What can we learn from neutrinoless double beta decay experiments?

What can we learn from neutrinoless double beta decay experiments?

John N. Bahcall,*Hitoshi Murayama,^†and C. Pen˜a-Garay^‡

School of Natural Sciences, Institute for Advanced Study, Princeton, New Jersey 08540, USA 共Received 7 April 2004; published 26 August 2004兲

We assess how well next-generation neutrinoless double beta decay and normal neutrino beta decay experi- ments can answer four fundamental questions.共1兲 If neutrinoless double beta decay searches do not detect a signal, and if the spectrum is known to be inverted hierarchy, can we conclude that neutrinos are Dirac particles?共2兲 If neutrinoless double beta decay searches are negative and a next-generation ordinary beta decay experiment detects the neutrino mass scale, can we conclude that neutrinos are Dirac particles?共3兲 If neutrino- less double beta decay is observed with a large neutrino mass element, what is the total mass in neutrinos?共4兲 If neutrinoless double beta decay is observed, but next-generation beta decay searches for a neutrino mass only set a mass upper limit, can we establish whether the mass hierarchy is normal or inverted? We base our answers on the expected performance of next-generation neutrinoless double beta decay experiments and on simulations of the accuracy of calculations of nuclear matrix elements.

DOI: 10.1103/PhysRevD.70.033012 PACS number共s兲: 14.60.Pq

I. INTRODUCTION

A new generation of double beta decay experiments will be undertaken with unprecedented accuracy. In approxi- mately the same time frame, it will become possible to make much more precise measurements of, or set constraints on, the mass of neutrinos emitted in ordinary beta decay. The results of these next-generation experiments will be impor- tant for understanding the physics of weak interactions.

If neutrinoless double beta decay is observed, then one can conclude 关1兴 immediately that neutrinos are Majorana particles without messing around with detailed calculations and qualifications of the kind discussed in this paper. 共We will not consider alternative interpretations, such as R-parity violation关2–4兴, which can probably be verified or excluded at high-energy colliders. The violation of the lepton number is clear in either case.兲 The community of physicists can and will celebrate if double beta decay is observed.

In this paper, we provide quantitative estimates of how well we can answer four other fundamental questions about neutrinos using the assumed results of the next generation of neutrinoless double beta decay experiments and normal beta decay experiments.

Our principal results are summarized in Table I.

A. How can we estimate the uncertainties in calculated nuclear matrix elements?

The uncertainty in the calculated nuclear matrix elements for neutrinoless double beta decay will constitute the princi- pal obstacle to answering some basic questions about neutri- nos. The essential problem is that the correct theory of nuclei is QCD, a notoriously difficult theory with which to do cal-

culations for nuclei with several nucleons. For neutrinoless double beta decay, the situation is even more severe because double beta candidates involve systems with A⬃50 to A

⬃100 and even larger. Very attractive next-generation ex- periments have been proposed for a number of different iso- topes, including ⁴⁸Ca关6兴, ⁷⁶Ge 关5,7,8兴,¹⁰⁰Mo关9,10兴, ¹¹⁶Cd 关11兴,¹³⁰Te 关12,13兴, ¹³⁶Xe 关14–16兴,¹⁵⁰Nd 关17兴, and ¹⁶⁰Gd 关18,19兴.

In the foreseeable future, it does not seem possible to derive in a direct and controlled manner from QCD nuclear matrix elements for large A. Thus there is no way of quanti- fying with absolute confidence the range of uncertainties in nuclear matrix elements calculated with different theoretical models or approximations.

In the absence of being able to derive the errors directly from QCD, we assume that the published range of calculated matrix elements defines a plausible approximation to the un- certainty in our knowledge of the matrix elements. We do not, for example, favor a particular calculation because it happens to give better agreement with the inferred matrix element for two-neutrino double beta decay共in the rare cases where this decay has been observed兲. We have no way of knowing for sure what the improved agreement for the two- neutrino case implies for the neutrinoless double beta decay matrix element and whether, indeed, the agreement in a spe- cial case is accidental or not.¹

We recognize that different individuals may regard the calculated range of nuclear matrix elements as either too nar- row or too broad to reflect the actual uncertainty. However, we do not know of any way to settle objectively and conclu- sively whether our estimate of the uncertainty is pessimistic or optimistic in any particular case.

*Email address: [email protected]

†On leave of absence from Department of Physics, University of California, Berkeley, California 94720, USA. Email address:

[email protected]

‡Email address: [email protected]

1Fukugita and Yanagita 关20兴 note that the nuclear levels that are important for neutrinoless double beta decay are typically at exci- tation energies of order 10 MeV, while for two neutrino double beta decay the characteristic excitation energies are lower, a few MeV.

Thus even if the lower excitation states are correctly described, there is no guarantee that the higher excitation states are also cor- rectly described.

1550-7998/2004/70共3兲/033012共12兲/$22.50 70 033012-1 ©2004 The American Physical Society

B. Some definitions

The neutrino mass matrix element that appears in neu- trinoless double beta decay 关21–23兴 is given by

兩具^mee␯ 典兩⫽me

1

冑^T1/2F_N⫽me

冑

F_N⫽G^0␯

冏

冉

冊

冏

For specificity, we consider a neutrinoless double beta de- cay experiment with sensitivity to T_1/2F_Nthat is exemplified by what is expected for the Majorana experiment 关5兴 共see also, compilation in Ref.关22兴兲. We will consider that a num- ber N_{ex p} of neutrinoless double beta decay experiments are performed with the expected Majorana sensitivity s.

Our results are, however, general. If the experiments that actually are or could be performed have a different sensitiv- ity, then our results should be rescaled by

N_{ex p}⬘ ⫽Nex ps/s⬘^. 共3兲 If a specific neutrinoless double beta decay experiment suc- cessfully detects a signal, then a greatly increased exposure with the same detector will not improve much the confidence

with which one can answer the questions raised in this paper.

