMTs. Energy calibration is performed using ²¹⁴Bi $\beta$’s and $\gamma$’s from ²²²Rn ($\tau=5.5$ days), introduced during the initial filling of the IB with Xe-LS, and 2.225 MeV $\gamma$’s from capture of spallation neutrons by protons. The detector Monte Carlo (MC) simulation is based on Geant4 Agostinelli2003 ; Allison2006 and is tuned to reproduce the timing and charge distributions observed in the data. The estimated energy and vertex resolutions in the Xe-LS are 6.7%/$\sqrt{E({\rm MeV})}$ and 13.7 cm/$\sqrt{E({\rm MeV})}$, respectively.

During the earlier period in KamLAND-Zen 800, we observed an increasing trend in the number of low photo-detection efficiency PMTs caused by age-related gain decrease. The resulting degradation in energy resolution could increase the background from $2\nu\beta\beta$ tails. In March 2020, we started a continuous campaign modifying the input circuitry of the KamLAND front-end electronics (FEE) to counter the PMT gain decrease. The 1 p.e. gain and timing offsets due to the circuit modification are corrected for each PMT based on photon counting with high-statistics background events. Owing to these restoration efforts, we have succeeded in preventing the degradation of the energy resolution by a factor of 1.2, which would have lead to a twofold increase of the $2\nu\beta\beta$ background.

Figure 1: Vertex distribution of candidate singles data (SD) events (black points) in the $0\nu\beta\beta$ energy window $2.35<E<2.70\,{\rm MeV}$. The thick black and red lines indicate the shape of the IB and the 1.57-m-radius spherical volume, respectively. The dot-dashed line indicates the nylon belt suspending the IB. The thin dashed lines illustrate the shape of the equal-volume spherical half-shells, which compose the 2.5-m-radius spherical fiducial volume. The high-count region at the IB bottom within the yellow solid line indicates the hot spot and is removed through a 0.7 m radius cut.

We apply the following cuts to select $\beta\beta$ decay events: (i) The reconstructed vertex must be within 2.5 m of the detector center and 0.7 m away from the bottom hot spot on the IB, which is outlined in Fig. 1. (ii) Muons and events within 2 ms after muons are rejected. (iii) Bi-Po decay pairs are eliminated by a delayed coincidence tag, which requires the time and distance between the prompt Bi and delayed Po decay candidates to be less than 1.9 ms and 1.7 m, respectively, and by rejecting events with a double pulse identification within a single event acquisition window. Those cuts remove $(99.95\pm 0.03)$% of ²¹⁴Bi-²¹⁴Po events, and $(97.9\pm 0.5)$% of ²¹²Bi-²¹²Po. (iv) Reactor $\overline{\nu}_{e}$’s identified by a delayed coincidence of positrons and neutron-capture $\gamma$’s are rejected. (v) Poorly reconstructed events are rejected. These events are tagged using a vertex-time-charge discriminator which measures how well the observed PMT time-charge distributions match those expected based on the reconstructed vertex. The overall selection inefficiency is less than 0.1%.

Background sources external to the Xe-LS are dominated by radioactive impurities on the IB film. The contamination levels of ²³⁸U and ²³²Th are $(4.3\pm 0.2)\times 10^{-12}$ g/g and $(2.4\pm 0.1)\times 10^{-11}$ g/g, respectively, which are roughly a factor of 10 smaller than those measured on the previous IB Gando2016 . The reference calculations for ²³⁸U and ²³²Th mentioned here assume secular equilibrium for comparison with the previously reported values. In the earlier period of the dataset, we found an increase in the background rate at the IB bottom, possibly due to the settling of dust particles containing radioactive impurities. To avoid this possible background, we tag and remove this high-background period from the dataset using machine learning algorithms, as discussed in Ref. Abe2023a ; Li2023 .

