Gauged Lµ − Lτ Symmetry at the Electroweak Scale

Julian Heeck∗ and Werner Rodejohann†

Max–Planck–Institut für Kernphysik,
Postfach 103980, D–69029 Heidelberg, Germany

The extension of the Standard Model by a spontaneously broken abelian gauge group based on
the Lµ − Lτ lepton number can resolve the long-standing discrepancy between experimental and
theoretical values for the magnetic moment of the muon. It furthermore naturally generates µ-τ
symmetric lepton mixing, introduces neutrino nonstandard interactions, and the associated gauge
boson Z 0 serves as a mediator to the right-handed neutrino sector. A detailed fit to electroweak data
is performed to identify the allowed values for the mass of Z 0 and its mixing with the Standard Model
Z. An economical new scalar sector is constructed that spontaneously breaks Lµ − Lτ and leads to
experimental consequences such as lepton flavor violation and collider signatures. Furthermore we
discuss the nonabelian extension to an SU (2)0 , particularly the neutrino sector.
arXiv:1107.5238v2 [hep-ph] 10 Oct 2011

I. INTRODUCTION

Extensions of the Standard Model (SM) of particle physics could either add new particles or represen-
tations, or extend the gauge sector. The enormous precision with which the SM has been tested in the
last decades, plus the various theoretical consistency conditions which have to be obeyed, require careful
addition of new physics. One particularly popular approach is the addition of an abelian gauge symmetry
U (1)0 . If this symmetry is broken (to avoid an additional force with infinite range), a massive Z 0 boson is
present, with model-dependent mass and couplings to the SM particles [1]. In this paper we focus on one
class of highly interesting U (1)0 models: within the particle content of the SM, it is possible to gauge one
of the three differences of lepton flavors Le − Lµ , Le − Lτ or Lµ − Lτ , without introducing an anomaly
[2, 3]. This surprising feature has lead to a number of works analyzing the consequences of one of those
broken symmetries [4–8]. In particular, Lµ − Lτ should be preferred over the other two combinations, be-
cause in the limit of conserved symmetry, the neutrino mass matrix is automatically µ-τ symmetric, and
predicts one degenerate neutrino pair. The necessary breaking of the symmetry will split their masses
and generate small departures from µ-τ symmetry, thereby rendering the neutrino phenomenology in
agreement with data. In contrast, if Le − Lµ or Le − Lτ are to be gauged, the neutrino mass matrix has
in the symmetry limit a structure far away from the one necessary to reproduce the experimental results.
This intimate connection of flavor and gauge symmetry is rather unique. It is worth stressing that gauged
Lµ − Lτ gives a lepton mixing structure close to observation without the usual complications of flavor
symmetries (see [9] for recent reviews), in which one typically involves nonrenormalizable terms including
a plethora of “flavon fields”, arranges for their proper vacuum expectation value (VEV) alignment by
additional input, and adds additional symmetries to avoid unwanted terms in the Lagrangian.
An important property of gauged Lµ − Lτ is that it does not act on first-generation leptons, but only
on muons and tauons. In this respect, further motivation for this model (and the main focus of previous
discussions of this gauge group [6, 7]) stems from the anomalous magnetic moment of the muon. This
measured quantity exhibits a 3.2σ difference to the theoretically predicted value, which can be explained
by the loop-contribution of a heavy ZL0 µ −Lτ gauge boson.
In the present work we study the phenomenology of gauged Lµ − Lτ in the regime in which the
anomalous magnetic moment of the muon is explained. In Section II we determine the currently allowed
parameter space for a generic U (1)Lµ −Lτ as determined by electroweak precision data, and also comment
on collider physics aspects of heavy Z 0 bosons. We propose a new and economic scalar sector to break
the symmetry spontaneously, and study the resulting neutrino sector in Section III. The scalar potential
and the Higgs spectrum is analyzed in Section IV. Section V is devoted to an extension from U (1)Lµ −Lτ
to an SU (2)0 that also acts on the electron. An overview over the used notation concerning nontrivial
gauge group representations is delegated to Appendix A, while Appendix B briefly discusses leptons in
reducible representations of SU (2)0 . Finally, we conclude in Section VI.

∗ Electronic address: [email protected]

† Electronic address: [email protected]
 2

II. GAUGE SECTOR

Extending the gauge group of the SM GSM ≡ SU (3)C ×SU (2)L ×U (1)Y by U (1)Lµ −Lτ leads to possible
Z–Z 0 mixing, even without a scalar charged under both U (1) groups [1]. This is due to kinetic mixing,
0
i.e. a gauge-invariant term ∼ Z µν Zµν in the Lagrange density L = LSM + LZ 0 + Lmix , with Z µν being
the gauge field strength tensor. If a scalar transforms nontrivially under SU (2)L × U (1)Y × U (1)Lµ −Lτ
and acquires a VEV, a mass-mixing term ∼ Z µ Zµ0 can also be generated, so the most general Lagrangian
after breaking SU (2)L × U (1)Y × U (1)Lµ −Lτ to U (1)EM takes the form:
1 1 a aµν 1 2 ê µ ê aµ a
LSM = − B̂µν B̂ µν − Ŵµν Ŵ + M̂Z Ẑµ Ẑ µ − j B̂µ − j Ŵ ,
4 4 2 ĉW Y ŝW W µ
1 0 0µν 1 02 0 0µ
LZ 0 = − Ẑµν Ẑ + M̂Z Ẑµ Ẑ − ĝ 0 j 0µ Zµ0 , (1)
4 2
sin χ 0µν
Lmix = − Ẑ B̂µν + δ M̂ 2 Ẑµ0 Ẑ µ .
2
The currents are defined as
 1 X 
jYµ = −
X 
L` γ µ L` + 2 `R γ µ `R + QL γ µ QL + 4 uR γ µ uR − 2 dR γ µ dR ,

3
`=e,µ,τ quarks

aµ
X σa X σa (2)
jW = L` γ µ L` + QL γ µ QL ,
2 2
`=e,µ,τ quarks

j 0µ = µ̄γ µ µ + ν̄µ γ µ PL νµ − τ̄ γ µ τ − ν̄τ γ µ PL ντ ,

with the left-handed SU (2)-doublets QL and L` and the Pauli matrices σ a . We also define the electric
current jEM ≡ jW3
+ 12 jY and the weak neutral current jNC ≡ 2jW 3
− 2ŝ2W jEM . We adopt the notation of
Ref. [10] with gauge-eigenstates {Â, Ẑ, Ẑ 0 } connected to the mass-eigenstates {A, Z1 , Z2 } via:
1 −ĉW sin ξ tan χ −ĉW cos ξ tan χ
    
 A
 Ẑ  = 0 cos ξ + ŝW sin ξ tan χ ŝW cos ξ tan χ − sin ξ  Z1  , (3)
sin ξ cos ξ
Ẑ 0 0 cos χ cos χ Z2

or, inverted:
    
A 1 0 ĉW sin χ Â
Z1  = 0 cos ξ −ŝW cos ξ sin χ + sin ξ cos χ  Ẑ  . (4)
Z2 0 − sin ξ cos ξ cos χ + ŝW sin ξ sin χ Ẑ 0
2b
Here the mixing angle ξ is defined as tan 2ξ = a−c with

δ M̂ 2
a ≡ M̂Z2 , b ≡ ŝW tan χM̂Z2 + ,
cos χ (5)
1  2 2 2 2 2

c≡ M̂ Z ŝW sin χ + 2ŝW sin χδ M̂ + M̂ Z 0 .
cos2 χ
The gauge boson couplings to fermions are hence changed to
e
ejEM Â + jNC Ẑ + g 0 j 0 Ẑ 0 →
2sW cW
1 −ĉW sin ξ tan χ −ĉW cos ξ tan χ
  
A (6)
ejEM , 2ŝWeĉW jNC , g 0 j 0 0 cos ξ + ŝW sin ξ tan χ ŝW cos ξ tan χ − sin ξ  Z1  .


sin ξ cos ξ
0 cos χ cos χ Z2
In the following we will for simplicity set χ = 0, as the mass mixing already shows all qualitative effects
of mixing and will be induced in our specific model in Section IV. A nonzero χ results in an additional
coupling of Z2 to the electromagnetic current which will change the constraints on sin ξ given below [1].
The mass eigenstate Z1 was studied extensively at LEP, so we know its axial and vector couplings
to leptons to a high precision.1 Compared to the SM case, the Z1 couplings become nonuniversal due

1 µ ψ ψ 5
ψ γ µ (gV
P
The axial and vector couplings can be obtained by rewriting jX Xµ = ψ − gA γ )ψ Xµ .
 3

e− , p, p µ+ , τ + µ+ , τ +

Z, γ

Z′
e+ , p, p µ− , τ − µ− , τ −

Figure 1: Detection process for an unmixed Z 0 in electron-positron (LEP) or proton-(anti-)proton (LHC and
Tevatron) collisions.

to the small admixture of the j 0 current. For a small mixing angle ξ this additional coupling to g 0 ξ j 0µ
modifies, for example, the well-measured asymmetry parameter A` = 2gV` gA `
/((gV` )2 + (gA
` 2
) ) for muons
and tauons:
   
4sW cW /e 0 4sW cW /e
Aµ → Aµ 1 − g 0 ξ , Aτ
→ A τ
1 + g ξ . (7)
1 − 4s2W 1 − 4s2W

Depending on the sign of g 0 ξ, we expect a hierarchy Aµ < Ae < Aτ or Aτ < Ae < Aµ (the SM
predicts Ae = Aµ = Aτ ), neither of which is observed [11]. This can be used to estimate a 3σ constraint
|g 0 ξ| . 10−3 .
In the unmixed case (χ = 0 = δ M̂ 2 ), the main constraint on the model stems from the anomalous
magnetic moment of the muon, since the Z 0 contributes a term [12]

Z1
g 02 1 2m2µ x2 (1 − x)
∆aµ = dx
4π 2π m2µ x2 + MZ2 0 (1 − x)
(
0 (8)
02
g 1 1 for M  mµ ,
Z0
'
4π 2π 2m2µ /3MZ2 0 for MZ 0  mµ ,

which can be used to soften the longstanding 3.2σ disagreement between experiment (aexpµ = 116592089×
10−11 ) and theoretic predictions (aSM
µ = 116591834 × 10 −11
) [6, 11] (note however the uncertainty in
the hadronic contributions to aµ [13]). The appropriate values for the case MZ 0  mµ lie around
MZ 0 /g 0 ∼ 200 GeV:
 2
200 GeV
∆aµ ' 236 × 10−11 . (9)
MZ 0 /g 0

The authors of Ref. [6] also derive a direct detection limit of roughly MZ 0 > 50 GeV, based on ALEPH
measurements of the process e+ e− → µµZ 0∗ → 4µ [14] (see Fig. 1) and under the assumption MZ 0 /g 0 ∼
200 GeV. A similar analysis using LEP2 data with higher luminosity ∼ 0.7 fb−1 [15] gives a bound of the
same order due to the low rate of 4` final states at LEP2 energies. We will comment on Tevatron and
LHC prospects of this process in Sec. II B.
Interestingly, the nonuniversality can also lead to nonstandard neutrino interactions (NSIs), which are
usually parametrized by the nonrenormalizable effective Lagrangian
√ fP  ¯ µ
Leff