For a single detector, the uncertainty will be dominated by the nuclear factor of that nucleus. Measurements with differ- ent nuclei will be required to improve the statistical signifi- cance of the answers to questions about the nature and prop- erties of neutrinos. On the other hand, suppose the search for neutrinoless double beta decay is negative with a given de- tector. Then an increase in the exposure time by a factor N_exposure is equivalent to performing N_exposure new experi- ments that have the identical sensitivity.

The neutrino mass element 兩具^mee␯ 典兩 is related to the fun- damental neutrino parameters by the expression

兩具^mee␯ 典兩⫽兩m1兩Ue1

2 兩e^i␾1⫹m2兩Ue2

2 兩e^i␾2⫹m3兩Ue3 2 兩兩, 共4兲 where m_i are the mass eigenvalues of the Majorana neutri- nos, U is the lepton mixing matrix, and ␾i are relative Ma- jorana phases. Normal 共inverted兲 hierarchy corresponds to the ratio between mass eigenvalues 共labeled in increasing mass eigenvalue m₁⬍m2⬍m3) given by m₃/m₂⬎m2/m₁ (m₃/m₂⬍m2/m₁). If hierarchies are indistinguishable, what happens when ⌬mi j

2Ⰶm1

2? Then, the mass scheme is called degenerate.

C. The dispersion in calculated nuclear matrix elements The dispersion of the calculated nuclear matrix elements obtained by different theoretical methods is large. For ex- ample, a compilation of 20 different calculations 关5,24,25兴 for⁷⁶Ge spans the range 2.7⫻10^⫺15⫺2.9⫻10^⫺13yr^⫺1.

Figure 1 shows the distribution of ⁷⁶Ge nuclear factors binned in a logarithmic scale. In our analyses of how much TABLE I. Answers to some questions about the potential of neutrinoless double beta decay experiments.

Answers refer to a C.L. of 99.73 % C.L. for the assumed probability distributions. We adopt a sensitivity s equal to what is projected for the Majorana experiment关5兴 共if another reference sensitivity s⬘is assumed, the required number of experiments should be scaled by N_{ex p}⬘ ⫽Nex ps/s⬘). If the answer for an inverted neutrino mass hierarchy is different from the answer for a normal mass hierarchy共see Fig. 2兲, we show in parentheses the answer for a normal mass hierarchy.

Section Assumptions Question N_expat 99.73 % C.L.

II No detected neutrinoless double␤ decay Dirac ? 230 (⬁) III lightest mass scale (1⫾0.05 eV),

No neutrinoless double␤ decay Dirac ? 1

III lightest mass scale (0.35⫾0.07 eV), No neutrinoless double␤ decay

Dirac ? 5共6兲

III lightest mass scale (0.3⫾0.1 eV),

No neutrinoless double␤ decay Dirac ? 16 (⬁)

IV Neutrinoless double␤ decay:

T_1/2(⁷⁶Ge)⫽(3.2⫾0.2)⫻10²⁵yr

Total mass ? 关0.46,9.56兴 共关0.48,9.58兴兲 IV Neutrinoless double␤ decay:

T1/2(⁷⁶Ge)⫽(1.⫾0.1)⫻10²⁶yr

Total mass ? 关0.24,8.34兴 共关0.28,8.40兴兲 IV Neutrinoless double␤ decay:

T_1/2(⁷⁶Ge)⫽(3.2⫾0.5)⫻10²⁶yr

Total mass ? 关0.08,5.68兴 共关0.16,6.06兴兲

V Detected neutrinoless double␤ decay Hierarchy ? No

V Detected neutrinoless double␤ decay, private communication: m⫽0

Hierarchy ? Yes

we can learn about different fundamental neutrino questions, we will also consider F_N as a random variable in linear and logarithmic scales of the constant and the Gaussian probabil- ity distributions. For the Gaussian distribution, we will adopt the central value of the F_Ninterval as the mean and one third of the radius of the interval covered by calculated values of F_N as the standard deviation. The lowest nuclear factor F_N shown in Fig. 1 corresponds to a recent calculation关26兴 that used a self-consistent renormalized quasiparticle random phase approximation. We do not know of any rigorous argu- ment that would exclude this recent calculation while includ- ing the other calculations shown in the figure.

For the numerical calculations given in this paper, we used the distribution of calculated nuclear factor for ⁷⁶Ge because this nucleus is the one for which we found the larg- est number of published calculations of F_N. We performed Kolmogorov-Smirnov tests to test if the distributions of F_N that were calculated for other double beta decay candidates (⁸²Se,¹³⁰Te,¹³⁶Xe) are consistent with the distribution shown in Fig. 1 of ⁷⁶Ge. Table 2 of Ref.关22兴 compiles a list of six calculations 关27兴 for these nuclei. The Kolmogorov- Smirnov tests show that we cannot reject at 95 % C.L., for any of the nuclei ⁸²Se,¹³⁰Te, or ¹³⁶Xe, the hypothesis that the distribution of calculations of F_Ngiven in Table 2 of Ref.

关22兴 is the same distribution as shown in Fig. 1 for ⁷⁶Ge. We also checked that the distribution of the six calculations listed in Table 2 of Ref.关22兴 for ⁷⁶Ge is consistent with the distribution of 20 calculations of F_N used in the present work.

The fact that the uncertainty in the nuclear matrix element plays a major role in our ability to resolve fundamental ques- tions in neutrinoless double beta decay experiments is well known 共see for example the famous reviews in Ref. 关23兴兲.

