Nucl - Phys.B v.687 PDF

Nuclear Physics B 687 (2004) 3–30
www.elsevier.com/locate/npe
Constraining Z from supersymmetry breaking

O.C. Anoka a , K.S. Babu a , I. Gogoladze a,b,1
a Department of Physics, Oklahoma State University, Stillwater, OK 74078, USA
b Department of Physics, University of Notre Dame, Notre Dame, IN 46556, USA
Received 30 January 2004; accepted 10 March 2004
Abstract
We suggest and analyze a class of supersymmetric Z models based on the gauge symmetry
U (1)x = xY − (B − L), where Y is the Standard Model hypercharge. For 1 < x < 2, the U (1)x
D-term generates positive contributions to the slepton masses, which is shown to solve the tachyonic
slepton problem of anomaly mediated supersymmetry breaking (AMSB). The resulting models are
very predictive, both in the SUSY breaking sector and in the Z sector. We find MZ = 2–4 TeV and
the Z–Z mixing angle ξ 0.001. Consistency with symmetry breaking and AMSB phenomenology
renders the Z “leptophobic”, with Br(Z → + − ) (1–1.6)% and Br(Z → q q̄) 44%. The
lightest SUSY particle is either the neutral Wino or the sneutrino in these models.
 2004 Elsevier B.V. All rights reserved.
PACS: 12.60.-i; 12.60.Cn; 12.60.Fr; 12.60.Jv; 13.38.Dg; 14.60.St; 14.70.Pw
1. Introduction
One of the simplest extensions of the Standard Model (SM) is obtained by adding a
U (1) factor to the SU(3)C × SU(2)L × U (1)Y gauge structure. Such U (1) factors arise
quite naturally when the SM is embedded in a grand unified group such as SO(10), SU(6),
E6 , etc., [1,2]. While it is possible that such U (1) symmetries are broken spontaneously
near the grand unification scale, it is also possible that some of the U (1) factors survive
down to the TeV scale. In fact, if there is low-energy supersymmetry, it is quite plausible
that the U (1) symmetry is broken along with supersymmetry at the TeV scale. The Zχ
and Zψ models arising from SO(10) → SU(5) × U (1)χ and E6 → SO(10) × U (1)ψ are
E-mail addresses: anoka@okstate.edu (O.C. Anoka), babu@okstate.edu (K.S. Babu), gogoladze.1@nd.edu

(I. Gogoladze).
1 On a leave of absence from: Andronikashvili Institute of Physics, GAS, 380077 Tbilisi, Georgia.
0550-3213/$ – see front matter  2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2004.03.009
4 O.C. Anoka et al. / Nuclear Physics B 687 (2004) 3–30
two popular extensions which have attracted much phenomenological attention [1–8]. Z
associated with the left–right symmetric extension of the Standard Model does not require
a grand unified symmetry. Other types of U (1) symmetries, which do not resemble the ones
with a GUT origin, are known to arise in string theory, in the free-fermionic construction
as well as in orbifold and D-brane models [9–11]. Gauge kinetic mixing terms of the
[12] which will be generated through renormalization group flow below the
type B µν Zµν
unification scale can further disguise the couplings of the Z .
The properties of the Z gauge boson—its mass, mixing and couplings to fermions—
associated with the U (1) gauge symmetry are in general quite arbitrary [13]. This is
especially so when the low-energy theory contains new fermions for anomaly cancellation.
In this paper we propose and analyze a special class of U (1) models wherein the Z
properties get essentially fixed from constraints of SUSY breaking. We have in mind the
anomaly mediated supersymmetric (AMSB) framework [14,15]. In its minimal version,
with the Standard Model gauge symmetry, it turns out that the sleptons of AMSB
become tachyonic. We suggest the U (1) symmetry, identified as U (1)x = xY − (B − L),
where Y is the Standard Model hypercharge, as a solution to the negative slepton mass
problem of AMSB. This symmetry is automatically free of anomalies with the inclusion
of right-handed neutrinos. It is shown that the D-term of this U (1)x provides positive
contributions to the slepton masses, curing the tachyonic problem. The consistency of
symmetry breaking and the SUSY spectrum points towards a specific set of parameters
in the Z sector. For example, 1 < x < 2 is needed for the positivity of the left-handed
and the right-handed slepton masses. Furthermore, the U (1)x gauge coupling, gx , is
fixed to be between 0.4–0.5. The resulting Z is found to be “leptophobic” [16] with
Br(Z → + − ) (1–1.6)% and Br(Z → q q̄) 44%.
AMSB models are quite predictive as regards the SUSY spectrum. The masses of the
scalar components of the chiral supermultiplets in AMSB scenario are given by [14,15]

2 φj 1 2 ∂ φ ∂ φ
m φ = Maux β(Y ) γφij + β(g) γφij , (1)
i 2 ∂Y ∂g
where summations over the gauge couplings g and the Yukawa couplings Y are assumed.
φ
γφij are the one-loop anomalous dimensions, β(Y ) is the beta function for the Yukawa
coupling Y , and β(g) is the beta function for the gauge coupling g. Maux is the vacuum
expectation value of a “compensator superfield” [14] which sets the scale of SUSY
breaking. The gaugino mass Mg , the trilinear soft supersymmetry breaking term AY and
the bilinear SUSY breaking term B are given by [14,15]
β(g) β(Y )
Mg = Maux , AY = − Maux , B = −Maux (γHu + γHd ). (2)
g Y
We see that the SUSY masses are completely fixed in the AMSB framework once the
spectrum of the theory and Maux are specified.
The negative slepton mass problem arises in AMSB because in Eq. (1) the gauge beta
φ
functions for SU(2)L and U (1)Y are positive, γφij are negative, and the Yukawa couplings
are small for the first two families of sleptons. In our Z models, there are additional
positive contributions from the U (1)x D-terms which render these masses positive.
O.C. Anoka et al. / Nuclear Physics B 687 (2004) 3–30 5
In Ref. [17] the negative slepton mass problem of AMSB has been solved with explicit
Fayet–Iliopoulos terms added to the theory. In contrast, in our models, the D-term is
calculable, which makes the Z sector more predictive. We find MZ = 2–4 TeV and the
Z–Z mixing angle ξ 0.001. Constraints from the electroweak precision observables are
satisfied, with the Z model giving a slightly better fit compared to the Standard Model.
Other attempts to solve the negative slepton mass problem of AMSB generally assume
TeV-scale new physics [18–20] or a universal scalar mass of non-AMSB origin [21]. In
Ref. [20] we have shown how a non-Abelian horizontal symmetry which is asymptotically
free solves the problem. Some of the techniques we use here for the symmetry breaking
analysis are similar to Ref. [20].
The plan of the paper is as follows. In Section 2 we introduce our model. In Section 3 we
analyze the Higgs potential of the model. In Section 4 we present formulas for the SUSY
spectrum. Section 5 contains our numerical results for the SUSY spectrum as well as for the
Z mass and mixing. In Section 6 we analyze the partial decay modes of the Z . In Section 7
we analyze other experimental test of the model. Here we show the consistency of our
models with the precision electroweak data. Section 8 has our conclusions. In Appendix A
we give the relevant expressions for the beta functions, anomalous dimensions as well as
for the soft masses.
2. U (1)x model
We present our model in this section. We consider adding an extra U (1) gauge group to
the Standard Model gauge structure of MSSM. The model is then based on the gauge group
SU(3)C ⊗ SU(2)L ⊗ U (1)Y ⊗ U (1)x , where the U (1)x charge is given by the following
linear combination of hypercharge Y and B − L:
U (1)x = xY − (B − L). (3)

The particle content of the model and the U (1)x charge assignment are shown in Table 1.
Besides the MSSM particles, the model has new particles {νic , ν c , ν̄ c , S+ andS− } which are
all singlets of the Standard Model gauge group.
In order for L̃i and ẽic sleptons to have positive mass-squared from the U (1)x D-term,
the charges of Li and eic must be of the same sign. This is possible only for 1 <
x < 2. We shall confine to this range of x, which is an important restriction on this
class of models. The νic fields are needed for U (1)x anomaly cancellation. S+ and S−
are the Higgs superfields responsible for U (1)x symmetry breaking. The ν c + ν̄ c pair
facilitates symmetry breaking within the AMSB framework. The superpotential of the
model consistent with the gauge symmetries is given by:
Table 1
Particle content and charge assignment of the U (1)x model. Here i = 1–3 is the family index
Superfield Qi uci dic Li eic Hu Hd νic νc ν̄ c S+ S−
x −1 − 2x 1 x +1 − x2 + 1 x − x2
U (1)x 6 3 3 +3 3 3 x−1 2 −1 −1 1 2 −2
W = (Yu )ij Qi Hu ucj + (Yd )ij Qi Hd djc + (Yl )ij Li Hd ejc + µHu Hd

3
+ µ S+ S− + fνic νic νic S+ + fν c ν c ν c S+ + hν̄ c ν̄ c S− + Mν c ν c ν̄ c . (4)
i=1
Here i, j = 1, 2, 3 are the family indices. The mass parameters µ and µ are of order TeV,
which may have a natural origin in AMSB [14]. In general, one can write additional mass
terms of the form Mi νic ν̄ c in the superpotential. Such terms will have very little effect on
the symmetry breaking analysis that follows. We forbid such mass terms by invoking a
discrete symmetry (such as a Z2 ) which differentiates ν c from νic .
Small neutrino masses are induced in the model through the seesaw mechanism.
However, the νic fields, which remain light to the TeV scale, are not to be identified as
the traditional right-handed neutrinos involved in the seesaw mechanism. The heavy fields
which are integrated out have U (1)x -invariant mass terms. Specifically, the following
effective non-renormalizable operators emerge after integrating out the heavy neutral
lepton fields:
Yν2ij
Lνeff = Li Lj Hu Hu S− . (5)
MN2
Here MN represents the masses of the heavy neutral leptons. For MN ∼ 109 GeV and
S− ∼ TeV, sub-eV neutrino masses are obtained. Note that we have not allowed neutrino
Dirac Yukawa couplings of the form hνij Li νjc Hu , which would generate Majorana masses
of order MeV for the light neutrinos. We forbid such terms by a global symmetry G, either
discrete or continuous. In our numerical examples we shall assume this symmetry to be
non-Abelian, with νic transforming as a triplet (for example, G can be O(3), S4 , A4 , etc.).
Such a symmetry would imply that fνic in Eq. (4) are equal for i = 1–3.
3. Symmetry breaking
The scalar potential (involving Hu , Hd , S+ , S− fields) of the model is given by:

2 2
V = MH u
+ µ2 |Hu |2 + MH d
+ µ2 |Hd |2 + MS2+ + µ 2 |S+ |2

+ MS2− + µ 2 |S− |2 + Bµ(Hu Hd + h.c.) + B µ (S+ S− + h.c.)
1 2 1
+ g12 + g22 |Hu |2 − |Hd |2 + g22 |Hu Hd |2
8 2
2
1 x x
+ gx2 |Hu |2 − |Hd |2 + 2|S+ |2 − 2|S− |2 , (6)
2 2 2
where the last term is the U (1)x D term. The B and the B terms for the model are given
by
B = −(γHu + γHd )Maux and B = −(γS+ + γS− )Maux , (7)
where the γ ’s are the one-loop anomalous dimensions given in Appendix A, Eqs. (A.6),
(A.7), (A.11) and (A.12).
We parameterize the VEVs of Hu , Hd , S+ and S− as

0 υd
Hu = , Hd = , S+ = z, S− = y. (8)
υu 0
In minimizing the potential, we have to keep in mind the fact that the VEVs of S+ and
S− should be much larger than the VEVs of Hu and Hd for a consistent picture. In
addition, the VEV of S+ should be greater than the VEV of S− in order for the D-term
contribution to the slepton masses to be positive. We have checked explicitly that all the
above-mentioned conditions are satisfied at the local minimum for a restricted choice of
model parameters. The physical Higgs bosons as well as the sleptons acquire positive mass-
squared, while generating a Z mass and Z–Z mixing angle consistent with experimental
constraints.
Minimization of the potential leads to the following conditions:
2Bµ
sin 2β = , (9)
2µ2 + MH
2 + M2
u Hd
2 − M 2 tan2 β
MZ2 MH Hu x 2 gx2 υ 2 xgx2 u2 cos 2ψ
= −µ2 + d
− − , (10)
2 tan2 β − 1 4 cos 2β
−2B µ
sin 2ψ = , (11)
2µ 2 + MS2+ + MS2−
MZ2 2
MS2− − MS2+ tan2 ψ x 2 gx2 υ 2 xgx2 υ 2 cos 2β
= −µ + + − . (12)
2 (tan2 ψ − 1) 4 cos 2ψ

Here MZ2 = x 2 gx2 υ 2 /2 + 8gx2 u2 , tan β = υu /υd , tan ψ = z/y, υu2 + υd2 = υ = 174 GeV

and z2 + y 2 = u.
To see the consistency of symmetry breaking, we need to calculate the Higgs boson
mass-squared and establish that they are all positive. We parameterize the Higgs fields (in
the unitary gauge) as

H + sin β υd + √1 (φ1 + i sin βφ3 )
Hu = , H d = 2 ,
υu + √1 (φ2 + i cos βφ3 ) H − cos β
2
1 1
S+ = z + √ (φ4 + i cos ψφ5 ), S− = y + √ (φ6 + i sin ψφ5 ). (13)
2 2
The CP-odd Higgs bosons {φ3 , φ5 } have masses given by
2Bµ 2B µ
m2A = , m2A = − . (14)
sin 2β sin 2ψ
The mass matrix for the CP-even neutral Higgs bosons {φ1 , φ2 , φ4 , φ6 } is given by
 
(M2 )11 (M2 )12 −2xgx2 υd z 2xgx2 υd y
2  (M2 )12 (M2 )22 2xgx2 υu z −2xgx2 υu y 
M cp-even =   −2xg 2 υd z 2xg 2 υu z
, (15)
x x (M2 )33 (M2 )34 
2xgx2 υd y −2xgx2 υu y (M2 )34 (M2 )44
where
1 2 2 2
M2 11
= m2A sin2 β + MZ2 cos2 β +
x gx υ cos2 β , (16)
2
2 1
M 12 = −m2A sin β cos β − MZ2 sin β cos β − x 2 gx2 υ 2 sin β cos β, (17)
2
2 1
M 22 = m2A cos2 β + MZ2 sin2 β + x 2 gx2 υ 2 sin2 β , (18)
2
2
M 33 = m2A cos2 ψ + 8gx2 z2 , (19)
2
M 34 = −m2A sin ψ cos ψ − 8gx2 yz, (20)
2
M 44 = m2A sin2 ψ + 8gx2 y 2 . (21)
It is instructive to analyze the effect of the U (1)x D-term on the mass of the lightest MSSM
Higgs boson h. Consider the upper left 2 × 2 subsector of the CP-even Higgs boson mass
matrix. It has eigenvalues given by

1 2 x 2 gx2 υ 2 x 2 gx2 υ 2 2
λ1,2 = mA + MZ2 + ∓ m2A + MZ2 + − 4m2A MZ2 cos2 2β
2 2 2
2 2 2 1/2
x gx υ
− 4m2A cos2 2β . (22)
2
From Eq. (22) we obtain an upper limit on mh :

x 2 gx2 υ 2
mh + MZ2 | cos 2β|. (23)
2
The mixing between the doublets and the singlets will reduce the upper limit further. In
fact, we find this mixing effect to be significant.
The lower 2 × 2 subsector of Eq. (15) has eigenvalues

1 2 2 2
λ1,2 = 8gx u + m2A ∓ 8gx2 u2 + m2A − 4m2A 8gx2 u2 cos2 2ψ . (24)
2
From Eq. (24) we obtain an upper bound of the lightest Higgs mass for the SU(2) singlet
sector:
mh mA | cos 2ψ|. (25)
The above upper limit on mh is affected only minimally by the mixing between the doublet
and the singlet Higgs fields.
As in the MSSM, the mass of the charged Higgs boson H ± is given by
m2H ± = m2A + MW
2
. (26)
We now turn to the supersymmetric fermion masses. The (Majorana) mass matrix of the
neutralinos {B̃, W̃3 , H̃d0 , H̃u0 , B̃ , S̃+ , S̃− } is given by
 υd υu 
M1 0 −√ g1 √ g1 0 0 0
2 2
 0 υd
√ −√ υu 
 M2 g2 g2 0 0 0 
 2 2 
−√ υd √υd
−µ − υd
√ 
 g1 g 2 0 xgx 0 0 
 2 2 2 
M =
(0)  √υu
g1 − g2 υu
√ −µ 0 υu
√ xgx 0 0 ,

 2 2 2
√ √ 
 0 0 − √υd
xg υu
√ xg M 2 2g z −2 2g y 
 x x 1 x x 
 2 2
√ 
 0 0 0 0 2 2gx z 0 µ 
√
0 0 0 0 −2 2gx y µ 0
(27)
where M1 , M1 and M2 are the gaugino masses for U (1)Y , U (1)x and SU(2)L . The physical
neutralino masses mχ̃ 0 (i = 1–7) are obtained as the eigenvalues of this mass matrix. We
i
denote the diagonalizing matrix as O:
OM(0)O T = diag{mχ̃ 0 , mχ̃ 0 , mχ̃ 0 , mχ̃ 0 , mχ̃ 0 , mχ̃ 0 , mχ̃ 0 }. (28)
1 2 3 4 5 6 7
In the basis {W̃ + , H̃u+ }, {W̃ − , H̃d− } the chargino (Dirac) mass matrix is

M2 g2 υd
M(c) = . (29)
g2 υu µ
This matrix is diagonalized by a biunitary transformation V ∗ M(c) U −1 = diag{mχ̃ ± , mχ̃ ± }.
1 2
The Z–Z mixing matrix is given by
2
MZ γ MZ2
M2Z –Z = , (30)
γ MZ2 MZ2
where
−xgx υ2 2 x 2 gx2 υ 2
γ= , MZ2 = g1 + g22 , MZ2 = + 8gx2 u2 . (31)
g12 + g22 2 2
The physical mass eigenstates Z1 and Z2 with masses MZ1 , MZ2 are
Z1 = Z cos ξ + Z sin ξ, (32)
Z2 = −Z sin ξ + Z cos ξ, (33)
where

1 2 2
MZ2 1 ,Z2 =
MZ + MZ2 ∓ MZ2 − MZ2 + 4γ 2 MZ4 . (34)
2

The Z–Z mixing angle ξ is given by

1 2γ MZ2
ξ = arctan −γ MZ2 /MZ2 . (35)
2 MZ2 − MZ2
in the Lagrangian [12,13].

We have ignored kinetic mixing of the form B µν Zµν
The masses of the heavy right-handed neutrinos are given by
mνic = fνic z, (36)

where i = 1–3 is the family index. The fourth right-handed neutrino mixes with the ν̄ c
νc
field forming two Majorana fermions. The masses are the eigenvalues of the mass matrix

fν c z Mν c
Mν ν̄ =
c c , (37)
Mν c hy
where Mν c is the mass parameter that appears in the superpotential of Eq. (4). We denote
the eigenstates of this matrix as ω1 , ω2 and the mass eigenvalues as mω1 and mω2 .
4. The SUSY spectrum
4.1. Slepton masses
The slepton mass-squareds are given by the eigenvalues of the mass matrices

m2˜ mei (AYli − µ tan β)
l
Ml˜2 = i
, (38)
mei (AYli − µ tan β) m2ẽc
i
where i = e, µ, τ , and

2
Maux 3 3 x 2
m2l˜ = Y l β(Y l ) − g 2 β(g 2 ) + g1 β(g1 ) + 2 1 − g x β(gx )
i (16π 2 ) i i
2 10 2

1 x
+ m2ei + − + sin2 θW cos 2βMZ2 + 2gx2 1 − z2 − y 2 , (39)
2 2
2
Maux 6
mẽ2c = 2Y li β(Y li ) − g 1 β(g 1 ) + 2(x − 1) 2
gx β(g x )
i (16π 2 ) 5

+ m2ei − sin2 θW cos 2βMZ2 + 2gx2 (x − 1) z2 − y 2 . (40)
The SUSY soft masses are calculated from the RGE given in Appendix A (Eqs. (A.15),
(A.21)). Note the positive contribution from the U (1)x D-terms in Eqs. (39) and (40),
given by the terms +2gx2 (1 − x/2)(z2 − y 2 ) and +2gx2 (x − 1)(z2 − y 2 ). There are also
negative contributions proportional to β(gx ), but in our numerical solutions, the positive
D-term contributions are larger than the negative contributions. We seek solutions where
z = S+ and y = S− are much larger than υu , υd , of order TeV, with z y.
The left-handed sneutrino masses are given by

2
Maux 3 3 x 2
mν̃L =
2
− g2 β(g2 ) − g1 β(g1 ) − 2 1 − gx β(gx )
i (16π 2 ) 2 10 2

1 x 2
+ cos 2βMZ2 + 2gx2 1 − z − y2 . (41)
2 2
4.2. Squark masses
The mixing matrix for the squark sector is similar to the slepton sector. The diagonal
entries of the up and the down squark mass matrices are given by

Q̃ 1 x 1 2
m2Ũ = m2soft Q̃i + m2Ui + 4MW 2
− MZ2 cos 2β + 2gx2 − z − y2 ,
i i 6 6 3

Ũ c 2 2 2x 1 2
m2Ũ c = m2soft ic + m2Ui − MW − MZ2 cos 2β + 2gx2 − + z − y2 ,
i Ũi 3 3 3

Q̃ 1 x 1 2
m2D̃ = m2soft Q̃i + m2Di − 2MW 2
+ MZ2 cos 2β + 2gx2 − z − y2 ,
i i 6 6 3

D̃ c 1 x 1 2
m2D̃ c = m2soft ic + m2Di + MW 2
− MZ2 cos 2β + 2gx2 + z − y2 . (42)
i D̃i 3 3 3
Here mUi and mDi are quark masses of different generations, i = 1, 2, 3. The squark soft
masses are obtained from the RGE as
2 Q̃i Maux 2 1 3 8
msoft Q̃ = 2
Yui β(Yui ) + Ydi β(Ydi ) − g1 β(g1 ) − g2 β(g2 ) − g3 β(g3 )
i 16π 30 2 3
2
x 1
−2 − gx β(gx ) , (43)
6 3
2 Ũic Maux 2 8 8
msoft c = 2
2Yui β(Yui ) − g1 β(g1 ) − g3 β(g3 )
Ũi 16π 15 3
2
2x 1
−2 − + gx β(gx ) , (44)
3 3
2 D̃ic Maux 2 2 8

x 1 2

msoft c = 2Ydi β(Ydi ) − g1 β(g1 ) − g3 β(g3 ) − 2 + gx β(gx ) .
D̃i 16π 2 15 3 3 3
(45)
4.3. Heavy sneutrino masses
The heavy right-handed sneutrinos (ν̃ic ) split into scalar (ν̃is

c c
) and pseudoscalar (ν̃ip )
components with masses given by
2
Maux
m2ν̃ c = 2
4fνic β(fνic ) − 2gx β(gx ) − 2gx2 z2 − y 2 + 2µ fνic y
is (16π )
+ 4fν2c z2 + 2fνic Aνi z, (46)
i
2
Maux
m2ν̃ c = 2
4fνic β(fνic ) − 2gx β(gx ) − 2gx2 z2 − y 2 − 2µ fνic y
ip (16π )
+ 4fν2c z2 − 2fνic Aνi z. (47)
i
c
As for the fourth heavy sneutrino, there is mixing between the ν̃ c and the ν̄˜ fields. This
leads to two 2 × 2 mass matrices, one for the scalars, and one for the pseudoscalars. They
are given by
Bν c ν̄ c
m2ν̃ c 2Mν c (fν c z + hy + 2 )
s
Mν̃2c = Bν c ν̄ c , (48)
s 2Mν c (fν c z + hy + 2 ) m2˜ c
ν̄ s
m2ν̃ c 2Mν c (fν c z + hy + Bν c ν̄ c
p 2 )
Mν̃2c = Bν c ν̄ c , (49)
p 2Mν c (fν c z + hy + 2 ) m2˜ c
ν̄ p
where
2
Maux
m2ν̃ c = 2
4fν c β(fν c ) − 2gx β(gx ) − 2gx2 z2 − y 2 + 2µ fν c y + 4fν2c z2
s (16π )
+ 2fν c Aν c z + Mν2c , (50)
Maux
2
m2ν̃ c = 2
4fν c β(fν c ) − 2gx β(gx ) − 2gx2 z2 − y 2 − 2µ fν c y + 4fν2c z2
p (16π )
− 2fν c Aν c z + Mν2c , (51)
2
Maux
m2˜ c = 2
4hβ(h) − 2gx β(gx ) + 2gx2 z2 − y 2 + 2µ hz + 4h2 y 2
ν̄ s (16π )
+ 2hAh y + Mν2c , (52)
2
Maux
m2˜ c = 2
4hβ(h) − 2gx β(gx ) + 2gx2 z2 − y 2 − 2µ hz + 4h2 y 2
ν̄ p (16π )
− 2hAh y + Mν2c , (53)
B ν c ν̄ c = −Maux (γ + γ ).
νc ν̄ c (54)
Here s (p) stands for scalar (pseudoscalar). The beta functions, gamma functions and the A
terms are given in Appendix A, Eqs. (A.16)–(A.22). We shall denote the mass eigenstates
of the scalars as ω̃1s , ω̃2s with masses m2ω̃ , m2ω̃ , and the pseudoscalars as ω̃1p , ω̃2p with
1s 2s
masses m2ω̃ , m2ω̃ .
1p 2p
5. Numerical results for the spectrum
As inputs at MZ we choose the central values (in the MS scheme) [22]

1
α3 (MZ ) = 0.119, sin2 θW = 0.23113, . α(MZ ) =
(55)
127.922
We keep the top quark mass fixed at its central value, Mt = 174.3 GeV. We follow the
procedure outlined in Ref. [20] to determine the parameter tan β and the lightest MSSM
Higgs boson mass mh . The gauge couplings and the top quark Yukawa coupling are
evolved from the lower momentum scale to Q = 1 TeV, where the Higgs potential is
minimized. We use the Standard Model beta functions for this evolution. In determining
the top quark Yukawa coupling Yt (mt ), we use 2-loop QCD corrections to convert the
physical mass Mt into the running mass mt (mt ).
Table 2
Sparticle masses in Model 1 (x = 1.3) for the choice Maux = 56.398 TeV, tan ψ = −1.295, u = 2.054 TeV,
fν c = 0.28, fν c = 0.28, h = 0.921, gx = 0.41, Mν c = 1 TeV and Mt = 174.3 GeV. This corresponds to
i
tan β = 4.39, µ = −0.977 TeV, µ = 0.214 TeV, yb = 0.03
Particles Symbol Mass (TeV)
Neutralinos {mχ̃ 0 , mχ̃ 0 , mχ̃ 0 , mχ̃ 0 } {0.175165, 0.517, 0.980, 0.980}
1 2 3 4
Neutralinos {mχ̃ 0 , mχ̃ 0 , mχ̃ 0 } {0.206, 1.644, 3.278}
5 6 7
Charginos {mχ̃ ± , mχ̃ ± } {0.175171, 0.983}
1 2
Gluino M3 1.239
Neutral Higgs bosons {mh , mH , mA } {0.121, 0.793, 0.792}
Charged Higgs bosons mH ± 0.796
R-H sleptons {mẽR , mµ̃R , mτ̃1 } {0.215, 0.215, 0.205}
L-H sleptons {mẽL , mµ̃L , mτ̃2 } {0.249, 0.249, 0.257}
Sneutrinos {mν̃e , mν̃µ , mν̃τ } {0.220, 0.220, 0.220}
R-H down squarks {md̃ , ms̃R , mb̃ } {1.284, 1.284, 1.284}
R 1
L-H down squarks {md̃ , ms̃L , mb̃ } {1.186, 1.186, 1.028}
L 2
R-H up squarks {mũR , mc̃R , mt˜ } {1.098, 1.098, 0.644}
1
L-H up squarks {mũL , mc̃L , mt˜ } {1.184, 1.184, 1.099}
2
R-H scalar neutrinos {mν̃sc } (i = 1–3) 0.605
i
R-H pseudoscalar neutrinos {mν̃pc } (i = 1–3) 0.413
i
c
Heavy scalar neutrino (ν̃ c , ν̄˜ ) {mω̃1s , mω̃2s } {1.142, 3.644}
c
Heavy pseudoscalar neutrino (ν̃ c , ν̄˜ ) {mω̃ps , mω̃2p } {0.595, 1.439}
R-H neutrinos {mν c } 0.455
i
Heavy neutrinos (ν c , ν̄ c ) {mω1 , mω2 } {0.933, 1.635}
For the lightest Higgs boson mass of MSSM we use the 2-loop radiatively corrected
expression for m2h = (m2h )o + ∆m2h , where ∆m2h is given in Ref. [23].
We present numerical results for two models: Model 1 with x = 1.3, and Model 2 with
x = 1.6. In Model 1, the left-handed sleptons are heavier than the right-handed sleptons,
while the reverse holds for Model 2.
The value of Maux should be in the range Maux = 40–100 TeV if the SUSY particles
are to have masses in the range 100 GeV–2 TeV. In Table 2, corresponding to Model 1, we
choose Maux = 56.398 TeV. In Table 7 (for Model 2) we choose Maux = 59.987 TeV.
We have included the leading radiative corrections [24] to M1 , M2 and M3 in our
numerical study. In Model 1 we find M1 : M2 : M3 = 3.0 : 1 : 7.1. The minimization
conditions (Eqs. (9) and (10)) fix tan β = 4.39 in this model. The choice of gx = 0.41,
fνic = fν c = 0.28, and h = 0.921 are motivated by the requirements of consistent symmetry
breaking with S+ S− υu , υd , and the positivity of slepton masses. We find that the
model parameters are highly constrained. Only small deviations from the choice in Table 2
are found to be consistent.
From Table 2 we see that the lightest Higgs boson of the MSSM sector has mass
of 121 GeV. The lightest SUSY particle is the neutralino χ̃10 , which is approximately
a neutral Wino. This is a candidate for cold dark matter [25]. Note that χ̃10 is nearly
Table 3
Z mass and Z–Z mixing angle in Model 1 for the same set
of input parameters as in Table 2
Z boson mass MZ 2.383 TeV
Z–Z mixing angle ξ 0.001
Table 4
Eigenvectors of the neutralino mass matrix in Model 1. The unitary matrix O in Eq. (86) is the transpose of this
array
Fields χ̃10 χ̃20 χ̃30 χ̃40 χ̃50 χ̃60 χ̃70
B̃ −0.003 0.998 0.051 0.025 0.000 −0.001 0.000
W̃30 −0.997 0.001 −0.052 −0.058 0.000 0.002 0.000
H̃d0 0.078 0.054 −0.703 −0.704 −0.002 0.030 0.001
H̃u0 −0.004 0.019 −0.707 0.706 0.001 −0.042 0.016
B̃ 0.000 0.000 −0.004 −0.023 −0.026 −0.612 −0.790
S̃+ 0.000 0.000 −0.011 0.039 −0.597 0.642 −0.479
S̃− 0.000 0.000 −0.009 0.026 0.802 0.458 −0.382
Table 5
Eigenvectors of the chargino mass matrix in Model 1, where U , V are the unitary matrices that diagonalize the
chargino mass matrix (V ∗ M (c) U −1 = Mdiag )
(c)
U11 U12 U21 U22 V11 V12 V21 V22

0.994 0.110 −0.110 0.994 1.000 0.006 −0.006 1.000
mass degenerate with the lighter chargino χ̃1± (which is approximately the charged Wino).
The mass splitting mχ̃ 0 − mχ̃ ± = 180 MeV, where the bulk (173 MeV) arises from finite
1 1
electroweak radiative corrections [26], not shown in Table 2.
In the U (1)x sector, there is a relatively light neutral Higgs boson h with a mass of
60 GeV. This occurs since the parameter tan ψ = z/y is close to 1—a requirement for
consistent symmetry breaking (see Eq. (25)). h is an admixture of S+ and S− , and as
such has no direct couplings to the Standard Model fields. Its mass being below 100 GeV
is fully consistent with experimental constraints. The phenomenology of such a weakly
coupled light neutral Higgs boson will be discussed in Section 7.
The mass of the Z gauge boson and the Z–Z mixing angle are listed in Table 3 (for
Model 1). In Section 7 we show that these values are compatible with known experimental
constraints.
Table 4 lists the eigenvectors of the neutralino mass matrix. These will become relevant
in discussing the decays of the Z gauge boson. Tables 5 and 6 give the eigenvectors of the
chargino and the CP-even Higgs bosons, which will also be used in the study of Z decays.
Tables 7–11 are analogous to Tables 2–6, except that they now apply to Model 2 (with
x = 1.6). In this case, tan β = 5.83 and mh = 126 GeV. Here the right-handed sleptons
are heavier than the left-handed sleptons. In fact, in this Model, the LSP is the left-handed
sneutrino. This can also be a candidate for cold dark matter in the AMSB framework, as
Table 6
Eigenvectors of the CP-even Higgs boson mass matrix in Model 1. This array corresponds to X used in Eqs. (82)–
(84) and (109) of the text
Fields h h H H
Hd0 0.226 −0.025 0.974 −0.007
Hu0 0.967 −0.110 −0.227 0.027
S+ −0.050 −0.612 −0.010 −0.790
S− 0.104 0.783 −0.008 −0.613
Table 7
Sparticle masses in Model 2 (x = 1.6) for the choice Maux = 59.987 TeV, tan ψ = −1.202, u = 2.697 TeV,
fν c = 0.4, fν c = 0.4, h = 1.0, gx = 0.45, M1 = 2.197 TeV, Mν c = 1 TeV and Mt = 174.3 GeV. This
i
corresponds to tan β = 5.83, µ = −1.046 TeV, µ = −0.505 TeV, yb = 0.06
Particles Symbol Mass (TeV)
Neutralinos {mχ̃ 0 , mχ̃ 0 , mχ̃ 0 , mχ̃ 0 } {0.185.851, 0.550, 1.049, 1.050}
1 2 3 4
Neutralinos {mχ̃ 0 , mχ̃ 0 , mχ̃ 0 } {0.498, 2.840, 4.539}
5 6 7
Charginos {mχ̃ ± , mχ̃ ± } {0.185855, 1.051}
1 2
Gluino M3 1.298
Charged Higgs bosons mH ± 0.630
R-H sleptons {mẽR , mµ̃R , mτ̃1 } {0.383, 0.383, 0.385}
L-H sleptons {mẽL , mµ̃L , mτ̃2 } {0.213, 0.213, 0.210}
Sneutrinos {mν̃e , mν̃µ , mν̃τ } {0.174, 0.174, 0.174}
R-H down squarks {md̃ , ms̃R , mb̃ } {1.370, 1.370, 1.369}
R 1
L-H down squarks {md̃ , ms̃L , mb̃ } {1.267, 1.267, 1.087}
L 2
R-H up squarks {mũR , mc̃R , mt˜ } {1.031, 1.031, 0.406}
1
L-H up squarks {mũL , mc̃L , mt˜ } {1.264, 1.264, 1.1141}
2
R-H scalar neutrinos {mν̃sc } (i = 1–3) 1.583
i
R-H pseudoscalar neutrinos {mν̃pc } (i = 1–3) 1.129
i
c
Heavy scalar neutrino (ν̃ c , ν̄˜ ) {mω̃1s , mω̃2s } {1.852, 4.700}
c
Heavy pseudoscalar neutrino (ν̃ c , ν̄˜ ) {mω̃ps , mω̃2p } {1.398, 2.586}
R-H neutrinos {mν c } 0.829
i
Heavy neutrinos (ν c , ν̄ c ) {mω1 , mω2 } {1.174, 2.070}
the decay of the moduli fields and the gravitino will produce ν̃Li with an abundance of the
right order [18,27].
6. Z decay modes and branching ratios
The Z gauge boson of our model has substantial coupling to the quarks. With its mass
in the range 2–4 TeV, it will be produced copiously at the LHC via the process pp → Z .
The reach of LHC is about 5 TeV for a Z with generic quark and lepton couplings [28].
Table 8
Z mass and Z–Z mixing angle in Model 2 for the same set
of input parameters as in Table 7
Z boson mass MZ 3.433 TeV
Z–Z mixing angle ξ 0.00068
Table 9
Eigenvectors of the neutralino mass matrix in Model 2. The unitary matrix O in Eq. (86) is the transpose of this
array
Fields χ̃10 χ̃20 χ̃30 χ̃40 χ̃50 χ̃60 χ̃70
B̃ −0.001 0.998 −0.052 0.023 0.000 0.000 0.000
W̃30 −0.997 0.002 0.053 −0.052 0.000 −0.001 0.000
H̃d0 −0.074 −0.052 −0.703 0.705 −0.002 0.011 0.001
H̃u0 0.000 −0.020 −0.707 −0.707 −0.001 −0.021 0.016
B̃ 0.000 0.000 0.006 −0.004 0.023 0.0563 0.826
S̃+ 0.000 0.000 0.011 0.018 −0.648 −0.620 0.441
S̃− 0.000 0.000 0.007 0.017 0.761 −0.546 0.350
Table 10
Eigenvectors of the chargino mass matrix in Model 2, where U , V are the unitary matrices that diagonalize the
chargino mass matrix (V ∗ M (c) U −1 = Mdiag )
(c)
U11 U12 U21 U22 V11 V12 V21 V22

0.994 0.105 −0.105 0.994 1.000 0.000 −0.000 1.000
Table 11
Eigenvectors of the CP-even Higgs boson mass matrix in Model 2. This array corresponds to X used in Eqs. (82)–
(84) and (109) of the text
Fields h h H H
Hd0 0.176 0.002 0.984 0.005
Hu0 0.984 0.010 −0.176 −0.025
S+ −0.012 −0.640 0.007 −0.768
S− −0.023 0.768 0.006 −0.640
Our model will then be directly tested at the LHC. Once produced, the Z will decay
into various channels. It is important to identify the dominant decay modes of the Z and
calculate the corresponding branching ratios. This is what we do in this section. We will
see that our Z is almost leptophobic, with Br(Z → e+ e− ) = (1–1.6)%. Direct limits on
such a Z are rather weak, however, the Z–Z mixing which occurs in our models at the
level of 0.001 does provide useful constraints.
We now turn to the dominant 2-body decays of Z . In this analysis we can safely ignore
the small Z–Z mixing for the most part.
The Lagrangian for Z coupling to the Standard Model fermions can be written as
L = gx f¯γ µ (vf − af γ5 )f Zµ . (56)
The Z decay rate into a fermion–antifermion pair is then

2 2
m2f
¯ gx2 m f m f
Γ (Z → f f ) = Cf MZ vf 1 + 2 2 + af 1 − 4 2

2 2
1−4 2 .
12π MZ MZ MZ
(57)
Here Cf = 3(1) for quarks (leptons), MZ is the Z
mass and gx is the U (1)x gauge
coupling. The vector and the axial-vector couplings (vf , af ) are related to the U (1)x
charges of the fermions as
1
vf = Q(fL ) + Q(fR ) , (58)
2
1
af = Q(fL ) − Q(fR ) . (59)
2
Here Q is the U (1)x charge of fL (listed in Table 1) and Q(fR ) = −Q(fLc ).
The decay width for Z → ν̄Li νLi and Z → ν̄ic νic are:
gx2 2
Γ (Z → ν̄Li νLi ) = Q MZ , (60)
24π νLi
m2ν c 3/2
gx2 2
Γ (Z → ν̄i νi ) =
c c
Q c MZ 1 − 4 2i
. (61)
24π νi MZ
There is mixing between the heavy vector-like ν c and the ν̄ c (cf. Eq. (37)), with the
mass eigenstates (ω1 , ω2 ) given by
c
ν cos θν c sin θν c ω1
= . (62)
ν̄ c − sin θν cos θν
c c ω2
Since Qν̄ c = −Qν c , the Lagrangian for the Z coupling to these neutrino is given by
gx
L = Qν c (cos 2θν c ω̄1 γ µ γ5 ω1 − cos 2θν c ω̄2 γ µ γ5 ω2 − sin 2θν c ω̄1 γ µ γ5 ω2
2
− sin 2θν c ω̄2 γ µ γ5 ω1 )Zµ . (63)
This leads to the decay rates

gx2 m2ω1 3/2
Γ (Z → ω1 ω1 ) = MZ Qν c cos 2θν c 1 − 4 2
2 2
, (64)
24π MZ

g2 m2ω 3/2
Γ (Z → ω2 ω2 ) = x MZ Q2ν c cos2 2θν c 1 − 4 22 , (65)
24π MZ
Γ (Z → ω1 ω2 )

gx2 (m2ω1 + m2ω2 ) (m2ω1 − m2ω2 )2 mω1 mω2
= MZ Qν c sin 2θν 1 −

2 2 c − −3
12π 2MZ2 2MZ4 MZ2

(mω1 + mω2 )2 (mω1 − mω2 )2
× 1− 1 − . (66)
MZ2 MZ2
Here mω1 (mω2 ) are the masses of the physical Majorana fermions.
The Z interaction with the sfermions is described by the Lagrangian
↔
L = igx (vf ± af )f˜L,R
∗
∂ µ f˜L,R Z .
µ
(67)
The rate for the decay Z to sfermions is given by
m2˜ 3/2
˜∗ ˜ gx2 fL,R
Γ (Z → fL,R fL,R ) = Cf MZ (vf ± af ) 1 − 4 2
2
, (68)
48π MZ
where the + (−) sign is for the left- (right-)handed sfermions and mf˜L,R is the left-
(right-)handed sfermion mass. vf and af are as given in Eqs. (58) and (59).
In the top squark sector, there is non-negligible mixing between the left and the right-
handed sfermions. This leads to the following modification of the Lagrangian:
↔ ↔ ↔
L = igx (vf ± af cos 2θf˜ )f˜1,2
∗
∂ µ f˜1,2 − af sin 2θf˜ (f˜1∗ ∂ µ f˜2 + f˜2∗ ∂ µ f˜1 ) Z µ ,
(69)
where θf˜ is the left–right sfermion mixing angle. The decay rate is given by
m2˜ 3/2
gx2
Γ (Z → f˜1,2
∗ ˜ f
f1,2 ) = Cf MZ (vf ± af cos 2θf˜ )2 1 − 4 21,2 , (70)
48π MZ
Γ (Z → f˜1∗ f˜2 )

gx2 (m2 + m2 ) (m2 − m2 )2 3/2
= Cf MZ (af sin 2θf˜ )2 1 + 2 1 2 2 + 1 4 2 . (71)
48π MZ MZ
c
The ν̃ c and ν̄˜ splits into two scalar and two pseudoscalar which mix (see Eqs. (48) and
(49)). The mass eigenstate ω̃is and ω˜ip are given as
c
ν̃s cos θωs sin θωs ω̃1s
c = , (72)
ν̄˜ s − sin θωs cos θωs ω̃2s
c
ν̃p cos θωp sin θωp ω̃1p
c = . (73)
ν̄˜ p − sin θωp cos θωp ω̃2p
The Lagrangian for the Z coupling to the scalar–pseudoscalar pair is given by:
↔
L = gx (Qν c cos θωs cos θωp + Qν̄ c sin θωs sin θωp )ω̃1s ∂ µ ω̃1p
↔
+ (Qν c sin θωs sin θωp + Qν̄ c cos θωs cos θωp )ω̃2s ∂ µ ω̃2p
↔
+ (Qν c cos θωs sin θωp − Qν̄ c sin θωs cos θωp )ω̃1s ∂ µ ω̃2p
↔
+ (Qν c sin θωs cos θωp − Qν̄ c cos θωs sin θωp )ω̃2s ∂ µ ω̃1p Z µ . (74)
This leads to the decay rate
(m2ωis + m2ωjp ) (m2ωis − m2ωjp )2 3/2
gx2 2
Γ (Z → ω̃is ω̃jp ) = Q 1−2 + , (75)
48π ij MZ2 MZ4
where Qij is identified with the appropriate coupling to ω̃is ω̃jp term in the Lagrangian of
Eq. (74).
The supersymetric partners of νic split into a scalar (ν̃is
c c
) and a pseudoscalar (ν̃ip ). The

decay of Z to these fields is similar to those analyzed in Eq. (75):
(m2ν̃ c + m2ν̃ c ) (m2ν̃ c − m2ν̃ c )2 3/2
g2
Γ Z → ν̃is ν̃ip = x Q2ν c 1 − 2 +
c c is ip is ip
, (76)
48π i MZ2 MZ4
where mν̃isc and mν̃ipc are the masses of the scalar and the pseudoscalar.
The Lagrangian for the Z coupling to the charged Higgs bosons is given by
↔
L = igx QHd sin2 β − QHu cos2 β H + ∂ µ H − Z µ
+ gx (QHd + QHu ) sin β cos βMW (Wµ+ H − + Wµ− H + )Z µ , (77)
where QHd (QHu ) is the U (1)x charge of Hd (Hu ) field. The decay rates of to Z H +H −
and W ± H ∓ are given by

+ − gx2 2 m2H ± 3/2
Γ (Z → H H ) = MZ QHd sin β − QHu cos β
2 2
1−4 2 , (78)
48π MZ
Γ (Z → W ± H ∓ )
2 − m2 ) 2 − m 2 )2
gx2 (5MW H±
(MW H±
= MZ (QHd + QHu )2 1 + 2 +
192π MZ2 MZ4

(M 2 + m2 ± ) (MW 2 − m 2 )2
× 1 − 2 W 2 H + 4
H±
. (79)
MZ MZ
Here mH ± is the mass of the H ± Higgs boson and MW is the mass of the W -boson.
The ZW + W − coupling of the Standard Model will induce, through Z–Z mixing, a
Z W + W − coupling. The decay of Z to a pair of W + W − is found to be [29]

Γ (Z → W + W − )
4 2 3/2
g22 M4 M2 MW MW
= cos2 θW sin2 ξ MZ Z4 1 + 20 W + 12 1 − 4 . (80)
192π MW MZ2 MZ4 MZ2
We now discuss the decays of Z → Zh, ZH, Zh , ZH as well as Z → hA, h A etc.
The relevant Lagrangian is

4
L = 2gx MZ (QHd cos βX1i − QHu sin βX2i )Z µ Zµ Hi
i=1

4
↔
− gx (QHd sin βX1i + QHu cos βX2i )Z µ Hi0 ∂ µ A
i=1

4
↔
− gx (QS+ cos ψX3i + QS− sin ψX4i )Z µ Hi0 ∂ µ A , (81)
i=1
where Hi0 (= h, h , H, H ) are the neutral CP-even Higgs bosons, mHi are the masses
of the corresponding Higgs boson, QS+ (QS− ) is the U (1)x charge of the S+ (S− ) field
and Xij are the matrix elements of the unitary matrix that diagonalizes the CP-even mass
matrix of Eq. (15). The decay rates are then
g2
Γ Z → ZHi0 = x MZ (QHd cos βX1i − QHu sin βX2i )2
48π

(5MZ2 − m2Hi ) (MZ2 − m2Hi )2
× 1+2 +
MZ2 MZ4

(MZ2 + m2Hi ) (MZ2 − m2Hi )2
× 1 − 2 + , (82)
MZ2 MZ4
gx2
Γ (Z → Hi A) = MZ (QHd sin βX1i + QHu cos βX2i )2
48π

(m2A + m2Hi ) (m2A − m2Hi )2 3/2
× 1−2 + , (83)
MZ2 MZ4
gx2
Γ (Z → Hi A ) = MZ (QS+ cos ψX3i + QS− sin ψX4i )2
48π

(m2A + m2Hi ) (m2A − m2Hi )2 3/2
× 1−2 + , (84)
MZ2 MZ4
where mA and mA are the pseudoscalar Higgs boson masses.
We parameterize the interactions between the neutralinos (χ̃10 , χ̃20 , . . . , χ̃70 ) and the Z
boson as

gij χ̃¯ i γ µ γ5 χ̃j0 Zµ .
0
L= (85)
i,j
Here the coupling gij is obtained from the eigenvectors of the neutralino mass matrix of
Eq. (27) as
 
0 0 0 0 0 0 0
0 0 0 0 0 0 0 
 
0 0 −x 0 0 0 0 
gx  2 
 T
ĝ = O  0 0 0 x
0 0 0 O , (86)
2  2 
0 0 0 0 0 0 0 
 
0 0 0 0 0 2 0 
0 0 0 0 0 0 −2
with gij = (ĝ)ij . Here O is the orthogonal matrix that diagonalizes the neutralino mass
matrix. The Z partial decay rates into neutralinos is found to be

gii2 m2i 3/2
Γ Z → χ̃i χ̃i =
0 0
MZ 1 − 4 2 , (87)
6π MZ
Table 12
Decay modes for Z in Model 1 for the parameters used in Table 2. The total decay width is Γ (Z → all) =
97.68 GeV
Decay modes of Z Width (GeV)
Z → {ūu, c̄c, t¯t} {4.75, 4.75, 4.64}
Z → d̄d(s̄s, b̄b) 9.59
Z → ēe(µ̄µ, τ̄ τ ) 1.13
Z → νeL νeL (νµL νµL , ντ L ντ L ) 0.65
Z → νeR νeR (νµR νµR , ντ R ντ R ) 4.19
Z → ω̄1 ω1 0.50
Z → {χ̃1 χ̃3 , χ̃1 χ̃4 , χ̃2 χ̃4 , χ̃3 χ̃4 , χ̃3 χ̃5 , χ̃4 χ̃5 , χ̃5 χ̃5 , χ̃5 χ̃6 } {0.01, 0.01, 0.01, 3.38, 0.01, 0.05, 3.34, 5.65}
Z → {χ̃2+ χ̃2− , χ̃1+ χ̃2− , χ̃1− χ̃2+ } {3.36, 0.02, 0.02}
Z → ũ∗R ũR (c̃R
∗ c̃ )
R 0.13
∗ ∗ ∗ t˜ }
Z → {t˜R t˜R , t˜L t˜R , t˜R {0.88, 0.13, 0.13}
L
∗ ẽ (µ̃∗ µ̃ , τ̃ ∗ τ̃ )
Z → ẽL 0.30
L L L L L

Z → ẽR ∗ ẽ (µ̃∗ µ̃ , τ̃ ∗ τ̃ ) 0.23
R R R R R
Z → ν̃eL
∗ ν̃ (ν̃ ∗ ν̃ , ν̃ ∗ ν̃ )
eL µL µL τ L τ L 2.52
Z → ν̃1s
c ν̃ c {ν̃ c ν̃ c , ν̃ c ν̃ c }
1p 2s 2p 3s 3p 1.94

Z → ω̃1s ω̃1p 0.36
Z → Zh 1.11
Z → {hA , H A, h A } {0.03, 0.47, 0.62}
Z → H +H − 0.46
Z → W + W − 1.08
Z → W ± H ∓ 0

(gij + gj i )2 (m2i + m2j ) (m2i − m2j )2 mi mj
Γ Z → χ̃i0 χ̃j0 = MZ 1 − − − 3
12π 2MZ2 2MZ4 MZ2

(mi + mj )2 (mi − mj )2
× 1− 1− (i = j ) (88)
MZ2 MZ2
where mi are the neutralino masses. (Here our result disagrees with Eq. (48) of Ref. [3] by
a factor of 2.)
The Lagrangian for the couplings of Z to the charginos is given by [3]
1 ¯± µ
2
L = gx χ̃ i γ (vij + aij γ5 )χ̃j± Zµ . (89)
2
i,j =1
The Z decay rate into the chargino pair is then

± ∓ gx2 2 (m2i + m2j ) (m2i − m2j )2
Γ Z → χ̃i χ̃j = MZ vij + aij 1 −

2
−
48π 2MZ2 2MZ4

mi mj
+ 3 vij2 − aij2
MZ2

(mi + mj )2 (mi − mj )2
× 1− 1− . (90)
MZ2 MZ2
Table 13
Decay modes for Z in Model 2 for the parameters used in Table 7. The total decay width is Γ (Z → all) =
229.93 GeV
Decay modes of Z Width (GeV)
Z → {ūu, c̄c, t¯t} {15.00, 15.00, 14.86}
Z → d̄d(s̄s, b̄b) 20.90
Z → ēe(µ̄µ, τ̄ τ ) 3.69
Z → νeL νeL (νµL νµL , ντ L ντ L ) 0.37
Z → νeR νeR (νµR νµR , ντ R ντ R ) 6.19
Z → {ω̄1 ω1 , ω̄1 ω2 } {1.41, 0.06}
Z → {χ̃1 χ̃3 , χ̃1 χ̃4 , χ̃2 χ̃4 , χ̃3 χ̃4 , χ̃3 χ̃5 , χ̃4 χ̃5 , χ̃5 χ̃5 , χ̃5 χ̃6 } {0.03, 0.03, 0.03, 10.99, 0.01, 0.04, 1.63, 6.64}
Z → {χ̃2+ χ̃2− } {10.96}
Z → ũ∗L ũL (c̃L
∗ c̃ )
L 0.02
∗
Z → ũR ũR (c̃R ∗ c̃ ) 3.80
R
∗ t˜ , t˜∗ t˜ , t˜∗ t˜ }
Z → {t˜R {5.93, 0.45, 0.45}
R L R R L
Z → d̃L
∗ d̃ (s̃ ∗ s̃ , b̃∗ b̃ )
L L L L L 0.02
∗ d̃ (s̃ ∗ s̃ , b̃∗ b̃ )
Z → d̃R 3.77
R R R R R
Z → ẽL ẽL (µ̃∗L µ̃L , τ̃L∗ τ̃L )
∗ 0.18
Z → ẽR
∗ ẽ (µ̃∗ µ̃ , τ̃ ∗ τ̃ )
R R R R R 1.54
∗ ν̃ (ν̃ ∗ ν̃ , ν̃ ∗ ν̃ )
Z → ν̃eL 4.54
eL µL µL τ L τ L

Z → ν̃1sc ν̃ c {ν̃ c ν̃ c , ν̃ c ν̃ c } 1.04
1p 2s 2p 3s 3p
Z → ω̃1s ω̃1p 0.91
Z → Zh 2.96
Z → {hA , H A, h A } {0.01, 2.38, 0.60}
Z → H +H − 2.38
Z → W + W − 2.81
Z → W ± H ∓ 0
Here mi is the chargino mass, vij and aij are given in terms of the charges QHu , QHd
and the matrices U and V which diagonalize the chargino mass matrix Eq. (29), can be
explicitly written as [3]
v11 = QHd V12

2
− QHu U12
2
, (91)
a11 = QHd V21
2
+ QHu U21
2
, (92)
v12 = v21 = QHd V12 V11 − δQHu U12 U11 , (93)
a12 = a21 = QHd V12 V11 + δQHu U12 U11 , (94)
v22 = QHd V11
2
− QHu U11
2
, (95)
a22 = QHd V22 + QHu U22 ,
2 2
(96)
where δ = sgn(mχ̃ ± ) × sgn(mχ̃ ± ).
1 2
In Table 12 we present the partial decay rates of Z to two fermions and to two scalars
in Model 1. The total width of Z is 106 GeV (this ignores three body decays, which
are more suppressed). One sees from Table 7 that the Z decays dominantly to q q̄ with
Br(Z → q q̄) 43.93%. On the other hand, Br(Z → e+ e− ) 1.16% in this case. Thus
this Z is leptophobic. We also see that Z → χ̃i0 χ̃j0 and Z → χ̃i± χ̃j∓ are significant.
There are also non-negligible decays into two Higgs particles, with Z → h A being
the dominant mode in this class. The decay of Z into sfermions is a new production
channel for supersymmetric particles. Decays into sneutrino pairs is the dominant mode
in this category, with Br(Z → ν̃L ν̃L ) 7.74%. The signature will be pp → Z →
ν̃Li ν̃Li → − − + + − +
i i χ̃1 χ̃1 , where the sneutrino decays into i χ̃1 , with the subsequent decay
χ̃1± → χ̃10 + π ± , etc.
In Table 13 we list the Z partial decay rates in Model 2. Br(Z → e+ e− ) 1.60% in
this case. Other features are very similar to the case of Model 1 (Table 7).
7. Other experimental signatures
In this section we discuss experimental signatures of the model other than Z decays.
7.1. Z decay and precision electroweak data
The Z–Z mixing angle and the direct coupling of Z to the Standard Model fermions
leads to modification of Z decays. Precision electroweak data from LEP and SLC can be
used to constrain such a Z in the mass range of a few TeV. Typically one finds the Z–Z
mixing angle ξ bounded to be less than a few ×10−3 [4], which is satisfied in our models.
The mixing of Z with Z shifts the mass of the Z boson from its SM value, while leaving
the W mass unaffected. This leads to a positive shift in the ρ parameter:

M2
ρ = ρSM 1 + ξ 2 Z2 . (97)
MZ
The partial decay width Γ (Z → f f¯) is modified to
αMZ
Γ (Z → f f¯) = 2
12 sin θW cos2 θW

× (gV cos ξ + κvf sin ξ )2 + (gA cos ξ + κaf sin ξ )2 , (98)
where
2gx sin θW cos θW
gV = T3 − 2q sin2 θW , gA = T3 , κ= , (99)
e
with q being the electric charge of the fermion. vf and va are given in Eqs. (58) and (59).
Partial widths of the Z will deviate from the Standard Model values owing to the shift
in the coupling of Z to fermions as well as due to a change in the derived value of sin2 θW .
We define
Γ (Z → f f¯)
∆f = − 1. (100)
Γ (Z → f f¯)SM
We use sin2 θW SM
= 0.23113 (the best fit in the Standard Model) for evaluating Γ (Z →
f f¯)SM . We do not perform a global fit to the available data, but we present a specific fit
which is at least as good as the Standard Model and perhaps slightly better. We choose to
set ∆ = 0, which yields sin2 θW = 0.230717 in Model 1. With this value of sin2 θW we
find
{∆u , ∆d , ∆ν } = {0.00100, 0.00171, 0.00206} (Model 1). (101)

This leads to the following modifications of decay widths:

Γhad = Γhad
SM
+ ∆d 2ΓdSM + ΓbSM + 2∆u ΓuSM = 1.74545 GeV, (102)
Γinv = (1 + ∆ν )Γinv
SM
= 502.793 MeV, (103)
Γhad
R = = 20.7744. (104)
Γ (Z → + − )
We see that Γhad is closer to the experimental value of 1.7444 GeV compared to the
Standard Model value of 1.7429 GeV. Similarly R is closer to the experimental value
(20.767 ± 0.025) than the Standard Model value (20.744). On the other hand, Γinv is
somewhat worse than the Standard Model fit (501.76 MeV) to be compared with the
experimental value of (499.0 ± 1.5 MeV). This deviation is still within acceptable range.
Here for our numerical fits we used the central values ΓdSM = 0.383185 GeV, ΓbSM =
0.375926 GeV and ΓcSM = ΓuSM = 0.300302 GeV, and Γhad SM = 1.7429 GeV [22].
The predicted value of MW is modified as

MZ2 1 − sin2 θW
MW = 1+ξ 2 SM
MW = 80.4427 GeV, (105)
MZ2 1 − sin2 θW SM
where MW SM
= 80.391 GeV is used. This value is closer to the direct measurement
MW = 80.446 than the Standard Model value.
In Model 2 we find, following the same procedure, sin2 θW = 0.230783, ∆d = 0.00131,
∆u = 0.00089, ∆ν = 0.00138 and Γhad = 1.74493 GeV, Γinv = 502.453 MeV, R =
20.7682, MW = 80.4356 GeV.
The radiative correction parameter in µ decay, ∆r, is slightly different in our model
compared to the Standard Model. In the on-shell scheme we have
2 sin2 θ
MW W (1 − ∆r)SM
= . (106)
2
(MW 2
sin θW )SM (1 − ∆r)
We obtain ∆r = 0.03501 (in Model 1) using the Standard Model value of ∆r = 0.0355 ±
0.0019. Clearly, such a shift is consistent with experimental constraints ((∆r)exp =
0.0347 ± 0.0011).
7.2. Z mass limit
The direct limit on the mass of Z with generic couplings to quarks and leptons is
MZ > 600 GeV. There is also a constraint on MZ from the process e+ e− → µ+ µ− . LEP
II has set severe constraints on lepton compositeness [22,30] from this process. We focus
on one such amplitude, involving all left-handed lepton fields. In our model, the effective
Lagrangian for this process is

x 2 1
Leff = −gx2 1 − 2
(e¯L γµ eL ) µ¯L γ µ µL . (107)
2 MZ
Comparing with Λ− LL (eeµµ) > 6.3 TeV [22], we obtain MZ /gx (1 − x/2)2.51 TeV.
For gx = 0.41 (0.45) and x = 1.3 (1.6) this implies MZ 361 (226) GeV. For the choice
of parameters in Tables 2 and 7, the above constraint is easily satisfied.
7.3. h → h h decay
Since the neutral Higgs boson h is lighter than the Standard Model Higgs h, the decay
h → h h can proceed for part of the parameter space. The decay rate is given by

Ghh
2 2
1 − 4 mh ,
Γ (h → h h ) = (108)
8πmh m2h
where
(g12 + g22 ) 2
G2hh = √ (υd X11 − υu X21 ) X12 − X22
2
4 2

+ 2(υd X12 − υu X22 )(X11 X12 − X21 X22 )
g2
+ √x 2(4X31 X32 − 4X41 X42 − xX11 X12 + xX21X22 )
4 2
× (−xυd X12 + xυu X22 − 4yX42 + 4zX32 )
2
− 4X32 − 4X42
2
− xX122
+ xX22
2

× (xυd X11 − xυu X21 + 4yX41 − 4zX31) . (109)
Here X is the unitary matrix that diagonalizes the CP-even Higgs mass matrix of Eq. (15).
In principle this can compete with the dominant decay h → bb̄. However, we find that in
Model 1 of Table 2 the decay is kinematically suppressed, while in Model 2 of Table 7
due to the small admixture of h in S+ , S− , this decay is suppressed: Γ (h → h h ) =
1.48 × 10−7 GeV (see Table 11). It is worth noting that if the mixings are as large as
in Table 6 and if the decay is kinematically allowed, then Γ (h → h h ) ∼ 0.1 MeV is
possible. Once produced, the dominant decays of h will be h → b b̄ and h → cc̄ with
comparable partial widths, as can be seen from Hu0 and Hd0 components in h (see Table 6).
7.4. Signatures of SUSY particles
The supersymmetric particles, once produced in pp (pp̄) collisions, will decay into
the LSP. The LSP is χ̃10 (the neutral Wino) in Model 1 while it is the scalar neutrino
ν̃L in Model 2. In Model 1, χ̃10 is nearly mass degenerate with the lightest chargino
χ̃1± , with a mass splitting of about 180 MeV. The decay χ̃10 → π ± χ1∓ will then occur
within the detector. At the Tevatron Run 2 as well as at the LHC, the process pp̄ (or pp)
→ χ̃10 + χ̃1± will produce these SUSY particles. Naturalness suggest that mχ̃ 0 , mχ̃ ±
1 1
300 GeV (corresponding to mgluino 2 TeV). Strategies for detecting such a quasi-
degenerate pair has been carried out in Refs. [31,32]. In the case where the LSP is the
left-handed sneutrino, the decay χ̃1± → ± ν̃L will be allowed. In this case χ̃10 will decay
dominantly to χ̃10 → ν̃L νL .
8. Conclusions
We have suggested in this paper a new class of supersymmetric Z models motivated

by the anomaly mediated supersymmetry breaking framework. The associated U (1)
symmetry is U (1)x = xY − (B − L), where Y is the Standard Model hypercharge. For
1 < x < 2, the charges of the lepton doublets and the lepton singlets have the same sign.
This implies that the U (1)x D-term can induce positive masses for both the doublet and the
singlet sleptons and can cure the tachyonic problem of AMSB. We have shown explicitly
that this is indeed possible in this class of models. In achieving this, the parameters of the
model get essentially fixed. We have found that MZ = 2–4 TeV and the Z–Z mixing angle
ξ 0.001. The phenomenologically viable Z turns out to be leptophobic—with Br(Z →
+ − ) (1–1.6)%. The dominant decay of Z is to q q̄ pair with Br(Z → q q̄) 44%.
Decays into supersymmetric particles and Higgs bosons are also significant.
In Tables 2 and 7 we present our spectrum for two models, Model 1 (with x = 1.3) and
Model 2 (with x = 1.6). The lightest SUSY particle is the neutral Wino (Model 1) or the
sneutrino (Model 2). The partial decay widths of Z are listed in Tables 12 and 13. These
models are compatible with precision electroweak data, with the Z models giving slightly
better fits to the data than the Standard Model. This Z should be within reach of LHC.
The correlations between the Z decays and the supersymmetric spectrum should make
this class of models distinguishable from other Z models.
Acknowledgements
This work is supported in part by DOE Grant No. DE-FG03-98ER-41076 and an award
from the Research Corporation. The work of I.G. was supported in part by the National
Science Foundation under Grant PHY00-98791.
Appendix A
In this appendix we give the one-loop anomalous dimension, beta-function and the soft
SUSY breaking masses for the various fields in our model.
A.1. Anomalous dimensions
The one-loop anomalous dimensions for the fields in our model are:

j 3 2 3 2 x 2 2
2
16π γLij = (Yl Yl† )j i − δi g + g +2 1− gx , (A.1)
10 1 2 2 2

j 6 2
16π 2 γeijc = 2(Yl† Yl )ij − δi g + 2(−1 + x)2 gx2 , (A.2)
5 1

j 1 2 3 2 8 2 x 1 2 2
16π 2 γQij = (Yd Yd† )j i + (Yu Yu† )j i − δi g1 + g2 + g3 + 2 − gx ,
30 2 3 6 3
(A.3)
2
j 8 8 2 1
16π 2 γUij = 2(Yu† Yu )ij − δi g2 + g2 + 2 x + gx2 , (A.4)
15 1 3 3 3 3

† j 2 2 8 2 x 1 2 2
16π γDij = 2(Yd Yd )ij − δi
2
g + g +2 + gx , (A.5)
15 1 3 3 3 3

3 3 x 2 2
16π 2 γHd = 3Yd23 + Yl23 − g12 − g22 − 2 − gx , (A.6)
10 2 2

3 2 3 2 x 2 2
16π γHu = 3Yu3 − g1 − g2 − 2 −
2 2
gx , (A.7)
10 2 2
16π 2 γνic = 4fν2c − 2gx2 , (A.8)
i
16π 2 γν c = 4fν2c − 2gx2 , (A.9)

16π γν̄ c = 4h − 2gx2 ,
2 2
(A.10)
3
16π γS+ = 2
2
fν2c + 2fν2c − 8gx2 , (A.11)
i
i=1
16π 2 γS− = 2h2 − 8gx2 . (A.12)
A.2. Beta function
The beta functions for the Yukawa couplings appearing in the superpotential, Eq. (4),
are:

Yd3 7 2 16 2 (4 + 2x + 7x 2) 2
β(Yd3 ) = 6Y 2
d3 + Y 2
u3 + Y 2
l3 − g − 3g 2
− g − g x ,
16π 2 15 1 2
3 3 9
(A.13)
2)
Yu3 13 16 (4 − 10x + 13x
β(Yu3 ) = 6Yu23 + Yd23 − g12 − 3g22 − g32 − gx2 ,
16π 2 15 3 9
(A.14)
Yl3 9 2
β(Yl3 ) = 4Yl3 + 3Yd3 − g1 − 3g2 − 4 − 6x + 3x gx ,
2 2 2 2 2
(A.15)
16π 2 5
fνe
β(fνe ) = 2
10fν2e + 2fν2µ + 2fν2τ + 2fν2c − 12gx2 , (A.16)
16π
fνµ
β(fνµ ) = 2
10fν2µ + 2fν2e + 2fν2τ + 2fν2c − 12gx2 , (A.17)
16π
fντ
β(fντ ) = 2
10fν2τ + 2fν2µ + 2fν2e + 2fν2c − 12gx2 , (A.18)
16π
fν4
β(fν c ) = 2
10fν2c + 2fν2µ + 2fν2τ + 2fν2e − 12gx2 , (A.19)
16π
h
β(h) = 10h − 12g 2
x . (A.20)
16π 2
The gauge beta function of our model are
gi3
β(gi ) = bi , (A.21)
16π 2
where bi = (33/5, 1, −3, (11x 2 − 16x + 26)) for i = 1, 2, 3, x.
A.3. A terms
The trilinear soft SUSY breaking terms are given by

β(Y )
AY = − Maux , (A.22)
Y
where Y = (Yui , Ydi , Yli , fνic , fν c , h).
A.4. Gaugino masses
The soft masses of the gauginos are given by:

β(gi )
Mi = Maux , (A.23)
gi
where i = 1, 2, 3, x, corresponding to the gauge groups U (1)Y , SU(2)W , SU(3)C , U (1)x
with β(gi ) given as in Eq. (A.21) with Mx = M1 .
A.5. Soft SUSY masses
The soft masses of the squarks and the sleptons are given in the text. For the Hu , Hd ,
ν c , S+ , S− fields they are:
2 Hu Maux 2 3 3
2
x

m̃soft H = 3Yu3 β(Yu3 ) − g1 β(g1 ) − g2 β(g2 ) − 2 gx β(gx ) ,
u 16π 2 10 2 2
(A.24)
2 Hd Maux 2 3 3
m̃soft H = 3Yd3 β(Yd3 ) + Yl3 β(Yl3 ) − g1 β(g1 ) − g2 β(g2 )
d 16π 2 10 2
2
x
−2 − gx β(gx ) , (A.25)
2
2 S+ Maux 2 3
m̃soft S = 2 f c
νi β(f c
νi ) + 2fν c β(fν c ) − 8gx β(gx ) , (A.26)
+ 16π 2
i=1
S−2
Maux
m̃2soft S−
=
2hβ(h) − 8g x β(gx ) , (A.27)
16π 2
2 νic Maux 2
m̃soft ν c = 2
4fνic β(fνic ) − 2gx β(gx ) , (A.28)
i 16π
2 ν c Maux 2
m̃soft ν c = 2
4fν c β(fν c ) − 2gx β(gx ) , (A.29)
16π
2 ν̄ c Maux 2
m̃soft ν̄ c = 2
4hβ(h) − 2gx β(gx ) . (A.30)
16π
References
[1] For a review see, J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193.
[2] M. Cvetič, P. Langacker, hep-ph/9707451.
[3] T. Gherghetta, T.A. Kaeding, G.L. Kane, Phys. Rev. D 57 (1998) 3178, hep-ph/9701343.
[4] J. Erler, P. Langacker, T. Li, Phys. Rev. D 66 (2002) 15002, hep-ph/0205001;
J. Erler, P. Langacker, Phys. Rev. Lett. 84 (2000) 212, hep-ph/9910315.
[5] F. Del Aguila, Acta Phys. Pol. B 25 (1994) 1317, hep-ph/9404323.
[6] G. Altarelli, et al., Mod. Phys. Lett. A 5 (1990) 495.
[7] C. Burgess, et al., Phys. Rev. D 49 (1994) 6115.
[8] E. Ma, Phys. Rev. Lett. 89 (2002) 041801, hep-ph/0201083.
[9] S. Chaudhuri, S.-W. Chung, G. Hockney, J. Lykken, Nucl. Phys. B 456 (1995) 89, hep-ph/9501361;
G. Cleaver, M. Cvetič, J.R. Espinosa, L.L. Everett, P. Langacker, J. Wang, Phys. Rev. D 59 (1999) 055005,
hep-ph/9807479.
[10] A.E. Faraggi, M. Masip, Phys. Lett. B 388 (1996) 524, hep-ph/9604302.
[11] D.M. Ghilencea, L.E. Ibanez, N. Irges, JHEP 0208 (2002) 016, hep-ph/0205083.
[12] B. Holdom, Phys. Lett. B 166 (1986) 196;
B. Holdom, Phys. Lett. B 259 (1991) 329.
[13] K.S. Babu, C.F. Kolda, J. March-Russell, Phys. Rev. D 57 (1998) 6788, hep-ph/9710441.
[14] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155.
[15] G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
[16] K.S. Babu, C.F. Kolda, J. March-Russell, Phys. Rev. D 54 (1996) 4635, hep-ph/9603212.
[17] I. Jack, D.R.T. Jones, Phys. Lett. B 482 (2000) 167, hep-ph/0003081.
[18] A. Pomarol, R. Rattazzi, JHEP 9905 (1999) 013, hep-ph/9903448;
Z. Chacko, M.A. Luty, I. Maksymyk, E. Ponton, JHEP 0004 (2000) 001, hep-ph/9905390;
B.C. Allanach, A. Dedes, JHEP 0006 (2000) 017, hep-ph/0003222.
[19] I. Jack, D.R.T. Jones, Phys. Lett. B 473 (2000) 102, hep-ph/9911491;
I. Jack, D.R.T. Jones, R. Wild, Phys. Lett. B 535 (2002) 193, hep-ph/0202101;
N. Arkani-Hamed, H. Murayama, Y. Nomura, D.E. Kaplan, JHEP 0102 (2001) 041, hep-ph/0012103;
R. Harnik, H. Murayama, A. Pierce, JHEP 0208 (2002) 034, hep-ph/0204122;
B. Murakami, J.D. Wells, Phys. Rev. D 68 (2003) 035006, hep-ph/0302209;
A.E. Nelson, N.J. Weiner, hep-ph/0210288;
E. Katz, Y. Shadmi, Y. Shirman, JHEP 9908 (1999) 015, hep-ph/9906296.
[20] O.C. Anoka, K.S. Babu, I. Gogoladze, hep-ph/0312176.
[21] J.L. Feng, T. Moroi, L. Randall, M. Strassler, S. Fang Su, Phys. Rev. Lett. 83 (1999) 1731, hep-ph/9904250;
J.L. Feng, T. Moroi, Phys. Rev. D 61 (2000) 095004, hep-ph/9907319.
[22] Particle Data Group Collaboration, K. Hagiwara, et al., Phys. Rev. D 66 (2002) 1.
[23] M. Carena, J.R. Espinosa, M. Quiros, C.E.M. Wagner, Phys. Lett. B 355 (1995) 209, hep-ph/9504316.
[24] T. Gherghetta, G.F. Giudice, J.D. Wells, Nucl. Phys. B 559 (1999) 27, hep-ph/9904378;
J.F. Gunion, S. Mrenna, Phys. Rev. D 62 (2000) 015002, hep-ph/9906270.
[25] T. Moroi, L. Randall, Nucl. Phys. B 570 (2000) 455, hep-ph/9906527;

G.D. Kribs, Phys. Rev. D 62 (2000) 015008, hep-ph/9909376;
B. Murakami, J.D. Wells, Phys. Rev. D 64 (2001) 015001, hep-ph/0011082;
P. Ullio, JHEP 0106 (2001) 053, hep-ph/0105052.
[26] D.M. Pierce, J.A. Bagger, K. Matchev, R.J. Zhang, Nucl. Phys. B 491 (1997) 3, hep-ph/9606211.
[27] A. Masiero, S. Pascoli, Int. J. Mod. Phys. A 17 (2002) 1723.
[28] A. Djouadi, M. Dittmar, A. Nicollerat, hep-ph/0307020;
A. Djouadi, M. Dittmar, A. Nicollerat, CMS, Technical proposal, CERN/LHCC/94–38;
ATLAS Collaboration, ATLAS detector and physics performance technical design report, CERN/LHCC/99–
14.
[29] V. Barger, K. Whisnant, Phys. Rev. D 36 (1987) 3429;
M.M. Boyce, M.A. Doncheski, H. Konig, Phys. Rev. D 55 (1997) 68, hep-ph/9607376.
[30] E.J. Eichten, K.D. Lane, M.E. Peskin, Phys. Rev. Lett. 50 (1983) 811.
[31] F. Paige, J.D. Wells, hep-ph/0001249.
[32] D.K. Ghosh, P. Roy, S. Roy, JHEP 0008 (2000) 031, hep-ph/0004127;
H. Baer, J.K. Mizukoshi, X. Tata, Phys. Lett. B 488 (2000) 367, hep-ph/0007073.
On deviations from bimaximal neutrino mixing

P.H. Frampton a , S.T. Petcov b,c,1 , W. Rodejohann b,c
a Department of Physics and Astronomy, University of North Carolina, Chapel Hill, NC 27599-3255, USA
b Scuola Internazionale Superiore di Studi Avanzati, Via Beirut 2–4, I-34014 Trieste, Italy
c Istituto Nazionale di Fisica Nucleare, Sezione di Trieste, I-34014 Trieste, Italy
Received 2 February 2004; accepted 16 March 2004
Abstract
The PMNS neutrino mixing matrix UPMNS is in general a product of two unitary matrices
Ulep and Uν arising from the diagonalization of the charged lepton and neutrino mass matrices,
†
UPMNS = Ulep Uν . Assuming that Uν is a bimaximal mixing matrix, we investigate the possible
forms of Ulep . We identify three possible generic structures of Ulep , which are compatible with the
existing data on neutrino mixing. One corresponds to a hierarchical “CKM-like” matrix. In this case
relatively large values of the solar neutrino mixing angle θsol , and of |Ue3 |2 ≡ |(UPMNS )e3 |2 , are
typically predicted, tan2 θsol 0.42, |Ue3 |2 0.02, while the atmospheric neutrino mixing angle
θatm can deviate noticeably from π/4, sin2 2θatm 0.95. The second corresponds to one of the
mixing angles in Ulep being equal to π/2, and predicts practically maximal atmospheric neutrino
mixing sin2 2θatm 1. Large atmospheric neutrino mixing, sin2 2θatm 0.95, is naturally predicted
by the third possible generic structure of Ulep , which corresponds to all three mixing angles in Ulep
being large. We focus especially on the case of CP-non-conservation, analyzing it in detail. We show
how the CP-violating phases, arising from the diagonalization of the neutrino and charged lepton
mass matrices, contribute to the measured neutrino mixing observables.
1. Introduction
After the spectacular results obtained in the experimental studies of neutrino oscillations
in the last two and a half years or so (see, e.g., [1,2] for a summary), understanding the
E-mail address: werner@sissa.it (W. Rodejohann).

1 Also at: Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 1784 Sofia,
Bulgaria.
doi:10.1016/j.nuclphysb.2004.03.014
32 P.H. Frampton et al. / Nuclear Physics B 687 (2004) 31–54
detailed structure and the origin of neutrino masses and mixing is of prime importance.
Progress in understanding the origin of the Pontecorvo–Maki–Nakagawa–Sakata (PMNS)
mixing matrix in the weak charged lepton current [3] can lead to a complete solution of
the fundamental problem regarding the structure of the neutrino mass spectrum, which can
be with normal or inverted hierarchy, or of quasi-degenerate type. It can also help gain
significant insight on, or even answer the fundamental question of, CP-violation in the
lepton sector.
The fact that, according to the existing data, the atmospheric neutrino mixing angle is
maximal, or close to maximal [4], θatm ∼ π/4, the solar neutrino mixing angle is relatively
large [2,5], θ ∼ π/5.4 (sin2 θ ∼ = 0.30), and that the mixing angle θ limited by the
CHOOZ and Palo Verde experiments [6,7] is small, sin2 θ < 0.074 (99.73% C.L.) [2],
suggests that the PMNS neutrino mixing matrix,2 UPMNS , can originate from a 3×3 unitary
mixing matrix having a bimaximal mixing form, Ubimax . In the case of CP-invariance in
the lepton sector one has
 √1 √1

0
2 2
 1 
Ubimax =  −2 √1  .
1
 2 2  (1)
1
2 −2 1 √1
2
The assumption of exact equality UPMNS = Ubimax implies θ = π/4, which is ruled out at
more than 5 s.d. by the existing solar neutrino data [2,5]. However, the deviation of θ from
π/4 can be described by a relatively small mixing parameter [8]: λ = sin(π/4 − π/5.4) ∼ =
0.20. Thus, the neutrino mixing matrix can have the form
UPMNS = Uλ† Ubimax . (2)
It is natural to suppose that Uλ† and Ubimax in Eq. (2) arise from the diagonalization of the
charged lepton and neutrino mass matrices, respectively.
The existing data show also that [2,4] m2 |m2A |, where m2 > 0 and m2A
are the neutrino mass squared differences driving the solar and atmospheric neutrino
oscillations. This fact and the preceding considerations suggest further that Ubimax can
arise from the diagonalization of a neutrino mass term of Majorana type having a specific
symmetry. In the absence of charged lepton mixing (Uλ = 1) this symmetry could
correspond to the conservation of the non-standard lepton charge [9] L = Le − Lµ − Lτ ,
where Ll , l = e, µ, τ , are the electron, muon and tauon lepton charges. The indicated
symmetry cannot be exact: it has to be broken mildly by the neutrino mass term in order
to ensure that m2 = 0, and by the charged lepton mass term (Uλ = 1), and/or by the
neutrino mass term, in order to guarantee that θ = π/4.
The possibility expressed by Eq. (2) and described above has been discussed in the
context of grand unified theories by many authors (see, e.g., [10]). It has also been
investigated phenomenologically first a long time ago (in a different context) in [9]
and more recently in [11–13]. Assuming that Eq. (2) holds, we perform in the present
article a systematic study of the possible forms of the matrix Uλ† in Eq. (2), which are
2 Throughout this article the case of three flavour neutrino mixing is considered.
P.H. Frampton et al. / Nuclear Physics B 687 (2004) 31–54 33
compatible with the existing data on neutrino mixing and oscillations. The case of CP-
non-conservation is of primary interest and is analyzed in detail. We show, in particular,
how the CP-violating phases, arising from the diagonalization of the neutrino and charged
lepton mass matrices, can influence the form of Uλ and can contribute in the measured
CP-conserving and CP-violating observables, related to neutrino mixing. The analysis
presented here can be considered as a continuation of the earlier studies quoted, e.g., in
[10–13]. However, it overlaps little with them.
2. The deviations from bimaximal mixing
We will employ in what follows the standard parametrization of the PMNS matrix:
 
c12 c13 s12 c13 s13
UPMNS =  −s12 c23 − c12 s23 s13 eiδ c12 c23 − s12 s23 s13 eiδ s23 c13 eiδ 
s12 s23 − c12 c23 s13 eiδ −c12s23 − s12 c23 s13 eiδ c23 c13 eiδ

× diag 1, eiα , eiβ , (3)
where we have used the usual notations cij = cos θij , sij = sin θij , δ is the Dirac CP-
violation phase, α and β are two possible Majorana CP-violation phases [14,15]. If we
identify the two independent neutrino mass squared differences in this case, m221 and
m231 , with the neutrino mass squared differences which induce the solar and atmospheric
neutrino oscillations, m2 = m221 > 0, m2A = m231 , one has: θ12 = θsol , θ23 = θatm ,
and θ13 = θ . The ranges of values of the three neutrino mixing angles, which are allowed
at 1 s.d. (3 s.d.) by the current solar and atmospheric neutrino data and by the data from
the reactor antineutrino experiments CHOOZ and KamLAND, read [2,16]:
0.35 (0.27) tan2 θsol ≡ tan2 θ12 0.52 (0.72),
|Ue3 |2 = sin2 θ13 < 0.029 (0.074),
sin2 2θatm ≡ sin2 2θ23 0.95 (0.85). (4)
As is well known, the oscillations between flavour neutrinos depend on the Dirac phase δ,
but are insensitive to the Majorana CP-violating phases α and β [14,17]. Information about
these phases can be obtained, in principle, in neutrinoless double beta decay ((ββ)0ν -
decay) experiments [18–20].
Let us define the matrix Uλ through Eq. (2), UPMNS ≡ Uλ† Ubimax , where the matrix
Ubimax is given by Eq. (1). The latter would have coincided with the PMNS matrix if the
neutrino mixing were exactly bimaximal. The bimaximal neutrino mixing can be obtained,
e.g., by exploiting the flavor neutrino symmetry corresponding to the conservation of the
non-standard lepton charge L = (Le − Lµ − Lτ ) [9] (see also [21,22]). The deviations
of the neutrino mixing from the exact bimaximal mixing form can be described, as was
shown in [8], by a real parameter λ ∼ 0.2, which was introduced in the following “flexible”
parametrization of three elements of UPMNS :

1 1
Ue2 = (1 − λ), Ue3 = Aλ ,n
Uµ3 = 1 − Bλm eiδ , (5)
2 2
where A and B are real parameters of order one. The integer numbers m and n can be
chosen according to the improved limits on, or precise values of, sin2 θ13 and sin2 2θ23 . The
fact that solar neutrino mixing is now confirmed (at more than 5 s.d.) to be non-maximal
means that λ = 0 and the best-fit value of tan2 θsol = 0.43 [2] corresponds to λ 0.23.
Taking the purely phenomenological parametrization (5) at face value and expressing
the PMNS matrix in terms of λ, A and B, one can solve Eq. (2) for Uλ in order to obtain the
physical ranges of values of the parameters involved. It was found in [8] that the structure
of Uλ can at leading order be the unit matrix plus corrections of order λ.
In the present article we perform a systematic study of the possible forms of Uλ allowed
by the current data. The neutrino mixing phenomenology can be described, for example,
by Uλ having a “CKM-like” structure, as has been discussed in [8,12,13]. Here we show,
in particular, that the restriction to a hierarchical form for Uλ is not necessary. Counter-
intuitive forms of Uλ , and/or the presence of CP-violating phases, can produce naturally
the observed deviation from π/4 of θsol as well as the requisite smallness of |Ue3 |.
3. The case of CP-conservation
In this case Uλ is an orthogonal matrix. We will use the standard parametrization (see
, θ , and θ , and define sin θ ≡ λ .
Eq. (3)) for Uλ . Let us denote the angles in Uλ as θ12 23 13 ij ij
We shall treat λ12 , λ23 , and λ13 as free parameters without assuming any hierarchy relation
between them. In the case of CP-conservation under discussion one can limit the analysis
to the case 0 θij π , and correspondingly to 0 λij 1. The results for λij < 0 can be
obtained formally from those derived for λij > 0 by making the change (i) λ23 → −λ23 ,
and/or (ii) λ13 → −λ13 as long as one keeps λ12 > 0. The change λ12 → −λ12 should be
done simultaneously with the change λ13 → −λ13 and in this case only the “solutions”
with |λ13 | > |λ12 | should be considered. In the latter case both signs of λ23 are possible.
Using Eqs. (2) and (1), we have determined the regions of values of λ12 , λ23 and λ13
for which the values of the neutrino mixing angles θ12 = θsol , θ23 = θatm , and θ13 = θ ,
lying within their 1σ and 3σ ranges, Eq. (4), can be reproduced. The results are shown
graphically in Fig. 1. One can clearly identify three different cases:
(i) all λij 0.35 (“small”);

(ii) λ23 = +1 and λ12,13 0.35 (“small”);
(iii) all λij 0.40 (“large”).
We shall discuss next these three cases in detail. The resulting structures of Uλ will turn
out to be rather different in the three cases.
3.1. Small λij
Multiplying Uλ† by Ubimax and expressing tan2 θsol , |Ue3 |2 and sin2 2θatm in terms of the
three small parameters λij , one finds:
Fig. 1. Scatter plot of the three λ parameters for the 3σ (left column) and 1σ (right column) allowed ranges of
values of the neutrino mixing parameters given in Eq. (4). Conservation of CP is assumed.
|Ue1 |2
tan2 θsol =
|Ue2 |2
√ √
1 − 2 2(λ12 − λ13 ) + 4(λ12 − λ13 )2 − 2 2λ23 (λ12 + λ13 ) + O λ3 ,

λ12 + λ13 λ23 3
|Ue3 |
√ − √ (λ12 − λ13 ) + O λ
,
2 2
4|Uµ3 | |Uτ 3 |2
2
sin2 2θatm ≡ 1 − 4λ223 + O λ3 . (6)
(1 − |Ue3 | )
2 2
The key quantities are therefore λ23 and the sum and the difference of λ12 and λ13 . To
leading order in the λ-parameters, the deviation from maximal solar neutrino mixing and
zero Ue3 are proportional to λ, while the atmospheric neutrino mixing is close to maximal
with only quadratic corrections to (1 − sin2 2θatm). The requirement of sin2 2θatm 0.85
(0.95) leads to order λ2 to the restriction λ23 0.19 (0.11).
Assuming that the oscillation parameters lie within their 1σ allowed ranges and
neglecting terms of order λ2 , one finds from Eq. (6) that the term (λ12 − λ13 ) has to be
positive and rather large (∼ 0.2) in order to ensure the deviation of θsol from π/4. At
the same time (λ12 + λ13 ) has to be smaller than ∼ 0.2 in order to satisfy the limit on
|Ue3 |. These two conditions imply (for λ12,13 > 0) a hierarchy of the form λ12
λ13 . The
remaining parameter is λ23 (λ12 + λ13 )/2 ∼ λ12 /2.
In Fig. 2 we show scatter plots of the values of λij , obtained by requiring that the
corresponding values of neutrino mixing parameters tan2 θsol , sin2 2θatm and |Ue3 |2 lie
within their 1σ allowed ranges, Eq. (4). One can see from the left panels in Fig. 2
that λ12 (0.21–0.26), λ23 0.10 and λ13 0.03. Thus, one has λ23 λ12 /2 and
λ13 λ212 /2. In Fig. 2 we also display the resulting correlations between the neutrino
mixing parameters for which the 1σ allowed ranges were used. From the panels in the
right column one sees that tan2 θsol 0.42 and |Ue3 |2 0.017. Most of the points tend to
lie at rather large values of tan2 θsol and |Ue3 |2 . Taking the 3σ range in Eq. (4) leads to the
lower limits of tan2 θsol 0.35 and of |Ue3 |2 0.003.
We consider next two representative (and to a certain degree typical) forms of Uλ . A
viable possibility is the existence of “hierarchical” relations between λ12 , λ23 and λ13 . One
can have, for instance, λ23 ∼ λ212 and λ13 ∼ λ312 , with λ12 ≡ λ and, e.g., λ ∼ (0.20–0.30).
This implies the following form of Uλ :
 
1 − λ2 /2 λ λ3

Uλ ∼  −λ 1 − λ2 /2 λ2  + O λ4 . (7)
λ3 −λ 2 1
The structure of Uλ is close to that of a diagonal matrix, and is similar to the structure of
the CKM matrix. Given the hierarchy in the charged lepton masses, the CKM-like form of
Uλ , Eq. (7), is rather natural. One has in this case:
√ √
tan2 θsol 1 − 2 2λ + 4λ2 − 2 2λ3 ,
λ
|Ue3 | √ ,
2
sin2 2θatm 1 − 4λ4 (8)
plus terms of higher order in λ. Consequently, this scenario “predicts” solar neutrino
mixing and |Ue3 |2 close to their currently allowed maximal values, and atmospheric
neutrino mixing close to maximal.
More specifically, the hierarchy of the λ-parameters we are considering, λ12
λ23,13 ,
leads, as it follows from Eqs. (6) and (8), to the interesting correlation:
tan2 θsol 1 − 4|Ue3 | + 8|Ue3 |2 − 8|Ue3 |3 . (9)

Fig. 2. Scatter plot of the λ and neutrino oscillation parameters in the case of “small” λij for the 1σ allowed
ranges of values of the neutrino mixing parameters given in Eq. (4). Conservation of CP is assumed.
The upper limit of tan2 θsol 0.52 (0.72) implies in this case a lower limit on |Ue3 |2
0.027 (0.006).3 This explains qualitatively the lower limit on |Ue3 |2 as observed in Fig. 2
and mentioned earlier. Another correlation related to the CKM-like form of Uλ , Eq. (7),
sol implies a significant lower limit on |Ue3 | in the case of CKM-like form of
3 That the upper limit on tan2 θ 2
Uλ was noticed also in [12].
reads:
sin2 2θatm 1 − 16|Ue3 |4 . (10)
Thus, the deviations from maximal atmospheric neutrino mixing are determined in the case
of the hierarchical relations λ12 ≡ λ
λ23,13 , λ23 ∼ λ2 , by the magnitude of |Ue3 |4 . For,
e.g., |Ue3 |2 = 0.029 (0.05), we have sin2 2θatm 0.975 (0.96).
The alternative possibility is that of a mild “hierarchy” between λ12 and λ23 . We can
have, for instance, λ23 λ12 /2, λ12 ≡ λ. Taking also λ13 λ3 λ12 , one finds:
√ √ √
tan2 θsol 1 − 2 2λ + (4 − 2 )λ2 − 2 2λ3 ,
λ
|Ue3 | √ ,
2
sin 2θatm 1 − λ2 1 − 2|Ue3 |2 .
2
(11)
The atmospheric neutrino mixing angle θatm can deviate more from π/4 than in the
hierarchical case: for |Ue3 |2 = 0.05 we have now sin2 2θatm 0.90. This provides a
possibility to distinguish the case of λ23 λ12 /2, λ12,23
λ13 , from the hierarchical one
λ13 ∼ λ312 < λ23 ∼ λ212 < λ12 considered above. For λ ∼ 0.24 one finds from Eq. (11)
tan2 θsol 0.43,
which is√quite encouraging since values of λ ∼ 0.24 could be interpreted
in terms of mµ /mτ 0.105/1.77 0.24.
3.2. The case of λ23 = 1
If λ23 = 1 and λ12,13 are “small”, λ12,13 0.35, we find:

√
tan2 θsol 1 − 2 2(λ12 + λ13 ) + 4(λ12 + λ13 )2 + O λ3 ,
λ12 − λ13
Ue3 √ + O λ3 ,
2
1 2
sin 2θatm 1 − λ212 + λ213 + λ12 λ13 λ212 − λ213 + O λ5 .
2
(12)
4
The correlations of λ12 and λ13 as well as of the mixing parameters for λ23 = 1 are
shown in Fig. 3. We used the 1σ allowed ranges of the oscillation parameters, Eq. (4).
As Fig. 3 shows, sin2 2θatm is practically 1 in this case, sin2 2θatm 0.9965. The lower
limit is reached for relatively large |Ue3 |2 0.025 and relatively small tan2 θsol 0.32.
However, such small deviations of sin2 2θatm from 1 are extremely difficult to measure.4
For λ23 = 1 and “small” λ12,13 , Uλ has the following structure:
 
1 − 12 (λ212 + λ213 ) λ12 λ13
 
Uλ  −λ13 −λ12 λ13 1 − 12 λ213  + O λ3 . (13)
λ12 −1 + 12 λ212 0
All three possibilities λ12
λ13 , λ13
λ12 and λ12 ∼ λ13 are allowed. In any of these
three cases the 23 submatrix of Uλ displays an antidiagonal structure, as seen in Eq. (13).
4 The requisite precision may be achieved in experiments at neutrino factories [23].