NSI = −2 2GF εαβ f γ P f [ν̄α γµ PL νβ ] . (10)

Note that this Lagrangian, when written in a gauge-invariant way, introduces charged-lepton flavor vi-
olation, which usually is a big problem in phenomenological studies of NSIs. Effective four-fermion
interactions can be obtained from Eq. (1) by integrating out the (heavy) mass eigenstate Z2 after per-
forming the transformation from Eq. (6). Since the analytical expression for the NSIs are only marginally
more complicated with a nonzero χ, we will include it in the next two equations. The effective Lagrangian
for Z2 interactions takes the form
2
−1

0 cos ξ 0 e
Leff
Z2 = g j − eĉ W cos ξ tan χ jEM + (ŝW cos ξ tan χ − sin ξ) j NC . (11)
2M22 cos χ 2ŝW ĉW

Expanding the square we only need the terms linear in g 0 for the NSIs, because the others are either
diagonal in flavor space and hence do not influence neutrino oscillations (these are just the SM terms with
 4

g 0 = 0), or do not involve e, u or d and hence do not couple neutrinos to “matter” (the terms quadratic in
g 0 ). Integrating out Z1 gives similar terms, so after adding up the different contributions to the effective,
Earth-like matter NSI ε⊕ eV uV dV ⊕ ⊕ ⊕
αβ = εαβ + 3 εαβ + 3 εαβ , we obtain the two parameters εµµ and ετ τ = −εµµ with

−g 0
 2
cos2 ξ
   
e 1 1 sin ξ
ε⊕
µµ = √ cos ξ sin ξ − + ŝW tan χ + . (12)
4 2GF cos χ ŝW ĉW M12 M22 M12 M22

The analogous parameter for solar matter ε = εe + 2 εu + εd vanishes, which shows that in neutral
matter the potential is generated by the neutrons.2 While this is similar to gauged B − L symmetries,
the nonuniversality of our model could make these NSIs observable in neutrino oscillations. For instance,
taking χ = 0, g 0 ' 1/4, M2 ' 50 GeV and ξ ' 4 × 10−3 can generate

ε⊕
µµ ' 10
−2
–10−3 , (13)

which is testable in future facilities [16] and can and still resolve the magnetic moment of the muon ∆aµ .
The mixing angle ξ obviously has to be not too tiny for such an effect to be observable. In addition, the
mass of the Z 0 should not be too heavy: for M2  M1 , ξ  1 and χ = 0 Eq. (12) can be simplified
−g 0 ξ e
to ε⊕ √ 1 0
µµ ' 4 2GF sW cW M12 ' −1.5 g ξ, where the constraint g ξ . 10
0 −3
by Z-pole measurements (even
stronger limits are given below) suppresses the NSI parameter. Since the two gauge boson masses enter
with opposite sign, the NSI parameters will be even smaller in the limit M1 ∼ M2 , only M2 < M1 can
lead to NSI values closer to the current limit.
Nevertheless, there is allowed parameter space of the model allowing for testable NSI, providing a
complementary way to probe such nonuniversal gauge bosons. Note further that this renormalizable
realization of NSI parameters does, in contrast to the effective approaches as in Eq. (10), not suffer from
charged-lepton flavor violation decays, due to the diagonal structure of the NSI parameters εαβ in flavor
space, as imposed by our symmetry. Consequently, lepton flavor violation will enter into our model only
via the symmetry breaking sector, which we will discuss in Sec. IV.

A. Fit to Electroweak Precision Data

Using a recent version (April 2010) of the Fortran program GAPP (Global Analysis of Particle Prop-
erties) [17], which we modified to take the ZL0 µ −Lτ boson into account, we can fit the U (1)Lµ −Lτ model
– with a mass around the electroweak scale – to a vast amount of electroweak precision data, including
radiative corrections of the Standard Model. In the following we will set the kinetic mixing angle χ to
zero, as its inclusion will not alter the discussed phenomenology qualitatively [1]. Its effect would be most
interesting for symmetry breaking schemes that do not generate δ M̂ 2 (e.g. U (1)0 breaking via SM-singlet
scalars), which is not the case for the model discussed in Section IV.
We leave the Higgs sector unspecified for the analysis as to not introduce more parameters, but re-
strict it to singlets and doublets under SU (2)L , i.e. we leave the ρ parameter untouched. Since the
modified couplings of Z1 compared to the SM only involve the mixing angle ξ in combination with the
coupling constant g 0 (see Eq. (6)), while Z2 mainly contributes to ∆aµ (see Eq. (8)), we use the common
convention [1] of giving limits on the two quantities g 0 sin ξ and M2 /g 0 .
As fit parameters, we used the conventional set of masses (Z, Higgs, top-, bottom- and charm-quark)
and couplings (strong coupling constant αs and the radiative contribution of the lightest three quarks to
(3)
the QED coupling constant ∆αhad 3 ), listed in Table I. We also enforced the direct 95% C.L. exclusion
limit mH > 114.4 GeV given by LEP; since a Higgs mass ∼ 90 GeV is favored by an unconstrained fit,
mH lies at its lower bound. Except for M2 and sin ξ, we will not bother calculating the errors on the
best-fit values, since they are not of interest here.
The best-fit values for the SM parameters hardly change with the addition of the Z 0 . As can be seen,
the reduced χ2red ≡ χ2min /Nd.o.f. decreases from 43.8/44 ' 0.995 to 36.4/42 ' 0.867 with the addition of
the two effective parameters M2 /g 0 and g 0 sin ξ, a significant improvement. Marginalizing over sin ξ we
can visualize the narrowness of the χ2 -minimum (Fig. 2 (left)).

2 Correspondingly, the kinetic mixing angle χ—describing a coupling to the electromagnetic current—gives only minor
contributions to ε⊕ 2
αβ and can not generate them without δ M̂ . This can be seen from Eq. (12) in the limit ξ, χ  1 or,
more general, by using the parameters (M̂i , δ M̂ 2 , χ) instead of (Mi , ξ, χ).
3 Contributing to the on-shell coupling via α(MZ ) = α/[1 − ∆α(MZ )].
 5

SM SM+Z 0
M1 [GeV] 91.1877 91.1877
mt [GeV] 164.0 164.0
mb [GeV] 4.199 4.200
mc [GeV] 1.270 1.278
αs 0.1183 0.1185
(3)
∆αhad (1.8 GeV) 5.75 × 10−3 5.72 × 10−3
mH [GeV] 114.4 114.4
M2 /g 0 [GeV] − 219.6
g 0 sin ξ − −2.5 × 10−4
χ2min /Nd.o.f. 43.8/44 36.4/42

Table I: Fit parameters and their best-fit values in an analysis with/without Z 0 . The masses denote pole masses
in the MS-scheme.
1000
46
99%

44 800
99%

M2 g' @GeVD
42 90%
600
Χ2

40
90%
400 68.27%

38 68.27%

200 ‰
36

0 200 400 600 800 1000 1200 1400 -1.5 -1.0 -0.5 0.0 0.5 1.0

M2 g' @GeVD g' sinHΞL @10-3 D

Figure 2: Left: χ2 as a function of M2 /g 0 – marginalized over sin ξ – and the 68.27%, 90%, and 99% C.L. limits
for one parameter. Right: χ2 contours (68.27%, 90%, and 99% C.L. for two parameters) in the M2 –sin ξ plane.
The cross marks the best-fit values (g 0 sin ξ, M2 /g 0 ) = (−2.5 × 10−4 , 219.6 GeV).

In Fig. 2 (right) we show the contours ∆χ2 = 2.30, 4.61 and 9.21, corresponding to 68.27%, 90% and
99% C.L. for 2 parameters. The best-fit value at g 0 sin ξ = −2.5 × 10−4 and M2 /g 0 = 219.6 GeV is shown
as well.
As can be seen, there is a preferred area for a Z 0 around M2 /g 0 = 200–300 GeV, mainly constrained
by ∆aµ . Performing a separate minimization for each of the parameters (marginalizing over the others),
we derive the following 90% C.L. bounds:
160 GeV . M2 /g 0 . 560 GeV ,
(14)
−0.0008 < g 0 sin ξ < +0.0003 .

Going back to the NSI parameters in Eq. (12) once more, we can see that ε⊕ 0
µµ is maximal for M2 /g at
0
the lowest bound and |g sin ξ| at the largest bound, resulting in the expression
|ε⊕
µµ | . 4 × 10
−4
/g 02 . (15)

Consequently, sizable NSI of order 10−2 can be generated at the edge of the allowed 90% C.L. parameter
space, e.g. for g 0 ∼ 0.25–0.35, without being in conflict with the direct detection limit.

B. Detection Possibilities at the LHC

The direct detection of the unmixed Z 0 has already been discussed in Refs. [6, 7], where the most
interesting process has been identified:4

4 A 3µ + ν final state is also of interest, see Ref. [6].

 6

pp, p̄p → Z, γ → 2µ Z ′ → 4µ

LHC (14 TeV)

100 LHC (7 TeV)
Tevatron (1.96 TeV)

10

σ [fb]
1

0.1

0.01
50 100 150 200 250 300 350 400 450 500
MZ ′ [GeV]

Figure 3: Cross section for the process pp(pp) → Z, γ → 2µ Z 0 → 4µ with the cuts pT > 10 GeV and |η| < 2.5 for
different center-of-mass energies.

pp → Z ∗ X → µµZ 0 X , (16)

i.e. Z 0 -radiation off final-state muons (or tauons, see Fig. 1). This makes the final states 4µ, 4τ and 2µ2τ
especially interesting since the invariant mass distribution of the lepton-pairs can be checked for a Breit-
Wigner peak of Z 0 . The inclusion of Z–Z 0 mixing does not change these prospects of direct detection,
because the smallness of the mixing angle ξ as constrained in Eq. (14) reduces the Drell-Yan-production of
the Z 0 by a factor of ξ 2 compared to Z (further suppressed by a possibly higher Z 0 mass). Consequently,
the rates for ``¯ production become nonuniversal at level ξ 2 . 10−6 , unlikely to be observed.
Figure 3 shows the cross section for the √ process (16) with four muons in the final state (using tauons
makes no real difference) for the energies s = 7 TeV and 14 TeV, as calculated with CompHEP [18].
The cross section is shown for g 0 = 1 and scales like g 02 . The expected integrated −1
√ luminosity of ∼ 05 fb 5
until 2012 corresponds to a discovery limit around MZ 0 = 100 GeV at the LHC ( s = 7 TeV) for g = 1.
This is on par with that of Tevatron, due to their higher luminosity of about 10 fb−1 . Should the LHC be
able to gather 10 fb−1 until their shutdown in 2013, this limit can be pushed to about MZ 0 = 130 GeV,
which is√still not enough to access the interesting parameter space of the global fit (14). In the final LHC
stage ( s = 14 TeV, L = 100 fb−1 ) we can probe the model up to MZ 0 = 350 GeV, so this solution to
the anomalous magnetic moment of the muon can be partly tested within the next couple of years. A
combined analysis of 4µ, 4τ and 2µ2τ final states can be used to increase statistics and improve these
discovery limits. √ √
The discovery potential for a linear e+ e− -collider with center-of-mass energy s = 0.5 TeV ( s =
1 TeV) has been calculated in Ref. [6] to be MZ 0 = 300 GeV (MZ 0 = 500 GeV) for the coupling constant
g 0 = 1. A muon collider would, of course, be the ideal√experiment to test this model, since precision
measurements could tighten the bounds on Z 0 even for s < MZ 0 .