Reference关28兴 is the most recent example with which we are familiar of a systematic analysis that assumes a small uncer- tainty in the nuclear matrix elements for neutrinoless double beta decay experiments 共for the nuclear physics discussion see Ref.关29兴兲. The discussion in Ref. 关28兴 assumes the cor- rectness of the renormalized quasiparticle random phase ap-

proximation共RQRPA兲 that leads to the lowest nuclear factor in Fig. 1. Readers who are optimistic regarding the validity of current calculational methods for calculating nuclear ma- trix elements in neutrinoless double beta decay may prefer the conclusions of Ref.关28兴 instead of the more conservative conclusions of the present paper.

The position adopted in this paper is that the RQRPA could be accurate, or some other calculational scheme could be more accurate, but we will not know for sure how precise any approximation is until calculations can be done in a con- trolled manner using QCD. Our attitude is consistent with the point of view expressed in the recent discussion of the RQRPA and QRPA approximations in Ref. 关29兴. These au- thors summarized their analysis with the statement 关29兴:

‘‘Even though we cannot guarantee this basic method 关RQRPA兴 is trustworthy, we have eliminated, or at least greatly reduced, the arbitrariness commonly present in pub- lished calculations.’’ In other words, the recommended pre- scription results in a small dispersion in calculated nuclear matrix elements, which may or may not be close to the true value.

The reader will chose what to believe based upon the reader’s convictions about the accuracy of the calculations of nuclear matrix elements. We believe that the burden of proof is on the person drawing conclusions that depend on the size of the nuclear matrix elements. The conclusions must be sup- ported by a proof that the matrix elements are equal to the QCD values within the stated errors.

Our goal is to provide, for the reader’s consideration, an alternative viewpoint to the one that is usually adopted in discussing neutrinoless double beta decay experiments. As far as we know, there is no previous systematic, quantitative study to evaluate the impact of the uncertainty in the nuclear matrix element for different assumed probability distribu- tions. Recently, it has been demonstrated that it is not prac- tical to detect in neutrinoless double beta decay experiments neutrino CP violation arising from Majorana phases关30,31兴.

D. How do we determine how many experiments are required?

For each question about neutrino properties that we ad- dress, we make specific assumptions about what is or is not observed experimentally. Depending on the particular ques- tion we are addressing, we will assume that the neutrino masses satisfy a normal or an inverted hierarchy, as illus- trated in Fig. 2. We will also make assumptions regarding the observation, or nonobservation, of a neutrino mass in ordi- nary共tritium兲 beta-decay.

Figure 3 shows the relationship between the neutrinoless double beta decay mass element 兩具^mee␯ 典兩 and the smallest neutrino mass m关32–34兴. This figure plays a key role in our discussion; we will return to Fig. 3 in Secs. II, IV, and V.

For a given set of assumptions as described above, we compute the different probability distributions that are im- plied by the assumed experimental constraints. In the final step of our analysis, we combine the computed probability distributions in order to determine how many experiments FIG. 1. Distribution of ⁷⁶Ge nuclear factor results. A compila-

tion of 20 different calculations spans in the range 2.7⫻10^⫺15

⫺2.9⫻10^⫺13yr^⫺1关5兴.

are required to answer a stated question at a specific confi- dence level.

E. What is the bottom line?

Some readers will only care about the bottom line. How many neutrinoless double beta decay experiments are re- quired in order to determine whether neutrinos are Majorana or Dirac particles?²What fraction of the closure mass of the universe do neutrinos constitute? Can we establish whether the neutrino masses satisfy a normal or an inverted mass hierarchy?

Table I summarizes our numerical results. We state in column 2 of Table I the different assumptions that we have made about future experiments. In column 3, we give abbre- viated names to the questions that we have asked. Finally, in column 4, we present a brief summary of our answers to the different physical questions about neutrinos. The reader in- terested in the details of how a specific question was an- swered can look in the section of this paper that is listed in column 1 of Table I.

F. Outline of this paper

In Sec. II, we show that an impractically large number of neutrinoless double beta decay experiments would be re- quired to show that neutrinos are Dirac particles if next gen- eration experiments do not reveal neutrinoless double beta decay. We show in Sec. III that nonobservation of neutrino- less double beta decay taken together with a measurement in ordinary beta decay of the lowest neutrino mass that is near the present upper limit共e.g., ⬃1 eV) would be sufficient to

show that neutrinos are Dirac particles. However, if the neu- trino mass is as low as 0.3 eV or lower, then many neutrino- less double beta decay experiments would be required to show that neutrinos are Dirac particles. We present in Sec. IV the allowed ranges in the total mass in neutrinos if neutrino- less double beta decay is detected at different possible half- lives. Finally, we show in Sec. V that even if neutrinoless double beta decay is observed in next generation experi- ments we nevertheless will not be able to decide from beta decay experiments alone whether the mass hierarchy is nor- mal or inverted. We discuss our principal results in Sec. VI.

II. ARE NEUTRINOS DIRAC PARTICLES?

NO NEUTRINOLESS DOUBLE BETA DECAY AND INVERTED HIERARCHY

In this section, we assume that next generation experi- ments 关16兴 will not observe neutrinoless double beta decay.

Figure 3 shows that it is much easier to observe neutrinoless double beta decay if the neutrino mass hierarchy is inverted.

If the hierarchy is normal, then the neutrino mass matrix element, 兩具^mee␯ 典兩, can be unobservably small even if neutri- nos are Majorana particles, making it impossible to decide for a normal hierarchy whether neutrinos are Dirac are Ma- jorana. Hence, we concentrate our numerical calculations in this section on the case in which the mass hierarchy is known to be inverted from long baseline experiments 关36,37兴 or from some other measurement.

2Note that we are referring to the dominant neutrino masses rel- evant to the currently observed neutrino oscillation. Even if the dominant masses are Dirac, there may be much smaller Majorana masses not relevant to neutrino oscillation, sometimes called pseudo-Dirac. We do not distinguish Dirac and pseudo-Dirac neu- trinos in this paper.