Solar neutrinos are an intrinsic background source for KamLAND-Zen. Based on neutrino flux calculations using the standard solar model Serenelli2011 , the elastic scattering (ES) of ⁸B solar neutrinos on electrons in the Xe-LS is estimated to be $(4.9\pm 0.2)\times 10^{-3}$ (ton day)^-1, including the effect of three-flavor neutrino oscillations. The charged-current (CC) interactions on ¹³⁶Xe produce $e^{-}$, ¹³⁶Cs, and $\gamma$’s from the excited states, and the subsequent decays of ¹³⁶Cs ($\tau=19.0$ days, $Q=2.548$ MeV) create a background peak around 2.0 MeV in visible energy, mostly overlap** with the resolution tail of $2\nu\beta\beta$ decays. The interaction rate is expected to be $(0.8\pm 0.1)\times 10^{-3}$ (ton day)^-1 based on the cross section calculated in Ref. Ejiri2014 ; Frekers2013 .

Cosmic-ray muons produce neutrons and radioactive isotopes through nuclear spallation of carbon and xenon, which decay by emitting $\beta$’s or $\gamma$’s. To remove these events, we apply cuts based on space and time correlation with muon and neutron capture $\gamma$ events. In the $0\nu\beta\beta$ window, decays of ¹⁰C ($\tau=27.8$ s, $Q=3.65$ MeV) and ⁶He ($\tau=1.16\,{\rm s}$, $Q=3.51\,{\rm MeV}$) dominate the muon spallation backgrounds. To reduce these short-lived backgrounds, we remove events within 150 ms after muons, and events reconstructed within 1.6 m of neutron vertices for 180 s. In addition, we reject remaining background events by the “shower” likelihood method based on reconstructed muon shower profiles Li2014 ; Li2015a ; Li2015b ; Zhang2016 ; Abe2023b . The overall rejection efficiencies for ¹⁰C and ⁶He are $>$$99.7$% and $(97.3\pm 1.5)\%$, respectively. To reduce the ¹³⁷Xe ($\tau=5.5\,{\rm min}$, $Q=4.17\,{\rm MeV}$) background, we remove events reconstructed within 1.6 m for 27 min of vertices identified as neutron captures on ¹³⁶Xe, which produce high energy $\gamma$’s ($Q=4.03\,{\rm MeV}$). This cut removes $(74\pm 7)\%$ of ¹³⁷Xe.

The products of muon spallation with lifetimes greater than $\mathcal{O}$(100 s), denoted as “long-lived products”, are attributed to the decay of heavy isotopes produced by xenon spallation. Xenon spallation can be tagged by detecting multiple neutrons. To characterize the long-lived products, we define a likelihood ratio, $R_{L}=L_{\rm spa}/(L_{\rm spa}+L_{\rm acc})$. Here $L_{\rm spa}$ and $L_{\rm acc}$ are the probability density functions (PDFs) for long-lived muon-spallation pairs and accidental pairs, respectively. These PDFs are constructed as a function of neutron multiplicity, distance to neutron vertices, and time interval from preceding muons. The cut value on $R_{L}$ is optimized using an estimate with MC simulation tools. FLUKA BOHLEN2014211 ; Ferrari:898301 is used to calculate the spallation isotope yields, and Geant4 to calculate the radioactive decays and energy spectra of the isotopes, considering their sequential decay chain. The predicted backgrounds are primarily from ¹³²I, ¹³⁰I, ¹²⁴I, ¹²²I, ¹¹⁸Sb, ¹¹⁰In, ⁸⁸Y, and from other isotopes in smaller amounts (Table IX in Ref. Abe2023b ). The total background rate in the $0\nu\beta\beta$ window is $30.0\pm 2.2$ (ton yr)^-1 in total, in units of ¹³⁶Xe exposure. Events that are not classified as coming from long-lived backgrounds are referred to as “singles data” (SD), and the others are referred to as “long-lived data” (LD).

Figure 2: Energy spectra of selected $\beta\beta$ candidates within a 1.57-m-radius spherical volume drawn together with best-fit backgrounds, the $2\nu\beta\beta$ decay spectrum, and the 90% C.L. upper limit for $0\nu\beta\beta$ decay of (a) singles data (SD), and (b) long-lived data (LD). The LD exposure is about 10% of the SD exposure.