III. SIMPLE MODEL FOR THE NEUTRINO SECTOR

We extend the fermion content of our theory, without introducing anomalies, by three right-handed
neutrinos in the representations of GSM × U (1)Lµ −Lτ

N1 ∼ (1, 1, 0)(0) , N2 ∼ (1, 1, 0)(+1) , N3 ∼ (1, 1, 0)(−1) . (17)

5 Here we define discovery by 10 Z 0 -induced events.

 7
c
The gauge-invariant Yukawa couplings are (with NiT C −1 Nj = −NiT CNj = −N i Nj )

1 µ2
−L ⊃ − NiT C −1 (MR )ij Nj + λe1 τ3
H Le H̃N1 + λH Lµ H̃N2 + λH Lτ H̃N3 + h.c. , (18)
2
where MR has the Lµ − Lτ -symmetric structure
 
X 0 0
MR =  0 0 Y  . (19)
 
0 Y 0

Electroweak symmetry breaking (H → (0, √1 (h

2
+ v))T ) generates the bilinear terms
  
  mνe 0 0 N1
L ⊃ − νe νµ ντ  0 mνµ 0  N2  + h.c. , (20)
  
0 0 mντ N3
√
where we introduced the Dirac mass matrix mD with the entries mνi ≡ −λiH v/ 2. Invoking the seesaw
mechanism [19] in the form of X, Y  mνi results in the low-energy mass matrix
 2 
mνe
0 0
 X
Mν ' −mD M−1 T
R mD = −  · 0
mνµ mντ 
, (21)
Y
· · 0

while the mass matrix of the charged leptons is diagonal because electron, muon and tauon carry different
U (1)0 charges. We stress here that X, Y as well as mνµ , mντ and mνe are allowed by the symmetry,
and hence expected to be of similar magnitude each. The eigenvalues of Eq. (21) are −m2νe /X and
±mνµ mντ /Y and are therefore naturally of similar magnitude, i.e. there is at most a mild hierarchy
between the neutrino masses. This is what observations seem to tell us, as the neutrino mass hierarchy
is much weaker than the one of charged leptons or quarks. The atmospheric mixing angle θ23 associated
with this mass matrix is maximal, while the other two mixing angles θ13 and θ12 are zero, and hence will
only be induced by breaking the U (1)Lµ −Lτ symmetry. The degenerate neutrino pair ±mνµ mντ /Y will
also be split by the breaking.
The phenomenology of texture zeros in neutrino mass matrices (like in Eq. (21)) has been discussed,
for example, in Ref. [20], where a classification for the different structures is given. Most importantly, an
analysis shows that Mν can have at most two texture zeros to be phenomenologically successful. There
is, of course, no unique way to break the U (1)Lµ −Lτ symmetry spontaneously. Even the restriction to
a seesaw-I implementation allows for various models with very different phenomenology. The choice to
break the symmetry in the neutrino mass matrix either in the Dirac matrix mD or in the right-handed
matrix MR fixes at least the GSM quantum numbers of the scalar fields to SU (2)L doublets or singlets,
respectively. Let us focus
qP on doublets for a moment: To resolve the magnetic moment of the muon ∆aµ ,
we need MZ 0 /g 0 = 02 2 0
j Yj vj & 200 GeV, where Yj denotes the U (1)Lµ −Lτ charge of the scalar φj
with VEV vj . Introducing just one additional doublet would result in a large Z–Z 0 mixing angle, so we
have to introduce another doublet with opposite Y 0 (this model was used in Ref. [6]). With mild fine-
tuning (or an additional symmetry) we can ensure the smallness of the off-diagonal Z–Z 0 -mass element
0 2
δ M̂ 2 ∼ gZ gP(v1 −v22 ). However, since all the doublets contribute to MZ and MW ± , we have the additional
constraint doublets vj2 ' (246 GeV)2 . This leaves at most 140 GeV for the VEV of the Standard Model
Higgs doublet, which is however the only doublet that couples to the top-quark. From these remarks it
is clear that this three-doublet model cannot describe the whole interesting parameter space of Eq. (14).
Breaking U (1)Lµ −Lτ solely in the right-handed sector via SM-singlets is very simple and allows for
an arbitrary large Z 0 mass, but has little interesting phenomenology, because these scalars dominantly
couple to the heavy neutrinos and Z 0 , both already difficult to probe. We therefore opt for a combined
breaking, using doublets and singlets, as this model displays numerous interesting effects that can be
tested experimentally. To that effect, we introduce another scalar doublet φ = (φ+ , φ0 )T ∼ (1, 2, +1)(−1)
and one SM-singlet S ∼ (1, 1, 0)(−1), leading to the following additional neutrino interactions
µ1 c c
−L ⊃ + λe3 12 13
φ Le φ̃N3 + λφ Lµ φ̃N1 + λS N 1 N2 S + λS N 1 N3 S + h.c. (22)
 8

µ1 0
If φ0 acquires a VEV it generates entries in mD (d ≡ −λe3 0
φ hφ i, f ≡ −λφ hφ i), while a nonzero hSi
12 13
modifies MR via entries s ≡ λS hSi and t ≡ λS hSi:
   
mνe 0 d X s t
mD =  f mνµ 0  , MR =  s 0 Y  , (23)
   
0 0 mντ t Y 0
so the low-energy neutrino mass matrix in linear order takes the form
m2
 
m νe f mν d m νe m νµ t mνe mντ s
− Xνe − X − Yµ + XY XY
mν mν 
Mν '  · − µY τ  , (24)

0
· · 0
where we used s, t  X, Y and d, f  mνα . It is important to note that the small parameters s and t
do not spoil the validity of the seesaw mechanism, since the light masses are still always suppressed by
X or Y . The remaining texture zeros break U (1)Lµ −Lτ by two units and are therefore filled by terms
of order two in our perturbative expansion. We can reduce the number of parameters by one with the
introduction of
mντ mνµ X
a= , b= , c=1−k = , (25)
mνe mνe Y
s t d f
ε= , α= , δ= , γ= . (26)
Y Y mνe mνe
Small breaking of U (1)Lµ −Lτ and quasidegenerate masses correspond to ε, α, δ, γ, k  1 and a b ' 1, so
we decompose
 
 2  1 0 0
mνe
Mν / = M0ν + ∆M = − 0 0 a b + ∆M , (27)
 
Y
0 ab 0
with the symmetric perturbation matrix (without any approximations)
 
−δ ε (−2 + δ ε) b α (1 + δ ε) + γ (−1 + δ ε) + b δ(k − 1) a ε(1 − δ ε)
∆M =  · −(γ − b α)2 a (b α ε + ε γ + b k) , (28)
 
· · −a2 ε2

which shifts the eigenvalues of M0ν from −1 and ±a b(1 − k) to

λ1 ' −1 + 2δ ε ,
1 1 1 (29)
λ2,3 ' ±a b(1 − k) ∓ a ε γ ∓ a b α ε − b2 α2 − a2 ε2 + b α γ − γ 2 .
2 2 2
Assuming a scale m2νe /Y ∼ 0.1 eV, the atmospheric mass-squared difference ∆m2atm ' 2.4 × 10−3 eV2 will
be generated by |a b| = 1 + O(0.1), while the solar one goes quadratic in the small parameters in Eq. (26)
and hence one needs them to be of order O(0.05). The mixing angle θ13 will be small but nonzero, since
it is linear in the small parameters and can be further increased by proper values of a and b:
1
sin θ13 ' √ (a ε + γ + b (δ − α)) ∼ 10−1 –10−3 , (30)
2(1 + a b)
√
in agreement with recent T2K results [21] and global fits [22]. The deviation from sin θ23 = 1/ 2 on the
other hand is quadratic in the small parameters and typically of order O(10−2 ):
1
∆ sin θ23 ' √ a2 ε2 − b2 α2 + 2 b α γ − γ 2 .


(31)
4 2
These deviations from µ-τ symmetry can be checked by future experiments, but are, of course, not specific
to this model. As a consequence of the partial or quasidegeneracy the absolute neutrino masses are rather
large, which, due to the Majorana nature of the light neutrinos, allows for neutrinoless double β-decay in
the reach of upcoming experiments. Since there are no particularly model-specific predictions, we omit a
further discussion of 0νββ and direct mass-measurement experiments.
From Sec. II A we already know the favored values for the U (1)Lµ −Lτ breaking scale s, t ∼ hSi ∼
MZ 0 /g 0 ∼ 200 GeV, which puts the NR scale in the range X, Y ' s/ε ∼ 1–10 TeV.
 9

IV. DETAILS OF THE SCALAR SECTOR

As already mentioned in Section III, we introduce the scalar fields

! !
h+ φ+
H= ∼ (1, 2, +1)(0) , φ= ∼ (1, 2, +1)(−1) , S ∼ (1, 1, 0)(−1) , (32)
h0 φ0

which adds up to 10 real scalar fields. Four of these will serve as the longitudinal modes of the W ± ,
Z and Z 0 bosons, so we will end up with 6 physical scalar fields (instead of one in the SM). Since we
introduce an additional Higgs doublet, the phenomenology of the scalars will be similar to the usual
two-Higgs-doublet models (2HDM, see Ref. [23] for a recent review). The general potential for our fields
can be written as
V = − µ1 |H|2 + λ1 |H|4 − µ2 |φ|2 + λ2 |φ|4 − µ3 |S|2 + λ3 |S|4
+ δ1 |H|2 |φ|2 + δ2 |H † φ|2 + δ3 |H|2 |S|2 + δ4 |φ|2 |S|2 (33)
√ 
− 2|µ|eiκ H † φS + h.c. .