FIG. 2. Normal and inverted neutrino mass hierarchy. The larger splitting is ⌬matm

2 ⯝2⫻10^⫺3 eV²; the smaller splitting is ⌬msolar 2

⯝7⫻10^⫺5 eV². The hierarchy is referred as ‘‘Degenerate’’ if the square of the smallest mass is much larger than either ⌬matm

2 or

⌬msolar

2 .

FIG. 3. The neutrinoless double beta decay mass element 兩具^mee␯典兩 versus the lowest neutrino mass m. The regions allowed at 90 % C.L. by existing neutrino oscillation data are shown for a nor- mal neutrino hierarchy 共NH兲, an inverted hierarchy 共IH兲, and de- generate neutrinos共D兲 共see Fig. 2 for an explanation of the different hierarchical arrangements of neutrino masses and Sec. V for a de- scription of how the allowed regions were computed兲. The hatched area shows the parameter space that can be excluded by the Katrin experiment关35兴 if no evidence for a neutrino mass is detected in tritium beta decay. The three dashed lines labeled共a兲, 共b兲, and 共c兲 refer to three possible positive results for a next generation neu- trinoless double beta decay search and are discussed in Secs. IV and V. The dotted horizontal line near 10^⫺2 eV illustrates the sensitivity that is expected for the Majorana experiment 关5兴. For an original version of this figure, see Ref.关32兴.

For definiteness and in order to minimize the number of required experiments, we assume that the data are free of all background and that there are no candidate neutrinoless double beta decay events. The decay constant␭ then satisfies an exponential probability decay function 共pdf兲 correspond- ing to the Poisson probability that no events are observed.

Given that we know that there is an inverted mass hierar- chy for neutrinos, how many neutrinoless double beta decay experiments would be needed to establish that neutrinos are Dirac particles at a given C.L.? We shall see that in this case 230 neutrinoless double beta-decay experiments are required in order to establish that neutrinos are Dirac particles at a C.L. equivalent to 3␴. If we admitted that there is a possi- bility that the neutrino hierarchy is normal, then an essen- tially infinite number of experiments would be required.

In order to calculate the required number of experiments, we first compute the probability distribution function of the neutrino mass element 兩具^mee␯ 典兩 given by Eq. 共1兲. This pdf depends upon the assumed distribution of the nuclear matrix element F_N.

Figure 4 illustrates the dependence of the probability dis- tribution function of兩具^mee␯ 典兩 on the pdf of the nuclear factor F_N. We show the calculated pdfs for 兩具^mee␯ 典兩 that follow

from Eq. 共1兲 for different assumptions about the pdf of FN: a Gaussian 共full line兲, a constant probability spanning the entire range of calculated F_N 共dashed line兲, and a pdf equal to the actual reported distribution of F_N共dashed line兲. In the computations shown in the upper and lower left-hand corners of Fig. 4, the value of F_N is treated as the random variable with the illustrated pdf, while in the upper and lower right- hand corners, the pdfs were calculated treating the logarithm of F_Nas the random variable. The two lower panels of Fig. 4 are similar to the two upper panels except for the fact that the lower panels refer to the pdfs computed assuming ten equivalent experiments 共equal sensitivity兲 have been per- formed instead of just one experiment.

We next concentrate on the neutrino mass element兩具^mee␯ 典^兩 as a function of the neutrino parameters. In the case of in- verted hierarchy, the appropriate expression for 兩具^mee␯ 典^{兩 is} given by

兩具^mee␯ 典兩IH⫽兩m sin²␪13

⫹cos²␪13共cos²␪_䉺冑^m2⫹⌬matm

2 ⫺⌬m_䉺²e^i␾1

⫹sin²␪_䉺冑^m2⫹⌬matm

2 e^i␾2)兩, 共5兲

where m is the mass of the lowest mass eigenstate;⌬m_䉺² and

⌬matm

2 are mass-squared splittings; and␪_䉺^and␪13are mix- ing angles determined by solar, atmospheric, reactor, and K2K experiments 关38兴. We have computed numerically the pdf of the neutrino mass element that corresponds to Eq.共5兲.

In this computation, we used Gaussian distributions for themass-squared splittings and mixing angles, with mean values and standard deviations given by ⌬m_䉺²⫽(7.1⫾0.7)

⫻10^⫺5 eV², ⌬matm

2 ⫽(2.0⫾0.4)⫻10^⫺3eV², sin²␪_䉺⫽0.30

⫾0.03, and sin²␪13⫽0.008⫾0.02 关38–40兴. In the latter case, we truncate the Gaussian distribution to include only positive values. We assumed constant probability distributions for the lightest mass m 共in logarithmic scale, with 10^⫺6⬍m

⬍2.3 eV) and the phases␾1 and␾2 共in linear scale兲.

FIG. 4. Probability distributions of兩具^mee␯典兩 in future generation neutrinoless double beta decay experiments. On the right side of each of the four panels, we plot the pdf of兩具^mee␯典兩, which is ob- tained from neutrino oscillation data by using Eq. 共5兲 共dashed- dotted line兲. On the left side of each of the four panels of the figure, we show probability distribution functions for 兩具^mee␯典^{兩 assuming} that next generation experiments do not detect neutrinoless double beta decay. The plotted pdfs were obtained by making different assumptions regarding the pdf of the nuclear factor F_Nthat appears in Eq.共1兲. For the left-hand-side panels, we assume that FNfollows a Gaussian distribution 共full line兲, a constant distribution 共dotted line兲, or the actual computed distribution of values of FNcomputed by different nuclear theorists共dashed line兲. For the right-hand-side panels, we assumed that log F_N follows these same distributions.

The upper pair of panels corresponds to a single experiment with a sensitivity equal to what is expected for the Majorana experiment 关5兴. The lower panels correspond to simulated results for ten experi- ments each with a sensitivity equal to the anticipated sensitivity of the Majorana experiment关5兴.