Fig. 3. Scatter plot of the two λ parameters and the oscillation parameters in the case of λ23 = 1 for the 1σ range
of the neutrino mixing parameters given in Eq. (4). Conservation of CP is assumed.
3.3. Large λij
It is also possible that all λij = sin θij are relatively large, λij 0.40. From Fig. 1 one
sees that in this case λ12 (0.92–0.99) (0.85–0.99), λ13 (0.57–0.77) (0.47–0.85), and
λ23 (0.57–0.92) (0.40–0.99), when the 1 s.d. (3 s.d.) ranges of the oscillation parameters
> 0, whereas cos θ
are used. It turns out that, cos θ23 12,13 can take both√signs.
We limit our discussion to the case of all λij being “large”: λij 1/ 2. We can use the
√
deviations of λ12 from 1, and the deviations of λ13,23 from 1/ 2, as small parameters to
get convenient expressions for tan2 θsol , |Ue3 |2 and sin2 2θatm :
1 1
λ12 ≡ 1 − 12
2
, λ13 ≡ √ 1 + 13
2
, λ23 ≡ √ 1 + 23
2
, (14)
2 2
2 0.08 (0.15), 2 0.30 (0.40) and 2 0.09 (0.20), where the 1 s.d. (3 s.d.)
where 12 23 13
ranges of the oscillation parameters were used. One obtains for sufficiently small ij :

tan2 θsol 1 − 412 + 812
2
+ O 3 ,

Ue3 232
− 12 + O 3 ,

sin2 2θatm 1 + O 4 . (15)
The terms of higher order in we do not give can be sizable. The scatter plots do not
show any new important correlation. In particular, sin2 2θatm 0.95 is possible also in the
case under discussion. Both 23 and 13 can be zero, whereas 12 cannot. No hierarchy is
implied. If 12
13,23 , we get |Ue3 | 12 (1 − 12
2 /4) + O( 4 ), and
tan2 θsol 1 − 4|Ue3 | + 8|Ue3 |2 − 16|Ue3|3 . (16)

√
For all λij “large”, λij 1/ 2, Uλ has the form:
 2 + 2
12 2 
12 √1 − √ 13 √1 + √13
2 2 2 2
 
 −1 2 + 2
12 −1 2 − 2 ) 
Uλ  √ − 12
√ + √ 23 + 12 + 12 (12
2 − 2 − 2 ) 1
+ 12 (23 13 
 2 2 2 2 13 23 2 
2 − 2
12 23 −1
√1 − √ + √ 12
2 − 12 + 12 (12
2 − 2 + 2 )
13 23
1
2 − 12 (23
2 + 2 )
13
2 2 2

+ O 3 . (17)
All entries in Uλ , except the 11 entry, are large.
Let us summarize the results obtained under the assumption of CP-conservation:
• If all λij are small, one typically has tan2 θsol 0.42 and |Ue3 |2 0.02. The matrix
Uλ has a “CKM-like” structure. One can have sin2 2θatm 0.95, i.e., sin2 2θatm can
deviate noticeably from 1.
• If λ23 = 1 and λ12,13 are small, one practically has sin2 2θatm 1, the possible
deviations from 1 being exceedingly small. Any deviation of sin2 2θatm from 1 would
imply rather large |Ue3 |2 and relatively small tan2 θsol . The 23 block of the matrix Uλ
has an antidiagonal (quasi-Dirac
√ like) form.
• If all λij are large, λij 1/ 2, the matrix Uλ displays the unusual structure given in
Eq. (17), with all elements, except the 11 element, being large. Also in this case we
can have sin2 2θatm 0.95.
3.4. Implied forms of the charged lepton mass matrix
The matrix Uλ arises from the diagonalization of the charged lepton mass matrix mlep .
Having the approximate form of Uλ and using the knowledge of the charged lepton masses
me , mµ and mτ , we can derive the structure of two different matrices related to mlep . The
first one is the matrix mlep m†lep , which is diagonalized by Uλ , i.e.,
diag2
mlep m†lep = Uλ mlep Uλ† , (18)
diag
where mlep is a diagonal mass matrix containing the charged lepton masses me , mµ
and mτ . The second matrix is
diag
Uλ mlep UλT . (19)
Under the assumption of mlep being symmetric, which is realized in some GUT theories,
the matrix in Eq. (19) coincides with the charged lepton mass matrix.
Consider first the matrix mlep m†lep . In the case of “small” λij , we find that
 2 
me + m2τ λ213 + m2µ λ212 m2τ λ13 λ23 + m2µ λ12 m2τ λ13 − m2µ λ12 λ13
 
mlep m†lep  · m2µ + m2τ λ223 m2τ λ23 − m2µ λ12 λ13 
· · m2τ
(20)
plus terms of order λ3 . Thus, mlep m†lep is close to a diagonal matrix. In the special case of
λ13 = λ23 0 we get:
 2 
me + m2µ λ2 m2µ λ 0
 
mlep m†lep  m2µ λ m2µ 0 . (21)
0 0 m2τ
The zero entries can be terms of order m2e . The matrix mlep m†lep has clearly a hierarchical
structure.
When λ23 = 1 and λ12,13 are “small”, one obtains:
 2 
me + m2τ λ213 + m2µ λ212 m2τ λ13 −m2µ λ12
 
mlep m†lep  m2τ λ13 m2τ m2µ λ12 λ13  . (22)
−m2µ λ12 m2µ λ12 λ13 m2µ
In contrast to the previous case, now the 22 element is equal to m2τ and the 33 element is
given by m2µ . The 12 block of mlep m†lep is hierarchical, while the 23 block has “inverted
hierarchy-like” form.
Finally, if all λij are “large”, we can expect that mlep m†lep is to a good approximation a
“democratic” matrix. Indeed, one finds:
 √ √ 
1 1/ 2 1/ 2
m 2 √
mlep m†lep τ  1/ 2 1/2 1/2  (23)
2 √
1/ 2 1/2 1/2
plus terms of order .
From the definition of the two characteristic mass matrices in Eqs. (18) and (19) it
diag
follows that the expressions for Uλ mlep UλT in the three cases we are discussing can be
obtained from the corresponding expressions of mlep m†lep given above by replacing m2l
with ml in the latter.
4. CP-non-conservation
4.1. General considerations
It is well known that if the neutrinos with definite mass are Majorana particles, the
PMNS matrix contains 6 physical parameters, three mixing angles, and three CP-violating
phases. The latter can be divided in one “Dirac phase”, δ, and two “Majorana phases”, α
and β. The Majorana phases appear only in amplitudes describing lepton number violating
processes in which the total lepton charge changes by two units. In general, as we have
already discussed, one has
†
UPMNS = Ulep Uν , (24)
where Ulep and Uν two 3 × 3 unitary matrices: Ulep arises from the diagonalization of the
charged lepton mass matrix, while Uν diagonalizes the neutrino Majorana mass term. Any
3 × 3 unitary matrix contains 3 moduli and 6 phases and can be written as [24]
U = ei P Ũ Q, (25)
where P ≡ and Q ≡
diag(1, eiφ1 , eiφ2 ) diag(1, eiρ1 , eiρ2 )
are diagonal phase matrices
having 2 phases each, and Ũ is a unitary “CKM-like” matrix containing 1 phase and
3 angles. The charged lepton Dirac mass term, mlep , is diagonalized by a bi-unitary
transformation:
diag
mlep = UL† mlep UR , (26)
diag
where UL,R are 3 × 3 unitary matrices and mlep is the diagonal matrix containing
the masses of the charged leptons. Casting UL,R in the form (25), i.e., UL,R =
eiL,R PL,R ŨL,R QL,R , we find
mlep = ei(R −L ) Q†L ŨL† PL† mlep PR ŨR QR .

diag
(27)
diag
The term PL† mlep PR contains only 2 relative phases, which can be associated with the
right-handed charged lepton fields. The three independent phases in ei(R −L ) Q†L can
be absorbed by a redefinition of the left-handed charged lepton fields. Therefore, Ulep is
effectively given by ŨL and contains three angles and one phase.
The neutrino mass matrix mν is diagonalized via
diag
mν = UνT mν Uν . (28)
The unitary matrix Uν can be written in the form (25). It is not possible to absorb phases
in the neutrino fields since the neutrino mass term is of Majorana type [14,15]. Thus,
† †
UPMNS = Ulep Uν = eiν Ũlep Pν Ũν Qν . (29)
The common phase ν has no physical meaning and we will ignore it. Consequently, in
the most general case, the elements of UPMNS given by Eq. (29) are expressed in terms of
six real parameters and six phases in Ũlep and Uν .5 Only six combinations of those—the
three angles and the three phases of UPMNS , are observable, in principle, at low energies.
Note that the two phases in Qν are “Majorana-like”, since they will not appear in the
probabilities describing the flavour neutrino oscillations [14,17]. Note also that if Ulep = 1,
the phases in the matrix Pν can be eliminated by a redefinition of the charged lepton fields.
5 Using a different representation of the unitary matrices U

lep and Uν , the authors of [13] came to the
conclusion that UPMNS is expressed in terms of six real parameters and seven phases of Ulep and Uν . Writing
the unitary matrices in the form of Eq. (25) allows to reduce the number of phases by one.
The requirement of bimaximality of Ũν implies that Ũν is real and given by Eq. (1). In
this case the three angles and the Dirac phase in the PMNS matrix UPMNS will depend in a
complicated manner on the three angles and the phase in Ũlep and on the two phases in Pν .
The two Majorana phases will depend in addition on the parameters in Qν .
It should be emphasized that the form of UPMNS given in Eq. (29) is the most general
one. A specific model in the framework of which the bimaximal Ũν is obtained, might
imply symmetries or textures, along the lines of, e.g., [31], in mν , which will reduce
the number of independent parameters in Uν . We will comment in Section 4.6 on such
possibilities.
In the scheme with three massive Majorana neutrinos under discussion there exist three
rephasing invariants related to the three CP-violating phases in UPMNS , δ, α and β [25–29].
The first is the standard Dirac one JCP [25], associated with the Dirac phase δ:
∗ ∗
JCP = Im{Ue1 Uµ2 Ue2 Uµ1 }. (30)
It determines the magnitude of CP-violation effects in neutrino oscillations [26]. Let us

note that if Ulep = 1 and Ũν is a real matrix, one has JCP = 0.
The two additional invariants, S1 and S2 , whose existence is related to the Majorana
nature of massive neutrinos, i.e., to the phases α and β, can be chosen as [27,29] (see also
[19]):6
∗ ∗
S1 = Im{Ue1 Ue3 }, S2 = Im{Ue2 Ue3 }. (31)
If S1 = 0 and/or S2 = 0, CP is not conserved due to the Majorana phases β and/or (α − β).

The effective Majorana mass in (ββ)0ν -decay, |m |, depends, in general, on S1 and S2
[19] and not on JCP . Let us note, however, even if S1,2 = 0 (which can take place if, e.g.,
|Ue3 | = 0), the two Majorana phases α and β can still be a source of CP-non-conservation
in the lepton sector provided Im{Ue1 Ue2 ∗ } = 0 and Im{U U ∗ } = 0 [29,30].
µ2 µ3
Let us denote the phase in Ũlep by ψ. We will include it in Ũlep in the same way this
is done for the phase δ in Eq. (3). We will also use Pν = diag(1, eiφ , eiω ) and Qν ≡
diag(1, eiρ , eiσ ). This means that the Dirac phase δ, which has observable consequences
in neutrino oscillation experiments, is determined only by the phases ψ, φ and ω. The
Majorana phases in UPMNS , α and β, receive contributions also from the two remaining
phases ρ and σ . Allowing the phases δ, α, and β to vary between 0 and 2π , permits to
constrain (without loss of generality) the mixing angles θij to lie between 0 and π/2. In
Fig. 4 we show the result of a random search for the allowed regions of the values of the
parameters λij . The regions present in Fig. 1 are again allowed. Clustering of points around
these regions is also observed. The effect of the phases is that values of the parameters,
located in areas between the regions corresponding to the case of CP-conservation are
allowed. Let us consider the physical observables of interest in the three main regions of
the parameter space we have identified earlier.
6 We assume that the fields of massive Majorana neutrinos satisfy Majorana conditions which do not contain
phase factors.
Fig. 4. Scatter plot of the three λ parameters and the CP-violating phases for the 1σ allowed ranges of values of
the neutrino mixing parameters given in Eq. (4).
4.2. Small λij
Let us choose λ12 = λ, λ23 = Aλ and λ13 = Bλ with A, B real and of order one.7 In
this case tan2 θsol , Ue3 and sin2 2θatm are given by
7 Hierarchical λ correspond in this parametrization to sufficiently small A and/or B.

ij
√ √
tan2 θsol 1 − 2 2(cφ − Bcω−ψ )λ + 2 2(cφ − Bcω−ψ )2 − 2A(cω + Bcφ−ψ ) λ2 ,

1 (1 − B 2 )Acω−φ
Ue3 (1 + B + 2Bcω−φ−ψ ) λ −
2 2
λ ,
2 1 + B 2 + 2Bcω−φ−ψ
sin2 2θatm 1 − 4A2 cω−φ λ2 (32)
plus terms of order O(λ3 ). We introduced the obvious notation cφ = cos φ, sω−ψ =
sin(ω − ψ), etc. Note that the possibility of all three λij being of the same order was
not possible in case of conservation of CP. The rephasing invariant JCP , which controls the
magnitude of CP-violating effects in neutrino oscillations, has the form:
λ
JCP √ sφ + Bsω−ψ + λ(sω−2φ + Bs2ω−φ−ψ )A + O λ3 . (33)
4 2
Thus, not only the phase ψ of Ũlep , but also the phases ω and φ from Pν contribute to JCP .
Using Eqs. (32) and (33) it is not difficult to convince oneself that the magnitude of JCP is
controlled by the magnitude of |Ue3 |. Indeed, if, e.g., cω−φ−ψ = −1 and B√= 1, so that the
term ∼ λ in the expression for Ue3 vanishes and to leading order Ue3 − 2Acω−φ λ2 , the
term ∼ λ in JCP also vanishes and we have JCP ∼ Ue3 .
The invariants S1 and S2 which (for |Ue3 | = 0) determine the magnitude of the effects
of CP-violation associated with the Majorana nature of massive neutrinos, read:
λ
S1 (sφ+σ + Bsω−ψ+σ ),
2
λ
S2 (sφ−ρ+σ + Bsω−ψ−ρ+σ ), (34)
2
where we gave only the terms of order λ. We see that all five phases in Ũlep , Pν and Qν , and
not only the phases ρ and σ in Qν , contribute to S1 and S2 . In the leading order expressions
for S1,2 , Eq. (34), the five phases enter in three independent combinations. Moreover, in the
case of ρ = σ = 0, which is realized if the bimaximal mixing structure of Uν is associated
(in the limit of Ulep = 1) with the approximate conservation of L = Le − Lµ − Lτ (see
Section 4.6), we have to leading order in λ:
S1 S2
JCP √ √ . (35)
2 2 2 2
Thus, the magnitude of the CP-violating effects in neutrino oscillations is directly related in
this case to the magnitude of the CP-violating effects associated with the Majorana nature
of neutrinos. To leading order, both types of CP-violating effects are due to two phases, φ
and (ω − ψ). We will consider a model in which ρ = σ = 0 in Section 4.6.
In Fig. 5 we show the correlations between the observables tan2 θsol , |Ue3 |2 and
2
sin 2θatm , when a hierarchy between the three λ parameters of the form we have
considered in Section 3.2, λ12 ≡ λ, λ13 = λ3 and λ23 = λ2 , is assumed. The 1σ allowed
ranges of tan2 θsol , |Ue3 |2 and sin2 2θatm , were used to constrain λ and the CP-violating
phases. As it follows from Fig. 5, tan2 θsol can take values in the whole 1σ interval allowed
by the data, Eq. (4). This is in sharp contrast to the case of CP-conservation, in which the
correlation between the values |Ue3 | and the deviation of tan2 θsol from 1 leads, as Fig. 5
Fig. 5. Scatter plot of the oscillation parameters for the case of hierarchical λij , i.e., λ13 = λ12 λ23 = λ312 . The
various possible regions and correlations are shown both in the case of CP-violation (bright green areas in web
version) and CP-conservation (dark red areas in web version). The 1σ allowed ranges of values of the neutrino
mixing parameters given in Eq. (4) were used.
shows, to the lower limit tan2 θsol 0.48. We find also that if CP is not conserved, the
following lower bound holds: |Ue3 |2 0.018. There are interesting correlations between
tan2 θsol , |Ue3 |2 and sin2 2θatm . For instance, tan2 θsol 0.40 implies |Ue3 |2 0.026 and
sin2 2θatm 0.975.
4.3. The case of λ23 = 1
This case corresponds to “small” λ12,13 , λ12,13 0.35. Introducing λ12 = λ and λ13 =
Bλ, with B real, we find:
√
tan2 θsol 1 − 2 2(cω + Bcφ−ψ )λ + 4(cω + Bcω−ψ )2 λ2 ,

1
Ue3 λ (1 + B 2 − 2Bcω−φ+ψ ) + O λ3 ,
2
1 2
sin2 2θatm 1 − 1 − B 2 − 2Bcω−φ−ψ λ4 . (36)
4
The formulae for the mixing parameters can actually be obtained from the corresponding
formulae for all λij small, Eq. (32), by setting A = 0 and making the change B → −B,
φ ↔ ω. The rephasing invariant JCP reads:
λ
JCP √ (sω − Bsφ−ψ ) + O λ3 . (37)
4 2
The Majorana counterparts to JCP , S1 and S2 , are given by
λ λ
S1 − (sω+σ − Bsφ−ψ+σ ), S2 − (sω−ρ+σ − Bsφ−ψ−ρ+σ ). (38)
2 2
√
For ρ = σ = 0 one finds that to order λ the relation S1 = S2 = −2 2JCP is valid. Rel-
atively small ω is favored in this case. Furthermore, atmospheric neutrino mixing is pre-
dicted to be close to maximal, to be more precise, the lower limit sin2 2θatm 0.97 holds.
4.4. Large λij

√
We consider λij 1/ 2 and choose again as small expansion parameters the ij
introduced in Eq. (14). For the oscillation observables one finds
cos(ω + φ)/2 2 + 2cω+φ + cω+ψ − cφ+ψ 2
tan2 θsol 1 − 4 12 + 2 12 ,
cos(ω − φ)/2 cos2 (ω − φ)/2
(ω − φ) (ω − φ)
Ue3 sin + 12 cos ,
2 2

sin2 2θatm 1 + O 4 . (39)
The atmospheric neutrino mixing is again very close to maximal. The term ∼ sin(ω − φ)/2
in Ue3 is not suppressed by positive powers of ij . This suggests that (ω − φ)/2 should be
relatively small. For ω = φ the expressions for the three rephasing invariants, associated
with CP-non-conservation, take a rather simple form:
sφ sφ−ψ 2
JCP − 12 + , (40)
4 4 23
and
sψ−σ sσ 2 sψ+ρ−σ sρ−σ 2
S1 − √ 12 − √ 23 , S2 √ 12 − √ 23 . (41)
2 2 2 2
In the case of σ = ρ = 0 we have S1 −S2 , but there is no simple relation between S1,2
and JCP .
Note that in all three cases under study the connection between S1 , S2 and JCP is
different.
4.5. Special cases
It is instructive to examine the interplay of parameters when some of the three phases
in Ũlep and Pν are zero. It turns out that there are no drastic consequences in the cases of
“small” hierarchical λij , of λ23 = 1, and when all λij are “large”. Interesting correlations
happen, however, in the case of “small” λij when the three λij are of the same order. To
illustrate this we choose again λ12 = λ, λ23 = Aλ and λ13 = Bλ with A, B√ real and of
order one. We fix λ = mµ /mτ 0.24 and vary A, B between 0.4 and 1/ λ. In doing
so we observe that values of A and B slightly below 0.5 are favored, indicating a mild
hierarchy in λij . Inspecting Eq. (32) one finds that the dependence on ω of the oscillation
parameters is rather weak. If ω = ψ one finds from Eq. (32) that tan2 θsol will be too large,
tan2 θsol 0.58. We show in Fig. 6 the scatter plots of the correlations between tan2 θsol and
|Ue3 |2 , and between |Ue3 |2 and JCP , which turn out to be the most interesting ones. The 1σ
allowed ranges of values of the neutrino mixing parameters given in Eq. (4) were used. If
the charged lepton mass Lagrangian conserves CP, i.e., if ψ = 0, there exists a lower bound
on |Ue3 |2 0.005. In this case |JCP | can be relatively large, for instance, |JCP | 0.035 for
|Ue3 |2 = 0.025, but is also allowed to be zero.
If CP is conserved by the neutrino mass term, i.e., if φ = ω = 0, a simple correlation
between tan2 θsol and |Ue3 |2 exists again: smaller |Ue3 |2 implies smaller tan2 θsol (see
Fig. 6). A similar limit on |Ue3 |2 as in the case of ψ = 0 applies. Most remarkably,
the area of allowed values of JCP “bifurcates” as a function of |Ue3 |2 and for values of
|Ue3 |2 0.008, JCP = 0 is no longer allowed. For |Ue3 |2 0.010, for instance, one has
|JCP | 0.01. This can be understood as follows. If ω = φ = 0 holds, the requirement of
vanishing JCP (see Eq. (33)) is fulfilled for ψ = 0. Then, one sees from Eq. (32) that to
order λ, the equality
1−B
tan2 θsol 1 − 4 |Ue3 | (42)
1+B
holds. For the values of B considered, B = (0.40–2.0), this equality is incompatible with
the 1σ upper bounds on8 tan2 θsol and |Ue3 |2 .
The “bifurcation” happens also when φ = ψ = 0 holds (Fig. 6). The lower limit on
|Ue3 |2 in this case is roughly by a factor of 4 smaller than in the preceding ones. Though
not shown here, there can also be interesting constraints on sin2 2θatm when some of the
phases vanish. If, e.g., ω = φ = 0 holds, one sees from Eq. (32) that 1 − sin2 2θatm receives
a correction of order λ2 0.06. It can be shown that typically sin2 2θatm 0.98 in this
case.
4.6. Specific models
We shall consider next specific models in which Eq. (24) arises. We will also investigate
how the CP-violating phases appear in the matrix Uν .
It is well known that when Ulep = 1, bimaximal neutrino mixing can be generated by
the following neutrino mass matrix [9,21]
 
0 1 1
m
mν = √  1 0 0  . (43)
2
1 0 0
CP = 0 for the minimal allowed value of |Ue3 | 0.006. As can

8 Note that, as is shown in Fig. 6, one has J 2
be seen from Eq. (33), the latter is reached in the case under discussion for the CP-conserving value of ψ = π ,
for which JCP vanishes.
Fig. 6. Scatter plot of the correlations between tan2 θsol and |Ue3 |2 , and between JCP and |Ue3 |2 , when the λij
are of the same order and some of the CP-violating phases, ψ , φ and ω, are zero. In the last row we display
the CP-violating cases corresponding to ω = 0 (bright green areas in web version) and φ = 0 (dark red area in
web version). The value λ12 = 0.24 and 1σ allowed ranges of values of the neutrino mixing parameters given in
Eq. (4) were used.
The latter has eigenvalues of 0 and ±m. This matrix has a flavour symmetry which
corresponds to the conservations of the lepton charge L = Le − Lµ − Lτ . One can
√ lift
the degeneracy of the two mass eigenvalues by adding a small perturbation m/ 2 in
the µτ entry. This yields a mass matrix known from the Zee model [32] (see also, e.g.,
[33–35]).
It is well known that there is no CP-violation in the Zee model since one can absorb all
three possible phases in the charged lepton fields. As we will see, this is true only when√the
charged lepton mass matrix is diagonal. One finds that in this case m2A m2 (1 − / 2 )
√
and m2 2m2 . This fixes
1 m2 1 7 · 10−5 eV2

√ √ 0.025, (44)
2 m2A 2 2 · 10−3 eV2
where we used the best-fit values found in the recent analyzes in [2,4]. For = 0 the√solar
mixing is slightly reduced from its maximal value [34,35]: we have tan2 θsol 1 − / 2
0.98. This is incompatible even with the 5σ allowed range of values of tan2 θsol (see, e.g.,
[2]).
In order to examine the implications of existence of phases in mν , we will assume that
the three non-vanishing entries in mν are complex and will denote the phases of these
entries—the eµ, eτ and µτ , by α , β and γ , respectively. The neutrino mass Lagrangian
then has the form:
m
L = − √ ν̄eL νµR c
e−iα + ν̄eL ντcR e−iβ + ν̄µL ντcR e−iγ + h.c., (45)
2 2
c
where νlR ≡ C ν̄eL
T , C being the charge conjugation matrix. We can redefine the flavor
, l = e, µ, τ . Choosing
neutrino fields as follows: νlL = eial νlL
1
ae = − (α + β − γ ),
2
1
aµ = − (α − β + γ ),
2
1
aτ = − (β − α + γ ), (46)
2
:
we get in terms of the fields νlL
m c

L = − √ ν̄eL νµR + ν̄eL ντ cR + ν̄µL

ντ cR + h.c. (47)
2 2
Now the neutrino mass matrix contains only real entries and is diagonalized by an
orthogonal transformation with an orthogonal real matrix O. The matrix O has to a good
approximation the bimaximal mixing form, Eq. (1). The mass eigenstate Majorana fields
χj , j = 1, 2, 3, are related to the fields νL ≡ (νeL , ν , ν )T via ν = Oχ , where χ ≡
µL τ L L L L
(χ1L , χ2L , χ3L )T . Since νL can be written as P̃ν† νL , where νL ≡ (νeL , νµL , ντ L )T contains
the original flavor neutrino fields, and P̃ν ≡ diag(eiae , eiaµ , eiaτ ), we have νL = P̃ν OχL .
We can factor out one of the three phases in P̃ν , i.e.,

P̃ν = e−(i/2)(α +β −γ ) diag 1, e−i(γ −β ) , e−i(γ −α ) ≡ eiφν Pν . (48)
Thus, comparing the resulting formula with Eq. (29), we see that in this special case
of only three phases in mν , for the phase matrix Qν holds: Qν = 1. Consequently, the
characteristic leading order relations between S1 , S2 and JCP as discussed in Sections 4.2–
4.4 will hold in the three generic structures of Uλ discussed above. In terms of the notations
used earlier in this section we have φ = (β − γ ) and ω = (α − γ ). Thus, there are
altogether 3 CP-violating phases in Ũlep and Uν in the specific model we are considering.
In the case of α = β , one has φ = ω.
The mixing parameters tan2 θsol , Ue3 and sin2 2θatm are functions of the 3 angles and
1 phase in Ũlep and of the 2 phases in Pν . The formulae and results we have derived in
Sections 4.2–4.5 are valid in the specific model under discussion if one sets ρ = σ = 0 and
takes into account the fact that φ = (β − γ ) and ω = (α − γ ). If Ulep = 1, the phases in
Pν are unphysical and can be absorbed by the charged lepton fields, CP is conserved in the
lepton sector and JCP = S1 = S2 = 0.
The neutrino mass term, Eq. (45), leads to a neutrino mass spectrum with inverted
hierarchy: m2 −m1 , m3 m2 , and m2 = m22 − m21 > 0, m2A = m22 − m23
m2
m2 . The effective Majorana mass measured in (ββ)0ν -decay, |m |, is given
approximately by (see, e.g., [19])
√

2mλ|1 − Bei(α −β −ψ) |, case I,
|m | m2A
Ue1
2 2
− Ue2 √ (49)
2mλ|1 + Bei(β −α −ψ) |, case II,
where case I and case II refer respectively to λ12 = λ, λ23 = Aλ, λ13 = Bλ, and to λ12 = λ,
λ23 = 1, λ13 = Bλ. Note that |m | does not vanish in the limit of = 0 [9]. Comparing
the expressions for |m | with that for JCP ,

 1
√ (s + Bsα −γ −ψ )λ, case I,
4 2 β −γ
JCP (50)
 1
√ (s − Bsβ −γ −ψ )λ, case II,
4 2 α −γ
one finds that all three phases (α − γ ), (β − γ ) (or φ, ω) and ψ contribute to both, the
“Dirac” phase δ and to the single Majorana phase that enters into the expression for |m |
in the case of neutrino mass spectrum with inverted hierarchy [18,20].
If we set β = α , only ψ contributes to |m |, while JCP is a non-trivial function of the
two phases (α − γ ) ≡ φ and (α − γ − ψ) ≡ (φ − ψ). If, however, β = γ , then√ φ = 0,
ω = (α − β ), and we have to leading order in λ in case I: JCP λBsα −β −ψ /(4 2 )
√ √
S1 /(2 2 ) S2 /(2 2 ). The effective Majorana mass as given in Eq. (49) can be written
as:
|m | m| cos 2θsol − i8JCP |. (51)
The same relation holds in case II when α = γ . Thus, quite remarkably, in these cases the
deviation of |m | from the minimal value it can have in the case of neutrino mass spectrum
with inverted hierarchy (see, e.g., [19]), min(|m |) m cos 2θsol, is determined by JCP .
A similar analysis can be performed and analogous results can be obtained if the small
term ∼ , generating the requisite splitting between the two non-zero eigenvalues of the
matrix (43), is present in one of the diagonal entries in Eq. (43).
5. Conclusions
Assuming that the three flavour neutrino mixing matrix UPMNS a product of two unitary
matrices Ulep and Uν arising from the diagonalization of the charged lepton and neutrino
†
mass matrices, UPMNS = Ulep Uν , and that apart from possible CP-violating phases, Uν has
a bimaximal mixing form, Eq. (1), we have performed in the present article a systematic
†
study of the possible forms of the matrix Ulep which are compatible with the existing data
on neutrino mixing and oscillations. The three mixing angles in Ulep were treated as free
parameters without assuming any hierarchy relation between them. The case of CP-non-
conservation was of primary interest and was analyzed in detail.
We have found that there exist three possible generic structures of Ulep , which are
compatible with the existing data on neutrino mixing. One corresponds to a hierarchical
“CKM-like” matrix. In this case relatively large values of the solar neutrino mixing angle
θsol , and of |Ue3 |2 ≡ |(UPMNS )e3 |2 , are typically predicted, tan2 θsol 0.42, |Ue3 |2 0.02,
while the atmospheric neutrino mixing angle θatm can deviate noticeably from π/4,
sin2 2θatm 0.95. The second corresponds to one of the mixing angles in Ulep being equal
to π/2, and predicts practically maximal atmospheric neutrino mixing sin2 2θatm 1.
Large atmospheric neutrino mixing, sin2 2θatm 0.95, is naturally predicted by the third
possible generic structure of Ulep , which corresponds to all three mixing angles in Ulep
being large.
†
The parametrization of the 3 flavour neutrino mixing matrix as UPMNS = Ulep Uν is
useful especially for studying the case of CP-non-conservation. We have found that, in
†
general, in addition to the three Euler-like angles, Ulep contains only one CP-violating
phase, while Uν includes four CP-violating phases and can be parametrized as follows:
Uν = Pν Ubimax Qν , where Pν and Qν are diagonal phase matrices containing two phases
each, and Ubimax is the bimaximal mixing matrix, Eq. (1). The two phases in Qν are
Majorana-like—the flavour neutrino oscillation probabilities do not depend on them. In
general, the three angles and the Dirac CP-violating phase UPMNS in the PMNS matrix
UPMNS depend in a complicated manner on the three angles and the phase in Ulep and
on the two phases in Pν . Finally, the two Majorana CP-violating phases in UPMNS are
functions of the all the five CP-violating phases in Ulep , Pν , and Qν , and of the three
angles in Ulep .
We have derived approximate and rather simple expressions for the observables
tan2 θsol , |Ue3 | and sin2 2θatm , as well as for three rephasing invariants of the UPMNS matrix,
JCP and S1,2 , associated respectively with CP-violation due to the Dirac phase and due to
the two Majorana phases in UPMNS . This is done for each of the three possible generic
structures of Ulep . In certain specific cases we find that simple relations between JCP and
S1,2 hold (see, e.g., Eq. (35)).
†
Finally, we considered a simple model in which UPMNS is expressed in terms Ulep Uν
with Uν = Pν Ubimax Qν . In this model the neutrino mass term is assumed to have the same
form (when we set Ulep = 1) as that of the Zee model. One finds that in this case Qν = 1,
while Pν contains two CP-violating phases. Expressions for the earlier indicated neutrino
mixing observables and for the effective Majorana mass in (ββ)0ν -decay, |m |, are given.
We find that if the CP-violating phases in Pν satisfy a certain condition, |m | depends in

a simple way on JCP (Eq. (51)).
It would be interesting to generalize to the matrix Ulep Pν the idea of unitarity triangles,
which has been successfully used in analyzing CP-violation in the quark sector, particularly
in B-meson decays. This could provide a helpful graphical representation of CP-violation
in neutrino oscillations.
Acknowledgements
It is a pleasure to thank S. Choubey and A. Smirnov for helpful discussions. W.R.

would like to acknowledge with gratefulness the hospitality of the Physics Department
at Dortmund University, where part of this study was done. This work was supported
in part by the EC network HPRN-CT-2000-00152 (W.R.), by US Department of Energy
Grant DE-FG02-97ER-41036 (P.H.F.), and by the Italian INFN under the program “Fisica
Astroparticellare” (S.T.P.).
References
[1] S. Pascoli, S.T. Petcov, hep-ph/0308034.

[2] A. Bandyopadhyay, et al., hep-ph/0309174, Phys. Lett. B, submitted for publication.
[3] B. Pontecorvo, Zh. Eksp. Teor. Fiz. 33 (1957) 549;
B. Pontecorvo, Zh. Eksp. Teor. Fiz. 34 (1958) 247;
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
[4] Y. Hayato, Talk presented at International Europhysics Conference on High-Energy Physics (HEP 2003),
Aachen, Germany, 17–23 July 2003, http://eps2003.physik.rwth-aachen.de/transparencies/07/index.php.
[5] S.N. Ahmed, et al., SNO Collaboration, nucl-ex/0309004.
[6] M. Apollonio, et al., CHOOZ Collaboration, Phys. Lett. B 466 (1999) 415.
[7] F. Boehm, et al., Phys. Rev. Lett. 84 (2000) 3764.
[8] W. Rodejohann, hep-ph/0309249, Phys. Rev. D, submitted for publication.
[9] S.T. Petcov, Phys. Lett. B 110 (1982) 245.
[10] R. Barbieri, et al., JHEP 9812 (1998) 017;
G. Altarelli, F. Feruglio, JHEP 9811 (1998) 021;
S. Davidson, S.F. King, Phys. Lett. B 445 (1998) 191;
R.N. Mohapatra, S. Nussinov, Phys. Rev. D 60 (1999) 013002;
C.H. Albright, S.M. Barr, Phys. Lett. B 461 (1999) 218;
Y. Nomura, T. Yanagida, Phys. Rev. D 59 (1999) 017303;
A.S. Joshipura, S.D. Rindani, Eur. Phys. J. C 14 (2000) 85;
R.N. Mohapatra, A. Perez-Lorenzana, C.A. de Sousa Pires, Phys. Lett. B 474 (2000) 355;
A. Aranda, C.D. Carone, P. Meade, Phys. Rev. D 65 (2002) 013011;
R. Kitano, Y. Mimura, Phys. Rev. D 63 (2001) 016008;
K.S. Babu, S.M. Barr, Phys. Lett. B 525 (2002) 289;
H.S. Goh, R.N. Mohapatra, S.P. Ng, Phys. Lett. B 542 (2002) 116.
[11] Z.Z. Xing, Phys. Rev. D 64 (2001) 093013.
[12] C. Giunti, M. Tanimoto, Phys. Rev. D 66 (2002) 053013.
[13] C. Giunti, M. Tanimoto, Phys. Rev. D 66 (2002) 113006.
[14] S.M. Bilenky, J. Hosek, S.T. Petcov, Phys. Lett. B 94 (1980) 495.
[15] M. Doi, et al., Phys. Lett. B 102 (1981) 323.
[16] G.L. Fogli, et al., Phys. Rev. D 67 (2003) 093006.
[17] P. Langacker, et al., Nucl. Phys. B 282 (1987) 589.

[18] S.M. Bilenky, et al., Phys. Rev. D 54 (1996) 4432.
[19] S.M. Bilenky, S. Pascoli, S.T. Petcov, Phys. Rev. D 64 (2001) 053010.
[20] S. Pascoli, S.T. Petcov, L. Wolfenstein, Phys. Lett. B 524 (2002) 319;
W. Rodejohann, hep-ph/0203214;
S. Pascoli, S.T. Petcov, W. Rodejohann, Phys. Lett. B 549 (2002) 177.
[21] C.N. Leung, S.T. Petcov, Phys. Lett. B 124 (1983) 461.
[22] S.M. Bilenky, S.T. Petcov, Rev. Mod. Phys. 59 (1987) 671.
[23] C. Albright, et al., hep-ex/0008064;
M. Apollonio, et al., hep-ph/0210192.
[24] S. Pascoli, S.T. Petcov, W. Rodejohann, Phys. Rev. D 68 (2003) 093007.
[25] C. Jarlskog, Z. Phys. C 29 (1985) 491;
C. Jarlskog, Phys. Rev. D 35 (1987) 1685.
[26] P.I. Krastev, S.T. Petcov, Phys. Lett. B 205 (1988) 84.
[27] J.F. Nieves, P.B. Pal, Phys. Rev. D 36 (1987) 315;
J.F. Nieves, P.B. Pal, Phys. Rev. D 64 (2001) 076005.
[28] G.C. Branco, L. Lavoura, M.N. Rebelo, Phys. Lett. B 180 (1986) 264.
[29] J.A. Aguilar-Saavedra, G.C. Branco, Phys. Rev. D 62 (2000) 096009.
[30] J.F. Nieves, private communication.
[31] P.H. Frampton, S.L. Glashow, D. Marfatia, Phys. Lett. B 536 (2002) 79.
[32] A. Zee, Phys. Lett. B 93 (1980) 389;
A. Zee, Phys. Lett. B 95 (1980) 461, Erratum;
L. Wolfenstein, Nucl. Phys. B 175 (1980) 93;
S.T. Petcov, Phys. Lett. B 115 (1982) 401.
[33] P.H. Frampton, S.L. Glashow, Phys. Lett. B 461 (1999) 95.
[34] Y. Koide, Phys. Rev. D 64 (2001) 077301;
Y. Koide, hep-ph/0201250.
[35] P.H. Frampton, M.C. Oh, T. Yoshikawa, Phys. Rev. D 65 (2002) 073014.
Probing minimal 5D extensions of the Standard

Model: from LEP to an e+e− linear collider
Alexander Mück a , Apostolos Pilaftsis b , Reinhold Rückl a
a Institut für Theoretische Physik und Astrophysik, Universität Würzburg, Am Hubland,
97074 Würzburg, Germany
b Department of Physics and Astronomy, University of Manchester, Manchester M13 9PL, United Kingdom
Received 13 January 2004; received in revised form 20 February 2004; accepted 18 March 2004
Abstract
We derive new improved constraints on the compactification scale of minimal 5-dimensional (5D)
extensions of the Standard Model (SM) from electroweak and LEP2 data and estimate the reach of
an e+ e− linear collider such as TESLA. Our analysis is performed within the framework of non-
universal 5D models, where some of the gauge and Higgs fields propagate in the extra dimension,
while all fermions are localized on a S 1 /Z2 orbifold fixed point. Carrying out simultaneous multi-
parameter fits of the compactification scale and the SM parameters to the data, we obtain lower
bounds on this scale in the range between 4 and 6 TeV. These fits also yield the correlation of the
compactification scale with the SM Higgs mass. Investigating the prospects at TESLA, we show
that the so-called GigaZ option has the potential to improve these bounds by about a factor 2 in
almost all 5D models. Furthermore, at the center of mass energy of 800 GeV and with an integrated
luminosity of 103 fb−1 , linear collider experiments can probe compactification scales up to 20–
30 TeV, depending on the control of systematic errors.
1. Introduction
The Kaluza and Klein (KK) paradigm [1] that our world may realize more than
four dimensions has been a central theme of the last ten years [2–5]. The additional
dimensions have to be sufficiently compact to explain why they have escaped detection
so far, their allowed size is, however, highly model-dependent. For example, if gravity
is the only force that feels the existence of additional space dimensions, the size of the
E-mail address: mueck@physik.uni-wuerzburg.de (A. Mück).
doi:10.1016/j.nuclphysb.2004.03.019
56 A. Mück et al. / Nuclear Physics B 687 (2004) 55–75
compactification radius R could be as large as 10−3 mm [4], without being in conflict with
phenomenological limits from collider experiments [6] and cosmological constraints [7].
This bound gets much stronger, if fields charged under the Standard Model (SM) gauge
group propagate in the extra dimensions as well. Again, the actual value of the lower
bound on the compactification scale M = R −1 crucially depends on the details of the
model. If all SM fields experience the existence of one extra compact dimension as in
the so-called universal extra-dimensional scenario [2,5,8–10], the lower limit on M is
rather weak. Mainly from the KK loop contributions to the ρ parameter [8] and the decay
Z → b b̄ [10], one finds M 300 GeV. However, if some of the SM fields, in particular, the
fermions, are confined to the familiar 4-dimensional world, M is constrained much more
strongly, namely, M 4 TeV [11]. This order of magnitude increase of the bound in non-
universal 5-dimensional (5D) settings of the SM can be attributed to the fact that single KK
excitations couple at tree-level to light SM modes on the brane. In universal models these
couplings are forbidden by selection rules.
In this paper we improve the constraints on the compactification scale M derived
earlier [11] in a wide class of non-universal 5D models by taking the latest LEP2 data
into account. Furthermore, we carry out a multi-parameter global fit of M simultaneously
with the SM parameters. This allows us to properly include the correlations of the
compactification scale with the SM parameters, in particular, with the Higgs-boson mass
mH and the top-quark mass mt . In [12], it has been argued that the value of mH ,
bounded from below by direct Higgs searches (mHSM 114 GeV) [13] and from above
by perturbative unitarity (mHSM 1 TeV) [14], may have significant effects on the limits
on M derived by a global analysis of electroweak precision data, and vice versa. Finally,
we systematically investigate the sensitivity of future experiments at a 500–800 GeV linear
e+ e− collider such as TESLA. In this analysis, we also study the improvements which can
be expected from the so-called GigaZ option of TESLA, where the machine is operated at
the Z pole with a luminosity 100 times larger than that of LEP.
The general theoretical framework of our investigations is provided by 5D extensions
of the SM (5DSM) compactified on an S 1 /Z2 orbifold, where all fermions are localized on
one of the two orbifold fixed points. As far as the gauge fields are concerned, we consider
three possible scenarios: (i) all gauge fields propagate in the extra dimension, i.e., the bulk;
(ii) only the SU(2)L gauge bosons are bulk fields, while the U(1)Y gauge field is confined
to the orbifold fixed point where the fermions live; (iii) only the U(1)Y boson propagates in
the bulk, while the SU(2)L bosons are restricted to the brane. As has been shown in [11],
the above 5D models can be consistently quantized using appropriate 5D gauge-fixing
conditions that lead, after the KK reduction, to the known class of Rξ gauges.
Although it is not our intention to put forward an explicit string-theoretic construction
of these non-universal 5D models, we note that they may result from the intersection
of higher-dimensional p-branes [15,16] within the context of type I and type II string
theories [17]. The SU(2)L and U(1)Y gauge groups of the SM may be associated
with two separate intersecting higher-dimensional spaces (Dp branes) that have different
compactification radii. If the SU(2)L compactification radius is so small that the KK
states become heavy enough to decouple from the low-energy observables of interest, the
low-energy sector of such a model would effectively look like a scenario with SU(2)L
gauge bosons confined to a 3-dimensional brane. In this construction, all SM fermions are
A. Mück et al. / Nuclear Physics B 687 (2004) 55–75 57
assumed to be localized on the intersection of the SU(2)L and U(1)Y branes that constitutes
our observable 3-dimensional world. Localized brane interactions may also be induced by
radiative effects from KK bosons in the bulk. They are omitted in our analysis. Detailed
discussions on this topic may be found in [18].
The article is organized as follows. In Section 2 we summarize the main features of
the minimal non-universal 5D extensions of the SM studied in detail in [11], and sketch
the phenomenological consequences. Section 3 describes our approach to probing 5D
models and presents the constraints on the compactification scale M obtained from multi-
parameter fits to electroweak precision data, including error correlations. In Section 4 we
analyze the 5D effects on the total cross sections and asymmetries of fermion-pair and
W -pair production at LEP2 and evaluate the corresponding bounds on M. In addition,
we give combined electroweak and LEP2 bounds. The sensitivity to M at a future e+ e−
linear collider is discussed in Section 5. Section 6 highlights the main conclusions. Finally,
the novel couplings of the KK bosons entering W -pair production and Higgsstrahlung are
presented in Appendices A and B.
2. Theoretical framework
In this section, we briefly review the minimal 5D extensions of the SM under study.
Further details may be found in [11]. We start by considering the so-called bulk–bulk model
where all gauge fields propagate in the extra dimension. The gauge and Higgs sector of this
model is described by the 5D Lagrangian
1 1 a
L(x, y) = − BMN B MN − FMN F aMN + (DM Φ1 )† (D M Φ1 )
4 4

+ δ(y)(Dµ Φ2 )† D µ Φ2 , (2.1)
where BMN denotes the U(1)Y field strength and a
FMN (a = 1, 2, 3) the SU(2)L field
strength. The covariant derivative DM is defined by
g5 a a g
DM = ∂M − i AM τ − i 5 BM , (2.2)
2 2
where τ a denote the Pauli matrices, and analogously for Dµ . In (2.1) the Higgs potential,
gauge-fixing and ghost terms are omitted for brevity. Throughout the paper, the 5D Lorentz
indices are denoted with capital Roman letters, in the above M, N = 0, 1, 2, 3, 5, while for
the respective 4D indices Greek letters are used, in the above µ = 0, 1, 2, 3. The Higgs
doublet Φ2 is restricted to a brane at the orbifold fixed-point y = 0, while the doublet Φ1
propagates in the bulk. The zero-mode of Φ1 and Φ2 acquire the vacuum expectation
values
(VEV) v1 and v2 , respectively. As usual, we define tan β = v2 /v1 and v = v12 + v22 . In
the phenomenological analysis we will often focus on the cases sin β = 0 or sin β = 1.
Different minimal 5D extensions of the SM can be obtained by restricting either the
SU(2)L or the U(1)Y gauge boson to the brane at y = 0. In the first case, one has

1 1 a aµν
L(x, y) = − BMN B MN + δ(y) − Fµν F + (Dµ Φ)† (D µ Φ) , (2.3)
4 4
while in the second model

1 a 1
L(x, y) = − FMN F aMN + δ(y) − Bµν B µν + (Dµ Φ)† (D µ Φ) . (2.4)
4 4
We will refer to (2.3) as the brane–bulk model and to (2.4) as the bulk–brane model. Here,
any Higgs doublet has to be confined to the brane because of gauge invariance. In all the
above models, the fermionic degrees of freedom are localized on the brane.
Compactification and integration over the extra dimension is performed most easily
by expanding the bulk fields in KK modes which respect the symmetries of the S 1 /Z2
orbifold. The Z2 -parity of the bulk fields is chosen such that the theory is gauge-invariant at
the classical level and the light degrees of freedom (the zero modes of a Fourier expansion)
coincide with those of the SM spectrum.
The resulting effective 4-dimensional theory contains heavy KK vector and scalar
particles with masses that are multiples of the compactification scale M, in addition to
the usual SM degrees of freedom. Electroweak symmetry breaking by a VEV of a brane
Higgs field leads to mixing between different KK modes including the zero modes which
shifts the masses and the couplings to fermions. These shifts are especially important
for observables at the Z pole where effects from the exchange of heavier KK modes,
dominating at high energies, are negligible.
In our phenomenological analysis, we proceed as follows. The prediction of the 5DSM
for a given observable O5DSM is related to the SM prediction OSM by

O5DSM = OSM 1 + ∆5DSM O , (2.5)
where ∆5DSM
O is the tree-level effect due to the compactified extra dimension. The SM
radiative corrections are included in OSM . However, SM loop effects on ∆5DSM O as
well as KK loop effects are neglected. For compactification scales in the TeV range
this is well justified. The tree-level calculation of ∆5DSM
O is performed in terms of the
compactification scale M and the usual SM input parameters, that is the electromagnetic
fine structure constant α, the Fermi constant GF , and the Z-boson mass mZ(0) . The
index (0) indicates that the observed Z boson is to be identified with the lightest mode
of the corresponding KK tower. The remaining SM parameters mt , mH , and the strong
coupling constant αs (mZ ) enter in OSM , but do not influence the calculation of ∆5DSM
O in
this approximation.
Furthermore, one has to take into account that the tree relations between the SM input
parameters and other SM parameters like gauge couplings and VEVs are also affected by
the extra dimension. An exception is the fine structure constant
α = e2 /4π. (2.6)
To order 1/M 2 , one finds
(g 2 + g 2 )v 2
m2Z(0) = (1 + ∆Z X), (2.7)
4
πα
GF = √ 2
(1 + ∆G X), (2.8)
2 sin θW cos2 θW m2Z(0)
where
g5 e g e
g= √ = , g = √ 5 = (2.9)
2πR sin θW 2πR cos θW
are the usual 4D gauge couplings. Inverting (2.8), one obtains

cos2 θ̂W
sin θW = sin θ̂W 1 − 2
2 2
∆G X , (2.10)
sin θ̂W − cos2 θ̂W
where θ̂W denotes the SM value of the weak mixing angle following from
πα
GF = √ 2
. (2.11)
2 sin θ̂W cos2 θ̂W m2Z(0)
In the above, the shifts with respect to the SM relations are parameterized by the model-
dependent, but mass-independent coefficients ∆Z and ∆G and the dimensionless factor
2
π 2 mZ(0)
X= 1. (2.12)
3 M2
For the bulk–bulk, the brane–bulk, and the bulk–brane model one has, respectively,

∆Z = −sβ4 , −ŝW 2
, −ĉW
2
, (2.13)

2
ŝ 4
∆G = ĉW 2
1 − 2sβ2 − W 2 β
s , −ŝW
2
, −ĉW
2
, (2.14)
ĉW
where the abbreviations of the trigonometric functions are obvious. It is the shift in sin2 θW
given in (2.10) which provides very sensitive tests of the above 5D models as will be seen in
Section 5.1. Similarly for other observables, we will also expand the shift ∆5DSM
O in X and
keep only the linear term. Exact analytic expressions for 5D shifts in masses and couplings
to all orders in X are presented in [11].
After compactification, the couplings of the KK modes of the gauge bosons to the SM
brane fermions are determined by their SM quantum numbers. In the interaction or weak
basis, these couplings are generically given by
∞

√
Lint (x) = g Ψ̄ γ gV − gA γ Ψ A(0)µ + 2
µ 5
A(n)µ , (2.15)
n=1
where gV and gA are the usual vector and axial vector coupling constants, and A(n)µ
denotes the nth KK mode of a given gauge field. However, in the presence of a nonzero
VEV of a brane Higgs field, the KK Fourier-modes mix to form mass eigenstates Â(n)µ .
The couplings of these physical fields can then be parameterized as follows:
∞

Lint (x) = g(n) Ψ̄ γ µ gV (n) − gA(n) γ 5 Ψ Â(n)µ . (2.16)
n=0
√
For Â(n)µ = W(n)µ , one has gV (n) = gA(n) = 1 and g(n) = gW (n) /(2 2 ), where for the
observed W boson

gW (0) = g 1 − sβ2 ĉW
2
X, 1, 1 − ĉW
2
X (2.17)
in the bulk–bulk, brane–bulk, and bulk–brane model, respectively. In the brane–bulk

model, there are of course no higher KK modes of the W . Similarly for the Z modes,
one obtains gV (n) = T3f (n) − 2Qf (n) sW
2 , g
A(n) = T3f (n) , and g(n) = gZ(n) /2cW , where,
focusing on the observed Z boson,

gZ(0) = g 1 − sβ2 X, 1, 1 ,

T3f (0) = T3f 1, 1 − ŝW
2
X, 1 − ĉW
2
X ,

Qf (0) = Qf 1, 1 − X, 1 . (2.18)
Here, Qf and T3f are the fermion charge (Qe = −1) and 3-component of the weak isospin
(T3e = −1/2). The photon is not affected by electroweak √ symmetry breaking and, hence,
gV (n) = 1, gA(n) = 0, and e(n) = e for n = 0 and 2e for n 1. In the brane–bulk and
the bulk–brane models, the higher photon KK modes are absent. The coupling parameters
gW (n) , gZ(n) , T3f (n) , and Qf (n) for the higher KK modes (n 1) can also be obtained from
Appendix B of [11]. To first order in X, they are given by

√ 3 √ 3
gW (n) = g 2 1− s 2 2
ĉ X , no KK modes, 2 1 − ĉ 2
X ,
2π 2 n2 β W 2π 2 n2 W

√ 3
gZ(n) = g 2 1− s 2
X , 1, 1 ,
2π 2 n2 β

√ 1 2 3
T3f (n) = T3f 1, 2 sW 1 + ĉW 2
− ŝW X ,
2 π 2 n2

√ 1 2 3
2 cW 1 + ŝW 2
− ĉW X ,
2 π 2 n2

√
2 3 √ 3
Qf (n) = Qf 1, 1 − ŝW 2 2 X , 2 ĉW 2 2 X .
2
(2.19)
sW 2π n π n
3. Electroweak constraints revisited
In order to extract bounds on M from the available data, one can proceed in two ways.
For the set of precision observables specified in [11] (mostly from LEP), one may fix the
SM parameters at their current best fit values and calculate the 5DSM predictions O5DSM
using (2.5) and taking the SM expectations OSM from [13]. Then, one can perform a one-
parameter χ 2 -analysis for M or equivalently X defined in (2.12). The χ 2 -function is given
by
exp exp
χ 2 (X) = Oi − Oi5DSM Vij−1 Oj − Oj5DSM (3.1)
i,j
with the covariance matrix Vij = Oi ρij Oj , Oi being the measurement error of a
exp
given observable Oi and ρij being the matrix of correlation coefficients. This approach
is widely used in the literature. It has also been followed in our previous analysis [11].
However, in this approach possible correlations between the SM parameters and the size
of the extra dimension are ignored, and hence the bounds on M may be overestimated.
Table 1
2σ bounds on M in TeV derived from electroweak precision measurements [13]
Brane–bulk Bulk–brane Bulk–bulk Bulk–bulk
(brane Higgs) (bulk Higgs)
One-parameter fit 4.9 3.2 5.5 4.2
Multi-parameter fit 3.1 3.1 4.2 3.7
Therefore, it is interesting to follow a more general approach in which X is fitted

simultaneously with the SM parameters αem (mZ ), GF , mZ , αs (mZ ), mt , mH to the
data. The SM predictions OiSM in (2.5) are obtained from ZFITTER [19], while the
5D corrections ∆5DSM
O are taken from [11]. The multi-parameter minimization of χ 2 is
performed with the program MINUIT [20].
The bounds on X can be derived from (3.1) by either using Bayesian statistics1 or by
simply requiring
χ 2 = χ 2 (X) − χmin
2
< n2 (3.2)
for X not to be excluded at the nσ confidence level. In the above, χ 2 (X) is the minimum
for a given X with respect to all other fit parameters requiring mH 114 GeV, while χmin2
is the overall minimum in the physically allowed region X 0, mH 114 GeV. If the best
fit value of X is not too far in the unphysical region, both methods lead to similar results
and approximate well the results of the unified approach [21]. We will present the bounds
as obtained from (3.2).
Having described the methods, we will now discuss the resulting bounds on the
compactification scale M listed in Table 1. Superseding the one-parameter fit in [11], we
use the latest experimental data [13] and take into account the error correlations which
had previously been ignored. The error correlations have only a small effect, shifting the
bounds by no more than 0.2 TeV. However, the change in the data with respect to the data
of the year 2000 alters the bounds by as much as 1 TeV. This mainly is due to the large
shift of the experimental value for the forward-backward asymmetry A(0,b) FB which is now
found to be more than 3σ below the SM expectation. Only the bulk–bulk model with a
(0,b)
bulk Higgs predicts a smaller value of AFB than the SM. Correspondingly, the bound for
this model is lowered while the constraints on the other models become stronger.
In the multi-parameter fit, the correlations between M and the SM parameters reduce
the bounds as expected, the size of the effect varying from model to model. In the brane–
bulk model the effect is biggest lowering the bound by almost 40%. Here, the best fit
value is relatively far in the unphysical region. Thus, as a cross-check, we also performed
a Bayesian analysis leading to a bound which is 0.1 TeV below the bound from (3.2).
Most interesting is the correlation between the compactification scale and the mass
of the Higgs boson. This correlation is illustrated in Fig. 1. The data set used for this
analysis differs from the one used for the familiar blue-band plot [22]. Within the SM
1 For example, in Bayesian statistics with a flat prior in the physical region X 0 and a zero prior in the
X
unphysical region, the 95% (1.96σ ) c.l. bound X95 is given by 0.95 = 0 95 dX P (X)/ 0∞ dX P (X), where
P (X) = exp[−(χ 2 (X) − χmin
2 )/2].
Fig. 1. Contours of χ 2 = 1, 4, 9 (Eq. (3.2)) derived from multi-parameter fits to electroweak precision data. The
shaded regions of the parameter space correspond to mH < 114 GeV and/or M 2 < 0.
(X = 0 in Fig. 1) it leads to a best fit value mH = 100+70

−40 GeV and to the 2σ upper bound
mH < 280 GeV. As can be seen in Fig. 1, in all models with the Higgs field localized on
the brane, the existence of an extra dimension favors a heavier Higgs boson. This confirms
the observation for the bulk–bulk model in [12]. The effect is most pronounced in the bulk–
bulk model with a brane Higgs and in the brane–bulk model. Quantitatively, for M = 5 TeV,
the best fit values increase to mH = 170+105 +105
−60 GeV and mH = 155−60 GeV, respectively.
If the compactification scale is included in the multi-parameter fit, the 2σ upper bound
on mH is relaxed to 330 and 400 GeV, respectively.
4. LEP2 constraints
At energies above the Z pole, the virtual exchange of KK excitations of the SM

gauge bosons becomes dominant. As long as s M 2 , the main effects come from the
interference of zero with higher KK modes. These effects scale like s/M 2 in contrast to
the energy-independent modifications of masses and couplings, as is illustrated in Fig. 2 for
muon-pair production. The residual energy dependence of the impact of the latter on the
Fig. 2. The shift ∆5DSM

µ+ µ−
of the total cross section for muon-pair production in the bulk–bulk model with a brane
Higgs for M = 5 (a), 10 (b), and 20 (c) TeV. The dashed curves show the effects from the mixing of masses and
couplings only.
cross section (dashed curves in Fig. 2) is due to the transition from dominant Z exchange
to dominant photon exchange.
Focusing on LEP2, we have investigated the total cross sections for lepton-pair
production, hadron production, and Bhabha scattering. Forward–backward asymmetries
for muon and tau production as well as the heavy quark observables A(0,b) (0,c)
FB , AFB ,
Rb = σ (b b̄)/σ (had), and Rc = σ (cc̄)/σ (had) are included in the fits, although they do not
contribute noticeably to the bounds. For completeness, we have also investigated W + W −
production.
4.1. Fermion-pair production
The differential cross section for fermion-pair production is given by

dσ (e+ e− → f f ) Nf s ef 2 ef 2
= (1 + cos ϑ)2 MLL (s) + MRR (s)
d cos ϑ 128π
ef 2 ef 2
+ (1 − cos ϑ)2 MLR (s) + MRL (s) , (4.1)
where ϑ is the scattering angle between the incoming electron and the negatively charged
outgoing fermion, and Nf = 1(3) for leptons (quarks) in the final state. For the matrix
elements entering (4.1) one finds
∞ e
gα(n)
f
gβ(n)

ef Qe Qf 1
Mαβ (s) = 2
e(n) + (4.2)
n=0
s − m2γ (n) cos2 θW s − m2Z(n)
with the couplings
f gZ(n)
gL,R(n) = (gV (n) ± gA(n) ) (4.3)
2
Table 2
2σ bounds on M in TeV from a one-parameter fit derived from LEP2 data alone and from LEP2 and electroweak
precision data combined
µ+ µ− 2.0 1.5 2.5 2.5
τ +τ − 2.0 1.5 2.5 2.5
hadrons 2.6 4.7 5.4 5.8
e+ e− 3.0 2.0 3.6 3.5
W +W − 1.1 2.0 2.1 2.2
LEP2 combined 3.5 4.6 5.6 5.9
Electroweak
and LEP2 data 5.4 4.8 6.9 6.0
combined
derived from (2.16), (2.18), and (2.19). The expression (4.1) has also been used in [23].
However, the mixing effects in the couplings and masses included in (4.2) and (4.3) have
been neglected there.
The data taken by the LEP experiments √ for muon, tau, and
√hadron production is properly
combined in [22] for energies between s = 130 GeV and s = 207 GeV. In the hadronic
channel it is very important to take into account the large correlations between the data
at different energies. If they are ignored, the bounds on the compactification scale are
overestimated by as much as 3 TeV. On the other hand, correlations in the muon and tau
channels are extremely small and have little effect.
The bounds from a simple one-parameter analysis are summarized in Table 2. In the
muon and tau channel, the bulk–brane model is least restricted because essentially only
left-handed fermions interact with the KK modes. The best fit values turn out to lie always
in the physical region X 0. Hadron production puts more stringent bounds on the 5D
models because of the larger cross section. In this case, the brane–bulk model is least
restricted, since the hadronic cross section is dominated by left-handed quarks whose small
hypercharges suppress the interference effects with the U(1)Y KK modes. In this case, the
best fit values lie in the unphysical region X < 0.
The differential cross section for Bhabha scattering may conveniently be expressed as
dσ (e+ e− → e+ e− )
d cos ϑ
s ee 2 ee 2
= (1 + cos ϑ)2 MLL (s) + MLL
ee
(t) + MRR (s) + MRR
ee
(t)
128π
ee 2 ee 2
+ (1 − cos ϑ)2 MLR (s) + MRL (s)
ee 2 ee 2
+ 4 MLR (t) + MRL (t) , (4.4)
where t = −s(1 − cos ϑ)/2 and ee
Mαβ (s
or t) can be read off from (4.2). The total cross
section is calculated by integrating (4.4) over ϑ in the experimental ranges. Since the
Bhabha data of the four LEP experiments [24] has not yet been combined, possible
correlations of the different experiments cannot be accounted for, at least for the time
being.
As can be seen from Table 2, the bounds on M from Bhabha scattering are
approximately 1 TeV stronger than those from the other leptonic channels. This is due
to the large Bhabha cross section. On the other hand, the dominance of the t-channel
photon exchange, which is not affected by the presence of an extra dimension, reduces
the sensitivity of Bhabha scattering with respect to hadron production in almost all 5D
models.
4.2. W + W − production
The differential cross section for W + W − production reads [25]

dσ (e+ e− → W + W − )
d cos ϑ

1 s 3 s s2
= β β 2 ML2 (s) + MR2 (s) s 2 2 + sin2 ϑ − +
32πs mW (0) 4 4m2W (0) 16m4W (0)
2
s s s2
+ ML (t)t
2 2
+ β sin ϑ
2 2
+
4m2W (0) 16t 2 64m4W (0)

m2W (0) s
+ ML (t)ML (s)st 2 + 2 + β2 2
t mW (0)

s s s2
− β 2 sin2 ϑ + − , (4.5)
4t 8m2W (0) 16m4W (0)
where

β= 1 − 4m2W (0)/s, t = m2W (0) − s 1 − 1 − β cos ϑ /2,
and ϑ is the scattering angle between the electron and the negatively charged W boson.
Note, that mW (0) = mSM
W (1 + ∆mW X) with

2 ĉ 2 2 2 2 2 ĉ 2
1 4 2 ŝW ŝW ŝW ĉW ŝW
∆mW = s ŝ − W
1 − 2sβ − 2 sβ ,
2 4
, W
(4.6)
2 β W 2ĉ2W ĉW 2ĉ2W 2ĉ2W
for the bulk–bulk, the brane–bulk, and the bulk–brane model, respectively [11]. Further-
more, ML,R (s) and ML (t) are given by
∞ Q e gγ
e gZ
gα(n)

e (n) 3(n) 3(n) 1
Mα (s) = + ,
n=0
s − m2γ (n) cos θW s − m2Z(n)
2
gW (0)
ML (t) = , (4.7)
t
γ
while MR (t) = 0 as in the SM. A novel feature are the triple gauge-boson couplings g3(n)
Z
and g3(n) of the photon, the Z boson, and their respective KK modes with the W zero
modes. They are given in Appendix A along with the corresponding Feynman rules.
Fig. 3. Contours of χ 2 = 1, 4, 9 (Eq. (3.2)) derived from the combined analysis of LEP2 data and electroweak
precision measurements. The shaded regions of the parameter space correspond to mH < 114 GeV and/or
M 2 < 0.
As can be seen from Table 2, W -pair production at LEP2 provides relatively weak
bounds on M. This can be understood by realizing that the effects of KK exchange are
almost negligible due to the suppression of the interference of SM and KK exchange by an
additional factor X. This is a direct consequence of selection rules which forbid the triple
boson couplings of a single higher KK mode to zero modes for the gauge eigenstates such
that they can only be induced by mixing.
4.3. Combined bounds on the compactification scale M
The 2σ bounds on M found from a one-parameter fit to the combined LEP2 data are
listed in Table 2. The bounds range from 3.5 TeV for the brane–bulk model to 5.9 TeV
for the bulk–bulk model with a bulk Higgs. Furthermore, including also the electroweak
precision measurements, the bounds range from 4.8 TeV for the bulk–brane model to
6.9 TeV for the bulk–bulk model with a brane Higgs. For the bulk–bulk models our results
agree with the results of [23]. The best fit values always lie in the unphysical region X < 0.
Table 3
2σ bounds on M in TeV derived from the muon-, tau-, and hadron production at LEP2 combined with electroweak
precision measurements
One-parameter fit 5.0 4.5 6.4 5.5
Multi-parameter fit 3.6 4.3 5.4 5.2
For muon-pair, tau-pair and hadron production (including asymmetries and heavy-
quark data), where ZFITTER can be used to calculate the SM predictions, we have also
performed a multi-parameter fit. The resulting χ 2 contours are shown in Fig. 3. The slight
distortions from smooth contours are due to a discontinuity
√ in hadronic cross sections and
asymmetries found in ZFITTER version 6.36 at s = mt [26]. The corresponding 2σ
bounds on M are listed in Table 3 along with the bounds from the corresponding one-
parameter fit. The correlations between mH , mt , and M are similar to what has been found
from the precision observables alone in Section 3. They are weak in the bulk–brane model
and most sizable in the brane–bulk model.
A comparison of Tables 2 and 3 shows that in the one-parameter fit Bhabha scattering
and W + W − production increase the combined bounds by about 0.5 TeV. Thus, the bounds
from a multi-parameter fit to all data, including Bhabha scattering and W + W − production
can be estimated to lie between 4 TeV for the brane–bulk model and 6 TeV for the bulk–
bulk model with a brane Higgs.
5. Sensitivity at a linear collider
Having extracted the bounds on the compactification scale M from available data, we
will now estimate the reach at a future linear collider such as TESLA [27]. For illustration,
we investigate both the potential of the GigaZ option as well as the sensitivity at high
energy and luminosity.
5.1. GigaZ option
At the Z pole, the luminosity goal at TESLA is L = 5 × 1033 cm−2 s−1 which is
sufficient to produce 109 Z bosons in only 50–100 days of running [27]. This will increase
the LEP statistics by more than an order of magnitude. The most relevant improvement in
testing the compactification scale will come from the precise measurement of the left–right
(LR) asymmetry ALR . Since photon exchange and the exchange of higher KK modes can
be neglected on the Z peak, the LR asymmetry at tree level can be approximated by
2gV (0) gA(0)
ALR = , (5.1)
gV2 (0) + gA(0)
2
where gV (0) and gA(0) are the vector and axial vector couplings of the electron to the
Z boson given in (2.16). This asymmetry is very sensitive to shifts of the weak mixing
angle with respect to the SM value because of the small ratio gV /gA = (1 − 4 sin2 θW ).
Table 4
2σ bound on the compactification scale M in TeV which can be obtained with the GigaZ option at TESLA if the
measured value of ALR coincides with the SM expectation
ALR 12.4 6.8 14.2 12.4
Using (2.10) and (2.18) in order to express (5.1) in terms of the input parameters, one
obtains the 5DSM corrections ∆ALR = ∆Ae given in [11].
With the GigaZ option it will be possible to measure ALR with an absolute error of
about 10−4 [27]. However, the uncertainties in the fine structure constant and the Z mass
will each induce an additional error of about 10−4 which has to be added in quadrature.
Coincidence of the measured value of ALR with the SM expectation would then imply
the bounds on M shown in Table 4. As can be seen, except for the bulk–brane model, the
GigaZ option should allow to improve the existing bounds by at least a factor 2.
With excellent b-tagging, it will also be possible to considerably improve the mea-
surement of the final state coupling Ab and the cross-section ratio Rb . The experimental
error for the mass of the W boson can also be reduced significantly by a threshold scan.
However, the sensitivity of these observables to M is small and does not allow to explore
compactification scales beyond the bounds already known from available data.
√
5.2. Tests at s = 800 GeV
At high energies, the interference effects from the exchange of SM and KK modes
completely dominate the mixing effects as illustrated in Fig. 2. We consider the same
processes as in Section 4, except for W -pair production which is not very sensitive to an
extra dimension since the couplings of the W bosons to higher KK modes in the s-channel
are either forbidden or suppressed. For Bhabha scattering, an acceptance cut, | cos ϑ| < 0.9,
is included. Furthermore, we assume an integrated luminosity of 1000 fb−1 .
In addition, we also study Higgsstrahlung, the differential cross section of which is
given by
2
dσ (e+ e− → ZH ) s mZ(0)
= λ1/2 (s) 8 + λ(s)(1 − cos2 ϑ)
d cos ϑ 512π s
2
× ML (s) + MR (s) .
2
(5.2)
Here,

(mH + mZ(0))2 (mH − mZ(0))2
λ(s) = 1 − 1−
s s
is the familiar two-particle phase-space function, ϑ is the scattering angle between the
electron and the outgoing Z, and
∞ e ZH
gα(n) g(n) 1
Mα (s) = . (5.3)
cos2 θW s − m2Z(n)
n=0
√
Fig. 4. Expected sensitivity to M as a function of relative systematic error at s = 800 GeV and for an integrated
luminosity of 1000 fb−1 : (a) bulk–bulk model with brane Higgs, (b) bulk–bulk model with bulk Higgs, (c)
brane–bulk model, and (d) bulk–brane model.
e
The couplings gα(n) ZH is the effective coupling of the Z modes
are defined in (4.3) and g(n)
to the Higgs boson given in Appendix B. The integrated cross section following from (5.2)
in the SM limit can be found, for example, in [28].
The Higgsstrahlung process is certainly not a primary search channel3 for−1an extra
dimension
√ because of the limited experimental accuracy. For L dt = 10 fb and at
s = 800 GeV, the error on the total cross section for mH 115 GeV is expected to
be 5% [27]. Nevertheless, this channel is interesting for distinguishing between a brane
and a bulk Higgs in the bulk–bulk model. If the produced Higgs boson is the zero mode
of a bulk field, the KK selection rules forbid the coupling H(0)Z(0) Z(n) for n 1. Thus,
the absence of massive KK modes in the s-channel is a clear signal for a bulk Higgs. In
this case, this process is rather SM-like in contrast to a Higgs boson localized on the brane.
The phenomenology of 2-Higgs-doublet models with a bulk and a brane Higgs has been
investigated in [29].
In the following, the sensitivity to the presence of an extra dimension is estimated by
requiring
(OSM − O5DSM )2
χ2 = i i
4. (5.4)
(Oi )2
i
The statistical errors will be so small that the systematic errors become decisive [30,
31]. Since the latter are not reliably known at the present time, we show, in Fig. 4, the
sensitivity to the compactification scale M as a function of the relative systematic error.
Fig. 5. Combined sensitivity to the compactification scale M as a function of relative systematic error at
√
s = 800 GeV and for an integrated luminosity of 1000 fb−1 in the models bulk–bulk with brane Higgs (short
dashed), bulk–bulk with bulk Higgs (solid), brane–bulk (dashed), and bulk–brane (long dashed).
√
Fig. 6. Search limit as a function of the integrated luminosity at s = 800 GeV assuming a systematic error of
1.0% (left) and 0.1% (right). Combination of channels, and models as in Fig. 5.
At Osys 0.001 the statistical uncertainty begins to dominate and the sensitivity to the
compactification scale saturates.
The combined sensitivity of the search channels of Fig. 4 is presented in Fig. 5.
For bulk/brane (bulk/bulk) models, the sensitivity limit increases from 15 (20) TeV for
systematic errors at the 1% level to 35 (50) TeV for negligible systematic errors. The role
played by the statistics is illustrated in Fig. 6, where the combined sensitivity is plotted as
a function of integrated luminosity.
In addition to integrated cross sections, it is also interesting to study the effects of an
extra dimension on angular distributions. This may provide further handles to discriminate
between different models. For the muon and tau channel, the angular distribution in the
bulk–bulk model is almost completely SM-like. However, the brane–bulk and the bulk–
brane models lead to significant distortions of the angular distribution because of the almost
pure U(1)Y and SU(2)L nature of the heavy KK modes. In Bhabha scattering, the s- and
the t-channel are affected differently such that the angular distribution is also affected in
the bulk–bulk models.
Fig. 7. Deviation of the angular distribution (dσ/d cos ϑ)/σtot in muon-pair production (left) and Bhabha
scattering (right) from the SM predictions for M = 10 TeV in the bulk–bulk (solid), brane–bulk (dashed), and
bulk–brane (long-dashed) models.
For illustration, Fig. 7 shows the shift ∆5DSMϑ of (dσ/d cos ϑ)/σtot from the SM
prediction as defined in (2.5). If the angular distributions in the muon channel can be
measured with a precision better than 1% per bin (using ten bins), one can probe the
compactification scale M beyond 10 TeV for the brane–bulk and the bulk–brane model.
In Bhabha scattering, one can reach a similar scale also for the bulk–bulk models, while
the bulk–brane model is difficult to probe in this channel.
6. Conclusions
In this article, we have determined detailed and robust bounds on the compactification
scale M from electroweak precision data and LEP2 measurements of fermion and W pair
production. Our analysis includes correlations of experimental errors in the data. Moreover,
besides one-parameter fits, we have performed multi-parameter fits in order to include
correlations between the SM parameters and the compactification scale M. In addition,
we have estimated the sensitivity to M at a future e+ e− collider such as TESLA. For this
estimate, we have considered fermion-pair production and Higgsstrahlung.
If both the SU(2)L and U(1)Y fields are bulk fields, the existing data imply M > 5.5–
6 TeV where the range reflects the dependence on details of the Higgs sector. If the SU(2)L
or U(1)Y fields are confined to the brane where the fermions live the bounds are M > 4 TeV
and M > 5 TeV, respectively. Furthermore, we have shown that the presence of an extra
compact dimension relaxes the upper bound on the SM Higgs mass from 280 GeV in the
SM (for the data set used in Section 3) to 400 GeV and 330 GeV in the brane–bulk and the
bulk–bulk model with a brane Higgs, respectively.
At an e+ e− linear collider, the GigaZ option should √ allow to increase the sensitivity
to M by a factor 2 in almost all 5D models. At s = 800 GeV and for an integrated
luminosity of 1000 fb−1 the discovery potential will crucially depend on the control of
systematic errors. For a systematic uncertainty of 1% in each search channel, one will be
able to reach compactification scales in the range 15–20 TeV. For systematic uncertainties
smaller than the statistical uncertainties, the sensitivity limit is estimated to be in the range
M = 35–50 TeV.
Finally, for a sufficiently low compactification scale, M 10 TeV, Higgsstrahlung and

angular distributions of 2-fermion final states can be used to discriminate between different
5D models. In particular, Higgsstrahlung can be used to distinguish brane from bulk Higgs
bosons.
Acknowledgements
This work was supported in part by the Bundesministerium für Bildung und Forschung
(BMBF, Bonn, Germany) under contract 05HT1WWA2, the Studienstiftung des deutschen
Volkes, and PPARC grant number PPA/G/O/2000/00461. We wish to thank the participants
of the Heidelberg theory colloquium for an inspiring discussion and G. Quast for his help
with ZFITTER.
Appendix A. Kaluza–Klein W(0) W(0) Z(n) and W(0) W(0) γ(n) couplings
Here, we present the Feynman rules for the triple gauge boson vertices shown in Fig. 8.
+ − + −
In the gauge basis, only the W(0) W(0) Z(0) and W(0) W(0) γ(0) vertices exist while the vertices
+ − + −
W(0) W(0) Z(n) and W(0) W(0) γ(n) with n 1 are forbidden by KK selection rules [11,32].
However, in the mass eigenstate basis, couplings to heavy KK states are induced by the
diagonalization of the gauge-boson mass matrix. Below, we give these couplings to first
order in X. To this order the zero-mode couplings are unaffected, that is
γ
Z
g3(0) = g cos θW , g3(0) = e (A.1)
with g from (2.9). For the higher modes (n 1), one gets
√ ĉW 2 2 3
Z
g3(n) = 2e ŝW − ĉW
2
sβ 2 2 X, (A.2)
ŝW n π
γ √ 6
g3(n) = − 2 eĉW2 2
sβ 2 2 X (A.3)
n π
in the bulk–bulk model,
Fig. 8. Triple gauge boson couplings. The numbers in parenthesis denote the KK mode numbers.
√ 3
Z
g3(n) = 2 eĉW X (A.4)
n2 π 2
in the brane–bulk model, and
√ ĉ2 3
Z
g3(n) = − 2e W 2 2X (A.5)
ŝW n π
in the bulk–brane model. In the latter two models, the γ(n) modes for n 1 are absent.
Appendix B. Kaluza–Klein H(0) Z(0) Z(n) couplings
In the bulk–bulk model with a brane Higgs only, the H ZZ coupling can be derived in
the gauge basis from
∞ √
2
1 g2 v µ
µ
LHZZ (x) = 2
h Z(0) + 2 Z(n)
4 cW
n=1
∞ √

2
g mZ(0) Z µ µ
= 1− X h Z(0) + 2 Z(n) , (B.1)
2 cW 2
n=1
where h denotes the Higgs field on the brane and v its VEV. The second relation follows
from (2.7). In the bulk–bulk model with a bulk Higgs field, the KK selection rules forbid
the couplings of two zero modes to higher modes. Moreover, the gauge eigenstates coincide
with the mass eigenstates. Thus, for zero-mode final states, Higgsstrahlung is described by
µ
the same H(0)Z(0) Z(0) vertex as in the SM. In the brane–bulk or bulk–brane model, the Z(n)
tower coincides with the U(1)Y or SU(2)L KK modes for n 1, respectively. Because a
brane Higgs field breaks momentum conservation in the extra dimension, no selection rules
exist.
In the mass eigenstate basis, the Lagrangian (B.1) leads to the vertex shown in Fig. 9.
In summary, the effective couplings are given by
ZH
g(0) =g ZH
and g(n1) =0 (B.2)
in the bulk–bulk model with a bulk Higgs,

∆Z
g(0) = g 1 − 2 +
ZH
X ,
2

√ 3 ∆Z
g(n1) = 2 g 1 − 1 + 2 2 +
ZH
X , (B.3)
2n π 2
in the bulk–bulk model with a brane Higgs,
Fig. 9. The H ZZ vertex. The numbers in parenthesis specify the KK modes.


∆Z
ZH
g(0) = g 1 − 2ŝW
2
+ X ,
2

√ 2 3 ∆Z
ZH
g(n1) = 2 sW g 1 − ŝW2
+ 3ŝW −2 + X , (B.4)
2n2 π 2 2
in the brane–bulk model, and

∆Z
ZH
g(0) = g 1 − 2ĉW
2
+ X ,
2

√ 2 3 ∆Z
ZH
g(n1) = 2 cW g 1 − ĉW 2
+ 3ĉW −2 + X , (B.5)
2n2 π 2 2
√ ZH
in bulk–brane model. The factor 2 in g(n1) is the usual enhancement of couplings
between higher KK modes and brane fields. The factors sW and cW reflect the fact that
µ µ 3µ
for n 1 Z(n) is mainly B(n) or A(n) , respectively.
References
[1] T. Kaluza, Sitzungsber. d. Preuss. Akad. d. Wiss. Berlin (1921) 966;

O. Klein, Z. Phys. 37 (1926) 895.
[2] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
J.D. Lykken, Phys. Rev. D 54 (1996) 3693.
[3] E. Witten, Nucl. Phys. B 471 (1996) 135;
P. Hořava, E. Witten, Nucl. Phys. B 460 (1996) 506;
P. Hořava, E. Witten, Nucl. Phys. B 475 (1996) 94.
[4] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
[5] K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47.
[6] G.F. Giudice, R. Rattazzi, J.D. Wells, Nucl. Phys. B 544 (1999) 3;
T. Han, J.D. Lykken, R.-J. Zhang, Phys. Rev. D 59 (1999) 105006;
E.A. Mirabelli, M. Perelstein, M.E. Peskin, Phys. Rev. Lett. 82 (1999) 2236;
J.L. Hewett, Phys. Rev. Lett. 82 (1999) 4765;
T.G. Rizzo, Phys. Rev. D 59 (1999) 115010.
[7] L.J. Hall, D.R. Smith, Phys. Rev. D 60 (1999) 085008.
[8] T. Appelquist, H.C. Cheng, B.A. Dobrescu, Phys. Rev. D 64 (2001) 035002.
[9] K. Agashe, N.G. Deshpande, G.H. Wu, Phys. Lett. B 514 (2001) 309;
C. Macesanu, C.D. McMullen, S. Nandi, Phys. Rev. D 66 (2002) 015009;
T.G. Rizzo, Phys. Rev. D 64 (2001) 095010;
F.J. Petriello, JHEP 0205 (2002) 003;
A.J. Buras, M. Spranger, A. Weiler, Nucl. Phys. B 660 (2003) 225;
P. Bucci, B. Grzadkowski, Phys. Rev. D 68 (2003) 124002;
P. Dey, G. Bhattacharyya, hep-ph/0309110.
[10] J. Papavassiliou, A. Santamaria, Phys. Rev. D 63 (2001) 016002;
J.F. Oliver, J. Papavassiliou, A. Santamaria, Phys. Rev. D 67 (2003) 056002.
[11] A. Mück, A. Pilaftsis, R. Rückl, Phys. Rev. D 65 (2002) 085037.
[12] T. Rizzo, J. Wells, Phys. Rev. D 61 (2000) 016007;
A. Strumia, Phys. Lett. B 466 (1999) 107.
[13] Particle Data Group, K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001.
[14] B.W. Lee, C. Quigg, H.B. Thacker, Phys. Rev. D 16 (1977) 1519.
[15] J. Polchinski, Phys. Rev. Lett. 75 (1995) 4724.
[16] M. Berkooz, M.R. Douglas, R.G. Leigh, Nucl. Phys. B 480 (1996) 265.
[17] G. Aldazábal, L.E. Ibáñez, F. Quevedo, A.M. Uranga, JHEP 0008 (2000) 002;
G. Aldazábal, S. Franco, L.E. Ibáñez, R. Rabadán, A.M. Uranga, J. Math. Phys. 42 (2001) 3103;
G. Aldazábal, S. Franco, L.E. Ibáñez, R. Rabadán, A.M. Uranga, JHEP 0102 (2001) 047;
R. Blumenhagen, B. Körs, D. Lüst, T. Ott, Nucl. Phys. B 616 (2001) 3;
M. Cvetič, G. Shiu, A.M. Uranga, Phys. Rev. Lett. 87 (2001) 201801;
D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 530 (2002) 202;
L.L. Everett, G.L. Kane, S.F. King, S. Rigolin, L.T. Wang, Phys. Lett. B 531 (2002) 263;
D. Cremades, L.E. Ibáñez, F. Marchesano, JHEP 0207 (2002) 022;
G. Honecker, JHEP 0201 (2002) 025;
C. Kokorelis, JHEP 0208 (2002) 036;
J.R. Ellis, P. Kanti, D.V. Nanopoulos, Nucl. Phys. B 647 (2002) 235.
[18] H. Georgi, A.K. Grant, G. Hailu, Phys. Lett. B 506 (2001) 207;
G.V. Gersdorff, N. Irges, M. Quirós, Nucl. Phys. B 635 (2002) 127;
M. Carena, T.M.P. Tait, C.E.M. Wagner, Acta Phys. Pol. B 33 (2002) 2355.
[19] D. Bardin, et al., Z. Phys. C 44 (1989) 493;
D. Bardin, et al., Comput. Phys. Commun. 59 (1990) 303;
D. Bardin, et al., Nucl. Phys. B 351 (1991) 1;
D. Bardin, et al., Phys. Lett. B 255 (1991) 290;
D. Bardin, et al., Comput. Phys. Commun. 133 (2001) 229.
[20] F. James, M. Roos, Comput. Phys. Commun. 10 (1975) 343.
[21] G.J. Feldman, R.D. Cousins, Phys. Rev. D 57 (1998) 3873.
[22] LEP Collaborations ALEPH, DELPHI, L3, OPAL, LEP Electroweak Working Group, SLD Heavy Flavor
and Electroweak Groups, hep-ex/0112021.
[23] K. Cheung, G. Landsberg, Phys. Rev. D 65 (2002) 076003.
[24] ALEPH Collaboration, Eur. Phys. J. C 12 (2000) 183;
ALEPH Collaboration, ALEPH 99-018;
DELPHI Collaboration, Eur. Phys. J. C 11 (1999) 383;
DELPHI Collaboration, Phys. Lett. B 485 (2000) 45;
DELPHI Collaboration, DELPHI 2000-128;
DELPHI Collaboration, DELPHI 2001-094;
L3 Collaboration, Phys. Lett. B 479 (2000) 101;
OPAL Collaboration, Eur. Phys. J. C 2 (1998) 441;
OPAL Collaboration, OPAL PN 424 (2000);
OPAL Collaboration, OPAL PN 469 (2001).
[25] W. Alles, C. Boyer, A.J. Buras, Nucl. Phys. B 119 (1977) 125.
[26] Private communication with D. Bardin.
[27] TESLA Technical Design Report, DESY-2001-011, hep-ph/0106315.
[28] W. Kilian, M. Krämer, P. Zerwas, Phys. Lett. B 373 (1996) 135.
[29] A. Aranda, C. Balázs, J.L. Diaz-Cruz, Nucl. Phys. B 670 (2003) 90.
[30] S. Riemann, LC-TH-2001-007 in 2nd ECFA/DESY Study 1998–2001 (2001) 1451.
[31] D. Bourilkov, Contribution to LHC/LC Study Group Working Document, hep-ph/0305125.
[32] D. Dicus, C. McMullen, S. Nandi, Phys. Rev. D 65 (2002) 076007.
Propagators and running coupling from SU(2) lattice

gauge theory
J.C.R. Bloch a , A. Cucchieri b , K. Langfeld c , T. Mendes b
a DFG Research Center “Mathematics for Key Technologies”, c/o Weierstrass Institute for Applied Analysis
and Stochastics, Mohrenstrasse 39, D-10117 Berlin, Germany
b Instituto de Física de São Carlos, Universidade de São Paulo, C.P. 369, 13560-970 São Carlos, SP, Brazil
c Institut für Theoretische Physik, Universität Tübingen, D-72076 Tübingen, Germany
Received 24 December 2003; received in revised form 13 February 2004; accepted 18 March 2004
Abstract
We perform numerical studies of the running coupling constant αR (p2 ) and of the gluon and ghost
propagators for pure SU(2) lattice gauge theory in the minimal Landau gauge. Different definitions
of the gauge fields and different gauge-fixing procedures are used, respectively, for gaining better
control over the approach to the continuum limit and for a better understanding of Gribov-copy
effects. We find that the ghost–ghost–gluon vertex renormalization constant is finite in the continuum
limit, confirming earlier results by all-order perturbation theory. In the low momentum regime, the
gluon form factor is suppressed while the ghost form factor is divergent. Correspondingly, the ghost
propagator diverges faster than 1/p2 and the gluon propagator appears to be finite. Precision data for
the running coupling αR (p2 ) are obtained. These data are consistent with an IR fixed point given by
limp→0 αR (p2 ) = 5(1).
PACS: 11.15.Ha; 12.38.Aw; 12.38.Lg; 14.70.Dj
Keywords: Gluon propagator; Ghost propagator; Running coupling constant; SU (2) lattice gauge theory
1. Introduction
The non-perturbative study of non-Abelian gauge theories is of great importance for the
determination of infrared (IR) properties such as color confinement and hadronization.
These properties are encoded in the low momentum behavior of Yang–Mills Green’s
E-mail addresses: attilio@if.sc.usp.br (A. Cucchieri), kurt.langfeld@uni-tuebingen.de (K. Langfeld).
doi:10.1016/j.nuclphysb.2004.03.021
J.C.R. Bloch et al. / Nuclear Physics B 687 (2004) 76–100 77
functions. Derived from these Green functions, the so-called (renormalized) running
coupling constant “αR (p2 )” plays an important role for phenomenological studies and
model building. While at large momentum the running coupling decreases logarithmically
with momentum, it rapidly rises at the hadronic energy scale of several hundred MeVs
thus signaling the breakdown of the perturbative approach. Non-perturbative studies of
αR (p2 ) may be carried out analytically using the Dyson–Schwinger equations (DSEs) and
numerically through lattice simulations.
Whereas the high momentum behavior of the running coupling is uniquely determined
(to leading orders) and provided by perturbation theory, several definitions of the running
coupling in the low momentum regime are possible. All of them match with the
perturbative result at high energies. For example, the corrections to the Coulomb law of
the static-quark potential may be used to define the running coupling [1,2]. Alternatively,
the finite-size scaling has its imprint on the running coupling and may be used for a high-
precision measurement of the coupling [3,4]. The approach adopted in [5–7] is based on
extracting the running coupling directly from a vertex function. For a recent review of
lattice calculations for αR (p2 ) see [8].
In order to obtain Green’s functions for the fundamental degrees of freedom, gluons and
quarks, gauge fixing is necessary. Despite being gauge-dependent, these Green functions
play an important role for the phenomenological approach to hadron physics. Landau
gauge is a convenient choice for gauge fixing for several reasons. First of all, it is a
Lorentz-covariant gauge implying that 2-point functions only depend on the square of
the momentum transfer. Secondly, the renormalization procedure is simplified since the
ghost–ghost–gluon vertex renormalization constant Z̃1 is finite, at least to all orders of
perturbation theory. This result—obtained by Taylor [9]—is a particular feature of Landau
gauge and allows another definition of the running coupling constant, which only requires
the calculation of 2-point functions: let FR (p2 , µ2 ) and JR (p2 , µ2 ) denote the form factors
(for a renormalization point µ) of the gluon and the ghost propagator, respectively; the
running coupling is then defined by

αR p2 = αR µ2 FR p2 , µ2 JR2 p2 , µ2 (1)
(see Section 2.7 below).
Due to its usefulness for the description of the physics of hadrons, the non-perturba-
tive approach to low-energy Yang–Mills theory by means of the DSEs has attracted
much interest over the last decade [10,11]. The coupled set of continuum DSEs for the
renormalized gluon and ghost propagators in Landau gauge has been recently studied by
several groups [12–25]. In all cases it was found that the gluon and ghost form factors
satisfy simple scaling laws in the IR momentum range p 1 GeV
α β
FR p2 , µ2 ∝ p2 , JR p2 , µ2 ∝ p2 , (2)
where the remarkable sum rule holds for the IR exponents α and β:
α + 2β = 0. (3)
It is interesting that this result is rather independent of the truncation scheme under
consideration. These exponents may be determined from lattice simulations, assuming the
78 J.C.R. Bloch et al. / Nuclear Physics B 687 (2004) 76–100
parameterization α = 2κ and β = −κ. Note that for κ > 0 this implies a divergent ghost
form factor JR (p2 , µ2 ) in the IR limit and a vanishing gluon form factor FR (p2 , µ2 ) in
the same limit. Also, since the gluon propagator is given by D(p2 ) = F (p2 )/p2 , one gets
that D(0) is infinite or finite, respectively, if κ < 0.5 or κ 0.5. In the second case one has
D(0) = 0 for κ > 0.5 and D(0) finite and non-zero for κ = 0.5. The IR sum rule (3) also
implies that the running coupling (defined in Eq. (61) below) develops a fixed point in the
IR limit

lim αR p2 = αc = const. (4)
p→0
Note that this result is independent of the value of κ as long as the IR sum rule (3) is
satisfied.
The precise value of κ as well as the fixed-point value αc depend strongly on
the truncation of the Dyson–Schwinger tower of equations. In fact, depending on the
truncation, one finds 0.3 < κ < 1 in the four-dimensional case [12,14–25]. These studies
vary in their vertex ansatzes, angular approximations of the momentum loop integral, and
on the tensor structure considered. In Ref. [17] a new class of truncation schemes has been
introduced, which manifestly ensures the multiplicative renormalizability of the propagator
solutions. In this truncation, although the exact values of κ and αc depend on the details of
the truncation of the DSE tower, the value of αc is constrained to
2π 8π
< αc < (5)
Nc Nc
for SU(Nc ).
An IR-finite gluon propagator [26–31] and an IR-divergent ghost form factor [30,31]
are also obtained using numerical simulations in the minimal Landau gauge. The present
numerical data for the ghost propagator indicate a value of κ = −β smaller than 0.5, while
for the gluon propagator it is still under debate if κ = α/2 is equal to or larger than 0.5. In
both cases large finite-size effects in the IR region make an exact determination of these
exponents difficult. This is particularly evident in the gluon propagator case [26,27,29,32,
33], where one needs to go to very large lattices in order to have control over the infinite-
volume extrapolation. Our present data are consistent with κ of the order of 0.5 [34,35].
Let us stress that in the minimal Landau gauge, which is the gauge-fixing condition
used in numerical simulations (see Section 2.3 below), the gauge-fixed configurations
belong to the region of transverse configurations, for which the Faddeev–Popov operator
is non-negative. This implies a rigorous inequality [36–38] for the Fourier components of
the gluon field and a strong suppression of the (unrenormalized) gluon propagator in the
IR limit. At the same time, the Euclidean probability gets concentrated near the border
of this region, the so-called first Gribov horizon, implying the enhancement of the ghost
propagator at small momenta [38]. A similar result was also obtained by Gribov in [39].
Taylor’s finding is based upon the Faddeev–Popov quantization, thereby ignoring the
effect of Gribov copies, which are certainly present in an intrinsically non-perturbative
approach. The goal of the present paper is to confirm Taylor’s result (Z̃1 is finite to
all orders of perturbation theory) by the non-perturbative approach provided by lattice
simulations. In addition, a thorough study of the gluon and the ghost form factors is
performed. A focal point is the IR limit of the running coupling constant. Support for the
existence of the fixed point is found, and a first estimate of αc is provided from extensive
lattice simulations. We present two sets of simulations, carried out, respectively, in São
Carlos and in Tübingen, employing different definitions of the gauge fields and different
gauge-fixing procedures. We believe that the comparison of these two formulations
strengthens the significance of our findings.
The paper is organized as follows. In Section 2 we describe the lattice approach to
Yang–Mills Green’s functions. Section 3 contains the numerical setup for the simulations
carried out in São Carlos and in Tübingen. In Section 4 we report our data for the gluon
and ghost propagators and for the running coupling constant. Conclusions are left to the
final section.
Preliminary results have been presented in [34,35,40,41].
2. The lattice approach to Green’s functions
In this section we explain the two lattice setups used for the numerical evaluation of the
gluon and ghost propagators. We also recall the definition of the running coupling constant
considered in Ref. [12], which can be evaluated using these propagators.
2.1. Gluon field on the lattice
The action S of the continuum SU(2) Yang–Mills theory is formulated in terms of the
field strength
a
Fµν [A](x) = ∂µ Aaν (x) − ∂ν Aaµ (x) + g0 abc Abµ (x)Acν (x) (6)
and is given by

1
S= a
d 4 x Fµν [A](x)Fµν
a
[A](x). (7)
4
Here g0 is the bare coupling constant and Aaµ (x) is the continuum gauge field.
On the lattice the dynamical fields are SU(2) matrices Uµ (x), which are associated with
the links of the lattice, and the Wilson action is given by
1
S =β 1 − trc Pµν (x), (8)
x,µ>ν
2
where the plaquette is defined as
Pµν (x) = Uµ (x)Uν (x + eµ )Uµ† (x + eν )Uν† (x) (9)
and eµ is a unit vector in the positive µ direction. (Note that the trace extends over color
indices only.) The Wilson action is invariant under the gauge transformation
UµΩ (x) = Ω(x)Uµ (x)Ω †(x + eµ ), (10)
where Ω(x) are SU(2) matrices. The link variables Uµ (x) may be expressed in terms of
the continuum gauge field Aµ (x) by making use of the relation

Uµ (x) = exp iag0 Abµ (x)t b , (11)
where a is the lattice spacing, t b = σ b /2 are the generators of the SU(2) algebra and σ b
are the Pauli matrices. One can check that in the naive continuum limit a → 0 the Wilson
action reproduces the continuum action in Eq. (7) if
4 1
β= 2
= , (12)
g0 πα 0
where α0 is the bare coupling constant (squared).