The positivity of the potential gives constraints on the coefficients, since we have to ensure that there is a
minimum around which we can use perturbation theory. To this effect, one studies the different directions
in field-space (e.g. S = 0 and |φ|, |H| → ∞) to find a number of algebraic equations that ensure V > 0.
In the case of Eq. (33), the quartic part of the potential (the relevant part for large field values, i.e. the
limit H, φ, S → ∞) has the structure of a quadratic form as long as δ2 = 0 or the field direction satisfies
H † φ = 0, which then allows us to simply use Sylvester’s criterion for positive-definite quadratic forms to
determine the relevant conditions:
0 < λj , 0 < 4λ1 λ2 − δ12 ,
(34)
0 < 4λ1 λ2 λ3 + δ1 δ3 δ4 − λ1 δ42 − λ2 δ32 − λ3 δ12 .
Including δ2 yields the supplementary bound
0 < 4λ1 λ2 λ3 + (δ1 + δ2 )δ3 δ4 − λ1 δ42 − λ2 δ32 − λ3 (δ1 + δ2 )2 . (35)
Additional constraints come from the positivity of the scalar masses, which are however more intricate
and will not be explicitly derived here; neither will the bounds from perturbativity and unitarity, which,
in principle, give upper bounds on the couplings. Introducing the VEVs6
√ √ √
Re S → hSi ≡ vS / 2 , Re φ0 → hφ0 i ≡ vφ / 2 , Re h0 → v/ 2 , (36)
and minimizing the potential, gives three equations for µ1,2,3 , which we plug back into the potential.
To calculate the masses we will go to unitary gauge, i.e. eliminate the unphysical degrees of freedom, as
determined by the kinetic terms:
L ⊃ (Dµ H)† (Dµ H) + (Dµ φ)† (Dµ φ) + (Dµ S)† (Dµ S)
e e
= |∂µ h0 − i Zµ h0 − i √ Wµ− h+ |2
2sW cW 2sW
e e
+ |∂µ h+ − i (c2 − s2W )Zµ h+ − ieAµ h+ − i √ Wµ+ h0 |2
sW cW W 2sW
e e (37)
+ |∂µ φ0 + ig 0 Zµ0 φ0 − i Z µ φ0 − i √ Wµ− φ+ |2
2sW cW 2sW
e e
+ |∂µ φ+ + ig 0 Zµ0 φ+ − i (c2 − s2W )Zµ φ+ − ieAµ φ+ − i √ Wµ+ φ0 |2
sW cW W 2sW
+ |∂µ S + ig 0 Zµ0 S|2 .
Expanding the fields around the VEVs, we find the mass terms for the gauge bosons

2 e2 e2
MW = (v 2 + vφ2 ) , MZ2 = (v 2 + vφ2 ) ,
4s2W 4s2W c2W (38)
e
MZ2 0 = g 02 (vφ2 + vS2 ) , δ M̂ 2 = − g 0 vφ2 .
2sW cW

6 While we choose all VEVs real and positive for simplicity, it must be stressed that this is not the most general case.
 10

Small Z–Z 0 mixing demands a small VEV vφ , but since the mixing angle ξ in Eq. (5) is quadratic in the
VEVs, this only constrains vφ . 10 GeV (using the limits (14) and assuming MZ 0 > MZ , i.e. g 0 ∼ 1).
0
The main contribution
q to the Z mass has to come from vS , the anomalous magnetic moment ∆aµ
gives the constraint vS2 + vφ2 & 200 GeV. In the following, approximations are made with the scaling
vφ  vS ∼ v.
Aside from the mass terms, we also find the cross terms between gauge bosons and Goldstone bosons,
namely:
e √
L⊃− √ Zµ ∂ µ (vφ Im φ0 + v Im h0 ) + 2g 0 Zµ0 ∂ µ (vφ Im φ0 + vS Im S)
2sW cW
e (39)
− iWµ+ ∂ µ (vφ φ− + v h− ) + h.c.
2sW
We read off the Goldstone fields (not properly normalized)

G− ∼ vφ φ− + v h− , GZ ∼ vφ Im φ0 + v Im h0 , G0 ∼ vφ Im φ0 + vS Im S , (40)

which are not orthogonal. Using the gauge freedom to fix G− = GZ = G0 = 0 would result in physical
scalars with unconventional kinetic terms; instead of rotating the noncanonical kinetic terms, this can be
avoided also by first constructing a orthonormal basis from GZ and G0 . We define the physical field σ
via σ ∼ G0 × GZ ,7 then “rotate” GZ to G̃Z ≡ σ × G0 . These fields are connected to the gauge eigenstates
by a unitary transformation:
    
G0 cos θ sin θ 0 Im S
G̃Z  = − cos β sin θ cos β cos θ sin β  Im φ0  (41)
    
0
σ sin β sin θ − sin β cos θ cos β Im h

with the two angles

vφ v q 2 v
tan θ ≡ , tan β ≡ vS + vφ2 = . (42)
vS vS vφ vS sin θ

The expected scaling vφ  vS ∼ v implies sin θ, cos β  1. The unitary gauge, G0 = G̃Z = 0, leaves the
physical field σ, contributing to the potential through
   T    
Im S cos θ sin θ 0 0 sin β sin θ σ
Im φ0  = − cos β sin θ cos β cos θ sin β   0  = − sin β cos θ σ  . (43)
       
Im h0 sin β sin θ − sin β cos θ cos β σ cos β σ

The field σ consists mainly of Im φ, so the imaginary part of h0 is not zero as in the SM, but suppressed
by cos β. The charged Goldstone boson is easier to handle, we have
! ! ! !
φ− cos β − − sin β − G− G− → 0 − sin β − σ −
= −−−−−−−→ , (44)
h− sin β − cos β − σ− cos β − σ −

with the angle tan β − ≡ v/vφ = cos θ tan β ' tan β  1. The physical fields σ, σ ± , Re S ≡ S,
Re φ0 ≡ φ and Re h0 ≡ h are not mass-eigenstates. Setting, for simplicity, the CP-violating angle κ
in the potential (33) to zero, we can at least read off the masses for σ and σ ± :

|µ| v vS |µ| vφ vS |µ| v vφ

m2σ = + + , (45)
vφ v vS
|µ| v vS |µ| vφ vS 1
m2σ± = + − δ2 (vφ2 + v 2 ) . (46)
vφ v 2

In the approximation we will use extensively, vφ  vS ∼ v, only the first term contributes and mσ ' mσ± .
The masses mσ and mσ± increase for decreasing vφ , reminiscent of an inverse seesaw mechanism; useful

7 To make use of the cross product we identify G0 and GZ with vectors in the basis (Im S, Im φ0 , Im h0 ).
 11

h3 h4 S3 S4 φ3 φ4 σ4
σ2 h hφS σ2 φ σ2 S φ2 S 2 h2 S 2 h2 φ 2
σ 2 φ2 σ 2 h2 h2 S h2 φ φ2 h σ2 S 2 φ2 S
S2φ hS 2 σ+ σ− σ+ σ− h2 σ + σ − φ2 σ + σ − σ2 σ+ σ− hσ + σ −
φσ + σ − Sσ + σ − hφ σ + σ − S 2 σ+ σ−
Z2h W + W − φ Z 2 σ+ σ− ZAσ + σ − ZW − σ + φ A2 σ + σ − Z 2 h2
W Aσ φ W + W − σ 2 W + W − φ2
− +
ZW σ σ W Aσ σ W W σ σ ZW − σ + h
− + + − + − + −

W − Aσ + h W + W − h W + W − h2 Z 2 σ2 Z2φ Z 2 φ2
Z 02 φ2 Z 02 S 2 W − Z 0 σ + φ ZZ 0 σ + σ − AZ 0 σ + σ − W + Z 0 σ − σ Z 02 σ + σ −
W + Z 0 σ − Z 02 φ Z 02 S Z 02 σ 2 ZZ 0 σ 2 ZZ 0 φ ZZ 0 φ2
Aσ + σ − Zσ + σ − Zσh Zσφ W − σ+ σ W − σ+ h W − σ+ φ
Z 0 φσ Z 0 σS Z 0 σ+ σ−

Table II: Interaction vertices involving scalars among themselves (first row), with SM vector bosons (second row),
with the Z 0 boson (third row), and couplings to vector bosons involving derivatives (last row).

values for |µ| will be discussed below. The CP-even scalars share the symmetric mass matrix (in (S, φ, h)
basis)
|µ| v v
 
2λ3 vS2 + s φ −|µ|v + δ4 vφ vS −|µ|vφ + δ3 vvS
M2CP−even =  · 2λ2 vφ2 + |µ| dv vS −|µ|vS + (δ1 + δ2 )vvφ  , (47)
 
2 |µ| vφ vS
· · 2λ1 v + v

with the approximate eigenvalues (labeled according to the predominant field in the unmixed scenario)
q
m2h̃,S̃ ' λ1 v 2 + λ3 vS2 ± (λ1 v 2 − λ3 vS2 )2 + δ32 vS2 v 2 , (48)
|µ| v vS
m2φ̃ ' . (49)
vφ

For small cos β ' vφ /v, the fields σ, σ + and φ have degenerate masses. In contrast to other 2HDM, we
do not have a light pseudoscalar σ, because it has roughly the same mass as the charged scalar σ ± . In the
next section, we will see that the charged scalar mass is bounded from below, mσ± & 80 GeV, so there
are no decay modes of h into real σσ, σ + σ − or φφ, unless h has a mass of at least 160 GeV. S could,
in principle, have a mass low enough to allow h → SS, depending on λ3 . Decay channels as signatures
in collider experiments will be discussed below. We point out that without large mass splittings in the
scalar sector, the quantum corrections to the ρ parameter, due to scalar loops, will be small [24].
Obviously, the µ term in the potential is crucial for the generation of large scalar masses, without it,
we would end up with masses ∼ vφ , either below MZ (introducing new invisible decay channels for Z)
or above it (introducing too large of a Z–Z 0 -mixing angle). The potential in the (σ, σ ± , S, φ, h) basis is
ridiculously lengthy and will not be shown here. It involves the interaction terms given in Table II; also
shown are the interactions with the gauge bosons. Making |µ| and the δi small, e.g. |µ| ∼ vφ , results in
small mass mixing of order µ/v and δ3 ; for simplicity, we will work in zeroth order and treat S, φ and h
as mass eigenstates. The additional mass mixing can be of the same order as the mixing through β and
β − , consequently the combined mixing could be larger or smaller, depending on their relative sign in the
coupling, similar to usual 2HDM [23]. Since we are only performing order of magnitude approximations
in the scalar sector, we do not go into more detail.
Just as an aside, we mention that none of the scalars are stable. The scalars σ and σ − couple directly to
fermions (albeit weakly) and will decay through such channels. The scalars φ and S couple predominantly
to the heavy neutrino sector, but can in any way decay via a Z 0 Z 0 . So, without invoking some additional
discrete symmetries, this model provides no candidate for dark matter.
 12

A. Yukawa Interactions and Lepton Flavor Violation

The gauge-invariant Yukawa interactions of the two doublets H and φ and the singlet S to the leptons
and right-handed neutrinos Ni can be written as:
µµ
−L ⊃ + λee ττ
H Le HeR + λH Lµ HµR + λH Lτ HτR
µ2
+ λe1 τ3
H Le H̃N1 + λH Lµ H̃N2 + λH Lτ H̃N3
(50)
+ λeµ τe
φ Le φµR + λφ Lτ φeR
µ1 c c
+ λe3 12 13
φ Le φ̃N3 + λφ Lµ φ̃N1 + λS N 1 N2 S + λS N 1 N3 S + h.c.