TABLE II. No neutrinoless double beta decay plus inverted hi- erarchy. The table gives the number of neutrinoless double beta decay experiments with sensitivity to兩具^mee␯典兩 equal to what is pro- jected for the Majorana 关5兴 experiment that are required to show that neutrinos are not Majorana particles at 90, 95, 99, and 99.73 % C.L. if an inverted hierarchy is correct 共see Fig. 2兲. We consider different probability distributions of the nuclear factor, F_N: Gaussian, constant, or the actual distribution of 20 different calculations 共see Fig. 1兲; either using linear 共lin兲 or logarithmic 共log兲 scales.

FN pdf

N_expat 90 % C.L.

N_expat 95 % C.L.

N_expat 99 % C.L.

N_expat 99.73 % C.L.

actual, lin 11 21 81 230

actual, log 9 17 61 141

Gaussian, lin 3 4 8 13

Gaussian, log 16 23 50 83

constant, lin 4 7 21 45

constant, log 24 40 95 156

In order to compute the number of experiments re- quired to establish that neutrinos are Dirac particles, we must compute the joint probability P that follows from the unob- served neutrinoless double beta decay experiments 关see Eq. 共1兲兴 and from the expression for the neutrino mass

兩具^mee␯ 典兩 in terms of various neutrino oscillation parameters 关see Eq. 共4兲兴. The probability P that these two conditions are satisfied is given by the product of the probabilities of the two individual constraints, conveniently normalized.

Thus

P共Nex p兲⫽

冕

冕

冑 _冕

冑 _冕

In Table II we show the number of experiments needed to reject, at 90, 95, 99, and 99.73 % C.L. (3␴), the hypothesis that neutrinos are Majorana particles. In all cases, we find that many experiments are required in order to establish that neutrinos are not Majorana particles. However, there is a wide dispersion, from 13 experiments to 250 experiments at 3␴, in the number of experiments that is required depending on the assumed pdf of the nuclear factor F_N.³

For the bottom-line table, Table I, we adopt the most con- servative case in which the assumed probability distribution function for F_Nis given by the actual calculated distribution of F_N values. In any event, Table I shows that a few next- generation neutrinoless double beta decay experiments will not be able to answer the question of whether neutrinos are Dirac or Majorana particles unless neutrinoless double beta decay is actually observed.⁴

What could be the effect of future improvement in the neutrino oscillation parameters, such as sin²␪12 from a SNO study with neutral current detectors 共NCD兲 关43兴, and ⌬m23 2

from NuMI/MINOS 关44兴 and T2K experiments 关45兴? In curves labeled ‘‘Eq. 4’’ in Fig. 4, the two peaks correspond to the maximally constructive共right兲 and destructive 共left兲 case

in Eq. 共5兲, connected by a plateau due to the randomly as- signed complex phases. On the other hand, the tails above and below the peaks are mostly due to the uncertainties in sin²␪12and⌬m23

2 . It is clear that the improvements in mea- surements cannot change the situation qualitatively. We have performed the same analysis with twice as accurate measure- ments or with no errors at all, and found that the numbers in Table II cannot change more than 40 %. For example, if we assume that all of the neutrino oscillation parameters are known with infinite precision, the required number of experi- ments to obtain a 3␴ result is reduced from 230 to 156.

III. ARE NEUTRINOS DIRAC PARTICLES?

NO NEUTRINOLESS DOUBLE BETA DECAY, BUT NEUTRINO MASS MEASURED In this section, we make two assumptions.

共i兲 Next generation experiments 关16兴 do not observe neu- trinoless double beta decay.

共ii兲 Next generation beta decay experiments 关35兴 observe the neutrino mass scale.

The first assumption is identical to our first assumption in Sec. II. The second assumption assumes that an experiment with the expected sensitivity of the KATRIN experiment关35兴 will successfully identify a spectral distortion of the tritium beta decay energy spectrum that is due to a finite neutrino mass.

Given the measurement of a neutrino mass in the KATRIN experiment, how many double beta experiments would we need to establish that neutrinos are Dirac particles at a given C.L.?

We will consider three cases. First, m⫽1 eV, which is chosen because this value is close to the present upper bound for a neutrino mass in ordinary beta decay关46,47兴. Second, m⫽0.35 eV, which is chosen because this is the smallest mass that could be discovered at 5␴ in next-generation ex- periments that perform with the sensitivity of the KATRIN experiment 关35兴. Third, m⫽0.30 eV, which is chosen be- cause it is the smallest mass that could be discovered at 3␴ in a next-generation experiment with the expected KATRIN sensitivity 关35兴. We assume that m is normally distributed

3We checked our results by comparing with a conservative case.

We assumed that the lightest neutrino mass is zero, neglected␪13

and the solar mass splitting, and chose the Majorana phase to be␲.

In this special case, the neutrinoless double beta mass element has a lower limit 关32–34,39,41,42兴. We can compute straightforwardly the probability that the mass matrix element derived from negative searches is higher than the lower bound at a given confidence level.

As expected, this calculation indicates more experiments are re- quired than we found are necessary using the full probability distri- butions. For example, in the case ‘‘actual, lin’’ the calculation using the lower bound gives 14共550兲 required experiments at 90 % C.L.

(3␴).