For the gauge group SU(2), the link variables Uµ (x) can be given in terms of (real)
four-vectors of unit length
0 2 2
Uµ (x) = u0µ (x)1 + i uµ (x) · σ , uµ (x) + uµ (x) = 1, (13)
where 1 is a 2 × 2 identity matrix. By defining the lattice gluon field Abµ (x) as
Uµ (x) − Uµ† (x)

Abµ (x) = (14)
2i
one obtains

Abµ (x) = 2ubµ (x) = ag0 Abµ (x) + O a 3 (15)
in the naive continuum limit a → 0. This definition has been used for the numerical
simulations done in São Carlos.
Note that the gluon field Abµ (x) defined above changes sign under a non-trivial center
transformation Z2 of the link fields Uµ (x) → −Uµ (x). Recently, another identification of
the gluonic degrees of freedom in the lattice formulation was proposed [28]. In this case,
one first notices that the gluon field in continuum Yang–Mills theories transforms under
the adjoint representation of the SU(2) color group, i.e.,
aed ec
Aa
µ (x) = O ab
(x)A b
µ (x) + O (x)∂µ O dc , (16)
2

O ab (x) = 2 trc Ω(x)t a Ω † (x)t b , (17)
where Ω(x) ∈ SU(2) is a gauge transformation of the fundamental quark field and
O ab (x) ∈ SO(3). In view of the transformation properties in Eq. (16), one can identify the
continuum gauge fields Aaµ (x) with the algebra-valued fields of the adjoint representation
cd
Uµcd (x) = exp ag0 Abµ (x)tˆb , (18)
where tˆac
b = abc and the total anti-symmetric tensor abc is the generator of the SU(2)
group in the adjoint representation. On the lattice, the adjoint links Uµab (x) are obtained
from

Uµcd (x) = 2 trc Uµ (x)t c Uµ† (x)t d (19)
and the gluon field Aaµ (x) is given by
Abµ (x) = 2u0µ (x)ubµ (x), (20)

without summation over µ on the right-hand side. By expanding Eq. (18) in powers of the
lattice spacing a and by using Eqs. (13) and (19) one obtains

Abµ (x) = ag0 Abµ (x) + O a 3 . (21)
Clearly, the representation (20) is invariant under a non-trivial center transformation
Uµ (x) → −Uµ (x). This discretization of the gluon field has been used for the simulations
done in Tübingen.
It is well known that different discretizations of the gluon field lead to gluon
propagators equivalent up to a trivial (multiplicative) renormalization [42–45]. Also, this
proportionality constant between different discretizations of the gluon propagator may be
(partially) explained as a tadpole renormalization [46,47]. We point out, however, that it is
useful to disentangle the information carried by center elements and coset fields, defined
above, when the vacuum energy is investigated. In particular, it was found that—in the
continuum limit—the center elements provide a contribution to the gluon condensate [48,
49].
2.2. Tadpole improved gluon fields
The relation between lattice and continuum gluon fields relies on the expansion (see
Eq. (11))
Uµ (x) = 1 + iag0 Abµ (x)t b + · · · , (22)
where the ellipses denote higher order terms in the bare coupling constant g0 . The artificial
contributions of these higher order terms to loop integrals are called “tadpole” terms [50].
These terms are only suppressed by powers of g02 and are generically large in simulations
using moderate β values.
In order to remove the tadpole contributions from the observable of interest one can
redefine the relation between the link matrices and the continuum gluon field by using

Uµ (x) = u0,L 1 + iag0 Abµ (x)t b + · · · , (23)
where u0,L is given by the “meanfield” value

trc
u0,L = Uµ (x) (for arbitrary µ) (24)
2
with the links Uµ (x) fixed to the Landau gauge. Equivalently, one can use [50] a gauge-
invariant definition of the tadpole factor given by

1/4
trc
u0,P = Pµν (x) , (25)
2
where Pµν (x) is the plaquette defined in Eq. (9). Thus, the use of tadpole improvement in
the case of the standard definition of the lattice gluon field (15) gives

Abµ (x) = ag0 Abµ (x) + O a 3 = 2ubµ (x)/u0,P . (26)
In the case of the coset definition (20) of the gauge fields, the tadpole factors are
expressed in terms of the expectation values of the adjoint link and of the adjoint plaquette
and are given by

trc
u0,L =
ad
Uµ (x) , (27)
3

1 1/4 1 2 1/4
uad
0,P = tr P
c µν (x) = tr P
c µν (x) − 1 . (28)
3 3
Thus, the tadpole improved relation between the continuum gauge field Abµ (x) and the link
matrices is given in this case by

Abµ (x) = ag0 Abµ (x) + O a 3 = 2u0µ (x)ubµ (x)/uad
0,P . (29)
In the sections below we will stress the effect of tadpole improvement on the various
quantities considered in this work.
2.3. Minimal Landau gauge
The gluon and ghost propagators depend on the choice of the gauge. In order to maintain
contact with the Dyson–Schwinger approach and the results presented in the Introduction,
we consider the so-called minimal (lattice) Landau gauge. This gauge condition is imposed
by minimizing the functional

Sfix [Ω] = − trc UµΩ (x), (30)
x,µ
where UµΩ (x) is the gauge-transformed link (10). This minimizing condition corresponds
to imposing the transversality condition

(
· A)b (x) = Abµ (x) − Abµ (x − eµ ) = 0 ∀b and x, (31)
µ
which is the lattice formulation of the usual Landau gauge-fixing condition in the
continuum. Let us notice that the condition (31) is exactly satisfied by the lattice gauge field
only if the standard discretization (15) is considered, while for the discretization given in
(20) the above result is valid up to discretization errors of order a 2 . However, in both cases,
the gauge-fixing condition (30) implies that the continuum Landau gauge-fixing condition
∂ · A = 0 is satisfied up to discretization errors of order a 2 . In practice, we stop the gauge
fixing when the average value of [(
· A)b (x)]2 is smaller than 10−12 .
The minimizing condition (30) also implies that the Faddeev–Popov matrix is positive
semi-definite. In particular, the space of gauge-fixed configurations {UµΩ (x)} lies within
the first Gribov horizon, where the smallest (non-trivial) eigenvalue of the Faddeev–Popov
operator is zero. It is well known that, in general, for a given lattice configuration {Uµ (x)},
there are many possible gauge transformations Ω(x) that correspond to different local
minima of the functional (30), i.e., there are Gribov copies inside the first Gribov horizon
[36,37]. Thus, the minimizing condition given in Eq. (30) is not sufficient to find a unique
representative on each gauge orbit. A possible solution to this problem is to restrict the
configuration space of gauge-fixed fields UµΩ (x) to the so-called fundamental modular
region [38], i.e., to consider for each configuration {Uµ (x)} the absolute minimum of
the functional (30). From the numerical point of view this is a highly non-trivial task,
corresponding to finding the ground state of a spin-glass model [51]. On the other hand, if
local minima are considered, one faces the problem that different numerical gauge-fixing
algorithms yield different sets of local minima, i.e., they sample different configurations
from the region delimited by the first Gribov horizon (see, for example, [44] and references
therein). This implies that numerical results using gauge fixing could depend on the gauge-
fixing algorithm, making their interpretation conceptually difficult.
In the simulations done in São Carlos, a stochastic overrelaxation algorithm [52–54] was
used. The simulations done in Tübingen have employed the simulated annealing technique
described in detail in Ref. [30]. The problem of Gribov copies was not considered in either
case. Even though neither method is able to locate the global minimum of the gauge-fixing
functional, i.e., to restrict the gauge-fixed configuration space to the fundamental modular
region, a comparison of the propagators obtained using the two methods can provide an
estimate of the bias (Gribov noise) introduced by the gauge-fixing procedure. From this
comparison we have found that the data for the propagators are rather insensitive to the
particular choice of gauge-fixing algorithm, suggesting that the influence of Gribov copies
on the two propagators (if present) is at most of the order of magnitude of the numerical
accuracy. For the gluon propagator this result is in agreement with previous studies in
Landau gauge for the SU(2) and SU(3) groups in three [33] and four dimensions [31,44,
55]. A similar result has also been obtained for the gluon propagator in Coulomb gauge
[56]. On the other hand, a previous study of the ghost propagator in SU(2) Landau gauge
[31] has shown a clear bias related to Gribov copies in the strong-coupling regime. In
particular, data (in the IR region) obtained considering only absolute minima have been
found to be systematically smaller than data obtained using local minima. This result—
which has been recently confirmed in [57]—can be qualitatively explained. In fact, as said
above, the smallest non-trivial eigenvalue λmin of the Faddeev–Popov operator goes to
zero as the first Gribov horizon is approached. At the same time, one expects that global
minima (i.e., configurations belonging to the fundamental modular region) be “farther”
away from the first Gribov horizon than local minima. Thus, the absolute minimum
configuration should correspond to a value of λmin larger—on average—than the value
obtained in a generic relative minimum.1 Since the ghost propagator is given by the
inverse of the Faddeev–Popov matrix (see Section 2.5 below), this would imply a smaller
ghost propagator (on average) at the absolute minimum, as observed in Refs. [31,57]. The
analysis carried out in these references has shown that Gribov-copy effects are visible
only for the smallest non-zero momentum on the lattice, at least for the lattice volumes
considered, which are still relatively small. As explained below (see Section 4.3), in our
analysis we have not considered the data points corresponding to the smallest momenta.
2.4. Gluon propagator
The continuum gluon propagator in position space is given by

Dµνab
(x − y) = Aaµ (x)Abν (y) . (32)
1 This was checked numerically in Ref. [58].

Correspondingly, one can consider the (position space) lattice gluon propagator

Dµν
ab
(x − y) = Aaµ (x)Abν (y) , (33)
where Aaµ (x) is one of the lattice discretizations of the continuum gluon field discussed
above (see Eqs. (15) and (20)). At the leading order a these two quantities are related by
ab
g02 a 2 Dµν (x − y) = Dµν
ab
(x − y)/u20,P , (34)
where u0,P is the tadpole factor given in Eq. (25) (respectively, Eq. (28)) when considering
the lattice gluon field defined in Eq. (15) (respectively, Eq. (20)).
The lattice gluon propagator in momentum space is obtained by evaluating the Fourier
transform
1 ab 2πnµ
Dµν
ab
(p̂) = Dµν (x − y) exp i p̂ · (x̂ − ŷ) , p̂µ = , (35)
V x,y Nµ
where nµ labels the Matsubara modes in the µ direction, Nµ is the number of lattice points
in the same direction, x = x̂a, y = ŷa and V is the lattice volume. In order to minimize
discretization effects [59], we consider the gluon propagator as a function of the lattice
momentum p with components

p̂µ
pµ = 2 sin . (36)
2
It is also useful to introduce the lattice gluon form factor F (p̂2 ), defined as
F (p̂2 ) 1 aa
D(p̂) = , D(p̂) = D (p̂), (37)
p2 9 a,µ µµ
which is a measure of the deviation of the full propagator from the free one. Note that, in
Landau gauge, the propagator is diagonal in color space and transversal in Lorentz space.
The transversality condition (31) implies [31] that D(0) is not
given by D(p̂) at p̂ = 0. In
fact, for p̂ = 0 the previous equation becomes D(0) = (1/12) a,µ Dµµ aa (0).
In order to evaluate numerically the gluon propagator in momentum space it is useful to

employ the formula [31]
2 2
1 a
D(p̂) = Aµ (x) cos (p̂ · x̂) + Aµ (x) sin (p̂ · x̂) .
a
(38)
9V a,µ x x
In fact, by expanding the previous equation we obtain

1 a
D(p̂) = Aµ (x)Aaµ (y) cos p̂ · (x̂ − ŷ) , (39)
9V a,µ x,y
which is directly related to Eq. (35).

One can also evaluate the form factor F (p̂2 ) directly [28]. To this end we can consider,
without any loss of generality, a momentum transfer parallel to the time direction p̂ =
(0, 0, 0, p̂4) and define

t Aµ (x) = Aµ (x + e4 ) − Aµ (x), (40)
where e4 is the unit vector in the time direction. The form factor is then obtained from
2 2
1
F p̂2 =
t Aaµ (x) cos (p̂ · x̂) +
t Aaµ (x) sin (p̂ · x̂) . (41)
9V a,µ x x
By expanding the previous formula one can verify that the free part 1/p2 is canceled
exactly. This strongly suppresses the statistical noise in the high-momentum regime. Here,
we will present results that directly address the gluon propagator (38) and the gluon form
factor (41), evaluated, respectively, in São Carlos and in Tübingen.
2.5. Ghost propagator
The ghost propagator Gab (p̂) is uniquely defined once the gauge-fixing functional (30)
is specified. In fact, if we write the gauge-fixing matrix as

Ω(x) = exp iθ a (x)t a , (42)
with t a defined as in Section 2.1, then the gauge-fixing functional can be expanded with
respect to the angles θ a (x) and, at any local minimum of Sfix , we obtain
1 a
Sfix = S0 + θ (x)Mxy θ (y) + O θ 3 ,
ab b
(43)
2 x,y
a,b
where ab
Mxy is the so-called Faddeev–Popov operator. Note that the linear term in θ (x)
is absent by virtue of the minimizing gauge-fixing condition (see Eqs. (30) and (31)).
The expression of the Faddeev–Popov operator in terms of the gauge-fixed link variables
can be found in [38, Eq. (B.18)]. Note that the matrix Mxy ab obtained in this way is a
lattice discretization of the continuum Faddeev–Popov operator (−∂ + A) · ∂ and that this
discretization yields automatically the standard discretization Abµ (x) for the gluon field
given in Eq. (15).
The lattice ghost propagator G ab (p̂) is provided by the inverse Faddeev–Popov operator
ab
Mxy . Due to translation invariance, the lattice average of the inverse operator depends only
on (x − y). Thus, in momentum space we have
1 −1 ab
G ab (p̂) = M xy
exp −i p̂ · (x̂ − ŷ) . (44)
V x,y
Since the matrix Mxyab depends linearly on the link variables U (x), tadpole improvement
µ
applied to the ghost propagator implies a rescaling
G ab (p) → G ab (p̂)u0,P . (45)
Thus, at the leading order a one has
a 2 Gab (p) = G ab (p̂)u0,P , (46)
where Gab (p) is the continuum ghost propagator in momentum space.
The asymptotic behavior of the ghost propagator Gab (p̂) is known from perturbation
theory: it decreases as 1/p2 with additional logarithmic corrections. The 1/p2 behavior is
inherited from the free-theory case. The non-trivial information on the ghost propagator is
therefore encoded in the (continuum) form factor J (p̂2 ), which is defined by
J (p̂2 )
Gab (p̂) = δ ab G(p̂) = δ ab , (47)
p2
yielding the lattice form
J (p̂2 )
G ab (p̂) = δ ab . (48)
p2
Numerically, the lattice ghost propagator can be obtained by inverting the Faddeev–Popov
matrix Mxy ab . In the numerical simulations carried out in São Carlos this has been done
using a conjugate-gradient algorithm. On the contrary, in the simulations in Tübingen the

(lattice) ghost form factor J (p̂2 ) has been evaluated directly. To this end one can consider
the following set of linear equations (for a given set of link variables U )

Mxyab
[U ]ūb (y) = na cos p̂ · (x̂ − eµ ) − cos[p̂ · x̂] , (49)
y,b

ab
Mxy [U ]v̄ b (y) = na sin p̂ · (x̂ − eµ ) − sin[p̂ · x̂] , (50)
y,b
where na is an arbitrary unit vector that specifies the components of the ghost propagator
under investigation. We are considering momenta with a non-zero component only in the µ
direction. In fact, by solving these equations for ūb (y) and v̄ b (y) and using trigonometric
identities we find that the ghost form factor is given by
1
J p̂2 = cos p̂ · (ŷ − eµ ) − cos[p̂ · ŷ] nb ūb (y)
V y

+ sin p̂ · (ŷ − eµ ) − sin[p̂ · ŷ] nb v̄ b (y) (51)

1 b −1 ab a p̂µ
= n M xy
n 4 sin2 cos p̂ · (x̂ − ŷ) . (52)
V x,y 2
a,b
Note that, with our choice of momenta, the lattice momentum squared is given by p2 =
4 sin2 (p̂µ /2), implying that the free part 1/p2 of the ghost propagator exactly cancels out
and we are left with the form factor J (p̂2 ). The set of equations (49)–(50) has been solved
using a bi-conjugate gradient method for matrix inversion.
2.6. Renormalization
Renormalization of Yang–Mills theories in four dimensions implies that the bare

coupling acquires a dependence on the ultraviolet (UV) cutoff ΛUV given by
α0 → α0 (ΛUV /Λscale ). (53)
Thereby, the bare coupling constant is no longer the theory’s parameter. The Yang–Mills
scale parameter Λscale takes over the role of the only parameter of the theory. In the context
of (quenched) lattice-gauge-theory simulations of the string tension σ is widely used as

the generic low-energy scale. In this case, the cutoff dependence of the bare coupling is
implicitly given by the β dependence of σ a 2 (β) where β is related to the bare coupling
in (12) and ΛUV = π/a(β). In Section 3.3 we will derive this relation from lattice data
(obtained in Ref. [60]).
In addition, wave-function renormalization constants develop a dependence on ΛUV /
Λscale . The lattice bare form factors of the previous subsections, FB and JB , are related to
their continuum analogues (for very large β) by
β
FB = FB , JB = JB u0,P . (54)
4u20,P
These form factors depend on the momentum p2 and on the UV cutoff ΛUV (given in units
of the string tension) or, equivalently, on the lattice coupling β (see Section 4.3 below).
Thus, we can write

FB = FB p 2 , β , JB = JB p2 , β . (55)
The renormalized form factors are obtained upon multiplicative renormalization

FR p2 , µ2 = Z3−1 (β, µ)FB p2 , β , (56)

JR p2 , µ2 = Z̃3−1 (β, µ)JB p2 , β , (57)
using the renormalization conditions

FR µ2 , µ2 = 1, JR µ2 , µ2 = 1. (58)
(Notice that tadpole renormalization does not affect the calculation of FR , JR but may be
useful for the determination of the renormalization constants Z3 , Z̃3 .)
Clearly, similar relations hold also for the bare and renormalized gluon and ghost
propagators. In practice, the multiplicative renormalizability of the theory implies that a
rescaling of the data for each β value (independently of the lattice momentum) is sufficient
to let the form factors FB (p2 , β) and JB (p2 , β)—or equivalently the corresponding
propagators—fall on top of a single curve describing the momentum dependence of the
corresponding renormalized quantity.
2.7. Running coupling constant
Of great importance for phenomenological purposes is the running coupling strength

αR (p2 ) considered in Ref. [12]. In particular, this strength enters directly the quark DSE
and can be interpreted as an effective interaction strength between quarks [61]. This
running coupling strength is a renormalization-group-invariant combination of the gluon
and ghost form factors. In order to derive this combination we can start with the definition
of the ghost–ghost–gluon vertex renormalized coupling strength
Z3 (β, µ)Z̃32 (β, µ)
αR µ2 = α0 (ΛUV ), (59)
Z̃12 (β, µ)
where Z̃1 (β, µ) is the ghost–ghost–gluon vertex renormalization constant. In lattice

simulations, the UV-cutoff is related to β by ΛUV = π/a(β), where a is the lattice spacing.
Using Eqs. (56) and (57) we can express the renormalization constants Z3 (β, µ) and
Z̃3 (β, µ) in terms of the bare and renormalized form factors yielding
α0 (ΛUV ) 2 2 2
αR µ2 FR p2 , µ2 JR2 p2 , µ2 = 2 FB p , β JB p , β . (60)
Z̃1 (β, µ)
Note that the left-hand side of this relation is finite and independent of β by construction
and that the right-hand side depends on the renormalization scale µ only through the ghost–
ghost–gluon vertex renormalization constant Z̃1 (β, µ). It was found more than twenty
years ago by Taylor [9] that (in the continuum) Z̃1 is finite, independent of µ, at least to all
orders of perturbation theory. This finding will be confirmed by our lattice studies below.
case, the right-hand side of Eq. (60) is thus independent of µ. Then, by choosing
In this
µ = p2 and using the renormalization conditions (58), we find the final expression for
the running coupling strength, i.e.,

αR p2 = αR µ2 FR p2 , µ2 JR2 p2 , µ2 . (61)
Finally, let us notice that Eqs. (34) and (46) imply that tadpole renormalization does not
affect the renormalized coupling defined above.
3. Details of the numerical simulations
3.1. Setup
All our simulations used the standard Wilson action for SU(2) lattice gauge theory in
four dimensions with periodic boundary conditions. In order to check finite-volume effects
and verify scaling we consider several values of β and of the lattice volumes V = Ns3 × Nt .
The dependence of the lattice spacing on β can be inferred from a calculation of the string
tension σ in lattice units. Here, we will use the data for σ a 2 reported in Ref. [60] and a
linear interpolation of the logarithm of these data where needed.2 A value of the lattice
spacing in physical units was obtained using the value σ = [440 MeV]2 for the string
tension. We used Nconf independent configurations for the numerical evaluation of the
propagators.
Computations in São Carlos were performed on the PC cluster at the IFSC-USP (the
system has 16 nodes with 866 MHz Pentium III CPU and 256 MB RAM memory). All
runs in São Carlos started with a random gauge configuration and for thermalization we
use a hybrid overrelaxed (HOR) algorithm. The total computer time used for the runs
was about 50 days on the full PC cluster. In Table 1 we report the parameters used for the
simulations in São Carlos. For each β value three different lattice volumes were considered,
i.e., V = 144 , 204 and 264 . For the lattice volume V = 144 (respectively, V = 204 and 264 )
and for each β value we produced Nconf = 500 (respectively, 150 and 50) configurations.
2 For β = 2.15 an extrapolation of these data was necessary in order to obtain σ a 2 .

Table 1
Simulation parameters of the runs in São Carlos. Data from Ref. [60] were used to obtain the lattice spacing in
units of the string tension. Error bars (in parentheses) come from propagation of errors and indicate one standard
deviation on the last significant digit. The lattice volumes V and the number of configurations Nconf considered
are discussed in the text
β 2.2 2.3 2.4 2.5 2.6 2.7 2.8
σ a2 0.220(9) 0.136(2) 0.071(1) 0.0363(3) 0.018(1) 0.0103(2) 0.0055(3)
Table 2
Simulation parameters of the runs in Tübingen. Data from Ref. [60] were used to obtain the lattice spacing in
units of the string tension
β 2.15 2.2 2.3 2.375 2.45 2.525
σ a2 0.280(13) 0.220(9) 0.136(2) 0.083(2) 0.0507(8) 0.0307(5)
Ns3 Nt 163 × 32 163 × 32 163 × 32 163 × 32 163 × 32 163 × 32
Nconf 200 200 200 200 200 200
Table 3
Differences between the São Carlos and the Tübingen numerical approach
São Carlos Tübingen
Definition of gauge fields fundamental representation adjoint representation
Gauge fixing iterative stoch. overrelaxation simulated annealing
Number of lattice points finite-size control fixed
Computations in Tübingen were carried out at the local PC cluster where 4–12 nodes (1
Ghz Athlon) were used. The simulation parameters for the runs in Tübingen are listed in
Table 2.
Table 3 lists the differences between the São Carlos and the Tübingen approach.
Reconstructing the continuum gauge field from the link fields in different manners
(compare Eq. (15) with Eq. (20)) provides insight into the discretization errors. Employing
different gauge-fixing algorithms points out the effect of the Gribov ambiguities on the
propagators.
3.2. Determination of renormalization constants
In order to obtain the renormalized propagators and form factors, one needs to evaluate
the renormalization constants Z3−1 (β, µ) and Z̃3−1 (β, µ) defined in Eqs. (56) and (57),
respectively. Multiplicative renormalizability implies that one can “collapse” data obtained
at different β on a single curve. This can be done by using the matching technique
described in detail in Ref. [62, Section V.B.2], For instance, for the gluon form factor
this is equivalent to considering the quantity

FR p2 , µ2 = Z3−1 (β, µ)FB β, p2 , (62)
where the factor Z3−1 (β, µ) for each β is obtained from the matching technique. The
renormalization point, i.e., the µ-dependence, comes into play when the “single” curve
is rescaled to satisfy the condition

FR µ2 , µ2 = 1. (63)
The same procedure is applied to the ghost form factor. For our analysis we considered a
renormalization scale of µ = 3 GeV.
For the São Carlos data we checked for finite-size effects before applying the matching
technique.3 In particular, by comparing data at different lattice sizes and same β value,
we find (for each β) a range of momenta for which the data are free from finite-volume
corrections. We then perform the matching using data for these momenta and V = 264 .
3.3. Asymptotic scaling
As said in Section 2.6, the cutoff dependence of the bare coupling is implicitly given by
σ a 2 (β). Thereby, the string tension σ serves as the fundamental energy scale.
In principle,
√ perturbation theory predicts the β dependence of σ a 2 for the regime
a 1/ σ . In practice, large deviations of the measured function σ a 2 (β) from the
perturbative scaling are observed in the scaling region, which corresponds to the interval
β ∈ [2.15, 2.8] in our case. Clearly, the relation between the “measured” values for σ a 2
and perturbation theory is highly important for a careful extrapolation of the lattice data to
continuum physics. One goal of the present paper is to present this relation.
For this purpose, we perform a large β expansion of the lattice spacing in units of the
string tension, i.e.,
2
4π 2 2β1 4π 4π 2 d
ln σ a 2 = − β + 2 ln β + + c. (64)
β0 β0 β0 β0 β
The first two terms on the rhs of (64) are in accordance with 2-loop perturbation theory.
The term d/β represents higher order effects and the term c is a dimensionless scale factor
to the string tension. Parameters c and d are determined by fitting the formula (64) to the
lattice data reported in Ref. [60]. Using only data for β 2.3 we obtain
c = 4.38(9), d = 1.66(4), χ 2 /d.o.f. = 0.62. (65)
The corresponding fit is shown in Fig. 1. It appears that the truncation of the series (64) at
the 1/β level reproduces the measured values to high accuracy.
In order to illustrate the impact of the d-term correction in (64) on the estimate of Yang–
Mills scale parameters, we briefly consider the lattice scale parameter Λlat . This parameter
is implicitly defined at the 2-loop level by considering [50]

−1 β0 1 β1 1
αlat = ln + ln ln , (66)
4π a 2 Λ2lat 2πβ0 a 2 Λ2lat
where for the SU(N) gauge group
11 17 2 −1 2π
β0 = N, β1 = N , αlat = β. (67)
3 3 N
3 For more details see [33, Section III].

Fig. 1. The string tension in units of the lattice spacing: lattice data from [60] and the fit using Eq. (64).
Inverting Eq. (66) consistently up to 2-loop perturbation theory yields

· 4π −1 2β1 4π −1
ln a 2 Λ2lat = − αlat + 2 ln αlat . (68)
β0 β0 β0
−1
Then, using αlat = πβ from Eq. (67) and eliminating a by subtracting Eq. (64) from the
latter equation we find
2
Λlat 4π 2 d
ln = lim −c − . (69)
σ β→∞ β0 β
Thus, if we extrapolate to the continuum limit β → ∞ we obtain
√ √
Λlat = e−c/2 σ = 0.112(5) σ . (70)
Using the value σ = [440 MeV]2 one gets Λlat = 49(2) MeV. If one instead of the limit
√ the asymptotic scaling regime is reached for, e.g., β = 2.5, one gets
in (69) assumes that
Λlat = 0.0188(8) σ . This is the order of magnitude familiar from the literature. Hence, for
a scaling analysis of lattice results employing β ∈ [2.15, 2.8] the irrelevant term of order
1/β is still important.
4. Results
4.1. Renormalization constants Z3 , Z̃3 and Z̃1
Let us firstly focus on the renormalization constants Z3 and Z̃3 . As outlined in

Section 3.2, these constants are obtained from “matching” the lattice data from simulations
Fig. 2. The gluon and ghost renormalization constants, Z3 (β, µ) and Z̃3 (β, µ), for µ = 3 GeV (left panel), y axis
is arbitrarily scaled. The cutoff dependence of Z3 Z̃32 (right panel) is consistent with a finite Z̃1 (see Eqs. (74),
(75)). Figures corresponding to the Tübingen data only.
Fig. 3. The analogue of Fig. 2, for the São Carlos data. The fits shown on the left-hand side for the gluon (upper
curve) and ghost (lower curve) renormalization constants are done with γ as a free parameter. All fits neglect the
leftmost data point.
using different β values. Figs. 2 and 3 show the cutoff dependence of these constants,
respectively, for the Tübingen and São Carlos sets of data.
Using the results above, we can check that our data are consistent with the predictions
from perturbation theory. For large enough UV cutoff, one expects that the 1-loop behavior
is recovered, i.e.,
γ
Z3 (β, µ), Z̃3 (β, µ) ≈ b − ln σ a 2 + ω . (71)
Table 4
Expectation value of the adjoint plaquette (28) used for the tadpole improvement of the gluon fields derived from
the adjoint representation
β 2.15 2.2 2.3 2.375 2.45 2.525
uad
0,P 0.225(42) 0.241(43) 0.274(45) 0.297(46) 0.317(46) 0.336(47)
If only small lattice spacings are considered, ω is related to the ratio between the string
tension and the Yang–Mills scale parameter at 1-loop level
π 2σ
ω = ln . (72)
Λ21-loop
Here, we treat ω as a fit parameter and explore a range of lattice spacings where one would
already expect significant deviations from the 1-loop behavior. As shown in Figs. 2 and 3
(left panel), a good consistency with the known anomalous dimensions is observed.
For the Tübingen data, we find that ω ≈ 1.13 is a good choice for reproducing the
data for Z3 and Z̃3 simultaneously. It turns out that within the β range explored in the
Tübingen runs, tadpole improvement has a minor effect on the anomalous dimension. The
adjoint plaquette used for the tadpole improvement (see Eq. (34)) employing the adjoint
representation is listed in Table 4.
For the São Carlos data we have performed the fits with γ as a free parameter, leaving
out the data point with β = 2.2 (i.e., the leftmost point in Fig. 3). We obtain the following
values for the gluon and ghost cases
γgluon = 0.60(5) and γghost = 0.32(7). (73)

We see that the values are, respectively, consistent with 13/22 ≈ 0.59 and 9/44 ≈ 0.20
within error bars (but notice that there is a discrepancy of almost two standard deviations
for γghost ). In this case we have not succeeded in finding a value of ω describing the
behaviors for Z3 and Z̃3 simultaneously.
In order to interpret the product of ghost and gluon form factors as the running coupling
strength (see Section 2.7), it is of great importance that the ghost–ghost–gluon vertex
renormalization constant Z̃1 be finite in the continuum limit. For detecting the UV behavior
of Z̃1 , let us investigate the product
αR (µ2 ) 2
Z3 (β, µ)Z̃32 (β, µ) = Z̃ (β, µ), (74)
α0 (ΛUV ) 1
where (59) was used. The left-hand side of the latter equation can be directly obtained
from the numerical result for the renormalization constants Z3 (β, µ) and Z̃3 (β, µ). Note
that for large UV cutoff, one finds
1 Λ2
∝ ln 2 UV = − ln σ a 2 + const, (75)
α0 (ΛUV ) Λ1-loop
where the constant comprises cutoff (and therefore β) independent terms. The crucial point
is that if the product Z3 (β, µ)Z̃32 (β, µ) rises linearly with − ln(σ a 2 ) the additional factor
Z̃12 (β, µ) must be finite in the continuum limit (since the renormalized coupling αR (µ2 ) is
assumed finite). Our numerical findings for Z3 (β, µ)Z̃32 (β, µ) are also shown in Figs. 2 and
3 (right panel). The data nicely support Taylor’s findings, i.e., Z̃1 is cutoff- and therefore
µ-independent.
4.2. The running coupling constant
Once it is established that Z̃1 is finite, the momentum dependence of the running
coupling constant can be simply derived from the product (61)

αR p2 = αR µ2 FR p2 , µ2 JR2 p2 , µ2 . (76)
The overall normalization factor can be obtained by comparing the lattice data with the
well-known perturbative result, which is valid at high momentum. Here, we compare
with the 2-loop expression, which is known to be independent of the renormalization
prescription, i.e.,

4π 2β1 ln(ln x)
α2-loop x = p2 /Λ22-loop = 1− 2 , (77)
β0 ln x β0 ln x
with β0 and β1 given in Eq. (67). In order to obtain Λ2-loop and to fix the overall factor,
we fitted αR (µ)FR (p2 , µ2 )JR2 (p2 , µ2 ) to the 2-loop running coupling α2-loop(p) where
only momenta p pM were taken into account. Fitting parameters were αR (µ) and the 2-
loop perturbative scale Λ2-loop . Starting from a very low value pM we fit these parameters
while gradually increasing pM . For small values of pM , we observe that the functional
form of (77) tries to incorporate genuine non-perturbative effects by adjusting Λ2-loop ,
thus, introducing a spurious pM dependence to Λ2-loop . However, a plateau is reached for
the Tübingen data at pM ≈ 2 GeV indicating that the data are well reproduced by the
2-loop formula in this regime. We find in this case that
Λ2-loop = 0.95(15) GeV. (78)

For the São Carlos data we have cut the data at pM ≈ 2.5 GeV, which corresponds to a
large drop in the χ 2 /d.o.f. of the fit. (This also corresponds to reaching a relatively good
plateau for Λ2-loop obtained from the fit.) We obtain the value Λ2-loop = 1.2(1) GeV. The
two values are consistent within error bars.
Our final results for the running coupling constant are presented in Fig. 4. In the
IR region, the two data sets show a clear departure from the perturbative behavior and
suggest a finite value αc for the running coupling constant at zero momentum. We estimate
αc = 5(1). This value is in agreement with our previous fits for αR (p2 ) [34,35], and is
consistent with the DSE result of Ref. [25], αc ≈ 5.2, using κ = 0.5 as input.
4.3. Gluon and ghost form factors
Fig. 5 shows the Tübingen data for the gluon and the ghost form factors. The error
bars comprise statistical errors only. It turns out that one observes an additional scattering
of the data points which is not of statistical origin. This additional systematic noise is
Fig. 4. The running coupling in comparison with the results from perturbation theory: Tübingen data (left panel)
and São Carlos data (right panel). The momentum cutoff is pM = 2 GeV in the former case and pM = 2.5 GeV
in the latter.
Fig. 5. The gluon form factor FR (p) and the ghost form factor JR (p) as a function of the momentum transfer p
(Tübingen data).
pronounced when simulated annealing is used for gauge fixing and it afflicts especially
the small momentum range. We attribute this error to the residual uncertainty of gauge
fixing (Gribov noise). In particular, since the simulated annealing is capable of hopping
from one local minimum to the other, the algorithm is sensitive to the large-scale structure
of the minimizing functional. This hopping then produces a non-Gaussian noise which is
underestimated when one uses the standard Gaussian error propagation. For this reason,
we dropped the first three momentum points from the Tübingen data sets.
Fig. 6. The gluon form factor FR (p) (above) and the gluon propagator DR (p) (below) as a function of the
momentum transfer p (São Carlos data). Note the logarithmic scale in the second plot. Fit of gluon propagator
using the analogue of Eq. (79) has been done for momenta p 2 GeV and with Λ = 1.2 GeV.
In Figs. 6 and 7 we report the rescaled São Carlos data for the renormalized gluon
(respectively, ghost) form factor and the data for the corresponding propagators. We stress
that in the gluon case finite-size effects depend on whether we consider the full propagator
or the form factor. In fact, for the propagator these effects are larger in the IR region, while
for the form factor the effects are larger in the UV limit (due to the multiplication by p2 ).
Thus, the ranges of momenta (for each β) considered for the plots are different in the
two cases. Nevertheless, the matching factors obtained are in agreement. The difference in
finite-size effects between propagator and form factor is less pronounced when considering
the ghost propagator.
At sufficiently high momentum p pM (we found in Section 4.2 that pM ≈ 2 GeV is
an acceptable choice), the momentum dependence of the renormalized form factors should
be given by the formula
γ
2 2 2 2 p2 p2
FR p , µ , JR p , µ ≈ d2 (µ) α2-loop 1 + γ̄ α2-loop ,
Λ22-loop Λ22-loop
(79)
Fig. 7. The ghost form factor JR (p) (above) and the ghost propagator GR (p) (below) as a function of the
momentum transfer p (São Carlos data). Note the logarithmic scale in the second plot. Fit of ghost propagator
using the analogue of Eq. (79) has been done for momenta p 2 GeV and with Λ = 1.2 GeV.
where γ is the leading-order anomalous dimension of the gluon (respectively, ghost)

propagator, given by γ = 13/22 (respectively, γ = 9/44). The parameter γ̄ stems from
the next to leading order to the anomalous dimension and is scheme-dependent. At least
in the MS scheme, this parameter is small. Furthermore, Eq. (79) can be derived from
the renormalization-group equation using the 2-loop scaling functions β(gR ) and γA (gR ).
Hence, (79) originates from the resummation of an infinite set of 2-loop diagrams and, e.g.,
comprises the so-called “leading logs”.
Using the Tübingen data set, pM = 2 GeV and Λ2-loop = 950 MeV, we fitted γ̄ to the
gluon and ghost data, respectively. We find that these parameters are, indeed, small, i.e.,
γ̄gluon = −0.036(18), γ̄ghost = 0.011(10). (80)
Although the errors on these parameters are rather large, we find it encouraging that
the parameters appear with opposite signs. In the case that the product of form factors
FR (p2 )JR2 (p2 ) is, indeed, renormalization group invariant, one would expect that
γ̄gluon + 2γ̄ghost = 0. (81)
It is clear from the Figs. 2 and 5 that the high momentum tail is well reproduced by (79).
For the São Carlos data we have found that a cut at pM = 2GeV corresponds to a large
drop in the χ 2 /d.o.f. of the fits, both for the gluon and for the ghost cases. We start by
fitting the leading-order term only (i.e., ignoring γ̄ ). We get
Λ2-loop = 1.19(4) and Λ2-loop = 1.13(2), (82)
respectively, from the fits of the gluon and of the ghost propagator. These values are
consistent with the result Λ2-loop = 1.2(1), obtained in the previous section. We then
fix this value for Λ2-loop and perform fits with γ̄ as a free parameter. We obtain a good
description of the data (see Figs. 6 and 7), with values for γ̄ even consistent with zero.
As can be seen from our plots, the gluon form factor is suppressed in the low momentum
regime, while the ghost form factor is divergent. Correspondingly, the ghost propagator
diverges faster than 1/p2 and the gluon propagator appears to be finite. As mentioned in
the introduction, an IR-finite gluon propagator [26–31] and an IR-divergent ghost form
factor [30,31] were obtained before by separate studies.
A quantitative analysis of the IR behavior for the propagators—including the evaluation
of the exponent κ mentioned in the introduction—was already presented in Refs. [34,
35]. More thorough such analyses will be presented separately for the two sets of data
in Refs. [63,64].
5. Conclusions
For the first time, evidence from extensive lattice simulations is provided that the ghost–
ghost–gluon vertex renormalization constant Z̃1 is, indeed, finite in continuum field theory
(as found by Taylor using all orders perturbation theory). Also, our result is probably
not affected by Gribov ambiguities, since Z̃1 is obtained using data in the UV limit. It
therefore appears that the Gribov ambiguities (in the lattice approach and the Faddeev–
Popov quantization) do not afflict the renormalization of the vertex.
Also, we performed a thorough study of the gluon and the ghost form factors. Our data
favor the scenario of an IR finite (or even vanishing) gluon propagator while the ghost form
factor is singular in the IR limit.
Finally, we have obtained the running coupling constant over a wide range of momenta
using the data for gluon and ghost form factors. Our data are consistent with the existence
of an IR fixed point αc = 5(1). Note that this value is inside the interval given by the DSE
expression (5).
We stress that we compared our results for two slightly different lattice formulations,
obtaining consistent results in all cases considered.
Acknowledgements
J.C.R.B.’s contribution was funded by the Deutsche Forschungsgemeinschaft (Project

No. SCHM 1342/3-1) and by the DFG Research Center “Mathematics for Key Technolo-
gies” (FZT 86) in Berlin. K.L. greatly acknowledges the stimulating atmosphere and the
hospitality of the Institute for Theoretical Physics, University of Karlsruhe where parts
of the project were performed. The research of A.C. and T.M. is supported by FAPESP
(Project No. 00/05047-5). We thank C.S. Fischer for helpful comments on the manuscript.
References
[1] C. Michael, Phys. Lett. B 283 (1992) 103, hep-lat/9205010.

[2] G.S. Bali, K. Schilling, Phys. Rev. D 47 (1993) 661, hep-lat/9208028.
[3] M. Luscher, R. Sommer, P. Weisz, U. Wolff, Nucl. Phys. B 413 (1994) 481, hep-lat/9309005.
[4] G.M. de Divitiis, R. Frezzotti, M. Guagnelli, R. Petronzio, Nucl. Phys. B 422 (1994) 382, hep-lat/9312085.
[5] C. Parrinello, Phys. Rev. D 50 (1994) 4247, hep-lat/9405024.
[6] J. Skullerud, A. Kizilersu, JHEP 0209 (2002) 013, hep-ph/0205318.
[7] S. Furui, H. Nakajima, in: Wien 2000, Quark Confinement and the Hadron Spectrum, 2000, p. 275, hep-
lat/0012017.
[8] A. Cucchieri, in: Proceedings of Hadron Physics 2002, World Scientific, Singapore, 2003, p. 161, hep-
lat/0209076.
[9] J.C. Taylor, Nucl. Phys. B 33 (1971) 436.
[10] C.D. Roberts, S.M. Schmidt, Prog. Part. Nucl. Phys. 45 (2000) S1, nucl-th/0005064.
[11] R. Alkofer, L. von Smekal, Phys. Rep. 353 (2001) 281, hep-ph/0007355.
[12] L. von Smekal, R. Alkofer, A. Hauck, Phys. Rev. Lett. 79 (1997) 3591, hep-ph/9705242.
[13] L. von Smekal, A. Hauck, R. Alkofer, Ann. Phys. 267 (1998) 1, hep-ph/9707327;
L. von Smekal, A. Hauck, R. Alkofer, Ann. Phys. 269 (1998) 182, Erratum.
[14] D. Atkinson, J.C.R. Bloch, Phys. Rev. D 58 (1998) 094036, hep-ph/9712459.
[15] D. Atkinson, J.C.R. Bloch, Mod. Phys. Lett. A 13 (1998) 1055, hep-ph/9802239.
[16] D. Zwanziger, Phys. Rev. D 65 (2002) 094039, hep-th/0109224.
[17] J.C.R. Bloch, Phys. Rev. D 64 (2001) 116011, hep-ph/0106031.
[18] C. Lerche, Diploma thesis, University Erlangen-Nuremberg, June 2001 (in German).
[19] C. Lerche, L. von Smekal, Phys. Rev. D 65 (2002) 125006, hep-ph/0202194.
[20] C.S. Fischer, R. Alkofer, Phys. Lett. B 536 (2002) 177, hep-ph/0202202.
[21] C.S. Fischer, R. Alkofer, H. Reinhardt, Phys. Rev. D 65 (2002) 094008, hep-ph/0202195.
[22] R. Alkofer, W. Detmold, C.S. Fischer, P. Maris, hep-ph/0309077.
[23] R. Alkofer, W. Detmold, C.S. Fischer, P. Maris, hep-ph/0309078.
[24] D. Zwanziger, Phys. Rev. D 67 (2003) 105001, hep-th/0206053.
[25] J.C.R. Bloch, Few Body Syst. 33 (2003) 111, hep-ph/0303125.
[26] A. Cucchieri, Phys. Lett. B 422 (1998) 233, hep-lat/9709015.
[27] A. Cucchieri, Phys. Rev. D 60 (1999) 034508, hep-lat/9902023.
[28] K. Langfeld, H. Reinhardt, J. Gattnar, Nucl. Phys. B 621 (2002) 131, hep-ph/0107141.
[29] F.D. Bonnet, et al., Phys. Rev. D 64 (2001) 034501, hep-lat/0101013.
[30] H. Suman, K. Schilling, Phys. Lett. B 373 (1996) 314, hep-lat/9512003.
[31] A. Cucchieri, Nucl. Phys. B 508 (1997) 353, hep-lat/9705005.
[32] A. Cucchieri, D. Zwanziger, Phys. Lett. B 524 (2002) 123, hep-lat/0012024.
[33] A. Cucchieri, T. Mendes, A.R. Taurines, Phys. Rev. D 67 (2003) 091502, hep-lat/0302022.
[34] J.C.R. Bloch, A. Cucchieri, K. Langfeld, T. Mendes, Nucl. Phys. B (Proc. Suppl.) 119 (2003) 736, hep-
lat/0209040.
[35] K. Langfeld, J.C.R. Bloch, J. Gattnar, H. Reinhardt, A. Cucchieri, T. Mendes, in: Gargnano 2002, Quark
Confinement and the Hadron Spectrum, 2003, p. 297, hep-th/0209173.
[36] D. Zwanziger, Phys. Lett. B 257 (1991) 168.
[37] D. Zwanziger, Nucl. Phys. B 364 (1991) 127.
[38] D. Zwanziger, Nucl. Phys. B 412 (1994) 657.
[39] V.N. Gribov, Nucl. Phys. B 139 (1978) 1.
[40] A. Cucchieri, T. Mendes, D. Zwanziger, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 697, hep-lat/0110188.
[41] K. Langfeld, hep-lat/0204025.
[42] L. Giusti, M.L. Paciello, S. Petrarca, B. Taglienti, M. Testa, Phys. Lett. B 432 (1998) 196, hep-lat/9803021.
[43] A. Cucchieri, F. Karsch, Nucl. Phys. B (Proc. Suppl.) 83 (2000) 357, hep-lat/9909011.
[44] L. Giusti, M.L. Paciello, C. Parrinello, S. Petrarca, B. Taglienti, Int. J. Mod. Phys. A 16 (2001) 3487, hep-
lat/0104012.
[45] I.L. Bogolubsky, V.K. Mitrjushkin, hep-lat/0204006.
[46] A. Cucchieri, T. Mendes, in: Strong and Electroweak Matter, Copenhagen 1998, 1999, p. 324, hep-
lat/9902024.
[47] A. Cucchieri, in: Understanding Deconfinement in QCD, Trento 1999, 1999, p. 202, hep-lat/9908050.
[48] K. Langfeld, E.M. Ilgenfritz, H. Reinhardt, G. Shin, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 501, hep-
lat/0110024.
[49] K. Langfeld, E.M. Ilgenfritz, H. Reinhardt, A. Schafke, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 658, hep-
lat/0110055.
[50] G.P. Lepage, P.B. Mackenzie, Phys. Rev. D 48 (1993) 2250, hep-lat/9209022.
[51] E. Marinari, C. Parrinello, R. Ricci, Nucl. Phys. B 362 (1991) 487.
[52] A. Cucchieri, T. Mendes, Nucl. Phys. B 471 (1996) 263, hep-lat/9511020.
[53] A. Cucchieri, T. Mendes, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 811, hep-lat/9608051.
[54] A. Cucchieri, T. Mendes, Comput. Phys. Commun. 154 (2003) 1, hep-lat/0301019.
[55] J.E. Mandula, Phys. Rep. 315 (1999) 273.
[56] A. Cucchieri, D. Zwanziger, Phys. Rev. D 65 (2002) 014001, hep-lat/0008026.
[57] T.D. Bakeev, et al., hep-lat/0311041.
[58] A. Cucchieri, Nucl. Phys. B 521 (1998) 365, hep-lat/9711024.
[59] P. Marenzoni, G. Martinelli, N. Stella, Nucl. Phys. B 455 (1995) 339, hep-lat/9410011.
[60] J. Fingberg, U.M. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493, hep-lat/9208012.
[61] J.C.R. Bloch, Phys. Rev. D 66 (2002) 034032, hep-ph/0202073.
[62] D.B. Leinweber, et al., Phys. Rev. D 60 (1999) 094507, hep-lat/9811027;
D.B. Leinweber, et al., Phys. Rev. D 61 (2000) 079901, Erratum.
[63] J.C.R. Bloch, K. Langfeld, in preparation.
[64] A. Cucchieri, T. Mendes, in preparation.
Flavor hierarchy from extra dimension and gauge

threshold correction
Kiwoon Choi, Ian-Woo Kim, Wan Young Song
Department of Physics, Korea Advanced Institute of Science and Technology, Daejeon 305-701, South Korea
Received 16 August 2003; accepted 18 March 2004
Abstract
Dynamical quasi-localization of matter fields in extra dimension is an elegant mechanism to
generate hierarchical 4-dimensional Yukawa couplings. We point out that a bulk matter field whose
zero mode is quasi-localized can give a large Kaluza–Klein threshold correction to low energy
gauge couplings, which is generically of the order of ln(Yukawa)/8π 2 , so it can significantly affect
gauge coupling unification. We compute such threshold corrections in generic 5-dimensional theories
compactified on S 1 /Z2 × Z2 , and apply the result to grand unified theories on orbifold generating
small Yukawa couplings through quasi-localization.
1. Introduction
It has been noticed that extra dimension can provide an elegant mechanism to generate
hierarchical Yukawa couplings [1–14]. The quark and lepton fields can be quasi-localized
in extra dimension in a natural manner, and then their 4-dimensional (4D) Yukawa
couplings are determined by the wavefunction overlap factor e−MπR , where M is a
combination of mass parameters in higher-dimensional theory and R is the length of
extra dimension. This allows that hierarchical Yukawa couplings are obtained from
fundamental mass parameters having the same order of magnitude. Extra dimension
has been known to be useful also for constructing a natural model of gauge unification
[15–25]. Supersymmetric 4D grand unified theories (GUTs) successfully accommodate
gauge coupling unification, however suffer from some difficulties such as the doublet–
E-mail addresses: kchoi@hep.kaist.ac.kr (K. Choi), iwkim@hep.kaist.ac.kr (I.-W. Kim),

wysong@hep.kaist.ac.kr (W.Y. Song).
doi:10.1016/j.nuclphysb.2004.03.018
102 K. Choi et al. / Nuclear Physics B 687 (2004) 101–123
triplet splitting problem and the proton decay problem. These problems can be elegantly
solved in GUT on orbifold with gauge symmetry broken by boundary conditions [15,18]. It
is straightforward to implement the idea of dynamical quasi-localization in orbifold GUT
to get hierarchical Yukawa couplings as well as successful gauge unification [7,9].
In any GUT, heavy particle threshold effects at GUT symmetry breaking scale should
be taken into account for a precision analysis of low energy gauge couplings ga2 . In
conventional 4D GUT, those threshold corrections to 1/ga2 are generically of the order of
1/8π 2 and thus not so important, unless the model contains a large number of superheavy
particles which become massive as a consequence of GUT symmetry breaking or some of
superheavy masses are hierarchically different from each other [26]. As was pointed out
a long time ago, higher-dimensional field theory and/or string theory contain (infinitely)
many Kaluza–Klein (KK) or stringy modes, so can have a sizable threshold correction [27].
Therefore, it is essential to include stringy and/or KK threshold correction in the precision
analysis of low energy couplings in string and/or higher-dimensional field theories. In
this paper, we wish to examine the KK threshold corrections in generic 5D orbifold
field theories in which hierarchical 4D Yukawa couplings are generated through quasi-
localization. As we will see, the KK threshold corrections to 1/ga2 in such models are
generically of the order of ln(Yukawa)/8π 2 , so can significantly affect gauge coupling
unification. The outline of this paper is as follows. In Section 2, we discuss the KK
spectrum and zero-mode wavefunctions of generic scalar and spinor fields in 5D theory
compactified on S 1 /Z2 × Z2 . In Section 3, we discuss the 4D Yukawa couplings of those
quasi-localized scalar and fermion zero modes. In Section 4, we compute KK threshold
corrections in generic 5D theories on S 1 /Z2 × Z2 . In Section 5, we use these results to
derive the KK correction to the predicted value of the low energy QCD coupling constant
in 5D GUT on orbifold. We then construct a class of supersymmetric 5D orbifold GUTs
which generate hierarchical Yukawa couplings through dynamical quasi-localization while
keeping successful gauge coupling unification. Section 6 is the conclusion.
2. Kaluza–Klein analysis and quasi-localized zero modes
In this section, we analyze the KK wavefunctions and spectra of scalar and spinor fields
in 5D theory compactified on S 1 /Z2 × Z2 . Our major concern is the dynamical quasi-
localization of zero-mode wavefunctions which would result in hierarchical 4D Yukawa
couplings. We also consider the full KK spectra which will be relevant for the discussion
of KK threshold corrections to low energy gauge couplings. Throughout this paper, we will
use the convention for S 1 /Z2 × Z2 which is given by the following transformation of the
fifth spacetime coordinate y ≡ y + 4πR:
Z2 : y → −y, Z2 : y + πR → −y + πR. (1)

so the fundamental domain of S 1 /Z2 × Z2 corresponds to 0 y πR.
Let us first consider a 5D complex scalar field with an action:

− d 5 x DM φ zz D M φ zz ∗ + m2zz + 2µzz δ(y) − 2µzz δ(y − πR) φ zz φ zz ∗ , (2)
K. Choi et al. / Nuclear Physics B 687 (2004) 101–123 103
and the orbifold boundary conditions:

φ zz (−y) = zφ zz (y),

φ zz (−y + πR) = z φ zz (y + πR),
where z = ±1 and z = ±1. The equation of motion for each KK mode of this scalar field
is given by
2 zz 2 zz
∂y + ωn φn = 2 µzz δ(y) − µzz δ(y − πR) φnzz , (3)
where
2 2
ωnzz = Mnzz − m2zz (4)

for Mnzz denoting the 4D mass of φnzz . Upon ignoring unimportant normalization factor,
one easily finds the following KK wavefunctions on the fundamental domain:
µ++
φn++ = cos ωn++ y + ++ sin ωn++ y ,
ωn

φn+− = sin ωn+− (y − πR) ,

φn−+ = sin ωn−+ y ,

φn−− = sin ωn−− y , (5)

where ωnzz (= 0) are either real or pure imaginary constants determined by
(µ++ − µ )ω++
tan ωn++ πR = ++ 2 ++ n ,
(ωn ) + µ++ µ++
ω+−
tan ωn+− πR = − n ,
µ+−
ω −+
tan ωn−+ πR = n ,
µ−+

tan ωn−− πR = 0. (6)

Once ωnzz are determined, the KK mass spectrum of φ zz can be determined by the relation
zz 2 2
Mn = m2zz + ωnzz . (7)

Note that a pure imaginary ωnzz gives (Mnzz )2 smaller than m2zz , while a real ωnzz gives

(Mnzz )2 bigger than m2zz . Note also that there can be only one or no imaginary ωnzz
from (6).
Since we are interested in the shape of zero-mode wavefunction, let us consider the

conditions for φ zz to have a zero mode, i.e., M0zz = 0, independently of the value of the
orbifold radius R. Obviously, only φ ++ can have such a zero mode only when its bulk and
brane masses are tuned to satisfy
µ++ = µ++ , m2++ = µ2++ , (8)

which are indeed satisfied in supersymmetric theories [5,28]. The resulting zero mode (on
the fundamental domain) is given by
φ0++ = exp(µ++ y), (9)

so it is quasi-localized at y = 0 if µ++ < 0, while at y = πR if µ++ > 0. In a more general
case with µ++ = µ++ but m2++ = µ2++ , the KK spectra of φ ++ are given by
2 ++ 2 n2
M0++ = m2++ − µ2++ , Mn = m2++ + 2 , (10)
R
for n being a positive integer.
For φ +− , if µ+− πR < −1, there can be a mode lighter than m+− :
+− 2
φ0+− = sinh ξ(y − πR), M0 = m2+− − ξ 2 , (11)
where ξ is a real constant determined by
ξ
tanh(ξ πR) = .
|µ+− |
In the limit µ+− πR −1, M0+− is determined to be
+− 2 2
M0 = m+− − µ2+− + 4µ2+− e−2|µ+− |πR + O(e−4|µ+− |πR ) . (12)
So in supersymmetric case in which m2+− = µ2+− , the mass of φ0+−
is exponentially
suppressed in the limit µ+− πR −1. On the other hand, all other KK modes of φ +−
are heavier than m+− . Similarly, if µ−+ πR > 1, φ −+ can have a lighter mode:
−+ 2
φ0−+ = sinh(ξ y), M0 = m2−+ − ξ 2 ,
where
ξ
tanh(ξ πR) = .
µ−+
In the limit µ−+ πR 1, we have
−+ 2 2 2
M0 = m−+ − µ2−+ + 4µ2−+ e−2µ−+ πR + O(e−4µ−+ πR ) ,
yielding an exponentially small mass of φ0−+ in supersymmetric case with µ−+ πR 1.
Let us now consider a 5D spinor field with an action

− d 5 x i Ψ̄ zz γ M ∂M + Mzz (y) Ψ zz , (13)
and the orbifold boundary condition

Ψ zz (−y) = zγ5 Ψ zz (y),

Ψ zz (−y + πR) = z γ5 Ψ zz (y + πR),
where (y) = − (−y) = (y + 2πR) = 1 on the covering space of S 1 /Z2 × Z2 . Note that
in order for the action to be invariant under Z2 × Z2 , 5D spinor can have only a kink type

mass Mzz (y). The equation of motion for each KK mode of Ψ zz is given by
2 zz 2 2 zz zz
∂y + ωn ΨnL = −2Mzz δ(y) − δ(y − πR) ΨnL ,
2 zz 2 2 zz zz
∂y + ωn ΨnR = +2Mzz δ(y) − δ(y − πR) ΨnR , (14)
where γ5 ΨL,R = ±ΨL,R and
zz 2 zz 2
ωn = Mn − (Mzz )2 (15)
zz zz
for Mn denoting the 4D mass of Ψn . When compared to the scalar equation of motion
(3), this suggests that in supersymmetric case the bulk and brane scalar masses should
satisfy
µzz = µzz , µ2zz = m2zz = M2zz .
Obviously, Ψ ++ has a left-handed zero mode for any value of M++ :
++
Ψ0L (y) = exp(−M++ y) (16)
which is quasi-localized at either y = 0 (if M++ > 0) or y = πR (if M++ < 0). All other
left-handed modes of Ψ ++ are paired-up with right-handed modes to get a 4D Dirac mass
bigger than M++ :

n2
Mn++ = M2++ + 2 . (17)
R
Similarly, Ψ −− has a right-handed zero mode for any value of M−− and also the massive
modes with (Mn−− )2 = M2−− + n2 /R 2 . The wavefunction of zero mode is given by
−−
Ψ0R (y) = exp(M−− y), (18)
so the zero mode is quasi-localized at either y = 0 (if M−− < 0) or y = πR (if M−− > 0).
For Ψ +− , there is no zero mode. However, if M+− πR > 1, there can be two modes
lighter than M+− , while all other modes are heavier than M+− The wavefunctions of
light modes are given by
+−
Ψ0L (y) = sinh k(y − πR) , (19)
+−
Ψ0R (y) = sinh(ky), (20)
where k is a real constant determined by
k
tanh(kπR) = . (21)
M+−
Obviously, one of these two light modes is localized at y = 0, while the other is localized
at y = πR. In 4D viewpoint, these two modes are paired up to get a Dirac mass
+− 2
M0 = M2+− − k 2 . (22)
Note that in the limit M+− πR 1,
M0+− ≈ 2M+− exp(−M+− πR), (23)
so Ψ0+− can be arbitrarily light. Similarly, if M−+ πR < −1, Ψ −+ can have a light Dirac
mode with M0−+ ≈ 2|M−+ | exp(M−+ πR).
The dynamical quasi-localization of light modes and also the shape of full KK spectra
offer a possibility that massless brane fields can be considered as a large mass limit of
bulk fields.1 To see this, let us take the limit |Mzz | → ∞.2 In this limit, the left-handed
zero mode of Ψ ++ becomes a chiral brane fermion confined at y = 0 (if M++ → ∞)
or at y = πR (if M++ → −∞), while the right-handed zero mode of Ψ −− becomes a
chiral brane fermion at y = 0 (if M−− → −∞) or at y = πR (if M−− → ∞). For Ψ +− ,
+− +−
the light modes Ψ0L and Ψ0R become a massless brane fermion at y = 0 and y = πR,
−+ −+
respectively, in the limit M+− → ∞. Similarly, Ψ0L and Ψ0R become a massless brane
fermion at y = πR and y = 0, respectively, in the limit M−+ → −∞. All other KK modes

of Ψ zz are heavier than |Mzz |, and thus are decoupled in the limit |Mzz | → ∞. So, any
massless brane fermion can be considered as a bulk fermion in the large kink mass limit.
3. Yukawa hierarchy from quasi-localization
To discuss flavor hierarchy arising from quasi-localization, let us consider a generic 5D

theory containing arbitrary number of scalar and fermion fields:

S = − d 5 x DM φI D M φI∗ + m2I δI J + 2µI J δ(y) − 2µI J δ(y − πR) φI φJ∗

+ i Ψ̄A γ M DM + MA (y) ΨA + LY . (24)
Here LY stands for the Yukawa couplings between φI and ΨA and the orbifold boundary
conditions are given by
φI (−y) = zI φ(y), φI (−y + πR) = zI φI (y + πR),

ΨA (−y) = zA γ5 ΨA (y), ΨA (−y + πR) = zA γ5 ΨA (y + πR). (25)
Though the brane masses of scalar fields can have off-diagonal components in general, we
will assume for simplicity that they are diagonal: µI J = µI δI J and µI J = µI δI J . To get
hierarchical 4D Yukawa couplings, we also assume that Yukawa couplings in 5D theory
exist only at the fixed points, which is assured by 5D SUSY in supersymmetric theories.
Then the most general form of Yukawa couplings can be written as
λXP Q λXP Q
LY = δ(y) ϕ ψ ψ + δ(y − πR) 3/2 ϕX ψP ψQ + c.c., (26)
Λ3/2 X P Q Λ
where Λ denotes the cutoff scale of our 5D orbifold field theory, ϕX = {φI or φI∗ },
and ψP = { 12 (1 + γ5 )ΨA or 12 (1 + γ5 )ΨAc }. Note that λipq and λipq are dimensionless
parameters in our convention.
1 Here we are implicitly assuming that there is a 3-brane at each orbifold fixed point.
2 This infinite kink mass should be considered as a large mass comparable to the cutoff scale of the orbifold
field theory under consideration.
Let φi denote 5D scalar fields having a zero mode φ0i = exp(mi y), i.e., scalar fields with
zi = zi = 1 and µi = µi = mi , and Ψp denote 5D spinor fields having a chiral zero mode
ψ0p = exp(−zp Mp y), i.e., spinor fields with zp = zp = ±1. It is then straightforward to
find that the 4D Yukawa couplings of canonically normalized zero modes are given by

yipq = Z(mi )Z(−zp Mp )Z(−zq Mq ) λipq

+ Z(−mi )Z(zp Mp )Z(zq Mq ) λipq , (27)
where
2m 1
Z(m) = . (28)
Λ e2mπR − 1
Obviously, yipq can have very different values, depending upon the values of mi and Mp,q ,
even when all of the 5D parameters λipq and λipq have similar values.
A simple way to get hierarchical Yukawa couplings through quasi-localization is to
assume that all Yukawa couplings originate from a single fixed point, for instance from
y = πR. As a concrete example, let us consider the case that all Yukawa couplings
originate from y = πR and φi is a brane field confined at y = πR with Z(−mi ) ≈ 1.
Note that a brane scalar field at y = πR can be obtained from a bulk scalar field by taking
the limit m = O(Λ) while keeping µ = µ = m. Then for zp,q Mp,q −1/R, we have

4|Mp Mq |
yipq ≈ λipq , (29)
Λ2
while for zp,q Mp,q 1/R,

4|Mp Mq | −(zp Mp +zq Mq )πR
yipq ≈ e λipq . (30)
Λ2
The physical interpretation of this result is simple. If zp,q Mp,q 1/R, the corresponding
zero modes are quasi-localized at y = 0, so the Yukawa couplings are exponentially
suppressed as they originate from y = πR. On the other hand, for zp,q Mp,q −1/R,
the zero modes are localized at y = πR, so there is no exponential suppression in Yukawa
couplings.
If the fundamental theory at Λ is weakly coupled, simple dimensional analysis would
suggest that the dimensionless λipq are of order unity or less. However, a more interesting
possibility is that the theory is strongly coupled at Λ [29,30]. In 5D theories under
consideration, the standard model gauge fields live in bulk spacetime, so their 4D couplings
are given by 1/ga2 ≈ πR/g5a 2 , where g 2 denote the dimensionful 5D gauge couplings
5a
with mass dimension −1. In order for our 5D theory to be a useful framework, it must be
valid up to an energy scale significantly higher than the compactification scale 1/R, i.e.,
Λ 1/R. If Λ were comparable to 1/R, we would not have to consider a 5D theory as
an intermediate step going from the fundamental theory at Λ to the 4D effective theory
for low energy physics. We would rather go directly to the 4D effective theory from the
fundamental theory. On the other hand, if Λ 1/R as desired, we have g5a 2 ≈ πRg 2
a
1/Λ, implying that the fundamental theory at Λ is strongly coupled. As is well known,
only in strongly coupled scenario, GUT on orbifold can provide a meaningful prediction
for sin2 θW or the QCD coupling constant at the weak scale. In such strongly-coupled
scenario, dimensional analysis suggests
λipq = O(4π).
It is then straightforward to get hierarchical 4D Yukawa couplings ranging from the top
quark Yukawa coupling yt ≈ 1 to the electron Yukawa coupling ye ≈ 10−5 –10−6 within
the parameter range
ΛπR = O(102 ), |Mp |πR 7.

Note that, in strongly coupled scenario, if the Higgs boson and the left and right-handed
top quarks are all brane fields, the resulting top-quark Yukawa coupling would be too large,
yt = O(4π), so one needs to put some of those fields in bulk spacetime.
In supersymmetric 5D theories, the fermion kink masses Mp are related to the
graviphoton gauging [32,33]:
DM Ψp = ∂M Ψp + iMp (y)BM Ψp + · · · , (31)

where BM denotes the graviphoton. This suggests that it is a plausible assumption that Mp
are quantized (in an appropriate unit) as the conventional gauge charges are quantized. If
true, the resulting Yukawa couplings (30) would have the same form as those obtained
from the Frogatt–Nielsen mechanism which generates small Yukawa couplings using a
spontaneously broken U (1) flavor symmetry [31]. In the next section, we will see that the
KK threshold corrections to low energy gauge couplings are generically given by

1 ln(yipq )
∆ 2 =O , (32)
ga 8π 2
if the hierarchical 4D Yukawa couplings yipq are generated by quasi-localization.
Obviously then the KK threshold corrections can significantly affect the gauge coupling
unification.
4. Kaluza–Klein threshold correction to low energy gauge coupling
In this section, we discuss the 1-loop threshold correction to low energy gauge coupling
in 5D orbifold field theory. The bare action of bulk gauge fields can be written as

1 κa κa
Sbare = − d x 5
2
+ δ(y) + δ(y − πR) F aMN FMN a
, (33)
4g5a 4 4
and then the 4D gauge couplings at tree level are given by

1 πR
= 2 + κa + κa . (34)
ga2 tree g5a
One-loop corrections to low energy couplings can be computed by summing the
contributions from all KK modes. In generic 5D orbifold field theory, such computation
yields [34,35]

1 πR γa
= 2 + ΛπR + κa + κa
ga2 (p) g5a 24π 3

1 Λ
+ 2
∆ a (ln Λ, m, µ, µ , M, R) + b a ln
8π p

1 1 Λ
≡ 2
+ 2
∆ a (ln Λ, m, µ, µ , M, R) + b a ln , (35)
ga bare 8π p
where γa are the coefficients of (UV-sensitive) linearly divergent corrections, ∆a stand for
(UV-insensitive) logarithmically divergent or finite threshold corrections due to massive
KK modes, and ba are the standard 1-loop beta function coefficients due to zero modes.
Here (1/ga2 )bare correspond to the uncalculable bare couplings of the model, while ∆a
are unambiguously calculable within 5D orbifold field theory. Note that ∆a contain a
piece linear in ln Λ as well as a finite piece depending on the scalar and fermion mass
parameters m, µ, µ , M and also on the orbifold radius R. Here we assume that the cutoff
scale Λ is large enough compared to other mass parameters of the theory, thus ignore the
part suppressed by an inverse power of Λ. In the above expression, the renormalization
point p is assumed to be below the mass of the lightest massive KK mode, MKK , but far
above the masses of all zero modes which would be around the weak scale.
The KK threshold corrections in 5D theories on warped S 1 /Z2 × Z2 have been
discussed before in [33,35–39]. In [33,35], ∆a for supersymmetric 5D theories on warped
S 1 /Z2 × Z2 have been computed in the framework of 4D effective supergravity. In this
framework, ∆a could be obtained by computing the tree-level Kähler potential and also the
one-loop correction to holomorphic gauge kinetic functions which can be determined by
the chiral anomaly structure of 5D orbifold field theory [40]. The KK threshold corrections
for nonsupersymmetric 5D theories on warped S 1 /Z2 × Z2 have been computed in [37–39]
by directly evaluating all KK mode contributions in dimensional regularization scheme
[41], and it was confirmed that the results in supersymmetric limit agree with those of
[33,35]. In this paper, we compute the KK threshold corrections in generic 5D theories on
flat S 1 /Z2 × Z2 with quasi-localized zero modes using the method of [39].
To calculate one-loop gauge couplings at low energies, we integrate out all massive KK
modes and derive the one-loop effective action of gauge field zero mode Aµ :
Seff = Sbare + Γφ + ΓΨ + ΓA , (36)

where Sbare is given in (33), and the 1-loop contributions from 5D scalar φ, 5D spinor Ψ
and 5D vector AM are given by
1
iΓφ = − Trφ ln −D 2 + M 2 (φ) ,
2
1 µν
iΓΨ = TrΨ ln −D 2 + M 2 (Ψ ) + Fµν J1/2 + TrΨ0 ln(/
D),
2
1 µν 1
iΓA = − TrAµ ln −D 2 + M 2 (Aµ ) + Fµν J1 − TrA5 ln −D 2 + M 2 (A5 )
2 2

+ Trξ ξ̄ ln −D 2 + M 2 (ξ ) . (37)
Here TrΦ implies the functional trace for the 5D field Φ, so contains the summations
over the whole KK modes. M 2 (Φ) denotes the mass-square operator whose eigenvalues
µν
correspond to the KK mass spectrum of Φ, and Jj is the 4D Lorentz generator for
µν ρσ
spin j , normalized as tr(Jj Jj ) = C(j )(g µρ g νσ − g µσ g νρ ) where C(j ) = (0, 1, 2) for
j = (0, 1/2, 1). TrΨ0 denotes the functional trace for the zero mode of Ψ and Trξ ξ̄ is the
trace over the ghost fields.
The most convenient way to calculate TrΦ is to replace the KK summation by a contour
integration with an appropriately chosen pole function P (q). The pole function we will
use here has the form
N (q)
P (q) = , (38)
2N(q)
where N(q) has zeroes at q 2 = Mn2 (Φ) − m2Φ for Mn (Φ) denoting the nth KK mass
and mΦ denoting the bulk mass of Φ. Note that our pole function is different from the
pole function of [39,41] as the pole positions are shifted by m2Φ , which is mainly for the
simplicity of calculation. Using the analysis of Section 2, we find the following forms of
N -functions for 5D scalar and fermion fields:
1 2
Nφ ++ (q) = − q + µ++ µ++ sin(qπR) + µ++ − µ++ cos(qπR),
q
1
Nφ +− (q) = µ+− sin(qπR) + q cos(qπR) ,
q
1
Nφ −+ (q) = −µ−+ sin(qπR) + q cos(qπR) ,
q
1
Nφ −− (q) = sin(qπR),
q
1
NΨ ++ (q) = sin(qπR),
q
1
NΨ +− (q) = −M+− sin(qπR) + q cos(qπR) ,
q
1
NΨ −+ (q) = M−+ sin(qπR) + q cos(qπR) ,
q
1
NΨ −− (q) = sin qπR, (39)
q
where the scalar and fermion mass parameters µzz , µzz and Mzz are defined in (2) and

(13). The N -functions for 5D vector fields Azz
M are also easily found to be
1
NA++ (q) = sin(qπR),
q
NA+− (q) = cos(qπR), NA−+ (q) = cos(qπR),
1
NA−− (q) = sin(qπR), (40)
q
Fig. 1. Contour C1 in the complex q-plane. Bold dots represent poles at q 2 = Mn2 (Φ) − m2Φ , where Mn is the
KK mass eigenvalue and mΦ is the bulk mass of 5D field Φ. Note that a 4D state with Mn < mΦ appears as a
pole at the imaginary axis.

where the boundary conditions of Azz
M are given by
zz
µ (−y) = zAµ (y),
Azz µ (−y + πR) = z Aµ (y + πR),
zz
Azz
zz
y (−y) = −zAy (y),
Azz y (−y + πR) = −z Ay (y + πR).
zz
Azz
For the pole function defined as above, it is straightforward to find
µν
Tr ln −D 2 + M 2 (Φ) + Fµν Jj

dq d 4p a
= P (q) A (−p)Aaν (p)Ta (Φ)
2πi (2π)4 µ
C1

d 4 k g µν ((p + k)2 + q 2 + m2Φ ) − 12 (p + 2k)µ (p + 2k)ν
× d(j )
(2π)4 (k 2 + q 2 + m2Φ )((p + k)2 + q 2 + m2Φ )
− 2C(j )(p2 g µν − pµ pν )

d 4k 1
×
(2π)4 (k 2 + q 2 + m2Φ )((p + k)2 + q 2 + m2Φ )

d 4p
≡i Ga (p)Aaµ (−p)(p2 g µν − pµ pν )Aaν (p), (41)
(2π)4
where d(j ) = (1, 4, 4) and C(j ) = (0, 1, 2) for j = (0, 1/2, 1), and Ta (X) = Tr(Ta2 (X)) is
the Dynkin index of the gauge group representation X. Here the contour C1 is depicted in
Fig. 1. To regulate the divergent part of the above integral, we split the pole function into
two parts:
P (q) = P̃ (q) + P∞ (q), (42)
where P̃ → O(q −2 ) at |q| → ∞. Then P∞ is given by
A iπR
P∞ (q) = − − Im(q) , (43)
q 2
Fig. 2. The contour C1 on the upper half-plane can be deformed into C2 without touching any singularity. The
poles at the imaginary axis do not overlap with the branch cut unless there is a tachyon state.
where (x) = x/|x| and A is a real constant depending on the Z2 × Z2 parity of the
corresponding 5D field:
A = (−1/2, 0, 0, 1/2)
for Z2 × Z2 parity (ZΦ , ZΦ
) = (++, +−, −+, −−). With the decomposition (42), all UV
divergences appear in the contribution from P∞ in a manner allowing simple dimensional

regularization.
The 4D momentum integral d 4 k in (41) exhibits a branch cut on the imaginary axis
of q for p2 > 0. For the contribution from P̃ , one can change the contour as in Fig. 2
since the contribution from the infinite half-circle vanishes. Note that the poles of P̃ on
the imaginary axis do not overlap with the branch cut as long as there is no tachyon. After
integrating by part, we find that the part of Ga from P̃ is given by

Ta (Φ) 1
Ga (P̃ ) = d(j ) − 2C(j ) F (q)
8π 2 6
q→i∞
1

Ta (Φ) 1
− dx d(j )(1 − 2x) − 2C(j ) F (q)
2
,
8π 2 2 q=i x(1−x)p 2+m2 Φ
0
(44)
where
1 iπR
F (q) = ln N + A ln q + q.
2 2
In fact, F (q)|q→i∞ turns out to be vanishing in the cases we are now considering.
The contribution from P∞ includes the log divergence from the pole term 1/q. This
can be regulated by the standard dimensional regularization of 4D momentum integral,
d 4 p → d D p, yielding a 1/(D − 4) pole. On the other hand, the step-function contribution
from (Im(q)) involves a 5D momentum integral which is linearly divergent, but it simply
gives a finite result in dimensional regularization. Note that linearly divergent correction
to 1/ga2 depends highly on the used regularization scheme. Namely, the coefficient γa of
Eq. (35) is regularization scheme dependent, and γa = 0 in the dimensional regularization

scheme we are currently using. However, this does not have any special meaning since the
physical amplitudes are always expressed in terms of the scheme-independent combination
πR γa
2
+ ΛπR.
g5a 24π 3
Adding the divergent contribution from P∞ to the finite part Ga from P̃ , we obtain
1

Ta (Φ) 1 1
Ga = dx − d(j )(1 − 2x) + 2C(j )
2
ln N
8π 2 2 2 q=i x(1−x)p 2+m2 Φ
0
1

1 1
+A dx − d(j )(1 − 2x) + 2C(j )
2
. (45)
2 D−4
0
Using the above result, we find that the KK threshold corrections from a 5D complex
scalar φ, 5D Dirac fermion Ψ and 5D vector AM are given by
Λ
∆a (φ) + baφ ln
p
1
1 ++ u2 p 2
= Ta (φ ) ln Λ − 3 du F (u) ln Nφ ++ i + m++
2
6 4
0
1
+− u2 p 2
− 3Ta (φ ) du F (u) ln Nφ +− i + m2+−
4
0
1
−+ u2 p 2
− 3Ta (φ ) du F (u) ln Nφ −+ i + m2−+
4
0
1
u2 p 2
− Ta (φ −− ) ln Λ + 3 du F (u) ln Nφ −− i + m2−− ,
4
0
Λ
∆a (Ψ ) + baΨ ln
p
1
1 ++ u2 p 2
= Ta (Ψ ) −2 ln p + 3 du G(u) ln NΨ ++ i + M2++
3 4
0
1
+− u2 p 2
+ Ta (Ψ ) +3 du G(u) ln NΨ +− i + M2+−
4
0
1
−+ u2 p 2
+ Ta (Ψ ) +3 du G(u) ln NΨ −+ i + M2−+
4
0
1
u2 p 2
+ Ta (Ψ −− ) −2 ln p + 3 du G(u) ln NΨ −− i + M2−− ,
4
0
Λ
∆a (AM ) + baA ln
p
1

1 ++ iu
= Ta A M −23 ln Λ + 44 ln p + du K(u) ln NA++ p2
12 2
0
1
iu
+ Ta A+−
M du K(u) ln NA+− p2
2
0
1
iu
+ Ta A−+
M du K(u) ln NA−+ p2
2
0
1

iu
+ Ta A−−
M 23 ln Λ − 2 ln p + du K(u) ln NA−− p2 ,
2
0
φ ψ
where ba , baand baA are the beta function coefficients for the massless modes from φ, Ψ
and AM , respectively, and
1/2
F (u) = u 1 − u2 ,

2 1/2
−1/2
G(u) = u 1 − u − u 1 − u2 ,
1/2 −1/2
K(u) = −9u 1 − u2 + 24u 1 − u2 .