In unitary gauge we replace the scalars by the physical degrees of freedom σ (43), σ ± (44), S, φ and h.
Denoting sin β with sβ , sin β − with s− β etc., this becomes:
 
X
`` v 1 − +
−L ⊃ λH √ `L `R + √ `L `R (h + icβ σ) + cβ ν ` `R σ
`=e,µ,τ
2 2
 
e1 v 1 − −
+ λH − √ ν e N1 − √ ν e N1 (h − icβ σ) + cβ eL N1 σ + (e1 → µ2, τ 3)
2 2
 
eµ v φ 1 − +
+ λφ √ eL µR + √ eL µR (φ − isβ cθ σ) − sβ ν e µR σ + (eµ → τ e)
2 2 (51)
 
e3 v φ 1 − −
+ λφ − √ ν e N3 − √ ν e N3 (φ + isβ cθ σ) − sβ eL N3 σ + (e3 → µ1)
2 2
 
12 v S c 1 c
+ λS √ N 1 N2 + √ N 1 N2 (S + isβ sθ σ)
2 2
 
13 v S c 1 c
+ λS √ N 1 N3 + √ N 1 N3 (S − isβ sθ σ) + h.c. ,
2 2
which can be further simplified using `¯L `R (h+icβ σ)+h.c. = `(h+ic
¯ β γ5 σ)`, emphasizing the pseudoscalar
nature of σ. The coupling to quarks is of the same form as in the Standard Model (since they are singlets
under U (1)0 ):
i i
λij j j
X X
−L ⊃ d QL HdR + λij
u QL H̃uR + h.c. (52)
i,j=1,2,3 i,j=1,2,3

Diagonalization of the mass matrices goes through as usual, via bi-unitary transformations; we end up
with
0
−L ⊃ dL Dd dR H 0 + uL Du uR H + uL V Dd dR H + − dL V † Du uR H − + h.c. , (53)
with the matrices in generation space
√ √
Dd ≡ 2 diag(md , ms , mb )/v , Du ≡ 2 diag(mu , mc , mt )/v , (54)
and V the usual unitary Cabibbo-Kobayashi-Maskawa matrix of the Standard Model. In the SM, the
terms with H ± vanish in unitary gauge, while in our case we have the Yukawa interactions:
X md i X mu
i i
−L ⊃ dL diR (h + icβ σ) + uiL uiR (h − icβ σ)
i
v i
v
√ √ (55)
2X d i j + − 2 i
muj Vji∗ dL ujR σ − + h.c.
X
+ c−
β mj Vij uL dR σ − cβ
v i,j v i,j

The flavor-changing interactions are suppressed by the Yukawa couplings mq /v and the angle cos β − '
vφ /v, compared to those induced by W ± . The interaction of the charged scalars with the quarks is very
similar to the Two-Higgs-Doublet Model of Type I (2HDM-I) [25], where the parameter cos β − is denoted
by tan β. The corresponding bound mσ− & 80 GeV on a charged scalar with decay channels σ − → c̄s,
τ ν̄τ , set by LEP [26], applies. Additional contributions from σ ± to charged-current decays are already
well suppressed; for example, the decay τ → ντ σ − → ντ ν µ µ has the width
!4
1 1 c−
β
 m m 2
τ µ
Γ(τ → ντ ν µ µ) ≡ Γσ± ' m5τ , (56)
192 (2π)3 mσ− v2
 13

resulting in an additional branching ratio of Γσ± /Γtotal ' 10−11 (c− 4 4

β ) (80 GeV/mσ − ) , at least 7 orders
of magnitude below the current sensitivity [11].
There are two different kinds of Lepton Family number Violation (LFV) associated with this model,
we will discuss them in the following. Since we have chosen the charge Y 0 = −1 for our Higgs fields φ and
S, the Lµ − Lτ number of a process will only changed by one unit in the simplest Feynman diagrams,
i.e. we expect decays τ → eX and µ → eX, but not τ → µX.

1. LFV Mediated by Z 0

As can be seen immediately in Eq. (51), the VEV vφ introduces nondiagonal elements in the mass
matrix of the charged leptons:
eµ
 
λee
H v λ φ vφ 0
1 
Mleptons = √  0 λµµH v 0 . (57)

2 τe ττ
λ φ vφ 0 λH v

The mass eigenbasis is obtained by means of a bi-unitary diagonalization, i.e. `L → UL `L , `R → UR `R ,

with UL 6= UR . The relevant rotation matrices are
   
L R
1 θ12 0 1 0 θ13
 L
UL ' −θ12 1 0 , UR '  0 1 0  , (58)
  
R
0 0 1 −θ13 0 1
√ √
L
with θ12 ' λeµ R τe
φ vφ / 2mµ and θ13 ' λφ vφ / 2mτ . Since these matrices operate in flavor space, they do
not change the normal Z-boson gauge interactions, but the Z 0 coupling:
 
X 0
jµ0 Z 0µ = `i  1  γµ `i Z
0µ
 
i=L,R −1
  (59)
0
`i Ui†  1
X
0µ
→  Ui γµ `i Z ,
 
i=L,R −1

Since Ui† diag(0, 1, −1)Ui is not diagonal, the Z 0 introduces interactions like τ → eZ 0∗ → eµµ̄. This also
generates a coupling of Z 0 to electrons, suppressed by θL,R2
, and furthermore all the couplings become
0
chiral, i.e. the Z couples differently to left- and right-handed fermions. The same reasoning applies
to LFV mediated by neutral scalars, since they couple in a generation-dependent way as well. The
Z 0 -mediated LFV decays are

τ → eµµ, τ → e ν µ,τ νµ,τ , and µ → e ν µ,τ νµ,τ , (60)

R
the first of which can be probed in B-factories and leads to the constraint on θ13 [27]
4
Γ(τ → eµµ̄)

200 GeV !
R 2
' (θ13 ) < 1.6 × 10−7 , (61)
Γ(τ → µντ ν̄µ ) MZ 0 /g 0

so for vφ ∼ mτ and MZ 0 /g 0 ∼ 200 GeV we find the bound λτφe < 10−3 –10−4 for the Yukawa coupling. The
L
angle θ12 can not be probed in this way due to the challenging neutrinos in the final state. However, this
angle contributes to the PMNS mixing matrix via the charged current interactions, i.e. UPMNS = UL† Uν ,
L
which most importantly adds to s13 a term θ12 s23 , where s23 denotes sin θ23 of Uν . A relatively large
θ13 can in consequence be generated without a strongly broken U (1)Lµ −Lτ , simply due to the interplay
with the charged leptons (depending on the signs, a cancellation could occur as well). Since nothing in
the motivation for our model depends on λeµ τe
φ and λφ , we can make them arbitrarily small (and they can
still be larger than the Yukawa coupling of the electron).
 14

µ1
s−
β λφ c− e1
β λH
µ N1 e
σ − γ

Figure 4: Charged-scalar mediated lepton-flavor-changing radiative decay.

2. LFV via Loops

The second source of LFV stems from the charged scalars, inducing the decays µ → eγ and τ → eγ via
diagrams like Fig. 4, with a heavy right-handed neutrino in the loop. Since these decays involve the same
Yukawa coupling λ`i that generate the U (1)0 -breaking elements in the neutrino mass matrix, they better
not be too small in our model. Calculating the branching ratio of the decay µ → eγ in the approximation
mN  mσ+ , mµ , me , we find [28]
µ1 − e1
!2
s−
2
β λφ cβ λH γ2 m2νe /Y

1
BR(µ → eγ) ' ' ' 10−29 γ 2 , (62)
192π 2 2GF m2N 96π 2 Y

which is highly suppressed by the heavy neutrino mass and poses no bound on γ = f /mνe (see Eq. (26)).
We also see that the predicted LFV from the scalars is too low to be observed in any future experiment,
as opposed to the Z 0 -mediated processes.

3. Contribution to ∆aµ

The physical scalars contribute to the anomalous magnetic moment of the muon via loop diagrams.
Setting the LFV Yukawa couplings λeµ τe
φ = λφ = 0, only h, σ and σ
−
couple directly to the muon. The
−
one-loop contributions from the pseudoscalar σ and the charged σ are [12]
1
 !2 
4 Z − 2
−m c − 3

µ β x(1 x) c β x
∆a1−loop
µ = dx  + ; (63)
8π 2 v 2 mσ− 1 + (x − 1)m2µ /m2σ− mσ 1 − x + x2 m2µ /m2σ
0

however, the two-loop contribution of σ is also important due to a larger coupling of σ to heavy fermions
in the loop, which compensates the additional loop suppression (see Fig. 5). The dominant effect gives
[29]
 2 2
1 mf /mσ
α mµ 22 m 2 Z ln x(1−x)
f f
X
2−loop 2
∆aµ = c Ncolor Qf 2 dx 2 2 . (64)
8π 3 v 2 β mσ mf /mσ − x(1 − x)
f =t,b,τ 0

The combined one and two-loop contributions are shown in Fig. 5 (right) for the case cos β − ' cos β,
mσ ' mσ− , corresponding to the vφ  v limit we are interested in. As can be seen the effects are about 2
orders of magnitude too small to have any visible effect.

B. Signatures at the LHC

The effects of additional scalars in collider experiments, especially concerning the disentanglement of
different multi Higgs doublet models, have been reviewed in Ref. [30]; since our model is similar to the
2HDM-I in the decoupling limit, we expect similar signatures. The best candidate for observation will be
the scalar h, with couplings reduced by mass mixing (which goes roughly with µ/v, of the same order as
cos β), which we did not discuss before, and smaller branching ratios due to the additional decay modes
via the other scalars (h → σσ, σ + σ − , φφ, φS, SS) and in association with gauge bosons (h → ZW − σ + ,
AW − σ + , W − σ + , Zσ), most important for a heavy h. An analysis of the branching ratios of h will
therefore not suffice to distinguish our model from the 2HDM-I.
It is interesting to note that the Z–Z 0 mixing angle goes roughly quadratic in vφ (ξ ∼ vφ2 /vS2 from
Eq. (5)), while the scalar mixing is linear (θ ∼ vφ /vS , cos β ∼ sin θ from Eq. (42)). This suggests better
 15

Da Μ contours
µ µ
3.0
10-13
σ γ, Z 2.5
10-12

log mΣ @GeVD
2.0 10-11
1.5 10-10
τ, b, t 1.0

0.5

γ 0.0
-2.0 -1.5 -1.0 -0.5 0.0

log cos Β

Figure 5: Dominating two-loop Barr-Zee-type diagram contributing to ∆aµ (left), actual values for ∆aµ in the
approximation cos β − ' cos β, mσ ' mσ− (right).

direct detection prospects via Drell-Yan processes for the scalars than for Z 0 . Since the interactions of
σ, σ ± , φ and S with the leptons are suppressed not only by cos β, but also by their small Yukawa couplings,
whereas the gauge boson couplings scale with cos β, this sector will be the most interesting. For example
the decay channel h → Zσ (discussed in Ref. [31]) scales with cos β; the decay h → Z 0 σ is induced by
mass mixing of the scalars and thus goes roughly with µ/v. This could lead to interesting signatures,
since the invariant mass of the subsequently created leptons gives information about the virtual particles,
their angular distribution about the spin of the bosons and the rates of electrons, muons and tauons and
about the admixture of Z 0 over Z. Such an analysis would however require a lot of luminosity. In general,
the most dominant effect of the scalars and Z 0 will be the difference in the e, µ, τ rates due to Z 0 decays.
We mention the obvious fact that a future muon collider would be the ideal experiment to test this
model, basically in total analogy to Z measurements at LEP. Since the U (1)Lµ −Lτ symmetry in this
model connects the heavy right-handed neutrino sector to the SM, this would also open up a way to
probe for this special mechanism of neutrino mass generation. An analysis of these signatures lies outside
the realm of this work, but has been performed for a similar model (based on gauged B − L at the LHC)
in Ref. [32].