4We checked that our results do not depend very sensitively upon the assumption that equal decades in the lightest mass m are equally probable. We made instead the extreme assumption that the pdf of the mass m is equally distributed on a linear scale with 0⬍m

⬍2.3 eV. This optimistic assumption presumes that there is a 50 % chance that the lowest mass lies between 1.15 eV and 2.3 eV. Nev- ertheless,the required number of experiments at 3␴ is 81.

with a mean value of 1.0共0.35兲 关0.3兴 eV and standard devia- tion 0.05 共0.07兲 共0.10兲 eV for the three cases listed in the order given above.⁵

The last two cases, which are given as examples in the Majorana proposal关5兴, are separated by only 0.05 eV. How- ever, as we shall see in the discussion below, this small dif- ference in mass makes a large difference in the number of required experiments. The essential reason for this large dif- ference is that if the experiment shows that m⫽0.35

⫾0.07 eV, then we know that m is well separated from zero mass at 3␴. However, if m⫽0.30⫾0.10 eV, then at 3␴ ^the lightest mass could be zero.

We are now in a position to compute the required number of neutrinoless double beta decay experiments. Compared to the analysis done in Sec. II, we need only modify our analy- sis of Eq.共5兲, replacing the pdf assumed in Sec. II for m by the corresponding Gaussian distribution that represents one of the three cases listed above for a next-generation beta decay experiment. Moreover, in this section we do calcula- tions for both neutrino mass hierarchies, normal, and in- verted 关given by Eq. 共5兲兴. For a normal hierarchy, the neu- trino mass element can be written as

兩具^mee␯典兩NH⫽兩cos²␪13共m cos²␪_䉺⫹冑^m2⫹⌬m_䉺² sin²␪_䉺^ei␾1兲

⫹冑^m2⫹⌬matm

2 ⫹⌬m_䉺² sin²␪13e^i␾2兩. 共7兲

For the first case, m⫽1.0⫾0.05 eV, one experiment is sufficient to prove that neutrinos are Dirac particles at more than 3␴ ^{( P}⬎99.77 %).

Table III presents for the second and third cases the re- sults for different assumptions about the pdf of F_N. For the second case, m⫽0.35⫾0.07 eV, one experiment is sufficient to prove that neutrinos are Dirac particles at 95 % C.L., but six experiments are required to prove that neutrinos are not Dirac particles at 3␴^. For the third case, m⫽0.30

⫾0.10 eV, and assuming an inverted hierarchy, 2 experi- ments are sufficient to prove neutrinos are Dirac particles at 90 % C.L., but 16 experiments are required to prove that neu- trinos are not Majorana particles at 3␴^.

The differences between hierarchies are small in the first and second cases listed above because the mass scale m is assumed large compared with the solar and atmospheric mass splittings. In these two cases, the neutrino masses are essentially degenerate 共imagine Fig. 2 for the case in which m is much larger than either the solar or the atmospheric mass splitting兲.

For the third case, m⫽0.030⫾0.10 eV, there are large quantitative differences between the normal hierarchy and the inverted hierarchy at higher C.L. The reason for the dif- ference in behavior can be seen visually in Fig. 3. The mass matrix element兩具^mee␯ 典兩 in a normal hierarchy can be greatly reduced because of cancellations, while m is bounded from below in an inverted mass hierarchy. For a normal hierarchy and small values of m, 兩具^mee␯ 典兩 could be extremely small, orders of magnitude below the expected level of sensitivity of next-generation double beta decay experiments. There- fore, the corresponding entries in Table III require at 3␴^C.L.

an infinite number of next-generation double beta decay ex- periments to distinguish between Dirac and Majorana par- ticles.

IV. WHAT IS THE TOTAL MASS IN NEUTRINOS?

NEUTRINOLESS DOUBLE BETA DECAY DETECTED We suppose in this section that next-generation experi- ments关16兴 successfully detect neutrinoless double beta with a large neutrino mass matrix element 兩具^mee␯ 典兩.

We will compute in this section the pdf for the lowest mass eigenstate m using hypothesized results from next- generation neutrinoless double beta decay experiments.

Since we already know from existing experiments the pdf for

5It is possible that m²is normally distributed at the same signifi- cance rather than m. Then m⫽0.3⫾0.1 eV is replaced by m²

⫽0.09⫾0.03 eV², and the effective error in m is reduced. In this case, we find that the number of experiments required to conclude neutrinos are Dirac particles is slightly larger关differ at most in 1共2兲 experiment共s兲 at 99 共99.73兲 % C.L.兴 than in the case m⫽0.35

⫾0.07 eV shown in Table III. On the other hand, it remains true that the sensitivity to the Majorana character of neutrinos quickly runs out of steam below 0.3 eV.

TABLE III. No neutrinoless double beta decay plus measured neutrino mass in ordinary beta decay. What do we need to know to conclude in this case that neutrinos are Dirac particles? The format of the table is similar to Table II except that for Table III we assume that a neutrino mass, m, has been detected in ordinary beta decay.

The table gives results for two hypothesized cases, m⫽0.35

⫾0.07 eV 关35兴 and m⫽0.30⫾0.10 eV 关35兴. If the required number of experiments depends on whether the neutrino mass hierarchy is normal or inverted, then the result for the normal hierarchy is writ- ten in parentheses. The case m⫽1.0⫾0.05 eV is discussed in Sec.

III.

F_N pdf

N_expat 90 % C.L.

N_expat 95 % C.L.

N_expat 99 % C.L.

N_{ex p}at 99.73 % C.L.

m⫽0.35⫾0.07 eV

actual, lin 1 1 2 5共6兲

actual, log 1 1 2 3共4兲

Gaussian, lin 1 1 1 1

Gaussian, log 1 1 2 2

constant, lin 1 1 1 1

constant, log 1 1 2 4

m⫽0.3⫾0.1 eV

actual, lin 1 2 5共14兲 15(⬁)

actual, log 1 1共2兲 4共10兲 9(⬁)

Gaussian, lin 1 1 1共3兲 2(⬁)

Gaussian, log 1 2 5共13兲 10(⬁)

constant, lin 1 1 2共4兲 4(⬁)

constant, log 1共2兲 2共3兲 7共19兲 16(⬁)

the mass splittings⌬m_䉺² and⌬matm

2 , we can use these data together with the results for the lowest mass m to compute the cumulative pdf for the total mass in neutrinos.