We improved the neutron identification for post-muon events to enhance the tagging efficiency of the long-lived backgrounds. We identify these neutrons with a second, deadtime-free electronics system, called MoGURA Abe2023b . Previously, the neutron vertex was reconstructed using only PMTs with normal gain. In this analysis, we added low gain PMTs to the photon counting, achieving an increase in the mean number of PMT hits from $181$ to $201$ for 2.225 MeV $\gamma$’s from neutron capture, after subtracting afterpulse noise contributions. The resulting neutron tagging efficiency is estimated to be $(74.5\pm 0.4)\%$ based on the neutron capture time distribution, approximately 2% better than in the previous analysis. Our MC study shows that $(47.1\pm 8.7)\%$ of long-lived spallation backgrounds are classified as LD, whereas only $9.0$% of uncorrelated events are mis-classified.

The total livetime for SD and LD in KamLAND-Zen 800 is $1131$ days and $111$ days, respectively. The total exposure of SD, which is sensitive to the $0\nu\beta\beta$ signal, is $2097$ kg yr of ¹³⁶Xe, more than double the amount of the previous search Abe2023a . The $0\nu\beta\beta$ decay rate is estimated from a simultaneous likelihood fit to the binned energy spectra of SD and LD between 0.5 and 4.8 MeV. The binned energy spectra are made by dividing the 2.5-m-radius fiducial volume into 40 equal-volume bins (illustrated in Fig. 1), considering the time variation of event rates. The contributions from major backgrounds in the Xe-LS, such as ⁸⁵Kr, ⁴⁰K, ²¹⁰Bi, the ²²⁸Th-²⁰⁸Pb subchain of the ²³²Th series, and long-lived spallation products, are free parameters and are left unconstrained in the spectral fit. The contributions from the ²²²Rn-²¹⁰Pb subchain of the ²³⁸U series and short-lived spallation products are constrained by their independent measurements. The parameters of the detector energy response model are constrained by ²¹⁴Bi data. The energy spectral distortion parameter (a linear rescaling) for the long-lived spallation products is constrained by its estimated uncertainty (Fig. 10 in Ref. Abe2023b ).

Figure 3: Allowed region for the ¹³⁶Xe $0\nu\beta\beta$ rate and the long-lived spallation background rate in the energy region $2.35<E<2.70\,{\rm MeV}$ ($0\nu\beta\beta$ window). The red dot and contour lines correspond, respectively, to the best-fit and the $1\sigma$, 90%, 95%, 99% C.L. intervals for two degrees of freedom. The horizontal line indicates the MC-based prediction.

Figure 2 shows the energy spectra of selected candidate SD and LD events within a 1.57-m-radius spherical volume together with the best-fit curves. The exposure of ¹³⁶Xe for SD in this reduced volume is 1.13 ton yr. The best-fit background contributions are summarized in Table 1. The $0\nu\beta\beta$ signal best-fit is 0 events, indicating no event excess over the background expectation. We obtain a 90% confidence level (C.L.) upper limit on the number of $0\nu\beta\beta$ decays of $<$ $10.0$ events ($<$ $8.0$ events in the range $2.35<E<2.70\,{\rm MeV}$), which corresponds to a limit of $<$ $8.9$ (ton yr)^-1 in units of ¹³⁶Xe exposure, or $T_{1/2}^{0\nu\beta\beta}>3.4\times 10^{26}$ yr (90% C.L.). An analysis based on the Feldman-Cousins procedure PhysRevD.57.3873_FCmethod gives a slightly stronger limit of $4.3\times 10^{26}$ yr (90% C.L.), indicating a limited impact of the physical boundary on the $0\nu\beta\beta$ rate at low statistics. An MC simulation of an ensemble of experiments assuming the best-fit background spectrum rwithout a $0\nu\beta\beta$ signal indicates a median sensitivity of $2.3\times 10^{26}$ yr. The probability of obtaining a limit stronger than that reported here is 24%. In addition to the frequentist analyses above, we also performed a statistical analysis within the Bayesian framework, assuming a flat prior for $1/T_{1/2}^{0\nu\beta\beta}$. The Bayesian limit and median sensitivity are $3.4\times 10^{26}$ yr and $2.4\times 10^{26}$ yr (90% C.L.), respectively.