For the case that p2 is much smaller than the lowest nonzero KK mass, the above results
are simplified to yield
21 ++
∆a = Ta AM + Ta A−− M ) ln(ΛπR)
12

1 ++ Λ(em++ πR − e−m++ πR )
− Ta (Φ ) ln
6 2m++

1 1
− Ta (φ ++ ) ln (m++ + µ++ )(m++ − µ++ )em++ πR
6 2m++ Λ

−m++ πR

− (m++ − µ++ )(m++ + µ++ )e

1 (m+− + µ+− )em+− πR + (m+− − µ+− )e−m+− πR

− Ta (φ +− ) ln
6 2m+−
1 −+ (m−+ − µ−+ )em−+ πR + (m−+ + µ−+ )e−m−+ πR

− Ta (φ ) ln
6 2m−+
−m
1 −− Λ(e −− − e −− )
m πR πR
− Ta (φ ) ln
6 2m−−
M++ πR − e −M++ πR )
2 Λ(e
− Ta (Ψ ++ ) ln
3 2M++
2 2
− Ta (Ψ +− ) ln e−M+− πR − Ta (Ψ −+ ) ln eM−+ πR
3 3
M
2 Λ(e −− πR − e−M−− πR )
− Ta (Ψ −− ) ln , (46)
3 2M−−
where mzz , µzz and µzz denote the bulk and brane masses of φ, and Mzz is the kink
mass of Ψ . Here Φ ++ is a 5D complex scalar field having a zero mode, i.e., a scalar field
with µ = µ = m, and φ ++ stands for complex scalar fields without zero mode. The 4D
beta function coefficients ba are given by
11 ++ 1 −− 1 2 2
ba = − Ta AM + Ta AM + Ta (Φ ++ ) + Ta (Ψ ++ ) + Ta (Ψ −− ), (47)
3 6 3 3 3
which can be easily understood by noting that A++M gives a massless 4D vector, AM a
−−
massless real 4D scalar, and Ψ ±± a massless 4D chiral fermion. Then comparing the above
∆a to the expression of 4D Yukawa couplings in (27), one easily finds that generically

∆a = O ln(yipq ) (48)
if the 4D Yukawa couplings yipq are generated by quasi-localization.
It is straightforward to find an expression of ∆a for supersymmetric 5D theories using
the above result. In supersymmetric theories, there can be two type of bulk fields: vector
multiplet V containing a 5D vector AM , a Dirac-spinor λ and a real scalar Σ, and
hypermultiplet H containing a Dirac spinor Ψ and two complex scalars φ, φ . The mass
parameters and Z2 × Z2 boundary conditions of component fields are given by
zz z̃z̃

V zz = Azz M , λ (M = 0), Σ (µ = µ = m = 0) ,

Hzz = φ zz (µ = µ = m = −M), φ z̃z̃ (µ = µ = m = M), Ψ zz (M) , (49)
where z = −z̃ = ±1, z = −z̃ = ±1. Here m, µ and µ denote the bulk and brane masses
of scalar field, and M is the kink mass of Dirac fermion field. We then find

(∆a )SUSY = Ta (V ++ ) + Ta (V −− ) ln(ΛπR) − Ta (H++ )

Λ(eM++ πR − e−M++ πR )
× ln
2M++
−M+− πR
+−
− Ta (H ) ln e − Ta (H−+ ) ln eM−+ πR

−− Λ(eM−− πR − e−M−− πR )
− Ta (H ) ln (50)
2M−−
and the 4D beta function coefficients

(ba )SUSY = −3Ta (V ++ ) + Ta (V −− ) + Ta (H++ ) + Ta (H−− ). (51)
In fact, one can obtain (∆a )SUSY using the 4D effective supergravity method discussed in
[33,35]. We confirmed that the result from 4D effective supergravity agrees with the above
result which was obtained from a direct calculation of KK threshold correction.
With the above result on ∆a , one can obtain the low energy gauge couplings at p MKK
which are determined as (35) at one-loop approximation. In most of 5D orbifold field
theories, we have
MKK Λ
(52)
MW MKK
by many orders of magnitude, where MW is the weak scale and MKK is the lightest KK
mass. Then the dominant part of higher order corrections (beyond one-loop) to low energy
couplings at MW come from the energy scales below MKK , which can be systematically
computed within 4D effective theory. To include those higher order corrections, one can
start with the matching condition at MKK :

1 1 1 Λ
= + ∆a + ba ln , (53)
ga2 (MKK ) ga2 bare 8π 2 MKK
where ∆a are given by (46), and then subsequently perform two-loop renormalization
group (RG) analysis over the scales between MKK and MW . If there is a massive particle
with mass M between MKK and MW , one needs to stop at M to integrate out this massive
particle, which would yield a new matching condition at M. For a given model, one can
repeat this procedure to find the gauge couplings at MW , and compare the results with the
experimentally measured values.
In some case, there can be another large mass gap between the lightest KK mass MKK
and the next lightest KK mass MKK . For instance, as we have noticed in Section 2, Ψ +−
has two light KK modes with a Dirac mass 2Me−MπR when its kink mass MπR 1.
Then the lightest KK mass is given by MKK = 2Me−MπR , while the next lightest KK
mass MKK = 1/R which corresponds to the mass of the first KK mode of gauge fields. If
MπR is large enough, so that

MKK eMπR Λ
= = ΛR, (54)
MKK 2MR MKK
the next important higher order corrections would come from energy scales between MKK
and MKK . Those next important higher order corrections can be included by performing
the two-loop RG analysis starting from MKK . The corresponding matching condition at

MKK is given by

1 1 1 Λ
) = g2
ga2 (MKK
+
8π 2 a
∆ + ba ln
MKK
, (55)
a bare
where

Λ
ba = ba + δba , ∆a = ∆a − δba ln (56)
MKK
for ∆a given by (46). Here ba denote the one-loop beta function coefficients for zero
modes, while δba denote the coefficient for the lightest KK states.
5. Application to orbifold GUT
In the previous section, we have discussed one-loop gauge couplings in generic 5D

orbifold field theory, including the KK threshold correction ∆a as

1 1 1 Λ
= + ∆ a + b a ln (a = 1, 2, 3). (57)
ga2 (p) ga2 bare 8π 2 p
The bare couplings here consist of several pieces which are not calculable within orbifold
field theory:

1 πR γa
= 2 + ΛπR + κa + κa . (58)
ga2 bare g5a 24π 3
So although it is an well-defined relation between the bare parameters and measurable
quantities, (57) does not give any useful prediction unless additional information on bare
couplings is provided. In orbifold GUT which is strongly coupled at Λ, g5a 2 and γ are
a
universal as a consequence of unified gauge symmetry in bulk, and both κa and κa are of
the order of 1/8π 2 as the theory is strongly coupled at Λ [29,30]. We then have

1 1 1
= 2 +O (a = 1, 2, 3). (59)
ga2 bare gGUT 8π 2
With this information on bare couplings, one would be able to predict for instance the
value of QCD coupling constant at MZ in terms of the measured values of the electroweak
coupling constants at MZ and the KK threshold corrections a computed in the previous
section.
Let us discuss in more detail the effect of KK threshold on the predicted value of the
QCD coupling constant. To this end, it is convenient to consider
ηa
, (60)
a
αa (p)
where αa = ga2 /4π and ηa are the coefficients determined by

η3 = 1, ηa = 0, ηa ba = 0. (61)
a a
It is then straightforward to find

1 1 1
= + (η1 ∆1 + η2 ∆2 + ∆3 ), (62)
α3 (MZ ) α3 (MZ ) 0 2π
where

1 η1 η2
=− + + δLE (63)
α3 (MZ ) 0 α1 (M Z ) α2 (M Z)
for the experimentally measured electroweak coupling constants α1 (MZ ) = 0.0169 and
α2 (MZ ) = 0.0338, and
ηa ηa
δLE = − (64)
a
αa (MKK ) a
αa (MZ )
can be determined by the RG analysis below MKK once the zero-mode spectra are known.
Note that when expanded in powers of ∆a /8π 2 , δLE is independent of ∆a at leading order.
If zero-mode spectra correspond to the standard model (SM), we have
115 333
η1 = , η2 = − , (65)
218 218

1 1
= 15.3 + O (66)
α3 (MZ ) 0 π
for MKK = 1013 –1015 GeV. Here we consider a rather wide range of MKK to cover the
case that the lightest KK mass is suppressed by a small localization factor as MKK =
2Me−MπR . It turns out that the numerical result is insensitive to MKK , e.g., the variation
of 1/α3 is within O(1/π) even when MKK varies by several orders of magnitude. In the
above, we have used the approximation scheme to include two-loop RG evolution below
MKK , and then the uncertainty of O(1/π) is from the dependence of δLE on MKK as well
as from the piece higher order in a /8π 2 . For more interesting case that zero-mode spectra
correspond to the minimal supersymmetric standard model (MSSM),
5 12
η1 = , η2 = − , (67)
7 7

1 1
= 7.8 + O . (68)
α3 (MZ ) 0 π
Here we have assumed the superparticle masses MSUSY at 0.3 ∼ 1 TeV, and then the
uncertainty of O(1/π) is mainly from the variation of superparticle threshold effects at
MSUSY .
To see the importance of KK threshold corrections more explicitly, let us consider a
class of 5D SU(5) orbifold GUTs whose effective 4D theory is given by the MSSM. To
break SU(5) by orbifolding, Z2 × Z2 is embedded into SU(5) as
Z2 = diag(+1, +1, +1, +1, +1),

Z2 = diag(+1, +1, +1, −1, −1), (69)
leading to the following orbifold boundary conditions of the 5D vector multiplet V
containing the SU(5) gauge fields:
V = (8, 1)(++)
0 + (1, 3)(++)
0 + (1, 1)(++)
0 + (3, 2)(+−) (+−)
−5/6 + (3̄, 2)5/6 , (70)
where the SU(5) adjoint representation is decomposed into the representations of the SM
gauge group. Obviously then the bulk SU(5) is broken down to SU(3) × SU(2) × U (1) at
y = πR, while it is unbroken at y = 0.
The model contains matter hypermultiplets Fp (5̄), Fp (5̄), Tp (10) and Tp (10) (p =
1, 2, 3) with kink masses MFp , MFp , MTp and MTp , and also the Higgs hypermultiplets
H (5) and H (5̄) with kink masses MH and MH , where the numbers in bracket mean the
SU(5) representation. We assign the Z2 × Z2 parities of these hypermultiplets as
Z2 (Fp ) = Z2 (Fp ) = Z2 (Tp ) = Z2 (Tp ) = Z2 (H ) = Z2 (H ) = 1,

Z2 (Fp ) = −Z2 (Fp ) = Z2 (Tp ) = −Z2 (Tp ) = Z2 (H ) = Z2 (H ) = −1, (71)
and then the orbifold boundary conditions of matter and Higgs hypermultiplets are given
by
F = (3̄, 1)(+−) (++)

1/3 + (1, 2)−1/2 , F = (3̄, 1)(++) (+−)
1/3 + (1, 2)−1/2 ,
(++) (+−) (+−)
T = (3, 2)1/6 + (3̄, 1)−2/3 + (1, 1)1 ,
T = (3, 2)1/6 + (3̄, 1)−2/3 + (1, 1)1
(+−) (++) (++)
,
H = (3, 1)(+−) (++)
−1/3 + (1, 2)1/2 , H = (3̄, 1)(+−) (++)
1/3 + (1, 2)−1/2 . (72)
Then using (68) and (50), we find

1 1 1
= 7.8 + [∆gauge + ∆higgs + ∆matter ] + O , (73)
α3 (MZ ) 2π π
where
1 3
∆gauge = ln(πRΛ) (74)
2π 7π
corresponds to the KK threshold correction from the 5D vector multiplet,

9 sinh πRMH
∆higgs = ln + πRMH
14 πRMH

sinh πRMH
+ ln + πRMH , (75)
πRMH
is the correction from Higgs hypermultiplets, and

9 sinh πRMFp
∆matter = ln + πRMFp
14 p πRMFp
sinh πRM

Fp
− ln − πRMFp
πRMFp

3 sinh πRMTp
+ ln + πRMTp
2 p πRMTp
sinh πRM

Tp
− ln − πRMTp (76)
πRMTp
is the correction from matter hypermultiplets.
For ΛπR ≈ 102 , we have

1
∆gauge ≈ 0.8. (77)
2π
Then the predicted value of 1/α3 would be very close to the experimental value:

1
= 8.55 ± 0.15, (78)
α3 (MZ ) exp
if ∆higgs + ∆matter is negligible. In other words, the KK threshold correction from Higgs
and matter hypermultiplets should be negligible, i.e.,

1 1
(∆higgs + ∆matter ) O , (79)
2π π
in order for the 5D orbifold GUT under consideration to be consistent with observation.
Obviously, for the class of models under consideration, a hypermultiplet with MπR 1
gives a large threshold correction ∆(1/α3 ) = O(MπR/2π) which can make the model in-
consistent with the observation. For instance, in the model of [9], the Higgs zero modes
are quasi-localized at y = 0 by having MH πR = 11.5 and MH πR = 6.9, the 1st and 3rd
generation matters are brane fields at y = πR and y = 0, respectively, and the 2nd gen-
eration matters come from hypermultiplets with vanishing kink masses. This model then
gives 1/α3 (MZ ) = 10.9 which is too large to be consistent with the experimental value.
A simple way to avoid a too large ∆higgs is to assume that both of the Higgs
hypermultiplets have MπR −1, for instance MH πR ≈ MH πR ≈ −10. In this case,
the Higgs zero modes are localized at y = πR, and
1 9
∆higgs ≈ − ln MH MH πR ≈ −0.45, (80)
2π 14π
which is small enough not to spoil the successful gauge unification. However, for matter
hypermultiplets, to generate the hierarchical 4D Yukawa couplings through dynamical
quasi-localization, one needs to localize heavy and light generations at different locations.
This means that some kink masses should be positive, while some others are negative.
One then needs a nontrivial cancellation between the corrections from different matter
hypermultiplets in order for ∆matter ≈ 0. In this regard, an interesting possibility is that
MFp = MFp , MTp = MTp , (81)

which obviously lead to
1
∆matter = 0. (82)
2π
The relation (81) between hypermultiplet masses can be considered as a consequence of
global SU(2)H symmetry under which (Fp , Fp ) and (Tp , Tp ) transform as a doublet. This
SU(2)H symmetry is broken down to U (1)3B+L at y = πR by the orbifolding boundary
conditions (71). It is then straightforward to introduce a dynamics at y = πR which breaks
U (1)3B+L spontaneously down to the matter-parity (−1)3B+L . This is in fact necessary
to generate nonzero Majorana masses of neutrinos. Then the brane interactions at y = πR
Table 1
A set of hypermultiplet masses which give realistic fermion masses and CKM mixing while satisfying (79) for
successful gauge unification. Here n = 0, 1, 2
MTp /M0 MT /M0 MFp /M0 MF /M0
p p
(4, 2, 0) (4, 2, 0) (n, n, n) (n, n, n)
(5, 1, 0) (3, 3, 0) (2, 0, 1) (1, 2, 0)
(3, 3, 0) (5, 1, 0) (0, 2, 1) (2, 0, 1)
(6, 6, 0) (2, 4, 0) (4, 0, 2) (0, 4, 2)
(5, 1, 0) (3, 3, 0) (3, 1, 2) (1, 3, 2)
(3, 3, 0) (3, 3, 0) (1, 3, 2) (3, 1, 2)
(2, 4, 0) (6, 0, 0) (0, 4, 2) (4, 0, 2)
are constrained only by the SM gauge group, 4D N = 1 supersymmetry and the R-parity
(−1)3B+L+2s (s = spin).
The KK threshold corrections (75) and (76) to 1/α3 from Higgs and matter hypermul-
tiplets have a correlation with the 4D Yukawa couplings given by (27). So the condition
(79) for successful gauge unification provides some restriction on the possible forms of
4D Yukawa couplings. However still it is not so difficult to construct models to produce
the correct form of Yukawa couplings through quasi-localization, while satisfying (79). To
see this, consider a class of models with MH πR −1 and MH πR −1, in which the
Higgs zero modes are quasi-localized at y = πR. The quark and lepton Yukawa couplings
are assumed to arise from the following brane interactions at y = πR:

1

d x d 2 θ δ(y − πR) 3/2 λU
5
pq H Qp Uq + λpq H Qp Dq + λpq H Lp Eq , (83)
c D c E c
Λ
where the Higgs doublets H1 and H2 come from H and H , respectively, the lepton
doublets Lp are from Fp , the lepton singlets Epc are from Tp , the quark doublets Qp are
from Tp , the quark singlets Upc and Dpc are from Tp and Fp , respectively. Then according
to the discussion of Section 2, the physical Yukawa couplings of quarks and leptons are
given by

U
ypq = Z(MH )Z(MTp )Z(MTq ) λU pq ,

D
ypq = Z(MH )Z(MTp )Z(MFq ) λD pq ,

E
ypq = Z(MH )Z(MFp )Z(MTq ) λE pq , (84)
where
2M 1
Z(M) = . (85)
Λ e2MπR − 1
If we further assume that all hypermultiplet masses are quantized in an appropriate unit,
then the hypermultiplet masses tabulated in Table 1 give the correct quark and lepton
masses as well as the correct CKM mixing angles, while satisfying (79) for successful
gauge unification. Here the unit mass M0 is defined to given by e−M0 πR = Cabibbo
angle ≈ 0.2.
6. Conclusion
In this paper, we have examined the KK threshold corrections to low energy gauge
couplings ga2 from bulk matter fields whose zero modes are dynamically quasi-localized
to generate hierarchical 4D Yukawa couplings. We derived the explicit form of threshold
corrections in generic 5D orbifold field theory on S 1 /Z2 × Z2 , and found that the
corrections to 1/ga2 are of the order of ln(y)/8π 2 where y denotes 4D Yukawa couplings
generated by quasi-localization. So generically quasi-localization significantly affects
gauge coupling unification. We then applied the results to 5D orbifold GUT, and discussed
the conditions for a 5D GUT to generate hierarchical Yukawa couplings without spoiling
successful gauge coupling unification. Some examples of such 5D GUTs are presented in
Table 1.
Acknowledgements
This work is supported by KRF PBRG 2002-070-C00022.
References
[1] N. Arkani-Hamed, M. Schmaltz, Hierarchies without symmetries from extra dimensions, Phys. Rev. D 61
(2000) 033005.
[2] E.A. Mirabelli, M. Schmaltz, Yukawa hierarchies from split fermions in extra dimensions, Phys. Rev. D 61
(2000) 113011.
[3] G.R. Dvali, M.A. Shifman, Families as neighbors in extra dimension, Phys. Lett. B 475 (2000) 295.
[4] D.E. Kaplan, T.M. Tait, Supersymmetry breaking, fermion masses and a small extra dimension, JHEP 0006
(2000) 020.
[5] N. Arkani-Hamed, T. Gregoire, J. Wacker, Higher-dimensional supersymmetry in 4D superspace,
JHEP 0203 (2002) 055.
[6] D.E. Kaplan, T.M. Tait, New tools for fermion masses from extra dimensions, JHEP 0111 (2001) 051.
[7] M. Kakizaki, M. Yamaguchi, Proton decay, fermion masses and texture from extra dimensions in SUSY
GUTs, hep-ph/0110266.
[8] N. Haba, N. Maru, (S)fermion masses in fat brane scenario, Phys. Rev. D 66 (2002) 055005.
[9] A. Hebecker, J. March-Russell, The flavour hierarchy and see-saw neutrinos from bulk masses in 5d orbifold
GUTs, Phys. Lett. B 541 (2002) 338.
[10] Y. Grossman, G. Perez, Realistic construction of split fermion models, Phys. Rev. D 67 (2003) 015011.
[11] W.F. Chang, J.N. Ng, CP violation in 5D split fermions scenario, JHEP 0212 (2002) 077.
[12] R. Kitano, T.j. Li, Flavor hierarchy in SO(10) grand unified theories via 5-dimensional wave-function
localization, Phys. Rev. D 67 (2003) 116004.
[13] K. Choi, D.Y. Kim, I.W. Kim, T. Kobayashi, Supersymmetry breaking in warped geometry, hep-ph/0305024.
[14] C. Biggio, F. Feruglio, I. Masina, M. Perez-Victoria, Fermion generations, masses and mixing angles from
extra dimensions, hep-ph/0305129.
[15] Y. Kawamura, Triplet–doublet splitting, proton stability and extra dimension, Prog. Theor. Phys. 105 (2001)
999.
[16] G. Altarelli, F. Feruglio, SU(5) grand unification in extra dimensions and proton decay, Phys. Lett. B 511
(2001) 257.
[17] L.J. Hall, Y. Nomura, Gauge unification in higher dimensions, Phys. Rev. D 64 (2001) 055003;
L.J. Hall, Y. Nomura, Gauge coupling unification from unified theories in higher dimensions, Phys. Rev.
D 65 (2002) 125012;
L.J. Hall, Y. Nomura, A complete theory of grand unification in five dimensions, Phys. Rev. D 66 (2002)
075004.
[18] A. Hebecker, J. March-Russell, A minimal S(1)/(Z(2) × Z (2)) orbifold GUT, Nucl. Phys. B 613 (2001) 3.
[19] A. Hebecker, J. March-Russell, The structure of GUT breaking by orbifolding, Nucl. Phys. B 625 (2002)
128.
[20] L.J. Hall, H. Murayama, Y. Nomura, Wilson lines and symmetry breaking on orbifolds, Nucl. Phys. B 645
(2002) 85.
[21] T. Asaka, W. Buchmuller, L. Covi, Gauge unification in six dimensions, Phys. Lett. B 523 (2001) 199.
[22] L.J. Hall, Y. Nomura, T. Okui, D.R. Smith, SO(10) unified theories in six dimensions, Phys. Rev. D 65
(2002) 035008.
[23] R. Dermisek, A. Mafi, SO(10) grand unification in five dimensions: proton decay and the mu problem, Phys.
Rev. D 65 (2002) 055002.
[24] H.D. Kim, J.E. Kim, H.M. Lee, TeV scale 5D SU(3)W unification and the fixed point anomaly cancellation
with chiral split multiplets, JHEP 0206 (2002) 048.
[25] H.D. Kim, S. Raby, Unification in 5D SO(10), JHEP 0301 (2003) 056;
H.D. Kim, S. Raby, Neutrinos in 5D SO(10) unification, JHEP 0307 (2003) 014.
[26] S. Weinberg, Effective gauge theories, Phys. Lett. B 91 (1980) 51;
L.J. Hall, Grand unification of effective gauge theories, Nucl. Phys. B 178 (1981) 75.
[27] K. Choi, String unification and threshold effects, Phys. Rev. D 37 (1988) 1564;
V.S. Kaplunovsky, One loop threshold effects in string unification, Nucl. Phys. B 307 (1988) 145;
V.S. Kaplunovsky, One loop threshold effects in string unification, Nucl. Phys. B 382 (1992) 436, Erratum.
[28] T. Gherghetta, A. Pomarol, Bulk fields and supersymmetry in a slice of AdS, Nucl. Phys. B 586 (2000) 141.
[29] Z. Chacko, M.A. Luty, E. Ponton, Massive higher-dimensional gauge fields as messengers of supersymmetry
breaking, JHEP 0007 (2000) 036.
[30] Y. Nomura, Strongly coupled grand unification in higher dimensions, Phys. Rev. D 65 (2002) 085036.
[31] C.D. Froggatt, H.B. Nielsen, Hierarchy of quark masses, Cabibbo angles and CP violation, Nucl. Phys.
B 147 (1979) 277.
[32] A. Ceresole, G. Dall’Agata, General matter coupled N = 2, D = 5 gauged supergravity, Nucl. Phys. B 585
(2000) 143.
[33] K. Choi, H.D. Kim, I.W. Kim, Gauge coupling renormalization in orbifold field theories, JHEP 0211 (2002)
033.
[34] R. Contino, L. Pilo, R. Rattazzi, E. Trincherini, Running and matching from 5 to 4 dimensions, Nucl. Phys.
B 622 (2002) 227.
[35] K. Choi, H.D. Kim, I.W. Kim, Radius dependent gauge unification in AdS(5), JHEP 0303 (2003) 034.
[36] K. Agashe, A. Delgado, R. Sundrum, Gauge coupling renormalization in RS1, Nucl. Phys. B 643 (2002)
172.
[37] R. Contino, P. Creminelli, E. Trincherini, Holographic evolution of gauge couplings, JHEP 0210 (2002) 029.
[38] W.D. Goldberger, I.Z. Rothstein, Effective field theory and unification in AdS backgrounds, Phys. Rev. D 68
(2003) 125011.
[39] K. Choi, I.W. Kim, One loop gauge couplings in AdS(5), Phys. Rev. D 67 (2003) 045005.
[40] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Anomalies on orbifolds, Phys. Lett. B 516 (2001) 395;
C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Anomalies in orbifold field theories, Phys. Lett. B 525
(2002) 169.
[41] S. Groot Nibbelink, Dimensional regularization of a compact dimension, Nucl. Phys. B 619 (2001) 373;
R. Contino, A. Gambassi, J. Math. Phys. 44 (2003) 570.
Finiteness of quantum gravity coupled with matter

in three spacetime dimensions
Damiano Anselmi
Dipartimento di Fisica “E. Fermi”, Università di Pisa, and INFN, Pisa, Italy
Received 10 October 2003; accepted 18 March 2004
Abstract
As it stands, quantum gravity coupled with matter in three spacetime dimensions is not finite. In
this paper I show that an algorithmic procedure that makes it finite exists, under certain conditions.
To achieve this result, gravity is coupled with an interacting conformal field theory C. The Newton
constant and the marginal parameters of C are taken as independent couplings. The values of the other
irrelevant couplings are determined iteratively in the loop and energy expansions, imposing that their
beta functions vanish. The finiteness equations are solvable thanks to the following properties: the
beta functions of the irrelevant couplings have a simple structure; the irrelevant terms made with the
Riemann tensor can be reabsorbed by means of field redefinitions; the other irrelevant terms have,
generically, non-vanishing anomalous dimensions. The perturbative expansion is governed by an
effective Planck mass that takes care of the interactions in the matter sector. As an example, I study
gravity coupled with Chern–Simons U (1) gauge theory with massless fermions, solve the finiteness
equations and determine the four-fermion couplings to two-loop order. The construction of this paper
does not immediately apply to four-dimensional quantum gravity.
1. Introduction
Gravity is not power-counting renormalizable. This might mean that quantum field
theory is inadequate to quantize gravity or, more conservatively, that power-counting
renormalizability is not an essential feature of the theories that describe nature. At
the theoretical level, there exist power-counting non-renormalizable theories that can be
quantized successfully, such as the four-fermion models in three spacetime dimensions [1]
in the large-N expansion. Moreover, a theory that is not power-counting renormalizable
E-mail address: anselmi@df.unipi.it (D. Anselmi).
doi:10.1016/j.nuclphysb.2004.03.024
D. Anselmi / Nuclear Physics B 687 (2004) 124–142 125
does not necessarily violate fundamental physical principles and so it cannot be discarded
a priori.
In four dimensions, ’t Hooft and Veltman showed that pure gravity is finite to one-loop
order [2], but finiteness is spoiled by the coupling with matter. Goroff and Sagnotti showed
that gravity is not finite to two-loop order [3], even in the absence of matter. These results
depressed the hopes to find a finite theory of quantum gravity.
To some extent, the problem of finiteness is simpler in three spacetime dimensions. In
odd dimensions every theory is finite to one-loop order, because there are no logarithmic
one-loop divergences. So, the problem starts from two loops. Moreover, pure gravity in
three dimensions,

1 √
S= g R, (1.1)
2κ
propagates no graviton and is finite to all orders [4]. Indeed, since the Weyl tensor vanishes,
the Riemann tensor is a linear combination of the Ricci tensor and the scalar curvature. This
ensures that all possible counterterms can be reabsorbed by means of field redefinitions.
The issue of finiteness is non-trivial in three dimensions if gravity is coupled
with matter. In [5] I have proved that renormalization generates counterterms with
dimensionality greater than three, in general infinitely many. I recall here the main results
of that paper:
(1) the Lorentz–Chern–Simons term

1 a b c abc
ε µνρ
ωµ ∂ν ωρ + ωµ ων ωρ ε
a a
, (1.2)
3
is not induced by renormalization, so there exists a subtraction scheme where it is absent at
each order of the perturbative expansion, if it is absent at the classical level. This property
can be proved combining a power-counting analysis of the complete theory with properties
of the trace anomaly of the matter sector embedded in external gravity. It is important
that the Lorentz–Chern–Simons term is not turned on by renormalization, because three-
dimensional gravity with a Lorentz–Chern–Simons term, known as “topologically massive
gravity” [6], is physically inequivalent to the theory without it.
(2) I have then considered a specific model, gravity coupled with Chern–Simons U (1)
gauge theory and massless fermions and proved by explicit computation that a four-fermion
counterterm is induced by radiative corrections to the second order in the loop expansion
and first order in the κ expansion, namely,
5κg 4 nf e
− (ψ̄γ a ψ)2 . (1.3)
384π 2ε 4
The result (1.3) is written up to subleading corrections in 1/nf , where nf is the number
of complex two-component spinors. This counterexample is sufficient to conclude that, as
it stands, quantum gravity coupled with matter in three spacetime dimensions is not finite.
The purpose of this paper is to show that, under certain conditions, quantum gravity
coupled with matter in three spacetime dimensions can be quantized in a unique way as a
finite theory.
A sketch of the idea is as follows. Gravity is coupled with an interacting conformal field
theory C, subject to some restrictions. If λ denote an irrelevant coupling, i.e., the coupling
126 D. Anselmi / Nuclear Physics B 687 (2004) 124–142
multiplying an irrelevant Lagrangian term Oλ , then the beta function of λ a has a simple
structure. In particular, it is linear in λ:
βλ = λγλ + δλ . (1.4)
Here γλ is the anomalous dimension of Oλ , which depends only on the marginal couplings
of C, but not on the irrelevant couplings of the complete theory. Instead, δλ depends on
the marginal couplings C plus a finite number of irrelevant couplings, but not on λ itself.
The formula (1.4) is written in symbolic form. A precise treatment is presented in the next
section.
The finiteness equations βλ = 0 can be solved if γλ is nonzero or γλ and δλ are
simultaneously zero. I show that, generically, in three dimensions the finiteness equations
admit a solution, thanks to the properties of three-dimensional spacetime, in particular, the
absence of a propagating graviton. The Newton constant and the marginal couplings of C
are taken as independent couplings of the theory coupled with gravity. The values of the
other irrelevant couplings are uniquely determined solving the finiteness equations. This
can be done perturbatively.
The perturbative expansion in powers of the energy is valid for energies much smaller
than an effective Planck constant, obtained multiplying the Planck mass by a factor that
depends only on the matter subsector C.
After working out the general principles of this approach to finiteness, I illustrate the
quantization mechanism in the case of gravity coupled with Chern–Simons U (1) gauge
theory and massless fermions. I solve the finiteness conditions to the second order in the
loop expansion, first order in the κ expansion, and leading order in the 1/nf expansion.
The solution uniquely determines the values of the couplings multiplying the four-fermion
vertices.
The paper is organized as follows. In Section 2 I present the idea in the most
general terms, so that it can be applied, in principle, to every non-renormalizable theory.
Moreover, I study the conditions for finiteness (structure of the beta functions of the
irrelevant couplings, existence of solutions to the finiteness equations, etc.). In Section 3
I consider quantum gravity coupled with matter in three dimensions and show that the
finiteness equations admit generically one solution. In Section 4 I introduce the model
studied explicitly in the rest of the paper. I recall the regularization technique, some
renormalization properties, and the four-fermion divergent vertex calculated in Ref. [5].
In Section 5 I report the results concerning the two-loop self-renormalization of the four-
fermion vertices. In Section 6 I solve the finiteness equations and determine the values of
the irrelevant couplings that multiply the four-fermion vertices. The solution is contained
in formulas (6.2) and (8.1). In Section 7 I briefly discuss some obstacles that prevent a
straightforward generalization of the approach of this paper to quantum gravity in four
dimensions. Section 8 collects the conclusions and Appendix A contains some notation.
2. Structure of the beta functions of the irrelevant couplings and solutions of the
finiteness equations
I consider a generic power-counting non-renormalizable theory of interacting fields ϕ

in d dimensions, having a classical Lagrangian of the form

Ni
Lcl [ϕ] = L0 [ϕ, α] + κi λiI OiI (ϕ). (2.1)
i I =1
The first piece, L0 , denotes the power-counting renormalizable sector of the theory, with
couplings α. The theory L0 is assumed to be finite. For example, in the case of three-
dimensional quantum gravity coupled with matter, L0 is the sum of the free spin-2 kinetic
term and the Lagrangian of a conformal field theory C, which I call the matter sector of
the theory.
The objects OiI are a basis of (gauge-invariant) local Lagrangian terms with canonical
dimensionalities d + i in units of mass. The index i denotes the “level” of Oi
(irrelevant operators have positive levels, marginal operators have level 0 and relevant
operators have negative levels) and can be a non-negative integer or a half-integer.
The λiI denote a complete set of essential couplings, labelled by their level i plus
an index I that distinguishes the couplings of the same level (subject, in general, to
renormalization mixing). The essential couplings are the couplings that multiply a basis
of Lagrangian terms that cannot be renormalized away or into one another by means of
field redefinitions [7].
The parameter κ is an auxiliary constant with dimensionality −1 in units of mass.
Every λ is dimensionless. For simplicity, I assume also that the theory (2.1) does not
contain masses, the cosmological constant and super-renormalizable parameters (couplings
with strictly positive dimensionalities in units of mass), because they form dimensionless
quantities when they are multiplied by suitable powers of the irrelevant couplings. The beta
functions can depend non-polynomially on such dimensionless combinations, which adds
unnecessary complications to the treatment.
The redundancy of the constant κ is exhibited by the invariance of (2.1) under the scale
symmetry
λiI → Ω −i λiI , κ → Ωκ. (2.2)
2.1. Structure of the beta functions
The beta function of λiI transforms like λiI under the scale symmetry (2.2) and cannot
contain negative powers of the λ’s. Therefore, the structure of βiI is
Nj
niI
βiI = f{niI } (α) (λj J ) jJ , (2.3)
jJ
{niI j i J =1
jJ }
where the f{niI } (α)s are functions of the marginal couplings and the sum is performed over
jJ
the sets {niI
j J } of non-negative integers nj J such that
iI
Nj

j j J = i.
niI (2.4)
j i J =1
The constant κ, which is, by assumption, the only dimensionful parameter in the theory,
does not appear in the beta functions.
Due to (2.4), only a finite set of numbers niI j J can be greater than zero. This ensures that
the beta functions depend on the irrelevant couplings in a polynomial way. Special sets
j J } satisfying (2.4) are those where nj J is equal to one for j = i and some index J , zero
{niI iI
otherwise. It is useful to isolate this contribution from the rest, obtaining

Nj

Ni miI
βiI = γiI J (α)λiJ + δiI , δiI = f{miI } (α) (λj J ) jJ . (2.5)
jJ
J =1 {miI j <i J =1
jJ }
Now the sum is performed over the sets {miI

j J } of non-negative integers such that
Nj

j j J = i.
miI (2.6)
j <i J =1
The functions γiI J (α) are the entries of the matrix γi (α) of anomalous dimensions of the
operators OiI of level i. The second term of (2.5) collects the contributions of the operators
Oj J of levels j < i. Observe that (2.6) implies
Nj

j J 2,
miI (2.7)
j <i J =1
which means that the beta function of λi is at least quadratic in the irrelevant couplings
with j < i. A fortiori, the δiI vanish when all of the λiI vanish. Indeed, at λiI = 0 the
theory reduces to L0 [ϕ, α], which is finite by assumption. So, λiI = 0 ∀i, I must be a
trivial solution of the finiteness equations.
2.2. Finiteness equations
The finiteness equations are the conditions βiI = 0 for every i and I , namely,

Ni
γiI J (α)λiJ = −δiI . (2.8)
J =1
If γi denotes the Ni × Ni matrix having entries γiI J (α), let (γi |δi ) denote the Ni × (Ni + 1)
matrix obtained adding the column δiI to γi . The equation βiI = 0 admits solutions if and
only if the ranks of the matrices γi and (γi |δi ) are equal. Writing rank(γi ) = rank(γi |δi ) =
ni Ni , then the solution of βiI = 0 contains Ni − ni free parameters.
Simple situations in which (2.8) admits solutions are those in which the matrix γi
is invertible, or, if it is not invertible, suitable entries of the vector δi vanish. In some
cases a symmetry ensures that certain irrelevant operators have δ identically zero. I call
these Lagrangian terms protected. The beta functions of the protected operators can be
set to zero in a straightforward way. If a protected operator is finite, i.e., its anomalous
dimension vanishes, then its coupling λ remains unconstrained. Examples of protected
operators are the chiral operators in four-dimensional supersymmetric theories [8]. The
anomalous dimensions of the chiral operators are generically different from zero in N = 1
supersymmetric theories, but they can vanish in families of finite N = 2 and N = 4

theories. These cases are not of primary interest for the investigation of this paper. I briefly
come back to this issue in the next section, but more details can be found in Ref. [9].
It is convenient to isolate the protected operators from the rest and concentrate the
search for solutions of the finiteness equations in the remaining subclass of irrelevant
terms. For simplicity, it is also convenient to set the couplings of the protected operators to
zero even if their anomalous dimensions vanish. Indeed, it is always possible to turn those
couplings on at a later stage. This operation is studied in [9] and defines a protected finite
irrelevant deformation. In the rest of this section, I assume that the protected operators
are dropped from (2.1) and that the λi s refer only to the unprotected irrelevant operators,
unless otherwise specified.
2.3. Finite solutions
Suppose that there exists an integer or a half-integer > 0 such that the matrices γn
are invertible for every n > 1 and n = rank(γ ) < N . Then the finiteness equations (2.8)
admit a non-trivial solution with N − n free parameters.
If λI denote the solutions of the equations

N
γI J (α)λJ = 0, (2.9)
J =1
let
λj J = 0 for every j = n, n = integer, (2.10)

Nn
I J
−1
λnI = − γn δnJ for every n > 1. (2.11)
J =1
The solutions of (2.9) contain N − n free parameters, by assumption. Now, formula (2.5),
with the condition (2.6), and (2.10) imply δj J = 0 for every j = n. This ensures that the
finiteness equations βj J = 0 are trivially satisfied for j = n. Moreover, formula (2.5)
implies also δI = 0, and therefore the λI s solve βI = 0, i.e., the finiteness equations
(2.8) for i = . Finally, the existence of the solutions (2.11) is ensured by the invertibility
of the matrices γn for n > 1. The δnI s for n > 1 are determined recursively as functions
of λI and α, using formula (2.5).
Summarizing, the theory described by the Lagrangian

N ∞

Nn
−1 I J
L[ϕ] = L0 [ϕ, α] + κ λI OI (ϕ) − κ n γn δnJ OnI (ϕ) (2.12)
I n=1 I,J =1
is finite. Its independent couplings are α and the N − n free parameters contained in λI .
The beta functions are identically zero, but in general renormalization demands non-trivial
field redefinitions. The power-like divergences do not contribute to the RG equations and
so can be subtracted as they come, without adding new independent couplings.
The theory L[ϕ] is a finite irrelevant deformation of the theory L0 [ϕ, α]. The level
is called lowest level of the deformation, while the last sum in (2.12) is called queue of
the deformation. If N = n the solution is trivial (all of the λs vanish) and coincides with
L0 [ϕ, α], which is finite by assumption.
The inclusion of protected operators in the solution (2.9)–(2.11) is straightforward,
since it is sufficient to set their couplings to zero. As remarked above, if some protected
operators are finite, it is possible to consider more general solutions that contain one extra
independent parameter for each finite protected operator [9].
2.4. Sufficient conditions for the existence of a perturbative expansion
If C is a family of conformal field theories that become free when some marginal
parameter g tends to zero, then the theory coupled with gravity might not admit a smooth
g → 0 limit, due to the inverse matrices that appear in formula (2.11). However, if the
anomalous dimensions of the irrelevant couplings satisfy a certain boundedness condition,
it is possible to keep g small, but different from zero, and have a meaningful perturbative
expansion in powers of g and κeff E, where E is the energy scale and κeff is an effective
inverse Planck mass that depends on g. Basically, the absolute values of the anomalous
dimensions of the unprotected irrelevant operators should admit a strictly positive bound
from below.
The first non-vanishing irrelevant couplings are the λI s, namely the solutions of (2.9),
some of which can have arbitrary values. Let λ = maxI |λI |. Assume that there exists a
η > 0, depending on and g, and non-vanishing g-independent numbers cn , such that
−1 I J cn
γ < (2.13)
n
η
when g ∼ 0, for every n > 1 and every I , J . The quantity η generically tends to zero
IJ
when g tends to zero. Observe that perturbation theory ensures that γn (α) and δnI have
a smooth limit when g → 0 at λ fixed.
Under the assumption (2.13), when g → 0 the solutions (2.11) behave not worse than
λn
|λnI | ∼ c̃n , (2.14)
η(n−1)
for other g-independent numbers c̃n , constructed with the cn ’s. The behavior (2.14) can
be proved inductively in n. Indeed, if (2.14) is true for n < m, then (2.5), (2.7) and (2.11)
immediately imply that it is also true for n = m.
Let us compare the behavior of an irrelevant term of dimensionality d + n with the
behavior of the marginal terms of C, as functions of the energy scale E of a process. The
ratio between these two types of contributions behaves not worse than
1/ n
λ κE
an η ,
η
an being calculable numbers, that depend on the cn of (2.13). The perturbative expansion
in powers of κ (equivalently, in powers of the energy) is meaningful for energies E much
smaller than the effective Planck mass
1 η
≡ . (2.15)
κeff κλ1/

This up to the behavior of the numerical factors an , which cannot be predicted unless the
theory is solved. The constant λ can be set to one without loss of generality, since it always
appears in the combination κ λ .
In conclusion, the condition to have a consistent non-trivial finite irrelevant deformation
is that there exists a lowest level such that
0 < < ∞, n < N , η > 0. (2.16)
I have emphasized that η can depend on . Observe that the conditions (2.16) concern only
the renormalizable subsector L0 [ϕ, α] of the theory, and can be studied before turning the
irrelevant deformation on.
3. Application to quantum gravity in three dimensions
The discussion of the previous section was completely general. Applied, for example,
to quantum gravity in four dimensions, it shows that it is not possible to make it finite in a
simple way, because (2.16) does not hold (η = 0 for every lowest level ). I come back to
this at the end of this section. Other types of four-dimensional applications can be thought,
as shown, for example, in [9].
A situation where (2.16) does hold is the case of gravity coupled with matter in three
spacetime dimensions, with = 1 and η ∼ α, α denoting some marginal coupling of C. In
this section I discuss three-dimensional quantum gravity in general terms. In the rest of the
paper I consider an explicit model in detail.
I assume that the κ → 0 limit L0 [ϕ, α] is the sum of the free spin-2 kinetic term plus
the Lagrangian LC [ϕ, α] of the matter sector, which I take to be a conformal field theory C.
The theory C is subject to the restrictions (2.16), which I discuss below. The beta functions
of the marginal couplings of C are independent of the irrelevant couplings and determined
solely within the conformal field theory C, i.e., at κ = 0. Since the matter subsector of the
theory is conformal, the beta functions of the marginal couplings of C vanish also when
κ = 0.
The Einstein term
1 √
gR (3.1)
2κλ1
contains the spin-2 kinetic term and an irrelevant deformation of level i = 1. The coupled
theory can contain other irrelevant terms with i = 1, such as four-fermion terms.
I prefer the notation (3.1), keeping λ1 and κ in the denominator (λ1 is redundant and
can be set to one at the end) and expanding the dreibein around flat space as eµa = δµa + φµa .
The formulas of the previous section apply unchanged, because it is easy to prove that in
(2.3) only positive powers
of λ̄1 can appear. Instead, expanding the dreibein around flat
space as eµa = δµa + κ λ̄1 φµa , to eliminate κ and λ̄1 from the denominator of (3.1), the
three-graviton vertex is regarded as an irrelevant deformation of level i = 1/2.
I assume that the Lorentz–Chern–Simons term (1.2) is absent at the classical level and
that the subtraction scheme is such that this term remains absent also at the quantum
level [5].
The beta function of λ̄1 vanishes identically, because the Einstein term is non-
renormalized. The reason is that no denominator 1/κ can be generated by the Feynman
diagrams. This fact implies that the lowest level is at least equal to 1.
If the conformal field theory C is interacting and “generic”, then it is reasonable
to expect that the anomalous dimensions of the irrelevant deformations of C are non-
vanishing. This ensures that = 1 satisfies the restriction (2.16). I now discuss this point
in detail.
The set of irrelevant terms of the coupled theory can be split into three subsets:
(i) the irrelevant terms that belong to the matter sector, i.e., those that have a non-
vanishing flat-space limit;
(ii) the irrelevant terms that belong to the gravity sector, i.e., those that are constructed
with powers of the curvature tensors and their covariant derivatives, but contain no
matter fields;
(iii) the mixed terms.
It is convenient to analyze the finiteness equations separately within these subsets.
3.1. Sufficient condition for a solution
The simplest sufficient condition to have a non-trivial solution is that the following two
requirements be satisfied:
(a) All of the unprotected irrelevant operators of the conformal field theory C have non-
vanishing anomalous dimensions (this is a restriction on C);
(b) The subsets (ii) and (iii) are empty, apart from the Einstein term.
Now I study when these requirements can be met.

A necessary condition for (a) is that C be interacting, otherwise the irrelevant terms
of class (i) have vanishing anomalous dimensions. In most cases, this restriction is also
sufficient to ensure that all of the terms of class (i) have non-vanishing anomalous
dimensions.
Exact results proving the existence (or non-existence) of theories satisfying (a) are
not available, to my knowledge. Nevertheless, common experience with renormalization
theory suggests that almost all interacting conformal field theories are expected to
satisfy (a). I make a brief digression to illustrate some aspects of this issue.
Operators that have vanishing anomalous dimensions are called finite. To disprove (a)
it is necessary to exhibit examples of finite unprotected irrelevant operators in flat space.
Generically speaking, in renormalization, whenever a quantity can diverge (because it is
not protected by symmetries, power-counting, etc.), it does diverge. Therefore, a counter-
example can only be the product of a miraculous cancellation. The finite operators known
to me represent no obstacle to the solubleness of the finiteness equations, either because
they are not irrelevant, or because they are protected.
The simplest finite operators are associated with conserved (and anomalous) currents,
and the marginal deformations of C. However, these operators have level zero or negative,
so they are not irrelevant.
Examples of irrelevant finite operators of arbitrary positive levels are provided by the
chiral operators of N = 2 and N = 4 superconformal field theories in four dimensions [8].
However, these operators are protected. For concreteness, consider N = 4 supersymmetric
Yang–Mills theory. In the formalism of N = 1 superfields, this theory contains a vector
multiplet and three chiral multiplets Φ i . The fields Φ i have zero anomalous dimensions
and the chiral operators, for example,

Yi1 ...in Φ i1 · · · Φ in d2 θ,
are finite. (Here Yi1 ...in is a constant tensor.) Because of the non-renormalization
theorem [8], the chiral operators have also δ = 0. Therefore, their beta functions vanish
identically.
I stress that the anomalous dimensions depend on the marginal couplings of C and so, in
the worst case, if the anomalous dimension of an unprotected irrelevant operator vanishes,
it is expected to vanish only for some special values of the marginal couplings α. In this
sense, the requirement (a) can be viewed as a restriction on the conformal field theory C.
In summary, the present knowledge supports the statement that almost all interacting
conformal field theories satisfy (a).
Now it is necessary to discuss the existence of solutions of the finiteness equations in
the subsectors (ii) and (iii) listed above. Since the Lagrangian terms of class (ii) do not
contain matter fields, they are just the identity operator, from the point of view of C, and
can be studied embedding C in external gravity. This means that the anomalous dimensions
of the terms of class (ii) are zero and their finiteness equations cannot be solved in the way
described in the previous section. Therefore, the quantization procedure outlined above
does not work, unless class (ii) contains only the Einstein term.
Classes (ii) and (iii) are empty, apart from the Einstein term, precisely in three-
dimensional quantum gravity. In three dimensions the field equations express the Riemann
tensor in terms of the matter fields and so the unique independent Lagrangian term of
classes (ii) and (iii) is the Einstein term.
The Einstein term has i = 1. Other irrelevant terms of level 1, belonging to class (i), can
be present (four-fermion vertices, Pauli terms, and so on) and their matrix of anomalous
dimensions is in general non-vanishing. It is convenient to decompose the matrix γ1I J ,
I, J = 1, . . . , N1 , into
¯¯ ¯
(γ̃1 )I J (γ1 )I N1
(γ1 ) =
IJ
. (3.2)
0 0
Here the N1 th value of the indices I, J is conventionally associated with the Einstein term
¯¯
(λ1N1 ≡ λ̄1 ). The block (γ̃1 )I J , I¯, J¯ = 1, . . . , N1 − 1, denotes the matrix of anomalous
dimensions of the irrelevant terms of level 1 belonging to class (i). The N1 th row of the
matrix γ1 is zero, because the beta function of the Newton constant is identically zero.
Because of the discussion made above, the matrix γ̃1 can be assumed to be invertible.
This ensures that the rank of the matrix γ1 is equal to N1 − 1 and therefore = 1. So, the
finiteness equations admit a non-trivial solution with lowest level equal to 1. The coupled
theory contains only one arbitrary parameter, the Newton constant, besides the marginal
couplings of C.
Using the decomposition (3.2) the finiteness equations

N1
β1I = γ1I J (α)λ1J = 0,
J =1
split into
1 −1
N
¯¯
β1N1 = 0 and β1I¯ = γ̃1I J (α)λ1J¯ + δ̃1I¯ = 0,
J¯=1
¯
where δ̃1I¯ = (γ1 )I N1 λ̄1 . The beta functions of the level-1 operators belonging to the matter
sector have the same form as (2.5) and so their solutions have the form (2.11).
Finally, the finite theory of quantum gravity coupled with the conformal field theory C
has Lagrangian
1 √
L[ϕ] = g R + LC [ϕ, α]
2κ
N1 −1 ∞

Ni
−1 I¯J¯ −1 I J
−κ γ̃1 δ̃1J¯ O1I¯ (ϕ) − κi γi δiJ OiI (ϕ),
I¯,J¯=1 i=2 I,J =1
where λ̄1 has been set to 1. Renormalization requires non-trivial field redefinitions, but the
coupling constants are non-renormalized.
3.2. Existence of a perturbative expansion
In general, the anomalous dimensions of the unprotected irrelevant operators are non-
zero already at two-loop order (the one-loop diagrams converge in odd dimensions), so
the quantity η of (2.13) is typically of order α 2 ∼ g 4 , where α ∼ g 2 is a generic marginal
coupling of C (the power is fixed assuming that g multiplies a three-leg vertex, such as
/ ψ) that tends to zero in the free-field limit. The perturbative expansion is meaningful
ψ̄A
for energies E much smaller than the effective Planck mass
1 η
MP eff = = ∼ αMP . (3.3)
κeff κ λ̄1
In practice, the Planck scale is screened by the interactions of C and effectively reduced by
a factor 1/η. To cross the energy MP eff it is necessary to resum the perturbative expansion.
In summary, in three dimensions it is possible to define a procedure of quantization in
the presence of gravity, when the matter sector has η > 0. Since this restriction concerns
only the matter sector of the theory, it is possible to say which kind of matter can be
coupled to gravity before effectively coupling it to gravity. In the next sections I study
gravity coupled with Chern–Simons U (1) gauge theory and massless fermions.
The reason why the procedure described in this paper cannot be applied straightfor-
wardly to quantize four-dimensional gravity is that in four-dimensional gravity the class (ii)
contains infinitely many non-trivial terms, of arbitrarily high levels, and no symmetry pro-
tects them, i.e., they have γ = 0, δ = 0 [3]. Therefore, η = 0 for every candidate lowest
level < ∞.
4. Gravity coupled with Chern–Simons U (1) gauge theory with massless fermions
In the rest of the paper I illustrate the quantization procedure defined in the previous
sections in a concrete model, namely, three-dimensional gravity coupled with Chern–
Simons U (1) gauge theory with massless fermions. In this section I recall the basic
properties of this theory and the results of [5]. I work in the Euclidean framework.
In flat space, Chern–Simons U (1) gauge theory with massless fermions is described by
the Lagrangian
1 µνρ
Lcl = ψ̄Dψ
/ + ε Fµν Aρ , (4.1)
2g 2
where Dµ = ∂µ + iAµ is the covariant derivative in flat space. This theory is conformal,
because the beta function of g vanishes [10]. The anomalous dimension of ψ is different
from zero. I consider nf copies of complex two-component spinors. The renormalized
Lagrangian reads
1 µνρ
LR = Zψ ψ̄Dψ
/ + ε Fµν Aρ .
2g 2
The lowest-order values of the fermion renormalization constant and anomalous dimension
are given by the graph (c) of Fig. 1, up to subleading corrections in 1/nf :
g 4 nf 1 d ln Zψ g 4 nf
Zψ = 1 − , γψ = = .
384π 2ε 2 d ln µ 384π 2
This theory is taken as the conformal field theory C for the coupling with gravity.
4.1. Coupling with gravity
The Lagrangian is
1 1
L= eR + eψ̄Dψ
/ + 2 εµνρ Fµν Aρ + O(κ), (4.2)
2κ 2g
(a) (b) (c)
Fig. 1. One-loop gauge field and graviton gauge field self-energies.

√
where e = g. This theory is not finite [5], because a counterterm (1.3) is induced by
renormalization to the second order in the loop expansion and first order in the κ expansion.
So, it is necessary to include in (4.2) the irrelevant terms generated by renormalization. I
focus here on the irrelevant terms of dimensionality four, or level 1, which are
κeψ̄D
/ 2 ψ, κeFµν F µν , κεµνρ eρa Fµν ψ̄γ a ψ, κe(ψ̄ψ)2 , κe(ψ̄γ a ψ)2 . (4.3)
Only two of these are independent, e.g., the four-fermion vertices [5]. Up to O(κ 2 ), the
complete Lagrangian
1 1 λ κ λ κ
Lcl = / + 2 εµνρ Fµν Aρ + 1 e(ψ̄ψ)2 + 2 e(ψ̄γ a ψ)2 + O(κ 2 )
eR + eψ̄Dψ
2κ 2g 4 4
(4.4)
has the field equations
λ1 κ λ2 κ
Dψ
/ + (ψ̄ψ)ψ + (ψ̄γ a ψ)γ a ψ + O(κ 2 ) = 0, (4.5)
2 2
ig 2
Fµν + eεµνρ eρa ψ̄γ a ψ + O(κ 2 ) = 0, (4.6)
2
1 1 1 ←
→ 1 ←
→ 1 ←→
Rµν − gµν R + eµa ψ̄γ a D ν ψ + eνa ψ̄γ a D µ ψ − gµν ψ̄ D ψ
2κ 2 8 8 4
λ1 κ λ2 κ
− gµν (ψ̄ψ)2 − gµν (ψ̄γ a ψ)2 + O(κ 2 ) = 0.
8 8
Using the fermion field equation (4.5), the first term of the list (4.3) can be converted to
O(κ 2 ). Using the gauge field equation (4.6) the second and third terms of (4.3) can be
converted into the forth term of the same list, up to O(κ 2 ). So, the Newton constant and
the couplings λ1,2 make a complete set of essential couplings of level 1.
The gravitational field is defined expanding the dreibein eµa around flat space:
eµa = δµa + φµa , ωµa = εabc ∂ b φµc + O(φ 2 ).

I choose the symmetric gauge φµa = φaµ .
As remarked in Ref. [5], since the divergent parts of the diagrams are polynomial in the
number nf of fermions, it is convenient to concentrate the attention on the contributions
proportional to nf . These are given by the diagrams that contain one-fermion loop. At the
second loop order the diagrams containing two-fermion loops factorize into the product of
two one-loop subdiagrams, and are therefore convergent.
The gauge-fixing Lagrangian is
1 1
Lgf = (∂µ φµν )2 + (∂µ Aµ )2 + Lghost.
2ακ 2λg 2
The gauge parameters λ and α are kept throughout the calculations, because gauge-
independence provides a powerful check of the calculations. The U (1) field is conveniently
gauge-fixed in flat space.
The ghost part of the gauge-fixing Lagrangian can be ignored in the calculations of this
paper. Indeed, diagrams with external ghost legs do not contribute to the renormalization
of the four-fermion vertices, but belong to the gauge-trivial sector of the theory. Instead,
diagrams with internal ghosts must have, to the leading order in 1/nf , one ghost loop and
one-fermion loop. These diagrams necessarily factorize into two one-loop subdiagrams
and therefore converge.
A convenient regularization technique consists of modifying the propagators with an
exponential cut-off:

1 1 p2
→ 2 exp − 2 .
p2 p Λ
This can be done in a gauge invariant way to all orders [5]. Instead, the dimensional
regularization technique presents some difficulties, because of the ε tensor appearing in
the U (1) Chern–Simons term and because the trace of an odd number of Dirac matrices
does not always vanish. Nevertheless, for the purposes of this paper, it is consistent to
use the dimensional regularization framework, since the divergent parts of the two-loop
diagrams are made of simple poles 1/ε, if ε = 3 − D, and the residues of simple poles
can be evaluated directly in three dimensions. The conversion of the results to the cut-off
approach is performed by means of the replacement 1/ε → ln Λ2 /µ2 and the power-like
divergences are subtracted as they come.
The bare Lagrangian reads
LB 1 1
= RB + ψ̄BD / B ψB + 2 εµνρ FBµν ABρ
eB 2κ 2g eB
1 1 2
+ λ1B κ(ψ̄B ψB )2 + λ2B κ ψ̄B γ a ψB + O(κ 2 ). (4.7)
4 4
I have not written the regularizing terms explicitly. The relations between bare and
renormalized quantities read
λ1B = λ1 Z1 , λ2B = λ2 Z2 ,
1/2
AµB = Aµ + O(κ), ψB = Zψ ψ + O(κ), a
eµB = eµa + O(κ). (4.8)
4.2. Calculations
The calculation of the two-loop counterterms can be divided into two parts: the
contributions of type δ in (1.4), which have been computed in Ref. [5], and the self-
renormalization of the four-fermion terms, that is to say the contributions of type λγλ in
(1.4).
The counterterms have to be simplified using the field equations, to separate the
renormalization of the essential couplings from the field redefinitions. In the case at hand,
this means that the following replacements
ig 2
Rµνρσ → 0, Fµν → − eεµνρ eρa ψ̄γ a ψ, / → 0,
Dψ (4.9)
2
are allowed.
The one-loop gauge field self-energy, given by Fig. 1(a), is
nf 1
− 2 (1+ε)/2
δµν k 2 − kµ kν .
16 (k )
The graviton gauge field self-energy of Fig. 1(b) vanishes by spin conservation. This fact
reduces the number of two-loop diagrams.
4.3. Counterterms induced by gravity
The results of Ref. [5] are that for λ1,2 = 0, at the second order in the loop expansion,
first order in κ and leading order in 1/nf , renormalization requires the four-fermion
counterterm
grav 5κg 4 nf e
Lcounter = − (ψ̄γ a ψ)2 (4.10)
384π 2ε 4
and the field redefinition
5nf αg 2 κ inf g 4 κ(3 + 5α) a
Aµ → Aµ − 2
eεµνρ F νρ − eµ ψ̄γ a ψ.
768π ε 768π 2ε
Moreover, no Lorentz–Chern–Simons term (1.2) is generated.
5. Self-renormalization of the four-fermion vertices
The counterterms proportional to λ1,2 can be computed in flat space and are associated
with the anomalous dimensions of the four-fermion vertices. The set of diagrams can be
split into two subsets: the diagrams that have two external fermions and one or no external
gauge field (see Fig. 2); the diagrams that have four external fermions (see Fig. 3). The
diagrams are constructed with one four-fermion vertex, one-fermion loop and one or two
internal gauge field legs, respectively. The two-loop diagrams with one four-fermion vertex
and two external gauge field legs factorize into products of one-loop subdiagrams and
therefore converge.
5.1. Fermion self-energy and fermion gauge field vertex
The diagrams are shown in Fig. 2. The counterterms sum to

ig 2 λ2 nf κ ig 2 nf κ
Lcounter-1 = − e ψ̄D/ Dψ
/ + (λ1 − λ2 )εµνρ eρa Fµν ψ̄γ a ψ. (5.1)
192π 2 ε 192π 2 ε
Fig. 2. Fermion self-energy and fermion gauge field vertex.

Fig. 3. Renormalization of the four-fermion vertices.
5.2. Four-fermion counterterms
The graphs contributing to these counterterms are shown in Fig. 3 and give

g 4 nf κ 1 1
Lcounter-2 = e (2λ1 − λ2 ) (ψ̄γ ψ) + 3(5λ1 + 6λ2 ) (ψ̄ψ) .
a 2 2
(5.2)
192π 2 ε 4 4
6. Solution of the finiteness equations
It is now time to collect the results of Ref. [5] and this paper, solve the finiteness
equations, and determine the values of the irrelevant couplings λ1,2 that multiply the four-
fermion vertices.
6.1. Totals
The total four-fermion counterterms can be obtained summing (4.10), (5.1) and (5.2)
and using the replacements (4.9). The result is

g 4 nf κ 1 1
Lcounter = e (12λ1 − 10λ2 − 5) (ψ̄γ ψ) + 6(5λ1 + 6λ2 ) (ψ̄ψ) .
a 2 2
384π 2ε 4 4
The renormalization constants of the couplings λ1 and λ2 are obtained subtracting the
contribution associated with the fermion wave-function renormalization constant. The net
counterterm is then

g 4 nf κ 1 1
Lcounter-net = e (12λ1 − 8λ2 − 5) (ψ̄γ ψ) + 4(8λ1 + 9λ2 ) (ψ̄ψ) .
a 2 2
384π 2ε 4 4
(6.1)
Using (4.7) and (4.8) the bare couplings are
g 4 nf (8λ1 + 9λ2 ) g 4 nf (12λ1 − 8λ2 − 5)
λ1B = λ1 + , λ2B = λ2 + .
96π 2ε 384π 2ε
6.2. Solution of the finiteness equations
Finiteness demands that the counterterm (6.1) vanishes, whence

45 10
λ1 = , λ2 = − . (6.2)
172 43
In conclusion, the finiteness conditions admit one solution and uniquely determine the
values of the four-fermion couplings.
To couplings λ1,2 turn out to be g-independent. This is due to the fact that γλ and δλ
are of the same order in g. The irrelevant terms belonging to higher levels, however, are
expected to have δλ ∼ 1 and so the quantity η defined in (2.13) is expected to behave like
1/g 4 . The effective Planck mass is therefore ∼ g 4 /κ.
7. Applications to four dimensions
The quantization procedure defined in Sections 2 and 3 is meaningful for those theories
that have η > 0, where η is defined by Eq. (2.13). I have shown that three-dimensional
quantum gravity coupled with a generic interacting conformal field theory has the desired
properties. This is not the case of four-dimensional quantum gravity, coupled with matter
or not, because every candidate lowest level < ∞ has η = 0. Indeed, the beta functions
of the irrelevant terms made with the Riemann tensor and its derivatives, such as
√ ρσ µν αβ
g Rµν Rαβ Rρσ (7.1)
have the form (1.4) with γ = 0 and δ = 0 [3]. In three dimensions, a term like (7.1) can
be reabsorbed by means of field redefinitions, because there is no graviton, but in four
dimensions this is impossible.
I have made a certain number of attempts, not reported here, to try to circumvent
the difficulty of four-dimensional gravity. These will be probably collected in a separate
publication. It is certainly possible to modify the theory to have non-vanishing γ s for the
operators (7.1), for example, adding a cosmological constant. Then, however, it is not easy
to solve the finiteness equations. Moreover, other problems appear in the presence of a
cosmological constant in four dimensions. The difficulties might be just technical or hide
more conceptual aspects.
It is worth mentioning that even if the ideas of this paper do not extend immediately to
quantum gravity in four dimensions, a more general framework where they do might exist,
with potentially appealing implications. The quantization of gravity might be possible only
in the presence of interacting matter, for example, thanks to the existence of QCD. The
energy at which the effects of quantum gravity become relevant could be not the Planck
mass, but an effective Planck mass that takes care of the presence of matter. If the matter
is weakly interacting, the effective Planck mass could be considerably small. The limit in
which the interaction of the matter sector is switched off could be singular.
Other types of applications to four dimensions are possible, as shown, for example,
in [9]. Generalization to running theories are possible also, but more tricky.
8. Conclusions
In this paper I have shown that it is possible to give a quantization prescription that
ensures, under certain conditions, finiteness of quantum gravity coupled with matter in
three spacetime dimensions. The procedure is algorithmic and so it can be implemented
perturbatively. Gravity is coupled with an interacting conformal field theory C. The values
of the irrelevant couplings, apart from the Newton constant, are determined imposing
that their beta functions vanish. The finiteness equations have solutions thanks of the
properties of three-dimensional spacetime, in particular, the absence of a propagating
graviton, and because the unprotected irrelevant operators of C have, generically, non-
vanishing anomalous dimensions. A quantity η, defined by formula (2.13), characterizes
the strength of the interactions of the matter subsector. The expansion in powers of the
energy is valid for energies much smaller than the effective Planck mass ηMP .
In a concrete example, I have studied the Chern–Simons U (1) gauge theory with
massless fermions coupled with gravity and applied the iterative procedure of Sections 2
and 3 to compute the coefficients of the four-fermion vertices. The “classical” Lagrangian
of the finite theory defined by this quantization prescription is
1 1
L= eR + eψ̄Dψ/ + 2 εµνρ Fµν Aρ
2κ 2g
45 κ 10 κ
+ e(ψ̄ψ)2 − e(ψ̄γ a ψ)2 + O(κ 2 ) (8.1)
172 4 43 4
and has only two arbitrary parameters: the Chern–Simons coupling g and the Newton
constant κ. The action (8.1) is renormalizable as it stands, i.e., without adding new
parameters, but just redefining the fields. In this sense, it is finite.
The results of this paper might revive some hopes to find a finite theory of gravitational
interactions. Several aspects of the ideas applied here admit generalizations to four
dimensions [9]. However, the peculiarity of three dimensions is crucial to have a non-
vanishing effective Planck mass in the presence of gravity. Quantum gravity in four
dimensions does not fulfill this requirement in a straightforward way. For this reason,
the generalization of these ideas to quantum gravity in four dimensions demands further
insight.
Appendix A
Torsion, curvatures, covariant derivatives and connections are:

1
Dea = dea − ωab eb = 0, R a = dωa + εabc ωb ωc ,
2
Dµ Vν = ∂µ Vν − Γµν ρ
Vρ , ρ
Γµν = eρa ∂µ eνa + ωµab eνa eρb ,
1
ωµa = εabc ∂µ eνb − ∂ν eµb eνc − eµa εbcd ∂ρ eνb − ∂ν eρb eνc eρd ,
4
i a a
Dµ ψ = ∂µ ψ − ωµ γ ψ + iAµ ψ.
2
The Ricci tensor and scalar curvature are defined as Rµν = Rµρ
ab e ρb e a , R = R g µν , where
ν µν
µ µ λ Γ µ + Γ λ Γ µ and, of
R = ε R = Rµν dx dx /2, R νρσ = ∂σ Γνρ − ∂ρ Γνσ − Γνσ
ab abc c ab µ ν µ
λρ νρ λσ
course, gµν = eµa eνa .
References
[1] G. Parisi, The theory of nonrenormalizable interactions. I. The large N expansion, Nucl. Phys. B 100 (1975)
368.
[2] G. ’t Hooft, M. Veltman, One-loop divergences in the theory of gravitation, Ann. Inst. Poincarè 20 (1974)
69.
[3] M.H. Goroff, A. Sagnotti, The ultraviolet behavior of Einstein gravity, Nucl. Phys. B 266 (1986) 709.
[4] E. Witten, (2 + 1)-dimensional gravity as an exactly soluble system, Nucl. Phys. B 311 (1988) 46.
[5] D. Anselmi, Renormalization of quantum gravity coupled with matter in three dimensions, Nucl. Phys. B
687 (2004) 143, this issue, hep-th/0309249.
[6] S. Deser, R. Jackiw, S. Templeton, Topologically massive gauge theories, Ann. Phys. 140 (1982) 372.
[7] S. Weinberg, Ultraviolet divergences in quantum theories of gravitation, in: S. Hawking, W. Israel (Eds.),
An Einstein Centenary Survey, Cambridge Univ. Press, Cambridge, 1979.
[8] A good reference for supersymmetry in the language of superfields is: S.J. Gates Jr., W. Siegel, M. Rocek,
M.T. Grisaru, Superspace, or One-Thousand and One Lessons in Super Symmetry, Addison–Wesley,
Reading, MA, 1983.
[9] D. Anselmi, Consistent irrelevant deformations of interacting conformal field theories, JHEP 0310 (2003)
045, hep-th/0309251.
[10] A. Blasi, N. Maggiore, S.P. Sorella, Nonrenormalization properties of the Chern–Simons action coupled to
matter, Phys. Lett. B 285 (1992) 54, hep-th/9204045.
Renormalization of quantum gravity coupled with

matter in three dimensions
Damiano Anselmi
Dipartimento di Fisica “E. Fermi”, Università di Pisa, and INFN, Pisa, Italy
Received 10 October 2003; received in revised form 16 February 2004; accepted 18 March 2004
Abstract
In three spacetime dimensions, where no graviton propagates, pure gravity is known to be finite. It
is natural to inquire whether finiteness survives the coupling with matter. Standard arguments ensure
that there exists a subtraction scheme where no Lorentz–Chern–Simons term is generated by radiative
corrections, but are not sufficiently powerful to ensure finiteness. Therefore, it is necessary to perform
an explicit (two-loop) computation in a specific model. I consider quantum gravity coupled with
Chern–Simons U (1) gauge theory and massless fermions and show that renormalization originates
four-fermion divergent vertices at the second loop order. I conclude that quantum gravity coupled
with matter, as it stands, is not finite in three spacetime dimensions.
1. Introduction
Gravity is not power-counting renormalizable in dimensions greater than two. It is

known [1] that pure gravity in four dimensions is finite to the first loop order and that
one-loop finiteness is spoiled by the coupling with matter. Moreover, four-dimensional
gravity is not finite to the second loop order [2], even in the absence of matter.
In three dimensions there is no propagating graviton and pure gravity

1 √
gR(x) d3x (1.1)
2κ
is known to be finite to all orders [3]. A quick proof is based on the observation that the
counterterms vanish using the field equations of (1.1) and therefore can be reabsorbed by
means of field redefinitions. Indeed, in three dimensions the Weyl tensor is identically
E-mail address: anselmi@df.unipi.it (D. Anselmi).
doi:10.1016/j.nuclphysb.2004.03.023
zero and so the Riemann tensor is a linear combination of the Ricci tensor and the scalar
curvature:
R R
Rµνρσ = gµρ Rνσ − gµσ Rνρ − gνρ Rµσ + gνσ Rµρ − gµρ gνσ + gµσ gνρ . (1.2)
2 2
Every counterterm is proportional to Rµν or R, apart from the Lorentz–Chern–Simons
term

1
εµνρ ωµa ∂ν ωρa + ωµa ωνb ωρc εabc , (1.3)
3
which does not appear by parity invariance. By dimensional counting, the counterterms
are actually quadratic, at least, in Rµν –R and therefore can be reabsorbed by means of
covariant field redefinitions, with no renormalization of the Newton constant κ.
It is natural to inquire whether finiteness survives the coupling with matter in three
dimensions. The renormalization of the theory has chances to be non-trivial, even if no
graviton propagates. If the theory is finite, renormalization requires only field redefinitions,
but no running of the coupling constants. If the theory is not finite, then renormalization
generates infinitely many new coupling constants, as in four dimensions. In this paper
I study these issues.
First I analyze non-renormalization properties and standard arguments about finiteness.
I prove that there exists a subtraction scheme where no Lorentz–Chern–Simons term is
generated by radiative corrections. This ensures that gravity is not driven to the theory
known as “topologically massive gravity” [4]. However, the standard non-renormalization
arguments are not sufficiently powerful to ensure finiteness, because higher-dimensioned
operators can be generated by renormalization. To decide whether finiteness survives the
coupling with matter or not, it is necessary to perform an explicit computation in a specific
model. I consider quantum gravity coupled with Chern–Simons U (1) gauge theory and
massless fermions. This model is a good laboratory to explore ideas about finiteness and
renormalizability beyond power counting. I show that renormalization originates a four-
fermion divergent vertex
5κg 4 nf e
Ldiv = (ψ̄γ a ψ)2 (1.4)
384π 2ε 4
at the second order in the loop expansion and first order in the Newton constant κ. The
result (1.4) is written up to subleading corrections in 1/nf . I conclude that quantum gravity
coupled with matter, as it stands, is not finite in three spacetime dimensions.
The computation is two-loop, because in three dimensions every theory is finite to
the first loop order. By symmetric integration, an odd-dimensional theory has no one-
loop logarithmic divergence. Moreover, the power-like divergences are scheme artifacts
(they are automatically absent using the dimensional regularization technique) and have
no effects on the renormalization group. So, the problem of finiteness starts at two loops in
three dimensions.
At the classical level, the properties of gravity coupled with matter in three dimensions
have been widely studied, starting from Ref. [5]. At the quantum level, there have
been studies on quantum gravity of point particles [6], quantum cosmology and black-
hole quantum mechanics [7], topologically massive gravity [4], gravitating topological
matter [8], de Sitter quantum gravity [9], loop quantum gravity [10], dynamically
triangulated quantum gravity [11] and many other subjects.
In flat space, the renormalization of (2 + 1)-dimensional quantum field theory has been
studied at the perturbative level [12–14] and in the large N expansion [15–19]. Besides
the finiteness of pure gravity in three dimensions [3], there have been studies on the
renormalizability of quantum gravity near two dimensions [20]. The renormalization of
2 + 1 quantum gravity coupled with matter has attracted less attention, so far. The interest
of this research is that it can shed some light on the properties of renormalization beyond
power-counting.
The paper is organized as follows. In Section 2 I recall the properties of Chern–Simons
U (1) gauge theory with matter in flat space. In Section 3 I couple it with gravity. In
Section 4 I prove that no Lorentz–Chern–Simons term is induced by renormalization.
In Section 5 I introduce the two-loop computations of this paper, the organization of
counterterms and the calculational technique. In Section 6 I collect the results about four-
fermion vertices induced by gravity. Section 7 contains the conclusions. In Appendix A
I collect useful formulas and some remarks about the difficulties of the dimensional
regularization of the Chern–Simons term.
2. Chern–Simons U (1) gauge theory with massless fermions
In this section I recall some properties of Chern–Simons U (1) gauge theory with
massless fermions in flat space and fix the notation. I work in the Euclidean framework.
The Lagrangian reads
1 µνρ
Lcl = ψ̄/
Dψ + ε Fµν Aρ , (2.1)
2g 2
where Dµ = ∂µ + iAµ is the covariant derivative in flat space. This theory is conformal,
since the beta function of g vanishes [14], but the anomalous dimension of ψ is
different from zero. Precisely, (2.1) is a one-parameter family of conformal field theories,
parametrized by g. I use two-component complex spinors and consider nf copies of them.
The Dirac matrices are Hermitean and such that γµT = −γ2 γµ γ2 , where T means transpose.
The “time” coordinate is x3 .
2.1. Discrete symmetries
The parity, charge-conjugation and “time”-reversal transformations are

P1 : xµ → (−x1, x2 , x3 ), ψ → γ1 ψ, ψ̄ → −ψ̄γ1 ,
Aµ → (−A1 , A2 , A3 ), g 2 → −g 2 ,
C: xµ → xµ , ψ → γ2 (ψ̄)T , ψ̄ → −ψ T γ2 , Aµ → −Aµ ,
T: xµ → (x1 , x2 , −x3 ), ψ → γ3 ψ, ψ̄ → −ψ̄γ3 ,
Aµ → (A1 , A2 , −A3 ), g 2 → −g 2 ,
where ψ̄ = ψ † γ3 . Only C and CPT are true symmetries, since P and T change the sign
of g 2 .
2.2. Regularization
The ordinary dimensional regularization technique is not convenient for the theory (2.1),
because of the difficulties related to the ε tensor and the trace of an odd product of gamma
matrices. Some observations on this issue are collected in Appendix A. Nevertheless, for
the purpose of computing divergent parts of two-loop graphs, where only simple poles
appear, it is consistent to ignore this problem and work with the dimensional technique.
This reduces the effort and simplifies the algebra. Instead, it is necessary to use an
alternative regularization technique to prove properties valid to all orders in the perturbative
expansion. A standard choice is to modify the gauge field propagator with higher-derivative
terms in a gauge-invariant way and match the fermion loops with loops of Pauli–Villars
fields. This can be achieved with a regularized Lagrangian

1 2
LB = ψ̄BD/ B ψB + 2 εµνρ FBµν 1 − 2 ABρ , (2.2)
2gB Λ
and a regularized functional integration measure
p
[dψ̄][dψ][dA] D B + Mj )cj ,
det(/ cj = −1, cj Mj = 0, (2.3)
j j j
where p = 1, 2, . . . , and the Mj have to tend to infinity. The determinants in (2.3)

come from integrating out the Pauli–Villars fields. The superscripts B mean bare. Finally,
I identify

cj ln Mj /µ = − ln Λ/µ, cj Mj ln Mj /µ = bΛ, (2.4)
j j
b being an unspecified numerical factor. Here µ denotes the renormalization scale, but
the conditions (2.3) ensure that the identifications (2.4) are µ-independent, and therefore
consistent with renormalization-group invariance. It is also consistent to set b = 0, to kill
the linear divergence by default.
The regularized gauge-fixing terms are

1 2 2
Lgf = (∂µ Aµ ) 1 − 2 (∂ν Aν ) + C̄2 1 − 2 C.
2λg 2 Λ Λ
The ghosts decouple, as usual.
2.3. Renormalization
The renormalized Lagrangian reads

1 µνρ 2
LR = Zψ ψ̄/Dψ + 2 ε Fµν 1 − 2 Aρ + ΛδZΛ Zψ ψ̄ψ
2g Λ
and the renormalization constants have expansions
∞

Zψ = 1 + an (g, λ)(ln Λ/µ)n
n=1
(a) (b) (c)
Fig. 1. Simplest diagrams with the fermion bubble.
etc., where µ denotes the subtraction point, in the minimal subtraction scheme. Standard
Ward identities, combined with the properties of the Chern–Simons term, ensure that the
beta functions vanish, βg = βλ = 0 [14], and there is no need to insert renormalization
constants for the gauge field, ZA = Zg = Zλ = 1.
The perturbative results of this paper are written in the formalism of the dimensional
regularization technique and are easily converted to the cut-off regularization technique
defined above replacing 1/ε with ln Λ2 /µ2 and understanding that power-like divergences
are subtracted in the conformal scheme (i.e., the scheme that preserves conformal
invariance at the quantum level).
The lowest order values of the fermion renormalization constant and anomalous
dimension are given by the graph (c) of Fig. 1. Up to subleading corrections in 1/nf their
values are
g 4 nf 1 d ln Zψ g 4 nf
Zψ = 1 − , γψ = = .
384π 2ε 2 d ln µ 384π 2
3. Gravity coupled with Chern–Simons U (1) gauge theory and matter
The conventions for covariant derivatives Dµ , curvature R a , torsion Dea and spin
connection ωµa are given in Appendix A. The Lagrangian of gravity coupled with Chern–
Simons U (1) gauge theory and massless fermions reads
1 1
L= eR + eψ̄/Dψ + 2 εµνρ Fµν Aρ , (3.1)
2κ 2g
√
where e = g. The constant κ has dimensionality −1 in units of mass and serves as an
expansion parameter for the irrelevant couplings. The perturbative expansion is a double
expansion in powers of g and κE, where E is the energy scale.
The gravitational field is defined expanding the dreibein eµa around flat space:
eµa = δµa + φµa , ωµa = εabc ∂ b φµc + O(φ 2 ),

and choosing the symmetric gauge φµa = φaµ . It is convenient to gauge-fix both gravity
and the U (1) gauge field in flat space. The symmetric gauge is algebraic and so can be
imposed directly. The gauge-fixing sector of the theory is therefore
1 1
Lgf = (∂µ φµν )2 + (∂µ Aµ )2 + Lghost.
2ακ 2λg 2
The gauge parameters λ and α are kept throughout the calculations, because a powerful
way to check the results is to check the gauge-fixing independence of various quantities.
The ghost part of the gauge-fixing Lagrangian is derived in detail in Appendix A. The
ghosts do not contribute to the quantities calculated in this paper (this is proven in
Section 5).
3.1. Regularization
The theory (3.1) is power-counting non-renormalizable, and therefore, up to miraculous

cancellations (that the results of this paper exclude), divergences can be removed
introducing infinitely many new coupling constants, multiplying all possible irrelevant
operators. The vertices of the complete theory can contain arbitrarily many derivatives
and the regularized propagators should tend to zero faster than any power at large
momenta. The most convenient cut-off regularization framework for the coupled theory is
a Slavnov higher-derivative regularization where propagators are exponentially corrected.
For example, the Chern–Simons field is regularized with

1 µνρ 2
ε Fµν exp − Aρ + non-minimal. (3.2)
2g 2 Λ2
The D’Alembertian is the covariant one and non-minimal terms have to be fixed to ensure
that the integrated regularized Chern–Simons term is gauge-invariant. To the order 1/Λ2
we need to add
1
− εµνρ Fµν Rρα Aα .
2g 2 Λ2
It is immediate to prove that there exist appropriate non-minimal terms to all orders
in 1/Λ2 . Similar operations can be used to introduce appropriate exponentials in
the graviton and fermion propagators. However, the exponentials regularize only the
superficial divergences of diagrams with more than one loop. One-loop divergences and
subdivergences have to be regularized apart, for example, with the gauge-invariant Pauli–
Villars method of Faddeev and Slavnov [21].
The existence of a manifestly gauge-invariant regularization ensures the absence of
gauge anomalies to all orders in perturbation theory.
3.2. Result
The four-fermion divergent vertex (1.4) is generated at the second order in the loop
expansion and first order in the κ-expansion, up to subleading corrections in 1/nf .
Therefore, finiteness of three-dimensional gravity does not survive the coupling with
matter. To renormalize (1.4), it is first necessary to add new vertices and coupling constants
to the theory (3.1),
1 1 e e
L= eR + eψ̄/
Dψ + 2 εµνρ Fµν Aρ + κλ1 (ψ̄ψ)2 + κλ2 (ψ̄γ a ψ)2 + O(κ 2 ).
2κ 2g 4 4
(3.3)
Then, it is necessary to renormalize the new couplings by means of suitable renormaliza-

tion constants, λ1,2B = λ1,2 Z1,2 , and redefine the fields. The field redefinitions have the
form
1/2
AµB = Aµ + O(κ), ψB = Zψ ψ + O(κ), a
eµB = eµa + O(κ).
As in every non-renormalizable theory, the number of couplings is expected to grow
indefinitely with the order of the perturbative expansion in κ. Thus (3.3), as a fundamental
theory, is not physically predictive. Of course, it is still predictive as an effective field
theory.
4. Absence of the Lorentz–Chern–Simons term
The theory (3.1) is parity violating. A priori, renormalization might generate a Lorentz–
Chern–Simons counterterm (1.3). Now I prove that this does not happen. Basically,
this term is finite and therefore it is possible to set its renormalized coupling to zero
consistently.
The Lorentz–Chern–Simons term has dimensionality 3 in units of mass, so the
contributions to its renormalization must be O(κ 0 ). Two types of diagrams can contribute:
(i) diagrams with internal gravitons and gravitational ghosts; (ii) diagrams with no internal
gravitons nor gravitational ghosts.
By dimensional counting, the diagrams of type (i) can only be one-loop. Indeed,
higher-loop diagrams with internal gravitons and/or gravitational ghosts contribute to
the renormalization of Lagrangian terms with dimensionality greater than 3. One-loop
diagrams, on the other hand, have no logarithmic divergence in three dimensions.
The diagrams of type (ii) can be studied in external gravity, using properties of the trace
anomaly. For concreteness, I consider the case of Chern–Simons U (1) gauge theory with
massless fermions, but the generalization of the proof is immediate.
It is useful to include the Lorentz–Chern–Simons term in the renormalized Lagrangian
of the theory embedded in external gravity. In general, the Lorentz–Chern–Simons term
has to be multiplied by a coupling constant ζ plus a counterterm ∆ζ ,
1 µνρ
LR = eZψ ψ̄/
Dψ + ε Fµν Aρ
2g 2

1 a b c abc
+ (ζ + ∆ζ ) ε µνρ
a a
+ Lgf .
3
The regularizing terms are not written explicitly. Obviously, Zψ and ∆ζ depend only on g
and the gauge-fixing parameter λ, but not on ζ , because the Lorentz–Chern–Simons term
is just the identity operator, in external gravity. Now I prove that ∆ζ = 0.
The stress tensor is expressed as a functional derivative of the action,
eνa (x) δS eµa (x)g νρ (x) δS
Tµν (x) = + . (4.1)
2e(x) δeµa (x) 2e(x) δeρa (x)
In the differentiation, the gauge-fixing and ghost terms can be ignored, since they add
gauge-exact contributions, which do not affect the physical correlation functions. Because
of the non-local regularization (3.2), the functional differentiation of (4.1) is involved. It is

convenient to focus the attention on the integrated stress tensor

e(x)Tµν (x) d3x,
and in particular, the integrated trace

e(x)Θ(x) d x = e(x)Tµµ (x) d3x.
3
(4.2)
Inside (4.2), the differentiation (4.1) simplifies considerably. For example, it is possible
to treat the spin connection, Kristoffel symbols and curvatures as constants, because the
functional differentiation of objects such as ∂µ gνρ and ∂µ eνa produces total derivatives,
which are killed by the spacetime integration of (4.2). In practice, the operation (4.2)
reduces to a constant Weyl rescaling. Since the theory depends on a unique scale, at the
bare level, namely, the cut-off Λ, the result is easily proved to be

∂S
e(x)Θ(x) d x = −2 eZψ ψ̄/
3
Dψ − Λ , (4.3)
∂Λ B
where the subscript means that the bare fields and coupling constants are kept fixed in
the Λ-differentiation. Since the renormalization constants depend only on ln Λ/µ, the
expression (4.3) can be easily converted into a renormalized equivalent,

∂S ∂S
e(x)Θ(x) d3x = −2 eZψ ψ̄/ Dψ − Λ −µ . (4.4)
∂Λ ∂µ
Now, consider a convergent correlation function
G(x1 . . . xn , y1 . . . ym , z1 . . . zm )
= Aµ1 (x1 ) · · · Aµn (xn )ψ̄(y1 ) · · · ψ̄(ym )ψ(z1 ) · · · ψ(zm ). (4.5)
The Λ∂/∂Λ derivative of a renormalized correlation function is zero in the Λ → ∞ limit,
by definition. Such derivative is equal to the insertion of −Λ∂S/∂Λ, so the first term
of (4.4) can be ignored when the integral of Θ is inserted inside these correlation functions.
The result is

∂S
e(x)Θ(x) d3x = −2 eZψ ψ̄/ Dψ − µ .
∂µ
Since the theory is conformal in flat space, the couplings do not run, apart possibly from ζ ,
and so the partial ln µ derivatives of the renormalization constants can be replaced with
total ln µ derivatives, up to terms proportional to the ζ beta function βζ = −d∆ζ /d ln µ.
The result is

e(x)Θ(x) d x = −2(1 + γψ ) eZψ ψ̄/
3
Dψ

1 a b c abc
+ βζ ε µνρ
a a
.
3
The integral signs can be removed up to total derivatives. The only ambiguity is a term
∂µ (ψ̄γ µ ψ), which however cannot appear, because it violates the symmetry under charge
conjugation (see Appendix A).
The result is

1
Θ(x) = −2(1 + γψ )[Eψ ] + βζ e−1 εµνρ ωµa ∂ν ωρa + ωµa ωνb ωρc εabc , (4.6)
3
where

1 ←
→ 1 δl S δr S
/ ψ = e−1 ψ̄
[Eψ ] = Zψ ψ̄ D + ψ ,
2 2 δ ψ̄ δψ
and δl , δr denote the left and right functional derivatives, respectively. The operator
[Eψ ](x) is proportional to the fermion field equation and therefore is finite. An immediate
proof is that inserting [Eψ ](x) in the correlation function (4.5) simply multiplies it by
1
m

δ(x − yi ) + δ(x − zi ) . (4.7)
2e(x)
i=1
This result is standard and follows from a functional integration by parts.

Moreover, the second term of (4.6) should simply not be there, because the unintegrated
Lorentz–Chern–Simons term is not Lorentz-invariant, while Θ is. Therefore, ζ does not
run:
βζ = 0, ∆ζ = constant. (4.8)
The constant can be moved inside ζ , so it is safe to write ∆ζ = 0.
Having proved that the renormalized coupling ζ does not run, it is meaningful to set
it to zero. This means that the subtraction scheme can be adapted in such a way that the
Lorentz–Chern–Simons term is absent at each order of the perturbative expansion. It is
worth mentioning that if the Lorentz–Chern–Simons term is treated within the minimal
subtraction scheme (or any generic scheme), finite contributions can survive and have to
be removed by hand. These facts have been recently confirmed by a number of explicit
two-loop computations [22].
To conclude, there exists a modified subtraction scheme where the finite part of the
Lorentz–Chern–Simons term is identically zero. This is important, because the theory
with ζ = 0, known as “topologically massive gravity” [4], is physically inequivalent to
the theory with ζ = 0. In the rest of the paper I focus on the theory with ζ = 0. On the
other hand, it is easy to prove that a small non-zero ζ does not change the two-loop results
of the next sections and does not affect the conclusion that quantum gravity coupled with
matter, as it stands, is not finite in three dimensions.
The arguments of this section are completely general and apply to every theory of matter
coupled with gravity. The generalization is straightforward.
Observe that Lorentz invariance is crucial in the derivation. The point is that the
unintegrated trace operator Θ(x) is Lorentz-invariant, but the unintegrated Lorentz–
Chern–Simons term is not. A similar argument proves that the beta function of the Chern–
Simons coupling g is zero [14]: the gauge invariance of Θ and the gauge non-invariance of
the unintegrated U (1) Chern–Simons term are not compatible with a running of g. Instead,
the invariance under diffeomorphisms is not helpful in this kind of reasonings, since the
unintegrated operator Θ(x) is not invariant under diffeomorphisms.
The second crucial point is the possibility to reduce to the theory in external gravity.
This is a lucky situation. The arguments based on the trace of the energy–momentum
tensor cannot be applied if gravity is dynamical, where the “energy–momentum tensor”
(by which I mean the derivative (4.1) of the action with respect to the metric) vanishes
identically using the field equations. Other definitions of the stress tensor for quantized
gravity are more tricky to use.
Finally, it is known that pure gravity can be related to a Chern–Simons theory in three
dimensions [3]. However, Witten’s arguments for finiteness are based on the possibility to
µ
express the action in a form that does not contain the inverse dreibein ea , nor the inverse
µν
metric tensor g . This is impossible if gravity is coupled with propagating matter.
These remarks explain why gravity coupled with matter can be not finite despite the fact
that there exists no graviton.
5. Two-loop calculations
In this section I describe the general setting of the two-loop computations.
5.1. Four-fermion vertices
I focus on the irrelevant terms of dimensionality four in units of mass, which are
D 2 ψ,
eψ̄/ eFµν F µν , εµνρ eρa Fµν ψ̄γ a ψ, e(ψ̄ψ)2 , e(ψ̄γ a ψ)2 . (5.1)
Only the last two terms are independent, as I now prove.
Considering the presence of irrelevant terms, necessary for renormalization, the most
general field equations have the form
ig 2
/ = O(κ),
Dψ Fµν + eεµνρ eρa ψ̄γ a ψ = O(κ), (5.2)
2
1 1 1 ←
→ 1 ←
→ 1 ←
→
Rµν − gµν R + eµa ψ̄γ a D ν ψ + eνa ψ̄γ a D µ ψ − gµν ψ̄ D / ψ = O(κ).
2κ 2 8 8 4
(5.3)
The first counterterm of (5.1) vanishes using the fermion field equation, up to higher
orders in κ, so it can be removed by means of a field redefinition. The second and third
counterterms in (5.1) are equal to the forth of (5.1) up to terms proportional to the field
equation of the gauge field (5.2) and terms of dimensionality greater than four. Finally, it is
immediate to prove, using Fierz identities, that the independent four-fermion vertices are
precisely the ones listed in (5.1).
The second term in (5.1) is an ordinary gauge field kinetic term. It has to be removed
with a field redefinition of the form
Aµ → Aµ + aκeεµνρ F νρ + bκeµa ψ̄γ a ψ, (5.4)
where a and b are numerical coefficients. A theory with a propagating gauge field is
physically inequivalent to (3.1)–(3.3). Moreover, power-counting has to be reconsidered
and the calculations have to be repeated using the complete gauge field propagator. Here I
stick to the theory (3.1)–(3.3).
For the calculations of this paper, the use of field equations to simplify the counterterms
amounts in practice to the replacements
ig 2
Rµνρσ → 0, Fµν → − eεµνρ eρa ψ̄γ a ψ, / → 0.
Dψ (5.5)
2
Collecting these observations, the two-loop counterterms have the form
grav κ κ
Lcounter = c e(ψ̄ψ)2 + d e(ψ̄γ a ψ)2 + O(κ 2 ), (5.6)
4 4
and the values of the numerical coefficients c and d have to be determined with an
explicit computation. Since the one-loop diagrams are convergent in three dimensions,
subdivergences are absent at two loops and the divergent parts are simple poles 1/ε or
simple logs ln Λ2 /µ2 .
5.2. Reduction of the number of diagrams
Observe that the counterterms (5.6) are necessarily polynomial in the number nf of
fermions. It is immediate to check the at the second loop order they are at most linear
in nf . A quadratic contribution in nf would come from two fermion loops. Two fermion
loops can be connected only by a four fermion vertex, otherwise the diagram is either not
one-particle irreducible or not two-loop. Then, however, the diagram factorizes into the
product of two one-loop diagrams, which are convergent in three dimensions.
The number of two-loop diagrams contributing to the four-fermion vertices is high. It is
convenient to concentrate on the contributions proportional to nf , given by the diagrams
that contain one fermion loop. Fermion loops with two external legs are shown in Fig. 1
and appear frequently as subdiagrams of the two-loop diagrams. Fermion loops with three
gauge field legs or one graviton leg and two gauge field legs can also appear inside the
two-loop diagrams.
As anticipated in Section 3, the ghosts do not contribute to the results of this paper.
Indeed, the relevant Feynman diagrams do not have external ghost legs (diagrams with
external ghost legs affect only the gauge-trivial sector of the theory). Diagrams with
internal ghost legs giving linear contributions in nf must have a ghost loop and a
fermion loop. Arguing as above, these diagrams factorize into the product of two one-loop
subdiagrams and therefore converge.
The one-loop self-energy of the gauge field is given by Fig. 1(a) and is equal to
nf 1
− 2 (1+ε)/2
(δµν k 2 − kµ kν ).
16 (k )
The ε-dependence in the power of k is kept, because it affects the pole parts of the two-loop
diagrams that contain the fermion bubble as a subdiagram.
Obvious considerations based on spin conservation imply that the graviton gauge field
self-energy of Fig. 1(b) is identically zero. This fact can be immediately checked with an
explicit calculation.
5.3. Calculations
The divergent parts of the diagrams can be evaluated with the techniques that follow.
First, the diagrams are contracted with external momenta, Dirac matrices, Kronecker
tensors and ε tensors in all possible ways, and traced in spinor indices. Then, the results of
these operations are differentiated a sufficient number of times with respect to the external
momenta, to arrive at dimensionless integrals, and the external momenta are set to zero.
Scalar products of internal momenta in the numerators are converted into sums of squares,
using, for example,
1 2
p·q = p + q 2 − (p − q)2 .
2
After a number of such algebraic manipulations, the calculation is reduced to a set of
integrals of the form

dD p dD q 1
, (5.7)
(2π) (2π) [p ] [q ] [(p − q)2 ]c
D D 2 a 2 b
where D = 3 − ε and a, b, c are integers such that a + b + c = 3. It is convenient to imagine

that the fermions have a mass, to avoid IR divergences at zero external momenta, and in
some diagrams it is also useful to give fictitious masses to the U (1) field and the graviton.
The unique non-trivial contributions comes from the two-loop “master” integral

dD p dD q 1 1
= + finite part. (5.8)
(2π) (2π) p q (p − q)
D D 2 2 2 32π 2 ε
The other integrals (5.7) are convergent. Indeed, if a, b, c are not all equal to one, then at
least one of them is zero or negative, so there are only two denominators. Integrals with
two denominators factorize, eventually after a translation, into the product of two integrals
of the form

dD p pµ1 · · · pµn
(2π)D [p2 ]m
which are convergent. Also the integral (5.8) can be reduced to the product of two integrals,
using the technique of partial integration [23]

dD p dD q ∂ pµ
0= ,
(2π)D (2π)D ∂pµ p2 q 2 (p − q)2
but this operation factorizes a 1/ε.
The manipulations described so far are reversible, in the sense that it is possible to
reconstruct the structure of the divergent parts of the diagrams, using the fact that they are
local in the external momenta.
6. Four-fermion vertices induced by gravity
The counterterms of dimensionality 4 induced by gravity are proportional to the Newton

constant κ and can therefore be computed at λ1 = λ2 = 0.
Fig. 2. Two-loop self-energy of the gauge field with an internal graviton.
Three classes of diagrams contribute: the gauge field two-point function, the fermion
gauge field three-point function and the fermion four-point function. The divergent
diagrams contain one internal graviton leg, one fermionic loop and zero, one or two internal
gauge field legs, respectively. There is no contribution to the fermion self-energy, since the
potentially relevant diagrams contain the subdiagram of Fig. 1(b). In the figures, diagrams
are depicted up to permutations of external legs and reversions of the fermion arrows. It is
not necessary to compute diagrams with two external fermion legs plus two external gauge
field legs, because they do not give new gauge-invariant contributions. To the order O(κ) it
is not even necessary to consider diagrams with external graviton legs. Such diagrams
either contribute to the gauge-trivial sector of the Lagrangian or factorize a curvature
tensor. Then, using (1.2) and the graviton field equation of (5.2), they can be converted
into O(κ 2 ) counterterms.
6.1. Gravitational contribution to the self-energy of the gauge field
The graphs that have an internal graviton leg and contribute to the two-loop self-energy
of the gauge field are shown in Fig. 2. The counterterms associated with these graphs sum
up to
5nf κα
Lcounter-1 = − eFµν F µν . (6.1)
384π 2 ε
Observe that (6.1) is proportional to the gauge-fixing parameter α and should therefore be
cancelled by some other contribution (see below).
6.2. Gravitational corrections to the fermion gauge field vertex
The divergent parts of these graphs, which are shown in the first half of Fig. 3, are
subtracted by the counterterm
iκg 2 nf
Lcounter-2 = − (3 + 10α)εµνρ eρa Fµν ψ̄γ a ψ. (6.2)
768π 2 ε
This is a Pauli term, but using the field equations (5.5) it can be converted into a four-
fermion vertex.
6.3. Gravitational contribution to the fermion four-point function
The four-fermion counterterms that cancel the poles of the graphs shown in the second
half of Fig. 3 are
κg 4 nf e
Lcounter-3 = (1 + 10α) (ψ̄γ a ψ)2 . (6.3)
384π 2 ε 4
Fig. 3. Two-loop gravitational corrections to the fermion gauge field vertex and to the four-fermion vertices.
Summing the three contributions (6.1)–(6.3) and using the substitutions (5.5), the
α-dependence drops out and we obtain
grav 5κg 4 nf e
Lcounter = Lcounter-1 + Lcounter-2 + Lcounter-3 = − (ψ̄γ a ψ)2
384π 2ε 4
plus terms proportional to the field equations. The field redefinition that reabsorbs the terms
proportional to the field equations reads
5nf αg 2 κ inf g 4 κ(3 + 5α) a
Aµ → Aµ − eε µνρ F νρ
− eµ ψ̄γ a ψ.
768π 2ε 768π 2ε
7. Conclusions
In this paper I have studied the renormalization of three-dimensional quantum gravity

coupled with matter. Using standard arguments it is possible to show that the Lorentz–
Chern–Simons term is not renormalized and therefore there exists a subtraction scheme
where it is identically absent. Instead, irrelevant counterterms cannot be excluded a priori.
I have performed a two-loop computation in a concrete model, gravity coupled with
Chern–Simons U (1) gauge theory and massless fermions, to show that it is not finite.
The calculation can be simplified in various ways, but involves a considerable number of
diagrams. A good source of checks is the gauge-fixing independence of the final result.
A four-fermion counterterm
5κg 4 nf e
− (ψ̄γ a ψ)2
384π 2ε 4
is turned on by renormalization. Therefore, finiteness is violated at the second order in the

loop expansion and first order in the κ expansion.
Acknowledgement
I am grateful to P. Menotti for drawing my attention to references on 2 + 1 quantum

gravity.
Appendix A
In this appendix I recall some basic formulas, useful to fix the notation and simplify the
analysis of the graphs. I also comment on the difficulties to treat the Chern–Simons form
in the context of the dimensional regularization technique.
A.1. Curved space convention
Torsion, curvatures, covariant derivatives and connections are

1
Dea = dea + εabc ωb ec = 0, R a = dωa + εabc ωb ωc ,
2
Dµ Vν = ∂µ Vν − Γµν
ρ
Vρ , ρ
Γµν = eρa ∂µ eνa + εabc ωµa eνb eρc ,

1
ωµa = εabc ∂µ eνb − ∂ν eµb eνc − eµa εbcd ∂ρ eνb − ∂ν eρb eνc eρd ,
4
i
Dµ ψ = ∂µ ψ − ωµa γ a ψ + iAµ ψ.
2
The Ricci tensor and scalar curvature are defined as Rµν = Rµρ
ab e ρb e a , R = R g µν , where
ν µν
1 ab µ ν
R ab = εabc R c = Rµν dx dx ,
2
µ λ µ
R µ νρσ = ∂σ Γνρ
µ
− ∂ρ Γνσ
µ
− Γνσ
λ
Γλρ + Γνρ Γλσ ,
and of course gµν = eµa eνa .
A.2. Ghosts
The ghosts are: C for U (1), C µ for diffeomorphisms and C ab = −C ba for Lorentz
rotations. Indices are raised and lowered with δµa . The BRST variations of the fields read
sAµ = ∂µ C − Aρ ∂µ C ρ − C ρ ∂ρ Aµ , seµa = −eρa ∂µ C ρ − C ρ ∂ρ eµa − C ab eµb ,

sC = −C ρ ∂ρ C, sC ab = −C ac C cb − C ρ ∂ρ C ab , sC ρ = −C σ ∂σ C ρ .
µ
It is necessary to introduce antighosts C̄, C̄ a and C̄ µa = −δb δνa C̄ νb . The ghost Lagrangian
reads

Lghost = ∂µ C̄ ∂µ C − Aρ ∂µ C ρ − C ρ ∂ρ Aµ

+ C̄ µa − ∂µ C̄ a eρa ∂µ C ρ + C ρ ∂ρ eµa + C ab eµb .