We will use ⁷⁶Ge as an illustrative case. The Heidelberg- Moscow experiment关48兴 provides a lower limit on the half- life共we remind the reader that there is a claim of 4␴ ^detec- tion in Ref. 关49兴兲,

T_1/2⬎1.9共3.1兲⫻10²⁵ yr 共8兲 at 90 % C.L. 共68 % C.L.兲.

We consider three feasible cases with positive neutrino- less double beta detection:共a兲 T1/2⫽(3.2⫾0.2)⫻10²⁵yr, 共b兲 T_1/2⫽(1.⫾0.1)⫻10²⁶yr, and 共c兲 T_1/2⫽(3.2⫾0.5)

⫻10²⁶yr, corresponding to 373, 118, and 37.3 events ex- pected in the parameter region of interest in the Majorana experiment 关5兴. The expected background in a deep under- ground experiment is 5.5 events, although background could be different by a factor of two. Systematic errors are ex- pected to be a few percent and to be dominated by energy resolution, the segmentation cut, and the pulse-shape dis- crimination acceptance. Our results are not significantly af- fected by including systematic errors of a few percent be- cause of the dominant contribution of the uncertainty in the nuclear factor F_N.

We computed numerically the pdf of the neutrino mass element 兩具^mee␯ 典兩 given by Eq.共1兲 for all three values of T1/2

listed above and for all six possible distributions of the nuclear factor F_Nthat were discussed in Secs. II and III. The determination of the neutrino mass element can be used to extract the probability distribution P(m) of the lightest mass eigenstate by extracting with the help of Eqs.共5兲 and 共7兲. In order to find P(m), we compute

P共m兲⫽

冕

冕

冑 _冕

冑 _冕

where P₁(兩具^mee␯ 典兩) is the pdf of the neutrino mass element given by Eq.共1兲 and P2(兩具^mee␯ 典兩,m) is the pdf of the neutrino mass element given by Eq. 共7兲 or 共5兲, respectively, for a normal or an inverted neutrino mass hierarchy.

Figure 5 shows the computed pdfs of the lightest mass eigenstate in the case of normal and inverted neutrino mass hierarchies for different assumptions regarding the pdf of the nuclear factor F_N. For illustrative purposes, we assumed in making the figure that T_1/2⫽(3.2⫾0.5)⫻10²⁶ yr 共case c above兲. The two other lifetimes considered above 共case a and case b兲 result in pdfs with very similar shapes, but shifted relative to Fig. 5 to larger values of m, the lightest neutrino mass.

We can also extract from our analysis the allowed ranges

of the total mass in neutrinos at a given C.L. Table IV pre- sents the allowed ranges for the total mass in neutrinos for different assumptions regarding the pdfs of the nuclear factor F_N and for the three values of the half-life for neutrinoless double beta decay assumed above共cases a–c兲.

Figure 6 shows the cumulative probabilities for the total mass in neutrinos M ( M⫽m1⫹m2⫹m3). The results are illustrated for different assumptions regarding the pdf of F_N and for both normal and inverted neutrino mass hierarchies.

In constructing Fig. 6, we assumed that T_1/2⫽(3.2⫾0.5)

⫻10²⁶ yr 共case c above兲. For the shorter half-lives corre- sponding to cases a and b above, the cumulative probabilities have very similar shapes, but are shifted to larger values of the total neutrino mass.

FIG. 5. Probability distributions of the lightest mass eigenstate m in a future generation neutrinoless double beta decay experi- ments. For illustration, we assumed a measured half life T_1/2(⁷⁶Ge)⫽3.2⫾0.5⫻10²⁶yr. We obtained the distributions with the aid of Eq.共9兲 共see text for details兲. For the left hand panels, we assume that the probability distribution function for the nuclear fac- tor F_Nthat appears in Eq.共1兲 satisfies a Gaussian distribution 共full line兲, a constant distribution 共dotted line兲, or the distribution of computed nuclear factor calculations by different theoretical groups 共dashed line兲. The right hand panels correspond to assuming that log F_Nfollows those distributions. The upper panels correspond to a normal neutrino mass hierarchy 关Eq. 共7兲兴, while the lower panels correspond to an inverted hierarchy关Eq. 共5兲兴.

V. NORMAL OR INVERTED MASS HIERARCHY?

NEUTRINOLESS DOUBLE BETA DECAY DETECTED, BUT NEUTRINO MASS NOT MEASURED We make two assumptions in this section.

共i兲 Next-generation experiments 关16兴 will observe neutrino- less double beta decay.

共ii兲 Next-generation ordinary beta decay experiments 关35兴 will not detect the neutrino mass scale.

These assumptions are the opposite of what we postulated in Sec. III.

In this section, we answer the following question. Given the detection of neutrinoless double beta decay and the non- detection of a neutrino mass in normal beta decay, can we determine if the neutrino mass hierarchy is normal or in- verted?

In order to answer this question, we computed the ␹² distribution as a function of the different neutrino variables, including neutrino oscillation data where available for⌬m_䉺² ,

⌬matm

2 , ␪_䉺^,␪13, the lightest mass m, the Majorana phases

␾1 and ␾2, and the neutrinoless mass 兩具^mee␯ 典兩. For an in- verted共normal兲 neutrino mass hierarchy, we imposed Eq. 共5兲 关Eq. 共7兲兴. We then marginalized over all variables except m and兩具^mee␯ 典^兩).

TABLE IV. Allowed ranges of the total mass in neutrinos for different assumed measurements of the half-life of ⁷⁶Ge to neutrinoless double beta decay. We consider different probability distributions of the nuclear factor: Gaussian, constant, or the actual distribution of 20 different theoretical calculations, using either a linear共lin兲 or a logarithmic 共log兲 scale for FN. In general, the results are different for normal and for inverted neutrino mass hierarchies. The results for the normal hierarchy are written in parentheses.