Figure 4: Effective Majorana neutrino mass $\left<m_{\beta\beta}\right>$ as a function of the lightest neutrino mass $m_{\rm lightest}$. The dark shaded regions are predictions based on best-fit values of neutrino oscillation parameters for (a) the normal ordering (NO) and (b) the inverted ordering (IO), and the light shaded regions indicate the $3\sigma$ ranges calculated from oscillation parameter uncertainties DellOro2014 ; NuFIT2020 . The horizontal lines indicate 90% C.L. upper limits on $\left<m_{\beta\beta}\right>$ with ¹³⁶Xe from KamLAND-Zen (this work), considering an improved phase space factor calculation Kotila2012 ; Stoica2013 and phenomenological NME calculations: shell model Menendez2018 ; PhysRevC.93.024308horoi ; PhysRevC.101.044315.coraggio2020 ; PhysRevC.105.034312.coraggio2022 (dot-dashed lines), quasiparticle randomphase approximation (QRPA) PhysRevC.87.064302.mustonen ; PhysRevC.91.024613.Hyv ; PhysRevC.98.064325.simko2018 ; PhysRevC.97.045503fang ; PhysRevC.102.044303terasaki (dotted lines), energy-density functional (EDF) theory PhysRevLett.105.252503.rodriguez ; PhysRevLett.111.142501lopez ; PhysRevC.95.024305song (solid lines), interacting boson model (IBM) PhysRevC.91.034304barea ; PhysRevD.102.095016deppisch (dashed lines). (c) The corresponding limits for ¹³⁶Xe, ⁷⁶Ge Agostini2020 , and ¹³⁰Te Adams2022 . Three theoretical predictions in the IO region are also shown: (A)Harigaya2012 , (B)Asaka2020 , (C)Asai2020 .

Table 1: Summary of the estimated and best-fit background contributions in the energy region $2.35<E<2.70\,{\rm MeV}$ within the 1.57-m-radius spherical volume. In total, $59$ events were observed.

[h] Background        Estimated       Best-fit ¹³⁶Xe $2\nu\beta\beta$ - $27.60$ Residual radioactivity in Xe-LS ²³⁸U series $0.08\pm 0.01$ $0.08$ ²³²Th series - $1.35$ External (Radioactivity in IB) ²³⁸U series - $6.61$ ²³²Th series - $0.04$ Neutrino interactions ⁸B solar $\nu$ $e^{-}$ ES $3.57\pm 0.09$ $3.57$ Spallation products Long-lived $17.82\pm 1.30$ ^$\dagger$ $19.39$ ¹⁰C $0.00\pm 0.05$ $0.00$ ⁶He $0.54\pm 0.27$ $0.56$ ¹³⁷Xe $0.72\pm 0.60$ $0.72$

• Estimation based on the spallation MC study. This event rate constraint is not applied in the spectral fit.

Figure 3 shows the allowed region of the ¹³⁶Xe $0\nu\beta\beta$ rate and the long-lived spallation background rate from a combined fit of the KamLAND-Zen 400 and 800 datasets, giving a limit of

$$T_{1/2}^{0\nu\beta\beta}>3.8\times 10^{26}\,\textrm{yr~{}~{}~{}(90\% C.L.)}.$$ (1)

The best-fit long-lived spallation background rate is $30.2\pm 4.5$ (ton yr)^-1indicating good consistency between the MC-based prediction and the LD analysis. This combined analysis has a sensitivity of $2.6\times 10^{26}$ yr, and the probability of obtaining a stronger limit is $29$%. The lower limit on $T_{1/2}^{0\nu\beta\beta}$ can be converted to an upper limit on $\langle m_{\beta\beta}\rangle$ assuming that the $0\nu\beta\beta$ decay mechanism is dominated by exchange of light Majorana neutrinos. Assuming the axial coupling constant $g_{A}\simeq 1.27$, we obtain an upper limit of

$$\langle m_{\beta\beta}\rangle<(28-122)\,\textrm{meV~{}~{}~{}(90\% C.L.)},$$ (2)