Contracted indices may appear both as subscripts or superscripts in Euclidean flat
space. We have two ghost sectors: U (1) and gravitational (diffeomorphisms plus Lorentz
symmetry). The two sectors have a diagonal quadratic Lagrangian, but mix due to a vertex
of the form C̄AC ρ .
A.3. Propagators
The gauge field propagator is

i kρ kµ kν
Aµ (k)Aν (−k) free = − g 2 εµνρ 2 + g 2 λ 4 .
2 k k
The graviton propagator reads

φµν (p)φρσ (−p) free
κ 1 κ
= (δµρ δνσ + δµσ δνρ − 2δµν δρσ ) + 4 (δµν pρ pσ + pµ pν δρσ )
2 p2 p

1 κ
+ α− (δµρ pν pσ + δνρ pµ pσ + δµσ pν pρ + δνσ pµ pρ )
2 p4
pµ pν pρ pσ
− 3ακ .
p6
A.4. Difficulties of the dimensional regularization technique in curved space
Here I collect some observations about the difficulties to define a consistent dimensional
regularization for the Chern–Simons term in curved space. If the theory contains two-
component spinors, it is possible to define the tensor
i
E abc = − tr γ a γ b γ c ,
2
where γ a are the dimensionally continued Pauli matrices. If the trace is assumed to be
cyclic, the E tensor is set to zero by the dimensional regularization [24]. However, if the
theory contains two-component fermions, the D = 3 limit of E abc should be the ordinary
ε tensor. In curved space the situation worsens. Since the Pauli matrices are constant and
covariantly constant, so is the E tensor, assuming that it exists: ∂µ E abc = Dµ E abc = 0.
The Bianchi identity following from these equations is
Rµν E + Rµν
ad dbc cd dab
E + Rµν
bd dca
E = 0. (A.1)
To define the propagator of the U (1) gauge field, it would be useful to have an “inverse”
of the E tensor, for example, an E tensor satisfying
1 abc
E abc E mnp = δ . (A.2)
3! mnp
However, contracting (A.1) with the E tensor it is immediate to obtain

(D − 3)Rµν
ab
= 0, (A.3)
which implies that either the dimension of spacetime is exactly equal to 3 or the spacetime
is flat.
ad replaced by E adm . This follows
Moreover, an identity similar to (A.1) holds with Rµν
immediately from the definition (A.2). Then,
(D − 3)E abc = 0.
This implies that the E tensor does not exist in D dimensions.
An alternative approach is to split the D-dimensional spacetime into the tensor product
of a three-dimensional spacetime and a (−ε)-dimensional spacetime, as is commonly done
in four dimensions to define the matrix γ5 and the tensor εµνρσ [24]. Let µ, µ̄, µ̃ denote the
D-dimensional, three-dimensional and (−ε)-dimensional spacetime indices, respectively.
The kinetic Lagrangian of the U (1) field lives in the three-dimensional subspace. The
(−ε)-dimensional component Aµ̃ of the U (1) gauge field appears only in the Dirac
term and thus has no kinetic term. A way to treat a situation like this can be found in
Ref. [18], using the large-nf expansion. The missing kinetic term is provided by the
fermion bubble, which is leading in the large-nf expansion. Alternatively, it is possible
to add Fµν2 multiplied by 1/Λ, where Λ is a further cut-off, that is sent to infinity after
ε → 0.
References
[1] G. ’t Hooft, M. Veltman, One-loop divergences in the theory of gravitation, Ann. Inst. H. Poincaré 20 (1974)
69.
[2] M.H. Goroff, A. Sagnotti, The ultraviolet behavior of Einstein gravity, Nucl. Phys. B 266 (1986) 709.
[3] E. Witten, (2 + 1)-dimensional gravity as an exactly soluble system, Nucl. Phys. B 311 (1988) 46.
[4] S. Deser, R. Jackiw, S. Templeton, Topologically massive gauge theories, Ann. Phys. 140 (1982) 372.
[5] S. Deser, R. Jackiw, G. t’ Hooft, Three-dimensional Einstein gravity: dynamics of flat space, Ann. Phys. 152
(1984) 220.
[6] G. ’t Hooft, Nonperturbative two particle scattering amplitudes in (2 + 1)-dimensional quantum gravity,
Commun. Math. Phys. 117 (1988) 685.
[7] S. Carlip, Quantum Gravity in 2 + 1 Dimensions, Cambridge Univ. Press, Cambridge, 1998.
[8] S. Carlip, J. Gegenberg, Gravitating topological matter in 2 + 1 dimensions, Phys. Rev. D 44 (1991) 424.
[9] J.E. Nelson, T. Regge, 2 + 1 quantum gravity, Phys. Lett. B 272 (1991) 213.
[10] L. Freidel, E.R. Livine, C. Rovelli, Spectra of length and area in 2 + 1 Lorentzian loop quantum gravity,
Class. Quantum Grav. 20 (2003) 1463.
[11] J. Ambjorn, J. Jurkiewicz, R. Loll, 3D Lorentzian, dynamically triangulated quantum gravity, Nucl. Phys. B
(Proc. Suppl.) 106 (2002) 980, hep-lat/0201013.
[12] L.V. Avdeev, G.V. Grigorev, D.I. Kazakov, Renormalization in Abelian Chern–Simons field theories with
matter, Nucl. Phys. B 382 (1992) 561.
[13] L.V. Avdeev, D.I. Kazakov, I.N. Kondrashuk, Renormalizations in supersymmetric and nonsupersymmetric
non-Abelian Chern–Simons field theories with matter, Nucl. Phys. B 391 (1993) 333.
[14] A. Blasi, N. Maggiore, S.P. Sorella, Nonrenormalization properties of the Chern–Simons action coupled to
matter, Phys. Lett. B 285 (1992) 54, hep-th/9204045.
[15] G. Parisi, The theory of non-renormalizable interactions. I. The large-N expansion, Nucl. Phys. B 100 (1975)
368.
[16] D.J. Gross, Applications of the renormalization group to high-energy physics, in: R. Balian, J. Zinn-Justin
(Eds.), Methods in Field Theory, Les Houches, Session XXVIII, North Holland, Amsterdam, 1976.
[17] B. Rosenstein, B. Warr, S.H. Park, Dynamical symmetry breaking in four-fermion interaction models, Phys.
Rep. 205 (1991) 59.
[18] D. Anselmi, Large-N expansion, conformal field theory and renormalization-group flows in three
dimensions, JHEP 0006 (2000) 042, hep-th/0005261.
[19] D. Anselmi, “Integrability” of RG flows and duality in three dimensions in the 1/N expansion, Nucl. Phys.
B 58 (2003) 440, hep-th/0210123.
[20] H. Kawai, Y. Kitazawa, M. Ninomiya, Renormalizability of quantum gravity near two dimensions, Nucl.
Phys. B 467 (1996) 313, hep-th/9511217.
[21] L.D. Faddeev, A.A. Slavnov, Gauge Fields, Introduction to Quantum Theory, Benjamin/Cummings, 1980,
Section 4.4.
[22] F. Landolfi, S. Benvenuti, unpublished.
[23] K.G. Chetyrkin, F.V. Tkachov, Integration by parts: the algorithm to calculate beta functions in 4 loops,
Nucl. Phys. B 192 (1981) 159.
[24] J.C. Collins, Renormalization, Cambridge Univ. Press, Cambridge, 1984.
Giant hedge-hogs: spikes on giant gravitons

Darius Sadri a,b , M.M. Sheikh-Jabbari a
a Department of Physics, Stanford University, 382 via Pueblo Mall, Stanford, CA 94305-4060, USA
b Stanford Linear Accelerator Center, Stanford, CA 94309, USA
Received 12 January 2004; accepted 16 March 2004
Abstract
We consider giant gravitons on the maximally supersymmetric plane-wave background of type IIB
string theory. Fixing the light-cone gauge, we deduce the low energy effective light-cone Hamiltonian
of the three-sphere giant graviton. At first order, this is a U (1) gauge theory on R × S 3 . We place
sources in this effective gauge theory. Although non-vanishing net electric charge configurations are
disallowed by Gauss’ law, electric dipoles can be formed. From the string theory point of view these
dipoles can be understood as open strings piercing the three-sphere, generalizing the usual BIons
to the giant gravitons (BIGGons). Our results can be used to give a two-dimensional (worldsheet)
description of giant gravitons, similar to Polchinski’s description for the usual D-branes, in agreement
with the discussions of hep-th/0204196.
1. Introduction
Giant gravitons [1] were first discussed in the context of (m − 2)-branes moving on
the sphere in an AdSn × S m background, where it was observed that such particles blow up
inside S m , losing their point-like structure, and where their size was related to their angular
momentum. In fact they are branes which couple to background form fields as dipoles,
in contrast to the manner in which flat branes couple to form fields; they carry zero net
form field charge, but a non-vanishing dipole moment. It is this dipole coupling that is
responsible for their blowing up. In the AdS5 × S 5 background, they are three-dimensional
branes. Initial interest in these objects arose from their connection to non-commutative
physics, and the scaling of their size with angular momentum was recognized as a hallmark
E-mail addresses: darius@stanford.edu (D. Sadri), jabbari@itp.stanford.edu (M.M. Sheikh-Jabbari).
doi:10.1016/j.nuclphysb.2004.03.013
162 D. Sadri, M.M. Sheikh-Jabbari / Nuclear Physics B 687 (2004) 161–185
of non-commutativity. The giant gravitons preserve the same supercharges as the graviton
multiplet [2], and are 1/2 BPS, forming short representations of the superalgebra.
Giant gravitons which expand into the AdS part of the space–time have also been
constructed, and they carry the same quantum numbers as sphere giant gravitons [2,3].
The vibration spectrum of small fluctuations for giant gravitons which have expanded in
either the AdS or the sphere directions have been studied [4], where it was found that
the masses were independent of the radius and angular momentum of the giant graviton,
depending only on the curvature scale of the background.
BPS solutions of type IIB supergravity describing giant gravitons carrying angular
momentum along the sphere are available [5], and collections of giant gravitons act as
external sources which give rise to extremal limits of charged black holes (superstars),
where the horizon coincides with the singularity (which is hence naked) in the AdS
component of the space–time. Solutions of eleven-dimensional supergravity (again BPS)
characterizing giant gravitons on AdS7 × S 4 and AdS4 × S 7 appeared in [6], where they also
found to contain naked singularities, sourced by giant gravitons interpreted as spherical M2
and M5 branes.
The AdS/CFT duality suggests that giant gravitons should correspond to some
operators in a dual conformal field theory, where these operators are chiral primary. The
dual operators (for both sphere and AdS giants) have been constructed [7–11], and some
correlation functions have also been computed. The sphere giant gravitons correspond to
operators constructed from determinants and subdeterminants of the scalar fields in the
N = 4 super-Yang–Mills theory (e.g., see footnote 4). The determinants are associated
with maximum size giant gravitons, carrying the maximum angular momentum on the S 5 .
Similarly there have been proposals for the operators dual to giant gravitons inside AdS [8].
A new maximally supersymmetric type IIB supergravity solution (“the” plane-wave),
arising as the Penrose limit of AdS5 × S 5 [12], has attracted much interest in the literature,
largely because of its connection to the AdS/CFT duality [13], and the fact that the Green–
Schwarz superstring action, in light-cone gauge, is exactly solvable [14,15]. This plane-
wave/super-Yang–Mills duality is a specification of the usual AdS/CFT correspondence in
the Penrose limit; it states that strings on a plane-wave background are dual to a particular
large R-charge sector of N = 4, D = 4 superconformal U (N) gauge theory.1 The study
of giant gravitons was then extended from AdS5 × S 5 to the plane-wave, and interesting
issues stemming from the nature of the dual operators were addressed [9,17], in particular
the question of open strings in the dual gauge theory.
In a different line of pursuit, Callan, Maldacena and Gibbons [18,19] considered the
low energy effective theory for a single brane, which gives rise via the Dirac–Born–
Infeld action, to a U (1) gauge theory, and showed that electric point charges could be
interpreted as end-points of fundamental strings ending on the brane, and the dual magnetic
charges as D-strings similarly ending on the brane. They demonstrated the profile these
strings took and showed that they are in fact BPS solutions. These BPS solutions of the
linearized (Maxwell) equations match the solutions of the full non-linear Born–Infeld
theory equations, hence called BIons. A similar setup was argued to hold for other BPS
1 A review can be found in [16].

D. Sadri, M.M. Sheikh-Jabbari / Nuclear Physics B 687 (2004) 161–185 163
brane junctions, giving a characterization in terms of local configurations of fields in the

effective description of the brane in terms of a gauge theory.
We study giant gravitons on the plane-wave appearing as the Penrose limit of AdS5 × S 5 ,
with a particular focus on the behaviour of charges in the worldvolume gauge theory and
their interpretation in terms of open strings. We find solutions which generalize the usual
BIons to giant gravitons, allowing an open string worldsheet description of such giant
gravitons. The outline of our paper follows: in Section 2 we present the low energy effective
theory describing giant gravitons on the plane-wave, working in light-cone gauge. We find
two zero-energy configurations (vacua), corresponding to a zero-size giant graviton and
one of finite size, with radius given in terms of the string coupling gs , the light-cone
momentum p+ , and a scale µ for measuring energies. We then analyze the spectrum of
fluctuations, writing their eigenfrequencies and eigenmodes. We find agreement between
the physical modes and those of N = 4 super-Yang–Mills on R × S 3 . Higher order
corrections are studied, extracting the effective coupling of the theory, and the relation
of this coupling to the dual BMN gauge theory parameters is discussed. In Section 3,
we analyze the behaviour of the worldvolume theory when gauge fields are turned on,
presenting the spectrum of the gauge field. In Section 4, we turn our attention to the
BIon solutions on giant gravitons (BIGGons), explicitly solving for the scalar and gauge
field configurations, and interpret them via energy considerations as fundamental strings
piercing the giant gravitons. The supersymmetry and stability of the configurations is also
addressed. In a final section, we summarize our conclusions and outline possible future
directions for pursuit. An Appendix A is included, summarizing the harmonics which
appear in the main body of the paper.
2. Giant gravitons in the plane-wave background
In this section we focus on the (3 + 1)-dimensional Dirac–Born–Infeld action in

the plane-wave background. First we note that due to symmetries of the plane-wave
background, in particular translational symmetry along the light-like directions x + and
x − [16] (similar to the case of strings on the same background [14]), fixing the light-
cone gauge will simplify considerably the action. The zero energy solutions to the light-
cone Hamiltonian in the sector with light-cone momentum µp+ is a sphere of radius
R 2 = µp+ gs . This sphere is a giant graviton [1]. It is worth noting that fixing the light-
cone gauge, in the language of rotating (orbiting) branes of [1,3], corresponds to going to
the rest frame of the giant graviton.
We also study fluctuation modes of the giant gravitons by expanding the light-cone
Hamiltonian about the zero energy solutions. The frequencies of these modes, as in the
case of giant gravitons on the AdS5 × S 5 background [4], are independent of their radius.
We next turn on the fermions and work out the full fermionic terms of the light-cone
Hamiltonian and the frequencies of their small fluctuations. We also briefly discuss higher
order interaction terms in the Hamiltonian and the fact that they may be analyzed in a
systematic perturbation expansion with the effective coupling geff (2.40).
2.1. Low energy effective dynamics in light-cone gauge
The low energy effective action for a D-brane is
S = SDBI + SCS , (2.1)

with the Dirac–Born–Infeld action

SDBI = −Tp d p+1 ζ e−φ − det(Gµ̂ν̂ + Bµ̂ν̂ + Fµ̂ν̂ ), (2.2)
where hatted Greek indices are used for the worldvolume coordinates ranging from zero
to p. We have set 2πα = 1; factors of α can be reintroduced on dimensional grounds
when necessary. We will consider D3-branes, for which p = 3 and the dilaton background
is constant, in which case gs = eφ . We first consider the case where, in addition to the
constant dilaton, only the metric is turned on, and drop (consistently) the other forms. The
gauge field Fµ̂ν̂ , however, would be considered in Section 3. Our metric conventions are
those of Polchinski [20]; we work with a mostly plus metric for the worldvolume and target
space. Note that the physical tension for this D-brane is Tp /gs . The Chern–Simons term
describing the coupling to the background RR four form is

SCS = q C4 , (2.3)
with q the charge of the brane. For BPS configurations the charge and tension are equal.
Gµ̂ν̂ is the pullback of the space–time metric onto the worldvolume of the brane, and C4
is the pullback of the RR four-form. They are given by
Gµ̂ν̂ = ∂µ̂ Xµ ∂ν̂ Xν gµν (2.4)

and
1
C4 = ∂µ̂0 Xµ0 · · · ∂µ̂3 Xµ3 Cµ0 ...µ3 dζ µ̂0 ∧ · · · ∧ dζ µ̂3 . (2.5)
4!
The Xµ give the embedding coordinates of the brane in the target space–time, i.e.,
µ = 0, . . . , 9, and ζ µ̂ are local coordinates on the brane worldvolume. Cµ0 ...µ3 is the space–
time RR four-form coupling to the worldvolume.
We are working in a background specified by the maximally supersymmetric type IIB
plane-wave (here we will follow the notation and conventions of [16])

ds 2 = −2dx + dx − − µ2 x i x i + x a x a (dx + )2 + dx i dx i + dx a dx a , (2.6a)
4 4
F+ij kl = µij kl , F+abcd = µabcd . (2.6b)
gs gs
From (2.6b) it is easy to read off the RR four-form potential C, as F = dC, and we have
µ µ
C+ij k = − ij kl x l , C+abc = − abcd x d , (2.7)
gs gs
which has the virtue of maintaining the translational symmetry along x + . We have chosen
our coordinates to make manifest the SO(4) × SO(4) symmetry of the transverse directions,
labeling the two SO(4)’s with i, j = 1, 2, 3, 4; a, b, c, d = 5, 6, 7, 8. For a more detailed

discussion on the isometries of the background we refer the reader to [16].
We separate the space and time indices on the brane worldvolume as ζ = (τ = σ 0 , σ r ),
with p, q, r = 1, 2, 3, the space indices. We will fix the light-cone gauge, setting
X+ = τ. (2.8)
In order to ensure that the above solution for X+
is maintained by the dynamics, we
should use a part of the gauge symmetries of the DBI action, which are the area preserving
diffeomorphisms on the brane worldvolume, to set
G0r = −∂r X− + ∂τ XI ∂r XI = 0. (2.9)
We have used upper-case indices to denote all eight transverse coordinates, where I =
(i, a) = 1, 2, . . . , 8.
Next we note that the background (2.6) is X− independent, (it is a cyclic coordinate),
and hence the momentum conjugate to X− , the light-cone momentum p+ , is a constant of
motion:

∂L 1 √ 1 − det Grs
p+ = − = − G00 − det G = − . (2.10)
∂∂τ X− gs gs G00
To obtain (2.10) we have used the fact that G0r = 0 implies G00 = 1/G00 . The light-cone
Hamiltonian P − (i.e., momentum conjugate to X+ ) is then found to be
∂L
P− ≡ −
∂∂τ X+
1 √ 1
= − G00 − det G ∂τ X− + µ2 XI XI − rps C+I J K ∂r XI ∂p XJ ∂s XK
gs 6
1
= p+ ∂τ X− + µ2 XI XI − rps C+I J K ∂r XI ∂p XJ ∂s XK . (2.11)
6
Using (2.10) we can solve G00 and hence ∂τ X− for p+ and det Grs :
det Grs
G00 = −2∂τ X− − µ2 XI XI + ∂τ XI ∂τ XI = − . (2.12)
(p+ gs )2
Inserting ∂τ X− from (2.12) into the light-cone Hamiltonian and noting that the momenta
conjugate to XI are
∂L
PI = = p+ ∂τ XI , (2.13)
∂∂τ XI
we obtain the light-cone Hamiltonian density
1
Hl.c. = + P I P I + V Xi , Xa , (2.14)
2p
where
µ2 p + 2 1
V Xi , Xa = Xi + Xa2 + + 2 det Grs
2 2p gs
µ ij kl i j k l
− X X , X , X + abcd Xa Xb , Xc , Xd . (2.15)
6gs
In the above,
Grs = ∂r Xi ∂s Xi + ∂r Xa ∂s Xa ,
and the brackets are “Nambu brackets” defined as
{F, G, K} = pqr ∂p F ∂q G∂r K, (2.16)
where the antisymmetrization is with respect to worldvolume coordinates. It is worth
noting that as a result of light-cone gauge fixing the square-root in the DBI action has
disappeared (see (2.15)). This will help us perform a more detailed analysis of the light-
cone Hamiltonian. We should also keep in mind that in the light-cone gauge, ∂r X− are
totally determined in terms of XI through (2.9), i.e.,
−PI ∂r XI + p+ ∂r X− ≈ 0, (2.17)
where ≈ is the “weak” equality, meaning that (2.17) should hold on the solutions of the
equations of motion of the light-cone Hamiltonian.
2.2. Zero energy configurations
We now search for classical minima of the light-cone Hamiltonian, and expand the
potential V (Xi , Xa ) around these vacua to find the spectrum of small fluctuations about
the vacua. First we note that if we set Xa = 0, then
1 i j k i j k
det Grs = det(∂r Xi ∂s Xi ) = X ,X ,X X ,X ,X , (2.18)
3!
and hence the potential becomes a perfect square

i a 1 + i 1 j k l 2
V X , X = 0 = + µp X − ij kl X , X , X . (2.19)
2p 6gs
The above potential has a minimum at

µp+ gs ij kl Xl = Xi , Xj , Xk . (2.20)
Eq. (2.20) has two solutions, one is the “trivial” vacuum, Xi = 0, and the other one is a
three-sphere of radius R, where [13]2
R 2 = µp+ gs . (2.21)
In other words if we set

4
X = Rx ,
i i
xi2 = 1, (2.22)
i=1
it is easy to check that ij kl x i = {x j , x k , x l }. This three-sphere is a giant graviton.
2 Note that all lengths are measured in units of the string scale α . Recovering the α -factors, we have
R 2 /α = (µp + α )gs .
Both of these vacua are zero energy configurations. We could have easily found another
minimum (zero energy configuration) corresponding to a three-sphere grown in the Xa
directions sitting at Xi = 0. Note also that both of these vacua are 1/2 BPS; they annihilate
all the dynamical supercharges of the background. In other words, all the fermionic
generators of the PSU(2|2) × PSU(2|2) × U (1) superalgebra would kill these states.3
2.3. Spectrum about the vacua
We now study the spectrum of small fluctuations about these vacua. To do so, we expand
the theory about the vacua to second order in fluctuations.
2.3.1. Spectrum about X = 0 vacuum

In this case the det Grs and the bracket terms would not contribute to the quadratic
Hamiltonian; they appear in the interactions, and the quadratic parts of the Hamiltonian are
(2) 1 1 µ2 p + 2 µ2 p + 2
HX=0 = P i Pi + Pa Pa + Xi + Xa . (2.23)
2p+ 2p+ 2 2
Therefore, there are eight modes, all with frequency µ, that is the modes are particles of
mass µ. Of course for a generic low energy state one may excite many of these modes.
2.3.2. Spectrum about the three-sphere vacuum

If we parameterize the small fluctuations in the Xi directions by Y i , i.e., Xi = Rx i + Y i ,
the quadratic Hamiltonian becomes

1 1 1 R2 2
(2)
HX=R = + Pi Pi + + Pa Pa + + µp+ Yi − ij kl x j , x k , Y l
2p 2p 2p 2gs

1 2 R 4
+ + µp+ Xa2 + 2 ∂r Xa ∂s Xa g0rs , (2.24)
2p gs
where g0rs is the inverse of the metric on a unit three-sphere. The bracket can be used to
obtain generators of SO(4) rotations along the three-sphere, explicitly:
1
Lij Φ ≡ (xj ∂i − xi ∂j )Φ = − ij kl x k , x l , Φ . (2.25)
2
In terms of Lij the Hamiltonian takes a simple form
(2) 1 1 1 2 +
j 2
HX=R = Pi Pi + P a Pa + µ p Yi + Lij Y
2p+ 2p+ 2

1 1
+ µ2 p+ Xa2 − Xa Lij Lij Xa . (2.26)
2 2
3 Although both of these vacua are BPS, it has been argued that the X = 0 vacuum might be unstable under
certain quantum corrections [1,3]. These arguments were finally confirmed for the case of spherical membranes
and fivebranes in M-theory in a detailed analysis of the matrix theory describing M-theory on the maximally
supersymmetric eleven-dimensional plane-wave background, the BMN matrix theory. It has been shown that the
X = 0 vacuum for the membrane/fivebrane case is in fact a finite size fivebrane/membrane [21]. For the case of
spherical three-branes, to the authors’ knowledge the issue is not yet fully answered.
The normal modes for the Y i directions about this vacuum satisfy the eigenvalue equation
Lij Y j = λYi ,
with masses given by
M 2 = µ2 (1 + λ)2 . (2.27)
The eigenvectors are vector spherical harmonics of the form
Yli = Sii1 ...il x i1 · · · x il , l 0, (2.28a)

l−1
Ỹli = x i S̃i1 ...il−1 x i1 · · · x il−1 − S̃ii1 ...il−2 x i1 · · · x il−2 , l 1, (2.28b)
2l
where S and S̃ are symmetric traceless SO(4) tensors. The eigenvectors (2.28) correspond
to the eigenvalues λ = l, −(l + 2), respectively. Both of these modes, although in different
SO(4) representations, would have the same mass:
Mi = µ(l + 1). (2.29)

Physically these two modes correspond to geometric fluctuations of the brane in the radial
directions.
Of the modes describing the five directions X− , Y i , there remain three zero modes
Ŷli = Aii1 ...il x i1 · · · x il , l 1, (2.30)

where A is symmetric in all lower indices and antisymmetric in the first upper and first
lower index (and hence xi Ŷli = 0 for any l). These eigenvectors are associated with the
eigenvalue λ = −1. These zero modes are not physical and correspond to gauge degrees of
freedom associated with the area preserving diffeomorphisms on the three-sphere.
The masses for Xa fluctuations can be easily obtained, noting that the eigenvalues for
the SO(4) Casimir L2 , are −2l(l + 2), with the corresponding Xa
Xla = Sia1 ...il x i1 · · · x il , l 0. (2.31)

The masses are

Ma2 = µ2 l(l + 2) + 1 = µ2 (l + 1)2 . (2.32)
As we see, all the modes, Y i ’s and Xa ’s, have the same mass. This is a direct
result of the supersymmetry algebra of this background, which as discussed in [16] is
PSU(2|2) × PSU(2|2) × U (1), and the fact that the light-cone Hamiltonian commutes
with the supercharges; as a result all the states in the same supermultiplet should have the
same mass. This is in contrast with the eleven-dimensional plane-wave superalgebra [22].4
4 The Xi = Rx i vacuum, being a 1/2 BPS state, should, in the dual N = 4, D = 4 gauge theory, be
represented by a chiral primary operator. And in our case, since we are working in the plane-wave background,
it should be a BMN [13] type operator. The corresponding operators have been introduced and studied in [7–9,
11,17]. Let Zji , i, j = 1, 2, . . . , N , be one of the three complex scalar fields of an N = 4, D = 4, U (N ) gauge
2.3.3. Fermionic modes

For completeness we also work out the spectrum of fluctuations for the fermionic modes
about both vacua. The fermionic contributions to the DBI and CS parts of the action
can be found using superspace coset techniques which make the superalgebra manifest
(see, for example, [23], and [15], and also [21,24] for a similar treatment for the case of
the membrane and fivebrane on the eleven-dimensional maximally supersymmetric plane-
wave). After fixing κ symmetry in light-cone gauge [25], the new contributions to the
potential in the Hamiltonian are quadratic in the fermions, and are given by
V ψ = µψ † αβ ψαβ
2 † αβ ij δ i j
+ + ψ σ α X , X , ψδβ + ψ † αβ σ ab α δ Xa , Xb , ψδβ
p gs
+ µψ † α̇ β̇ ψα̇ β̇
2 † α̇ β̇ ij δ̇ i j
+ + ψ σ α̇ X , X , ψδ̇ β̇ + ψ † α̇ β̇ σ ab α̇ δ̇ Xa , Xb , ψδ̇ β̇ .
p gs
(2.34)
We have chosen to decompose the SO(4) × SO(4) fermions in terms of representations of
the two SU(2)’s appearing in each SO(4), following the notation of [16], making manifest
theory (for more on conventions and notations see [16]). Then [7]
5 1 j j ···j k ···k i1 i2 iJ
OJS = NJ i i ...i k ...k 1 2 J J +1 N Zj1 Zj2 · · · ZjJ , (2.33)
J !(N − J )! 1 2 J J +1 N
is the operator dual to a giant graviton grown in the S 5 direction (the normalization factor NJ2 = (N−J
N!
)!
is
S 5 S 5
chosen so that OJ ŌJ = 1).
1
i1 i i
OJAdS = Zi Z 2 · · · Zi J ,
J! σ (1) iσ (2) σ (J )
σ ∈ SJ
with SJ being the permutation group of length J , is proposed to describe giant gravitons grown in the AdS
directions [8]. Note that in the plane-wave case (after the Penrose limit) the two giant gravitons grown in S 5 and
in AdS essentially become indistinguishable.
Using the above operators one may construct the dual gauge theory operators corresponding to the fluctuation
modes of the giant graviton we have studied here. This can be done by insertion of “impurities” in the sequence
of Z’s, much like what has been done in [13] for strings. For example, if φa , a = 1, 2, 3, 4 denote the other four
scalars of the N = 4, D = 4 gauge multiplet, then the dual gauge theory operators for l = 0, 1 states of (2.31)
are [9,11]
Xa 1
OJ l=0 = NJ +1
(J + 1)!(N − J − 1)!
i
× i1 i2 ...iJ +1 kJ +2 ...kN j1 j2 ···jJ +1 kJ +2 ···kN (φ a )j1 Zj2 Zj3 · · · ZjJ +1 ,
i i i
1 2 3 J +1
Xa 1
OJ l=1 = NJ +2
(J + 2)!(N − J − 2)!
i
× i1 i2 ...iJ +2 kJ +3 ...kN j1 j2 ···jJ +2 kJ +3 ···kN (φ a )j1 Zj2 Zj3 · · · (φ b )jll · · · ZjJ +2 .
i i i i
1 2 3 J +2
Clearly these operators have ∆ − J = 1, 2, respectively. Similarly, one may construct higher l excitations by more
insertions of φ’s.
the fermion representations under the PSU(2|2) × PSU(2|2) part of the superalgebra of the
maximally supersymmetric plane-wave we are considering. The explicit mass terms in V ψ
for the fermions come from a shift in Gτ τ arising from fermionic contributions to the super-
vielbein and the terms involving Nambu brackets from the Chern–Simons terms. Following
the notation of [16], the fermions ψ are spinors of two different SU(2)’s, one coming from
the decomposition of each of the two SO(4)’s into SU(2) × SU(2); in other words, ψ above
carries two spinor indices (sitting in the same chirality representations). More details on
our spinor conventions can be found in [16]. Note that in the potential (2.34), the two sets
of fermions with dotted and undotted indices do not couple to each other.
Expanding the potential around the three-sphere solution, setting Xi = Rx i and Xa = 0,
the quadratic part of the potential (2.34) becomes, after using (2.21) and (2.25)
ψ
V(2) = µ ψ † αβ ψαβ − ψ † αβ σ ij α δ ij kl Lkl ψδβ

+ µ ψ † α̇ β̇ ψα̇ β̇ − ψ † α̇β̇ σ ij α̇ δ̇ ij kl Lkl ψδ̇ β̇ . (2.35)
The spectrum of small fluctuations around this vacuum is given by solutions of the
eigenvalue equation

ij kl σ ij α β Lkl ψβ = λψα , (2.36)
with similar equations for the other modes. We have for clarity suppressed one of the
indices since it is a bystander in the eigenvalue equation.
The frequencies (masses) are then given by
ω = µ|1 − λ|, (2.37)

where λ is the eigenvalue corresponding to the excitation mode. The eigenfunctions and
corresponding eigenvalues (suppressing the inactive spinor index) are

ψαl = θαi1 ...il + j i1 kl σ kl α β θβj i2 ...il x i1 · · · x il , λ = −l,
kl β i (2.38)
ψ̃α = lθαi1 ...il + (l + 2) 1
l i j kl σ α θβj i2 ...il x 1 · · · x l , λ = l + 2,
i
where l 0 and θ , carrying the spinorial index, forms a totally symmetric traceless
representation of SO(4) in the indices j, i1 , . . . , il .5 Therefore both ψ̃αl and ψαl excitations
have the same mass, |ω|, equal to µ(l + 1). As it is clear from (2.35), fermions ψα̇ β̇
would also have the same mass. Hence all the bosonic and fermionic excitations about
the Xi = Rx i vacuum (Yli , Ỹli , Xla ; ψαβ
l , ψ̃ l , ψ l , ψ̃ l ) have the same mass, as expected
αβ α̇ β̇ α̇ β̇
from the PSU(2|2) × PSU(2|2) × U (1) superalgebra, and fall into the same supermultiplet.
However, these modes do not complete the multiplet (as there are two more fermions than
5 It is straightforward but tedious to check that these are indeed eigenfunctions of (2.36), making use of the
identity
ij β kl ρ 1 ρ
σ α σ β = − δα δ ik δ j l − δ il δ j k + i ij kl
4

+ 2 δ ik σ j l α ρ + δ j l σ ik α ρ − δ il σ j k α ρ − δ j k σ il α ρ .
bosons). These two extra bosonic modes correspond to a U (1) gauge field living on the
giant graviton; we will come back to this point in Section 3.1.
The frequencies of small perturbations around the zero size giant graviton are simply
given by µ, arising from the explicit mass term in the potential. The masses are then the
same as the bosonic spectrum. In the X = 0 vacuum, as opposed to the spherical vacuum,
the number of scalar and fermionic excitations are both eight, i.e., there are no gauge field
modes.
Finally we would like to point out that, although we do not explicitly show it here, all the
modes, about both vacua, fall into a BPS (short) multiplet of the PSU(2|2) × PSU(2|2) ×
U (1) superalgebra. A study of representation theory of this superalgebra is an interesting
open problem in need of a thorough analysis.
2.4. Interaction terms
So far we have only considered the quadratic terms around each of the two vacua. One
may study the theory perturbatively about the spherical or X = 0 solution. The purpose
of this section is to find the effective coupling about these vacua and discuss under what
conditions the expansion around these vacua can be trusted.
Let us first consider the spherical vacuum. Expanding (2.14) about the Xi = Rx i
solution, we obtain the interaction terms which are from cubic up to sixth order in
fluctuations Y i or Xa . In order to read the coupling constant, however, we should redefine
(rescale) the fluctuations so that the quadratic part of the Hamiltonian takes the standard
† †
canonically normalized form of i h̄ωl al al , where al is the corresponding creation
operator and ωl is the mass of the mode, which in our case is µ(l + 1). For this we need to
rescale Y i and Xa as
1
Y i , Xa → Y i , Xa . (2.39)
µp+
As can be seen from (2.34), for the fermions no rescaling is needed. It is straightforward
to see that the cubic term is suppressed by a factor of geff , and likewise terms of order n in
n−2
fields are accompanied by a factor of geff , where
1 1
geff = √ = . (2.40)
µp+ g s R µp+
(Note that energy is measured in units of µ and hence one should take out a factor of µ
from the potential. This can be done systematically if we scale time with 1/µ.)
One might rewrite geff in terms of the BMN gauge theory parameters, J , N and gYM2 ,
where [16]
2 N
1 gYM J2
= ≡ λ , g2 ≡ .
(µp+ )2 J2 N
√
Then R 2 = λ g2 and geff 2 = 1/g . Noting that g is the genus counting parameter for
2 2
strings on plane-waves, (2.40) suggests that our giant graviton theory is somehow S-dual
to string theory on the plane-wave.6
One may repeat the same analysis for the X = 0 vacuum, for which we should use the
same scaling as above and hence we again end up with the same coupling as (2.40). We
caution that the above coupling should be thought of as a “bare” coupling and in a properly
quantized system this coupling may be dressed with some other factors of µp+ and also
this dressing factor can be different for different vacua. In this respect the situation is quite
similar to the membrane case which was analyzed in detail in [24]. However, in our case
we do not know how to quantize the Nambu brackets.
3. Gauge theory on giant gravitons
We are now ready to include the contribution from the gauge fields on the brane. The
analysis of Section 2 is modified slightly in this case when Fµ̂ν̂ = (dA)µ̂ν̂ is turned on.
The equations giving the metric on the brane as the pullback of the plane-wave space–time
metric are of course unaffected, but the determinant appearing in the DBI action receives
contributions from the gauge field strength. We write
Mµ̂ν̂ = Gµ̂ν̂ + Fµ̂ν̂ ,

where as before, Gµ̂ν̂ is the pullback of the space–time metric given by Eq. (2.4). As a part
of the light-cone gauge fixing, as we did in Section 2.1, we set G0r = 0 and hence
M00 = G00 , M0r = F0r = −Fr0 ≡ Er , Mrs = Grs + Frs , (3.1)

where Er is the electric field and G00 is still given by (2.12). The Chern–Simons terms
and also the fermionic contributions (2.34) are unaffected by the appearance of the gauge
fields. The contribution of the gauge field to the momentum conjugate to X− results in a
modification of (2.10), as
1 00 √
p+ = − M − det M, (3.2)
gs
but now M 00
= 1/M00 because of the off-diagonal electric field appearing in Mµ̂ν̂ ; it
becomes
det(Grs + Frs )
M 00 = . (3.3)
det M
6 It is interesting to note that the three point function of O S 5 (2.33) is given by [17]
J
−r(1−r)
S5 S5 S5 g2
OJ ŌrJ Ō(1−r)J e−g2 r(1−r)/2 ∼ e eff , (2.41)
where 0 r < 1 and by we mean that the result is presented after the BMN limit, i.e., J, N → ∞ and
J 2 /N = fixed. (2.41) shows that the above three point function corresponds to tunneling between two different
giant graviton states which is a non-perturbative effect in the giant graviton theory.
The determinant appearing in the action is also modified

det M = det(Grs + Frs ) G00 + Er Grs Es . (3.4)
Using (3.2), we can write
det(Grs + Frs )
− = G00 + Er Grs Es . (3.5)
(p+ gs )2
The expression for the momentum conjugate to X+ (the light-cone Hamiltonian) remains
as in (2.11), but with p+ now given by (3.2), i.e.,
1
P − = p+ ∂τ X− + µ2 XI XI − rps C+I J K ∂r XI ∂p XJ ∂s XK .
6
The momenta conjugate to X (2.13) are unaffected. We can solve for ∂X− in terms of XI
I
and their conjugate momenta PI , and also Er and Frs , using (3.2), (3.3) and (3.5). Since
the computations are very similar to those of Section 2.1 we do not repeat them here.
Gathering all the terms, the total light-cone Hamiltonian density becomes
PI2 1 1 1
Hl.c. = +
+ µ2 p+ XI2 + + 2 det(Grs + Frs ) + p+ Er Grs Es
2p 2 2p gs 2
µ ij kl i j k l b c d
+ X X ,X ,X + abcd a
X X ,X ,X . (3.6)
6gs
3.1. Spectrum of small fluctuations of the gauge field
From (3.6) it is readily seen that X = 0 and Xi = Rx i (and of course together with
Er = Frs = 0) are still the only zero energy configurations. Then one may expand the
theory about each of these vacua. The spectrum of Xi , Xa modes is the same as those we
studied in Section 2.3. (This statement is also true for fermionic modes. In fact one can
show that the full supersymmetric version of the Hamiltonian (3.6) is obtained by adding
(2.34) to (3.6). This in particular means that fermions do not directly couple to gauge fields.
The latter could be understood by noting that we are only dealing with a U (1) gauge theory
where fermions sit in the adjoint representation, i.e., they are neutral.)
For the X = 0 vacuum, there are no gauge field contributions, because the gauge field
terms only appear in quartic or higher powers. For example, the induced metric Grs is
second order in X fluctuations and hence the Er Grs Es term is at least quartic. This is
compatible with our earlier discussions in Section 2.3.3, that the fluctuations of Xi ’s, Xa ’s
and the corresponding fermionic modes, complete a PSU(2|2) × PSU(2|2) × U (1) (short)
multiplet. In this case, gauge fields only couple to “scalar” bosonic modes with the “bare”
coupling given by (2.40).
As for the Xi = Rx i vacuum, expanding (3.6) up to second order in all fluctuations we
obtain
(2) (2) 1
Hl.c. = Hl.c. + Er g0rs Es + µ2 Br g0rs Bs , (3.7)
2µgs
(2)
where Hl.c. is given in (2.14), g0rs is the (inverse) metric on the unit three-sphere and
rp (2)
g0 Bp = rqs Fqs is the magnetic field. It is worth noting that Hl.c. (plus (2.34)) is exactly
the Hamiltonian for an N = 4 U (1) gauge theory on R × S 3 (the latter action may be found
in [26]). Among the fields in the four-dimensional N = 4 gauge multiplet the gauge field,
four scalars, the Xa modes, and 8 fermions (which are in the correct representation, e.g.,
see [16]) are explicit. The other two scalar modes, however, are a combination of Y i ’s.
Although the explicit expressions defining these two scalars in terms of Y i ’s is not so
simple, as we discussed in Section 2.3.2, Y i ’s only contain two physical modes, with the
masses equal to the other four scalar, Xa modes, and also the fermions. Moreover, the
above argument would imply that the SO(4) symmetry rotating Xa ’s among each other,
can be generalized to SO(6) including these other two scalar modes, giving rise to the full
R-symmetry group of the N = 4 gauge theory.
In the same manner that we argued that the fluctuation modes of Section 2.3.2 complete
a PSU(2|2) × PSU(2|2) × U (1) multiplet, we should also work out the spectrum of the
gauge fields, i.e., photons, on the three-sphere, and show that the two polarizations of the
photon have the same “mass” as the fermions and scalars. Let us start with the equation of
motion for the gauge field, Aµ̂ :

1
gµ̂ν̂ ∇ − (∇µ̂ ∇ν̂ + ∇ν̂ ∇µ̂ ) Aν̂ = 0.
0 2
2
We choose to work in Coulomb gauge, i.e., A0 = 0, ∇µ̂ Aµ̂ = 0.7 Next we note that ∇r and
the gradient on the unit three-sphere do not commute. In fact they commute to the Riemann
tensor. For a unit sphere Rrs = 2grs , where Rrs is the Ricci tensor, and we have
2
ω − µ2 ∇r2 + 1 As = 0. (3.8)
If we take Ar to be in the spin l representation of SO(4), i.e., ∇ 2 Alr = l(l + 2)Alr , the
spectrum of the two photon polarizations is obtained to be
ω = µ(l + 1), l 1. (3.9)

As we see the “mass” for photons has a purely geometric origin. (The same is also true
for fermions.) This is in contrast with that of scalars, where we have an explicit mass
term. Eq. (3.9), as we expected from the superalgebra arguments, leads exactly to the
same spectrum as scalars and fermions. These two photon modes together with the other
excitations studied in Section 2.3.2 complete a multiplet of the PSU(2|2) × PSU(2|2) ×
U (1) algebra.
4. BIGGons: BIons on giant gravitons
In this section we focus on a U (1) gauge theory on R × S 3 and study static, BPS charge
configurations. We cannot have non-zero electric net charge (S 3 is compact), however,
higher-pole configurations are allowed. We study the dipole configurations in some detail.
7 We would like to point out that working with the light-cone gauge in the bulk, as we have done here, does
not necessarily imply that for the worldvolume gauge theory we have also fixed the same gauge. In fact these are
two independent gauge symmetries and the U (1) gauge theory should be fixed separately.
Looking for BPS configurations, we are forced to turn on the scalar fields as well. This is a
direct generalization of Callan–Maldacena [18] and Gibbons’ BIons [19] argument to the
“compact” branes. These configurations, from the point of view of an observer far away
in the bulk, have the interpretation of (fundamental) strings piercing the S 3 . The points
of attachment (the north and south pole of the three-sphere), carry positive and negative
electric charge. The open strings, locally, can be thought of as Polchinski’s open strings
ending on the brane (giant graviton in our case) with Dirichlet boundary conditions [27].
As evidence that this configuration is really a string we show that the energy is proportional
to the distance to the brane (at least for far distances).
4.1. Solutions
We would now like to consider placing charges on the giant graviton. We take for the
embedding coordinates of a unit three-sphere
x1 = sin ψ sin θ cos φ,

x2 = sin ψ sin θ sin φ,
x3 = sin ψ cos θ,
x4 = cos ψ, (4.1)
with 0 ψ, θ π and 0 φ 2π . The metric on the three-sphere in the coordinate
system (4.1) is
ds 2 = dψ 2 + sin2 ψ dΩ22 , (4.2)

√
with dΩ22 = (dθ 2 + sin2 θ dφ 2 ) the metric on the unit two-sphere, and det g =
√
sin2 ψ sin θ . The Laplacian acting on a scalar field Φ is ∇ 2 Φ = √det
1
∂ ( det g g µν ∂ν Φ),
g µ
which on the three-sphere for Φ’s with only ψ dependence becomes
1
∇ 2 Φ(ψ) = 2
∂ψ sin2 ψ∂ψ Φ(ψ) . (4.3)
sin ψ
The three-sphere is compact, and hence the giant graviton cannot support single charges.
It does, however, support dipoles (as well as higher poles) with vanishing total charge.
Consider a dipole with two opposite charges placed at the two poles (ψ = 0, π ). The charge
density of such a configuration, as seen by the gauge theory, is
Q
ρ= 2
δ(ψ) − δ(ψ − π) δ(cos θ )δ(φ). (4.4)
sin ψ
A few words about the normalizations of the fields are in order: the scalar and gauge
fields appearing in the Hamiltonian (2.24) and (3.7) are not canonically normalized. The
normalization of the gauge field in (3.7) is such that the gauge theory action carries an
√
YM = µ gs , a convenient choice for studying gauge
overall factor of 1/gYM2 , where g
theories on curved backgrounds. Canonical normalization of the gauge field can be

√
achieved by taking Aµ → µ gs Aµ . The coupling of the gauge field to the charges in the
gauge theory is of the form J · A, with J0 the charge density ρ, which carries the same units
as the charge Q since the angular coordinates are dimensionless. The scalar fields in (2.24),
as stated in (2.39), can be normalized canonically by taking X → Rgeff Φ. This is the
normalization in which the scalar field couples to the sources in the same way as the gauge
field, via (4.4). Noting that the energy is measured in units of µ (a choice of scale for the
time coordinate), we find that the charges, as seen from the DBI action (2.2), are measured
√ √
in units of µ gs , so q = Qµ gs , with Q dimensionless. It is the canonical fields (scalar
and gauge) that are sourced by Q. The fact that the gauge and scalar fields enter the
Hamiltonian with different scales is an artifact of the choice of relative normalizations
of the fields in the DBI action (2.2). In fact, the normalizations in the DBI action are
chosen to reproduce the correct field normalizations in the gauge theory for the case of a
flat background, but in a curved background do not reproduce the standard normalizations
for a gauge theory coupled to a fixed curved metric, as is the case for the SYM theory on
R × S 3 . We also remind the reader that α µp+ gs is a dimensionless quantity, so in our
units, where we have set α = 1/2π, µp+ and geff are both dimensionless. From this point
on we shall deal only with canonically normalized fields.
We study the electrostatic problem for the gauge field in a gauge where A0 = Λ. The
equation of motion for the gauge field, arising from the quadratic Hamiltonian (3.7), is
simply Poisson’s equation, which in appropriate units requires ∇ 2 Λ = −ρ. We take as our
ansatz a field Λ(ψ) which is a constant along the two angular directions θ, φ. The solution
is
Λ(ψ) = Q cot ψ. (4.5)

As in the discussion of [18], exciting the gauge field alone would not result in a BPS
configuration (supersymmetry implies a relation between the profile of the gauge field and
the other fields in the N = 4 supermultiplet). The solution can be made BPS by turning on
a non-trivial profile for the scalars, keeping a constant vanishing background fermion field.
To find the scalar profile, it will prove useful to consider the quadratic part of the
Hamiltonian, expanded around the spherical vacuum, restricted to field configurations
which depend only on the radial direction. The potential for the scalar field describing
the radial profile can then be written
1 2
VΦ = Φ + (∇S 3 Φ)2 , (4.6)
2
where the square is with respect to the metric on the three-sphere of unit radius. The spikes
cannot go off to infinity in just any direction because of the potential from the plane-wave,
but they can extend off to infinity along X− , which can be taken as the radial direction.
The equation of motion for the field Φ in the gauge theory, in the presence of a dipole
charge configuration is simply
Q
∇S23 Φ − Φ = 2
δ(ψ) + δ(ψ − π) δ(cos θ )δ(φ). (4.7)
sin ψ
The right-hand side represents the sourcing of the scalar field Φ by the charges, as required
by supersymmetry. The fact that the two sources contribute with the same sign can be seen
(as we will show below) from requiring that the solutions of these equations, together with
the electric field, form a BPS configuration.
Fig. 1. Two dipole configurations. The left one has an interpretation of two fundamental strings extending off the
giant graviton to infinity. The configuration on the right is unstable and has no such string interpretation. The sign
of the charges are indicated and the arrows denote the direction of flux of the electric field.
The equation of motion for the scalar field (4.7) is solved by taking (see the Appendix A
for more details)
Q
Φ(ψ) = . (4.8)
sin ψ
Measured with respect to the origin in spherical coordinates, the profile of the spike is
given by R ± Φ, with R the radius of the giant graviton (see Fig. 1). We would like to note
that in our case the profile of Φ and the Coulomb potential Λ, (4.8) and (4.5) are different;
this should be contrasted with the usual BIon case [18]. Near the north pole (with ψ ≈ 0),
the solution for the scalar field and gauge field are similar, while at the south pole (ψ ≈ π ),
they differ by a sign, and for ψ away from the poles, each solution interpolates smoothly
between the solutions in the two regions. The scalar field is blind to the sign of the charges
which source it, while the gauge field is not.
The size of the throat, as seen by the DBI action, is Rgeff , the coupling for the
canonically normalized fields. Explicitly, the solution for the radial direction X is

geff
X=R 1±Q . (4.9)
sin ψ
Interestingly, the corrections to the shape of the strings arising from non-linearities are
suppressed relative to the tree-level shape by the same scale that sets the throat size,
i.e., geff . This basically means that the perturbative expansion of the light-cone Hamiltonian
(up to quadratic order) is a good one as long as the size of the throat is much smaller than
the giant graviton itself. The non-linearities and interaction terms can in fact modify the
form of the potential around a given vacuum such that one solution, explicitly the solution
corresponding to the choice of minus sign in (4.9), is destabilized, as happens to the dipole
Fig. 2. A quadrapole configuration of charges and the associated fundamental strings on the giant graviton.
solution (diagram on right of Fig. 1), where the strings enter the interior and meet. The
energy of this arrangement can be lowered by moving the end-points of the strings closer
to each other, and the charges at the ends would eventually annihilate, leaving behind
a giant graviton with no strings attached. For the dipole where the strings extend off to
infinity, i.e., the solution with plus sign in (4.9), the layout of the strings at opposite ends
(the diagram on the left of Fig. 1) is in fact a minimum, and remains so even when the
interactions are included, with the interactions only modifying the profile of the string
at the junction. Higher pole solutions can be analogously constructed (see Fig. 2 for a
quadrapole, and Fig. 3 for a more general “hedge-hog” configuration). When the coupling
is of order the separation of the strings attached to the giant graviton, the end-points can
meet and the string can separate from the giant graviton. From the gauge theory point of
view, this corresponds to the charges annihilating each other. For finite coupling (and hence
finite throat size), there will be a maximum number of strings which can attach to the giant
graviton, and hence a maximum pole configuration in the gauge theory. The smallest size
which can be effectively probed by the open strings is set by geff , and this leads to fuzziness
of the giant gravitons in view of an open string probe.
One may also consider configurations which are sourced by magnetic dipoles (and
higher poles). Such configurations correspond to S-dual solutions to the ones considered
above, where the strings ending on the giant graviton are D-strings. Dyonic configurations
with both electric and magnetic sources can also be envisioned. More general configura-
tions, where several giant gravitons are coincident, can also be constructed, and by analogy
to the general case of coincident D-branes, would give rise a non-Abelian gauge theory on
their worldvolume.
4.2. Energy
We would now like to give an interpretation to the spikes we found as solutions of

the quadratic Hamiltonian in Section 4.1. To do so we consider the energy of such a
configuration. To find the energy density, we use the solutions (4.5) and (4.8) for the gauge
Fig. 3. A generic giant graviton with multiple spikes, suggesting the hedge-hog title.
field configuration and radial profile of the giant graviton in the Hamiltonian density, then
integrate the density to find the total energy. For the dipole, there are two solutions, one
for which the spike extends off to infinity away from the giant graviton, and one where
the spikes enter the interior and join (see Fig. 1). We consider both configurations. The
resultant energy E, in units of µ, for the first configuration is
E = 4πQ2 cot , (4.10)

and the scalar and gauge fields contribute equally to the energy. We have integrated along
the ψ direction from to π/2. The serves as a cut-off, since the total energy would
diverge; we are interested in the scaling of this energy with length as the cut-off is removed.
The π/2 captures one string (the other string would give an equal contribution). For
small , the result scales as
E ∼ Φ() (4.11)
up to some fixed numerical coefficients. In other words, the energy per unit length is the
same as the tension of the fundamental string.8 We expect also that the spectrum of small
fluctuations for the string should reproduce the spectrum of massive modes of the open
fundamental string in the plane-wave background [9].
The profile where the “strings” enter into the interior of the giant graviton is given by
R − Φ, with Φ the same as for the outgoing strings, but the range of ψ is now restricted
such that R − Φ is limited to only one “hemisphere” inside the giant graviton. At the lower
cut-off for ψ, the string joins onto the other string originating at the other charge. In other
words, sin−1 (geff ) ψ π − sin−1 (geff ) and only when the size of the throat is very small
8 In our units, the string tension T ∼ 1.

Fig. 4. Two giant gravitons of different radii, connected by “strings”. In the limit that the radii are equal and the
giant gravitons become coincident, the gauge symmetry of the worldvolume theory is enhanced.
compared to the size of the giant graviton (i.e., when geff 1) this spike can be interpreted
as a fundamental string. For any finite value of geff , the profile never reaches that of a
fundamental string.
In any case, as we have already discussed, this configuration is unstable (metastable).
The dipole where the strings run off to infinity is stabilized by the fact that the endpoints
of the strings have their boundary conditions fixed at infinity, and any small perturbation
of their junction increases the length of the string and hence the energy.
Given a total light-cone momentum p+ , one may distribute it among some number
of giant gravitons, that is a configuration of concentric giant gravitons. One may wonder
whether in the limit when two of these giant gravitons become coincident, analogously to
the case of D-branes, one should expect enhancement of the U (1) gauge symmetry to U (2).
That is possible if the metastable spikes (strings) depicted in Fig. 4 become massless in the
coincident limit, a fact which is confirmed by our energy analysis. The life-time of these
spikes depends on geff as well as the difference in the radii of the two giant gravitons.
√
Based on energy arguments we expect it to be proportional to the inverse of gs as well
as the difference of inverse radii squared of the two giant gravitons.
4.3. Supersymmetry
The spin connection one-forms on the three-sphere can be deduced from the metric (4.2)
Ω 12 = − cos ψ dθ, Ω 23 = − cos θ dφ, Ω 31 = cos ψ sin θ dφ, (4.12)
and the supersymmetry variation of the gaugino for the Abelian theory is [28,29]

1 1
δ λ = Fµν Γ µν − ∂µ Φ m Γ m Γ µ − Φm Γ m Γ µ ∇µ . (4.13)
2 2
The derivatives on the fields are not gauge covariant since these fields transform in the
adjoint of U (1), and hence are neutral. Also, ∇µ = ∂µ + 14 Ωµab Γab , is the covariant
derivative with the spin connection Ωµab .9 Solutions in a given background, for which
this variation vanishes, are the Killing spinors, and the number of such solutions gives
the amount of supersymmetry preserved by the background. For the solutions we are
considering, the deformation of the sphere is independent of the angular directions along
an S 2 of the S 3 , and preserves an SU(2) symmetry of the SO(4) invariant vacuum.
We make use of the solution for the electric field in terms of the scalar potential, where
Q
F0ψ = −∂ψ Λ(ψ) = , (4.14)
sin2 ψ
and Φ m Γ m = Φ(ψ)Γ r with r designating the radial direction in the transverse directions,
and with Φ = Q/sin ψ . The Killing spinor equation for this background is [26]
1
∇µ = Γ r Γµ , (4.15)
2
and the square of the Killing spinors are the Killing vectors. The Killing spinor
equation (4.15) has the maximal number of solutions [28]. For these Killing spinors, the
condition for supersymmetry (4.13) reduces to10
Q
2
Γ 0 + Γ˜ r = 0, (4.16)
sin ψ
with = Γ ψ , and
Γ˜ r = cos ψΓ r + sin ψΓ ψ
is a rotated Dirac matrix of unit norm, i.e., (Γ˜ r )2 = 1. This implies that half the
supersymmetries of the background plane-wave remain unbroken by the presence of the
D-brane, and the dipole configuration of the giant graviton state with the two spikes
piercing it is 1/2 BPS, i.e., it preserves eight supercharges.
The Γ˜ r matrix at ψ = 0 is Γ r and hence the supersymmetry condition (4.16) is
essentially that of the usual BIon [18] with, say a positive charge. At ψ = π , however,
Γ˜ r = −Γ r reducing (4.16) to a usual BIon with negative charge and the term proportional
to Φ in (4.13) makes it possible to have a smooth supersymmetric transition form a positive
charge to a negative charge.
There also exists a solution of the quadratic Hamiltonian for which Φ = −Q/sin ψ ,
but with the gauge field configuration unchanged. The condition for the existence of
supersymmetry for this configuration is
Q
2
Γ 0 − Γ˜ r = 0, (4.17)
sin ψ
which differs from (4.16) by the relative sign between the time-like and radial Dirac
matrices, but preserves precisely the same amount of supersymmetry as the original
configuration, and is also 1/2 BPS. One should note that being BPS does not necessarily
9 Note that the indices a, b are with respect to the orthonormal tangent frame, while µ, ν are curved indices
on the worldvolume and m ranges over the six SO(6) components.
10 With our metric conventions, (Γ 0 )2 = −1, with the other Dirac matrices squaring to one.
guarantee the stability of the solution, particularly when the interaction terms in the light-
cone Hamiltonian are taken into account.11
5. Outlook and future directions
In this paper we have analyzed some aspects of the worldvolume theory of giant
gravitons on the plane-wave background. Working out the spectrum of small fluctuations
of the giant three-sphere, we argued that they fall into (short) multiplets of the PSU(2|2) ×
PSU(2|2) × U (1) algebra. One of the interesting features of the three-brane light-cone
Hamiltonian (2.14) is the natural appearance of Nambu brackets (2.16). In this point of
view “quantization” of Nambu bracket (2.16) would provide us with a natural quantization
of the theory living on the giant graviton. In the membrane case the corresponding Nambu
bracket is essentially a Poisson bracket and its quantization is possible by replacing the
bracket with matrix commutators [30]. This quantization of Possion brackets from the
membrane point of view can be regarded as discretization of the worldvolume, which also
leads to the “non-commutative” (non-Abelian) structure of the BFSS matrix model. In the
same trend quantization of the three-sphere giant graviton theory may provide us with
an answer to the puzzle of finding a holographic description of type IIB string theory on
the plane-wave background, which supposedly is a matrix theory [31] (for a summary of
discussions on the matter see Section 9 of [16]).
As another aspect of the gauge theory living on giant gravitons, we studied static
configurations which source the gauge fields and also the scalar fields. The basic building
blocks of such objects are dipole configurations with the largest possible dipole moment
being proportional to the size of the giant graviton. We argued that the BPS dipole
configurations, from the bulk viewpoint, can be understood as open strings ending on
the giant graviton. These are open strings with their two ends on the north and south
pole of the three-sphere. It is evident from our construction that these open strings
satisfy Dirichlet boundary conditions in the directions transverse to the brane, a natural
expectation generalizing Polchinski’s D-brane picture [27]. We also argued that it is
possible to have dipole configurations with the spike going inside the three-sphere. These
states, although being metastable, are responsible for enhancing the gauge symmetry when
two concentric giant gravitons become coincident.
As we discussed, since at finite geff the size of the throat of the spikes is finite, one would
physically expect to have an upper limit on the highest multiple moment. In other words
there is a minimum area which can be probed using these open strings and also there is a
minimum size dipole moment. This suggests that the fuzzy three sphere [32] is the right
description of the quantized giant graviton [31]. A description of multiple coincident giant
gravitons in terms of a non-commutative three-sphere defined as a Hopf fibration over a
fuzzy two-sphere is given in [33].
11 In principle the statement that BPS configurations are protected should be taken with a grain of salt. It is
possible that some multiplets which are BPS at a given value of coupling combine into a long (ordinary non-BPS)
multiplet and receive corrections. For explicit examples and more detailed discussion on this point see [22].
As a direct generalization of our giant hedge-hog configurations one can consider

circular D-strings in the AdS3 × S 3 background (or the corresponding Penrose limit [34]).
In that case, however, we expect that similar to the flat D-string case [35], the spike touches
the giant circle to form a “string junction”. This leads to a pair of three string junctions, two
of the legs of each are connected and make a deformed half circle. This construction can
then be generalized to junctions of (p, q) strings and string networks [36] in plane-wave
backgrounds [37].
Acknowledgements
We would like to thank Keshav Dasgupta, Michal Fabinger, Sergey Prokushkin, and
especially Leonard Susskind for fruitful discussions. The work of M.M.Sh.-J. is supported
in part by NSF grant PHY-9870115 and in part by funds from the Stanford Institute for
Theoretical Physics. The work of D.S. is supported by the Department of Energy, Contract
DE-AC03-76SF00515.
Appendix A. SO(4) harmonics in terms of usual Ylm ’s
The Laplacian on the three-sphere in the coordinate system we have adopted is

1 1
∇S23 = 2
∂ψ sin2 ψ∂ψ + ∇S22 , (A.1)
sin ψ sin2 ψ
where ∇S22 = sin1 θ ∂θ (sin θ ∂θ ) + sin1 θ ∂φ2 . One may use (A.1) to write SO(4) harmonics in
terms of the SO(3) Ylm ’s. Explicitly, let us consider the (source free) equation of motion
for the Coulomb potential Λ:
∇S23 Λ(ψ, θ, φ) = 0. (A.2)

Separating variables as Λ(ψ, θ, φ) = Λl (ψ)Ylm (θ, φ), (A.2) can be cast in the form
1 1
2
∂ψ sin2 ψΛl − l(l + 1)Λl = 0. (A.3)
sin ψ sin2 ψ
After the change of variable u = cot ψ, (A.3) takes the form

1 + u2 Λl − l(l + 1)Λl = 0, (A.4)
where Λ = du d
Λ. For l = 0, (A.4) is simply solved by Λ0 = u = cot ψ (the solution we
have already discussed as the dipole (4.5)), and for l = 1, Λ1 = 1 + u2 = 1/ sin2 ψ. For
general l, (A.4) can be solved using a series expansion for Λl (u)
l+1
Λl (u) = a k uk ,
k=0
where al+1 = 1, al = 0 and

(k + 1)(k + 2)
ak = ak+2 , 0 k l − 1.
l(l + 1) − k(k − 1)
Similarly, solutions to the equation for the scalar field, namely ∇S23 Φ − Φ = 0 can be
decomposed as
Φ = Φl (ψ)Ylm (θ, φ).

Taking v = 1/ sin ψ and dvd
Φ = Φ , then

v 2 v 2 − 1 Φl + vΦl − l(l + 1)v 2 + 1 Φl = 0. (A.5)
For the l = 0 case, as we discussed in (4.8), the solution is Φl=0 = v, and for general l, as
in the previous case, (A.5) may be solved using Taylor expansion techniques, inserting

l+1
Φl (v) = bk v k
k=0
into the equation. It turns out that (A.5) has only solutions for even l with b0 = 0, bl+1 = 1,
and
(k + 1)2
bk = − bk+2 , 1 k l − 1.
l(l + 1) − k(k − 1)
The fact that (A.5) has (polynomial) solutions only for even l can physically be understood
by noting that the source term for the scalars is a sum of delta-functions (whereas that of
the Coulomb potential is an alternating sum, so that the total net charge is zero).
Finally, we would like to mention that in our expansions the 2l+1 -poles of SO(4) are
related to Ylm (i.e., 2l -pole of SO(3)). For example, our “dipole” configurations correspond
to the l = 0 case. Also note that the dipole configuration can be thought of as a Dirac string
on the sphere where ψ = 0 corresponds to the monopole and ψ = π corresponds to the
end of the Dirac string tail, which in the flat space language is at infinity.
References
[1] J. McGreevy, L. Susskind, N. Toumbas, Invasion of the giant gravitons from anti-de Sitter space, JHEP 0006
(2000) 008, hep-th/0003075.
[2] M.T. Grisaru, R.C. Myers, O. Tafjord, SUSY and Goliath, JHEP 0008 (2000) 040, hep-th/0008015.
[3] A. Hashimoto, S. Hirano, N. Itzhaki, Large branes in AdS and their field theory dual, JHEP 0008 (2000)
051, hep-th/0008016.
[4] S.R. Das, A. Jevicki, S.D. Mathur, Vibration modes of giant gravitons, Phys. Rev. D 63 (2001) 024013,
hep-th/0009019.
[5] R.C. Myers, O. Tafjord, Superstars and giant gravitons, JHEP 0111 (2001) 009, hep-th/0109127.
[6] F. Leblond, R.C. Myers, D.C. Page, JHEP 0201 (2002) 026, hep-th/0111178.
[7] V. Balasubramanian, M. Berkooz, A. Naqvi, M.J. Strassler, Giant gravitons in conformal field theory,
JHEP 0204 (2002) 034, hep-th/0107119.
[8] S. Corley, A. Jevicki, S. Ramgoolam, Exact correlators of giant gravitons from dual N = 4 SYM theory,
Adv. Theor. Math. Phys. 5 (2002) 809, hep-th/0111222.
[9] V. Balasubramanian, M.X. Huang, T.S. Levi, A. Naqvi, Open strings from N = 4 super-Yang–Mills,
JHEP 0208 (2002) 037, hep-th/0204196.
[10] P. Ouyang, Semiclassical quantization of giant gravitons, hep-th/0212228.
[11] D. Berenstein, Shape and holography: studies of dual operators to giant gravitons, Nucl. Phys. B 675 (2003)
179, hep-th/0306090.
[12] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, A new maximally supersymmetric background of
IIB superstring theory, JHEP 0201 (2002) 047, hep-th/0110242;
M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry,
Class. Quantum Grav. 19 (2002) L87, hep-th/0201081;
M. Blau, J. Figueroa-O’Farrill, G. Papadopoulos, Penrose limits, supergravity and brane dynamics, Class.
Quantum Grav. 19 (2002) 4753, hep-th/0202111.
[13] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp-waves from N = 4 super-Yang–Mills,
JHEP 0204 (2002) 013, hep-th/0202021.
[14] R.R. Metsaev, Type IIB Green–Schwarz superstring in plane wave Ramond–Ramond background, Nucl.
Phys. B 625 (2002) 70, hep-th/0112044.
[15] R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in plane wave Ramond–Ramond
background, Phys. Rev. D 65 (2002) 126004, hep-th/0202109.
[16] D. Sadri, M.M. Sheikh-Jabbari, The plane-wave/super-Yang–Mills duality, hep-th/0310119.
[17] H. Takayanagi, T. Takayanagi, Notes on giant gravitons on pp-waves, JHEP 0212 (2002) 018, hep-
th/0209160.
[18] C.G. Callan, J.M. Maldacena, Brane dynamics from the Born–Infeld action, Nucl. Phys. B 513 (1998) 198,
hep-th/9708147.
[19] G.W. Gibbons, Born–Infeld particles and Dirichlet p-branes, Nucl. Phys. B 514 (1998) 603, hep-th/9709027.
[20] J. Polchinski, String Theory, Vol. II: Superstring Theory and Beyond, Cambridge Univ. Press, Cambridge,
1998.
[21] J. Maldacena, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Transverse fivebranes in matrix theory, JHEP 0301
(2003) 038, hep-th/0211139.
[22] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Protected multiplets of M-theory on a plane wave,
JHEP 0209 (2002) 021, hep-th/0207050.
[23] R. Kallosh, J. Rahmfeld, A. Rajaraman, Near horizon superspace, JHEP 9809 (1998) 002, hep-th/9805217.
[24] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Matrix perturbation theory for M-theory on a P P -
wave, JHEP 0205 (2002) 056, hep-th/0205185.
[25] R.R. Metsaev, Supersymmetric D3 brane and N = 4 SYM actions in plane wave backgrounds, Nucl. Phys.
B 655 (2003) 3, hep-th/0211178.
[26] M. Blau, Killing spinors and SYM on curved spaces, JHEP 0011 (2000) 023, hep-th/0005098.
[27] J. Polchinski, Dirichlet-branes and Ramond–Ramond charges, Phys. Rev. Lett. 75 (1995) 4724, hep-
th/9510017.
[28] K. Okuyama, N = 4 SYM on R × S 3 and pp-wave, JHEP 0211 (2002) 043, hep-th/0207067.
[29] H. Nicolai, E. Sezgin, Y. Tanii, Conformally invariant supersymmetric field theories on S P × S 1 and super
P -branes, Nucl. Phys. B 305 (1988) 483.
[30] W. Taylor, M(atrix) theory: matrix quantum mechanics as a fundamental theory, Rev. Mod. Phys. 73 (2001)
419, hep-th/0101126.
[31] M.M. Sheikh-Jabbari, Plane-wave matrix string theory, in preparation.
[32] S. Ramgoolam, On spherical harmonics for fuzzy spheres in diverse dimensions, Nucl. Phys. B 610 (2001)
461, hep-th/0105006.
[33] B. Janssen, Y. Lozano, D. Rodriguez-Gomez, Nucl. Phys. B 669 (2003) 363, hep-th/0303183.
[34] J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NS–NS and R–R plane wave
backgrounds, JHEP 0204 (2002) 021, hep-th/0202179.
[35] K. Dasgupta, S. Mukhi, BPS nature of 3-string junctions, Phys. Lett. B 423 (1998) 261, hep-th/9711094.
[36] A. Sen, String network, JHEP 9803 (1998) 005, hep-th/9711130.
[37] D. Sadri, Giant string junctions, in preparation.
Nuclear Physics B 687 [FS] (2004) 189–219
Semiclassical particle spectrum of double

sine-Gordon model
G. Mussardo a,b , V. Riva a,b,∗ , G. Sotkov c
a International School for Advanced Studies, Via Beirut 1, 34100 Trieste, Italy
b Istituto Nazionale di Fisica Nucleare, Sezione di Trieste, Italy
c Instituto de Física Teorica, Universidade Estadual Paulista, Rua Pamplona 145, 01405-900 Sao Paulo, Brazil
Received 27 February 2004; accepted 1 April 2004
Abstract
We present new theoretical results on the spectrum of the quantum field theory of the double
sine-Gordon model. This non-integrable model displays different varieties of kink excitations and
bound states thereof. Their mass can be obtained by using a semiclassical expression of the matrix
elements of the local fields. In certain regions of the coupling-constants space the semiclassical
method provides a picture which is complementary to the one of the form factor perturbation theory,
since the two techniques give information about the mass of different types of excitations. In other
regions the two methods are comparable, since they describe the same kind of particles. Furthermore,
the semiclassical picture is particularly suited to describe the phenomenon of false vacuum decay,
and it also accounts in a natural way the presence of resonance states and the occurrence of a phase
transition.
PACS: 11.10.Kk; 11.15.Kc; 11.27.+d
Keywords: Non-integrable quantum field theories; Form factors; Bound states
* Corresponding author. Postal address: Valentina Riva, International School for Advanced Studies, via
Beirut 4, 34014 Trieste, Italy. Tel.: +39-040-3787516; fax: +39-040-3787528.
E-mail addresses: mussardo@sissa.it (G. Mussardo), riva@sissa.it (V. Riva), sotkov@ift.unesp.br
(G. Sotkov).
doi:10.1016/j.nuclphysb.2004.04.003
190 G. Mussardo et al. / Nuclear Physics B 687 [FS] (2004) 189–219
1. Introduction
As a natural development of the studies on integrable quantum field theories, there has
been recently an increasing interest in studying the properties of non-integrable quantum
field theories in 1 + 1 dimensions, both for theoretical reasons and their application
to several condensed-matter or statistical systems. However, contrary to the integrable
models, many features of these quantum field theories are still poorly understood: in most
of the cases, in fact, their analysis is only qualitative and even some of their basic data,
such as the mass spectrum, are often not easily available. Although one could always rely
on numerical methods to shed some light on their properties, it would be obviously useful
to develop some theoretical tools to control them analytically. In this respect, there has
been recently some progress, thanks to two different approaches.
The first approach, called the form factor perturbation theory (FFPT) [1,2], is best
suited to deal with those non-integrable theories close to the integrable ones. It permits, in
particular, to obtain quantitative predictions on their mass spectrum, scattering amplitudes
and other physical quantities. As any other perturbation scheme, it works finely as far as
the non-integrable theory is an adiabatic deformation of the original integrable model,
i.e., when the two theories are isospectral. This happens when the field which breaks
the integrability is local with respect to the operator which creates the particles. If, on
the contrary, the field which moves the theory away from integrability is non-local with
respect to the particles, the resulting non-integrable model generally displays confinement
phenomena and, in this case, some caution has to be taken in interpreting these perturbative
results.
The second approach, known as semiclassical method and based on the seminal work
of Dashen, Hasslacher and Neveu [3], is on the other hand best suited to deal with those
quantum field theories (integrable or not) having kink excitations of large mass in their
semiclassical limit. Under these circumstances, in fact, once the non-perturbative classical
solutions are known, it is relatively simple to determine the two-particle form factors
on the kink states of the basic fields of the theory and to extract the spectrum of the
excitations from their pole structure [4–6]. Although this method is restricted to work in
a semiclassical regime, it permits however to analyze non-integrable theories in the whole
coupling-constants space, even far from the integrable points.
An interesting non-integrable model where both approaches can be used is the so-called
double sine-Gordon model (DSG). Its Lagrangian density is given by
1
L = (∂µ ϕ)2 − V (ϕ), (1.1)
2
with

µ λ β
V (ϕ) = − 2 cos βϕ − 2 cos ϕ + δ + C, (1.2)
β β 2
where C is a constant that has be chosen such that to have a vanishing potential energy of
the vacuum state. The classical dynamics of this model has been extensively studied in the
past by means of both analytical and numerical techniques (see [7] for a complete list of
the results), while its thermodynamics has been studied in [8] by using the transfer integral
method [9], and with the path integral technique (see for instance [10]).
G. Mussardo et al. / Nuclear Physics B 687 [FS] (2004) 189–219 191
With λ or µ equal to zero, the DSG reduces to the ordinary integrable sine-Gordon (SG)
model with frequency β or β/2 respectively. Hence the DSG model with a small value of
one of the couplings can be regarded as a deformation of the corresponding SG model and
studied, therefore, by means of the FFPT [2]. On the other hand, for β → 0, irrespectively
of the value of the coupling constants λ and µ, the DSG model reduces to its semiclassical
limit. Despite the non-integrable nature of the DSG model, its classical kink solutions are—
remarkably enough—explicitly known [7,8] and therefore the semiclassical method can be
successfully applied to recover the (semi-classical) spectrum of the theory. As we will see
in the following, the two approaches turn out to be complementary in certain regions of
the coupling constants, i.e., both are needed in order to get the whole mass spectrum of the
theory, whereas in other regions they provide the same picture about the spectrum of the
excitations.
Apart from the theoretical interest in testing the efficiency of the two methods on this
specific model where both are applicable, the study of the DSG is particularly important
since this model plays a relevant role in several physical contexts, either as a classical non-
linear system or as a quantum field theory. At the classical level, its non-linear equation
of motion can be used in fact to study ultra-short optical pulses in resonant degenerate
medium or texture dynamics in He3 (see, for instance, [11] and references therein).
As a quantum field theory, depending on the values of the parameters λ, µ, β, δ in its
Lagrangian, it displays a variety of physical effects, such as the decay of a false vacuum
or the occurrence of a phase transition, the confinement of the kinks or the presence of
resonances due to unstable bound states of excited kink–antikink states. Moreover, it finds
interesting applications in the study of several systems, such as the massive Schwinger field
theory or the Ashkin–Teller model [2], as well as in the analysis of the O(3) non-linear
sigma model with θ term [12], i.e., the quantum field theory relevant for understanding the
dynamics of quantum spin chains [13,14]. The DSG model also matters in the investigation
of other interesting condensed matter phenomena, such as the soliton confinement of spin-
Peierls antiferromagnets [15], the dynamics of the spin chains in a staggered external field
or the electron interaction in a staggered potential [16].
Motivated by the above combined theoretical and physical interests, a thorough study
of the spectrum of the DSG model seems therefore to be particularly interesting and in this
paper we present the results of such analysis.
The paper is organized as follows: in Section 2 we briefly recall the basic formulas of
the form factor perturbation theory whereas in Section 3 we remind the basic results of the
semiclassical method. Section 4 is devoted to the semiclassical analysis of the spectrum
of the DSG model and its comparison with the results coming from FFPT. Section 5
deals with the analysis of false vacuum decay. In Section 6 we discuss the occurrence
of resonance phenomena in the DSG in relation with analogous effects observed in the
classical scattering of kink states. Our conclusions are in Section 7. The paper also contains
several appendices. In Appendix A we compute the kink mass corrections by using the
FFPT, in Appendix B we collect the relevant expressions of the semiclassical Form Factors,
Appendix C is devoted to the analysis of neutral states in comparison with the sine-Gordon
model, and in Appendix D we discuss the basic results in a closely related model, i.e., the
double sinh-Gordon model.
2. Form factor perturbation theory
The method of the form factor perturbation theory (FFPT) [1,2] permits to analyze a
non-integrable quantum field theory when its action A is represented by a deformation of
an integrable one A0 through a given operator Ψ :

A = A0 + g d 2 x Ψ (x). (2.1)
One of the first consequences of moving away from integrability is a change in the
spectrum of the theory: the first order corrections to the mass of the particle a belonging to
the spectrum of the unperturbed theory is in fact given by

δm2a = 2gFaΨā (iπ) + O g 2 , (2.2)
where the particle–antiparticle form factor of the operator Ψ (x), defined by the matrix
element1

FaΨā (θ1 − θ2 ) = 0Ψ (0)a(θ1 )ā(θ2 ) , (2.3)
is introduced. The mass correction (2.2) may be finite or divergent, depending on the
locality properties of the operator Ψ (x) with respect to the particle a. The situation was
clarified in [2] and it is worth recalling the main conclusion of that analysis.
In integrable theories, the form factors of a generic scalar operator O(x) can be
determined due to the simple form assumed by the Watson equation [17,18] and for the
two-particle case, one has
FaOā (θ ) = Sabāb̄ (θ )Fb̄b
O
(−θ ), (2.4)
FaOā (θ + 2iπ) = e −2iπγO,a O
Fāa (−θ ), (2.5)
where θ = θ1 − θ2 . In the first equation, expressing the discontinuity of the matrix element
across the unitarity cut, Sabāb̄ (θ ) is the elastic two-body scattering amplitude. In the second
equation, expressing the crossing symmetry of the form factor, the explicit phase factor
e−2iπγO,a is inserted to take into account a possible semi-locality of the operator which
interpolates the particle a (i.e., any operator ϕa such that 0|ϕa |a = 0) with respect to
the operator O(x).2 When γO,a = 0, there is no crossing symmetric counterpart to the
unitarity cut but when γO,a = 0, there is instead a non-locality discontinuity in the plane
of the Mandelstam variable s, with s = 0 as branch point.3 In the rapidity parameterization
there is however no cut because the different Riemann sheets of the s-plane are mapped
onto different sections of the θ -plane; the branch point s = 0 is mapped onto the points
θ = ±iπ which become therefore the locations of simple annihilation poles. The residues
at these poles are given by [18] (see also [19])

−i Res FaOā (θ ) = 1 − e∓2iπγO,a 0|O|0. (2.6)
θ=±iπ
1 We adopt the standard parameterization of the on-shell two-dimensional momenta given in terms of the
rapidity, p (0) = m cosh θ , p (1) = m sinh θ .
O ,ā = −γO ,a .
2 Consistency of Eq. (2.5) requires γ
3 The Mandelstam variable s is expressed by s = (p + p )2 = 4m2 cosh2 (θ/2).
a ā a
In a sine-Gordon model with frequency β, an exponential operator Ψα = eiαϕ has a semi-

locality index with respect to the soliton s of the theory given by γα,s = α/β whereas it
has a vanishing semi-locality index with respect to the breather particles [2]. This implies
that, taking the sine-Gordon action as the integrable A0 and Ψα as the perturbing operator,
the formula (2.2) can be safely applied to compute the first order correction to the mass of
the breathers, whereas a divergence may appear in an analogous computation of the mass
correction of the solitons. This divergence has to be seen as the mathematical signal that
the solitons of the original integrable model no longer survive as asymptotic particles of
the perturbed theory, i.e., they are confined.
3. Semiclassical method
The semiclassical quantization of a field theory defined by a potential V (φ) consists in

identifying a classical background φcl (x), which satisfies the equation of motion
∂µ ∂ µ φcl + V (φcl ) = 0, (3.1)
and in applying to it various well established techniques, like the path integral formalism
[20] or the solution of the field equations in classical background [3], usually called the
DHN method (for a systematic review, see [21]).
The procedure is particularly simple and interesting if one considers classical field
solutions φcl (x) in 1 + 1 dimensions which are static “kink” configurations interpolating
between degenerate minima of the potential, and whose quantization gives rise to a particle-
like spectrum.
A remarkable result, due to Goldstone and Jackiw [4] (see also [5] for a non-relativistic
context), is that the classical background φcl (x) has the quantum meaning of Fourier
transform of the form factor of the basic field φ(x) between kink states. The technique to
derive this result relies on the Heisenberg equation of motion satisfied by the quantum field
φ(x) together with the basic hypothesis that the kink momentum is very small compared to
its mass.4 In [6], we have refined the original argument overcoming its serious drawback of
being formulated non-covariantly in terms of the kink space-momenta. This was possible
thanks to the use of the rapidity variable θ of the kink states (and considering it as very
small), instead of the momentum.
The final result is the expression of the semiclassical form factor between kink states as
the Fourier transform of the classical kink background, with respect to the Lorentz invariant
rapidity difference θ ≡ θ1 − θ2 :

p1 |φ(0)|p2 ≡ f (θ ) ≡ Mcl da eiMcl θa φcl (a), (3.2)
4 The mass of kink state is inversely proportional to the coupling constant, considered small in the
semiclassical regime, and therefore the kink is a heavy particle in this limit.
where Mcl is the classical energy of the kink.5 Having a covariant formulation, it is possible
to express the crossed channel form factor through the variable transformation θ → iπ − θ :
F2 (θ ) ≡ 0|φ(0)|p1 , p̄2 = f (iπ − θ ). (3.3)
The analysis of this quantity provides a direct information about the spectrum of the theory.
Its dynamical poles, in fact, located at θ ∗ = i(π − u) with 0 < u < π , coincide with the
poles of the kink–antikink S-matrix, and the relative bound states masses can be then
expressed as
u
m(b) = 2Mcl sin . (3.4)
2
It is worth stressing that this procedure for extracting the semiclassical bound states masses
is remarkably simpler than the standard DHN method of quantizing the corresponding
classical backgrounds, because in general these solutions depend also on time and have a
much more complicated structure than the kink ones. Moreover, in non-integrable theories
these backgrounds could even not exist as exact solutions of the field equations: this
happens for example in the φ 4 theory, where the DHN quantization has been performed on
some approximate backgrounds [3].
In order to compute the first quantum corrections to the masses, one has to quantize
semiclassically the theory around the classical solution by splitting the field as φ(x, t) =
φcl (x) + η(x)e−iωt and finding the eigenvalues ωi of the stability equation [3,21]
2

−∂x + V φcl (x) ηi (x) = ωi2 ηi (x). (3.5)
With these, the semiclassical energy levels are build as
1

E{ni } = Ecl + h̄ ni + ωi + O h̄2 , (3.6)
2
i
and, in particular, the particles masses are given by the ground state of these levels
h̄
E0 ≡ E{ni =0} = Ecl + ωi + O h̄2 . (3.7)
2
i
In the following, we will not include these corrections in our results, since the analytical
solution of the stability equation (3.5) in the case of the DSG model is still missing.
Nevertheless, these corrections are not necessary in the cases in exam, because we consider
kink particles, for which the classical energy is the term of leading order in the coupling,
and their bound states, for which expression (3.4) already encodes the first semiclassical
corrections (see [3]).
However, the eigenvalues ωi play an important role in the case of unstable particles,
since the fingerprint of instability is precisely the imaginary nature of some of these
frequencies. Hence, although we will obtain real values for the masses of all the considered
particles, we will always keep in mind that many of these masses receive imaginary
contributions coming from some of the ωi .
5 Along the same lines, it is possible to prove that the form factor of an operator expressible as a function of φ
is given by the Fourier transform of the same function of φcl . For instance, the form factor of the energy density
dφ
operator ε can be computed performing the Fourier transform of εcl (x) = 12 ( dxcl )2 + V [φcl ].
4. Semiclassical analysis of DSG particle spectrum
The double sine-Gordon model is defined by the potential

µ λ β
Vδ (ϕ) = − 2 cos βϕ − 2 cos ϕ + δ + C, (4.1)
β β 2
with the constant C chosen such that the vacuum state has a vanishing potential energy. We
will study this theory in a regime of small β, where the semiclassical results are expected
to give a valuable approximation of the spectrum.6 At the quantum level, the different
renormalization group trajectories originating from the Gaussian fixed point described
by the kinetic term 12 (∂µ ϕ)2 of the Lagrangian (1.1) are labeled by the dimensionless
scaling variable η = λµ−(8π−β /4)/(8π−β ) which simply reduces to the ratio η = µλ in
2 2
the semiclassical limit. When λ or µ are equal to zero, the DSG model coincides √with
√
an ordinary sine-Gordon model with coupling β or β/2, and mass scale µ or λ/4,
respectively.
Since for general values of the couplings the potential (4.1) presents a 4π
β -periodicity, it
was noticed in [22] that one has an adiabatic perturbation of an integrable model only if the
λ = 0 theory is regarded as a two-folded sine-Gordon model. This theory is a modification
of the standard sine-Gordon model, where the period of the field φ is defined to be 4π β ,
instead of 2π
β [23]. As a consequence of this new periodicity assignment, such a theory has
two different degenerate vacua |k, with k = 0, 1 and |k + 2 ≡ |k, which are defined by
k|φ|k = 2πβ k. Hence it has two different kinks, related to the classical backgrounds by
the formula
2kπ 4
cl
Kk,k+1 (x) = + arctan emx , k = 0, 1, (4.2)
β β
and two corresponding antikinks, related to the classical solutions by the expression
2kπ 4
cl
Kk+1,k (x) = + arctan e−mx , (4.3)
β β
2(k + 1)π 4
= − arctan emx , k = 0, 1. (4.4)
β β
Finally, in the spectrum there are also two sets of kink–antikink bound states bn(l) , with
l = 0, 1 and n = 1, . . . , [ 8π
ξ ].
The flow between the two limiting sine-Gordon models (with frequency β or β/2,
respectively) displays a variety of different qualitative features, including confinement and
phase transition phenomena, depending on the signs of λ and µ, and on the value of the
relative phase δ. However, it was observed in [2] that the only values of δ which lead to
6 By applying the stability conditions found in [2] to this model, they reduce to the condition β 2 < 8π .
Hence, for these values of β and, in particular in the semiclassical limit β → 0, the potential (4.1) is stable under
renormalization and no countertems have to be added.
inequivalent theories are those given by |δ| π2 . Furthermore, in virtue of the relations

π
Vδ φ + , λ, µ = Vδ+π/2 (φ, λ, −µ),
β
Vδ (−φ, λ, µ) = V−δ (φ, λ, µ), (4.5)
we can describe all the inequivalent possibilities keeping µ positive and the relative phase
in the range 0 δ π2 . The sign of the coupling λ, instead, simply corresponds to a
shift or a reflection of the potential, without changing its qualitative features. As we are
going to show in the following, the case δ = π2 displays peculiar features, while a common
description is possible for any other value of δ in the range 0 δ < π2 .
In closing this discussion on the general properties of the DSG model, we would like to
mention that the possibility of writing exact classical solutions for all the different kinds
of topological objects in this model finds a deep explanation in the relation between the
trigonometric potential (4.1) and power-like potentials. In fact, defining
nπ 4
ϕ= ± arctan Y, n = 0, 1, 2, 3, (4.6)
β β
one can easily see that the first order equation which determines the kink solution

1 dϕ 2 µ λ β
= − 2 cos βϕ − 2 cos ϕ+δ +C (4.7)
2 dx β β 2
is mapped into the equation for Y

1 dY 2
= U (Y ), (4.8)
2 dx
where U (Y ) describes various kinds of algebraic potentials, depending on the values of
n, δ and C. The δ = 0 case was analyzed in [24] and its classical solutions are very
simple because U (Y ) only contains quartic and quadratic powers of Y . It is easy to see
that a similar situation also occurs in the δ = π2 case; for instance, choosing n = 1 and
λ2
C = − β12 (µ + 8µ ), one obtains the quartic potential
2
(4µ + λ)2 4µ − λ 2
U (Y ) = Y −1 , (4.9)
128µ 4µ + λ
which has the well-known classical background

4µ + λ λ2 x
Y (x) = tanh µ − . (4.10)
4µ − λ 16µ 2
For generic δ, instead, also cubic and linear powers of Y appear, making more complicated
the analysis of the classical solutions.
4.1. δ = 0 case
It is convenient to start our discussion with the case δ = 0. This case, in fact, displays
those topological features which are common to all other models with 0 < δ < π2 , but
it admits a simpler technical analysis, due to the fact that parity invariance survives the
deformation of the original SG model. As we will see explicitly, in this case the results of
the FFPT and the Semiclassical Method are complementary, since they describe different
kinds of excitations present in the theory.
Fig. 1 shows the shape of this DSG potential in the two different regimes, i.e., (i) 0 <
λ < 4µ and (ii) λ > 4µ. The absolute minimum persists in the position 0 (mod 4π β ) for any
values of the couplings, while the other minimum at 2π 4π
β (mod β ) becomes relative and
disappears at the point λ = 4µ. The breaking of the degeneration between the two initial
vacua in the two-folded SG causes the confinement of the original SG solitons, as it can
be explicitly checked by applying the FFPT. The linearly rising potential, responsible for
the confinement of the SG solitons, gives rise then to a discrete spectrum of bound states
whose mass is beyond 2MSG , where MSG is the mass of the SG solitons [2,15].
The disappearing of the initial solitons represents, of course, a drastic change in the
topological features of the spectrum. At the same time, however, a stable new static
kink solution appears for λ = 0, interpolating between the new vacua at 0 and 4π β . The
existence of this new topological solution is at the origin of the complementarity between
the FFPT and the semiclassical method. By the first technique, in fact, one can follow
adiabatically the deformation of the SG breathers masses: these are neutral objects that
persist in the theory although the confinement of the original kinks has taken place. It
is of course impossible to see these particle states by using the semiclassical method,
since the corresponding solitons, which originate these breathers as their bound states,
have disappeared. semiclassical method can instead estimate the masses of other neutral
particles, i.e., those which appear as bound states of the new stable kink present in the
deformed theory.
Fig. 1. DSG potential in the case δ = 0.