V. EXTENSION TO SU (2)Lµ −Lτ

Nonabelian family symmetries based on SU (2)H or SU (3)H (“horizontal symmetry”) have been dis-
cussed extensively in the literature [33], although mainly with focus on the quark sector. An extension
from U (1)Lµ −Lτ to a nonabelian group is natural since it includes the electron into the symmetry. It also
forces the kinetic mixing angle χ to be zero at tree-level, because the field strength tensor of the SU (2)0
gauge bosons is not a gauge-invariant object. Reference [3] also contains discussions of an SU (2) exten-
sion of U (1)Lµ −Lτ , but with no emphasis on the neutrino structure. Constructing a three-dimensional
SU (N ) SU (N ) SU (N )
representation with a diagonal generator Tdiag (µ) = −Tdiag (τ ) and Tdiag (e) = 0 is possible for
N = 2 and N = 3. Since SU (3) can be seen as an extension of SU (2) we will not consider it in the
following. The extension from U (1)Lµ −Lτ to SU (2)0 remains anomaly-free even without right-handed
neutrinos, partly because the SU (2) only has real and pseudoreal representations.8 For the SU (2)0 , we
have two possibilities concerning the representation of electron, muon and tauon:
(i) irreducible: e, µ and τ form an SU (2)0 triplet,
(ii) reducible: e transforms as a singlet and (µ, τ ) form a doublet.
The latter case once again treats the electron differently than muon and tauon, furthermore it is not
possible to implement a seesaw-I mechanism, so we discuss it only briefly in Appendix B. In the following
we will therefore use the GSM × SU (2)0 representations

L ≡ (Lµ , Le , Lτ ) ∼ (1, 2, −1)(3) , `R ≡ (µR , eR , τR ) ∼ (1, 1, −2)(3) . (65)

8 The only possible anomaly is SU (2)0 -SU (2)0 -U (1)Y , which vanishes as long as the charged leptons are in the same SU (2)0
representation.
 16

We will also refer to the SU (2)0 as “leptospin” for convenience later on. Since U (1)Lµ −Lτ is the SU (2)0
subgroup generated by
 
1 0 0
Tz(3) = 0 0 0  , (66)
 
0 0 −1

we expect a possible breaking pattern SU (2)0 → U (1)Lµ −Lτ → nothing, which might still resolve ∆aµ and
explain the neutrino mixing angles. In the next sections we will comment on the difficulties concerning
this task.
It proves convenient for the most part to work in the flavor basis (µ, e, τ ), as already used in Eq. (65)
and (66); however, to make the neutrino mass matrices look more familiar, the transformation back to
the usual (e, µ, τ ) basis can be performed via
 
0 1 0
L → UL , M → U MU , U = 1 0 0 , (67)
 
0 0 1

where the matrix U satisfies U = U −1 = U T .

A. Lepton Masses

The allowed mass terms for the charged leptons are generated by

L ⊃ YH LH`R , (68)

which gives me = mµ = mτ . To break this symmetry we introduce an SU (2)0 triplet ∆ and a pentet
(leptospin-2) Σ, with the same GSM = SU (3)C × SU (2)L × U (1)Y quantum numbers as the standard
Higgs H, i.e.:

H ∼ (1, 2, +1)(1) , ∆ ∼ (1, 2, +1)(3) , Σ ∼ (1, 2, +1)(5) . (69)

In matrix notation, we have (see App. A for a short collection of used representations)
   √ + √ ++ 
∆0 ∆ + 0 Σ0 3Σ 6Σ
1  1 √ √
∆ = √ ∆− 0 ∆+  , Σ = √  3 Σ− −2 Σ0 − 3 Σ+  , (70)
 
2 6 √ √
0 ∆− −∆0 6 Σ−− − 3 Σ− Σ0

where the superscript denotes the Lµ − Lτ charge of the SU (2)L doublet, not the electric charge. In fact,
all of the following discussion is focused on flavor space, the SU (2)L contractions will not be used. Since
the two leptospin-1 fields L and `R can couple to leptospin-0, 1 and 2, the most general allowed Yukawa
couplings are given by

L ⊃ L (YH H + Y∆ ∆ + YΣ Σ) `R , (71)

so if the fields acquire VEVs that leave U (1)Lµ −Lτ intact (i.e. only ∆0 and Σ0 ), we get the masses
√ √
mµ = YH hHi + Y∆ h∆0 i/ 2 + YΣ hΣ0 i/ 6 ,
√
me = YH hHi − 2YΣ hΣ0 i/ 6 , (72)
√ √
mτ = YH hHi − Y∆ h∆0 i/ 2 + YΣ hΣ0 i/ 6 .

To get the charged-lepton masses right we need all three VEVs, the small electron mass is the result of
a fine-tuned cancellation. Specifically, we have

YH hHi = (mµ + me + mτ )/3 ' 0.6 GeV ,

√
Y∆ h∆0 i = (mµ − mτ )/ 2 ' −1.2 GeV , (73)
√
YΣ hΣ0 i = (mµ − 2 me + mτ )/ 6 ' 0.8 GeV .
 17

Since all these SU (2)L doublets contribute to MW and MZ , we have the boundary condition hHi2 +
h∆0 i2 + hΣ0 i2 ' (174 GeV)2 , and because hHi gives the mass to the top-quark, it will be the largest of
these three VEVs; for approximations, we will use h∆0 i, hΣ0 i ∼ O(10)GeV. Seeing that this breaking
scheme leaves U (1)Lµ −Lτ as an exact symmetry, there will not be any mixing of Z with the SU (2)0 gauge
bosons Xi at tree-level. The kinetic terms for the charged leptons obviously lead to LFV:
/ + i`R D`
L ⊃ iLDL / R, (74)
(3)
with covariant derivative Dµ = ∂µ − ig 0 Xj Tj . It proves convenient to define the two gauge fields
√
X ± ≡ (X1 ∓ iX2 )/ 2, the SU (2)0 gauge interactions then take the form
+ − + −
L/g 0 ⊃ µX
/ 3µ − τ X
/ 3 τ + eX
/ τ + eX
/ µ + µX
/ e + τX
/ e, (75)
which generate the process τ → e µ e at tree-level (plus other, less constrained decays involving neutrinos).
The branching ratio for this process is less than 1.5 × 10−8 [11], leading to a constraint
MX ± /g 0 & O(10) TeV . (76)
Since such a high breaking scale can not be realized with SU (2)L doublets, it is necessary to introduce
more scalar fields that break SU (2)0 but do not contribute to MZ and MW ± . Fortunately, this fits into
the neutrino mass generation via seesaw. Once again we will use the breaking scale to find the NR -scale,
in analogy to Section III.

1. Majorana Masses

Introducing right-handed neutrinos N ≡ (Nµ , Ne , Nτ ) ∼ (1, 1, 0)(3), conveniently written as

√ !
Ne / 2 Nµ
N= √ , (77)
Nτ −Ne / 2
c c c c
allows for the SU (2)0 -invariant mass term tr(N N ) = N e Ne + N µ Nτ + N τ Nµ , leading to a Majorana
mass matrix
 
0 0 1
MR = mR 0 1 0 . (78)
 
1 0 0
Note that the eigenvalues of MR are degenerate. As far as the allowed Yukawa couplings go, the coupling
c
of the symmetric bilinear N i Nj to a leptospin-1 field vanishes,9 so we introduce another leptospin-2 field
Ω ∼ (1, 1, 0)(5), transforming as a singlet under GSM . Since it carries no other quantum numbers, we
can choose the fields real, i.e. Ω is an hermitian matrix:
 √ + √ ++ 
Ω0 3Ω 6Ω
1 √ √
Ω = √  3 Ω− −2 Ω0 − 3 Ω+  with (Ω− )† = Ω+ , (Ω−− )† = Ω++ . (79)

6 √ −− √ − 0
6Ω − 3Ω Ω
This allows for the Yukawa terms
 c c
 √ + √ ++   
Ne Nµ 0 Ω0 3Ω 6Ω Ne Nµ 0
YΩ  c c  √ √
L ⊃ √ tr N τ 0 N µ   3 Ω− −2 Ω0 − 3 Ω+  Nτ 0 Nµ 
 
2 6 c c √ √
0 N τ −N e 6 Ω−− − 3 Ω− Ω0 0 Nτ −Ne
√ √   (80)
6 Ω−− − 3 Ω− Ω0 −Nµ
YΩ  c  √ √ + 

c c −
= √ −N µ N e N τ − 3 Ω 2Ω 0
3 Ω   Ne  ,

2 6 √ √
Ω0 3 Ω+ 6 Ω++ Nτ
where the first line shows explicitly the gauge invariance and the second line the symmetric nature of the
coupling. A nonzero VEV hΩ0 i can be used to break the degeneracy of the Ne and Nµ,τ masses and lead
to the general U (1)Lµ −Lτ invariant Majorana mass matrix (19).

9 The coupling of three leptospin-1 fields uses the SU (2)0 invariant antisymmetric symbol εijk .
 18

2. Dirac Neutrino Masses

In direct analogy to the charged-lepton masses we have

 
˜ + ỸΣ Σ̃ N R ,
L ⊃ L ỸH H̃ + Ỹ∆ ∆ (81)

which leads to a diagonal Dirac-matrix mD with nondegenerate eigenvalues after ∆0 → h∆0 i, Σ0 → hΣ0 i.
For the definition of the tilde-fields see App. A.

B. Masses for the Gauge Bosons

The Lagrangian for the SU (2)0 -charged scalars:

1
L ⊃ tr (Dµ ∆)† Dµ ∆ + tr (Dµ Σ)† Dµ Σ + tr (Dµ Ω)† Dµ Ω ,



(82)
2
results in mass-terms for Xi after SU (2)0 -breaking via ∆0 , Σ0 and Ω0 :
2 02 0 2 02 0 2 02 0 2 2
MX ± = 2 g h∆ i + 6 g hΣ i + 3 g hΩ i , MX = 0. (83)
3

Because of the constraint (76), the VEV hΩ0 i should be around 10 TeV, which is fine for an NR -scale
around 100 TeV. We mention that MX3 can be pushed arbitrarily high via the VEV of an SU (2)0 doublet,
without affecting any of the discussed lepton phenomenology.