F_Npdf M 共eV兲 at 90 % C.L. M 共eV兲 at 95 % C.L. M 共eV兲 at 99 % C.L. M 共eV兲 at 99.73 % C.L.

T_1/2⫽(3.2⫾0.2)⫻10²⁵yr

actual, lin 关0.63,4.70兴 共关0.65,4.73兴兲 关0.56,5.96兴 共关0.58,5.99兴兲 关0.48,8.58兴 共关0.50,8.60兴兲 关0.46,9.56兴 共关0.48,9.58兴兲 actual, log 关0.64,4.22兴 共关0.66,4.24兴兲 关0.57,5.36兴 共关0.59,5.39兴兲 关0.49,7.84兴 共关0.51,7.88兴兲 关0.46,9.41兴 共关0.48,9.43兴兲 Gaussian, lin 关0.62,2.07兴 共关0.63,2.09兴兲 关0.59,2.35兴 共关0.60,2.36兴兲 关0.54,3.09兴 共关0.55,3.10兴兲 关0.51,3.93兴 共关0.52,3.94兴兲 Gaussian, log 关0.96,5.17兴 共关0.98,5.17兴兲 关0.84,6.04兴 共关0.85,6.04兴兲 关0.64,7.98兴 共关0.66,7.98兴兲 关0.53,9.24兴 共关0.55,9.24兴兲 constant, lin 关0.55,2.81兴 共关0.57,2.83兴兲 关0.52,3.61兴 共关0.54,3.63兴兲 关0.48,5.90兴 共关0.50,5.92兴兲 关0.46,7.88兴 共关0.48,7.90兴兲 constant, log 关0.63,6.45兴 共关0.65,6.48兴兲 关0.57,7.74兴 共关0.59,7.75兴兲 关0.51,9.45兴 共关0.53,9.45兴兲 关0.48,9.83兴 共关0.50,9.83兴兲 T_1/2⫽(1.0⫾0.1)⫻10²⁶yr

actual, lin 关0.34,2.76兴 共关0.37,2.81兴兲 关0.30,3.63兴 共关0.33,3.69兴兲 关0.26,6.23兴 共关0.29,6.30兴兲 关0.24,8.34兴 共关0.28,8.40兴兲 actual, log 关0.34,2.46兴 共关0.38,2.52兴兲 关0.30,3.24兴 共关0.34,3.29兴兲 关0.26,6.20兴 共关0.30,6.29兴兲 关0.24,8.06兴 共关0.28,8.11兴兲 Gaussian, lin 关0.34,1.16兴 共关0.36,1.18兴兲 关0.32,1.32兴 共关0.34,1.34兴兲 关0.29,1.73兴 共关0.31,1.76兴兲 关0.28,2.21兴 共关0.30,2.24兴兲 Gaussian, log 关0.53,2.93兴 共关0.56,2.94兴兲 关0.46,3.43兴 共关0.49,3.45兴兲 关0.34,4.65兴 共关0.38,4.66兴兲 关0.28,5.70兴 共关0.32,5.71兴兲 constant, lin 关0.29,1.56兴 共关0.33,1.61兴兲 关0.28,2.01兴 共关0.31,2.07兴兲 关0.26,3.35兴 共关0.29,3.43兴兲 关0.24,4.66兴 共关0.28,4.73兴兲 constant, log 关0.34,3.80兴 共关0.37,3.86兴兲 关0.31,4.68兴 共关0.34,4.73兴兲 关0.27,6.33兴 共关0.30,6.36兴兲 关0.25,7.46兴 共关0.29,7.47兴兲 T_1/2⫽(3.2⫾0.5)⫻10²⁶yr

actual, lin 关0.14,1.45兴 共关0.22,1.60兴兲 关0.12,1.95兴 共关0.19,2.13兴兲 关0.09,3.63兴 共关0.17,3.93兴兲 关0.08,5.68兴 共关0.16,6.06兴兲 actual, log 关0.15,1.29兴 共关0.22,1.42兴兲 关0.12,1.72兴 共关0.20,1.85兴兲 关0.09,3.31兴 共关0.17,3.56兴兲 关0.08,4.51兴 共关0.16,4.67兴兲 Gaussian, lin 关0.16,0.63兴 共关0.21,0.67兴兲 关0.15,0.72兴 共关0.20,0.76兴兲 关0.12,0.95兴 共关0.18,1.00兴兲 关0.09,1.20兴 共关0.17,1.27兴兲 Gaussian, log 关0.27,1.62兴 共关0.32,1.66兴兲 关0.22,2.61兴 共关0.28,1.94兴兲 关0.13,2.61兴 共关0.22,2.64兴兲 关0.10,3.21兴 共关0.19,3.25兴兲 constant, lin 关0.13,0.79兴 共关0.19,0.91兴兲 关0.11,1.02兴 共关0.18,1.16兴兲 关0.09,1.74兴 共关0.17,1.93兴兲 关0.08,2.46兴 共关0.16,2.67兴兲 constant, log 关0.16,2.03兴 共关0.22,2.16兴兲 关0.14,2.53兴 共关0.20,2.65兴兲 关0.10,3.51兴 共关0.18,3.60兴兲 关0.09,4.18兴 共关0.17,4.26兴兲

FIG. 6. The cumulative probability that the total mass in neutri- nos is less than M. For illustration, we assumed a measured neu- trinoless double beta decay half-life T_1/2(⁷⁶Ge)⫽3.2⫾0.5

⫻10²⁶yr. We calculated the cumulative probability by integrating Eq.共9兲 共see text for details兲. The organization of the panels and the notation are the same as for Fig. 5.