Fig. 2. Kink solution (4.11).
4π
This new kink solution, interpolating between 0 and β , is given explicitly by

2π 4 λ
ϕK (x) = + arctan sinh(mx) , (4.11)
β β λ + 4µ
where
λ
m2 = µ + (4.12)
4
is the curvature of the absolute minimum. Interestingly enough [7], this background admits
an equivalent expression in terms of the superposition of two solitons of the unperturbed
sine-Gordon model, centered at the fixed points ±R
ϕK (x) = ϕSG (x + R) + ϕSG (x − R), (4.13)

where ϕSG (x) = β4 arctan[emx ] are the usual sine-Gordon solitons with the deformed mass
parameter (4.12) whereas their distance 2R is expressed in terms of the couplings by

1 4µ
R = arccosh + 1.
m λ
By looking at Fig. 2, it is clear that this background, in the small λ limit, describes the two
confined solitons of SG, which become free in the λ = 0 point, i.e., where R → ∞.
The classical energy of this kink is given by

16m λ 4µ
MK = 2 1 + √ arctanh , (4.14)
β 4µ(λ + 4µ) λ + 4µ
and in the λ → 0 limit it tends to twice the classical energy of the sine-Gordon soliton, i.e.,
√
16 µ
MK −→ , (4.15)
λ→0 β 2
therefore confirming the above picture. In the µ → 0 limit, the asymptotic value of the
above expression is instead the mass of the soliton in the sine-Gordon model with coupling
β/2. The expansion for small µ

√
8 λ/4 µ 32
MK −→ 2
+ 2 √ + O µ2 , (4.16)
µ→0 (β/2) β 3 λ
gives the first order correction which is in agreement with the result of the FFPT in the
semiclassical limit (see Appendix A).
The bound states created by the kink (4.11) and its antikink can be obtained by looking
at the poles of the semiclassical form factors of the fields ϕ(x) and ε(x), reported in
Appendix B, and their mass are given by7

(n) m MK
m(K) = 2MK sin n , 0<n<π . (4.17)
2MK m
For small µ we easily recognize the perturbation of the standard breathers in sine-Gordon
with β/2:
2 2 2
(n) 64 λ β 2 µ 32 β β
m(K) −→ 2 sin n + √ sin n + n cos n
µ→0 β 4 64 3 λ β2 64 64
2
+O µ , (4.18)
while the expansion of the bound states masses for small λ
√ 2
(n) 32 µ β
m(K) −→ 2
sin n
λ→0 β 32
2 2
1 λ λ 32 β λ β
+ √ 1 − ln sin n + n ln cos n
8 µ 16µ β 2 32 16µ 32
2
+O λ
deserves further comments: in fact, although the above masses have well-defined
asymptotic values, they do not correspond however to any state of the unperturbed SG
theory. The reason is that the classical background (4.11) in the λ → 0 limit does not
describe any longer a localized single particle. This implies that its Fourier transform
cannot be interpreted as the two-particle form factor and, consequently, its poles cannot
be associated to any bound states.
A technical signal of the disappearing of the above mentioned bound states in the λ → 0
limit can be found by computing the three particle coupling among the kink, the antikink
and the lightest bound state. The residue of the kink–antikink form factor on the pole
corresponding to the lightest bound state b(1) has to be proportional to the one-particle form
factor 0|φ|b(1) through the semiclassical 3-particle on-shell coupling of kink, antikink
and elementary boson gk k̄b :
g
Res F2 (θ ) = i √ k k̄b (1) 0φ b(1) . (4.19)
θ=θ1 2 2M∞ mb
7 Due to parity invariance, the dynamical poles of the form factor of ϕ between kink states only give the bound
states with n odd. The even states can be obtained from the form factor of the energy operator (x).
Fig. 3. Neutral states coming from: (a) solitons confinement, (b) deformations of SG breathers, (c) bound states
of the kink (4.11).
√
Since the one-particle form factor takes the constant value 1/ 2, at leading order in β we
get

1 16m 3 λ 4µ
gK K̄b = √ 1+ √ arctanh . (4.20)
4 λ β 4µ(λ + 4µ) λ + 4µ
The divergence of the coupling as λ → 0 indicates that the considered scattering processes
cannot be seen anymore as a bound state creation, i.e., the corresponding bound state
disappears from the theory. A general discussion of the same qualitative phenomenon for
the ordinary sine-Gordon model can be found in [25], where the disappearing from the
theory of a heavy breather at specific values of β is explicitly related to the divergence or
to the imaginary nature of the three particle coupling among this breather and two lightest
ones.
Summarizing, in this model we have three kinds of neutral objects, i.e., meson particles.
The first kind (a) is given by the bound states originating from the confinement potential of
the original solitons. These discrete states have masses above the threshold 2MSG , where
MSG is the mass of the SG solitons, and merge in the continuum spectrum of the non-
confined solitons in the λ → 0 limit [2,15]. The second kind (b) is represented by the
deformations of SG breathers, that can be followed by means of the FFPT and have masses,
for small λ, in the range [0, 2MSG ]. Finally, the third kind (c) is given by the bound states
(4.17) of the stable kink of the DSG theory and they have masses in the range [0, 4MSG].
All these mass spectra are drawn in Fig. 3.
Obviously, due to the non-integrable nature of this quantum field theory not all these
particles belong to the stable part of its spectrum. Apart from a selection rule coming from
the conservation of parity, decay processes are expected to be simply controlled by phase-
space considerations, i.e., a heavier particle with mass Mh will decay in lighter particles of
masses mi satisfying the condition

Mh mi . (4.21)
i
Hence, to determine the stable particles of the theory, one has initially to identify the
lightest mesons of odd and even parity with mass m− and m+ (m− < m+ ), respectively.
Then, the stable particles of even parity are those with mass m below the threshold 2m−
whereas the stable particles of odd parity are those with mass m < m− + m+ . For instance,
in the µ → 0 limit we know that the only stable mesons are those given by the particles (c),
as confirmed by the expansion (4.18). Hence, in this limit no one of the other neutral
particles is present as asymptotic states. For the mesons of type (a), this can be easily
understood since they are all above the threshold dictated by the lightest neutral particle.
The situation is more subtle, instead, for the states (b). However, their absence in the theory
with µ → 0 clearly indicates that at some particular value of λ even the lightest of these
objects acquires a mass above the threshold 2m(1) (1)
(K) , with m(K) given by (4.17). Analogous
analysis can be done for other values of the couplings so that the general conclusion is that
most of the above neutral states are nothing else but resonances of the DSG model.
In addition to the above scenario of kink states and bound state thereof, in the region λ <
4µ there is another non-trivial static solution of the theory, defined over the false vacuum
placed at ϕ = 2π β . It interpolates between the two values β and β − β arccos(1 − λ/2µ),
2π 4π 2
and then it comes back. Its explicit expression is given by

4π 4 λ
ϕB (x) = − arctan cosh(mf x) , (4.22)
β β 4µ − λ
where
λ
m2f = µ − (4.23)
4
is the curvature of the relative minimum. Similarly to the kink (4.11), it admits an
expression in terms of a soliton and an antisoliton of the unperturbed SG model

ϕB (x) = ϕSG (x + R) + ϕSG −(x − R) , (4.24)
where now ϕSG (x) = β4 arctan[emf x ] are the sine-Gordon solitons with the deformed mass
parameter (4.23) whereas their distance 2R is now given by

1 4µ
R= arcsinh − 1. (4.25)
mf λ
In the small λ limit, it is clear that this background describes the confined soliton and
antisoliton of the SG model, which become free in the λ = 0 point, i.e., where R → ∞.
The classical background (4.22) is not related to any stable particle in the quantum
theory. This can be directly seen from Eq. (3.5); in fact, Lorentz invariance always
implies the presence of the eigenvalue ω02 = 0, with corresponding eigenfunction η0 (x) =
d
dx ϕcl (x). However, in the case of the solution (4.22) the eigenfunction η0 clearly displays a
node, which indicates that the corresponding eigenvalue is not the smallest in the spectrum.
Hence, there must be a lower eigenvalue ω−1 2 < 0, with a corresponding imaginary part of
the mass relative to this particle state. Furthermore, the instability of (4.22) can be related
to the theory of false vacuum decay [26,27]: due to the deep physical interest of this topic,
we will discuss it separately in Section 5.
Fig. 4. Bounce-like solution (4.22).
Fig. 5. DSG potential in the case δ = π3 .
4.2. Comments on generic δ case
We have already anticipated that the qualitative features of the theory relative to δ = 0
case are common to all other theories associated to the values of δ in the range 0 < δ < π2 .
This can be clearly understood by looking at the shape of the potential, which is shown in
Fig. 5 for the case δ = π3 .
In contrast to the δ = 0 case, parity invariance is now lost in these models, and the
minima move to values depending on the couplings. Furthermore, in addition to the change
in the nature of the original vacuum at 2πβ , which becomes a relative minimum by switching
on λ, there is also a lowering of one of the two maxima. These features make much more
complicated the explicit derivation of the classical solutions, as we have mentioned at the
beginning of the section.
However, it is clear from Fig. 5 that the excitations of these theories share the same
nature of the ones in the δ = 0 case. In fact, the original SG solitons undergo a confinement,
while a new stable topological kink appears, interpolating between the new degenerate
minima. Hence, the analysis performed for δ = 0 still holds in its general aspects, i.e., also
in these cases the spectrum consists of a kink, antikink, and three different kinds of neutral
particles.
4.3. δ = π
2 case
The value δ = π2 describes the peculiar case in which no confinement phenomenon takes
place, since the two different vacua of the original two-folded SG remain degenerate also in
the perturbed theory. As a consequence, the original SG solitons are also asymptotic states
in the perturbed theory. By means of the Semiclassical Method we can then compute their
bound states, which represent the deformations of the two sets of breathers in the original
two-folded SG. Hence, in this specific case FFPT and semiclassical method describe the
same objects, and their results can be compared in a regime where both β and λ are small.
Fig. 6 shows the behavior of this DSG potential. There are two regions, qualitatively
different, in the space of parameters, the first given by 0 < λ < 4µ and the second given by
λ > 4µ. They are separated by the value λ = 4µ which has been identified in [2] as a phase
transition point. We will explain how this identification is confirmed in our formalism.
Let us start our analysis from the coupling constant region where λ < 4µ. Switching on
λ, the original inequivalent minima of the two-folded sine-Gordon, located at φmin = 0,
β (mod β ), remain degenerate and move to φmin = −φ0 , β + φ0 (mod β ), with
2π 4π 2π 4π
φ0 = 2
β
λ
arcsin 4µ . The common curvature of these minima is
1 λ2
m2 = µ − . (4.26)
16 µ
Correspondingly there are two different types of kinks, one called “large kink” and
β + φ0 , the other called “small
interpolating through the higher barrier between −φ0 and 2π
kink” and interpolating through the lower barrier between 2πβ + φ0 and − φ0 . Their
4π
β
classical expressions are explicitly given by

π 4 4µ + λ m
ϕL (x) = + arctan tanh x (mod 4π), (4.27)
β β 4µ − λ 2

3π 4 4µ − λ m
ϕS (x) = + arctan tanh x (mod 4π). (4.28)
β β 4µ + λ 2
Fig. 6. DSG potential in the δ = π2 case.

With the notation previously introduced, the vacuum structure of the corresponding
quantum field theory consists of two sets of inequivalent minima, denoted by |0 and
|1, identified modulo 2, i.e., |a + 2n ≡ |a. The spontaneous breaking of the symmetry
T : ϕ → 2π − ϕ selects one of these minima as the vacuum. If we choose to quantize the
theory around |0, the admitted quantum kink states are |L = |K0,1 and |S̄ = |K0,−1 ,
with the corresponding antikink states |L̄ = |K1,0 and |S = |K−1,0 , and topological
charges
βφ0
QL = −QL̄ = 1 + ,
π
βφ0
QS = −QS̄ = 1 − . (4.29)
π
Multikink states of this theory satisfy the selection rule coming from the continuity of
vacuum indices and are generically given by

Kα α (θ1 )Kα α (θ2 ) · · · Kα α (θn−2 )Kα α (θn−1 ) . (4.30)
1 2 2 3 n−2 n−1 n−1 n
The leading contributions to the masses of the large and small kink are given by their
classical energies, which can be easily computed

8m λ π λ
ML,S = 2 1 ± ± arcsin . (4.31)
β 16µ2 − λ2 2 4µ
The expansion of this formula for small λ is given by
√
8 µ λ π
ML,S −→ 2 ± 2 √ + O λ2 , (4.32)
λ→0 β β µ
and the first order correction in λ coincides with the result of FFPT in the semiclassical
limit (see Appendix A).
Since two different types of kink |L and |S are present in this theory, one must be
careful in applying Eq. (3.2) to recover the form factors of each kink separately. In fact,
one could expect that both types of kink contribute to the expansion over intermediate
states used in [4] to derive the result. For instance, starting from the vacuum |0 located at
φmin = −φ0 there might be the intermediate matrix elements 0 S̄|O|L0 and 0 L|O|S̄0 .
However, if O is a non-charged local operator, it easy to see that these off-diagonal
elements have to vanish for the different topological charges of |L and |S. Hence, the
expansion over intermediate states diagonalizes and one recovers again Eq. (3.2).
Therefore, from the dynamical poles of the form factor of ϕ on the large and small
kink–antikink states, reported in Appendix B, we can extract the semiclassical masses of
two sets of bound states:

m ML
m(n)
(L) = 2M L sin nL , 0 < nL < π , (4.33)
2ML m

(n) m MS
m(S) = 2MS sin nS , 0 < nS < π . (4.34)
2MS m
Expanding for small λ, we can see that these states represent the perturbation of the two
sets of breathers in the original two-folded sine-Gordon model:
√ 2 2 2
(n) 16 µ β λ 1 β n β
m(L,S) −→ 2
sin n ± 2π √ 2
sin n − cos n
λ→0 β 16 µ β 16 16 16
2
+O λ . (4.35)
A discussion of these results, in comparison with previous studies of this model [22], is
reported in Appendix C.
Concerning the stability of the above spectrum, for λ < 4µ the only stable bound states
(n) (1)
are the ones with m(L,S) < 2m(S) ; for λ close enough to 4µ, however, the small kink creates
no bound states, hence the stability condition becomes m(n) (1)
(L) < 2m(L) .
In the limit λ → 4µ, φ0 tends to πβ , the two minima at 2πβ + φ0 and β − φ0 coincide and
4π
the small kink disappears, becoming a constant solution with zero classical energy. All the
large kink bound states masses collapse to zero, and in this limit all dynamical poles of the
large kink form factor disappear. This is nothing else but the semiclassical manifestation
of the occurrence of the phase transition present in the DSG model (see [2]).
In the second coupling constant region, parameterized by λ > 4µ, there is only one
minimum at fixed position − πβ (mod 4π β ), with curvature
λ
m2 = − µ. (4.36)
4
There is now only one type of kink, given by

π 4 λ
ϕK (x) = + arctan sinh(mx) . (4.37)
β β λ − 4µ
Its classical mass, expanded for small µ, is again in agreement with FFPT (see
Appendix A)

16m λ π λ − 8µ
MK = 2 1 + √ − arcsin (4.38)
β 4 µ(λ − 4µ) 2 λ
√
8 λ/4 µ 32
−→ 2
− 2 √ + O µ2 . (4.39)
λ→0 (β/2) β 3 λ
The bound states of this kink (see Appendix B for the explicit form factor) have masses

(n) m MK
m(K) = 2MK sin n , 0<n<π . (4.40)
2MK m
For small µ, these states are nothing else but the perturbed breathers of the sine-Gordon
model with coupling β/2:
2 2 2
(n) 64 λ β 2 µ 32 β β
m(K) −→ 2 sin n − √ 2
sin n + n cos n + O µ2 .
µ→0 β 4 64 3 λ β 64 64
In closing the discussion of the δ = π/2 case, it is interesting to mention another model
which presents a similar phase transition phenomenon, although in a reverse order. This
is the double sinh-Gordon model (DShG), discussed in Appendix D. The similarity is due
to the fact that also in this case a topological excitation of the theory becomes massless
at the phase transition point, but the phenomenon is reversed, because in DSG the small
kink disappears when λ reaches the critical value, while in DShG a topological excitation
appears at some value of the perturbing coupling.
5. False vacuum decay
The semiclassical study of false vacuum decay in quantum field theory has been
performed by Callan and Coleman [26], in close analogy with the work of Langer [27].
The phenomenon occurs when the field theoretical potential U (ϕ) displays a relative
minimum at ϕ+ : this classical point corresponds to the false vacuum in the quantum theory,
which decays through tunneling effects into the true vacuum, associated with the absolute
minimum ϕ− (see Fig. 7).
The main result of [26] is the following expression for the decay width per unit time
and unit volume:
−1/2
Γ B
−B/h̄ det [−∂ + U (ϕ)]
2
= e det[−∂ 2 + U (ϕ )] 1 + O(h̄) , (5.1)

V 2π h̄ +
specialized here to the case of two-dimensional space–time. Omitting any discussion of
the determinant, about which we refer to the original papers [26], we will present here an
explicit analysis of the coefficient B.
It has been shown that B coincides with the Euclidean action of the so-called “bounce”
background ϕB :
∞
1 dϕB 2
B = SE = 2π dρ ρ + U (ϕB ) . (5.2)
2 dρ
0
This classical solution is the field-theoretical generalization of the path of least √
resistance
in quantum mechanical tunneling; it only depends on the Euclidean radius ρ = τ 2 + x 2
Fig. 7. Generic potential for a theory with a false vacuum.

and satisfies the equation

d 2 ϕB 1 dϕB
2
+ = U [ϕB ], (5.3)
dρ ρ dρ
with boundary conditions
dϕB
lim ϕB (ρ) = ϕ+ , (0) = 0. (5.4)
ρ→∞ dρ
Although in general one does not know explicitly the bounce solution, it is possible to
set up some approximation to extract a closed expression for the coefficient B. The so-
called “thin wall” approximation consists in viewing the potential U (ϕ) as a perturbation
of another potential U+ (ϕ), which displays degenerate vacua at ϕ± and a kink ϕK (x)
interpolating between them. The small parameter for the approximation is the energy
difference ε = U (ϕ+ ) − U (ϕ− ).
In this framework, one can qualitatively guess that the bounce has a value ϕ(0) very
close to ϕ− , then it remains in this position until some vary large ρ = R and finally it moves
quickly towards the final value ϕ+ . For ρ near R, the first-derivative term in Eq. (5.3) can
be neglected; if in addition one also approximates U with U+ , then one can express the
unknown bounce solution as [26]
ϕ , ρ R,
−
ϕB (ρ) = ϕK (ρ − R), ρ ≈ R, (5.5)
ϕ+ , ρ R.
Since the bounce has to represent the path of least resistance, the parameter R, free up to
this point, can be fixed by minimizing the action
SE = −πR 2 + 2πRMK , (5.6)
which is given by the sum of a volume term and a surface term. Hence, the condition
dSE
dR = 0 is realized by the balance of these two different terms in competition, and it finally
gives
MK M2
R= ⇒ B = π K . (5.7)
ε ε
In the DSG model, however, we know explicitly the bounce background in the thin
wall regime (here we have ε = β2λ2 ), without any approximation on the potential. This is
given by the solution (4.22) with x replaced by ρ, that can be directly used to estimate the
decay width. Unfortunately the integral in (5.2) does not admit a simple expression to be
expanded for small λ, but it is clear from Eq. (4.24) and Fig. 4 that the leading contribution
is given by

R+r 2
dϕSG 8π 16µ
SE 2πR dx (R − x) log , (5.8)
dx β2 λ
R−r
with R given by (4.25). This behavior in λ does not agree with the general prediction
(5.7). The reason can be traced out in the fact that Eq. (4.24) explicitly realizes the relation
between the bounce and the kink of the unperturbed theory, but in a more sophisticated
way than (5.5). In fact, the mass parameter mf of the SG kink ϕSG is dressed to be the
one of the DSG theory, and the parameter R is not free, since (4.22) is already the result of
a minimization process, being a solution of the Euler–Lagrange equations. The thin wall
approximation can be still consistently used because R is very big for small λ, while the
crucial difference is that the volume term is now missing from the action, since the value
ϕB (0) = ϕ1 is the so-called classical turning point (see Fig. 7), degenerate with the false
vacuum. It is worth noting that the path of least resistance in quantum mechanics precisely
interpolates between the false vacuum and the turning point.
Up to the determinant factor, our result for the leading term in the decay width is then
2
Γ 4 λ 8π/β 16µ
2 log . (5.9)
V β 16µ λ
It will be interesting to investigate whether the above mentioned difference with the
prediction (5.7) is a particular feature of the DSG model or it appears for a generic potential
if one improves the approximate description of the bounce along the lines discussed here.
6. Other kind of resonances
The appearance of resonances in the classical scattering of the double sine-Gordon

kinks has been extensively studied with numerical techniques, and a complete picture of
this phenomenon can be found in [7]. In this work, the key ingredient for the presence of
resonances was identified in the presence of a discrete eigenvalue, besides the zero mode, of
the small oscillations around the kink background. This eigenvalue, called “shape mode”,
represents an internal excitation of the kink [3,21].
This mechanism can be easily interpreted also in our formalism, but unfortunately in the
case of the double sine-Gordon model we were not able to solve analytically the stability
equation around the kink backgrounds. Hence, we will limit ourselves to the discussion
of the same phenomenon in a simpler theory, the φ 4 field theory in the broken symmetry
phase8
g 4 m2 2 m4
V (φ) = φ − φ + . (6.1)
4 2 4g
The standard kink background of this theory is given by

m mx
φcl (x) = √ tanh √ , (6.2)
g 2
√ 3
with classical energy M = 2 3 2 mg . The small oscillations (3.5) around this solution have,
in addition to the usual translational mode ω0 = 0, another discrete eigenvalue
3
ω12 = m2 , (6.3)
2
8 The main features of the numerical analysis performed in [7] for the DSG model were indeed previously
recognized in this simpler theory [28].
which represents an internal excitation of the kink [3,21]. This feature, quite crucial for
the analysis performed in [28], has a counterpart in our formalism. In fact, it was shown
by Goldstone and Jackiw [4] that, performing the Fourier transform of the corresponding
eigenfunction η1 (a), one can write the form factor of the field φ between asymptotic states
containing a simple and an excited kink. Furthermore, also this result, as the previous one
relative to the form factors of the elementary kinks, can be refined in terms of the rapidity
variable, so that one obtains a covariant expression that can be analytically continued in
the crossed channel. Since in this case the eigenfunction is

sinh mx
√
2
η1 (x) = − mx ,
2 √
(6.4)
2 cosh
2
for the corresponding form factor we have
Mπ M(iπ − θ )
0φ(0)p̄2 p1∗ = −i
, (6.5)
61/4 m5/2 cosh √π M(iπ − θ )
2m
where the p1∗ denotes the momentum of the excited kink state. The dynamical poles of this
object correspond to bound states with masses

(n) 2 2 3 g
mb∗ = 4M(M + ω1 ) sin (2n + 1) + ω12 . (6.6)
8 m2
The states with

8 m2 4M 2 − ω12 4 m2
arcsin < 2n + 1 < π (6.7)
3 g 4M(M + ω1 ) 3 g
have masses in the range
(n)
2M < mb∗ < 2M + ω1 , (6.8)
and, therefore, they can be seen as resonances in the kink–antikink scattering.
Since the numerical analysis done in [28] is independent of the coupling constant,9
a quantitative comparison with our semiclassical result is rather difficult, due to the
dependence on g of (6.7). However, the presence of many resonance states seen at classical
level is qualitatively confirmed to persist also in the quantum field theory at small g, i.e.,
in its semiclassical regime, according to (6.7).
Back to the double sine-Gordon model, the shape mode with the relative resonances has
been numerically observed for the small kink (4.28) in the δ = π2 case, and for the kink
(4.11) in the case δ = 0. Our analysis is in agreement with these results, and it adds another
possibility for the small kink case. In fact, since in this regime it is also present the large
kink, which has higher mass, the resonances seen in the small kink–antikink scattering are
(n)
related both to their excited bound states with masses mS ∗ in the range
2MS < m(n)

S ∗ < 2MS + ω̃1 , (6.9)
9 Classically, in fact, one can always rescale the field and eliminate the coupling constant g.
and to the large kink–antikink bound states with masses in the range
2MS < m(n)

L < 2ML , (6.10)
where m(n)
L are given by (4.33).
7. Conclusions
When available, the semiclassical method is an efficient tool for studying the mass
spectrum of an integrable or a non-integrable theory. In the last case, it may be
complementary to the form factor perturbation theory or it may provide results comparable
with this method. We have applied both techniques for analyzing the mass spectrum of
the non-integrable quantum field theory given by the double sine-Gordon model, for few
qualitatively different regions of its coupling-constants space. This model appears to be
an ideal theoretical playground for understanding some of the relevant features of non-
integrable models. By moving its coupling constants, it shows, in fact, different types
of kink excitations and confinement phenomena, a rich spectrum of meson particles,
resonance states, false vacuum decay and the occurrence of a phase transition. In light
of the many applications it finds in condensed matter systems, it would be interesting to
investigate further its properties.
Acknowledgements
The authors thank G. Delfino for valuable discussions. Two of us (G.M. and V.R.) would
also like to thank the Instituto de Fisica Teorica in San Paulo for the warm hospitality
during the period of their staying, when this work was done. We also acknowledge
interesting discussions with Z. Bajnok, L. Palla, G. Takacs and F. Wagner. Moreover, G.M.
thanks them for the nice hospitality during his visit to Budapest. This work was partially
supported by the Italian COFIN contract “Teoria dei Campi, Meccanica Statistica e Sistemi
Elettronici” and by the European Commission TMR programme HPRN-CT-2002-00325
(EUCLID). G.S. thanks FAPESP for the financial support.
Appendix A. Kink mass corrections in the FFPT
In this appendix we compute by means of the FFPT the corrections to the kink masses
in the semiclassical limit, which is relevant for a comparison with our results.
For small λ, we have to consider the DSG model as a perturbation of the two-folded
sine-Gordon [22]. In the δ = π/2 case, the perturbing operator is Ψ = sin β2 φ. Its form
factors between the vacuum and the two possible kink–antikink asymptotic states are
obtained at the semiclassical level by performing the Fourier transform of sin[ β2 Kk,k+1
cl (x)]
cl
[6], with Kk,k+1 (x) given by Eq. (4.2). Hence we obtain
8π 1
FKΨ (θ ) = (−1)k . (A.1)
k,k+1 ,K̄k,k+1 β 2
cosh 2 (θ − iπ)
4π
β
The first order correction in λ to the kink masses is then
λ 1 Ψ λ π
δMKk,k+1 = 2 F (iπ) = (−1)k 2 √ , (A.2)
β MK Kk,k+1 ,K̄k,k+1 β µ
in agreement with the correction to the classical masses (4.32), since K0,1 is associated
with the large kink, and K1,2 with the small one.
In the δ = 0 case, instead, we can explicitly see how the solitons disappear from the
spectrum as soon as λ is switched on. The form factor of the operator Ψ = cos β2 φ has, in
fact, a divergence at θ = iπ
8π 1
FKΨ (θ ) = −i (−1)k . (A.3)
k,k+1 ,K̄k,k+1 β2 sinh 4π2 (iπ − θ )
β
The other interesting regime to explore is the small µ limit. In the case δ = 0, this can
be seen as the perturbation of the SG model at coupling β̃ = β/2 by means of the operator
Ψ = cos 2β̃ϕ. The semiclassical form factor is
16 32 iπy
Ψ
FK, K̄
(θ ) = 2
1 − 2y 2 , (A.4)
3 β sin iπy
where we have defined y = 16
β2
(iπ − θ ). The corresponding mass correction is given by
µ 1 Ψ µ 16 1
δMK = F (iπ) = 2 √ , (A.5)
β 2 MK K,K̄ β 3 λ/4
in agreement with (4.16).
The case δ = π2 can be described by shifting the original SG field as ϕ → ϕ + πβ . In this
way the perturbing operator becomes −Ψ and we finally obtain the same mass correction
but with opposite sign, as in (4.38).
Appendix B. Semiclassical form factors
In this appendix we explicitly present the expressions of the two-particle form factors,
on the asymptotic states given by the different kinks appearing in the DSG theory, of the
operators ϕ(x) and ε(x), the last one defined by

1 dϕ 2
ε(x) ≡ + V ϕ(x) .
2 dx
These matrix elements are obtained by performing the Fourier transforms of the
corresponding classical backgrounds, as indicated in (3.2) and (3.3). We use the notation:

FKΨK̄ (θ ) = 0Ψ (0)K(θ1 )K̄(θ2 ) ,
with θ = θ1 − θ2 .
For the kink (4.11) in the δ = 0 case we have

ϕ 4π 2
4π 1 cos α MmK (iπ − θ )
FK K̄ (θ ) = MK δ MK (iπ − θ ) + i
, (B.1)
β β iπ − θ cosh π MK (iπ − θ )
2 m
where

λ + 4µ
α = arccosh ,
λ
while m and MK are given by (4.12) and (4.14), respectively, and
128π m3 MK
FKε K̄ (θ ) = −
β2 λ
M

1 d sinh (arccosh c) 2mK (iπ − θ )
×
√
sinh π M2m (iπ − θ )
K dc c2 − 1
M

2 sinh π d c sinh (arccosh c) 2mK (iπ − θ )
−
√ ,
cosh π MmK (iπ − θ ) − 1 dc c2 − 1
(B.2)
where c = 1 + 8µ
λ .
For the large kink (4.27) in the δ = π
case (with λ < 4µ) we have
2
M
ϕ 2π 2
4π 1 sinh α mL (iπ − θ )
FLL̄ (θ ) = ML δ ML (iπ − θ ) + i
, (B.3)
β β iπ − θ sinh π ML (iπ − θ )
m
where

4µ + λ
α = 2 arctan ,
4µ − λ
while m and ML are given by (4.26) and (4.31), respectively, and
M

8π m3 ML 1 d sinh (arccos c) mL (iπ − θ )
FLL̄ (θ ) = 2
ε

√ ,
β µ sinh π ML (iπ − θ ) dc 1 − c2
m
(B.4)
where c = − 4µ
λ
.
For the small kink (4.28) in the δ = π
case (with λ < 4µ) we have
2
M
ϕ 6π 2
4π 1 sinh α mS (iπ − θ )
FS S̄ (θ ) = MS δ MS (iπ − θ ) + i
, (B.5)
β β iπ − θ sinh π MS (iπ − θ )
m
where

4µ − λ
α = 2 arctan ,
4µ + λ
while m and MS are given by (4.26) and (4.31), respectively, and
8π m3 MS 1
FSεS̄ (θ ) =
β 2 µ sinh π MS (iπ − θ )
m
M

d sinh (arccos c) mS (iπ − θ )
× √ , (B.6)
dc 1 − c2
where c = 4µλ
.
Finally, for the kink (4.37) in the δ = π
2 case (with λ > 4µ) we have

ϕ 2π 2
4π 1 cos α MmK (iπ − θ )
FK K̄ (θ ) = MK δ MK (iπ − θ ) + i
, (B.7)
β β iπ − θ cosh π MK (iπ − θ )
2 m
where

λ − 4µ
α = arccosh ,
λ
while m and ML are given by (4.36) and (4.38), respectively, and
M

128π m3 MK 1 d sinh (arccos c) 2mK (iπ − θ )
FKε K̄ (θ ) = − 2
√
β λ sinh π M2m (iπ − θ )
K dc 1 − c2
M

2 sinh π d c sinh (arccos c) 2mK (iπ − θ )
−
√ ,
cosh π MmK (iπ − θ ) − 1 dc 1 − c2
(B.8)
8µ
where c = 1 − λ .
Appendix C. Neutral states in the δ = π

2 case
The semiclassical results reported in the text, i.e., Eqs. (4.33), (4.34) and (4.35), pose
an interesting question about the nature of neutral states in the DSG model at δ = π2 . It
should be noticed, in fact, that the first order correction in λ obtained by the Semiclassical
Method does not match with the results reported in [22] where, by using the FFPT and
an extrapolation of numerical data, the authors concluded that this correction was instead
identically zero.10 It is worth discussing this problem in more detail.
10 It is worth stressing that the linear correction (4.35) in λ is very small even for finite values of β (it is easy to
check, indeed, that the first term of its expansion is 24π ( β 2 )2 ) and somehow compatible with the numerical data
16
given in [22].
In the standard sine-Gordon model, the breathers |bn , with n odd (or even), are defined
as the bound states of odd (or even) combinations of |K K̄ and |K̄K, where K represents
the soliton and K̄ the antisoliton. The combinations |K K̄ ± K̄K are eigenstates of the
parity operator P : φ → −φ, which commutes with the Hamiltonian and acts on the soliton
transforming it into the antisoliton. The above mentioned identification of the bound states
relies on a very peculiar feature of the sine-Gordon S-matrix in the soliton sector [29],
whose elements are defined as
K(θ1 )K̄(θ2 ) = ST (θ12 )K̄(θ2 )K(θ1 ) + SR (θ12 )K(θ2 )K̄(θ1 ), (C.1)

K(θ1 )K(θ2 ) = S(θ12 )K(θ2 )K(θ1 ), (C.2)
K̄(θ1 )K̄(θ2 ) = S(θ12 )K̄(θ2 )K̄(θ1 ). (C.3)
In fact, both the transmission and the reflection amplitudes ST (θ ) and SR (θ ) display poles
at θn∗ = i(π − nξ ), with residua which are equal or opposite in sign depending whether n
is odd or even. Hence, the diagonal elements
S− (θ ) = ST (θ ) − SR (θ ) , (C.4)
2
1
S+ (θ ) = ST (θ ) + SR (θ ) (C.5)
2
have only the poles with odd or even n, respectively, and for each n there is only one bound
state with definite parity.
However, this is a special feature of the sine-Gordon model which finds no counterpart,
for instance, in other problems with a similar structure. As an explicit example, one can
consider the (RSOS)3 scattering theory, which displays a 3-fold degenerate vacuum and
two types of kink and antikink with the same mass. The central vacuum is surrounded by
two other minima, as in the sine-Gordon case, and this gives the possibility to define both
a kink–antikink state and an antikink–kink state around it. The minimal scattering matrix,
given in [30], can be dressed with a CDD factor to generate bound states. It is easy to
check that the common poles in the transmission and reflection amplitudes have in this
case different residua, giving rise to two distinct bound states, degenerate in mass, over the
central vacuum.
(0) (1)
Hence, if we call |bn the bound states of kink–antikink and |bn the bound states of
antikink–kink, in general we have to consider them as two distinct excitations, and if they
have the same mass we can build two other states from their linear combinations
(±) |bn(0) ± |bn(1)

b = √ . (C.6)
n
2
The peculiarity of the sine-Gordon model is the removal of this double multiplicity due to
(+) (−)
the fact that the states |b2n+1 and |b2n decouple from the theory. This feature is shared
also by the two-folded version of the model, since the kink scattering amplitudes have the
same analytical form as in SG [23].
In the two-folded SG there are two different kink states |K−1,0 and |K0,1 (see
Section 4 and Ref. [23] for the notation), and the parity P , which is still an exact symmetry
of the theory, acts on them transforming the kink of one type into the antikink of the other
type:
P : |K0,1 → |K0,−1 , |K−1,0 → |K1,0 . (C.7)
If we quantize the theory around the vacuum |0, we can define |bn(0) as the bound states of
(1)
|K0,1 K1,0 , and |bn as the bound states of |K0,−1 K−1,0 . These degenerate states, which
transform under P as

P : bn(0) → bn(1) , bn(1) → bn(0) , (C.8)
(±)
can be still organized in parity eigenstates |bn , and the particular dynamics of the
problem causes the decoupling of half of them from the theory. Furthermore, it is easy
to see that the form factors of an odd operator between two of these states has to vanish in
virtue of the relation

β β
0 sin φ bn(±) bn(±) = 0P −1 P sin φ P −1 P bn(±)bn(±)
2 2
β (±) (±)
= − 0 sin φ bn bn ,
2
leading to the FFPT result that the breathers receive a zero mass correction at first order
in λ, as it is claimed in [22].
However, FFPT can be applied by taking into account the nature of neutral states in
the DSG model, where the addition to the Lagrangian of the term − βλ2 sin β2 ϕ spoils the
invariance under P . The kinks |K−1,0 and |K0,1 are deformed into the small and large
kinks |S and |L, respectively, which are not anymore degenerate in mass and cannot
be superposed in linear combinations. Hence, the neutral states present in the theory are
(L) (S) (0) (1)
|bn and |bn , deformations of |bn and |bn respectively. In virtue of the general
considerations presented above, one can see that this interpretation does not lead to any
(+) (−)
drastic change in the spectrum. In fact, the states |b2n+1 and |b2n have no reason to
decouple in the DSG theory, but they have to carry a coupling which is a function of λ
adiabatically going to zero in the two-folded SG limit.
(0) (1)
A proper use of the FFPT on |bn and |bn reproduces indeed the situation described
by (4.35), in which the two sets of breathers receive mass corrections including also odd
terms in λ, but with opposite signs. This is easily seen by considering the P transformations
in the two-folded SG model:

β (0) (0) −1 β β
0 sin φ bn bn = 0 P P sin φ P −1 P bn(0)bn(0) = − 0 sin φ bn(1)bn(1) ,

2 2 2
which gives, at first order in λ,
δm(L) = −δm(S) , (C.9)
in agreement with our semiclassical result (4.35). It is worth noting that also with this
interpretation the total spectrum of the DSG model remains unchanged under the action
of P , which corresponds to the transformation λ → −λ. In fact, the two types of kinks
and breathers are mapped one into the other. This is consistent with the observation that P ,
although it is not anymore a symmetry of the perturbed theory, simply realizes a reflection
of the potential, hence the total spectrum should be invariant under it.
Presently the above symmetry considerations seem to us the correct criterion to define
the neutral states, and find confirmation in our semiclassical result (4.35). However, the
available numerical data presented in [22] pose a challenge to this interpretation and
further studies are needed to solve this interesting and delicate problem. In fact, although
δm(L) and δm(S) are not forced to vanish by symmetry arguments, there is in principle the
possibility that both of them are identically zero in the complete quantum computation.
This could follow from the use of the exact kink masses entering Eqs. (4.33) and (4.34),
together with a proper shift of the semiclassical pole in the form factors, due to higher order
contributions. The exact cancellation of the linear corrections is a very strong requirement,
in support of which we have presently no indication in the theory, but a careful analysis of
this point is nevertheless an interesting open problem.
Appendix D. Double sinh-Gordon model
Among the different qualitative features taking place in perturbing integrable models, a
situation particularly interesting is the one in which the perturbation is adiabatic for small
values of the parameters but nevertheless a qualitative changes in the spectrum occurs by
increasing its intensity.
This is indeed the situation in the δ = π2 case of DSG model, where we have two
types of kinks for small λ, but at λ = 4µ one of them disappears from the spectrum. This
phenomenon is obviously unaccessible by means of FFPT, hence the semiclassical method
is the best tool to describe it.
Here we consider another interesting example of this kind, realized by the double sinh-
Gordon model (DShG). In this case the phenomenon is even more evident, because in the
unperturbed sinh-Gordon model there are no kinks at all, but just one scalar particle, while
perturbing it, at some critical value of the coupling a kink and antikink appear, i.e., there is
a deconfinement phase transition of these particles.
The DShG potential, shown in Fig. 8, is expressed as

µ λ β
V (ϕ) = cosh βϕ − cosh ϕ . (D.1)
β2 β2 2
In the regime λ < 4µ the qualitative features are the same as in the unperturbed
sinh-Gordon model. At λ = 4µ, however, the single minimum splits in two degenerate
minima, which for λ > 4µ are located at ϕ± = ± β2 arccos h 4µλ
. A study of the classical
thermodynamical properties of the theory in this regime has been performed in [31] with
the transfer integral method.
Fig. 8. DShG potential.
The kink interpolating between the two degenerate vacua is

4 λ − 4µ m
ϕK (x) = arctanh tanh x , (D.2)
β λ + 4µ 2
with
λ2 − 16µ2
m2 = .
16µ
Its classical mass is given by

8m 2λ λ − 4µ
MK = 2 −1 + arctanh . (D.3)
β λ2 − 16µ2 λ + 4µ
From the form factor of ϕ on the kink–antikink asymptotic state, expressed as
λ MK

π 1 sin arccosh 4µ m (iπ − θ )

F2 (θ ) = −i
, (D.4)
β iπ − θ sinh π MmK (iπ − θ )
we derive the bound states spectrum

(n) m MK
m(K) = 2MK sin n , 0<n<π . (D.5)
2MK m
All the kink–antikink bound states disappear from the theory at a certain value λ∗ > 4µ
such that π MmK |λ∗ = 1, and the kink becomes a constant solution with zero classical energy
when λ → 4µ. This is the semiclassical manifestation of a phase transition, analogous
to the one observed in DSG with δ = π2 . As we have already anticipated, here the
phenomenon occurs in a reverse order, since in this case a kink appears in the theory by
increasing the coupling λ.
References
[1] G. Delfino, G. Mussardo, P. Simonetti, Nucl. Phys. B 473 (1996) 469.

[2] G. Delfino, G. Mussardo, Nucl. Phys. B 516 (1998) 675.
[3] R.F. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 10 (1974) 4130;
R.F. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 11 (1975) 3424.
[4] J. Goldstone, R. Jackiw, Phys. Rev. D 11 (1975) 1486.
[5] Y.S. Tyupkin, V.A. Fateev, A.S. Shvarts, Sov. J. Nucl. Phys. 22 (1976) 321.
[6] G. Mussardo, V. Riva, G. Sotkov, Nucl. Phys. B 670 (2003) 464.
[7] D.K. Campbell, M. Peyrard, P. Sodano, Physica D 19 (1986) 165.
[8] C.A. Condat, R.A. Guyer, M.D. Miller, Phys. Rev. B 27 (1983) 474.
[9] J.A. Krumhansl, J.R. Schrieffer, Phys. Rev. B 11 (1975) 3535.
[10] R. Giachetti, P. Sodano, E. Sorace, V. Tognetti, Phys. Rev. B 30 (1984) 4014.
[11] R.K. Bullough, P.J. Caudrey, H.M. Gibbs, Double sine-Gordon model, in: Solitons, in: Topics in Current
Physics, vol. 17, Springer, Berlin, 1980.
[12] D. Controzzi, G. Mussardo, Phys. Rev. Lett. 92 (2004) 021601.
[13] F.D.M. Haldane, Phys. Lett. A 93 (1983) 464;
F.D.M. Haldane, Phys. Rev. Lett. 50 (1983) 1153;
F.D.M. Haldane, J. Appl. Phys. 57 (1985) 3359;
I. Affleck, Nucl. Phys. B 257 (1985) 397;
I. Affleck, F.D.M. Haldane, Phys. Rev. B 36 (1977) 5291.
[14] I. Affleck, Quantum theory methods and quantum critical phenomena, in: Fields, Strings and Critical
Phenomena, Les Houches XLIX, 1988.
[15] I. Affleck, Soliton confinement and the excitation spectrum of spin-Peierls antiferromagnets, in: A. Skjeltorp,
D. Sherrington (Eds.), Dynamical Properties of Unconventional Magnetic Systems, in: NATO ASI Series E,
vol. 349, Kluwer Academic, 1998;
I. Affleck, cond-mat/9705127.
[16] M. Fabrizio, A.O. Gogolin, A.A. Nersesian, Nucl. Phys. B 580 (2000) 647.
[17] M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 445.
[18] F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, World Scientific,
1992.
[19] V.P. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 6 (1991) 3419.
[20] C.G. Callan, D.J. Gross, Nucl. Phys. B 93 (1975) 29;
J.L. Gervais, A. Jevicki, B. Sakita, Phys. Rev. D 12 (1975) 1038;
J.L. Gervais, A. Jevicki, Nucl. Phys. B 110 (1976) 113.
[21] R. Rajaraman, Solitons and Instantons, North-Holland, Amsterdam, 1982.
[22] Z. Bajnok, L. Palla, G. Takacs, F. Wagner, Nucl. Phys. B 601 (2001) 503.
[23] Z. Bajnok, L. Palla, G. Takacs, F. Wagner, Nucl. Phys. B 587 (2000) 585.
[24] J.D. Gibbon, N.C. Freeman, R.S. Johnson, Phys. Lett. A 65 (1978) 380.
[25] C.J. Goebel, Prog. Theor. Phys. Suppl. 86 (1986) 261.

[26] S. Coleman, Phys. Rev. D 15 (1977) 2929;
C.J. Callan, S. Coleman, Phys. Rev. D 16 (1977) 1762.
[27] J.S. Langer, Ann. Phys. 41 (1967) 108.
[28] D.K. Campbell, J.F. Schonfeld, C.A. Wingate, Physica D 9 (1983) 1.
[29] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[30] Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 497.
[31] S. Habib, A. Khare, A. Saxena, Phys. Rev. Lett. 79 (1997) 3797;
S. Habib, A. Khare, A. Saxena, Physica D 123 (1998) 341.
The algebraic Bethe ansatz for the Osp(2|2) model

with open boundary conditions
Guang-Liang Li a , Kang-Jie Shi b , Rui-Hong Yue b
a Department of Applied Physics, Xi’an Jiaotong University, Xi’an 710049, China
b Institute of Modern Physics, Northwest University, Xi’an 710069, China
Received 31 December 2003; received in revised form 13 February 2004; accepted 18 March 2004
Abstract
We obtain four different diagonal reflecting matrices by solving the reflection equation of the
Osp(2|2) model. At the same time, we solve the model with open boundary condition by using
the algebraic Bethe ansatz. The procedure of constructing the multi-particle state and achieving the
eigenvalue of the transfer matrix and corresponding Bethe equations is presented in detail.
PACS: 05.50.+q; 75.10.Hk; 75.10.Jm
Keywords: Osp(2|2) model; Algebraic Bethe ansatz
1. Introduction
The exactly solvable models [1] have an important application in condensed matter
physics. For example, the t–J model and Hubbard model can be used to describe the
strongly correlated electrons system. In order to investigate the thermodynamic properties
of an integrable model, such as specific heat, boundary effects, we should solve the model
at first.
In the procedure of solving an integrable model, the method often used is the algebraic
Bethe ansatz (ABA) [1,2]. The ABA can be easily applied to the six-vertex type model,
however, to the non-six-vertex type model, it is proved to be rather complex. On the base
of the Tarasov’s work [3], Martins and Ramos successfully solved some non-six-vertex
type models with period boundary conditions by using the ABA [4–6]. In their work, they
E-mail address: leegl@mail.xjtu.edu.cn (G.-L. Li).
doi:10.1016/j.nuclphysb.2004.03.022
G.-L. Li et al. / Nuclear Physics B 687 [FS] (2004) 220–256 221
throw the light on how the ABA works in dealing with this kinds of models. After their
work, some attempts have been made to solve the non-six-vertex type models with open
boundary conditions by the ABA [7–10].
Much work has been done on the exact solution of the integrable open boundary models
[11–30]. However, only a few of non-six-vertex type models are solved by using the BAB.
In Ref. [7], the Izergin–Korepin (IK) model with open boundary conditions is solved by the
means of the ABA. The procedure of solving the IK model which include constructing the
multi-particle state and achieving the eigenvalue of the transfer matrix and corresponding
Bethe equations is presented. The conclusions for the multi-particle state are proved based
on a reasonable assumption. In Refs. [8–10], the general multi-particle states and the
eigenvalue of the transfer matrix and the corresponding Bethe equations for the open
boundary Hubbard-like models are proposed with the help of the ABA. In the view of the
work of Refs. [8–10], here we are going to solve the Osp(2|2) model, one of Hubbard-like
models, by using ABA under the open boundary conditions.
The Osp(2|2) model [31,32], has been solved by the means of the analytical Bethe
ansatz under the period boundary conditions in Ref. [32]. The model with open boundary
conditions has not been solved by using the ABA so far. In this paper, we will present
the procedure of constructing the multi-particle state and achieving the eigenvalue of the
transfer matrix and corresponding Bethe equations for the model in detail by using the
ABA. Our results show that the ABA also can be successfully applied to the Hubbard-like
models with the open boundary conditions.
The paper is organized as follow. In Section 2 we introduce graded Osp(2|2) model
and give four diagonal K± matrices of graded reflection equation which determines the
nontrivial boundary terms in the Hamiltonian. In Section 3, we solve the model by using
the ABA. The n-particle states is constructed explicitly and the eigenvalue and the Bethe
equations for the transfer matrix are derived in detail. The summary and some discussions
of our main results are included in Section 4. In Appendices A–D, some coefficients and
necessary calculations are provided.
2. The vertex model and integrable boundary conditions
The R-matrix with f bbf grading for the Osp(2|2) model [32] is
 
−f1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 f3 0 0 f6 0 0 0 0 0 0 0 0 0 0 0 
 0 0 f3 0 0 0 0 0 f6 0 0 0 0 0 0 0 
 0 −f2 −f9 −f10 −f8 0 
 0 0 0 0 0 0 0 0 0 0 
 0 f6 0 0 f3 0 0 0 0 0 0 0 0 0 0 0 
 0 0 
 0 0 0 0 f4 0 0 0 0 0 0 0 0 0 
 0 0 0 f9 0 0 f5 0 0 f7 0 0 f9 0 0 0 
 0 0 
R(u) =  
0 0 0 0 0 0 f3 0 0 0 0 0 f6 0
 0 0 f6 0 0 0 0 0 f3 0 0 0 0 0 0 0  (1)
 0 0 
 0 0 f10 0 0 f7 0 0 f5 0 0 f10 0 0 
 0 0 0 0 0 0 0 0 0 0 f4 0 0 0 0 0 
 0 
 0 0 0 0 0 0 0 0 0 0 f3 0 0 f6 0 
 0 0 0 −f8 0 0 −f9 0 0 −f10 0 0 −f2 0 0 0 
 0 
 0 0 0 0 0 0 f6 0 0 0 0 0 f3 0 0 
0 0 0 0 0 0 0 0 0 0 0 f6 0 0 f3 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 −f1
222 G.-L. Li et al. / Nuclear Physics B 687 [FS] (2004) 220–256
with
f1 (u) = (1 − u)(1 + u),
f2 (u) = −u(2 + u),
f3 (u) = u(1 + u),
f4 (u) = (1 + u)2 ,
f5 (u) = u2 ,
f6 (u) = 1 + u,
f7 (u) = 1 + 2u,
f8 (u) = 1,
f9 (u) = −u,
f10 (u) = u. (2)
The graded R-matrix satisfies the following properties
1
regularity: R12 (0) = ρ(0) 2 P12 ,
unitarity: R12 (u)R21 (−u) = ρ(u),
st1 st1
crossing-unitarity: R12 (u)R21 (−u − 2) = ρ(1 + u),
ab = (−1)p(a)+p(b)δ δ .
where P is the graded exchange operator defined by Pcd ad bc
The R-matrix also fulfill the graded Yang–Baxter equation (YBE) [1]
R12 (u − v)R13 (u)R23 (v) = R23 (v)R13 (u)R12 (u − v), (3)
R12 (u) = R(u) ⊗s 1, R23 (u) = 1 ⊗s R(u), etc. The ⊗s denotes the graded tensor product
(A ⊗s B)ab
cd = (−1)
(p(a)+p(c))p(b) a b
Ac Bd ,
the p(i) is the Grassmann parities with p(1) = p(4) = 1, p(2) = p(3) = 0, R21 = P12 ×
j
R12 P12 , sti denotes super-transposition in the space i with (Ast )ij = (−1)p(j )(p(i)+p(j ))Ai ,
ρ(u) = f1 (u)f1 (−u).
For an N × N square lattice, if we can find K± (u) which satisfy the following reflection
equations
1 2 2 1
R12 (u − v)K− (u)R21 (u + v)K− (v) = K− (v)R12 (u + v)K− (u)R21 (u − v), (4)
st1 ,st¯ 2 1 st1 ,st¯ 2 2 ¯
R21 (−u + v)K st 1
+ (u)R12 (−u − v − 2)K st 2
+ (v)
2 ¯ st1 ,st¯ 2 1 st1 st¯ 2
= K st 2
+ (v)R21 (−u − v − 2)K st 1
+ (u)R12 (−u + v), (5)
where Eq. (4) is called the reflection equation and Eq. (5) is called the dual reflection
1 2
equation, K ± (u) = K± (u) ⊗s 1, K ± (u) = 1 ⊗s K± (u), st¯ i stands for the inverse of super-
transposition of sti ,
st st¯ st¯ st
A = A = A,
then the transfer matrix t (u) defined as

t (u) = str K+ (u)U (u) (6)
satisfies [t (u), t (v)] = 0, by which we can obtain an one-parameter commutative family.
Here
U (u) = T (u)K− (u)T −1 (−u), (7)
T (u) = R01 (u)R02 · · · R0N (u), (8)
the space V0 is usually called the auxiliary space, the space V1 ⊗s V2 · · · ⊗s VN is called
the quantum space. The corresponding integrable open chain Hamiltonian takes the form
0

N−1
1 1 str K + (0)HN,0
H= Hk,k+1 + K − (0) + , (9)
2 str K+ (0)
k=1

where Hk,k+1 = Pk,k+1 Rkk+1 (u)|u=0 .
From Eqs. (4) and (5), we can see that, given a solution K− (u) of Eq. (4), the matrix
K+ (u) = K−
st
(−u − 1) (10)
satisfies Eq. (5).
Solving Eq. (4), we obtain (here we omit the procedure)
K− (u) = diag(1, 1, 1, 1), (11)
K− (u) = diag(1 − u, 1 + u, 1 + u, 1 − u), (12)

K− (u) = diag ζ − − u, ζ − − u, ζ − + u, ζ − + u , (13)

K− (u) = diag ζ − − u, ζ − + u, ζ − − u, ζ − + u . (14)
Then, by Eq. (10) we have
K+ (u) = diag(1, 1, 1, 1), (15)
K+ (u) = diag(2 + u, −u, −u, 2 + u), (16)

K+ (u) = diag ζ + + 1 + u, ζ + + 1 + u, ζ + − 1 − u, ζ + − 1 − u , (17)

K+ (u) = diag ζ + + 1 − u, ζ + − 1 − u, ζ + + 1 + u, ζ + − 1 − u , (18)
where ζ − and ζ + are arbitrary parameters.
3. The algebraic Bethe ansatz
3.1. The vacuum state
Firstly, we write the double-monodromy matrix (7) as

 
A(u) B1 (u) B2 (u) F (u)
 D (u) A11 (u) A12 (u) E1 (u) 
U (u) =  1 . (19)
D2 (u) A21 (u) A22 (u) E2 (u)
G(u) C1 (u) C2 (u) A2 (u)
With the help of Eqs. (3), (4), we can prove that U (u) Eq. (19) satisfy the following
equation
1 2 2 1
R12 (u − v)U (u)R21 (u + v)U (v) = U (v)R12 (u + v)U (u)R21 (u − v), (20)
1 2
where U (u) = U (u) ⊗s 1, U (u) = 1 ⊗s U (u). Now we introduce the vacuum state
⊗
sN
|0 = (1, 0, 0, 0)t , (21)
where t denotes the transposition. Acting the double-row monodromy matrix (19) on the
vacuum state, we can find
Di (u)|0 = 0, Ci (u)|0 = 0, G(u)|0 = 0 (i = 1, 2),
Bi (u)|0 = 0, Ei (u)|0 = 0, F (u)|0 = 0,
A12 (u)|0 = 0, A21(u)|0 = 0. (22)
Considering the definition of U (u) Eq. (7), we have
A(u)|0 = T (u)11 K− (u)1 T −1 (−u)11 |0 + T (u)12 K− (u)2 T −1 (−u)21 |0
+ T (u)13 K − (u)3 T −1 (−u)31 |0 + T (u)14 K − (u)4 T −1 (−u)41 |0, (23)
A11 (u)|0 = T (u)21 K− (u)1 T −1 (−u)12 |0 + T (u)22 K− (u)2 T −1 (−u)22 |0
+ T (u)23 K − (u)3 T −1 (−u)32 |0 + T (u)24 K − (u)4 T −1 (−u)42 |0, (24)
A22 (u)|0 = T (u)31 K− (u)1 T −1 (−u)13 |0 + T (u)32 K− (u)2 T −1 (−u)23 |0
+ T (u)33 K − (u)3 T −1 (−u)33 |0 + T (u)34 K − (u)4 T −1 (−u)43 |0, (25)
A2 (u)|0 = T (u)41 K− (u)1 T −1 (−u)14 |0 + T (u)42 K− (u)2 T −1 (−u)24 |0
+ T (u)43 K − (u)3 T −1 (−u)34 |0 + T (u)44 K − (u)4 T −1 (−u)44 |0. (26)
In above equations, the first term of Eqs. (24), (25) and the previous three terms of
Eq. (26) cannot be calculated directly but it can be worked out by using the following
method. Taking v = −u in the Yang–Baxter equation, we can get
T2−1 (−u)R12 (2u)T1 (u) = T1 (u)R12 (2u)T2−1 (−u). (27)
Taking special indices in this relation and applying both sides of this relation to the vacuum
state, we find

T (u)i1 T −1 (−u)1i |0 = f˜1 (u) T (u)ii T −1 (−u)ii − T −1 (−u)11 T (u)11 |0,

T (u)4i T −1 (−u)i |0 = f˜3 (u) T −1 (−u)i T (u)i − T (u)44 T −1 (−u)44 |0,
4 i i
T (u)41 T −1 (−u)14 |0

= f˜2 (u) T −1 (−u)11 T (u)11 − T (u)44 T −1 (−u)44

3
−1
+ f˜1 (u)f˜3 (u) T (−u)i T (u)i − T (u)4 T (−u)4 |0,
i i 4 −1 4
(28)
i=2
where i = 2, 3,
f6 (2u) f8 (2u)
f˜1 (u) = − , f˜2 (u) = ,
f1 (2u) f1 (2u)
f6 (2u)f8 (2u) − f6 (2u)f1 (2u)
f˜3 (u) = .
2f6 (2u)f6 (2u) − f1 (2u)(f4 (2u) + f7 (2u))
After defining two new operators
Ãab (u) = Aab (u) + f˜1 (u)A(u)δab , (29)

2
Ã2 (u) = A2 (u) − f˜2 (u)A(u) − f˜3 (u) Ãaa (u), (30)
a=1
and substituting Eq. (28) to Eq. (23)–(26), we obtain
2N
A(u)|0 = K− (u)1 −f1 (u) ρ(u)−N |0 = ω1 (u)|0, (31)

Ãaa (u)|0 = K− (u)a+1 + f˜1 (u)K− (u)1 f3 (u)2N ρ(u)−N |0
= ka− (u)ω2 (u)|0, (32)
Ã2 (u)|0

2

= K− (u)4 − f˜3 (u) K− (u)a+1 + f˜4 (u)K− (u)1 −f2 (u)2N ρ(u)−N |0
a=1
= ω3 (u)|0, (33)
where f˜4 (u) = −f˜2 (u) − 2f˜1 (u)f˜3 (u), k (u) = diag[k1 (u), k2 (u)] is a 2 × 2 diagonal
− − −
matrix. For the case of Eq. (11),

2N 2u 2N
ω1 (u) = −f1 (u) ρ(u)−N , ω2 (u) = f3 (u) ρ(u)−N ,
2u − 1
4u(1 + u) 2N −N
ω3 (u) = −f2 (u) ρ(u) ,
(1 + 2u)(3 + 2u)
k − (u) = diag[1, 1]. (34)
For the case of Eq. (12),
2N 2u2 2N
ω1 (u) = (1 − u) −f1 (u) ρ(u)−N , ω2 (u) = f3 (u) ρ(u)−N ,
2u − 1
4u2 (2 + u) 2N
ω3 (u) = − −f2 (u) ρ(u)−N ,
(1 + 2u)(3 + 2u)
−
k (u) = diag[1, 1]. (35)
2N 2u 2N
ω1 (u) = ζ − − u −f1 (u) ρ(u)−N , ω2 (u) = f3 (u) ρ(u)−N ,
2u − 1
4u(1 + u) 2N
ω3 (u) = ζ − + 1 + u −f2 (u) ρ(u)−N ,
(1 + 2u)(3 + 2u)

k − (u) = diag ζ − − u, ζ − − 1 + u . (36)

2N 2u 2N
ω1 (u) = ζ − − u −f1 (u) ρ(u)−N , ω2 (u) = f3 (u) ρ(u)−N ,
2u − 1
4u(1 + u) 2N
ω3 (u) = ζ − + 1 + u −f2 (u) ρ(u)−N ,
(1 + 2u)(3 + 2u)
−
k (u) = diag ζ − 1 + u, ζ − − u .
−
(37)
Considering Eqs. (29), (30), the transfer matrix (6) can be rewritten as

2
t (u) = w1 (u)A(u) + wa+1 (u)Ãaa (u) + w4 (u)Ã2 (u), (38)
a=1
with

2
w1 (u) = −K+ (u)1 − f˜1 (u) K+ (u)a+1 − f˜2 (u)K+ (u)4 ,
a=1
wa+1 (u) = K+ (u)a+1 − f˜3 (u)K+ (u)4 = w̃(u)ka+ (u),
w4 (u) = −K+ (u)4 , (39)
where k + (u) = diag[k1+ (u), k2+ (u)] is a 2 × 2 diagonal matrix. For the case of Eq. (15),
4u(1 + u)
w1 (u) = , w4 (u) = −1,
1 − 4u2
2(1 + u)
w̃(u) = , k + (u) = diag[1, 1]. (40)
(3 + 2u)
4(u − 1)(1 + u)2
w1 (u) = , w4 (u) = −2 − u,
1 − 4u2
2(1 + u)2
w̃(u) = − , k + (u) = diag[1, 1]. (41)
(3 + 2u)
4u(1 + u)(ζ + + u)
w1 (u) = , w4 (u) = 1 + u − ζ + ,
1 − 4u2
2(1 + u)
w̃(u) = , k + (u) = diag ζ + + 2 + u, ζ + − 1 − u . (42)
(3 + 2u)
4u(1 + u)(ζ + + u)
w1 (u) = , w4 (u) = 1 + u − ζ + ,
1 − 4u2
2(1 + u)
w̃(u) = , k + (u) = diag ζ + − 1 − u, ζ + + 2 + u . (43)
(3 + 2u)
3.2. The fundamental commutation relations
Rewriting Eq. (20) in component form

j k j
R(u− )ij11ij22 U (u)k11 R(u+ )l11k22 U (v)kl22 (−1)(p(j1 )+p(k1 ))p(j2 )
i j j
= U (v)ij22 R(u+ )j11 k22 U (u)k11 R(u− )kl11lk22 (−1)(p(j1)+p(k1 ))p(k2) , (44)
where the repeated indices sum over 1 to 4, u− = u − v, u+ = u + v, we can obtain the
following fundamental commutation relations
Ba (u)Bb (v) + δāb ξa g12 (u, v)F (u)A(v) + ξa g22 (u, v)F (u)Ãāb (v)

= r̂(u− )cd
ab Bd (v)Bc (u) + δd̄c ξd g1 (v, u)F (v)A(u) + ξd g2 (v, u)F (v)Ãd̄ c (u) ,
2 2
(45)
A(u)Ba (v)
= a11(u, v)Ba (v)A(u) + a21 (u, v)Ba (u)A(v) + a31 (u, v)Bd (u)Ãda (v)
+ ξa a41 (u, v)F (u)Dā (v) + a51 (u, v)F (u)Ca (v) + ξa a61 (u, v)F (v)Dā (u), (46)
Ãab (u)Bc (v)
fg fa
dg r̄(u− )bc Be (v)Ãdf (u) + R1 (u, v)bc Bf (u)A(v)
= r̃(u+ )ae A
fa
+ R2A (u, v)bd Bf (u)Ãdc (v) + δb̄c ξb R3A (u, v)Ea (u)A(v)
fa
+ ξb R4A (u, v)Ea (u)Ãb̄c (v) + ξf R5A (u, v)bc F (u)Df¯ (v)
fa
+ δab R6A (u, v)F (u)Cc (v) + ξf R7A (u, v)bc F (v)Df¯ (u)
fa
+ R8A (u, v)bc F (v)Cf (u), (47)
Ã2 (u)Ba (v)
= a13(u, v)Ba (v)Ã2 (u) + a23 (u, v)Ba (u)A(v) + a33 (u, v)Bd (u)Ãda (v)
+ ξā a43 (u, v)Eā (u)A(v) + ξd a53 (u, v)Ed (u)Ãd̄a (v) + ξa a63 (u, v)F (u)Dā (v)
+ a73 (u, v)F (u)Ca (v) + ξa a83 (u, v)F (v)Dā (u) + a93 (u, v)F (v)Ca (u), (48)
A(u)F (v)
= b11(u, v)F (v)A(u) + b21 (u, v)F (u)A(v) + b31 (u, v)F (u)Ãdd (v)
+ b41 (u, v)F (u)Ã2 (v) + ξd b51(u, v)Bd̄ (u)Bd (v) + b61 (u, v)Bd (u)Ed (v), (49)
Ãab (u)F (v)
= b12(u, v)F (v)Ãab (u) + δab b22 (u, v)F (u)A(v) + R1F (u, v)āb
cd
F (u)Ãd̄c (v)
+ δab b32 (u, v)F (u)Ã2 (v) + ξa R2F (u, v)cd
āb Bd (u)Bc (v)
+ R3F (u, v)ca
bd Bc (u)Ed (v) + b4 (u, v)Ea (u)Bb (v)
2
+ ξb b52(u, v)Ea (u)Eb̄ (v), (50)

Ã2 (u)F (v)
= b13(u, v)F (v)Ã2 (u) + b23 (u, v)F (u)A(v) + b33 (u, v)F (u)Ãdd (v)
+ b43 (u, v)F (u)Ã2 (v) + ξd b53(u, v)Bd̄ (u)Bd (v) + b63 (u, v)Bd (u)Ed (v)
+ b73 (u, v)Ed (u)Bd (v) + ξd b83 (u, v)Ed (u)Ed̄ (v), (51)
Da (u)Bb (v)
db Bc (v)Dd (u) + ξa c1 (u, v)Bā (v)Cb (u) + δab c2 (u, v)F (v)G(u)
= R1D (u, v)ac 1 1
+ ξa c31 (u, v)Bā (u)Cb (v) + c41 (u, v)Ea (u)Cb (v) + δab c51 (u, v)A(v)A(u)
+ c61 (u, v)A(v)Ãab (u) + δab c71 (u, v)A(u)A(v) + c81 (u, v)A(u)Ãab (v)
+ c91 (u, v)Ãab (u)A(v) + c10
1
(u, v)Ãad (u)Ãdb (v), (52)
Ca (u)Bb (v)
ab Bd (v)Cc (u) + ξa R2 (u, v)db Bc (v)Dd (u) + δāb ξa c1 (u, v)F (v)G(u)
= R1C (u, v)cd C āc 2
+ c22 (u, v)Ba (u)Cb (v) + c32 (u, v)Eā (u)Cb (v) + δa b̄ ξa c42 (u, v)A(v)A(u)
ab A(v)Ãd̄c (u) + δa b̄ ξa c5 (u, v)A(v)Ã2 (u)

+ ξa R3C (u, v)cd 2
+ δa b̄ ξa c62 (u, v)A(u)A(v) + ξa c72 (u, v)A(u)Ãāb (v)

+ ξa R4C (u, v)cd
ab Ãd̄c (u)A(v) + ξa R5 (u, v)ae Ãd̄c (u)Ãeb (v)
C cd
+ δa b̄ ξa c82 (u, v)Ã2 (u)A(v) + ξa c92 (u, v)Ã2 (u)Ãāb (v), (53)
Ba (u)Eb (v)
db Ec (v)Bd (u) + ξb R2 (u, v)a b̄ Bd (v)Bc (u) + δab e1 (u, v)F (v)A(u)
= R1be (u, v)ac be cd 1
+ R3be (u, v)cd

a b̄
F (v)Ãd̄c (u) + δab e21 (u, v)F (u)A(v) + R4be (u, v)ac
db F (u)Ãcd (v)
+ δab e31 (u, v)F (u)Ã2 (v), (54)

F (u)Ba (v)
= d11 (u, v)Ba (v)F (u) + d21 (u, v)F (v)Ba (u) + ξa d31 (u, v)F (v)Eā (u), (55)
Ba (u)F (v)
= d12 (u, v)F (v)Ba (u) + ξa d22 (u, v)F (v)Eā (u) + d32 (u, v)Ba (v)F (u)
+ ξa d42 (u, v)Eā (v)F (u), (56)
F (u)Ea (v)
= d13 (u, v)Ea (v)F (u) + ξa d23 (u, v)Bā (v)F (u) + ξa d33 (u, v)F (v)Bā (u)
+ d43 (u, v)F (v)Ea (u), (57)
Ea (u)F (v)
= d14 (u, v)F (v)Ea (u) + ξa d24 (u, v)Bā (v)F (u) + d34 (u, v)Ea (v)F (u), (58)
F (u)F (v) = F (v)F (u), (59)
ab
where all the repeated indices sum over 1 to 2 and the coefficients which have the form Rcd
in Eqs. (46)–(54) are not zeros when a = b = c = d, or a = c = b = d or a = d = b = c.
The explicit expressions of these coefficients are presented in Appendix A. Here we define
ā = 3 − a,

1, a = 1,
ξa = (60)
−1, a = 2,
f10 (u− )f3 (2v) f10 (u+ )
g12 (u, v) = , g22 (u, v) = − . (61)
f2 (u− )f1 (2v) f3 (u+ )
The r̂(u), r̃(u) and r̄(u) are given by
f3 (u)f6 (u) f10 (u)
r̂(u) = − r(u), r̃(u) = − r(u − 1),
f1 (u)f2 (u) f1 (u)
f6 (u)
r̄(u) = − r(u). (62)
f2 (u)
The r(u) is a 6-vertex R-matrix
   
u+2 0 0 0 a(u) 0 0 0
 0 u 2 0   0 b(u) c(u) 0 
r(u) =  = . (63)
0 2 u 0 0 c(u) b(u) 0
0 0 0 u+2 0 0 0 a(u)
3.3. The n-particle state
Let
Φnb1 ···bn (v1 , . . . , vn )
b2 ···bn

n
d ···d
= Bb1 (v1 )Φn−1 (v2 , . . . , vn ) + F (v1 ) 3
Φn−2 n
(v2 , . . . , v̌i , . . . , vn )
i=2
d2 ···dn n−2
× Sb2 ···bn (vi ; {v̌1 , v̌i })Λ1 (vi ; {v̌1 , v̌i })g21 (v1 , vi )A(vi )δb̄1 d2 ξb1
n
d3 ···dn
···dn
+ F (v1 ) Φn−2 (v2 , . . . , v̌i , . . . , vn ) T̃ n−2 (vi ; {v̌1 , v̌i })dc33···cn b̄1 c2
i=2
c2 ···cn
× Sb2 ···bn (vi ; {v̌1 , v̌i })g22 (v1 , vi )ξb1 , (64)
where
···dn d c di ci−1

n
Sbd11···b n
(vi ; {v̌i }) = r̂bd12cd21 (v1 − vi )r̂b23c32 (v2 − vi ) · · · r̂bi−1 bi (vi−1 − vi ) δd j b j
j =i+1
(65)
and
···dn

T̃ n (u; {vm })dc11···c
ab
n
h dn
= r̃h1 g1 (u + v1 )r̃hh21gd22 (u + v2 ) · · · r̃hnn−1
ad1
gn (u + vn )Ãhn fn (u)
fn gn fn−1 gn−1 f1 g1
× r̄fn−1 cn (u − vn )r̄fn−2 cn−1 (u − vn−1 ) · · · r̄bc1 (u − v1 ) (66)
···dn
with Sbd11···b n
(v1 ; v2 , . . . , vn ) = δd1 b1 ,[T̃ 0 (u; v1 , . . . , vn )]ab = Ãab (u) and the following
notations are used
{vm } = {v1 , . . . , vn }, {v̌i } = {v1 , . . . , v̌i , . . . , vn },

{v̌i , v̌j } = {v1 , . . . , v̌i , . . . , v̌j , . . . , vn },

n
Λnl (u; v1 , v2 , . . . , vn ) = a1l (u, vi ) (l = 1, 3).
i=1
The the general n-particle state is constructed as

Υn (v1 , . . . , vn ) = Φ b1 ···bn (v1 , . . . , vn )F b1 ···bn |0. (67)
n
We can check that the Φnb1 ···bn (v1 , . . . , vn ) own the property as below
b ···bi bi+1 ···bn
Φn 1 (v1 , . . . , vi , vi+1 , . . . , vn )F b1 ···bn |0
b ···ai ai+1 ···bn a a b1 ···bn
= Φn 1 bi+1 (vi − vi+1 )F
(v1 , . . . , vi+1 , vi , . . . , vn )r̂bii+1 i
|0. (68)
It is easy to verify the above property Eq. (68) at the case i = 1. When i = 1 we will need
the following conclusions Eqs. (69), (70) and Eq. (C.1) in Appendix C and some necessary
commutation relations Eqs. (45), (54)–(57), (59).
Enlightened by the previous work [4–10], we can infer that

A(u)Υn (v1 , . . . , vn )

= ω1 (u)Λn (u; v1 , . . . , vn )Υn (v1 , . . . , vn )
1

n
(1) n
(2)
+ a21 (u, vi )Ψn−1 (u, vi ; {vm })d1 d1 + a31 (u, vi)Ψn−1 (u, vi ; {vm })dd
i=1 i=1

n−1
n
(5)
+ A
δd̄1 e2 ξd1 H1,d 1
(u, vi , vj )Ψn−2 (u, vi , vj ; {vm })d1 e2
i=1 j =i+1

n−1
n
ef (6) fe
+ A
ξd1 H2,d 1
(u, vi , vj )d1 c2 Ψn−2 (u, vi , vj ; {vm })d1 c2
i=1 j =i+1

n−1
n
(7)
+ A
ξd1 H3,d 1
(u, vi , vj )Ψn−2 (u, vi , vj ; {vm })d1 d1
i=1 j =i+1

n−1
n

ed (8)
+ A
ξd1 H4,d 1
(u, vi , vj ) d 1 f Ψn−2 (u, vi , vj ; {vm })ed
f d1 , (69)
i=1 j =i+1

Ãaa (u)Υn (v1 , . . . , vn )

= Φnd1 ···dn (v1 , . . . , vn ) T̃ n (u; {vm })db11 ···d n
···bn aa F
b1 ···bn
|0

n
(1)
+ R1A (u, vi )dad1 a1 Ψn−1 (u, vi ; {vm })d1 d1
i=1

n
(2) n
(3)
+ R2A (u, vi )da Ψ (u, vi ; {vm })dd + R3A (u, vi )Ψn−1 (u, vi ; {vm })aa
ad n−1
i=1 i=1

n
(4)
+ R4A (u, vi )Ψn−1 (u, vi ; {vm })aa
i=1

n−1
n
(5)
+ Aaa
δd̄1 e2 ξd1 H1,d 1
i=1 j =i+1

n−1
n
ef (6) fe
+ Aaa
ξd1 H2,d 1
i=1 j =i+1

n−1
n
(7)
+ Aaa
ξd1 H3,d 1
i=1 j =i+1

n−1
n
(8)
+ Aaa
d1 f Ψn−2 (u, vi , vj ; {vm })f d1 ,
(u, vi , vj )ed ed
ξd1 H4,d 1
(70)
i=1 j =i+1

Ã2 (u)Υn (v1 , . . . , vn )

= ω3 (u)Λn3 (u; v1 , . . . , vn )Υn (v1 , . . . , vn )

n
(1) n
(2)
+ a23 (u, vi )Ψn−1 (u, vi ; {vm })d1 d1 + a33 (u, vi)Ψn−1 (u, vi ; {vm })dd
i=1 i=1

n
(3) n
(4)
+ a43 (u, vi )Ψn−1 (u, vi ; {vm })d̄1 d̄1 + a53 (u, vi )Ψn−1 (u, vi ; {vm })dd
i=1 i=1

n−1
n
(5)
+ A2
δd̄1 e2 ξd1 H1,d 1
i=1 j =i+1

n−1
n
ef (6) fe
ξd1 H2,d2 1 (u, vi , vj )d1 c2 Ψn−2 (u, vi , vj ; {vm })d1 c2
A
+
i=1 j =i+1

n−1
n
(7)
+ A2
ξd1 H3,d 1
i=1 j =i+1

n−1
n
(8)
+ A2
d1 f Ψn−2 (u, vi , vj ; {vm })f d1 ,
(u, vi , vj )ed ed
ξd1 H4,d 1
(71)
i=1 j =i+1
where
(1)
Ψ (u, vi ; {vm })f d
n−1 1
d2 ···dn ···dn
= Bf (u)Φn−1 (v1 , . . . , v̌i , . . . , vn )Sbd11···b n
(vi ; {v̌i })
b1 ···bn
× Λn−1
1 (vi ; {v̌i })ω1 (vi )F |0, (72)
(2)
Ψ (u, vi ; {vm })f d
n−1
e2 ···en e ···e
= Bf (u)Φn−1 (v1 , . . . , v̌i , . . . , vn ) T̃ n−1 (vi ; {v̌i })d2 ···dn dd
2 n 1
d ···d
× Sb11···bnn (vi ; {v̌i })F b1 ···bn |0, (73)
(3)
Ψ (u, vi ; {vm })ab
n−1
d2 ···dn ···dn
= Ea (u)Φn−1 (v1 , . . . , v̌i , . . . , vn )Sbd11···b n
(vi ; {v̌i })
b1 ···bn
× Λn−1
1 (vi ; {v̌i })ω1 (vi )ξb δb̄d1 F |0, (74)
(4)
Ψ (u, vi ; {vm })ab
n−1
e2 ···en
= Ea (u)Φn−1 (v1 , . . . , v̌i , . . . , vn ) T̃ n−1 (vi ; {v̌i })ed2 ···en
···d b̄d
2 n 1
d ···d
× Sb11···bnn (vi ; {v̌i })ξb F b1 ···bn |0, (75)
(5)
Ψ (u, vi , vj ; {vm })d e
n−2 1 2
e ···e ···en ···dn
= F (u)Φn−2
3 n
(v1 , . . . , v̌i , . . . , v̌j , . . . , vn )Sde22 ···d n
(vj ; {v̌i , v̌j })Sbd11···b n
(vi ; {v̌i })
b1 ···bn
× Λn−2
1 (vi ; {v̌i , v̌j })Λ1 (vj ; {v̌i , v̌j })A(vi )A(vj )F
n−2
|0, (76)
(6)
Ψ (u, vi , vj ; {vm }) f e
n−2 d1 c2
e3 ···en
= F (u)Φn−2 (v1 , . . . , v̌i , . . . , v̌j , . . . , vn ) T̃ n−2 (vi ; {v̌i , v̌j })ec33 ···en
···cn f¯e
···cn ···dn
× Sdc22 ···d n
(vj ; {v̌i , v̌j })Sbd11···b n
(vi ; {v̌i })Λn−2
1 (vj ; {v̌i , v̌j })
× A(vj )F b1 ···bn |0, (77)
(7)
Ψ (u, vi , vj ; {vm })f d
n−2 1
e3 ···en
= F (u)Φn−2 (v1 , . . . , v̌i , . . . , v̌j , . . . , vn ) T̃ n−2 (vj ; {v̌i , v̌j })ec33 ···en
···cn f¯c 2
···cn ···dn b1 ···bn
× Sdc22 ···d n
(vj ; {v̌i , v̌j })Sbd11···b n
(vi ; {v̌i })Λn−2
1 (vi ; {v̌i , v̌j })A(vi )F |0, (78)
(8)
Ψ (u, vi , vj ; {vm })ed
n−2 f d1
e3 ···en
= F (u)Φn−2 (v1 , . . . , v̌i , . . . , v̌j , . . . , vn ) T̃ n−2 (vi ; {v̌i , v̌j })ea33···en
···an d̄e
···an
···cn
× T̃ n−2 (vj ; {v̌i , v̌j })ac33···c n fc
Sdc22 ···d n
(vj ; {v̌i , v̌j })
2
···dn
× Sbd11···b n
(vi ; {v̌i })F b1 ···bn |0. (79)
Here, the coefficients Hl,d x (u, v , v ) (x = A, A , A , l = 1, 2, 3, 4) are listed in
1
i j aa 2
Appendix B.
Our conclusions Eqs. (69)–(71) can be verified directly by using mathematical
induction, however, it is a rather hard work for us to do. Generally speaking, as what the
algebraic Bethe ansatz assumes, for {v1 , . . . , vm }, m n, we can think that such “quasi”
n-particle states like BΦn−1 |0, EΦn−1 |0, BBΦn−2 |0, BEΦn−2 |0, EBΦn−2 |0,
F Φn−2 |0, FBΦn−3 |0 and so on are linearly independent, here we omit all the indices and
all the spectrum parameters in the “quasi” n-particle state keep the order {vi1 , vi2 , . . . , vik }
b3 ···bn
with i1 < i2 < · · · < ik . For example, B1 (v1 )B1 (v2 )Φn−2 (v3 , . . . , vn )F 11b3 ···bn |0 and
b ···b
3
B1 (v1 )B2 (v2 )Φn−2 n
(v3 , . . . , vn )F 12b3 ···bn |0 are thought to be linearly independent.
Then, by using the assumption, the property of Eq. (68) and some necessary relations,
we can prove that our conclusions Eqs. (69)–(71) still hold at the case n + 1 as we have
done in Ref. [7].
In calculating the one and two-particle states, we verified the assumption. We also
checked some conclusions obtained by the assumption in calculating the n-particle state.
3.4. The eigenvalue and Bethe equations
By using some relations, we change the Eqs. (69)–(71) into (see Appendix D)

A(u)Υn (v1 , . . . , vn )

= ω1 (u)Λn (u; v1 , . . . , vn )Υn (v1 , · · · , vn )
1

n
(1) n
(2)
+ a21 (u, vi )Ψn−1 (u, vi ; {vm })d1 d1 + a31 (u, vi)Ψ̃n−1 (u, vi ; {vm })dd
i=1 i=1

n−1
n
(5)
+ A
ξd1 H̃1,d 1
(u, vi , vj )Ψ̃n−2 (u, vi , vj ; {vm })d1
i=1 j =i+1

n−1
n
(6)
+ A
ξd1 H̃2,d (u, vi , vj ) Ψ̃ (u, vi , vj ; {vm })d
1 n−2 1
i=1 j =i+1

n−1
n
(7)
+ A
ξd1 H̃3,d 1
i=1 j =i+1

n−1
n
(8)
+ A
ξd1 H̃4,d 1
(u, vi , vj )Ψ̃n−2 (u, vi , vj ; {vm })d1 , (80)
i=1 j =i+1

Ãaa (u)Υn (v1 , . . . , vn )

= ω2 (u)Λn2 (u; {vm })Φnd1 ···dn (v1 , . . . , vn ) T n (u; {vm })db11 ···d n
···bn aa F
b1 ···bn
|0

n
(1)
+ R1A (u, vi )dad1 a1 Ψn−1 (u, vi ; {vm })d1 d1
i=1
n
(2) n
(3)
+ R2A (u, vi )da Ψ̃ (u, vi ; {vm })dd + R3A (u, vi )Ψn−1 (u, vi ; {vm })aa
ad n−1
i=1 i=1

n
(4)
+ R4A (u, vi )Ψ̃n−1 (u, vi ; {vm })aa
i=1

n−1
n
(5)
+ Aaa
ξd1 H̃1,d 1
i=1 j =i+1

n−1
n
(6)
+ Aaa
ξd1 H̃2,d (u, vi , vj ) Ψ̃ (u, vi , vj ; {vm })d
1 n−2 1
i=1 j =i+1

n−1
n
(7)
+ Aaa
ξd1 H̃3,d 1
i=1 j =i+1

n−1
n
(8)
+ Aaa
ξd1 H̃4,d (u, vi , vj ) Ψ̃ (u, vi , vj ; {vm })d , (81)
1 n−2 1
i=1 j =i+1

Ã2 (u)Υn (v1 , . . . , vn )

= ω3 (u)Λn3 (u; v1 , . . . , vn )Υn (v1 , . . . , vn )

n
(1) n
(2)
+ a23 (u, vi )Ψn−1 (u, vi ; {vm })d1 d1 + a33 (u, vi)Ψ̃n−1 (u, vi ; {vm })dd
i=1 i=1

n
(3) n
(4)
+
a4 (u, vi ) Ψn−1 (u, vi ; {vm })d̄1 d̄1 +
3
a53 (u, vi )Ψ̃n−1 (u, vi ; {vm })dd
i=1 i=1

n−1
n
(5)
+ A2
ξd1 H̃1,d 1
i=1 j =i+1

n−1
n
(6)
+ A2
ξd1 H̃2,d 1
i=1 j =i+1

n−1
n
(7)
+ A2
ξd1 H̃3,d 1
i=1 j =i+1

n−1
n
(8)
+ A2
ξd1 H̃4,d 1
(u, vi , vj )Ψ̃n−2 (u, vi , vj ; {vm })d1 , (82)
i=1 j =i+1
x (j = 1, 2, 3, 4) are given in Appendix D,

where the coefficients H̃j,d 1

T̃ n (u; {vm })db11 ···d
F b1 ···bn |0
···bn
n
ab

= ω2 (u)Λn2 (u; {vm }) T n (u; {vm })db11 ···d n
···bn ab F
b1 ···bn
|0, (83)

n
Λn2 (u; {vm }) = ρ1 (u − vi )ρ2 (u, vi ),
i=1
f10 (u + v)f6 (u − v)
ρ1 (u) = a(u)a(−u), ρ2 (u, v) = ,
f1 (u + v)f2 (u − v)
n ···dn

T (u; {vm })dc11···cn ab
n−1 n −
h d
= L(ũ, ṽ1 )ad h1 d2
h1 g1 L(ũ, ṽ2 )h2 g2 · · · L(ũ, ṽn )hn gn khn fn (u)
1
× L−1 (−ũ, ṽn )fnn−1n cn L−1 (−ũ, ṽn−1 )fn−1 · · · L−1 (−ũ, ṽ1 )bc11 1
f g n−1 f g f g
n−2 cn−1
(84)
with ũ = u − 12 , v˜i = vi − 12 , and

(2)
Ψ̃ (u, vi ; {vm })f d
n−1
1
ρ 2 (0)ω2 (vi ) n−1 e2 ···en
= 1 (d) Λ2 (vi ; {v̌i })Bf (u)Φn−1 (v1 , . . . , v̌i , . . . , vn )
T (vi )
2 ···en
× Sdde1 ···d n
(vi ; {v̌i })τ1n (vi ; {vm })db11 ···d n b1 ···bn
···bn F |0, (85)
(4)
Ψ̃ (u, vi ; {vm })ab
n−1
1
ρ12 (0)ω2 (vi ) e2 ···en
= Λn−1
2 (vi ; {v̌i })Ea (u)|Φn−1 (v1 , . . . , v̌i , . . . , vn )
T (b̄) (vi )
2 ···en
× Sdb̄e1 ···d n
(vi ; {v̌i })τ1n (vi ; {vm })db11 ···d n
···bn ξb F
b1 ···bn
|0, (86)
(5)
Ψ̃ (u, vi , vj ; {vm })d
n−2 1
e ···e d̄ e ···e ···dn