using an improved phase space factor calculation of $14.54\times 10^{-15}\,(\textrm{yr})^{-1}$ Kotila2012 ; Stoica2013 , and phenomenological nuclear matrix element (NME) calculations, summarized in Table 2. In this letter, we follow the current standard to use those phenomenological models, although recent theoretical work has produced new NME calculations belley2023ab ; Jokiniemi2023 incorporating both quenching physics Faessler_2008 ; PhysRevC.87.014315 and short-range contact PhysRevLett.120.202001 contributions, which are missing in the calculations in Table 2. Considering the calculations from Ref. belley2023ab ; Jokiniemi2023 gives ranges of upper limits on $\langle m_{\beta\beta}\rangle$ of $71-125$ meV (ab initio belley2023ab ), $25-60$ meV (pnQRPA Jokiniemi2023 ), and $44-138$ meV (shell model Jokiniemi2023 ). Figure 4 illustrates the allowed range of $\langle m_{\beta\beta}\rangle$ as a function of the lightest neutrino mass for the NME values in Table 2. The shaded regions include the uncertainties in mixing elements $U_{ei}$ and the neutrino mass splitting for each hierarchy. Also drawn are experimental limits from $0\nu\beta\beta$ decay searches in other nuclei Agostini2020 ; Adams2022 . The limit using ¹³⁶Xe by KamLAND-Zen has provided the most stringent tests of the effective Majorana mass in the IO region and several theoretical models Harigaya2012 ; Asaka2020 ; Asai2020 . It also provides the strongest constraint on $m_{\rm lightest}$ considering extreme cases of the combination of CP phases and the uncertainties from neutrino oscillation parameters DellOro2014 ; NuFIT2020 . We obtain a 90% C.L. upper limit of $m_{\rm lightest}<(84-353)\,{\rm meV}$.

In conclusion, we presented a $0\nu\beta\beta$ decay search with the complete KamLAND-Zen dataset including improvements in the rejection of muon-induced long-lived spallation backgrounds. This search places the most stringent constraints to date on $\left<m_{\beta\beta}\right>$, and reaches the inverted mass ordering region below 50 meV with half of the phenomenological nuclear matrix element calculations. The sensitivity is limited by background from xenon spallation products and $2\nu\beta\beta$ decay. In the near future, we plan to upgrade the KamLAND detector with more ¹³⁶Xe mass, enhanced energy resolution to minimize the leakage of the $2\nu\beta\beta$ tail into the $0\nu\beta\beta$ analysis window, and enhanced rejection of spallation products through improved neutron tagging and particle identification methods.

Table 2: Summary of commonly used phenomenological nuclear matrix elements ($M^{0\nu}$) for ¹³⁶Xe $0\nu\beta\beta$ decay and the corresponding 90% C.L. upper limits on the effective Majorana neutrino mass $\left<m_{\beta\beta}\right>$ from KamLAND-Zen data. The limits shown in Eq. (2) use the minimum and maximum.

	Ref.	$M^{0\nu}$	$\langle m_{\beta\beta}\rangle$ (meV)
Shell model	Menendez2018	2.28, 2.45	59.4, 55.3
	PhysRevC.93.024308horoi	1.63, 1.76	83.1, 77.0
	PhysRevC.101.044315.coraggio2020 ; PhysRevC.105.034312.coraggio2022	2.39	56.7
QRPA	PhysRevC.87.064302.mustonen	1.55	87.4
	PhysRevC.91.024613.Hyv	2.91	46.6
	PhysRevC.98.064325.simko2018	2.71	50.0
	PhysRevC.97.045503fang	1.11, 1.18	122, 115
	PhysRevC.102.044303terasaki	3.38	40.1
EDF theory	PhysRevLett.105.252503.rodriguez	4.20	32.3
	PhysRevLett.111.142501lopez	4.77	28.4
	PhysRevC.95.024305song	4.24	32.0
IBM	PhysRevC.91.034304barea	3.25	41.7
	PhysRevD.102.095016deppisch	3.40	39.9