C. Scalar Potential

One SU (2)0 singlet (H), one triplet (∆) and two leptospin-2 fields (Σ and Ω) result in a gauge-invariant
potential V (H, ∆, Σ, Ω) = V2 + V3 + V4 with

V2 = µ2H H † H + µ2∆ tr(∆† ∆) + µ2Σ tr(Σ† Σ) + µ2Ω tr(ΩΩ) ,

V3 = +µ1 Htr(Σ† Ω) + µ2 tr(∆† Ω∆) + µ3 tr(Σ† Ω∆) (84)
†
+ µ4 tr(Σ ΣΩ) + µ5 tr(ΩΩΩ) + h.c. ,

and finally some of the quartic interactions:

V4 = λH (H † H)2 + λ∆ (tr(∆† ∆))2 + λΣ (tr(Σ† Σ))2 + λΩ (tr(ΩΩ))2 + λ1 det(∆† ∆)

+ λ2 tr(Σ† ΣΣ† Σ) + λ3 tr(Σ† Σ† ΣΣ) + λ4 Htr(Σ† ΩΩ) + λ5 Htr(∆† ΩΩ)
+ λ6 Htr(∆† Σ† Σ) + λ7 Htr(∆† ∆† ∆) + λ8 Htr(Σ† Σ† Σ) + λ9 Htr(∆† ∆† Σ)
+ λ10 |H|2 tr(∆† ∆) + λ11 |H|2 tr(Σ† Σ) + λ12 |H|2 tr(ΩΩ) + λ13 tr(∆† ∆)tr(Σ† Σ) (85)
† † † †
+ λ14 tr(∆ ∆)tr(ΩΩ) + λ15 tr(Σ Σ)tr(ΩΩ) + λ16 tr(Σ ΣΩΩ) + λ17 tr(∆ ∆ΩΩ)
+ λ18 tr(∆† ∆Σ† Σ) + λ19 tr(∆† ΣΩΩ) + λ20 tr(∆† Σ∆† ∆) + λ21 tr(∆† ΣΣ† Σ)
+ λ22 tr(ΣΣ) tr(∆† ∆† ) + λ23 tr(ΣΣ∆† ∆† ) + . . . + h.c.

The potential is obviously very complicated to analyze, so we will only discuss the potential for Ω:

V (Ω) = µ2Ω tr(ΩΩ) + µ5 tr(ΩΩΩ) + λΩ (tr(ΩΩ))2 , (86)

It can be shown that one can eliminate either Ω++ or Ω+ via SU (2)0 gauge transformations (for hermitian
Ω), while making the other fields real (App. A 2). One can therefore study V (Ω) as a function of the two
real parameters Ω0 and Re Ω+ . For µ2Ω < 0, the potential has a minimum at
s
−µ2Ω 3 µ5
hΩ0 i = + √ + O(µ25 /µΩ ) , hΩ+ i = 0 , (87)
2λΩ 8 6 λΩ

where we assumed 0 < µ5  µΩ and a positive VEV. In unitary gauge (Ω± is eaten by X ± ), Ω0 receives
a mass MΩ2 0 ∼ λ hΩ0 i2 while Ω++ is comparatively light, MΩ2 ++ ∼ µ5 hΩ0 i.
 19

The above discussion was meant to show the possibility of the aforementioned breakdown SU (2)0 →
U (1)Lµ −Lτ → nothing, which can still accommodate the nice features of the pure U (1)Lµ −Lτ model,
namely a motivation for the maximal atmospheric mixing angle θ23 and the resolution of the magnetic
moment of the muon. To complete the model, i.e. break U (1)Lµ −Lτ , one would need to examine the
full scalar potential (84, 85), a task that goes beyond the scope of this paper. We merely point out
that the required VEVs of the U (1)Lµ −Lτ charged scalars need to be such that the off-diagonal parts
in the neutrino mass matrix mD are large enough to generate a viable UPMNS mixing matrix, while the
off-diagonal charged-lepton entries need to be small enough to allow a Z 0 with MZ 0 /g 0 ∼ 200 GeV without
large LFV.

VI. CONCLUSION

We constructed a viable extension of the Standard Model based on an additional gauge group
U (1)Lµ −Lτ . We discussed the most general low-energy Lagrangian for a broken U (1)Lµ −Lτ , includ-
ing mixing effects with the Z-boson, and identified the parameter space allowed by electroweak precision
measurements. The goodness-of-fit can be improved significantly with a Z 0 at the electroweak scale,
mostly due to the resolved anomaly of the muons magnetic moment. As a side effect of the nonuniversal
gauge coupling, nonstandard neutrino interactions are induced, potentially testable by future neutrino
oscillation experiments. To complete the model we introduced an economic scalar field sector that breaks
the additional gauge symmetry spontaneously, generating a viable neutrino mass matrix at tree-level,
which features nearly maximal mixing in the atmospheric sector and nonzero θ13 . Neutrino masses are
expected to be quasi- or partially degenerate and lead to testable neutrinoless double β-decay, whereas
the heavy right-handed neutrinos are light enough to be produced at a future muon collider via Z 0 gauge
interactions. The scalar sector of the theory is similar to other two-Higgs-doublet models, introducing a
small mixing between the physical scalars. Z 0 -mediated lepton family number violation can be tested in
upcoming experiments and distinguishes this model from others via its selected allowed modes.
The nonabelian extension of U (1)Lµ −Lτ to SU (2)0 naturally includes the electron into the symmetry
and allows for a breakdown that leaves U (1)Lµ −Lτ exact at the electroweak scale, maintaining the nice
features of the pure U (1)Lµ −Lτ model.

Acknowledgments

We thank Jens Erler for providing us with a recent version of GAPP. This work was supported by
the ERC under the Starting Grant MANITOP and by the DFG in the Transregio 27. JH acknowledges
support by the IMPRS-PTFS.

Appendix A: Field Transformations and Representations

A field ϕ in a particular representation of the gauge group GSM × U (1)Lµ −Lτ or GSM × SU (2)0 is
specified by the numbers


dim RSU (3) (ϕ), dim RSU (2) (ϕ), Y (ϕ) (Lµ (ϕ) − Lτ (ϕ)) or


(A1)
dim RSU (3) (ϕ), dim RSU (2) (ϕ), Y (ϕ) dim RSU (2)0 (ϕ) ,
SU (2)
respectively, where Y (ϕ) = 2 Q(ϕ) − 2 Tz (ϕ) denotes the hypercharge. To distinguish more easily
between charges and dimensions of representations, the dimensions are set in boldface.

1. SU (2)0 Representations

The SU (2)0 -triplet ∆ ∼ (1, 2, +1)(3) and pentet Σ ∼ (1, 2, +1)(5) can be written as vectors like
 T  T
∆ = −∆+ , ∆0 , ∆− , Σ = Σ++ , −Σ+ , Σ0 , Σ− , Σ−− , (A2)
 20

(s)
which transform like ∆ → U ∆, Σ → U Σ with U = exp(−iθj Tj ) and the SU (2)0 -generators for the
s-dimensional representation T (s) , explicitly:
 
  0 1 0 0 0
p
0 1 0 1
 p0 3/2 0 0
1 
Tx(3) = √ 1 0 1 , Tx(5) = 0
p
3/2 0 3/2 0 , (A3)
   
2  p 
0 1 0 0 0 3/2 0 1
0 0 0 1 0
 
  0 −i 0 0 0
p
0 −i 0 i
 p0 −i 3/2 0 0
1 
Ty(3) = √  i 0 −i , Ty(5) = 0 i 3/2
p
0 −i 3/2 0  , (A4)
   
2  p 
0 i 0 0 0 i 3/2 0 −i
0 0 0 i 0
 
  2 0 0 0 0
1 0 0 0 1 0 0 0 
 
Tz(3) = 0 0 0  , Tz(5) = 0 0 0 0 0  . (A5)
   
 
0 0 −1 0 0 0 −1 0 
0 0 0 0 −2

A more convenient representation is given by 3 × 3-matrices transforming like M → U M U † , which can

be obtained with the help of Clebsch-Gordan-coefficients:
   √ + √ ++ 
∆0 ∆ + 0 Σ0 3Σ 6Σ
1  1  √ √
∆ = √ ∆− 0 ∆+  , Σ = √  3 Σ− −2 Σ0 − 3 Σ+  , (A6)
 
2 − 0 6 √ −−
√ − 0
0 ∆ −∆ 6Σ − 3Σ Σ

where, as before, the superscript denotes the U (1)Lµ −Lτ charge of the field. The leptospin-1 field also
has a representation as a 2 × 2-matrix:
√ !
∆0 / 2 ∆+
∆= √ . (A7)
∆− −∆0 / 2

The weird sign in Eq. (A2) was chosen to make the matrix representations of ∆ and Σ more symmetric,
we could of course redefine ∆+ → −∆+ to shift the sign to the matrices. The correct mapping between
representations is only important when using both in the same Lagrangian to build invariants, e.g. to
show the equality
 
∗
−∆+
  1  
tr ∆†3×3 Σ ∆3×3 = − −∆+ , ∆0 , ∆− Σ  ∆0  . (A8)
 
2 −
∆

Other useful identities to build the scalar potential:

 2  
tr ∆†3×3 ∆3×3 = 2 tr ∆†3×3 ∆3×3 ∆†3×3 ∆3×3 , (A9)
   
= tr ∆†2×2 ∆2×2 ∆†2×2 ∆2×2 + 2 det ∆†2×2 ∆2×2 . (A10)

† † † †
One
 can also show that the invariants
 tr∆
 ∆ ∆∆ and tr∆ ∆ tr∆∆ can be expressed via
† † †
tr ∆2×2 ∆2×2 ∆2×2 ∆2×2 and det ∆2×2 ∆2×2 .
To form Yukawa couplings with the right-handed neutrinos, it is convenient to define an SU (2)L
doublet with opposite hypercharge, e.g. via H̃ = −iσ2 H ∗ . For the nontrivial SU (2)0 fields ∆ and Σ, the
corresponding definition is
˜ ≡ −iσ2 ε0 ∆∗ ε0 ∼ (1, 2, −1)(3) ,
∆ (A11)
 21

where ε0 acts on the SU (2)0 indices and takes the form

 
0 0 1
ε0 ∝ 0 −1 0 , (A12)
 
1 0 0
when ∆ (or Σ) is written as a 3 × 3-matrix.