= F (u)Φn−2
3 n
(v1 , . . . , v̌i , . . . , v̌j , . . . , vn )Sd21···d
3
n
n
(vj ; {v̌i , v̌j })Sbd11···b n
(vi ; {v̌i })
b1 ···bn
1 (vi ; {v̌i })Λ1 (vj ; {v̌j })ω1 (vi )ω1 (vj )F
× Λn−1 n−1
|0, (87)
(6)
Ψ̃ (u, vi , vj ; {vm })d
n−2 1
e ···e
= F (u)Φn−2
3 n
(v1 , . . . , v̌i , . . . , v̌j , . . . , vn )
e3 ···en
× Sdd̄21···d n
···dn
(vj ; {v̌i , v̌j })Scd11···cn
(vi ; {v̌i })τ1n (vi ; {ṽm })cb11···cn
···bn
1
b1 ···bn
× Λn−1 2 n−1
2 (vi ; {v̌i })ω2 (vi )ρ1 (0)Λ1 (vj ; {v̌j })ω1 (vj )F |0, (88)
(7)
Ψ̃ (u, vi , vj ; {vm })d
n−2 1
e ···e
= F (u)Φn−2
3 n
(v1 , . . . , v̌i , . . . , v̌j , . . . , vn )
d̄ e ···e ···dn c ···c
× Sd21···d
3
n
n
(vj ; {v̌i , v̌j })Scd11···cn
(vi ; {v̌i })τ1n (vj ; {ṽm })b11 ···bnn
1
b1 ···bn
× Λn−1 2 n−1
2 (vj ; {v̌j })ω2 (vj )ρ1 (0)Λ1 (vi ; {v̌i })ω1 (vi )F |0, (89)
(8)
Ψ̃ (u, vi , vj ; {vm })d
n−2 1
e ···e d̄ e ···e ···dn

= F (u)Φn−2
3 n
(v1 , . . . , v̌i , . . . , v̌j , . . . , vn )Sd21···d
3
n
n
(vj ; {v̌i , v̌j })Scd11···cn
(vi ; {v̌i })
a1 ···an n−1
× τ1n (vi ; {ṽm })ca11···c n−1
···an τ1 (vj ; {ṽm })b1 ···bn Λ2 (vi ; {v̌i })Λ2 (vj ; {v̌j })
n n
× ω2 (vi )ω2 (vj )ρ1 (0)F b1 ···bn |0. (90)

In Eqs. (85)–(90), we define

2
T (d1 ) (vi ) = kd+ (vi )r(2vi − 1)dd 1
d1 d ,
d=1
L(ũ, ṽ)ab
cd = rcd ab
(ũ + ṽ),
ab
rcd (ũ − ṽ)
L−1 (−ũ, ṽ)ab = ,
cd
ρ1 (ũ − ṽ)
···dn
τ1n (u; {ṽm })dc11···cn
n−1 n −
= ka+ (u)L(ũ, ṽ1 )ad h1 d2 h d
h1 g1 L(ũ, ṽ2 )h2 g2 · · · L(ũ, ṽn )hn gn khn fn (u)
1
× L−1 (−ũ, ṽn )fnn−1n cn L−1 (−ũ, ṽn−1 )fn−1 · · · L−1 (−ũ, ṽ1 )ac1 1 1 .
f g f
n−1 g f g
n−2 cn−1
(91)
Using Eqs. (69)–(71) and Eq. (38), we get

t (u)Υn (v1 , . . . , vn )

= w1 (u)ω1 (u)Λn (u; v1 , . . . , vn )Υn (v1 , . . . , vn )
1
+ w̃(u)ω2 (u)Λn2 (u; v1 , . . . , vn )Φnd1 ···dn (v1 , . . . , vn )τ1n (u; {ṽm })db11 ···d n b1 ···bn
···bn F |0

+ w4 (u)ω3 (u)Λn3 (u; v1 , . . . , vn )Υn (v1 , . . . , vn ) + u.t., (92)
where u.t. denotes the unwanted terms. If
τ1n (u; {ṽm })F b1 ···bn = Λ̃n1 (u; v1 , . . . , vn )F b1 ···bn (93)

and
−1 ω1 (vi )Λn−1
1 (vi ; {v̌i })
Λ̃n1 (vi ; v1 , . . . , vn ) = −ρ1 2 (0) β(vi ) (i = 1, . . . , n) (94)
ω2 (vi )Λn−1
2 (vi ; {v̌i })
we can get

t (u)Υn (v1 , . . . , vn )

= w1 (u)ω1 (u)Λn1 (u; v1 , . . . , vn )
+ w̃(u)ω2 (u)Λn2 (u; v1 , . . . , vn )Λ̃n1 (u; v1 , . . . , vn )

+ w4 (u)ω3 (u)Λn3 (u; v1 , . . . , vn ) Υn (v1 , . . . , vn )

= Λn (u; v1 , . . . , vn )Υn (v1 , . . . , vn ) , (95)
and easily check that all the unwanted terms vanish. Where
 2vi (3+2vi )

 , for Eq. (15),
 1−2vi
2(vi −1)(3+2vi )
β(vi ) = 2vi −1 , for Eq. (16), (96)


 2vi (ζ + +vi )(3+2vi )
1−2vi , for Eqs. (17), (18)
and all the unwanted terms are canceled out by the following three identities
d1 d
w1 (u)a21 (u, vi ) + 2d=1 wd+1 (u)R1A (u, vi )dd + w4 (u)a23 (u, vi )
β(vi ) = T (d1 )
(vi )
1
d1 d
,
w1 (u)a31 (u, vi ) + 2d=1 wd+1 (u)R2A (u, vi )dd 1
+ w4 (u)a 3 (u, v )
3 i
(97)
wd̄ +1 (u)R3 (u, vi ) + w4 (u)a4 (u, vi )
A 3
β(vi ) = T (d1 ) (vi ) 1 , (98)
wd̄1 +1 (u)R4A (u, vi ) + w4 (u)a53(u, vi )
4
4
xl
xl
wl (u)H̃1,d1 (u, vi , vj ) − wl (u)H̃2,d 1
(u, vi , vj ) β(vi )
l=1 l=1

4
x
− l
wl (u)H̃3,d 1
(u, vi , vj ) β(vj )
l=1

4
xl
+ wl (u)H̃4,d 1
(u, vi , vj ) β(vi )β(vj ) = 0 (99)
l=1
with x1 = A, x2 = A11 , x3 = A22 , x4 = A2 . Referring the conclusion of Ref. [12], we have
Λ̃n1 (u; v1 , . . . , vn )
(1) (1)

n1 (1) (1)
a(v − ũ)b(v + ũ)
= w1 (ũ)ω1 (ũ, {ṽm }) (1)
l
(1)
l
l=1 b(vl − ũ)a(vl + ũ)

n1
a(ũ − v (1) )a(v (1) + ũ + 2)
+ w2(1) (ũ)ω2(1) (ũ, {ṽm }) l
(1)
l
(1)
(100)
l=1 b(ũ − vl )a(ũ + vl )
with Bethe equations
ω2(1) (vj(1) , {ṽm })
(1) (1)
n1
a(vl(1) − vj(1) )b(vl(1) + vj(1) )
(1) (1)
= β vj (−1) (1) (1) (1) (1)
ω1 (vj , {ṽm }) l=1=j a(vj − vl )a(vj + vl + 2)
(j = 1, . . . , n1 ), (101)
where
c(2ũ) +
w1(1) (ũ) = k1+ (u) + k (u), w2(1) (ũ) = k2+ (u), (102)
a(2ũ) 2
n
ω1(1) (ũ, {ṽm }) = k1− (u) a(ũ + ṽi )a(ũ − ṽi )ρ1−1 (ũ − ṽi ),
i=1
! "
n
c(2ũ) −
k2− (u) − b(ũ + ṽi )b(ũ − ṽi )ρ1−1 (ũ − ṽi ), (103)
(1)
ω2 (ũ, {ṽm }) = k1 (u)
a(2ũ)
i=1
(1) (1) (1) (1)
b(2vj )c(ũ−vj ) b(2vj )a(2ũ+2)c(ũ+vj )
w1(1) (ũ) (1)
+ w2 (ũ)
a(2v (1))b(ũ−v (1) ) a(2v (1))a(2ũ)a(ũ+v (1) )
β (1) vj(1) =
j j j j
(1) (1)
. (104)
c(ũ+vj ) a(2ũ+2)c(ũ−vj )
w1(1) (ũ) + w2(1) (ũ)
a(ũ+vj(1) ) a(2ũ)b(ũ−vj(1) )
(1)
The explicit expression of β (1)(vj ) is
 (1)


2vj

 , for Eqs. (15), (16),

 2+2vj(1)


(1) (1)
2vj + ζ + 12 +vj
(1)
β (1) vj = (1) , for Eq. (17), (105)

 2+2vj ζ + − 32 −vj
(1)



 (1)
ζ + + 12 −vj
(1)


2vj
(1) + 5 (1) , for Eq. (18).
2+2vj ζ + 2 +vj
The explicit expressions of the eigenvalue and Bethe equations are
Λn (u; v1 , . . . , vn )
(ũ + 12 )(ũ + 32 ) n
(ũ + ṽi + 1)(ũ − ṽi + 1)
=
ũ(ũ + 1) (ũ + ṽi )(ũ − ṽi )
i=1
#
× W1− (ũ)W1+ (ũ)
$1 %2N
n1 (1) (1)
&
+ ũ (ũ − v̂l − 1)(ũ + v̂l − 1)
+ W2− (ũ)W2+ (ũ) 2
1
2 − ũ l=1 (ũ − v̂l(1) + 1)(ũ − v̂l(1) + 1)
(ũ + 12 )(ũ + 32 ) n
(ũ + ṽi + 1)(ũ − ṽi + 1)
+
(ũ + 1)(ũ + 2) (ũ + ṽi + 2)(ũ − ṽi + 2)
i=1
# $ 1 %
− + ( 2 + ũ)( 52 + ũ) 2N
× W4 (ũ)W4 (ũ) 1
( 2 − ũ)( 32 + ũ)
$1 %2N n1
&
− + 2 + ũ (ũ − v̂l(1) + 3)(ũ + v̂l(1) + 3)
+ W3 (ũ)W3 (ũ) 1 , (106)
2 − ũ
(1) (1)
l=1 (ũ − v̂l + 1)(ũ − v̂l + 1)
and
$1 %2N
n1 (1) (1)
− ṽi (ṽi − v̂ − 1)(ṽi + v̂ − 1)
α1− (ṽi )α1+ (ṽi ) 2
= l l
1
2 + ṽi (1) (1)
l=1 (ṽi − v̂l + 1)(ṽi + v̂l + 1)
(i = 1, . . . , n), (107)
n (v̂ (1) + ṽ − 1)(v̂ (1) − ṽ − 1)
i i
α2− v̂j(1) α2+ v̂j(1)
j j
(1) (1)
i=1 (v̂j + ṽi + 1)(v̂j − ṽi + 1)

n1
(v̂j(1) + v̂l(1) − 2)(v̂j(1) − v̂l(1) − 2)
= (1) (1) (1) (1)
(j = 1, . . . , n1 ) (108)
l=1=j (v̂j + v̂l + 2)(v̂j − v̂l + 2)
(1) (1)
respectively, where v̂l = vl + 1,


 1, for Eq. (11),
W1− (ũ) = 2 − ũ,
1
for Eq. (12), (109)

 ζ − − 1 − ũ, for Eqs. (13), (14),
2

 1, for Eq. (11),

 1 + ũ,
 for Eq. (12),
W2− (ũ) =
2
− (110)

 ζ − 2 − ũ, for Eq. (13),
1

 − 1
ζ − 2 + ũ, for Eq. (14),



1, for Eq. (11),

 1 + ũ, for Eq. (12),
W3− (ũ) =
2
− (111)

 ζ + + ũ, for Eq. (13),
3
 − 25

ζ − 2 − ũ, for Eq. (14),



1, for Eq. (11),
 (ũ+ 1 )(ũ+ 5 )
W4− (ũ) = − (ũ+ 3 ) , for Eq. (12),
2 2
(112)

 2

ũ + 32 + ζ − , for Eqs. (13), (14),

 −1, for Eq. (15),

 (ũ− 1 )(ũ+ 3 )
W1+ (ũ) =
2 2
, for Eq. (16), (113)
 (ũ+ 12 )


−ũ − 12 − ζ + , for Eqs. (17), (18),

 1, for Eq. (15),


 −ũ − 3 , for Eq. (16),
W2+ (ũ) =
2
+ (114)

 ζ + 2 + ũ, for Eq. (17),
1

 + 1
ζ + 2 − ũ, for Eq. (18),

 1, for Eq. (15),


 −ũ − 3 , for Eq. (16),
W3+ (ũ) =
2
+ (115)

 ζ − 2 − ũ, for Eq. (17),
3


ζ + + 52 + ũ, for Eq. (18),

 −1,
 for Eq. (15),
+
W4 (ũ) = −ũ − 2 ,
5
for Eq. (16), (116)

 ũ + 3 − ζ + , for Eqs. (17), (18),
2

 1, for Eqs. (11), (13),


 1 −ṽi
2
, for Eq. (12),
α1− (ṽi ) = 1
2 +ṽi (117)



 ζ − − 12 −ṽi
, for Eq. (14),
ζ − − 12 +ṽi

 (1 + ṽi ), for Eqs. (15), (17),

 1 −ṽi
 2
(1 + ṽi ), for Eq. (16),
α1+ (ṽi ) = 12 +ṽi (118)


 + 1
 (1 + ṽi ) ζ + 2 +ṽi , for Eq. (18),
+ 1
ζ + 2 −ṽi
 1, for Eqs. (11), (12),



 − 1 (1)
 ζ + 2 +v̂j , for Eq. (13),
α2− v̂j(1) = ζ − + 12 −v̂j
(1)
(119)



 ζ − − 32 −v̂j(1)
 (1) , for Eq. (14),
ζ − − 2 +v̂j
3
 1, for Eqs. (15), (16),




 + 1 (1)
(1)  ζ − 2 +v̂j , for Eq. (17),
α2+ v̂j = ζ + − 12 −v̂j
(1)
(120)



 ζ + + 32 −v̂j(1)
 (1) , for Eq. (18).
ζ + + 2 +v̂j
3
4. Conclusions
By solving the reflection equation of the graded Osp(2|2) model, we get four different
diagonal reflecting matrices. We exactly solve the model by using the algebraic Bethe
ansatz. The n-particle eigenstates is constructed explicitly and the eigenvalue of the
transfer matrix and the corresponding Bethe ansatz equations are derived in detail. With
the help of the linearly independent property of the “quasi” n-particle states as what the
algebraic Bethe ansatz assumes, we can prove our conclusions for the general n-particle
state.
Using the Bethe ansatz equations and energy spectrum, we can study the boundary
contributions to the thermodynamic quantities and the surface critical behavior of two-
dimensional polymers. We also can use the n-particle state to calculate the corresponding
form factor and correlation functions which have rather important applications in
condensed matter physics.
It seems that our procedure can be applied to all the Hubard-like model with open
boundary conditions. For the models such as Bn , Cn models, the same procedure can
be also used. For the model with non-diagonal reflecting matrices, it is an interesting
problem.
Acknowledgements
We thank Professor Shengli Zhang for his some useful discussions. This work is
supported by the National Natural Science Foundation of China under Grant No.10175050.
Appendix A
f1 (u− )f3 (u+ ) f6 (u− )f3 (2u)

a11 (v, u) = , a21 (v, u) = − ,
f3 (u− )f1 (u+ ) f3 (u− )f1 (2u)
f6 (u+ ) f6 (u− )f10 (u+ )
a31 (v, u) = − , a41 (v, u) = ,
f1 (u+ ) f3 (u− )f1 (u+ )
f8 (u+ ) f1 (u− )f10 (u+ )
a51 (v, u) = − , a61 (v, u) = − , (A.1)
f1 (u+ ) f3 (u− )f1 (u+ )
f6 (u+ )f3 (2u)f3 (2v)
aa = −
R1A (u, v)aa ,
f1 (u+ )f1 (2u)f1 (2v)
f6 (u+ )f3 (u− )f3 (2v)
āa = −
R1A (u, v)āa ,
f1 (u+ )f2 (u− )f1 (2v)
$ %
˜ f10 (u− ) f6 (u+ )f3 (2v)
R1A (u, v)āa
a ā = f1 (u) − ,
f2 (u− ) f1 (u+ )f1 (2v)
f6 (u− )f3 (2u)
aa =
R2A (u, v)aa ,
f3 (u− )f1 (2u)
f6 (u− )f5 (u+ )
āa = −
R2A (u, v)āa ,
f3 (u− )f3 (u+ )
$ %
R2A (u, v)āa = − f ˜1 (u) + f10 (u+ ) f6 (u− ) ,
a ā
f3 (u+ ) f3 (u− )
f (u
10 − 3 )f (2v) f10 (u+ )
R3A (u, v) = − , R4A (u, v) = ,
f2 (u− )f1 (2v) f3 (u+ )
f6 (u+ )f10 (u+ )f1 (u− )f3 (2u)
aa =
R5A (u, v)aa ,
f1 (u+ )f3 (u+ )f3 (u− )f1 (2u)
f5 (u+ ) − f5 (u− )
āa =
R5A (u, v)āa ,
f1 (u+ )f2 (u− )
(f10 (u− ) − f1 (u+ ) − f7 (v))f3 (2u)
a ā =
R5A (u, v)āa ,
f1 (u+ )f2 (u− )f1 (2u)
f10 (u+ )f6 (u− )f3 (2u)
R6A (u, v) = ,
f3 (u+ )f3 (u− )f1 (2u)
f3 (u− )f10 (u+ )f˜1 (u)
aa =
R7A (u, v)aa ,
f1 (u+ )f5 (u− )
f5 (u+ )f6 (u− )
āa = −
R7A (u, v)āa ,
f1 (u+ )f2 (u− )
$ %
f1 (u+ )f3 (2u) ˜1 (u) f10 (u+ )f6 (u− ) ,
R7A (u, v)āa
a ā = − 2 f
f6 (u+ )f1 (2u) f1 (u+ )f2 (u− )
f (u
10 + 6 −)f (u )
aa = −
R8A (u, v)aa ,
f1 (u+ )f10 (u− )
f10 (u+ )f3 (u− )

āa =
R8A (u, v)āa ,
f1 (u+ )f2 (u− )
f10 (u+ )f6 (u− )
a ā = 2
R8A (u, v)āa , (A.2)
f1 (u+ )f2 (u− )
f3 (u− )f3 (u+ ) f3 (u− )f10 (u+ )f10 (u+ )
a13 (u, v) = − ,
f2 (u− )f2 (u+ ) f2 (u− )f2 (u+ )f3 (u+ )
f3 (2v)f10 (u+ )f2 (2u)
a23 (u, v) = −f˜3 (u) ,
f1 (2v)f3 (u+ )f3 (2u)
f10 (u− )f2 (2u)
a33 (u, v) = f˜3 (u) ,
f2 (u− )f3 (2u)
f3 (2v)f10 (u+ )f2 (2u)
a43 (u, v) = f˜3 (u) ,
f1 (2v)f3 (u+ )f10 (2u)
f10 (u− )f2 (2u)
a53 (u, v) = −f˜3 (u) ,
f2 (u− )f10 (2u)
f6 (u− )f10 (u+ )f2 (2u)
a63 (u, v) = f˜3 (u)f10 (2u) ,
f2 (u− )f3 (u+ )f3 (2u)
f8 (u− )f2 (2u)
a73 (u, v) = f˜3 (u)f10 (2u) ,
f2 (u− )f3 (2u)
f6 (u− )f5 (u+ )f2 (2u)
a83 (u, v) = −f˜3 (u) ,
f2 (u− )f3 (u+ )f3 (2u)
f6 (u− )f5 (u+ )f2 (2u)
a93 (u, v) = −f˜3 (u) , (A.3)
f2 (u− )f3 (u+ )f10 (2u)
f1 (u− )f2 (u+ ) f8 (u− )f2 (2u)
b11 (v, u) = , b21 (v, u) = − ,
f2 (u− )f1 (u+ ) f2 (u− )f1 (2u)
f10 (u− )f6 (u+ )f2 (2u) f8 (u+ )
b31 (v, u) = −f˜3 (u) , b41 (v, u) = − ,
f2 (u− )f1 (u+ )f10 (2u) f1 (u+ )
f10 (u− )f3 (u+ ) f6 (u+ )
b51 (v, u) = , b61 (v, u) = − , (A.4)
f2 (u− )f1 (u+ ) f1 (u+ )
f1 (u− )f6 (u− )f2 (u+ )f10 (u+ )
b12 (u, v) = ,
f3 (u− )f10 (u− )f3 (u+ )f6 (u+ )
f8 (u− )f8 (u+ )f3 (2u)f3 (2v)
b22 (u, v) = 4 ,
f9 (u− )f6 (u+ )f1 (2u)f1 (2v)
f8 (u− )f8 (u+ )f3 (2u)
b32 (u, v) = ,
f10 (u− )f6 (u+ )f1 (2u)
f6 (u− )f2 (u+ ) f10 (u+ )
b42 (u, v) = , b52 (u, v) = ,
f3 (u− )f3 (u+ ) f3 (u+ )
f6 (v)f10 (u)
aa = 4
R1F (u, v)aa ,
f4 (u+ )f5 (u− )
f2 (−u− )f8 (u+ )f3 (2u)f10 (1 + v)
āa = 2
R1F (u, v)āa ,
f5 (−u− )f4 (u+ )f1 (2u)f7 (1 + v)
(2f7 (−u− ) + f1 (u+ ))f3 (2u)f10 (1 + v)

a ā = 2
R1F (u, v)āa ,
f5 (u− )f4 (u+ )f1 (2u)f7 (1 + v)
f2 (−u− )f8 (u+ )
aa =
R2F (u, v)aa ,
f5 (−u− )f6 (u+ )
f3 (−u− )f8 (u+ )f3 (2u)
āa =
R2F (u, v)āa ,
f5 (−u− )f6 (u+ )f1 (2u)
(f6 (−u− ) + f7 (−u))f8 (u+ )
a ā =
R2F (u, v)āa ,
f7 (−u)f10 (u− )f6 (u+ )
f8 (u− )f3 (2u)
aa =
R3F (u, v)aa ,
f10 (u− )f1 (2u)
f6 (u− )f5 (u+ )
āa = −
R3F (u, v)āa ,
f3 (u− )f3 (u+ )
(f10 (u+ ) + f10 (2u))f8 (u+ )
a ā =
R3F (u, v)āa , (A.5)
f7 (−u)f10 (u− )f6 (u+ )
f1 (u− )f2 (u+ )
b13 (u, v) = ,
f2 (u− )f2 (1 + u+ )
f8 (1 + u+ )f2 (2u)f2 (2v)
b23 (u, v) = − ,
f2 (1 + u+ )f2 (1 + 2u)f1 (2v)
f10 (u− )f10 (1 + u+ )f2 (2u)f3 (1 + 2v)
b33 (u, v) = ,
f2 (u− )f2 (1 + u+ )f2 (1 + 2u)f2 (1 + 2v)
f8 (u− )f2 (2u)
b43 (u, v) = − ,
f2 (u− )f2 (1 + 2u)
f3 (u− )f10 (1 + u+ )f6 (u)
b53 (u, v) = 2 ,
f2 (u− )f2 (1 + u+ )f2 (1 + 2u)
f10 (u− )f6 (u)
b63 (u, v) = 2 ,
f2 (u− )f2 (1 + 2u)
f3 (1 + 2u)f10 (1 + u+ )f3 (u− )
b73 (u, v) = − ,
f2 (1 + u+ )f2 (1 + 2u)f2 (u− )
f3 (1 + 2u)f10 (u− )
b83 (u, v) = − , (A.6)
f2 (1 + 2u)f2 (u− )
R(u+ )a+1,b+1
c+1,d+1
cd =
R1D (u, v)ab ,
f1 (u+ )
f6 (u− )f10 (u+ )
c11 (u, v) = ,
f3 (u− )f1 (u+ )
f3 (u− )f6 (u+ )
c21 (u, v) = ,
f3 (u− )f1 (u+ )
f6 (u− )f10 (u+ )
c31 (u, v) = − ,
f3 (u− )f1 (u+ )
f3 (u− )f8 (u+ )

c41 (u, v) = − ,
f3 (u− )f1 (u+ )
f6 (u+ ) f6 (u− )f3 (u+ )
c51 (u, v) = − f˜1 (u) ,
f1 (u+ ) f3 (u− )f1 (u+ )
f6 (u− )f3 (u+ )
c61 (u, v) = ,
f3 (u− )f1 (u+ )
f6 (u− )f3 (u+ ) f6 (u+ )
c71 (u, v) = f˜1 (v) − f˜1 (v)f˜1 (u) ,
f3 (u− )f1 (u+ ) f1 (u+ )
f6 (u+ ) f6 (u− )f3 (u+ )
c81 (u, v) = f˜1 (u) − ,
f1 (u+ ) f3 (u− )f1 (u+ )
f˜1 (v)f3 (u− )f6 (u+ )
c91 (u, v) = ,
f3 (u− )f1 (u+ )
f3 (u− )f6 (u+ )
1
c10 (u, v) = − , (A.7)
f3 (u− )f1 (u+ )
f10 (u− )f2 (−u+ )

c12 (u, v) = − ,
f2 (u− )f3 (−u+ )
f8 (u− )f5 (u+ )
c22 (u, v) = ,
f2 (u− )f1 (u+ )
f10 (u− )f10 (u+ )
c32 (u, v) = − ,
f2 (u− )f1 (u+ )
f1 (u− )f6 (u+ )f10 (2u)
c42 (u, v) = − ,
f3 (u− )f1 (u+ )f1 (2u)
f10 (u− )f2 (u+ )
c52 (u, v) = − ,
f2 (u− )f3 (u+ )
f8 (u− )f10 (2u)f6 (2v)
c62 (u, v) = ,
f10 (u− )f1 (2u)f1 (2v)
f8 (u− )f10 (2u)
c72 (u, v) = ,
f10 (u− )f1 (2u)
f10 (u+ )
c82 (u, v) = −f˜1 (v) ,
f3 (u+ )
f10 (u+ )
c92 (u, v) = ,
f3 (u+ )
R(u− )a+1,b+1
c+1,d+1 f3 (u+ ) − δad f6 (u+ )f10 (u− )c1 (u, v)
1
cd =
R1C (u, v)ab ,
f2 (u− )f3 (u+ )
f10 (u− )f6 (u+ ) D
cd =
R2C (u, v)ab cd − δad ,
R1 (u, v)ab (A.8)
f2 (u− )f3 (u+ )
R3C (u, v)ab

cd
a+1,b+1
f10 (u− )f6 (u+ ) 1 f10 (u+ )R(u− )c+1,d+1
= c6 (u, v)δad − ηad
f2 (u− )f3 (u+ ) f2 (u− )f3 (u+ )
f (u − )f (u + )
− f˜3 (u)
10 2
(1 − δab ),
f2 (u− )f3 (u+ )
R4C (u, v)ab
cd
f10 (u− )f6 (u+ ) 1 f10 (u− )R(u+ )a+1,b+1

c9 (u, v)δad − ηad f˜1 (v)
c+1,d+1
=
f2 (u− )f3 (u+ ) f2 (u− )f3 (u+ )
f10 (u+ )f2 (u− )
− f˜1 (v)f˜3 (u) (1 − δab ),
f2 (u− )f3 (u+ )
R5C (u, v)ab
cd
a+1,b+1
f10 (u− )f6 (u+ ) 1 f10 (u− )R(u+ )c+1,d+1
= c10 (u, v)δad + ηad
f2 (u− )f3 (u+ ) f2 (u− )f3 (u+ )
f (u )f
10 + 2 − (u )
− f˜3 (u) (1 − δab ), (A.9)
f2 (u− )f3 (u+ )
where ηad = 1 if a = d and ηad = −1 if a = d̄,
f10 (1 + u+ ) f9 (−u− )f10 (2u)
e31 (v, u) = , e21 (v, u) = ,
f2 (1 + u+ ) f2 (−u− )f1 (2u)
! "
f3 (2v) f9 (−u− )f10 (1 + u+ )f8 (u− ) f8 (u+ )f3 (1 + u+ )f10 (u− )
e1 (v, u) = −
1
+ ,
f1 (2v) f2 (−u− )f2 (1 + u+ )f6 (u− ) f10 (u+ )f2 (1 + u+ )f6 (u− )
f4 (u+ )f3 (1 + u+ )
aa =
R1be (v, u)aa ,
f3 (u+ )f2 (1 + u+ )
f2 (u+ )
āa =
R1be (v, u)āa ,
f5 (u+ )
f6 (u+ )f3 (1 + u+ )
a ā = −2
R1be (v, u)āa ,
f3 (u+ )f2 (1 + u+ )
f6 (u+ )f8 (u− )
aa = 2
R2be (v, u)aa ,
f3 (u+ )f6 (u− )
4f6 (v)
āa =
R2be (v, u)āa ,
f6 (2 + u+ )f10 (u+ )f6 (u− )f6 (1 − u− )
f6 (u+ )f10 (u− ) − 2f6 (1 + v)
a ā = −2
R2be (v, u)āa ,
f6 (2 + u+ )f10 (u+ )f6 (u− )f6 (1 − u− )
R̃ be (u, v)ab
cd =
cd
R3be (v, u)ab δab
r̃(u+ )aa āā f3 (u− )f3 (u+ )
cd r̃(u+ )āa − R̃ (u, v)dc r̃(u+ )aa
R̃ be (u, v)ab a ā be ab aa
+ (1 − δab ),
(r̃ 2 (u+ )aāaā − r̃ 2 (u+ )aa
aa )f3 (u− )f3 (u+ )
f3 (1 + 2u)f8 (u− )
aa =
R4be (v, u)aa ,
f2 (1 + 2u)f10 (u− )
f3 (u+ )f8 (u− )
āa =
R4be (v, u)āa ,
f3 (u+ )f10 (u− )
(3f6 (u) + f6 (v))f8 (u− )
a ā =
R4be (v, u)āa , (A.10)
f6 (u+ )f6 (2 + 2u)f10 (u− )
with
R̃ be (u, v)ab
cd
= f6 (u− )f3 (u+ )δad + ηad f6 (u− )f10 (u+ )g22 (u, v)r̂(−u− )ab cd

+ f6 (u+ )f3 (u− )b51 (v, u)g22 (u, v) r̂(−u− )āa
a ā
− r̂(−u− )aa āā (1 − δab ), (A.11)
f3 (u+ )f3 (u− )

d11 (u, v) = d14 (v, u) = ,
f2 (u+ )f1 (u− )
f6 (u− )
d21 (u, v) = d34 (v, u) = ,
f1 (u− )
f3 (u+ )f10 (u− )
d31 (u, v) = d24 (v, u) = − ,
f2 (u+ )f1 (u− )
f3 (u+ )f3 (u− )
d12 (u, v) = d13 (v, u) = − ,
f5 (u+ )f1 (u− )
f6 (u+ )f10 (u− )
d22 (u, v) = −d23 (v, u) = 2 ,
f2 (u+ )f1 (u− )
f8 (u+ )f6 (−u− ) f4 (u+ )f6 (u− )
d32 (u, v) = d43 (v, u) = − ,
f2 (u+ )f1 (−u− ) f2 (u+ )f1 (u− )
f8 (u+ )f10 (u− )
d42 (u, v) = −d33 (v, u) = 2 . (A.12)
f10 (u+ )f6 (u− )
Appendix B
A
H1,d (u, vi , vj )

1
= a41(u, vi ) c51 (vi , vj ) + c71 (vi , vj )

+ a51 (u, vi ) c42 (vi , vj ) + c62 (vi , vj ) + b21 (u, vi )g12 (vi , vj )

+ a11 (u, vi )a21 (u, vj )g12 (u, vi ) r̂(vi − u)dd̄1 dd̄1 − r̂(vi − u)d̄d̄1 dd1 , (B.1)
1 1 1 1
A
H2,d 1
(u, vi , vj )dd11 dd11

= a41(u, vi ) c61 (vi , vj ) + c91 (vi , vj )

+ a51 (u, vi ) R3C (vi , vj )dd11 dd11 + R4C (vi , vj )dd11 dd11
+ a11 (u, vi )a21 (u, vj )g22 (u, vi )r̂(vi − u)dd11 dd11 ,
A
H2,d 1
(u, vi , vj )dd1 d̄d̄1
1 1
C
= a5 (u, vi ) R3 (vi , vj )dd1 d̄d̄1 + R4C (vi , vj )dd1 d̄d̄1 + b31 (u, vi )g12 (vi , vj )
1
1 1 1 1
d d̄
− a11 (u, vi )a21 (u, vj )g22 (u, vi )r̂(vi − u)d1 d̄1 ,
1 1
A
H2,d 1
(u, vi , vj )dd̄1 dd̄1
1 1

= a41(u, vi ) c61 (vi , vj ) + c91 (vi , vj )

+ a51 (u, vi ) R3C (vi , vj )dd̄1 d̄d1 + R4C (vi , vj )dd̄1 dd̄1
1 1 1 1
+ b31 (u, vi )g12 (vi , vj ) + a11 (u, vi )a21 (u, vj )g22 (u, vi )r̂(vi − u)dd̄1 dd̄1 , (B.2)
1 1
A
H3,d 1
(u, vi , vj )
= a41(u, vi )c81 (vi , vj ) + a51 (u, vi )c72 (vi , vj ) + b21 (u, vi )g22 (vi , vj )

1 1 1 1
A
H4,d 1
= a41(u, vi )c10
1
(vi , vj ) + a51 (u, vi )R5C (vi , vj )dd11 dd11
d d
+ a11 (u, vi )a31 (u, vj )g22 (u, vi )r̂(vi − u)d11 d11 ,
A
H4,d 1
1 1
= a51(u, vi )R5C (vi , vj )dd1 d̄d̄1 + b31 (u, vi )g22 (vi , vj )

1 1
− a11 (u, vi )a31 (u, vj )g22 (u, vi )r̂(vi − u)dd1 d̄d̄1 ,

1 1
A
H4,d 1
1 1
d̄ d
= a41(u, vi )c10
1
(vi , vj ) + a51 (u, vi )R5C (vi , vj )d1 d̄1 + b31 (u, vi )g22 (vi , vj )
1 1
+ a11 (u, vi )a31 (u, vj )g22 (u, vi )r̂(vi − u)dd̄1 d̄d1 , (B.4)
1 1
Aaa
H1,d 1
(u, vi , vj )
d1 a 1 2
= R5A (u, vi )ad 1
c 5 (vi , vj ) + c 1
7 (vi , vj ) + R6
A
(u, vi ) c 4 (vi , vj ) + c 2
6 (vi , vj )
d1 a A
+ b22 (u, vi )g12 (vi , vj ) + r̃(u + vi )ad
da r̄(u − vi )ad1 R3 (u, vj )e1 (vi , u)
1
dg
+ r̃(u + vi )ad 1 2 A d̄1 d
dg r̄(u − vi )ad1 g1 (u, vi )R1 (u, vj )d d̄1

× r̂(vi − u)dd̄1 d̄d1 − r̂(vi − u)d̄d̄1 dd1
1 1 1 1
− r̃(u + vi )ad̄ d̄a1 r̄(u − vi )dad1 a1 g12 (u, vi )R1A (u, vj )dd1 d̄d̄1
1 1 1

× r̂(vi − u)dd̄1 d̄d1 − r̂(vi − u)d̄d̄1 dd1 , (B.5)
1 1 1 1
Aaa
H2,d 1
d1 a 1
= R5A (u, vi )ad 1
c6 (vi , vj ) + c91 (vi , vj )
d d d d
+ R6A (u, vi ) R3C (vi , vj )d11 d11 + R4C (vi , vj )d11 d11
dg
+ r̃(u + vi )ad d1 d d1 d1
dg r̄(u − vi )ad1 g2 (u, vi )R1 (u, vj )dd1 r̂(vi − u)d1 d1
1 2 A
− δa d̄1 r̃(u + vi )ad ad1 A d1 d1

ad1 r̄(u − vi )ad1 R3 (u, vj )R3 (vi , u)d1 d1 ,
1 be
Aaa
H2,d 1
1 1
C
= R6 (u, vi ) R3 (vi , vj )dd1 d̄d̄1 + R4C (vi , vj )dd1 d̄d̄1
A
1 1 1 1
d1 d̄1 2
+ R1F (u, vi )āa g1 (vi , vj )
dg 2
− r̃(u + vi )ad d̄1 d
dg r̄(u − vi )ad1 g2 (u, vi )R1 (u, vj )d d̄1 r̂(vi
1 A
− u)dd1 d̄d̄1
1 1
− r̃(u + vi )ad̄ d̄a1 r̄(u − vi )dad1 a1 g22 (u, vi )R1A (u, vj )dd1 d̄d̄1 r̂(vi − u)dd̄1 d̄d1
1 1 1 1 1
d1 a A d1 d̄1
+ r̃(u + vi )ad
da r̄(u − vi )ad1 R3 (u, vj )R3 (vi , u)d d̄ ,
be
Aaa
H2,d 1
1 1
d1 a 1
= R5 (u, vi )ad1 c6 (vi , vj ) + c91 (vi , vj )
A

+ R6A (u, vi ) R3C (vi , vj )dd̄1 dd̄1 + R4C (vi , vj )dd̄1 dd̄1 + R1F (u, vi )āa
d̄1 d1 2
g1 (vi , vj )
1 1 1 1
ad dg d̄ d d̄ d
+ r̃(u + vi )dg1 r̄(u − vi )ad1 g22 (u, vi )R1A (u, vj )d1d̄ r̂(vi − u)d1 d̄1
1 1 1
+ r̃(u + vi )ad̄ d̄a1 r̄(u − vi )dad1 a1 g22 (u, vi )R1A (u, vj )dd1 d̄d̄1 r̂(vi − u)d̄d̄1 dd1
1 1 1 1 1
d1 a A d̄1 d1
+ r̃(u + vi )ad
be
(B.6)
Aaa
H3,d 1
(u, vi , vj )
d1 a 1
= R5A (u, vi )ad c (vi , vj ) + R6A (u, vi )c72 (vi , vj )
1 8
d1 a A
+ b22 (u, vi )g22 (vi , vj ) + r̃(u + vi )ad
da r̄(u − vi )ad1 R4 (u, vj )e1 (vi , u)
1
dg
+ r̃(u + vi )ad 1 2 A d̄1 d
dg r̄(u − vi )ad1 g1 (u, vi )R2 (u, vj )d d̄1

× r̂(vi − u)dd̄1 d̄d1 − r̂(vi − u)d̄d̄1 dd1
1 1 1 1
− r̃(u + vi )ad̄ d̄a1 r̄(u − vi )dad1 a1 g12 (u, vi )R2A (u, vj )dd1 d̄d̄1
1 1 1

× r̂(vi − u)dd̄1 d̄d1 − r̂(vi − u)d̄d̄1 dd1 , (B.7)
1 1 1 1
Aaa
H4,d 1
d1 a 1
= R5A (u, vi )ad c (vi , vj ) + R6A (u, vi )R5C (vi , vj )dd11 dd11
1 10
dg
+ r̃(u + vi )ad d1 d d1 d1
dg r̄(u − vi )ad1 g2 (u, vi )R2 (u, vj )dd1 r̂(vi − u)d1 d1
1 2 A
− δa d̄1 r̃(u + vi )ad ad1 A d1 d1

ad1 r̄(u − vi )ad1 R4 (u, vj )R3 (vi , u)d1 d1 ,
1 be
Aaa
H4,d 1
1 1
= R6A (u, vi )R5C (vi , vj )dd1 d̄d̄1 + R1F (u, vi )dāa1 d̄1 g22 (vi , vj )
1 1
dg
− r̃(u + vi )ad d̄1 d d1 d̄1
dg r̄(u − vi )ad1 g2 (u, vi )R2 (u, vj )d d̄ r̂(vi − u)d d̄
1 2 A
1 1 1
− r̃(u + vi )ad̄ d̄a1 r̄(u − vi )dad1 a1 g22 (u, vi )R2A (u, vj )dd1 d̄d̄1 r̂(vi − u)dd̄1 d̄d1
1 1 1 1 1
d1 a A d1 d̄1
+ r̃(u + vi )ad
be
Aaa
H4,d 1
1 1
d1 a 1
= R5A (u, vi )ad c (vi , vj ) + R6A (u, vi )R5C (vi , vj )dd̄1 dd̄1 + R1F (u, vi )d̄āa1 d1 g22 (vi , vj )
1 10 1 1
dg 2
+ r̃(u + vi )ad d̄1 d
dg r̄(u − vi )ad1 g2 (u, vi )R2 (u, vj )d d̄1 r̂(vi
1 A
− u)dd̄1 dd̄1
1 1
+ r̃(u + vi )ad̄ d̄a1 r̄(u − vi )dad1 a1 g22 (u, vi )R2A (u, vj )dd1 d̄d̄1 r̂(vi − u)d̄d̄1 dd1
1 1 1 1 1
d1 a A d̄1 d1
+ r̃(u + vi )ad
be
(B.8)
A
H1,d2 1 (u, vi , vj ) = a63(u, vi ) c51 (vi , vj ) + c71 (vi , vj ) + b23 (u, vi )g12 (vi , vj )

+ a73 (u, vi ) c42 (vi , vj ) + c62 (vi , vj ) + a13 (u, vi )a43 (u, vj )e11 (vi , u)

1 1 1 1
A2
H2,d (u, vi , vj )dd11 dd11

1
= a63(u, vi ) c61 (vi , vj ) + c91 (vi , vj )

+ a73 (u, vi ) R3C (vi , vj )dd11 dd11 + R4C (vi , vj )dd11 dd11
+ a13 (u, vi )a23 (u, vj )g22 (u, vi )r̂(vi − u)dd11 dd11
− a13 (u, vi )a43 (u, vj )R3be (vi , u)dd11 dd11 ,
A2
H2,d 1
1 1
C
= a7 (u, vi ) R3 (vi , vj )dd1 d̄d̄1 + R4C (vi , vj )dd1 d̄d̄1 + b33 (u, vi )g12 (vi , vj )
3
1 1 1 1
− a13 (u, vi )a23 (u, vj )g22 (u, vi )r̂(vi − u)dd1 d̄d̄1 + a13 (u, vi )a43 (u, vj )R3be (vi , u)dd1 d̄d̄1 ,
1 1 1 1
A2
H2,d 1

1 1
= a63(u, vi ) c61 (vi , vj ) + c91 (vi , vj )

+ a73 (u, vi ) R3C (vi , vj )dd̄1 d̄d1 + R4C (vi , vj )dd̄1 dd̄1 + b33 (u, vi )g12 (vi , vj )
1 1 1 1
+ a13 (u, vi )a23 (u, vj )g22 (u, vi )r̂(vi − u)dd̄1 d̄d1

1 1
+ a13 (u, vi )a43 (u, vj )R3be (vi , u)dd̄1 d̄d1 , (B.10)

1 1
A2
H3,d 1
(u, vi , vj )
= a63(u, vi )c81 (vi , vj ) + a73 (u, vi )c72 (vi , vj )
+ b23 (u, vi )g22 (vi , vj ) + a13 (u, vi )a53 (u, vj )e11 (vi , u)

1 1 1 1
A2
H4,d 1
= a63(u, vi )c10
1
(vi , vj ) + a73 (u, vi )R5C (vi , vj )dd11 dd11
d d d d
+ a13 (u, vi )a33 (u, vj )g22 (u, vi )r̂(vi − u)d11 d11 − a13 (u, vi )a53 (u, vj )R3be (vi , u)d11 d11 ,
A2
H4,d 1
1 1
= a73(u, vi )R5C (vi , vj )dd1 d̄d̄1 + b33 (u, vi )g22 (vi , vj )

1 1
− a13 (u, vi )a33 (u, vj )g22 (u, vi )r̂(vi − u)dd1 d̄d̄1 + a13 (u, vi )a53 (u, vj )R3be (vi , u)dd1 d̄d̄1 ,
1 1 1 1
A d̄ d
H4,d2 1 (u, vi , vj )d1 d̄1
1 1
= a63(u, vi )c10
1
(vi , vj ) + a73 (u, vi )R5C (vi , vj )dd̄1 d̄d1 + b33 (u, vi )g22 (vi , vj )
1 1
+ a13 (u, vi )a33 (u, vj )g22 (u, vi )r̂(vi − u)dd̄1 d̄d1

1 1
+ a13 (u, vi )a53 (u, vj )R3be (vi , u)dd̄1 d̄d1 , (B.12)

1 1
where all the repeated indices sum over 1 to 2 except for a, d1 and we have checked that
x (u, v , v )d̄1 d1 x (u, v , v )d̄1 d1
x
H2,d (u, vi , vj )dd11 dd11 H2, d̄ i j d̄ d H2,d 1
i j d d̄
1
=− 1 1 1
= 1 1
, (B.13)
rdd11dd11 (ṽi − ṽj ) rdd1d̄d̄1 (ṽi − ṽj ) rd̄d1dd̄1 (ṽi − ṽj )
1 1 1 1
x (u, v , v )d̄1 d1 x (u, v , v )d̄1 d1
x
H4,d (u, vi , vj )dd11 dd11 H4, d̄ i j d̄ d H4,d 1
i j d d̄
1
=− 1 1 1
= 1 1
, (B.14)
rdd11dd11 (ṽi + ṽj ) rdd1d̄d̄1 (ṽi + ṽj ) rd̄d1dd̄1 (ṽi + ṽj )
1 1 1 1
where x = A, Aaa , A2 .
Appendix C

Ãa ā (u)Υn (v1 , . . . , vn )

= Φnd1 ···dn (v1 , . . . , vn ) T̃ n (u; {vm })db11 ···d n
···bn a ā F
b1 ···bn
|0

n
(1)
+
ād1 Ψn−1 (u, vi ; {vm })ād1
R1A (u, vi )āa
i=1

n
(2) n
(3)
+ R2A (u, vi )āa Ψ (u, vi ; {vm })āa + R3A (u, vi )Ψn−1 (u, vi ; {vm })a ā
āa n−1
i=1 i=1

n
(4)
+ R4A (u, vi )Ψn−1 (u, vi ; {vm })a ā
i=1

n−1
n
(5)
+ Aa ā
δae2 ξd1 H1,d 1
i=1 j =i+1

n−1
n
ag (6) fe
(u, vi , vj )āg Md1 c2 Ψn−2 (u, vi , vj ; {vm })d1 c2
Aa ā ef
+ ξd1 H2,d 1
i=1 j =i+1

n−1
n
(7)
+ Aa ā
ξd1 H3,d 1
(u, vi , vj )Ψn−2 (u, vi , vj ; {vm })ād1
i=1 j =i+1

n−1
n
A ag (8)
+ āg Mf d1 Ψn−2 (u, vi , vj ; {vm })f d1 ,
ξd1 H4,daā1 (u, vi , vj )ed ed
(C.1)
i=1 j =i+1
where Mcdab
= 1 when a = b = c = d, or a = c = b = d or a = d = b = c and Mcd
ab
= 0 for
other cases. The coefficients are
Aa ā
H1,d 1
(u, vi , vj )
d̄1 a 1
= −δad1 R5A (u, vi )ād 1
c 5 (vi , vj ) + c 1
7 (vi , vj )
d̄1 a A
+ r̃(u + vi )ad
da r̄(u − vi )ād1 R3 (u, vj )e1 (vi , u)
1
dg
+ r̃(u + vi )adgd̄1 r̄(u − vi )ād1 g12 (u, vi )R1A (u, vj )dd
d1 d
1
d1 d̄1 d1 d̄1
× r̂(vi − u)d̄ d − r̂(vi − u)d d̄
1 1 1 1
− r̃(u + vi )ad d̄1 a 2 d1 d̄1

d1 a r̄(u − vi )ād1 g1 (u, vi )R1 (u, vj )d1 d̄1
1 A

× r̂(vi − u)dd̄1 d̄d1 − r̂(vi − u)dd1 d̄d̄1 , (C.2)
1 1 1 1
Aa ā
H2,d 1
(u, vi , vj )āā āā
d̄1 a 1
c6 (vi , vj ) + c91 (vi , vj ) + R1F (u, vi )āāāā g12 (vi , vj )
dg
− δad1 r̃(u + vi )adgd̄1 r̄(u − vi )ād1 g22 (u, vi )R1A (u, vj )dd̄1d̄d r̂(vi − u)d̄d̄1 d̄d̄1
1 1 1
+ δa d̄1 r̃(u + vi )ad 1

d̄1 ā
r̄(u − vi )ād g (u, vi )R1A (u, vj )dd1 d̄d̄1 r̂(vi
d1 ā 2
1 2
− u)dd11 dd11
1 1
+ r̃(u + vi )aa āā r̄(u − vi )dād1 ā1 R3A (u, vj )R3be (vi , u)āā āā ,
Aa ā
H2,d 1
(u, vi , vj )āa
āa
dg
= δad1 r̃(u + vi )adgd̄1 r̄(u − vi )ād1 g22 (u, vi )R1A (u, vj )dd
d1 d
1
r̂(vi − u)āa
d̄ d 1 1
+ r̃(u + vi )ad 1 d̄1 a 2 A d̄1 d1

d1 a r̄(u − vi )ād1 g2 (u, vi )R1 (u, vj )d̄1 d1 r̂(vi − u)āa
d1 d̄1
d̄1 a A
− r̃(u + vi )ad
da r̄(u − vi )ād1 R3 (u, vj )R3 (vi , u)d d̄ ,
be āa
Aa ā a ā
H2,d 1
(u, vi , vj )āa
d̄1 a 1
c6 (vi , vj ) + c91 (vi , vj )
dg
d1 d
1
r̂(vi − u)ad̄ ād
1 1
+ r̃(u + vi )ad d̄1 a 2 d̄1 d1

d1 a r̄(u − vi )ād1 g2 (u, vi )R1 (u, vj )d̄1 d1 r̂(vi
1 A
− u)ad ād̄
1 1
d̄1 a A
+ r̃(u + vi )ad
be a ā
(C.3)
Aa ā
H3,d 1
(u, vi , vj )
d̄1 a 1 d̄1 a A
= −δad1 R5A (u, vi )ād c (vi , vj ) + r̃(u + vi )ad
1 8 da r̄(u − vi )ād1 R4 (u, vj )e1 (vi , u)
1
dg
d1 d
1
d1 d̄1 d1 d̄1
× r̂(vi − u)d̄ d − r̂(vi − u)d d̄
1 1 1 1
− r̃(u + vi )ad d̄1 a 2 d̄1 d1

d1 a r̄(u − vi )ād1 g1 (u, vi )R2 (u, vj )d̄1 d1
1 A

× r̂(vi − u)dd̄1 d̄d1 − r̂(vi − u)dd1 d̄d̄1 , (C.4)
1 1 1 1
Aa ā
H4,d 1
(u, vi , vj )āā āā
d̄1 a 1
= −δad1 R5A (u, vi )ād c (vi , vj ) + R1F (u, vi )āā āā g22 (vi , vj )
1 10
dg
− δad1 r̃(u + vi )adgd̄1 r̄(u − vi )ād1 g22 (u, vi )R2A (u, vj )dd̄1d̄d r̂(vi − u)d̄d̄1 d̄d̄1
1 1 1
+ δa d̄1 r̃(u + vi )ad 1

d̄1 ā
r̄(u − vi )ād g (u, vi )R2A (u, vj )dd1 d̄d̄1 r̂(vi
d1 ā 2
1 2
− u)dd11 dd11
1 1
+ r̃(u + vi )aa āā r̄(u − vi )dād1 ā1 R4A (u, vj )R3be (vi , u)āā āā ,
Aa ā
H4,d 1
(u, vi , vj )āa
āa
dg
= δad1 r̃(u + vi )adgd̄1 r̄(u − vi )ād1 g22 (u, vi )R2A (u, vj )dd
d1 d
1
r̂(vi − u)āa
d̄ d 1 1
+ r̃(u + vi )ad 1 d̄1 a 2 A d̄1 d1

d1 a r̄(u − vi )ād1 g2 (u, vi )R2 (u, vj )d̄1 d1 r̂(vi − u)āa
d1 d̄1
d̄1 a A
− r̃(u + vi )ad
be āa
A a ā
H4,daā1 (u, vi , vj )āa
d̄1 a 1
= −δad1 R5A (u, vi )ād c (vi , vj )
1 10
dg
d1 d
1
r̂(vi − u)ad̄ ād
1 1
+ r̃(u + vi )ad d̄1 a 2 d̄1 d1

d1 a r̄(u − vi )ād1 g2 (u, vi )R2 (u, vj )d̄1 d1 r̂(vi
1 A
− u)ad ād̄
1 1
d̄1 a A
+ r̃(u + vi )ad
da r̄(u − vi )ād1 R4 (u, vj )R3 (vi , u)d d̄ .
be a ā
(C.5)
In above equations all the repeated indices sum over 1 to 2 except for a, d1 .
Appendix D
Denote
ρ3 (u) = c(u)2 − b(u)2 ,

hg
X1c (vi , vj ) = T (h) (vj )rgh
cc̄
(ṽj − ṽi )rcc̄ (ṽj + ṽi ),
hg
X2c (vi , vj ) = T (h) (vj )rgh
cc̄
(ṽj − ṽi )rc̄c (ṽj + ṽi ),

2
2
ρ4 (vi , vj ) = X2c (vi , vj ) − X1c (vi , vj ). (D.1)
c=1 c=1
Before deducing Eqs. (85)–(90), we present the following four relations (the proofs are
omitted here)
···dn
Scd11···cn
(vi ; {v̌i })τ1n (vi ; {ṽm })cb11···cn
···bn
1 −1 ···dn c ···c
= ρ12 (0) T (d1 ) (vi ) T n−1 (vi ; {v̌i })dc2···c d Sb11 ···bnn (vi ; {v̌i }), (D.2)
2 n 1 c1
···cn
···en
Sce22···cn
(vj ; {v̌i , v̌j })Sdc11 ···d n
(vi ; {v̌i })τ1n (vi ; {ṽm })db11 ···d
···bn
n
1 −1 f1 h
= ρ1 (ṽi − ṽj )ρ12 (0) T (c1 ) (vi )rhc11 ef21 (ṽi + ṽj )rc d 1 (ṽi − ṽj )
2 1
e ···e c2 ···cn d1 ···dn
× T n−2 (vi ; {v̌i , v̌j })c3 ···cn h S (vj ; { v̌i , v̌j })Sb1 ···bn (vi ; {v̌i }), (D.3)
3 n 1 h1 d2 ···dn
···cn
···en
Sce22···cn
(vj ; {v̌i , v̌j })Sdc11 ···d n
(vi ; {v̌i })τ1n (vj ; {ṽm })db11 ···d
···bn
n
1 −1 h g
= ρ1 (ṽj − ṽi )ρ12 (0) T (h1 ) (vj )rgc11he21 (ṽj − ṽi )rh 1d 1 (ṽj + ṽi )
2 1
e ···e c2 ···cn d1 ···dn
× T n−2 (vj ; {v̌i , v̌j })c3 ···cn h S (vj ; { v̌i , v̌j })S 1 ···bn
(vi ; {v̌i }), (D.4)
3 n 2 c2 d2 ···dn b
···en
Sae22 ···a n
···an
(vj ; {v̌i , v̌j })Sca11···cn
(vi ; {v̌i })τ1n (vi ; {ṽm })cd11···c d1 ···dn
···dn τ1 (vj ; {ṽm })b1 ···bn
n n
−1 f1 h
= ρ1 (ṽi − ṽj )ρ1 (0) T (a1 ) (vi )T (f1 ) (vj )rha11fe21 (ṽi + ṽj )rf d 1 (ṽj + ṽi )
2 1
n−2 a ···a
× T n−2 (vi ; {v̌i , v̌j })ea3 ···en
···a h h
T (vj ; {v̌i , v̌j })c3···cn f
3 n 1 1 3 n 2 c2
c2 ···cn d1 ···dn

×S d2 ···dn
(vj ; {v̌i , v̌j })S (v ; {v̌i }).
b1 ···bn i (D.5)
(2)
With the help of relation Eq. (D.2), we can easily change the |Ψn−1 (u, vi ; {vm })f d and
(4) (2) (4)
|Ψ̃n−1 (u, vi ; {vm })ab into |Ψ̃n−1 (u, vi ; {vm })f d and |Ψ̃n−1 (u, vi ; {vm })ab , respectively.
Let e2 = c̄1 , from Eq. (D.3), we can get
c̄ h c2 ···cn d1 ···dn
rc1d 1 (ṽi − ṽj ) T n−2 (vi ; {v̌i , v̌j })ec3 ···en
S
···c c h d ···d (vj ; { v̌i , v̌j })Sb 1 ···bn
(vi ; {v̌i })
2 1 3 n 1 1 2 n
1
ρ1 (ṽi − ṽj )ρ1 (0) 2
c e3 ···en h2 ···hn
= c(ṽi + ṽj )Sh12 ···h (vj ; {v̌i , v̌j })Sdc̄11 ···d (vi ; {v̌i })
T (c1 ) (vi )ρ3 (ṽi + ṽj ) n n
c̄ e3 ···en h2 ···hn
− b(ṽi + ṽj )Sh12 ···h n
(vj ; {v̌i , v̌j })Sdc11···d n
(vi ; {v̌i }) τ1n (vi ; {ṽm })db11 ···d n
···bn . (D.6)
Here we should note that in Eqs. (D.6), (D.9), (D.12), the repeated indices c1 or a1 do not
sum. Using Eq. (D.6) and considering Eq. (B.13), we have
ef (6) fe
x
ξd1 H2,d 1
(6)
= ξd1 H̃2,d
x
(u, vi , vj ) Ψ̃ (u, vi , vj ; {vm })d , (D.7)
1 n−2 1
where
x
H̃2,d 1
(u, vi , vj )
1
=
T (d1 ) (vi )ρ2 (vi , vj )ρ3 (ṽi + ṽj )a11 (vj , vi )
! x (u, v , v )d1 d1 x
H2, (u, vi , vj )d̄1 d̄1 "
d̄ d̄
H2,d i j d1 d1 d̄1
× c(ṽi + ṽj ) 1
+ b(ṽi + ṽj ) 1 1
. (D.8)
a(ṽi − ṽj ) a(ṽi − ṽj )
Let e2 = c̄1 , from Eq. (D.4), we obtain
n−2 c ···c c̄ d2 ···dn
T (vj ; {v̌i , v̌j })ec3 ···en
S 2 n (v ; {v̌i , v̌j })Sb11 ···b
···c c c d ···d j n
(vi ; {v̌i })
3 n 1 2 2 n
1
ρ1 (ṽj − ṽi )ρ1 (0) c1
2
c e3 ···en h2 ···hn
= X2 (vi , vj )Sh12 ···h n
(vj ; {v̌i , v̌j })Sdc̄11 ···d n
(vi ; {v̌i })
ρ4 (vi , vj )
c̄ e3 ···en h2 ···hn
− X1c̄1 (vi , vj )Sh12 ···hn
(vj ; {v̌i , v̌j })Sdc11 ···d n
(vi ; {v̌i })
× τ1n (vj ; {ṽm })db11 ···d n
···bn . (D.9)
Then, by Eq. (D.9), we have
(7)
x
ξd1 H3,d 1
(7)
= ξd1 H̃3,d
x
1
(u, vi , vj )Ψ̃n−2 (u, vi , vj ; {vm })d1 , (D.10)
where
x
H̃3,d 1
(u, vi , vj )
1
=
ρ2 (vj , vi )ρ4 (vi , vj )a11 (vi , vj )

× X2d̄1 (vi , vj )H3,dx
1
(u, vi , vj ) + X1d̄1 (vi , vj )H3,
x
d̄
(u, vi , vj ) . (D.11)
1
Let e2 = ā1 , from Eq. (D.5), we achieve

ā1 h e ···e a ···a
rf d 1 (ṽj + ṽi ) T n−2 (vi ; {v̌i , v̌j })a3 ···an a T n−2 (vj ; {v̌i , v̌j })c3···cn f2 c2
2 1 3 n 1 h1 3 n
c ···c d ···d
× Sd2 ···dn (vj ; {v̌i , v̌j })Sb11···bnn (vi ; {v̌i })
!
2 n
ρ1 (ṽi − ṽj )ρ1 (0) c(ṽi + ṽj ) a1 e3 ···en a2 ···an
= (ā ) S (vj ; {v̌i , v̌j })Sgā11···g (vi ; {v̌i })
T 1 (vj )ρ3 (ṽi + ṽj ) T (ā1 ) (vi ) a2 ···an n
"
b(ṽi + ṽj ) ā1 e3 ···en a2 ···an
− (a ) Sa2 ···an (vj ; {v̌i , v̌j })Sga11···g n
(vi ; { v̌i })
T 1 (vi )
g ···g d ···d
× τ1n (vi ; {ṽm })d11···dnn τ1n (vj ; {ṽm })b11 ···bnn . (D.12)
Using Eq. (D.12) and considering Eq. (B.14), we have
(8)

d1 f Ψn−2 (u, vi , vj ; {vm })f d1
x
ξd1 H4,d 1
(u, vi , vj )ed ed
(8)
= ξd1 H̃4,d
x
(u, vi , vj ) Ψ̃ (u, vi , vj ; {vm })d , (D.13)
1 n−2 1
where
x
H̃4,d 1
(u, vi , vj )
1
=
T (d1 ) (vi )ρ2 (vi , vj )ρ2 (vj , vi )ρ3 (ṽi + ṽj )ρ1 (ṽi − ṽj )
! x d1 d1 x d̄1 d̄1 "
c(ṽi + ṽj ) H4,d1 (u, vi , vj )d1 d1 b(ṽi + ṽj ) H4,d̄1 (u, vi , vj )d̄1 d̄1
× + . (D.14)
T (d1 ) (vj ) a(ṽi + ṽj ) T (d̄1 ) (vj ) a(ṽi + ṽj )
It is easy to get
(5)
x
δd̄1 e2 ξd1 H1,d 1
(5)
= ξd1 H̃1,d
x
1
(u, vi , vj )Ψ̃n−2 (u, vi , vj ; {vm })d1 , (D.15)
where
x
H1,d (u, vi , vj )
x
H̃1,d (u, vi , vj ) = 1
. (D.16)
1
a11 (vi , vj )a11 (vj , vi )
References
[1] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1982.
[2] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation
Function, Cambridge Univ. Press, Cambridge, 1993.
[3] V.A. Tarasov, Theor. Math. Phys. 56 (1988) 793.
[4] P.B. Ramos, M.J. Martins, Nucl. Phys. B 474 (1996) 678.
[5] M.J. Martins, P.B. Ramos, Nucl. Phys. B 500 (1997) 579.
[6] M.J. Martins, P.B. Ramos, Nucl. Phys. B 522 (1998) 413.
[7] G.L. Li, K.J. Shi, R.H. Yue, Nucl. Phys. B 670 (2003) 401.
[8] X.W. Guan, J. Phys. A 33 (2000) 5391.

[9] A. Foerster, X.W. Guan, J. Links, I. Roditi, H.Q. Zhou, Nucl. Phys. B 596 (2001) 525.
[10] X.W. Guan, A. Foerster, U. Grimm, R.A. Römer, M. Schreiber, Nucl. Phys. B 618 (2001) 650.
[11] I.V. Cherednik, Theor. Math. Phys. 17 (1983) 77;
I.V. Cherednik, Theor. Math. Phys. 61 (1984) 911.
[12] E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
[13] H.J. de Vega, Int. J. Mod. Phys. A 4 (1989) 2371.
[14] V. Pasquier, H. Saleur, Nucl. Phys. B 316 (1990) 523.
[15] L. Mezincescue, R.I. Nepomechie, J. Phys. A 24 (1991) L17.
[16] L. Mezincescu, R.I. Nepomechie, Nucl. Phys. B 372 (1992) 597.
[17] A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611;
A. Foerster, M. Karowski, Nucl. Phys. B 408 (1993) 512.
[18] C. Destri, H.J. de Vega, Nucl. Phys. B 374 (1992) 692.
[19] S.M. Fei, R.H. Yue, J. Phys. A 27 (1994) 3715.
[20] H.J. de Vega, A. Gonzalez-Ruiz, Nucl. Phys. B 417 (1994) 553.
[21] C.M. Yung, M.T. Batchelor, Nucl. Phys. B 435 (1995) 430.
[22] R.H. Yue, H. Fan, B.Y. Hou, Nucl. Phys. B 462 (1996) 167.
[23] Y.P. Wang, Phys. Rev. B 56 (1997) 14045.
[24] A. Foerster, J. Links, A. Tonel, Nucl. Phys. B 552 (1999) 707.
[25] H. Frahm, A.A. Zvyagin, J. Phys. C 9 (1997) 9939.
[26] G. Bedürfitg, H. Frahm, J. Phys. A 32 (1999) 4585.
[27] R.I. Nepomechie, J. Phys. A 34 (2001) 9993.
[28] W.L. Yang, Y.Z. Zhang, Nucl. Phys. B 596 (2001) 495.
[29] J.P. Cao, H.Q. Lin, K.J. Shi, Y.P. Wang, Nucl. Phys. B 663 (2003) 487.
[30] R.I. Nepomechie, J. Phys. A 37 (2004) 433.
[31] T. Deguchi, A. Fujii, K. Ito, Phys. Lett. B 238 (1990) 242.
[32] M.J. Martins, P.B. Ramos, J. Phys. A 28 (1995) L525.
Bethe ansatz equations and exact S matrices for the

osp(M|2n) open super-spin chain
D. Arnaudon a , J. Avan b , N. Crampé a , A. Doikou a ,
L. Frappat a,1 , E. Ragoucy a
a Laboratoire d’Annecy-le-Vieux de Physique Théorique, LAPTH, CNRS, UMR 5108,
Université de Savoie, B.P. 110, F-74941 Annecy-le-Vieux cedex, France
b Laboratoire de Physique Théorique et Modélisation, CNRS UMR 8089, Université de Cergy,
5 mail Gay-Lussac, Neuville-sur-Oise, F-95031 Cergy-Pontoise cedex, France
Received 28 October 2003; received in revised form 6 February 2004; accepted 23 March 2004
Abstract
We formulate the Bethe ansatz equations for the open super-spin chain based on the super
Yangian of osp(M|2n) and with diagonal boundary conditions. We then study the bulk and boundary
scattering of the osp(1|2n) open spin chain.
 2004 Published by Elsevier B.V.
MSC: 81R50; 17B37
1. Introduction
The notion of the reflection equation associated with solutions of the Yang–Baxter
equation [1,2], goes back to the key works of Cherednik [3] and Sklyanin [4]. The subject
has recently attracted a great deal of activity as was summarised in [5] (and references
therein). More specifically, starting from a quantum R-matrix R(λ) depending on the
spectral parameter λ and satisfying the (super)-Yang–Baxter equation [1,2,6]
R12 (λ1 − λ2 )R13 (λ1 )R23 (λ2 ) = R23 (λ2 )R13 (λ1 )R12 (λ1 − λ2 ), (1.1)
E-mail address: arnaudon@lapp.in2p3.fr (D. Arnaudon).

1 Member of Institut Universitaire de France.
0550-3213/$ – see front matter  2004 Published by Elsevier B.V.

doi:10.1016/j.nuclphysb.2004.03.027
258 D. Arnaudon et al. / Nuclear Physics B 687 [FS] (2004) 257–278
one derives the reflection equation for an object K(λ) as
R12 (λ1 − λ2 )K1 (λ1 )R12 (λ1 + λ2 )K2 (λ2 )

= K2 (λ2 )R12 (λ1 + λ2 )K1 (λ1 )R12 (λ1 − λ2 ). (1.2)
We have proposed in [5] a classification of c-number solutions K(λ) of the reflection
equation (1.2) for rational (super) Yangian R-matrices [7,8] associated to the infinite series
so(m), sp(2n) and osp(m|2n). This classification entailed K matrices with purely diagonal,
anti-diagonal and mixed (diagonal, anti-diagonal) non-zero entries. The explicit values of
the K matrices were then used within the analytical Bethe ansatz formulation [5,9–12] for
the derivation of the spectrum and the bulk and boundary S-matrices for the so(m), sp(2n)
open spin chains.
There exists a substantial body of work on gl(m|n) super spin chains. Interest in
these systems stemmed from the existence of physically relevant particular cases such as
supersymmetric t–J and extended Hubbard models. They have been the object of many
studies. Supersymmetric t–J models were considered, e.g., in [13] (thermodynamical
aspects), [14] (diagonal boundary K matrices) and [15] (boundary S-matrix). Extended
Hubbard models were considered in [16] (closed chain) and in [17–19] (open chains with
integrable boundary conditions), whilst spin ladder systems associated to some sl(m|n)
superalgebras were obtained in [20,21]. General results for continuum limit of the gl(m|n)
super spin chains were derived in [22].
A natural alternative to these models with gl(m|n) underlying superalgebras is
provided by super spin chains with underlying osp(m|2n) superalgebras. A connection
to intersecting loop models and hence polymer field theories was pointed out in [23],
where the analytical Bethe ansatz equations were written for the closed spin chain.
Algebraic methods were used for some specific cases in [24,25] including nested Bethe
ansatz in [25]. Field theoretical limits were also considered in the literature: the exact
bulk osp(2|2) S-matrix was conjectured in [26] in the framework of disordered systems.
Investigation of the thermodynamics of osp(1|2n) closed spin chains was undertaken in
[27] using the thermodynamical Bethe ansatz formalism. An algebraic construction using
Birman–Wenzl–Murakami algebra then yielded conjectural S-matrices for field theoretical
osp(m|2n) models, and allowed a subsequent thermodynamical Bethe ansatz analysis of
their thermodynamical properties [28]. However a systematic thermodynamic treatment of
these models with more general boundaries is still missing.
Our purpose is to make an exhaustive study of the more complicated case of open spin
chains with osp(m|2n) underlying superalgebra and any integrable (diagonal at a first step)
boundary conditions. The strategy is to establish (insofar as the methods are available)
Bethe ansatz equations for ground state and excited states (note that there is no obvious
relation between closed spin chains Bethe ansatz equations and open spin chain Bethe
ansatz equations, particularly when non-trivial boundary conditions are involved); solve
them within the non-trivial string hypothesis (discussed in the closed case in [13,22]);
and use the results to obtain the S-matrix and thermodynamical quantities, with explicit
evaluation of the effect of boundary conditions.
This paper is our first step in this direction: using the analytical Bethe ansatz method, we
derive the Bethe ansatz equations for all orthosymplectic superalgebras, and all diagonal
D. Arnaudon et al. / Nuclear Physics B 687 [FS] (2004) 257–278 259
K matrices. Restricting ourselves then to osp(1|2n), we solve these equations in the

thermodynamic limit, we derive the ground state and low-lying excitations, and compute
explicitly the bulk and boundary S-matrices. Further generalisations will be left for future
investigations.
2. Bethe ansatz equations for the osp(M|2n) open spin chain
2.1. Conventions and notations
The Bethe ansatz equations will be derived here for the osp(M|2n) N -site open spin
chain with diagonal reflection conditions by means of the analytical Bethe ansatz method
(see, e.g., [9,10,12]). As customary to construct the open chain transfer matrix we introduce
the R-matrix which is a solution of the super-Yang–Baxter equation. We focus on the
osp(M|2n) invariant R-matrix given by [8]
R(λ) = λ(λ + iκ)1 + i(λ + iκ)P − iλQ, 2κ = θ0 (M − 2n − 2), (2.1)
where P is the (super)permutation operator (i.e., X21 ≡ P X12 P )

M+2n
P= (−1)[j ] Eij ⊗ Ej i (2.2)
i,j =1
and

M+2n
Q= (−1)[i][j ] θi θj E¯ı̄ ⊗ Ej i ≡ P t1 . (2.3)
i,j =1
For each index i, we have introduced a conjugate index
ı̄ = M + 2n + 1 − i. (2.4)
We also introduce a sign θi and a Z2 -grading [i] whose definition, due to the conventions
we adopt (see below), depend whether we consider the superalgebra osp(2|2n) or any other
osp(M|2n) superalgebra:
For osp(M|2n) superalgebras, M = 2:

+1 for 1 i M + n,
θi = (2.5)
−1 for M + n + 1 i M + 2n,
(−1)[i] = +1 for 1 i n and M + 1 i M + 2n, (2.6)
[i]
(−1) = −1 for n + 1 i n + M. (2.7)
We will associate to this choice the sign θ0 = −1.
For osp(2|2n) superalgebras:

+1 for 1 i n + 1 and i = 2n + 2,
θi = (2.8)
−1 for n + 2 i 2n + 1,
(−1)[i] = +1 for i = 1 and i = 2n + 2, (2.9)
[i]
(−1) = −1 for 2 i 2n + 1. (2.10)
The sign corresponding to this choice will be θ0 = +1.

The transposition t used in (2.3) and below is defined, for A = ij Aij Eij , by
ij
At = (−1)[i][j ]+[j ] θi θj Aij E¯ı̄ = At Eij . (2.11)
ij ij
The R-matrix (2.1) satisfies crossing and unitarity, namely,

t1
R12 (λ)R12 (−λ) = λ2 + κ 2 λ2 + 1 1, R12 (λ) = R12 (−λ − iκ). (2.12)
We finally define the super trace operation according to the Z2 -grading we have introduced:

M+2n
M+2n
Tr A = (−1)[j ] Ajj for A = Aij Eij . (2.13)
j =1 i,j =1
2.2. Transfer matrix and pseudo-vacuum
The open chain transfer matrix is given by [4]

t (λ) = Tr0 K0+ (λ)T0 (λ)K0− (λ)T̂0 (λ), (2.14)
where Tr0 denotes here the super trace (2.13) over the auxiliary space,
T0 (λ) = R0N (λ)R0,N−1 (λ) · · · R02 (λ)R01 (λ),
T̂0 (λ) = R10 (λ)R20 (λ) · · · RN−1,0 (λ)RN0 (λ), (2.15)
K0− (λ) is any solution of the super boundary Yang–Baxter equation
R12 (λ1 − λ2 )K1 (λ1 )R12 (λ1 + λ2 )K2 (λ2 )
= K2 (λ2 )R12 (λ1 + λ2 )K1 (λ1 )R12 (λ1 − λ2 ) (2.16)
and K0+ (λ) is a solution of a closely related reflection equation defined to be:
t t
R12 (λ2 − λ1 )K11 (λ1 )R12 (−λ1 − λ2 − 2iκ)K22 (λ2 )
= K2t2 (λ2 )R12 (−λ1 − λ2 − 2iκ)K1t1 (λ1 )R12 (λ2 − λ1 ). (2.17)
It is clear that any solution K − (λ) of (2.16), e.g., given in [5], gives rise to a solution
K + (λ) of (2.17), defined by K + (λ) = K − (−λ − iκ)t .
To determine the eigenvalues of the transfer matrix and the corresponding Bethe ansatz
equations, we use the analytical Bethe ansatz method [5,9,10]. We follow the same
procedure as in [5], by imposing certain constraints on the eigenvalues, deduced from the
crossing symmetry of the model, the symmetry of the transfer matrix, the analyticity of
the eigenvalues, and the fusion procedure for open spin chains. These constraints allow to
determine the eigenvalues by solving a set of coupled non-linear consistency equations or
Bethe ansatz equations.
We first describe the case with trivial boundaries, K − (λ) = K + (λ) = 1.
We recall that the fusion procedure for the open spin chain [12,29] yields the fused
transfer matrix
t˜(λ) = ζ (2λ + 2iκ)t (λ)t (λ + iκ) − ζ(λ + iκ)2N q(2λ + iκ)q(−2λ − 3iκ), (2.18)
Fig. 1. Distinguished Dynkin diagrams of the osp(M|2n) superalgebras.
where we define
ζ (λ) = (λ + iκ)(λ + i)(λ − iκ)(λ − i), q(λ) = θ0 (λ − iθ0)(λ − iκ). (2.19)

Note that the value of q(λ) is related to the specific choice of the position of the orthogonal
and the symplectic part in the R-matrix. We choose for the general case (apart from the
osp(2|2n) case) the symplectic part to be “outside” and the orthogonal part to be “inside”.
This formulation corresponds to a specific Dynkin diagram: the so-called distinguished
one (see Fig. 1). For the case of osp(2|2n) the distinguished Dynkin diagram has a special
form (see Fig. 1): it corresponds to the orthogonal part being “outside” and the symplectic
part being inside. These considerations justify the conventions we have adopted in (2.5),
(2.6) and (2.8), (2.9). From the crossing symmetry of the R-matrix (2.12) it follows that:
t (λ) = t (−λ − iκ). The transfer matrix with K − = K + = 1 is obviously osp(M|2n)
invariant, since the corresponding R-matrix (2.1) is osp(M|2n) invariant, namely
[R12 , U1 + U2 ] = 0, (2.20)
where U is any generator of the osp(M|2n) algebra. Finally, from the assumption
of analyticity of the eigenvalues, we require that no singularity appears in the Bethe
eigenvalues. The aforementioned set of constraints uniquely fixes the eigenvalues.
We now choose an appropriate pseudo-vacuum, which is an exact eigenstate of the

transfer matrix; it is the state with all “spins” up, i.e.,
 
1
N
0
|ω+ = |+i where |+ =  
 ...  ∈ C
M+2n
. (2.21)
i=1
0
Our choice of θ0 ensures that this state is always bosonic, whichever orthosymplectic
superalgebra we consider.
After some lengthy computation, we determine explicitly the action of the transfer
matrix as t (λ)|ω+ = Λ0 (λ)|ω+ , where Λ0 (λ) is given by the following expression

2n+M−2
Λ (λ) = a(λ)
0 2N
g0 (λ) + b(λ) 2N
(−1)[l+1] gl (λ)
l=1
+ c(λ) 2N
g2n+M−1 (λ) (2.22)
with
a(λ) = (λ + i)(λ + iκ), b(λ) = λ(λ + iκ), c(λ) = λ(λ + iκ − i). (2.23)
The expressions of the functions gl (λ) depend on the case we consider.
For the generic osp(M|2n) case, M = 2, they are given by (with M = 2m or M =
2m + 1):
λ(λ + iκ
− 2i )(λ + iκ)
gl (λ) = 2
i(l+1)
, l = 0, . . . , n − 1,
(λ + 2 )(λ + 2 )(λ +
iκ il
2 )
λ(λ + iκ2 − 2i )(λ + iκ)
gl (λ) = i(l+1)
, l = n, . . . , n + m − 1,
(λ + 2 )(λ + in − 2 )(λ + in −
iκ il
2 )
λ(λ + iκ)
gn+m (λ) = , if M = 2m + 1,
(λ + i n−m
2 )(λ + i
n−m+1
2 )
gl (λ) = g2n+M−l−1 (−λ − iκ), l = 0, 1, . . . , M + 2n. (2.24)
In this case, we have κ = n + 1 − M2 .
For the case of osp(2|2n), due to the different conventions, one has:
(λ + iκ
+ 2i )(λ + iκ)
g0 (λ) = 2
,
(λ + 2 )(λ + 2 )
iκ i
λ(λ + iκ2 + 2i )(λ + iκ)

gl (λ) = i(l+1)
, l = 1, . . . , n,
(λ + 2 )(λ + i − 2 )(λ + i −
iκ il
2 )
gl (λ) = g2n+1−l (−λ − iκ), l = 0, . . . , 2n + 1. (2.25)
We remind that in this latter case, κ = −n.
2.3. Dressing functions
From the exact expression for the pseudo-vacuum eigenvalue, we introduce the
following assumption for the structure of the general eigenvalues:

2n+M−2
Λ(λ) = a(λ)2N g0 (λ)A0 (λ) + b(λ)2N (−1)[l+1] gl (λ)Al (λ)
l=1
+ c(λ) 2N
g2n+M−1 (λ)A2n+M−1 (λ), (2.26)
where the so-called “dressing functions” Ai (λ) need now to be determined.
We immediately get from the crossing symmetry of the transfer matrix:
Al (λ) = A2n+M−l−1 (−λ − iκ), l = 0, . . . , M + 2n − 1. (2.27)

Moreover, we obtain from the fusion relation (2.18) the following identity, by a comparison
of the forms (2.26) for the initial and fused auxiliary spaces:
A0 (λ + iκ)A2n+M−1 (λ) = 1. (2.28)

Gathering the above two equations (2.27), (2.28) we conclude
A0 (λ)A0 (−λ) = 1. (2.29)

Additional constraints are then imposed on the “dressing functions” from analyticity
properties. Studying carefully the common poles of successive gl ’s, we deduce from
the form of the gl functions (2.24) that gl and gl−1 have common poles at λ = − il2 or
λ = −in + il2 , therefore from analyticity requirements

il il
Al − = Al−1 − , l = 1, . . . , n − 1,
2 2

il il
Al −in + = Al−1 −in + , l = n, . . . , n + m − 1. (2.30)
2 2
There is an extra constraint when M = 2m + 1, namely

ik ik
An+m −in + = An+m−1 −in + . (2.31)
2 2
Having deduced the necessary constraints for the “dressing functions”, we determine them
explicitly. The “dressing functions” Al are basically characterised by a set of parameters
(l)
λj with j = 1, . . . , M (l) , where the integer numbers M (l) are related to the diagonal
generators of osp(M|2n). These generators are defined as:

N
S (l)
= si(l) , s (l) = (ell − el¯l¯)/2, (ekl )ij = δik δj l . (2.32)
i=1
The precise identification of M (l) follows from the symmetry of the transfer matrix (see
also [9]):
S (l) = M (l−1) − M (l) , l = 1, . . . , n − 1, n + 1, . . . , n + m − 2, (2.33)

S (n)
=M (n−1)
− 2M (n)
, (2.34)
S (n+m−1) = M (n+m−2) − M (n+m−1) ,
S (n+m) = M (n+m−1) − M (n+m) , if M = 2m + 1, (2.35)
S (n+m−1) = M (n+m−2) − M (+) − M (−) ,
S (n+m) = M (+) − M (−) , if M = 2m (2.36)
and M (0) = N
2.
2.3.1. osp(2m + 1|2n)

The dressing functions take the form:
(1)

M λ + λ(1)
j −
i
2 λ − λ(1)
j −
i
2
A0 (λ) = (1) (1)
,
j =1 λ + λj + i
2 λ − λj + i
2
(l)

M λ + λ(l)
j +
il
2 + i λ − λ(l)
j +
il
2 +i
Al (λ) = (l) (l)
j =1 λ + λj + il
2 λ − λj + il
2
(l+1)
M (l+1) (l+1)
λ + λj + il
2 − i
2 λ − λj + il
2 − i
2
× (l+1) (l+1)
,
j =1 λ + λj + il
2 + i
2 λ − λj + il
2 + i
2
l = 1, . . . , n − 1,
(l)

M λ + λ(l)
j + in −
il
2 − i λ − λ(l)
j + in −
il
2 −i
Al (λ) = (l) (l)
j =1 λ + λj + in − il
2 λ − λj + in − il
2
(l+1)
M (l+1) (l+1)
λ + λj + in − il
2 + i
2 λ − λj + in − il
2 + i
2
× ,
j =1 λ + λ(l+1)
j + in − il
2 − i
2 λ − λ(l+1)
j + in − il
2 − i
2
l = n, . . . , n + m − 1,
(k)

M λ + λ(k)
j + in −
ik
2 + i λ − λ(k)
j + in −
ik
2 +i
An+m (λ) = (k) (k)
j =1 λ + λj + in − ik
2 λ − λj + in − ik
2
(k) (k)
λ + λj + in − ik
2 − i
2 λ − λj + in − ik
2 − i
2
× (k) (k)
, (2.37)
λ + λj + in − ik
2 + i
2 λ − λj + in − ik
2 + i
2
and Al (λ) = A2n+2m−l (−λ − iκ) for l > n + m, κ = n − m + 12 .

2.3.2. osp(2m|2n) with m > 1

The dressing functions are the same as in the previous case for l = 0, . . . , n + m − 3,
but
(k−2)
M λ + λ(k−2)
j + in − ik
2 λ − λ(k−2)
j + in − ik
2
An+m−2 (λ) =
j =1 λ + λ(k−2)
j + in − ik
2 + i λ − λ(k−2)
j + in − ik
2 +i
(+) (+) (+)

M λ + λj 2 + 2
ik 3i
+ in − λ − λj + in −
2 + 2
ik 3i
× (+) (+)
j =1 λ + λj + in − 2 + 2 λ − λj + in − ik2 + 2i
ik i
(−) (−) (−)

M λ + λj 2 + 2
ik 3i
+ in − λ − λj + in −
2 + 2
ik 3i
× (−)
,
j =1 λ + λj + in − 2 + 2
ik i
λ − λ(−)
j + in − 2 + 2
ik i
(+)

M λ + λ(+)
j + in −
2 + 2
ik 3i
λ − λ(+)
j + in − 2 + 2
ik 3i
An+m−1 (λ) = (+) (+)
j =1 λ + λj + in − 2 + 2 λ − λj + in − ik2 + 2i
ik i
(−) (−) (−)

M λ + λj + in − ik
2 − i
2 λ − λj + in − ik
2 − i
2
× , (2.38)
j =1 λ + λ(−)
j + in −
ik
2 + i
2 λ − λ(−)
j + in −
ik
2 + i
2
and Al (λ) = A2n+2m−l−1 (−λ − iκ) for l > n + m − 1, κ = n + 1 − m.
2.3.3. osp(2|2n)
As already mentioned, this case must be treated separately, because of the different
position of the orthogonal and symplectic parts in the R-matrix. Here the orthogonal part
of the R-matrix is considered to be “outside” and the symplectic part “inside” as opposed
to the previous cases. The corresponding dressing functions have the form
(1)

M λ + λ(1)
j −
i
2 λ − λ(1)
j −
i
2
A0 (λ) = (1) (1)
,
j =1 λ + λj + i
2 λ − λj + i
2
(l) (l) (l)

M λ + λj − il
2 λ − λj − il
2
Al (λ) =
j =1 λ + λ(l)
j +i −
il
2 λ − λ(l)
j +i −
il
2
(l+1)
M λ + λ(l+1)
j + 2 − 2
3i il
λ − λ(l+1)
j + 2 − 2
3i il
× (l+1) (l+1)
,
j =1 λ + λj + 2i − il2 λ − λj + 2i − il2
l = 1, . . . , n − 1,
(n) (n) (n)

M λ + λj − in
2 λ − λj − in
2
An (λ) =
j =1 λ + λ(n)
j +i −
in
2 λ − λ(n)
j +i−
in
2
(n+1)
M λ + λ(n+1)
j + 2i − in
2 λ − λ(n+1)
j + 2i − in
2
× (n+1) (n+1)
, (2.39)
j =1 λ + λj − in
2 λ − λj − in
2
and Al (λ) = A2n+1−l (−λ − iκ) for l > n, κ = −n.
2.4. Bethe ansatz equations
We define the function
λ+ ix
ex (λ) = 2
. (2.40)
λ− ix
2
From the analyticity requirements one obtains the Bethe ansatz equations which read as:
2.4.1. osp(2m + 1|2n)
(1)
(1) 2N
M
(1) (1) (1) (1)
e1 λi = e2 λi − λj e2 λi + λj
j =1,j =i
(2)

M
(2) (1) (2)
× e−1 λ(1)
i − λj e−1 λi + λj ,
j =1
(l)

M
(l) (l) (l)
1= e2 λ(l)
i − λj e2 λi + λj
j =1,j =i
(l+τ )
M (l) (l+τ ) (l) (l+τ )
× e−1 λi − λj e−1 λi + λj ,
τ =±1 j =1
l = 2, . . . , n + m − 1, l = n,
(n+1)
M
(n+1) (n)
1= e1 λ(n)
i − λj e1 λi + λ(n+1)
j
j =1
(n−1)
M
(n) (n−1) (n) (n−1)
× e−1 λi − λj e−1 λi + λj ,
j =1
(n+m)
M
(n+m) (n+m) (n+m) (n+m)
1= e1 λi − λj e1 λi + λj
j =1,j =i
M (n+m−1)
(n+m)
× e−1 λ(n+m)
i − λ(n+m−1)
j e−1 λi + λ(n+m−1)
j . (2.41)
j =1
In particular for M = 1 the Bethe ansatz equations become

(1)
(1) 2N
M
(1) (1) (1) (1)
e1 λi = e2 λi − λj e2 λi + λj
j =1,j =i
(2)

M
(2) (1) (2)
× e−1 λ(1)
i − λj e−1 λi + λj ,
j =1
M (l)
(l) (l) (l) (l)
j =1,j =i
(l+τ )
M (l+τ ) (l+τ )
× e−1 λ(l)
i − λj e−1 λ(l)
i + λj ,
τ =±1 j =1
l = 2, . . . , n − 1,
(n)

M
(n) (n) (n) (n) (n) (n) (n) (n)
1= e−1 λi − λj e−1 λi + λj e2 λi − λj e2 λi + λj
j =1,j =i
(n−1)
M
(n−1) (n−1)
× e−1 λ(n)
i − λj e−1 λ(n)
i + λj . (2.42)
j =1
2.4.2. osp(2m|2n) with m > 1

The first n + m − 3 equations are the same as in the previous case for M = 2m + 1, see
Eq. (2.41), but the last three equations are modified, and they become identical to the last
three equations of the so(2n + 2m) open spin chain, namely,
M (n+m−2)
(n+m−2) (n+m−2) (n+m−2) (n+m−2)
j =1,j =i
M (n+m−3)
(n+m−2)
× e−1 λ(n+m−2)
i − λ(n+m−3)
j e−1 λi + λ(n+m−3)
j
j =1
(τ )
M
) (n+m−2) )
× e−1 λ(n+m−2)
i − λ(τ
j e−1 λi + λ(τ
j ,
τ =± j =1
(τ )

M
(τ ) (τ ) (τ ) (τ )
j =1,j =i
M (n+m−2)
) (n+m−2) ) (n+m−2)
× e−1 λ(τ
i − λj e−1 λ(τ
i + λj , τ = ±. (2.43)
j =1
2.4.3. osp(2|2n)
(2)
(1) 2N M (1) (2) (1) (2)
e1 λi = e1 λi − λj e1 λi + λj ,
j =1
M (l)
(l) (l) (l)
1= e2 λ(l)
i − λj e2 λi + λj
j =1,j =i
(l+τ )
M (l) (l+τ ) (l) (l+τ )
× e−1 λi − λj e−1 λi + λj ,
τ =±1 j =1
l = 2, . . . , n − 1,
(n+1)
M
(n+1) (n+1)
1= e−2 λ(n)
i − λj e−2 λ(n)
i + λj
j =1,
(n−1)
M
(n) (n−1) (n) (n−1)
× e−1 λi − λj e−1 λi + λj
j =1
(n)

M
(n) (n) (n) (n)
× e2 λi − λj e2 λi + λj ,
j =1,j =i
(n+1)
M
(n+1) (n+1) (n+1) (n+1)
j =1,j =i
(n)

M
(n+1)
× e−2 λ(n+1)
i − λ(n)
j e−2 λi + λ(n)
j . (2.44)
j =1
In particular the equations for the osp(2|2) open chain are given by
(2)
2N M (1) (1) (2)
e1 λ(1)
i = e2 λi − λ(2)
j e2 λi + λj ,
j =1
M (1)
(1) (2) (1)
1= e−2 λ(2)
i − λj e−2 λi + λj
j =1
(2)

M
(2) (2) (2)
× e4 λ(2)
i − λj e4 λi + λj . (2.45)
j =1,j =i
Notice that there is a one-to-one correspondence between the distinguished Dynkin

diagrams (see Fig. 1) and the Bethe ansatz equations derived for each case. The Bethe
ansatz equations for osp(1|2n), osp(2|2n), osp(2m|2), osp(2m+1|2) can now be compared
with the corresponding results obtained in [25] for super spin chains with periodic
boundaries. Let us point out however that we derived explicitly the Bethe ansatz equations
for any osp(M|2n) open spin chain, and we expect the corresponding equations for a chain
with periodic boundary conditions to be “halved” (i.e., half of the factors in the products
should be missing) compared to the ones we found.
2.5. Non-trivial diagonal boundary conditions
Until now we have derived the Bethe ansatz equations for trivial boundary conditions,
namely K − = K + = 1. We shall now insert non-trivial boundary effects and then rederive
the modified Bethe ansatz equations. We choose K − to be one of the diagonal solutions
D1, D2, D3 found in Proposition 3.1 of [5]. We consider, for simplicity but without loss
of generality, K + = 1. Note that the pseudo-vacuum remains an exact eigenstate after this
modification. We rewrite the solutions D1, D2 and D3 of [5] in a slightly modified notation,
which we are going to use from now on.
D1. The solution D1 can be written in the following form
K(λ) = diag(α, . . . , α, β, . . . , β). (2.46)

The number of α’s is equal to the number of β’s, so that this solution exists only for the
osp(2m|2n) cases as stated in Proposition 3.1 of Ref. [5], and
α(λ) = −λ + iξ, β(λ) = λ + iξ, (2.47)

where ξ is the free boundary parameter.
D2. Solution D2 can be written in the osp(M|2n) case (M > 2) as
K(λ) = diag(1, . . . , 1, ᾱ, 1, . . . , 1, β̄, 1, . . . , 1) (2.48)

n M−2 n
and for the osp(2|2n) case as
K(λ) = diag(ᾱ, 1, . . . , 1, β̄) (2.49)

2n
with
−λ + iξ1 −λ + iξn
ᾱ(λ) = , β̄(λ) = , (2.50)
λ + iξ1 λ + iξn
where ξ1 and ξn are the boundary parameters which satisfy the constraint
ξ1 + ξn = κ − θ0 . (2.51)
Obviously, this solution does not exist for the osp(1|2n) superalgebras.
D3. Solution D3 has the form in the osp(M|2n) case (M = 2)
K(λ) = diag(β, . . . , β , α, . . . , α , β, . . . , β , α, . . . , α , β, . . . , β ). (2.52)

n−n1 n1 +m1 M−2m1 n1 +m1 n−n1
while for the osp(2|2n) case it takes the form

K(λ) = diag(α, α, . . . , α , β, . . . , β , α, . . . , α , α). (2.53)

n1 2n−2n1 n1
The osp(1|2n) case is recovered by taking M = 1 and m1 = 0 in (2.52).