2. Elimination of Ω++ from hΩi

In the discussion of the vacuum structure of the potential V (Ω) in Sec. V C we made use of the fact
that a VEV of Ω++ can be rotated away via SU (2)0 transformations. We will now briefly proof this
claim. We decompose the complex fields Ω++ and Ω+ into real and imaginary parts. It is clear that a
z-transformation can be used to make Ω++ real, so a general SU (2)0 transformation takes the form
      T
exp −iz̃ Tz(5) exp −iy Ty(5) exp −iz Tz(5) a, −(b + ic), d, b − ic, a . (A13)

Ignoring the z̃ transformation for now, the demand for a vanishing first component of the above vector
takes the form (split into real and imaginary part):
cos y (−2a cos z sin z) + sin y (c cos z − b sin z) = 0 , (A14)
√
a(3 + cos 2y) cos 2z + 6 d sin2 y + 2 sin 2y (b cos z + c sin z) = 0 . (A15)
The first equation can be readily solved for given z, so we plug the solution into the second equation to
obtain
√ √ 
−4b2 + 4c2 + 6 ad + a2 + 4(b2 + c2 ) cos 2z − a


6 d cos 4z + a cos 6z = 0 , (A16)

which can be shown to have real solutions by expressing cos nz through tan z ≡ t:
 √   √ 
!
f (t) ≡ c2 + 2a2 − b2 + 2c2 + 6 ad t2 + c2 − 2a2 − 2b2 + 6 ad t4 − b2 t6 = 0 . (A17)

f (t) has real zeros because f (0) > 0 and f (t → ∞) < 0. Hence we always find y, z to eliminate the first
component of Ω (and also the last one since Ω is hermitian). The final z̃ transformation can be used to
0
make the Ω+ component real.
An analogous conclusion can be reached concerning the elimination of Ω+ instead of Ω++ .

Appendix B: Different SU (2)0 Charge Assignments

Putting µ, e and τ in an SU (2)0 -triplet seems natural, but is not the only possibility. We will now
briefly discuss the other scheme, namely e ∼ 1, (µ, τ ) ∼ 2, i.e. the leptons form a reducible representation
1 ⊕ 2 under SU (2)0 (both left- and right-handed ones). Since the electron does not take part in the gauge
interactions, there are no dangerous LFV involving the electron on tree-level, so a low SU (2)0 breaking
scale is possible. The charged leptons now have masses me , mµ = mτ , so we need to break this symmetry
using a Higgs field with leptospin-1. Putting the right-handed neutrinos in the same reps., i.e. Ne ∼ 1,
(Nµ , Nτ ) ∼ 2 is problematic because only Ne can acquire a Majorana mass term, the invariant εab NaT Nb
vanishes due to symmetry. We have therefore no good zeroth-order mass matrix, but would have to
generate a proper MR via SU (2)0 breaking.
Taking the right-handed neutrinos once again as a leptospin-1 field brings back the Majorana ma-
trix (78), but of course does not allow a Dirac mass term νhHiN , so with this assignment, mD has to be
generated by SU (2)0 breaking. Both schemes provide bad starting points and seem unnatural, which is
why we will not discuss them further.
It is, of course, possible to build viable models using different neutrino mass generation schemes than
seesaw-I. In Refs. [34] the 1 ⊕ 2 representation was discussed in a similar context to build Le − Lµ − Lτ
symmetric neutrino mass matrices, using either seesaw-II or MR generation via VEVs, as discussed
above.

[1] P. Langacker, Rev. Mod. Phys. 81, 1199 (2009) [arXiv:0801.1345 [hep-ph]].
 22

[2] R. Foot, Mod. Phys. Lett. A 6, 527 (1991); R. Foot, X. G. He, H. Lew and R. R. Volkas, Phys. Rev. D 50,
4571 (1994) [arXiv:hep-ph/9401250].
[3] X. G. He, G. C. Joshi, H. Lew and R. R. Volkas, Phys. Rev. D 44, 2118 (1991).
[4] N. F. Bell, R. R. Volkas, Phys. Rev. D 63, 013006 (2000) [arXiv:hep-ph/0008177]; A. S. Joshipura and S. Mo-
hanty, Phys. Lett. B 584, 103 (2004) [arXiv:hep-ph/0310210]; A. Bandyopadhyay, A. Dighe and A. S. Jo-
shipura, Phys. Rev. D 75, 093005 (2007) [arXiv:hep-ph/0610263]; A. Samanta, JCAP 1109, 010 (2011)
[arXiv:1001.5344 [hep-ph]]; J. Heeck and W. Rodejohann, J. Phys. G 38, 085005 (2011) [arXiv:1007.2655
[hep-ph]]; H. Davoudiasl, H. S. Lee and W. J. Marciano, Phys. Rev. D 84, 013009 (2011) [arXiv:1102.5352
[hep-ph]].
[5] S. Choubey, W. Rodejohann, Eur. Phys. J. C 40, 259-268 (2005) [arXiv:hep-ph/0411190].
[6] E. Ma, D. P. Roy and S. Roy, Phys. Lett. B 525, 101 (2002) [arXiv:hep-ph/0110146]; E. Ma, D. P. Roy,
Phys. Rev. D 65, 075021 (2002).
[7] S. Baek, N. G. Deshpande, X. G. He and P. Ko, Phys. Rev. D 64, 055006 (2001) [arXiv:hep-ph/0104141];
S. Baek and P. Ko, JCAP 0910, 011 (2009) [arXiv:0811.1646 [hep-ph]].
[8] S. N. Gninenko and N. V. Krasnikov, Phys. Lett. B 513, 119 (2001) [arXiv:hep-ph/0102222].
[9] G. Altarelli, F. Feruglio, Rev. Mod. Phys. 82, 2701-2729 (2010) [arXiv:1002.0211 [hep-ph]]; H. Ishimori,
T. Kobayashi, H. Ohki, Y. Shimizu, H. Okada, M. Tanimoto, Prog. Theor. Phys. Suppl. 183, 1-163 (2010)
[arXiv:1003.3552 [hep-th]].
[10] K. S. Babu, C. F. Kolda and J. March-Russell, Phys. Rev. D 57, 6788 (1998) [arXiv:hep-ph/9710441].
[11] K. Nakamura et al. [Particle Data Group], J. Phys. G 37, 075021 (2010).
[12] J. P. Leveille, Nucl. Phys. B 137, 63 (1978); K. R. Lynch, Phys. Rev. D 65, 053006 (2002) [arXiv:hep-
ph/0108080].
[13] B. Malaescu, arXiv:1006.4739 [hep-ph].
[14] D. Buskulic et al. [ALEPH Collaboration], Z. Phys. C 66, 3-18 (1995).
[15] J. Abdallah et al. [DELPHI Collaboration], Eur. Phys. J. C 51, 503 (2007) [arXiv:0706.2565 [hep-ex]].
[16] See for instance J. Bernabeu et al., arXiv:1005.3146 [hep-ph].
[17] J. Erler, P. Langacker, Phys. Lett. B 456, 68-76 (1999) [arXiv:hep-ph/9903476]; J. Erler, arXiv:hep-
ph/0005084.
[18] A. Pukhov et al., arXiv:hep-ph/9908288; E. Boos et al. [CompHEP Collaboration], Nucl. Instrum. Meth. A
534, 250 (2004) [arXiv:hep-ph/0403113].
[19] P. Minkowski, Phys. Lett. B 67, 421 (1977); M. Gell-Mann, P. Ramond, and R. Slansky, in Supergravity,
p. 315, edited by F. van Nieuwenhuizen and D. Freedman, North Holland, Amsterdam, 1979; T. Yanagida,
Proc. of the Workshop on Unified Theory and the Baryon Number of the Universe, KEK, Japan, 1979; R.N.
Mohapatra and G. Senjanović, Phys. Rev. Lett. 44, 912 (1980).
[20] P. H. Frampton, S. L. Glashow and D. Marfatia, Phys. Lett. B 536, 79 (2002) [arXiv:hep-ph/0201008];
W. Grimus and L. Lavoura, J. Phys. G 31, 693 (2005) [arXiv:hep-ph/0412283].
[21] K. Abe et al. [T2K Collaboration], Phys. Rev. Lett. 107, 041801 (2011) [arXiv:1106.2822 [hep-ex]].
[22] G. L. Fogli, E. Lisi, A. Marrone, A. Palazzo and A. M. Rotunno, Phys. Rev. D 84, 053007 (2011)
[arXiv:1106.6028 [hep-ph]].
[23] G. C. Branco, P. M. Ferreira, L. Lavoura, M. N. Rebelo, M. Sher, J. P. Silva, arXiv:1106.0034 [hep-ph].
[24] W. Grimus, L. Lavoura, O. M. Ogreid, P. Osland, J. Phys. G 35, 075001 (2008) [arXiv:0711.4022 [hep-ph]].
[25] H. E. Haber, G. L. Kane, T. Sterling, Nucl. Phys. B 161, 493 (1979); M. Sher, Phys. Rept. 179, 273-418
(1989).
[26] [LEP Higgs Working Group for Higgs boson searches and ALEPH and DELPHI and L3 and OPAL Collab-
orations], arXiv:hep-ex/0107031; J. Abdallah et al. [DELPHI Collaboration], Eur. Phys. J. C 34, 399-418
(2004) [arXiv:hep-ex/0404012].
[27] K. Hayasaka et al., Phys. Lett. B 687, 139 (2010) [arXiv:1001.3221 [hep-ex]].
[28] L. Lavoura, Eur. Phys. J. C 29, 191-195 (2003) [arXiv:hep-ph/0302221].
[29] D. Chang, W. -F. Chang, C. -H. Chou, W. -Y. Keung, Phys. Rev. D 63, 091301 (2001) [arXiv:hep-
ph/0009292].
[30] V. Barger, H. E. Logan, G. Shaughnessy, Phys. Rev. D 79, 115018 (2009) [arXiv:0902.0170 [hep-ph]].
[31] S. de Visscher, J. -M. Gerard, M. Herquet, V. Lemaitre, F. Maltoni, JHEP 0908, 042 (2009) [arXiv:0904.0705
[hep-ph]].
[32] L. Basso, A. Belyaev, S. Moretti and C. H. Shepherd-Themistocleous, Phys. Rev. D 80, 055030 (2009)
[arXiv:0812.4313 [hep-ph]].
[33] See, for example, F. Wilczek and A. Zee, Phys. Rev. Lett. 42, 421 (1979); T. Yanagida, Phys. Rev. D 20,
2986 (1979); Y. Chikashige, G. Gelmini, R. D. Peccei and M. Roncadelli, Phys. Lett. B 94, 499 (1980);
E. Papantonopoulos and G. Zoupanos, Phys. Lett. B 110, 465 (1982); Z. G. Berezhiani, Phys. Lett. B 129,
99 (1983); R. Foot, G. C. Joshi, H. Lew and R. R. Volkas, Phys. Lett. B 226, 318 (1989); K. Bandyopadhyay,
D. Choudhury and U. Sarkar, Phys. Rev. D 43, 1646 (1991); D. S. Shaw and R. R. Volkas, Phys. Rev. D 47,
241 (1993) [arXiv:hep-ph/9211209].
[34] R. Kuchimanchi and R. N. Mohapatra, Phys. Rev. D 66, 051301 (2002) [arXiv:hep-ph/0207110]; R. Kuchi-
manchi and R. N. Mohapatra, Phys. Lett. B 552, 198 (2003) [arXiv:hep-ph/0207373]; K. L. McDonald and
B. H. J. McKellar, Phys. Rev. D 73, 073004 (2006) [arXiv:hep-ph/0603129].