Again, α and β are given by
α(λ) = −λ + iξ, β(λ) = λ + iξ, (2.54)
where ξ = (κ + 2m1 − 2n1 − 1)/2 has a fixed value, the integers m1 and n1 being restricted
to 0 n1 n and 0 m1 m (M = 2m or M = 2m + 1), respectively, n1 and m1 being
neither both zero nor taking maximal values simultaneously.
We now come to the explicit expression of the eigenvalues when K − is one of the
above mentioned solutions. We should point out that the dressing functions are related to
the bulk behaviour of the chain and thus they are form-invariant under changes of boundary
conditions. Indeed the only modifications in the expression of the eigenvalues (2.26) occur
in the gl functions, which basically characterise the boundary effects. We call the new gl
functions g̃l .
D1. As already mentioned, the solution D1 can only be applied to osp(2m|2n) with
m 1. In this case we have
g̃l (λ) = (−λ + iξ )gl (λ), l = 0, . . . , n + m − 1,
g̃l (λ) = (λ + iξ + iκ)gl (λ), l = n + m, . . . , 2n + 2m − 1, (2.55)
where gl (λ) are given by (2.24), (2.25). The system with such boundaries has a residual
symmetry sl(m|n), which immediately follows from the structure of the corresponding K
matrix.
D2. We have to separate the cases osp(M|2n) with M = 2 and osp(2|2n). In the
osp(M|2n) case with M = 2, one gets
(−λ + iξ1 − in)
g̃l (λ) = gl (λ), l = 0, . . . , n − 1, g̃n (λ) = gn (λ),
(λ + iξ1 )
(λ + iξ1 − i)
g̃l (λ) = gl (λ), l = n + 1, . . . , n + M − 2,
(λ + iξ1 )
(λ + iξ1 − i) (−λ + iξn − in + iM − 3i)
g̃n+M−1 (λ) = gn+M−1 (λ),
(λ + iξ1 ) (λ + iξn )
(λ + iξ1 − i) (λ + iξn − i)
g̃l (λ) = gl (λ), l = n + M, . . . , 2n + M − 1. (2.56)
(λ + iξ1 ) (λ + iξn )
In the osp(2|2n) case, the formulae are similar to the sp(2n) case:
(−λ + iξ1 )
g̃0 (λ) = g0 (λ),
(λ + iξ1 )
(λ + iξ1 + i) (λ + iξ1 + iκ)
g̃2n+1 (λ) = g2n+1 (λ),
(λ + iξ1 ) (−λ − iκ + iξ1 + i)
(λ + iξ1 + i)
g̃l (λ) = gl (λ), l = 1, . . . , 2n. (2.57)
(λ + iξ1 )
The functions gl (λ) are given by (2.24), (2.25). The system with such boundaries has a
residual symmetry osp(M − 2|2n).
D3. For the D3 solution we find the following modified g functions in the case of
osp(M|2n) with M = 2:
g̃l (λ) = (λ + iξ )gl (λ), l = 0, . . . , n − n1 − 1,
g̃l (λ) = (−λ + iξ − in + in1 )gl (λ), l = n − n1 , . . . , n + m1 − 1,

κ i
g̃l (λ) = λ + i − gl (λ), l = n + m1 , . . . , n + m − 1,
2 2
(λ + i κ2 − 2i )
g̃M+2n−1−l (λ) = g̃l (−λ − iκ), l = 0, . . . , n + m − 1, (2.58)
(λ + i κ2 + 2i )
and in the case of osp(2m + 1|2n)

κ i
g̃n+m (λ) = λ + i − gn+m (λ). (2.59)
2 2
In the osp(2|2n) case, the formulae become
g̃l (λ) = (−λ + iξ )gl (λ), l = 0, . . . , n1 ,

κ i
g̃l (λ) = λ + i + gl (λ), l = n1 + 1, . . . , 2n − n1 ,
2 2
(λ + i κ2 + 2i )
g̃l (λ) = (−λ − iκ − iξ ) gl (λ), l = 2n − n1 + 1, . . . , 2n + 1. (2.60)
(λ + i κ2 − 2i )
The system with such boundaries has a residual symmetry osp(M − 2m1 |2n − 2n1 ) ⊕
osp(2m1 |2n1 ).
We now formulate the Bethe ansatz equations for the general diagonal solutions. The
only modifications induced on Bethe ansatz equations are the following for each solution:
−1
D1. The factor −e2ξ +κ (λ) appears in the LHS of the (n + m)th Bethe equation.
D2. The factor −e2ξ1 −n (λ) appears in the LHS of the nth Bethe equation. The factor
−1
−e2ξ 1 −n−1
(λ) appears in the LHS of the (n + 1)th Bethe equation.
D3. The factor −e2ξ −(n−n1 ) (λ) appears in the LHS of the (n − n1 )th Bethe equation.
−1
The factor −e2ξ +2n1 −m1 −n (λ) appears in the LHS of the (n + m1 )th Bethe equation.
3. Scattering for the osp(1|2n) open spin chain
3.1. Low-lying excitations
Before we derive explicitly the bulk and boundary scattering amplitudes for the
osp(1|2n) case we first need to determine the ground state and the low-lying excitations of
the model. We recall that the energy is derived via the relation H = d
dλ t (λ)|λ=0 . It is given
by
(1)
1
M
1
E=− . (3.1)
2π (1) 2
j =1 (λj ) + 4
1
In what follows we write the Bethe ansatz equations for the ground state and the low-
lying excitations (holes) of the models under study. Bethe ansatz equations may in general
only be solved in the thermodynamic limit N → ∞. In this limit, a state is described in
particular by the density functions σ l (λ) of the parameters λ(l)
i .
3.1.1. osp(1|2)
Let us first consider the osp(1|2) case, for which the ground state consists of one filled
Dirac sea with real strings (all λi ’s real). The set of Bethe ansatz equations for the osp(1|2)
case takes the form
e1 (λi )2N e1 (λi )e− 1 (λi )

2

M
= e2 (λi − λj )e2 (λi + λj )e−1 (λi − λj )e−1 (λi + λj ). (3.2)
j =1
The reason why we study this case separately is basically because we wish to point out the
striking similarity between the latter Bethe ansatz equations (3.2) and the corresponding
equations appearing in the study of the SU(3) open spin chain with “soliton non-
preserving” boundary conditions [12]. This is indeed a remarkable connection, which
can presumably be extended to open spin chains with “soliton non-preserving” boundary
conditions, for higher rank algebras. In particular, we expect that for any SU(n) (n odd)
chain with “soliton non-preserving” boundary conditions the resulting Bethe ansatz
equations will have the same form as in the osp(1|n − 1) open spin chain with certain
diagonal boundaries. We hope to report on this in detail elsewhere.
We now study the low-lying excitations, which are holes in the filled Dirac sea. In order
to convert the sums into integrals, after taking the thermodynamic limit (N → ∞), we
employ the following approximate relation
∞ (l)
1 (l) 1 (l)
M ν
1
f λi = dλ f (λ)σ l (λ) − f λ̃i − f (0), (3.3)
N N 2N
i=1 0 i=1
where the correction terms take into account the ν (l) holes located at values λ̃(l)
i and the
halved contribution at 0+ . For osp(1|2) in particular ν (l) ≡ ν and λ̃(l)
i ≡ λ̃i .
We shall denote by fˆ(ω) the Fourier transform of any function f (λ). Once we take the
logarithm and the derivative of (3.2), we derive the densities from the equation
1
K̂(ω)σ̂ (ω) = â1 (ω) + F̂ (ω), (3.4)
N
ω
where a (λ) = i d
2π dλ ln e (λ) and â (ω) = e− 2 . Moreover
ω cosh 3ω
K̂(ω) = e− 2 4
,
cosh ω4

ν

F (λ) = a2 (λ) − a 1 (λ) + (a2 − a1 )(λ − λ̃i ) + (a2 − a1 )(λ + λ̃i ) . (3.5)
2
i=1
Finally Eq. (3.4) can be written as
1
σ (λ) = 2(λ) + Φ(λ), (3.6)
N
where
ˆ (ω) = â1 (ω)R̂(ω), Φ̂(ω) = R̂(ω)F̂ (ω), and R̂(ω) = K̂−1 (ω). (3.7)
cosh ω4
In particular the energy can be written in terms of hyperbolic functions as ˆ (ω) = .
cosh 3ω
4
3.1.2. osp(1|2n)
We recall that we can only consider the D3 solution in this case. The ground state
consists of n filled Dirac seas with real strings. With the help of relation (3.3) we derive
the densities that describe the state with ν (l) holes in the l sea from the equation
1 1
K̂(ω)σ̂ (ω) = â(ω) + F̂ (ω) + Ĝ(ω, ξ ), (3.8)
N N
where we have introduced
 
  σ 1 (λ)
2a1 (λ)  .. 
 0   . 
 
a(λ) = 
 ..  , σ (λ) =  σ l (λ)  . (3.9)
.   . 
 .. 
0
σ n (λ)
F (λ) is a n-vector as well with
(l)

ν

F (λ) = a1 (λ)δj 1 − a1 (λ) + a2 (λ) +
j
a2 λ − λ̃(l)
i + a2 λ + λ̃(l)
i δlj
i=1
ν (l)
(l) (l)
− a1 λ − λ̃i + a1 λ + λ̃i (δj,l+1 + δj,l−1 ),
i=1
j = 1, . . . , n − 1,
F n (λ) = a1 (λ) − a 1 (λ) + a2 (λ)

2
ν (l)

+ (a2 − a1 ) λ − λ̃(l)
i + (a2 − a1 ) λ + λ̃(l)
i δln
i=1
(l)

ν
(l) (l)
− a1 λ − λ̃i + a1 λ + λ̃i δl,n−1 (3.10)
i=1
and K̂ is a n × n matrix with entries given by

K̂ij (ω) = 1 + â2 (ω) δij − â1 (ω)(δi,j +1 + δi,j −1 ), i, j = 1, . . . , n − 1,
K̂nn−1 (ω) = K̂n−1n (ω) = â1 (ω), K̂nn (ω) = 1 − â1 (ω) + â2 (ω). (3.11)
Finally, the n-vector G carries all the explicit dependence on the boundary parameter ξ of
the D3 solution
Gj (λ) = a2ξ −ñ1 (λ)δj,ñ1 , (3.12)
where ñ1 = n − n1 . Solving equation (3.8) we find the densities σ i which describe a Bethe
ansatz state. The solution of (3.8) has the following form
1 1
σ (λ) = 2(λ) + Φ0 (λ) + Φ1 (λ, ξ ), (3.13)
N N
where and Φ0,1 are n-vectors (columns) with

n
ˆ i (ω) = R̂i1 (ω)â1 (ω), Φ̂0i (ω) = R̂ij (ω)F̂ j (ω),
j =1

n
Φ̂1i (ω, ξ ) = R̂ij (ω)Ĝj (ω, ξ ). (3.14)
j =1
R̂ = K̂−1 and j is the energy of a hole in the j sea; they are written in terms of hyperbolic
functions as
ω sinh(min(i, j ) ω2 ) cosh(n + 1
− max(i, j )) ω2
R̂ij (ω) = e 2 2
, i, j = 1, . . . , n, (3.15)
cosh(n + 12 ) ω2 sinh ω2
cosh(n + 1
− j ) ω2
ˆ j (ω) = 2
, j = 1, . . . , n. (3.16)
cosh(n + 12 ) ω2
3.2. Scattering
As already mentioned the main aim here is the derivation of the exact bulk and boundary
S-matrices. We follow the standard formulation developed by Korepin [30], and later by
Andrei and Destri [31]. One first implement the so-called quantisation condition,
2iNpl
e S − 1 λ̃l1 , λ̃l2 = 0, (3.17)
where pl is the momentum of the particle (in our case, the hole) with rapidity λ̃l1 . For the
case of ν (even) holes in l sea we insert the integrated density (3.13) into the quantisation
condition (3.17). We use the dispersion relation
1 d l
l (λ) = p (λ) (3.18)
2π dλ
λ̃
and the sum rule N 0 i dλ σ (λ) ∈ Z+ . We end up with the following expression for the
boundary scattering amplitudes:
λ̃1

α +l α −l = exp 2πN dλ σ l (λ) − 2 l (λ) (3.19)
0
with
α −l (λ, ξ ) = k0 (λ)k1 (λ, ξ ), α +l (λ) = k0 (λ), (3.20)

where α +l is realised just as the overall factor in front of the unit matrix at the left boundary
(recall that K + = 1, whereas K − is given by the solution D3). Moreover,
λ̃l1 l
λ̃1
l l
k0 λ̃1 = exp iπ dλ Φ0l (λ) , k1 λ̃1 , ξ = exp 2iπ dλ Φ1l (λ, ξ ) ,
0 0
(3.21)
with Φlgiven by (3.14), (3.10). We finally restrict ourselves to l = 1 in the first sea and we
write the latter expression in term of the Fourier transform of Φ 1 (3.14),
∞
1 dω 1 −iωλ
k0 (λ) = exp − Φ̂ (ω)e ,
2 ω 0
−∞
∞
dω 1 −iωλ
k1 (λ, ξ ) = exp − Φ̂ (ω, ξ )e . (3.22)
ω 1
−∞
Let us discuss first the form of the exact bulk S-matrix. It is easy to compute the
scattering amplitude between two holes in the first sea. The bulk scattering amplitude
comes from the contribution of the terms of Φ 1 given by Eqs. (3.10), (3.14), (3.15), with
argument λ ± λ̃j . After some algebra and using the following identity
∞ µω
1 dω e− 2 Γ µ+1
= ln µ+3
4
(3.23)
2 ω cosh ω2 Γ
0 4
we conclude that the hole–hole scattering amplitude is given by the expression

iλ−1 iλ −iλ iλ+1 1
tan π 2n+1 Γ 2n+1 Γ 2n+1 + 12 Γ −iλ+1 2n+1 Γ 2n+1 + 2
S0 (λ) = iλ+1 −iλ iλ iλ+1 −iλ+1 1 . (3.24)
tan π 2n+1 Γ 2n+1 Γ 2n+1 + 2 Γ 2n+1 Γ 2n+1 + 2
1
As a consistency check we compute one further eigenvalue of the S-matrix. In particular if

one considers the state with two holes in the first Dirac sea, and a two-string located at the
midpoint of the two holes, the corresponding eigenvalue is given by (see also [32])
Sb (λ) = e1 (λ)S0 (λ). (3.25)
The explicit bulk S-matrix, which is a solution of the super-Yang–Baxter equation has the
following structure
S0 (λ)
S(λ) = λ(λ + iκ)1 + i(λ + iκ)P − iλQ . (3.26)
(iλ + κ)(iλ + 1)
We now give the expressions for the boundary S-matrix, which follow from (3.21), (3.22),
and the duplication formula for the Γ function

1 1
2 2x−1
Γ x+ Γ (x) = π 2 Γ (2x). (3.27)
2
The overall factor k0 , (3.21), is given by

−iλ

iλ
Γ Γ + 3 Γ iλ
2n+1 + 1
2(2n+1) + 3
4
k0 (λ) = Y0 (λ) 2n+1
−iλ
2n+1 4
−iλ
Γ 2n+1 iλ
Γ 2n+1 + 3
4 Γ 2n+1 + 1
2(2n+1) + 3
4

−iλ

Γ 2n+1 + 1
2(2n+1) + 1
2
× iλ , (3.28)
Γ 2n+1 + 1
2(2n+1) + 1
2
where

sin π iλ
2n+1 + 1
2(2n+1) − 1
4 sin π iλ
2n+1 − 1
2(2n+1) + 1
2
Y0 (λ) = iλ iλ
sin π 2n+1 − 1
2(2n+1) + 1
4 sin π 2n+1 + 1
2(2n+1) − 1
2

tan π iλ
2n+1 + 1
4(2n+1) + 1
4
× . (3.29)
tan π iλ
2n+1 − 1
4(2n+1) − 1
4
The ξ dependent part for the D3 solution k1 , (3.21) reads

ξ
−iλ ξ
Γ iλ
2n+1 + 2n+1 + 1
2 Γ 2n+1 + 2n+1
k1 (λ, ξ ) = −iλ ξ iλ ξ
Γ 2n+1 + 2n+1 + 1
2 Γ 2n+1 + 2n+1
−iλ
ξ −ñ1 ξ −ñ1
Γ 2n+1 + 2n+1 + 1
2 Γ iλ
2n+1 + 2n+1 +1
× iλ ξ −ñ1 −iλ ξ −ñ1 , (3.30)
Γ 2n+1 + 2n+1 + 1
2 Γ 2n+1 + 2n+1 +1
where ξ = ξ − 12 has the fixed value found for D3, so that the boundary S-matrix satisfies
the reflection equation. Note that our solutions include the necessary CDD factors both for
the bulk and boundary matrices.
Acknowledgements
We are thankful to R.I. Nepomechie for useful suggestions. This work was supported
by the TMR Network EUCLID: “Integrable models and applications: from strings to
condensed matter”, contract number HPRN-CT-2002-00325.
References
[1] R.J. Baxter, Partition function of the eight-vertex lattice model, Ann. Phys. 70 (1972) 193;
R.J. Baxter, J. Stat. Phys. 8 (1973) 25;
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, San Diego, 1982.
[2] V.E. Korepin, New effects in the massive Thirring model: repulsive case, Commun. Math. Phys. 76 (1980)
165;
V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method, Correlation Functions and
Algebraic Bethe Ansatz, Cambridge Univ. Press, Cambridge, 1993.
[3] I.V. Cherednik, Factorizing particles on a half line and root systems, Theor. Math. Phys. 61 (1984) 977.
[4] E.K. Sklyanin, Boundary conditions for integrable quantum systems, J. Phys. A 21 (1988) 2375.
[5] D. Arnaudon, J. Avan, N. Crampé, A. Doikou, L. Frappat, E. Ragoucy, Classification of reflection matrices
related to (super) Yangians and application to open spin chain models, Nucl. Phys. B 668 (2003) 469,
math.QA/0304150.
[6] P.P. Kulish, Integrable graded magnets, J. Sov. Math. 35 (1986) 2648.
[7] V.G. Drinfel’d, Hopf algebras and the quantum Yang–Baxter equation, Sov. Math. Dokl. 32 (1985) 254;
V.G. Drinfel’d, A new realization of Yangians and quantized affine algebras, Sov. Math. Dokl. 36 (1988)
212.
[8] D. Arnaudon, J. Avan, N. Crampé, L. Frappat, E. Ragoucy, R-matrix presentation for (super)-Yangians
Y (g), J. Math. Phys. 44 (2003) 302, math.QA/0111325.
[9] V.I. Vichirko, N.Yu. Reshetikhin, Excitation spectrum of the anisotropic generalization of an SU 3 magnet,
Theor. Math. Phys. 56 (1983) 805;
N.Yu. Reshetikhin, A method of functional equations in the theory of exactly solvable quantum systems,
Lett. Math. Phys. 7 (1983) 205;
N.Yu. Reshetikhin, Sov. Phys. JETP 7 (1983) 691;
N.Yu. Reshetikhin, Integrable models of quantum one-dimensional magnets with O(n) and Sp(2k)
symmetry, Theor. Math. Phys. 63 (1985) 555;
N.Yu. Reshetikhin, The spectrum of the transfer matrices connected with Kac–Moody algebras, Lett. Math.
Phys. 14 (1987) 235.
[10] L. Mezincescu, R.I. Nepomechie, Analytical Bethe ansatz for quantum algebra invariant spin chains, Nucl.
Phys. B 372 (1992) 597.
(2)
[11] S. Artz, L. Mezincescu, R.I. Nepomechie, Spectrum of transfer matrix for Uq (Bn ) invariant A2n open spin
chains, Int. J. Mod. Phys. A 10 (1995) 1937, hep-th/9409130;
(2) (1) (1) (1)
S. Artz, L. Mezincescu, R.I. Nepomechie, Analytical Bethe ansatz for A2n−1 , Bn , Cn , Dn quantum
algebra invariant open spin chains, J. Phys. A 28 (1995) 5131, hep-th/9504085.
(1)
[12] A. Doikou, Fusion and analytical Bethe ansatz for the An−1 open spin chain, J. Phys. A 33 (2000) 4755;
A. Doikou, Quantum spin chain with “soliton nonpreserving” boundary conditions, J. Phys. A 33 (2000)
8797.
[13] K.J.B. Lee, P. Schlottman, Soluble one-dimensional narrow-band model with arbitrary spin S and possible
relevance to heavy-fermions and resonating valence bonds, J. Phys. Colloq. 49 (C8) (1988) 709;
P. Schlottman, Thermodynamics of the degenerate supersymmetric t–J model in one dimension, J. Phys.
C 4 (1992) 7565.
[14] A. González-Ruiz, Integrable open-boundary conditions for the supersymmetric t–J model. The quantum
group invariant case, Nucl. Phys. B 424 (1994) 468, hep-th/9401118.
[15] F.H.L. Eßler, The supersymmetric t–J model with a boundary, J. Phys. A 29 (1996) 6183, cond-
mat/9605180.
[16] F.H.L. Eßler, V. Korepin, K. Schoutens, Exact solution of an electronic model of superconductivity in 1 + 1
dimensions, cond-mat/9211001;
F.H.L. Eßler, V. Korepin, Spectrum and low lying excitations in a supersymmetric extended Hubbard model,
cond-mat/9307019.
[17] D. Arnaudon, Algebraic approach to q-deformed supersymmetric variants of the Hubbard model with pair
hoppings, JHEP 9712 (1997) 006, physics/9711001.
[18] X.Y. Ge, M.D. Gould, Y.-Z. Zhang, H.-Q. Zhou, Integrable eight-state supersymmetric U model with
boundary terms and its Bethe ansatz solution, cond-mat/9709308;
A. Bracken, X.Y. Ge, Y.-Z. Zhang, H.-Q. Zhou, Integrable open-boundary conditions for the q-deformed
supersymmetric U model of strongly correlated electrons, Nucl. Phys. B 516 (1998) 588, cond-
mat/9710141;
A. Bracken, X.Y. Ge, Y.-Z. Zhang, H.-Q. Zhou, An open-boundary integrable model of three coupled XY
spin chains, Nucl. Phys. B 516 (1998) 603, cond-mat/9710171;
Y.-Z. Zhang, H.-Q. Zhou, New integrable boundary conditions for the q-deformed supersymmetric U model
and Bethe ansatz equations, Phys. Lett. A 244 (1998) 427, cond-mat/9711238, and references therein.
[19] X.-W. Guan, A. Foerster, U. Grimm, R.A. Romer, M. Schreider, A supersymmetric Uq (osp(2|2))-extended
Hubbard model with a boundary, Nucl. Phys. B 618 (2001) 650, cond-mat/0106511.
[20] A. Foerster, K.E. Hibberd, J.R. Links, I. Roditi, Quantum spin ladder systems associated with su(2|2), cond-
mat/0010035.
[21] A.P. Tonel, A. Foerster, K. Hibberd, J. Links, Integrable generalize spin ladder models based on the SU(1|3)
and SU(3|1) algebras, cond-mat/0105302.
[22] H. Saleur, The continuum limit of SL(N |K) integrable super spin chains, Nucl. Phys. B 578 (2000) 552,
solv-int/9905007.
[23] M.J. Martins, B. Nienhuis, R. Rietman, An interesting loop model as a solvable super spin chain, Phys. Rev.
Lett. 81 (1998) 504, cond-mat/9709051.
[24] D. Arnaudon, C. Chryssomalakos, L. Frappat, Classical and quantum sl(1|2) superalgebras, Casimir
operators and quantum chain invariants, J. Math. Phys. 36 (10) (1995) 5262, q-alg/9503021.
[25] M.J. Martins, P.B. Ramos, The algebraic Bethe ansatz for rational braid-monoid lattice models, Nucl. Phys.
B 500 (1997) 579, hep-th/9703023.
[26] Z.S. Bassi, A. LeClair, The exact S-matrix for an osp(2|2) disordered system, Nucl. Phys. B 578 (2000)
577.
[27] Z. Tsuboi, A note on the osp(1|2s) thermodynamic Bethe ansatz equations, Int. J. Mod. Phys. A 17 (2002)
2351, cond-mat/0108358.
[28] H. Saleur, B. Wehefritz-Kaufmann, Integrable quantum field theories with OSP(m|2n) symmetries, Nucl.
Phys. B 628 (2002) 407, hep-th/0112095.
[29] L. Mezincescu, R.I. Nepomechie, Fusion procedure for open chains, J. Phys. A 25 (1992) 2533.
[30] V. Korepin, Direct calculation of the S matrix in the massive Thirring model, Theor. Math. Phys. 41 (1979)
953.
[31] N. Andrei, C. Destri, Dynamical symmetry breaking and fractionization in a new integrable model, Nucl.
Phys. B 231 (1984) 445.
[32] A. Doikou, R.I. Nepomechie, Bulk and boundary S matrices for the su(N ) chain, Nucl. Phys. B 521 (1998)
547, hep-th/9803118;
(1)
A. Doikou, R.I. Nepomechie, Duality and quantum algebra symmetry of the An−1 open spin chain with
diagonal boundary fields, Nucl. Phys. B 530 (1998) 641, hep-th/9807065.
On conformal field theory and stochastic

Loewner evolution
R. Friedrich a,b , J. Kalkkinen c
a Institut des Hautes Études Scientifiques, Le Bois-Marie, 35 Route de Chartres,
Bures-sur-Yvette F-91440, France
b Université Paris-Sud, Laboratoire de Mathématiques, Université Paris XI, F-91504 Orsay, France
c The Blackett Laboratory, Imperial College, Prince Consort Road, London SW7 2BZ, UK
Received 9 October 2003; received in revised form 16 March 2004; accepted 22 March 2004
Abstract
We describe Stochastic Loewner Evolution on arbitrary Riemann surfaces with boundary using
Conformal Field Theory methods. We propose in particular a CFT construction for a probability
measure on (clouded) paths, and check it against known restriction properties. The probability
measure can be thought of as a section of the determinant bundle over moduli spaces of Riemann
surfaces. Loewner evolutions have a natural description in terms of random walk in the moduli space,
and the stochastic diffusion equation translates to the Virasoro action of a certain weight-two operator
on a uniformised version of the determinant bundle.
PACS: 02.50.Ey; 05.50.+q; 11.25.Hf
MSC: 60D05; 58J52; 58J65; 81T40
Keywords: Probability theory; Conformal field theory
1. Introduction
The motivation for this article stems from the work of Lawler, Schramm and Werner [1]
in which they investigate on a purely probabilistic basis the “restriction property” of certain
probability amplitudes. This property can be phrased in terms of a class of stochastic
processes defined on proper simply connected domains of the complex plane, originally
E-mail addresses: rolandf@ihes.fr (R. Friedrich), jek@imperial.ac.uk (J. Kalkkinen).
doi:10.1016/j.nuclphysb.2004.03.025
280 R. Friedrich, J. Kalkkinen / Nuclear Physics B 687 [FS] (2004) 279–302
introduced in Ref. [2]. These processes—termed Stochastic Loewner (or Schramm–

Loewner) Evolutions (SLE)—proved to be of great value in providing a rigorous basis
for certain results in Conformal Field Theory (CFT) [3–6].
The precise relation of SLEs to CFT has remained, however, rather speculative. Some
aspects of it have been clarified building on the restriction property in Refs. [7,8], where an
interpretation of the parameters involved in the “Brownian bubble process” as the central
charge and the conformal weight of a highest weight representation of an underlying CFT
was obtained. This led further to an explicit calculation of the correlation functions of the
stress-energy tensor inserted at the boundary for trivial central charge c = 0. In this paper
we continue to investigate the properties of the objects found in [1] and propose a precise
CFT model for chordal SLE that to some extent generalises it. In a forthcoming paper [9]
we will expand on the mathematical side of this article. An introduction to the present
literature and to remaining challenges in particular in the case of percolation can be found
in [10,11].
Recall that a class of statistical mechanics models at the critical temperature can be
described in terms of CFT. We will investigate models defined on domains with boundaries,
or more generally on surfaces with boundaries (open string world-sheets). The choice of
appropriate boundary conditions will give rise to a long chordal domain wall, connecting
two points of the (same) boundary component. Since these fluctuating curves exist on
all lengths scales it is meaningful to ask for the limiting object: in the case of simply
connected planar domains it has been shown [2] that the scaling limit corresponds to a class
of conformally invariant probability measures, supported by these random curves. We will
show here how to relate, in the general case, such random curves to the vacuum expectation
values of certain nonlocal operators inserted on the (boundaries of) the manifold. The
proper treatment of random curves starting from a point in the bulk and connecting another
point on the boundary is related to order disorder lines which we will not treat in the present
article.
In the general situation we will show how specific densities associated to chordal curves
have the same covariance properties under conditioning as are known in the case of chordal
SLE on the upper half-plane.
The diffusion equation, which describes the stochastic process, can be thought of
as a real Schrödinger equation of a quantum mechanical particle on the boundary. The
Hamiltonian, which is a second-order hypo-elliptic differential operator, is the generator
of the stochastic process. At zero energy, that is to say when the Hamiltonian annihilates
the wave function, the probability density is a martingale.
The probability densities of SLE are conformal forms on the moduli space of Riemann
surfaces with boundary components and one marked point Mg,b,1 . They can be assumed
to belong to highest weight representations of the Virasoro algebra of the CFT, the weight
itself being determined by the variance of the driving Brownian motion, i.e., the diffusion
coefficient. From this point of view the martingale property corresponds to requiring that
the Hamiltonian shifts the conformal weight of the wave function by two, and that the
resulting representation is still of highest weight, and singular, so that it is equivalent to
zero in the physical Hilbert space. The fact that representations of the Virasoro algebra
were involved in the description of SLE curves was noticed already in [7,8,12].
R. Friedrich, J. Kalkkinen / Nuclear Physics B 687 [FS] (2004) 279–302 281
As the Loewner process changes the conformal class of the Riemann surface, it also
generates a random walk in the moduli space Mg,b,1 . We describe this in detail, and write
down the probability densities as sections of certain line bundles on this space. We show in
particular that the induced motion on the moduli space has a tangent vector field that lies
entirely in a rank-two vector bundle determined by the second-order differential operator.
Suppose we are given a differentiable path γt in the upper half-plane H. Then we can
make a polygonal approximation of it with (infinitesimal) straight lines. The Riemann
mapping theorem guarantees the existence of a conformal transformation that maps the
upper half-plane minus a first part of the Jordan curve onto the upper half-plane. This
map is, after a suitable normalisation, unique. A particularly important question is its
boundary behaviour. In our situation Carathéodory’s extension theorem tells us that it
extends continuously to the two-sided slit.
To illustrate the situation we give the inverse mapping for a slit of length tα α /(1 −α)α ,
extending in from 0 at an angle πα ∈ ]0, π[
α
−1 α
gt : H → H \ , z → (z + t) 1−α
z− t . (1)
1−α
The points −t and +αt/(1 − α) are mapped to 0, and 0 itself to the tip at tα α /(1 − α)α eiπα ;
this function is now normalised so that gt (z) → z when z tends to infinity. The exact
location of the two zeroes does not affect the present discussion.
Suppose we are given locally on the upper half-plane a conformally flat metric, say
eσ |dz|2 . Under the above coordinate transformation z → f (z, z̄) the conformal class—
and hence the complex structure—of a locally conformally flat metric changes according
to |dz|2 → |dz + µ dz̄|2 . (We omit here the arising Weyl factors.) The thereto associated
Beltrami differential µ(z, z̄) can be used to detect where the transformation ceases to be
conformal, and by what amount. Given the transformation f = gt−1 on X it can be readily
calculated from the defining equation ∂f = µ∂f ¯ . The Beltrami differential associated to
gt given in Eq. (1) can be argued to be a distribution of the form
α
µt (z, z̄) = −2πi (z − t)2 δ (2) (z − t), (2)
1−α
where the parameter t is related to the length of the slit. This distribution is, of course,
trivial when integrated with regular test functions. In our case it will appear with functions
with second-order poles precisely at z = t, so that it yields a finite result as discussed in
detail in Section 4.1.
The plan of the paper is as follows. In Sections 2 and 3 we recall some facts about
the probabilistic treatment of the chordal Loewner process on the upper half-plane, and
partition functions of statistical mechanics and Conformal Field Theory, respectively. If
not otherwise stated, we are always referring to the cordal case. In Section 4 we describe
the behaviour of CFT observables under the Loewner process, and propose a probability
density associated to a specific path on the Riemann surface. In Section 5 we rephrase this
in terms of determinant bundles over moduli spaces of Riemann surfaces in order to give
such a geometric description to the Loewner process in terms of these moduli spaces that
the Virasoro action becomes apparent. In the discussion Section 6 we draw the probabilistic
and the CFT discussions together.
2. Loewner process and restriction properties
Let us recall the definition of a chordal SLEκ in the upper half-plane H that starts from
0 and continues to infinity [11,13]:
Let (Ω, F , (Ft ), P ) be a standard filtered probability space that is complete and
continuous from the right. Given z ∈ H, t 0, define gt (z) by g0 (z) = z and
∂gt (z) 2
= . (3)
∂t gt (z) − Wt
√
Here Wt := κ Bt is the standard one-dimensional Brownian motion Bt , defined on
R+ × Ω with initial point 0 and with variance κ > 0. This means in particular that its
Itô-differentials satisfy (dWt )2 = κ dt.
Given the initial point g0 (z) = z, the ordinary differential equation (3) is well defined
until a random time τz when the right-hand side in (3) has a pole. Define the set Kt in the
closure of the upper half-plane as Kt := {z ∈ H: τz < t}.
The family (Kt )t 0 is an increasing family of compact sets, also called hulls, in the
closed upper half-plane H and gt is the uniformising map from H \ Kt onto H. It has been
shown in [4,14] that there exists a continuous process (γt )t 0 with values in H such that
H \ Kt is the unbounded connected component of H \ γ [0, t] with probability one. This
process is the trace of the SLEκ and it can be recovered from gt , and therefore from Wt , by
γt = lim gt−1 (z). (4)

z→Wt
z∈H
The constant κ characterises the nature of the resulting curves as classified in [14]. For
0 < κ 4 SLEκ traces over simple curves, for 4 < κ < 8 self-touching curves (curves with
double points, but without crossing its past) and, finally, if 8 κ the trace becomes space
filling.
Let us consider the function ft (z) := gt (z) − Wt that satisfies by virtue of Eq. (3) the
stochastic differential equation
2
dft (z) = dt − dWt . (5)
ft (z)
Consider k points xi , i = 1, . . . , k, on the real axis R which we view as the boundary of the
upper half plane ∂H, their images yi := ft (xi ) under the Loewner mapping, and a smooth
function F : Rk → R of the coordinates yi , i = 1, . . . , k. Then Ft := F (ft (x1 ), . . . , ft (xk ))
is a new stochastic process, and Itô’s formula and the ordinary differential equation (3)
yield, as in [7],

κ 2
dFt = dWt L−1 Ft + dt L−1 − 2L−2 Ft , (6)
2
where the differential operators Ln are given by

k
∂
Ln := −yjn+1 . (7)
∂yj
j =1
Similarly, one may construct slightly more general stochastic processes by pulling back
holomorphic differential forms on the upper half-plane H by the Loewner mapping ft ; the
same algebra as above goes indeed through provided we restrict to holomorphic functions
or, more generally, differential forms. The additional piece of information one requires is
the fact that the standard de Rham differential ddR acts in this situation as usual through

ddR ft (xi ) = ft∗ ddRyi x , (8)
i
and that the Itô differential d acts independently

−2
d ddR ft (xi ) = dt ⊗ ddR ft (xi ). (9)
ft (xi )2
Consider in particular a product of holomorphic differential forms with conformal weights
⊗h
hi , namely, ω ∈ i Ω (1,0)(H) i . These objects are conformal forms evaluated at k
distinct points on yi ∈ H. Pulling back these forms with the Loewner mapping1 ft : xi →
yi , we can define the stochastic process ωt := ft∗ ω. The components of the differential
form ωt obey still Eq. (6), but the operators Ln are now defined as

k
∂
Ln := −yjn+1 + hj (n + 1)yjn (10)
∂yj
j =1
in terms of the local coordinates yj of the j th factor. This is precisely the action of a
generator of the Witt algebra, Der K,
[Ln , Lm ] = (n − m)Ln+m , (11)
on a highest weight state, with conformal weights hj . The Witt algebra has a universal
one-dimensional central extension
0 → C → Vir → Der K → 0, (12)
known as the Virasoro algebra Vir. For a specific central element c1, its generators Ln ,
n ∈ Z obey the following commutation relations
c
[Ln , Lm ] = (n − m)Ln+m + n(n2 − 1)δn+m 1. (13)
12
Later in the Paper, in particular in Section 5.2, we shall see the Witt algebra acting on
certain objects and how this action can be extended to that of the Virasoro algebra. This
will amount to replacing Ln by Ln .
If we specialise now the above discussion to the case k = 1 where ω ∈ Ω (1,0)(X)⊗h is a
single holomorphic conformal form on the Riemann surface X = H with conformal weight
h and x ∈ X some choice of local coordinates on X, then the stochastic process ωt := ft∗ ω
obeys the stochastic differential equation

κ 2
dωt = dWt L−1 ωt + dt L − 2L−2 ωt . (14)
2 −1
1 Note that the operators L are expressed in terms of the coordinates y and not of x .
n i i
This result implies in particular that the process ωt happens to be a martingale for SLEκ ,
i.e.,

κ 2
L−1 − 2L−2 ωt (xi ) = 0 (15)
2
precisely when
(3κ − 8)(κ − 6)
c=− , (16)
2κ
6−κ
h= . (17)
2κ
Then the descendants of ω do indeed contain the required singular (null) vector in the
Verma module V2,1 .2 These are also the only values of these parameters for which the drift
term is a highest weight representation of the Virasoro algebra.
In the rest of the paper we shall be mostly interested in the behaviour of these differential
form-valued stochastic processes on the boundary ∂X of the Riemann surface X, which in
this section is simply the upper half-plane. In this, as well as the more general case, the local
coordinates x ∈ ∂X on the boundary components are real. It is important, nevertheless, that
the holomorphic analysis above goes through because this guarantees that we can express
these boundary objects in terms of the right bulk objects as required in the Schottky double
treatment of Boundary Conformal Field theory (BCFT).
We make now use of the fundamental fact that one can associate a second-order partial
differential operator to Itô diffusions, namely, its generator. For the stochastic differential
equation (5) we obtain
κ
H̄ = L2−1 − 2L−2 , (18)
2
in the notation of definition (10) with h = 0. The Feynman–Kac formula states that if we
define for ϕ ∈ C02 (Rn ) and V ∈ C 0 (Rn ) bounded from above
t

ψ(x, t) := Ex ϕ(Wt ) exp V (Ws ) ds , (19)
0
then we have
∂ψ(x, t)
= H̄(x) + V (x) ψ(x, t), (20)
∂t
ψ(x, 0) = ϕ(x), (21)
where H̄ is the generator of the process ωt .3 By choosing formally the potential function
as
2h
V (x) := 2 (22)
x
2 We consider here the highest weight vector submodule associated to this singular vector rather than the
quotient of the full Verma module V2,1 with it; the quotient construction with the values of c and h as in (16) and
(17) has been considered independently from a different point of view in Refs. [17,18] for the upper half-plane.
3 At this point we need to restrict to boundary points x ∈ ∂X, which are real.
we get the full Hamiltonian for arbitrary positive h

Ĥ := H̄(x) + V (x) (23)
κ 2
= L−1 − 2L−2 (24)
2
as it appeared in Eq. (18). This potential is of course not bounded at the source x = 0.
Nevertheless, the Fokker–Planck equation (20) gives the time evolution of the probability
density ψ(x, t), and can be interpreted as a real Schrödinger equation. In that context ψ is
the wave function of a quantum mechanical particle evaluated at time t and at the initial
position x.
2.1. Restriction property
Consider chordal SLEκ for κ < 4 which produces a simple curve γ : [0, ∞) → H with
γ (0) = 0, γ (0, ∞) ⊂ H, and γ (t) → ∞ as t → ∞. Let the hull A be a compact set A ⊂ H
such that A ∩R ⊂ R∗+ , and H \A is simply connected. For such a hull A, let φA : H \A → H
be the conformal map that preserves 0 and infinity such that φA
(∞) = 1, i.e., that it
is hydrodynamically normalised. Under this normalisation composition of uniformising

maps defines a pseudo-semigroup on hulls. Suppose that A and A
are hulls and let
A1 := ΦA−1 (A
). Then
φA1 = φA
◦ φA . (25)
Let us further define the set V∞ := {ω: γ [0, ∞) ∩ A = ∅} which is measurable and has
a positive probability. Given a specific outcome γ ∈ V∞ in this set, we may consider the
path γ̄ := φA ◦ γ (t). For all A as above and assuming γ [0, ∞) ∩ A = ∅, we say that SLEκ
satisfies the chordal restriction property if the conditional distribution of γ̄ is the same as
a time change of SLEκ . This can be summarised in the commutative diagram in Fig. 1.
SLEκ has indeed been shown to have this property for κ = 8/3 in Ref. [1].
On the other hand, restricting is the same as introducing a new probability measure QA
under which all paths that meet the hull A in a finite time form a set of measure zero. Using
the original probability measure P , the new measure QA can be simply defined as
P (G ∩ V∞ )
QA (G) := , where G ∈ F (26)
P (V∞ )
It is therefore a conditional probability. Furthermore, it is absolutely continuous with
respect to the original measure P , i.e., QA P on F∞ . We can therefore calculate the
Radon–Nikodým derivative of QA with respect to P on the sub-σ -algebra Ft and express
it in the form

dQA (ω)Ft = Mt (ω) dP (ω)Ft . (27)
This can actually also be extended to the range κ ∈ (0, 8/3). This requires, however,
surrounding the simple path γ with a cloud of Brownian bubbles Ξ with a certain intensity
λ as in Ref. [1], and posing then the following question: what is the probability that the
thus created hull around the path γ does not intersect the set A? The difference between
the original measure P on γ and the conditioned measure on γ can be expressed again
Fig. 1. The commutative diagram.
through the Radon–Nikodým derivative

t
λ
Yt := h
t (Wt )α exp hs (Ws ), Ws ds , (28)
6
0
where the bracket { , } is the Schwarzian derivative

f
(z) 3 f
(z) 2
f (z), z :=
− . (29)
f (z) 2 f
(z)
The condition for this process indeed to be a martingale is given in Proposition 5.3 of
Ref. [1]
(3κ − 8)(κ − 6)
λ= , (30)
2κ
6−κ
α= . (31)
2κ
Recall that we already found that a conformal form ωt of weight h was a martingale
Ĥ ωt = 0 provided (16) and (17) hold. These parameter values are consistent with the
above-found parameter values, provided α = h and λ = −c. Note, however, that the range
of the diffusion parameter κ ∈ (0, 8/3) restricts the central charge to negative values. This
was already noticed in [1,7].
3. Partition functions
Consider a statistical mechanics model on a discretisation of a surface X with states

getting values in the discrete set Z2 . At a critical temperature T = Tc the correlation length
Fig. 2. A typical phase boundary γ in a Z2 -model on a simply connected domain D.
diverges, and the correlation functions become conformal. The typical statistical mechanics
configuration below the critical temperature is patchwise constant, and the emerging
structure is that of Kadanoff’s droplet picture. The fact that the correlation length diverges
at criticality means that the structure becomes self-similar. One can nevertheless still
recognise a fractal phase boundary structure γ ⊂ X as depicted in the heavily simplified
Fig. 2.
If we now fix boundary conditions we force the phase boundaries to include in particular
a path that connects the points on the boundary where boundary conditions are changed.
In Fig. 2, for instance, this happens at 0 and ∞. Since the partition function with free
boundary conditions, Z, should provide a measure for the number of states in the physical
system, the partition function Zαβ with fixed boundary conditions αβ should provide a
measure for the paths forced by changes of boundary conditions. Consequently, the fraction
Zαβ
(32)
Z
should be the fraction of phase boundaries that include the paths forced by these changes
of boundary conditions among all possible phase boundaries included in the full partition
function. It can be heuristically considered, therefore, to be the probability that some path
connects these specified boundary points in the unconstrained theory.
The question of how to define properly statistical mechanics models on Riemann
surfaces is still, however, an open problem. For present purposes we translate the problem
to a question in CFT, where this issue has been addressed in the literature already.
At the critical temperature T = Tc , namely, the above statistical mechanics models
become scale invariant. We assume here that there should then exist a CFT that has the
same correlators as the statistical model. For a definition of a CFT, see for instance Segal’s
axioms in [15,16]. The partition function Z := 1 can clearly still be thought of as a
weighted sum over configurations as in the statistical mechanics case; the same is true
for the partition functions with different boundary conditions. Suppose we require the
boundary conditions |α̃i to apply on intervals labelled by i = 1, . . . , n. This can be done
technically by inserting the corresponding boundary operators φi := φαi αi+1 that change the
state |α̃i to the state |α̃i+1 at the endpoints of the intervals {x1 , . . . , xn } [19], for general
discussions see, e.g., Refs. [20,21]. This can be used to express the partition function of
the constrained theory in terms of the BCFT correlation function

n

Zα1 ...αn := Z φi (xi ) . (33)
i=1
On physical grounds it seems reasonable to expect that the (positive) ratio Zα1 ...αn /Z
should be bounded by 1; this issue should nevertheless be settled by carefully studying
candidate CFTs case by case. We do not need to assume that there were only one boundary
component. In what follows, we shall refer to these operator insertions collectively as
O(xi ) := ni=1 φi (xi ).
In CFT we cannot restrict configurations to phase boundaries as in the case of classical
statistical mechanics models, simply because configurations in a quantum field theory do
not lend themselves to classical treatment. The only well-defined restrictions are indeed
boundary conditions as discussed above.
Suppose we have inserted a boundary operator at a marked point O(0)X and wish
nevertheless to restrict to states that have a quantum mechanical analogue of a phase
boundary along some path γ that starts from the marked point and ends somewhere in
the bulk z ∈ X. This is actually possible if we first cut out the path from the domain where
the CFT is defined, thus in effect turning the path into a part of the boundary where we can
impose boundary conditions O(0)X\γ , and then move the operator insertion to the tip of
the cut-out path. The ratio
O(p)X\γ
(34)
O(0)X
can therefore be considered as the probability associated to the path γ occurring among
all the possible phase boundaries forced by the operator insertion O(0)X at the original
point on the boundary. We will next find this object for curves that correspond to Loewner
processes in (56) and (58).
4. Moduli under Loewner process
A Riemann surface X, i.e., the complex structure of the two-dimensional manifold X,

corresponds to a conformal class of metrics, which locally can be expressed in the
form ds 2 = eσ |dz|2 , where z is a local isothermal coordinate. In general, given a two-
dimensional manifold with such a metric we can deform it in the following way.
Deformations of a metric g on the Riemann surface X can be decomposed to local
reparametrisations given by a global vector field v ∈ Γ (T X) on the surface, local Weyl
rescalings given by a global function ϕ on X, and Teichmüller reparametrisations given by
the Beltrami differentials
µ ∈ Ω (−1,1)(X) := T (1,0)X ⊗ Ω (0,1)(X) (35)
and µ̄ ∈ Ω (1,−1)(X). Under these transformations a correlator

O := Φi (zi ) (36)
i
of (holomorphic) primary fields—of given spins si and scaling dimensions ∆i = 2hi −si —
inserted at points zi ∈ X transforms as [20]

1
δO = d2 z ∇z̄ v z + µ T (z)O + ∇z v z̄ + µ̄ T̄ (z̄)O
2πi
X

− ∆i ϕ(zi )O. (37)
i
The last term here arises from delta-functions due to insertions of Tzz̄ + Tz̄z , though
only when we choose to change the representative of the conformal class σ → σ + ϕ
as well. We use here conformally flat reference metric proportional to |dz|2 . Choosing the
transformations judiciously, this identity implies also the standard conformal Ward identity.
On manifolds with boundary, the diffeomorphisms generated by the vector field (v z , v z̄ )
are required to preserve the boundary; this necessitates also that only one independent
copy of Vir and Vir is preserved so that at the boundary x ∈ ∂X the stress-energy tensors
coincide T (x) = T̄ (x).
As the above deformations are all of the deformations we can perform in two dimensions
(conformal and complex structures being equivalent), then the Beltrami differentials
are the only true deformations of the moduli of the theory, and can be thought of as
(anti-)holomorphic vector fields on the tangent space of the moduli space (µ, µ̄) ∈ TX M.
More formally, the Beltrami differentials µ can be thought of as classes of the tangent
bundle valued first cohomology group µ, µ̄ ∈ H 1 (X, T X). In the case of the moduli
space of Riemann surfaces with boundary the moduli space has, again, only real analytic
structure.
The partition function Z depends, first of all, on the moduli of the Riemann surface X
and the details of the CFT defined on that surface [22]. It is therefore, in particular, locally
a function of the coordinates m, m̄ of the moduli space. In a nontrivial CFT c = 0 there
is a trace anomaly and the partition function depends on the choice of a representative
of the conformal class. In defining the partition function in this way we need to specify,
therefore, that the partition function Z(m, m̄) is evaluated, for instance, in the constant
curvature background metric gc in the conformal equivalence class of metrics we are
interested in. The dependence on the representative of the equivalence class arises through
the Liouville action SL (σ, gc ) so that if we know the partition function in the constant
curvature background metric Z(m, m̄), on general backgrounds g = eσ gc the physical
partition function becomes
Z(g) = ecSL (σ,gc ) Z(m, m̄). (38)
Neither is the partition function Z(m, m̄) in general a well-defined function on the
moduli space of Riemann surfaces, but rather a section of a (projective) line bundle on it.
In the case of closed surfaces this line bundle factorises Ec ⊗ Ēc to holomorphic and anti-
holomorphic parts. The line bundle Ec comes equipped with a connection, with respect to
which the partition function is covariantly constant. This can be seen also form Eq. (37) by
choosing O = 1 so that δO = δ1 = 0. This implies

1
∇µ Z := δµ Z − d2 z T (z)µ(z, z̄)Z = 0, (39)
2πi
X
Fig. 3. Changing the Riemann surface in Loewner process gt induces a random walk in the moduli space with
the tangent vector field given by the Beltrami differentials µt .
which tells us that the partition function is parallel transported along the vector field µ
on the moduli space with respect to the connection d + T . The holomorphic part can be
recognised as a tensor power of the standard determinant bundle
⊗c/2
Ec = detX , (40)
otherwise known as the inverse Hodge bundle Det1 = det−1 X .
In the case of Riemann surfaces with boundary components the holomorphic and the
anti-holomorphic sectors are related by complex conjugation. This means that the partition
function Z(m, m̄) with m = m̄∗ is actually a section of the emerging real-analytic bundle
|detX |⊗c → M. (41)
Boundary operators can be described4 in terms of bulk operators on the Schottky double
by insertions of operators and their complex conjugates on the original surface, and its
mirror image, respectively. This means in particular that if in the expectation values of
∗(1,0)
bulk operators inserted at the point z ∈ X transform as elements of (Tz X)⊗h for
some conformal weights h
, then the corresponding [20] expectation values of boundary
operators at z → x ∈ ∂X transform as elements of |Tx∗ ∂X|⊗h for some conformal weights
h that depend on the structure of the theory at hand.
4.1. Infinitesimal deformations
Eq. (39) above is an example of the natural pairing given by integration over the
Riemann surface X of Beltrami differentials µ ∈ TX (1,0) M and holomorphic quadratic
differentials ν ∈ ΩX(2,0)(X)

(ν, µ) := ν ∧ µ, (42)
X
4 At least in the unitary case.

as the stress-energy tensor ν = T (z)(dz)2 ∈ Ω (2,0)(X) is a locally defined quadratic

differential. Holomorphicity is required here for guaranteeing independence of the choice
of the representative of the Beltrami differential µ ∼ µ + ∂v. ¯ On manifolds with boundary
we must restrict to vector fields v that generate flows that leave the boundary invariant. In
(1,0)
this sense we can identify ΩX M Ω (2,0)(X).
Let us consider Loewner process along a parameterised
path γ ⊂ X, and suppose that
we have divided it in infinitesimal parts γ = i γi . By infinitesimal we simply mean that
the pertinent measure on each path γi —Hausdorff or Lebesgue—is arbitrarily small. In
what follows we denote this measure by dt, given a specific parameterisation of the original
curve.
Consider now in particular Loewner procedure over the first of these infinitesimal
installments γ0 ; the idea is to iterate the procedure over all of the infinitesimal contributions
to get a finite result, as we shall see: the Beltrami differential associated to cutting
the surface along the infinitesimal path γ0 can be thought of as an infinitesimal vector
µt ∈ TX M. Its length is proportional to the volume of the cut-out set, i.e., dt. It has support
only on the path γ0 , if anywhere. Given, as a test function, a cotangent vector from TX∗ M
represented by a smooth quadratic differential ν ∈ Ω (2,0)(X) the pairing is, therefore, of
the form

µt , ν(z)(dz)2 ∼ dt ν(γ0 ). (43)
Here ν(z) is assumed constant across the infinitesimal set γ0 . This means that the Beltrami
differential can be put formally in the form
dz̄ dt
µt := µt (z) = −2πiδ(γ0 ) , (44)
dz (dz)2
where the formal current
−2πiδ(γ0 ) dt = µt (z) d2z (45)

has distributional support on the cut-out path γ0 . The details of the distribution will depend
of how we choose to regularise it.
Under Loewner procedure the conformal class of the original Riemann surface X
changes as a function of the parameter t along the path γ ⊂ X. The resulting family of
Riemann surfaces Xt ⊂ M traces similarly over a certain path Γ ⊂ M in the moduli
space. The choice of regularisation of the formal current above amounts then to a choice
of parameterisation of this curve as we shall presently see.
To make the discussion indeed somewhat more concrete, let us return to Eq. (2) where
the Beltrami differential associated to the specific Loewner process gt on the upper half-
plane was found to be
α
µt (z, z̄) = −2πi (z − t)2 δ (2) (z − t). (46)
1−α
The distribution δ(γ0 ) in this case is
α
δ(γ0 ) = (z − t)2 δ (2) (z − t). (47)
1−α
In view of the regularisation procedure that will follow, we choose first to change the
parameterisation of the path Γ from t to t
where
α dt
= . (48)
1−α dt
This is precisely the definition of the parameter on the path Γ ⊂ M alluded to below
Eq. (44).
This distribution is zero when integrated with regular test functions; in our case it will
appear with functions with second order poles precisely at z = t
h 1
T (z)φh (t, t¯ ) ∼ φh (z, z̄) + ∂z φh (z, z̄), (49)
(z − t)2 z−t
so that it picks the coefficient of the leading pole in the operator product expansion,
provided the operator φh is inserted precisely at z = t. It is therefore useful to define the
operator T̂ that does just that, namely, picks the leading term in the Laurent expansion of
the stress-energy tensor with operator insertions at the specified point, for instance,
T̂ (z)φh (t) := hφh (t, t¯ ), (50)
for z = t, otherwise T̂ (z)φh (t) = 0. One could represent this operator, e.g., in terms of
contour integration of the standard operator product. All other operator insertions provide
a trivial result. Note also that the contribution coming from the operator insertion at z = t
will now be finite, and its precise value does indeed depend on the regularisation or other
details of the distribution δ(γ0 ).
We may now express the change of the correlator through

(µt , T )φh (t) regularised = dt
T̂ (t)φh (t) (51)
which is now to be looked upon as a one-form in TX∗ M. Note that the argument of φh
is a point on the boundary of the Riemann surface and the differential dt
refers to a
parameterisation of the path Γ ⊂ M.
The operator insertions we consider φh are precisely where we want the boundary
conditions in CFT to change—at the intersection of the considered path γt and the
boundary ∂X. After having performed the above iterative step, we should, therefore,
translate the operator insertion from the original point z = t to z = 0 where the rest of
the path intersects the boundary. This is to be seen as a part of what we mean by Loewner
procedure, but boils down to choosing to evaluate the field φh (z, z̄) precisely at z = 0; the
difference is indeed negligible as long as the infinitesimal path γ0 is small enough. This is
also reminiscent of the definition of stochastic integrals, where one chooses to evaluate the
integrand at the left point of each interval.
Inserting these results in Eq. (37) and not forgetting the anti-holomorphic sector
produces now

δO = −dt
T̂ γ0 (t) O + T̄ˆ γ0 (t) O , (52)
with the understanding that the right-hand side involves an insertion of O in the beginning
of the path and the left-hand side in the end of the path. In summary, the concrete analysis
has produced the following results:
(1) With an operator insertion at the intersection of the path γ and the boundary ∂X,
correlators transform by an infinitesimal but nontrivial amount;
(2) The transformation can be expressed in terms of a (finite) line integral of the stress-
energy tensor along the path γ ⊂ X; and,
(3) The integration parameter is determined in regularising Eq. (45); the integral is
invariant under reparameterisations as long as we change the volume measure on the
Riemann surface X on the right-hand side of this equation or the regularisation of the
distribution on the left-hand side as well.
Let us now return to the general discussion, assuming we have regularised the
distribution δ(γ0 ), introduced operators T̂ , and a parameterisation t of the path Γ ⊂ M.
Iterating this procedure for all infinitesimal paths γi is tantamount to exponentiating
the infinitesimal operation: the procedure leads indeed to (essentially) the standard path-
ordered exponential that appears in parallel transportations, which can in this case be
defined in a regularised form, in notation that will be explained below, as

0
P exp − T (γt ) dt := T(γi ) 1 − dt T̂ pi . (53)
γ i γi
Here the stress-energy tensor is evaluated at the starting point pi0 ∈ ∂γi of each infinitesimal
path γi , and the integral
itself reduces to the (Lebesgue or Hausdorff) measure of the
infinitesimal path γi dt. Now, in the limit where the paths are indeed taken arbitrarily
small, the operator product expansion of the factors in this formula with any operator
insertion φh (x) on the boundary x ∈ ∂X are never more singular than with insertions at
the beginning of each path pi0 ∈ ∂γi for the simple reason that this is the only place where
the infinitesimal path intersects the boundary: therefore, all operator product expansions
are dominated by whatever contribution arises from the starting points x = pi0 and, as was
shown above, these contributions are finite. We have stipulated a specific way of doing
this expansion by determining that the operator insertion φh , if any, be translated from the
beginning of the path pi0 to the end pi1 by inserting explicitly the translation operators

T(γi ) · φh pi0 := φh pi1 . (54)
What all of this amounts to is a specific regularisation of the formal, a priori perhaps
rather singular operator

P exp − T (γt ) dt φh . (55)

γ
There might be other ways of regularising this object—essentially a parallel transport of

operators along the path γ . It would be interesting to investigate further how different
regularisations affect the present discussion or under what conditions this formal product
should indeed converge. In particular, one should consider insertions of boundary operators
in terms of insertions of bulk operators together with their mirror image on the other half
of the Schottky double. We leave these issues, however, to later study.
4.2. Parallel transport
Let us consider now the correlation function

O(yi ) γ [0,t ] := P exp − ds T γ (s) + T̄ γ (s) O(yi ) , (56)
γ [0,t ]
associated to the path γ [0, t] that follows the path γ from time 0 until the time t. The
definition of the path-ordered exponential used here was the subject of Section 4.1; the
regularised expression we intend to use here was explicitly constructed in Eq. (53). In
particular, the exponential is to be expanded in a product of exponentials of infinitesimal
contributions along the path acting successively on O(yi ), and the operator insertion
should be transported along the path in the process of performing the integration. The
operator O was defined in Eq. (36) in terms of chiral primaries. In what follows we
shall choose to restrict to operator insertions on the boundary zi = yi ∈ ∂X, and choose
the operators themselves from the pertinent BCFT. These operators are, by construction,
invariant under complex conjugation and the associated correlators are thus real. As was
observed in the end of Section 4, they transform as elements of |Ty∗i ∂X|⊗hi under conformal
transformations.
Using the explicit Beltrami differential (2), it was possible in fact to show in Section 4.1
that the Loewner procedure along the path γ maps the correlator O to the above correlator
Oγ [0,t ] . What we see here is, therefore, its parallel transport with respect to the operator
valued connection T [23]. Physically one can think of this as the partition function in the
presence of energy density, or a current, distributed along the path γ .
The argument to this effect made use of the explicit form of the Beltrami differential µ
on H given in Eq. (2), and the fact that it always arises together with the correlator of the
stress-energy tensor T and the explicit operator insertion O: the operator product expansion
has precisely the quadratic pole needed to produce a nontrivial result. The infinitesimal
change in the correlator can be thought of as a suitably normalised differential on the
moduli space; cutting out parts of the path repeatedly leads to the path-ordered exponential
integrated along a finite path in the moduli space.
We can now define a probability density associated to any parameterised path in the
space of paths on the surface X Π(X, p; t), namely, let
Oγ [0,t ]
PX : Π(X, p; t) → R∗+ , γ → (57)
O
for each path in Π(X, p; t) that starts from a fixed point p ∈ ∂X and goes on until the final
parameter value t. This density has, first of all, the property that it is real and normalised
PX ∈ (0, 1] such that PX (p) = 1. Reality follows from the facts that the operator insertions
on the boundary are, by construction, real O∗ = O and that the conformal mapping
f : X \ γ → X preserves the boundary on the real axis. The fact that it is non-negative
follows from the assumption that O = 0 as a physical partition function, and the fact that
Oγ [0,t ] was constructed from it by exponentiation. It is bounded by its value PX (p) = 1
if we assume that the classical weak energy condition T + T̄ > 0 translates to a positivity
condition on the spectrum of the corresponding operator in CFT. Whether these (mild)
assumptions are satisfies depends on the details of the pertinent CFT model.
The above observations amount to the statement that we can consider PX (γ ) to be a

probability associated to a path γ . It is, however, not the probability of the path occurring
among all the paths of Π(X, p) but rather the probability of finding a path in a hull of
γ , perhaps, whose width is related to the structure of the CFT, and the central charge,
i.e., the diffusion coefficient in particular. The precise interpretation of this probability
density follows from stochastic considerations, and will be deferred to Section 6.
Suppose now that the simple curve γ [0, t] is an SLEκ process γt on a Riemann surface
X with 0 κ < 4. We can then likewise associate to this path the correlator

Oγ [0,t ] = P exp − dσH (s) T γ (s) + T̄ γ (s) O , (58)
γ [0,t ]
where σH is the Hausdorff measure of dimension [24]

κ
dimH (γ ) = min 2, 1 + . (59)
8
This extends the definition (56) to the case of fractal curves. It shows, again, how
the correlation functions of observables O transform under Loewner processes, and it
continues to be normalised as suggested above.
4.3. Conditioning correlators
A local conformal transformation ρ induces a transformation R(ρ) of the pertinent

CFT (Hilbert) space.5 The operator insertions of primary fields transform homogeneously,
whereas the stress-energy operator changes inhomogeneously
h
O(yi ) = R(ρ)O ρ(yi ) R(ρ)−1 ρ
(yi ) i , (60)
i
c
T (z) = R(ρ)T ρ(z) R(ρ)−1 + ρ(z), z 1, (61)
12
where { , } is the Schwarzian derivative (29). The correlator Oγ [0,t ] transforms therefore
as
h
ρ ∗ O(yi ) γ [0,t ] = ρ
(yi ) i
i

c
× exp − ds Re ρ γ (s) , γ (s) O(yi ) γ [0,t ] . (62)
6
γ [0,t ]
In order to condition the probability to paths that do not enter a given simply connected
domain A ⊂ X that touches the boundary, we need to investigate the behaviour of the
density PX under the pertinent diffeomorphism ρ : X \ A → X. In the simple case that this
5 This amounts to a representation of Aut O defined in Eq. (67) on the pertinent Hilbert space constructed
essentially by exponentiating the positive part of Virasoro algebra Vir0 , cf., e.g., Lemma 5.2.2 in Ref. [29].
diffeomorphism happens to be a conformal mapping or that the pertinent moduli space is

a discrete set of points as in the case of the upper half-plane H we find
PX\A
hi c
= ρ (yi ) exp − ds Re ρ γ (s) , γ (s) . (63)
PX 6
i γ [0,t ]
Comparing to the process Yt in the stochastic analysis (28), we see that PX and P
have precisely the same behaviour under conditioning, cf. Section 6. In comparing this
expression to Yt in stochastic analysis, one needs to take into account the fact that there
it was convenient to keep the origin of the complex plane fixed and let the intersection
Wt = γt ∩ ∂H move, whereas in the CFT analysis one kept the intersection fixed at the
origin and allowed the original zero to move.
In the general case where the diffeomorphism ρ : X\A → X changes the conformal
structure of the Riemann surface, it is difficult to give an explicit formula for the
transformation. For infinitesimal such a deplacement this is nevertheless possible, and
reduces clearly to insertions of the stress-energy tensor integrated with the Beltrami
differential associated to ρ in the correlators.
The correlator Oγ [0,t ] can be recognised as a section of a certain bundle Lh over the
moduli space of Riemann surfaces.6 In this context it is clear that PX (γ ) is consequently
a holonomy, or a Wilson line, of this section when parallel transported from the fibre at X
to the fibre over X\γ with respect to the connection d + T .
Recall that if the correlator O satisfies Ĥ O = 0, the operator creates a state in
the Verma module V2,1 . This module is closed under Virasoro action, which in turn
is generated by the stress-energy tensor T . Since the Loewner process involves only
insertions of the stress-energy tensor in the correlator, the final correlator Oγ [0,t ] has
to be that of an operator belonging to the same Verma module and satisfying the same
differential equation. This is true irrespective of the moduli of the Riemann surface, and
provides indeed an independent analytic characterisation of the correlators Oγ [0,t ] as
those sections of Lh that are annihilated by Ĥ .
In the next section we will describe this bundle Lh and the action of Ĥ in detail.
5. Determinant bundles
In this section we shall consider the geometry of the densities PX as sections of certain
bundles over moduli space. We shall pay particular attention to the explicit form of the
Virasoro action on these bundles because it translates directly to a description of the
Loewner process in the moduli space. We follow in much the notation of Ref. [29].
6 For the definition, see Eq. (66).

5.1. Virasoro action
The CFT partition functions on Riemann surfaces with b boundary components and of
genus g could be seen as a section of the determinant bundle

1 ∈ Γ Mg,b , |detX |⊗c (64)
as discussed in Section 4. We recall that the fibre at X ∈ Mg of the standard determinant
bundle Detj with j ∈ Z is

max
⊗j

max
⊗j ∗
H 0 X, ΩX ⊗ H 1 X, ΩX . (65)
The determinant bundle detX associated to a surface X is the inverse of the Hodge bundle
detX = Det−1
1 . Of determinant bundles it is known, for instance, that the Hodge bundle
⊗(6j 2 −6j +1)
Det1 generates [25] the Picard group of Mg , and that Detj Det1 [26].
For one (spin-0) operator insertion k = 1 at y1 = p with conformal weight h the
transformation rule (62) implies Oγ [0,t ] ∈ Γ (Lh ) where7
⊗h
Lh := |detX,p |⊗c ⊗ Tp∗ ∂X . (66)
Here we have used the pull-back bundle |detX,p | := ∗ |detX | where : Mg,1 → Mg .
The fibre of |detX,p |⊗c → Mg,1 at the Riemann surface X is twisted by a tensor power
of the modulus of the cotangent space of the surface at the marked point p ∈ ∂X. The
modulus appears, as was discussed in the end of Section 4, because insertions of boundary
operators can be described as insertions of bulk operators and their mirrors on the Schottky
double; as the mirror transforms by the complex conjugate ρ
(p̄)∗ of the transformation of
bulk field8 ρ
(p), the total effect is a transformation by a positive function |ρ
(p)|2 .
Let us denote by O = C[[z]] the set of formal Taylor series of a formal coordinate on
the unit disc z ∈ D, and K = C((z)) the set of formal Laurent series in the punctured disc
z ∈ D∗ . The automorphism group Aut O are changes of coordinates, representable in terms
of Taylor series of the form
∞

Aut O an z : a1 = 0 .
n
(67)
n=1
We associate to the marked point p a formal coordinate z ∈ D that gets its values in the
formal unit disk. The triples (X, p, z), where X is the Riemann surface, glue together to an
7 To be quite concrete, a point in the moduli space (X, p) ∈ M

g,1 determines a Riemann surface and a
marked point p ∈ X. As the fibre in |detX,p |⊗c consists of non-negative numbers v, we can give a point in the
total space as (v; X, p) ∈ |detX,p |⊗c . The transformation functions λ
ij in this bundle can be given in terms of
the transformation functions λij (cf. Eq. (38)) on detX,p by setting λ
ij = |λij |c . Twisting by |Tp∗ ∂X|⊗h means
that if the representative of the conformal class of (X, p) are different on two charts—i.e., we must perform a
conformal transformation ρ on the Riemann surfaces when changing charts in the moduli space—then the full
transition function in Lh is indeed |λij |c · |ρ
(p)|h .
8 Note that the conformal weight h above is the conformal weight of the associated boundary field, not that of
the bulk field and its mirror, as discussed in the end of Section 4.
Aut O-bundle
π : M̂g,1 → Mg,1. (68)
The action of Aut O represents changes of formal coordinates, and has the Lie algebra
Lie(Aut O) = Vir0 = zC[[z]]∂z . We can think of the fibres of this bundle, therefore, as
the set of all choices of formal coordinates around the marked point p ∈ X. M̂g,1 can also
be viewed as a bundle
π̂ : M̂g,1 → Mg , (69)
but we have to add to the structure group also the shifts of the marked point, generated
by ∂z . Changes of the conformal structure of the surface are generated at the fixed point
by the singular vector fields of the form z−n C[[z]]∂z ⊂ Vir<−1 for n > 0. All these formal
vector fields included in Vir<−1 , Vir0 , and C∂z form together the Witt algebra Der K.
The actual Virasoro algebra Vir is a central extension of this (12). One can actually show
that M̂g,1 carries a transitive action of Der K compatible with the Aut O action along the
fibres [27,28], cf. also [29].
The underlying structure here is the Harish–Chandra pair (Der K, Aut O) and a
flat vector bundle M̂g,1 → Mg,1 associated to it. This structure generalises [27,28]
to the Harish–Chandra pair (Vir, Aut O) as well: it has an associated flat vector
bundle π̂ ∗ Detj → M̂g,1 whose infinitesimal automorphisms Lie(Aut π̂ ∗ Detj ) form a
representation of the Virasoro algebra [27]. The flat connection there is of the form d+L−1 .
Note that even if the tensor powers of these bundles were not globally well-defined, the
underlying D-module is always well-defined.
The difference between the bundles Lh and π ∗ Lh is therefore that the latter extends
the former by keeping track of the formal coordinates near the marked point. Where the
curvature of (the holomorphic part of) Lh was related to the Picard group of the moduli
space, the natural connection on the bundle π ∗ Lh is flat.
5.2. Loewner process on the determinant bundle
As the Loewner process produces a nontrivial Beltrami differential, it generates a

motion in the pertinent moduli space Mg,1 . This involves deforming the surface, changing
the complex structure, and displacing the marked point p ∈ X on the surface. If we attach
a formal disc with coordinate z ∈ D, we see immediately that the above operations induce
actions of Vir0 , Vir<−1 , and Vir−1 ∼ C∂z on the disc. This means that the Loewner
process acts on the disc by Der K.
In the special case of the upper half-plane we can simply identify the formal (half) disc
with the Riemann surface X = H itself. In Section 5.1 where we encountered the action
of Der K through operators Ln , in particular the Loewner process was generated by the
second-order hypo-elliptic differential operator
κ
Ĥ := L2−1 − 2L−2 . (70)
2
In the case of general Riemann surfaces this Der K action then extends to a transitive action
on the bundle M̂g,1 → Mg,1 compatible with the structure group, as was recalled in the
previous section.
Fig. 4. The Loewner process changes the conformal structure near the marked point p which is here represented
by a deficit angle in the formal disc.
The CFT analysis leads us to consider sections of the line bundle Lh , which is a twisted
version of the standard determinant bundle defined on the moduli space Mg,1 . By using
the above defined projection π , we can construct the pull-back bundle π ∗ Lh on M̂g,1 . This
bundle carries now a transitive Virasoro action, and can be equipped with a flat connection
∇ = d + L−1 . In this way the Der K action is lifted to a Virasoro action in the quantum
theory. The generator of the Loewner process
κ
Ĥ := L2−1 − 2L−2 (71)
2
should therefore be seen naturally as a map
Ĥ : Γ (π ∗ Lh ) → Γ (π ∗ Lh+2 ). (72)
The sections of these pull-back bundles differ from the correlators suggested by the CFT
analysis only in that they also depend on the formal coordinate. This extra structure is
just enough to enable us to equip them with the appropriate Virasoro action and a flat
connection.
More precisely, recall that we could consider pull-backs by the Loewner mapping
ft∗ of conformal forms of weight h as stochastic processes ωt := ft∗ ω. The correlator
ωt = Oγ [0,t ] furnishes now an example of such a stochastic process. These objects are
defined as sections of Lh ; when we pull these sections back into the bundle π ∗ Lh we need
to specify their dependence on the formal coordinate z ∈ D. This can be done by requiring
that the resulting process π ∗ ωt is still a martingale, in the sense that it is annihilated by the
action of Ĥ . As Ĥ is of second order, this determines a rank-two subbundle ker Ĥ ⊂ π ∗ Lh
over Mg,1 . In this way we have been able to eliminate the apriorious dependence on the
formal coordinate z from the stochastic process π ∗ ωt , so that it can really be thought of
as an element of Γ (Mg,1 , ker Ĥ ), but have nevertheless been able to retain the Virasoro
action on it.
The condition that the pertinent action of Ĥ annihilates the correlator arose also when
we wanted to characterise the behaviour of the correlators under conditioning; even though
we could not present an explicit formula, the correlator Oγ [0,t ] and ρ ∗ Oγ [0,t ] both were
argued to be sections of ker Ĥ. We see now concretely, that the correlator may change
under conditioning, but these changes are restricted to remain in the properly defined
rank-two bundle ker Ĥ . The existence of a flat connection on this bundle means that
under deformations of the complex structure such as in Fig. 1 the diagram does indeed
commute, even though the precise holonomies may be more complicated than Yt . The only
requirement is now that the deformation path in the moduli space is contractible.
5.3. Probability distributions
Apart from acting on the stochastic processes ωt , the operator Ĥ also acts on the
probability distributions ψ(x, t) with z = x + iy. Where above Ĥ ωt = 0 implied that ωt
was a martingale, Ĥψ(x, t) = 0 implies that the probability distribution is constant in
time ∂t ψ = 0. We can therefore also identify the suitably normalised positive sections
ψ := π ∗ P(X, x, p) of the real line bundle P ⊂ ker Ĥ as constant in time probability
distributions that solve the diffusion equation in (20) or, in other words, as zero-energy
wave functions of the real Schrödinger equation.
Let us finally find the explicit form of a generic section ψ ∈ Γ (Mg,1 , P) of the real
subbundle P ⊂ ker Ĥ of the two-dimensional bundle in the case of only one marked point
and the formal coordinate z = x + iy defined in its local neighbourhood. This amounts to
finding the real solutions of the differential equation Ĥψ(z, t) = 0: in polar coordinates
z = ρeiϕ they are
ψ(ρ, ϕ) = Aρ a e−bϕ cos(aϕ + b ln ρ + δ), (73)

where A and δ are integration constants, and
√
κ −4 32 − κ 2
a= , b= . (74)
2κ 2κ
This means that in the range κ ∈ (0, 4) the solutions to this differential equation have a
singularity at z = 0 and are bounded by ρ a where a is negative and vanishes precisely
for κ = 4. In particular, when the Hausdorff dimension of the stochastic process on the
Riemann surface reaches dimH = 3/2 at κ = 4 and the curve becomes self-touching, the
function ψ reflects this accordingly by ceasing to be suppressed when ρ tends to infinity.
6. Discussion
We have found that much of the analytical structure of SLEκ can be represented in
terms of a CFT with a central charge dictated by κ. The use of this is the fact that the CFT
treatment extends SLE processes to arbitrary Riemann surfaces and κ 4; the drawback
is perhaps the fact that the involved CFT apparatus is still mostly heuristically defined.
Nevertheless, there is a correspondence between, for instance, the following objects:
SLEκ Eq. CFT Eq.

α, λ (30), (31) ←→ h, c (17), (16)
P (27) ←→ PX (γ ) (57)
ψ(x, t) (19) ←→ π ∗ P(X, p, z) (73)
Yt (28) ←→ ρ ∗ Oγ [0,t] /Oγ [0,t] (62)
Yt is a martingale (28) ←→ Ĥ Oγ [0,t] = 0 (15)
Most importantly, the measure of Section 2.1 for clouded paths γ ⊂ Ξ ⊂ X enjoys
precisely the same covariance properties as the correlation function PX (γ ) defined in (57)
when restricted to the upper half-plane. Furthermore, the commutativity of the diagram in
Fig. 1 in probability theory is equivalent in CFT to the fact that there exists a flat connection
in π ∗ Lh .
As the SLE and the CFT structures appear to be isomorphic it seems reasonable to
conjecture that the probability density does indeed coincide with the weight constructed in
CFT P (Ξ ) = PX (γ ). The latter construction extends SLEκ to arbitrary Riemann surfaces
and to κ > 8/3.
Note added
After the completion of this work the related preprints [30,31] were brought to
our attention. We would like to thank the Referee for pointing out that also Makarov
(unpublished) has proposed an extension of SLE to non-simply connected geometries.
Acknowledgements
We acknowledge many stimulating discussions with M. Kontsevich, as well as the kind

hospitality and support of the Institut des Hautes Études Scientifiques. R.F. acknowledges
further the discussions with K. Linde, V. Beffara, Y. Le Jan. W. Werner, he would like to
thank having drawn his attention to the restriction property very early, the discussions on
SLE and his help.
References
[1] G. Lawler, O. Schramm, W. Werner, Conformal restriction: the chordal case, math.PR/0209343.
[2] O. Schramm, Scaling limits of loop-erased random walks and uniform spanning trees, Israel J. Math. 118
(2000) 221.
[3] G. Lawler, O. Schramm, W. Werner, Values of Brownian intersection exponents I: half-plane exponents,
Acta Math. 187 (2001).
[4] G. Lawler, O. Schramm, W. Werner, Values of Brownian intersection exponents II: plane exponents, Acta
Math. 187 (2001).
[5] G. Lawler, O. Schramm, W. Werner, Values of Brownian intersection exponents III: two-sided exponents,
Ann. Inst. Henri Poincaré 38 (2002).
[6] S. Smirnov, Critical percolation in the plane: conformal invariance, Cardy’s formula, scaling limits, C. R.
Acad. Sci. Paris Ser. I Math. 333 (2001).
[7] R. Friedrich, W. Werner, Conformal fields, restriction properties, degenerate representations and SLE, C. R.
Acad. Sci. Paris Ser. I Math., math.PR/0209382.
[8] R. Friedrich, W. Werner, Conformal restriction, highest-weight representations and SLE, math-ph/0301018.
[9] Work in preparation.
[10] J. Cardy, Lectures on conformal invariance and percolation, math-ph/0103018.
[11] W. Werner, Lectures on random planar curves and Schramm–Loewner evolutions, Lecture Notes of the 2002
St.-Flour Summer School, Springer, submitted for publication.
[12] M. Bauer, D. Bernard, SLE martingales and the Virasoro algebra, hep-th/0301064.
[13] G. Lawler, Introduction to stochastic Loewner evolution, http://www.math.duke.edu/jose/papers.html.
[14] S. Rohde, O. Schramm, Basic properties of SLE curves, preprint, 2001.

[15] G.B. Segal, The definition of conformal field theory, in: Differential Geometrical Methods in Theoretical
Physics, in: NATO ASI Series, vol. 250, Kluwer, Dordrecht, 1987.
[16] E. Charpentier, K. Gawedzki, Wess–Zumino–Witten conformal field theory for simply laced groups at level
one, Ann. Phys. 213 (1992) 233.
[17] M. Bauer, D. Bernard, SLEκ growth processes and conformal field theories, Phys. Lett. B 543 (2002) 135,
math-ph/0206028.
[18] M. Bauer, D. Bernard, Conformal field theories of stochastic Loewner evolutions, Commun. Math. Phys. 239
(2003) 493, hep-th/0210015.
[19] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[20] P. Di Francesco, P. Mathieu, D. Senechal, Conformal Field Theory, Springer-Verlag, New York, 1997.
[21] I. Runkel, Boundary problems in conformal field theory, PhD thesis, King’s College, London, 2000.
[22] D. Friedan, S.H. Shenker, The analytic geometry of two-dimensional conformal field theory, Nucl. Phys.
B 281 (1987) 509.
[23] K. Ranganathan, H. Sonoda, B. Zwiebach, Connections on the state space over conformal field theories,
Nucl. Phys. B 414 (1994) 405, hep-th/9304053.
[24] V. Beffara, The dimension of the SLE curves, math.PR/0211322.
[25] E. Arbarello, M. Cornalba, The Picard groups of the moduli spaces of curves, Topology 26 (1987) 153.
[26] D. Mumford, Stability of projective varieties, Enseignement Math. 23 (1977) 39.
[27] M. Kontsevich, Virasoro algebra and Teichmüller spaces, Funktsional. Anal. Prilozhen. 21 (1987) 78.
[28] A.A. Beilinson, V.V. Schechtman, Determinant bundles and Virasoro algebras, Commun. Math. Phys. 118
(1988) 651.
[29] E. Frenkel, D. Ben-Zvi, Vertex Algebras and Algebraic Curves, in: Mathematical Surveys and Monographs,
vol. 88, AMS, Providence, MA, 2001.
[30] M. Bauer, D. Bernard, Conformal transformations and the SLE partition function martingale, math-
ph/0305061.
[31] M. Kontsevich, SLE, CFT, and phase boundaries, Arbeitstagung 2003, preprint MPI 2003 (60).
On vacuum energies and renomalizability

in integrable quantum field theories
Olalla Castro-Alvaredo a , Andreas Fring b
a Laboratoire de Physique, Ecole Normale Supérieure de Lyon, UMR 5672 du CNRS,
46 Allée d’Italie, 69364 Lyon cedex, France
b Centre for Mathematical Sciences, City University, Northampton Square, London EC1V 0HB, UK
Received 23 January 2004; received in revised form 12 March 2004; accepted 1 April 2004
Abstract
We compute for various perturbed conformal field theories the vacuum energies by means of
the thermodynamic Bethe ansatz. Depending on the infrared and ultraviolet divergencies of the
models, governed by the scaling dimensions of the underlying perturbed conformal field theory in
the ultraviolet, the vacuum energies exhibit different types of characteristics. In particular, for the
homogeneous sine-Gordon models we observe that once the conformal dimension of the perturbing
scalar field is smaller or greater than 1/2, the vacuum energies are positive or negative, respectively.
This behaviour indicates the need for additional ultraviolet counterterms in the latter case. At the
transition points we obtain an infinite vacuum energy, which is partly explainable with the presence
of several free fermions in the models studied.
 2004 Published by Elsevier B.V.
PACS: 11.10.-z; 11.55.Ds; 11.10.Kk
1. Introduction
According to the ideas developed first in [1] a large class of massive quantum field
theories in 1 + 1 space–time dimensions can be viewed as perturbed conformal field
theories (CFT) with Euclidian action

S = SCFT + λ d 2 x ϕ(x). (1.1)
E-mail addresses: ocastroa@ens-lyon.fr (O. Castro-Alvaredo), a.fring@city.ac.uk (A. Fring).
0550-3213/$ – see front matter  2004 Published by Elsevier B.V.

doi:10.1016/j.nuclphysb.2004.04.005
304 O. Castro-Alvaredo, A. Fring / Nuclear Physics B 687 [FS] (2004) 303–322
Here SCFT denotes a fixed point action, ϕ(x) a scalar field with (left, right) conformal
dimension (∆, ∆) and λ a coupling constant, scaling with (1 − ∆, 1 − ∆). The great virtue
of such theories is that very often they are integrable and can be solved exactly, that is to
all orders in perturbation theory. Since the original formulation various non-perturbative
techniques have been developed to study such theories with great success. Nonetheless,
once the CFT is well investigated one may also employ standard perturbative arguments
and unravel the meaning of certain types of behaviour in that more traditional language.
Accordingly, the vacuum expectation value of any local operator O can then be computed
as
∞
(−λ)n
O(z, z̄) = Z −1 d 2 z1 · · · d 2 zn O(z, z̄)ϕ(z1 , z̄1 ) · · · ϕ(zn , z̄n ) CFT .
n!
n=0
(1.2)
2
Here the normalization factor is in general Z = exp −λ d zϕ (z, z̄)CFT , with CFT
denoting the vacuum state related to SCFT . In quantum field theories such expressions are
plagued by various types of divergencies. First of all one encounters the infinities due to the
self-contraction of the fields, which can be regularized fairly easily by a normal ordering
prescription. Second, one might have ultraviolet (UV) singularities for (z − zi ) → 0.
Here the case O =ϕ will be important, for which we can approximate with the help of
standard CFT operator product expansion the integrals in (1.2) as ∼ dzi |z − zi |−2∆ .
Thus for ∆ < 1/2 the integrals in (1.2) remain finite, whereas for ∆ > 1/2 we require in
general counterterms to eliminate the divergencies. Third, one might have infrared (IR)
singularities for (z − zi ) → ∞. In the infinite plane it is usually an intricate issue to handle
them [2,3]. However, when formulating the theory from the very beginning on a cylinder
instead of an infinite plane the integrals in (1.2) will automatically be IR finite for ∆ > 0,
as the cylinder radius R constitutes a natural cut off. The fourth singularity occurring is
related to the fact, that even when the individual integrals in (1.2) are finite the entire series
will in general be IR divergent for large R.
Supposing now that one is able to compute (1.2) exactly, that is to all orders in
perturbation theory, the different types of renormalization quantities should be tractable
in that context. In fact, the thermodynamic Bethe ansatz (TBA) [4] is a method which
allows such identifications when O is taken to be the energy operator. The above mentioned
arguments hold when recalling [5] that this operator is proportional to the perturbing
field ϕ. Defining then for the ground state energy E(R) the scaling function c(R) =
−6RE(R)/π one encounters several types of general behaviours, which can all be brought
into the generic form
∞

c(r) = ceff + E0 r 2 + E0 r 2 ln r + En λn fn (r). (1.3)
n=1
Usually one uses the dimensionless parameter r = R/m with m being a mass scale and
E0 , E0 being finite real numbers. The function c(r) is normalized in such a way that
c(r = 0) coincides with the effective central charge ceff = c − 24∆min , with c being the
Virasoro central charge of the underlying ultraviolet conformal field theory and ∆min the
smallest conformal scaling dimension in the model. This constant c has the well-known
O. Castro-Alvaredo, A. Fring / Nuclear Physics B 687 [FS] (2004) 303–322 305
interpretation as the Casimir energy, which is the vacuum energy on the cylinder and
becomes zero when mapped onto the plane. Viewing (1.2) as resulting from a partition
function, the term E0 r 2 has to be present in (1.3), since thermodynamics dictates that
for large r the energy has to be proportional to the volume. In quantum field theoretic
terms both E0 r 2 and E0 r 2 ln r are related to renormalization issues, characterized by the
conformal dimension ∆ as described above. These terms are also needed in order to ensure
that limr→∞ c(r) = 0, which one expects for a purely massive model. Finally, the fn (r)
result from the integrals in (1.2) and takes on various general forms depending on the
regime in which ∆ is valued.
In this paper we will discuss more concretely the precise nature of the expansion (1.3).
We will first recall in Section 2 how the TBA can be used to compute the vacuum energies
and in the following sections we discuss the different regimes for different types of concrete
theories, the homogeneous sine-Gordon (HSG) models [6,7] and affine Toda field theories
(ATFT) [8,9]. These theories probe several regimes for ∆ and exhibit different types of
behaviours. In particular for the HSG-models, which are defined in the entire regime
0 < ∆ < 1, our results will be new. Our conclusions are stated in Section 6.
2. Vacuum energies from the TBA
Let us briefly recall the main steps of how vacuum energies may be computed [4]
(more details on the arguments can also be found in [10]) non-perturbatively with the
help of the TBA. One considers a relativistic theory in which the scattering matrices
Sij (θ ) for the particles of the type i, j with masses mi , mj are known as functions of
the rapidity θ . Then the entire TBA analysis can be formulated with only two inputs:
first the dynamical interaction, which enters via the logarithmic derivative of the S-matrix
ϕij (θ ) = −i d ln Sij (θ )/dθ and an assumption on the statistical interaction, which we
choose here to be of fermionic type. The thermodynamic Bethe ansatz equations are then
a set of coupled non-linear integral equations

rmi cosh θ = εi (θ, r) + ϕij ∗ ln 1 + e−εj (θ, r), (TBA)
j
where the pseudo-energies εi (θ, r) are the unknown quantities.

We denote as usual the
convolution of two functions by (f ∗ g)(θ ) := 1/(2π) dθ f (θ − θ )g(θ ). The scaling
parameter is related to the inverse temperature T as r = m/T , ml → ml /m, with m being
an overall mass scale. In [4] it was shown that when taking the sum and difference of the
derivatives d/dr(TBA) and d/dθ (TBA)/r one may derive a set of coupled linear integral
equations for the quantities
∂εi (θ, r) 1 ∂εi (θ, r)
i
ψ± (θ, r) = ± , (2.1)
∂r r ∂θ
respectively, namely

1

(θ, r) = mi e±θ +
j
ψ±i
ϕij ∗ εj ψ± (θ, r). (2.2)
e ±1
j
The strategy is now to solve first the equations (TBA) for εi (θ, r) and thereafter (2.2) for
ψ±i (θ, r). Once one has carried out the first step, one can already compute the scaling
function
∞
3r
c(r) = 2 mi dθ cosh θ Li (θ, r), (2.3)
π
i −∞
with Li (θ, r) = ln(1 + e−εi (θ,r) ). Concerning the status of analytical solutions for (2.2),
it is similar as for the TBA itself, that is only for free theories [10] a closed solution
was found and for interacting theories (2.2) was only solved in the extreme ultraviolet
limit. Numerical solutions exist even less. Once it is solved, one may compute the vacuum
expectation value of the trace of the energy–momentum tensor, i.e., vacuum energies
π2 d 1 i
T µµ = − c(r) = mi T+ + T−i (2.4)
3r dr 2
i
∞
1 1 i −θ

= mi dθ ψ+ (θ, r)e + ψ−
i
(θ, r)e θ
. (2.5)
2 1 + eεi (θ,r)
i −∞
In a parity invariant theory we have εi (θ, r) = εi (−θ, r) and consequently ψ+ i

(θ, r) =
ψ− (θ, r), T+ = T− = T such that matters simplify. We like to keep the treatment here
i i i i
generic for a while as we will also consider below the homogeneous sine-Gordon models,
which are not parity invariant.
There exists no systematic way to solve the equations (TBA) and (2.2) analytically,
albeit, numerically this is a solvable problem. Nonetheless, it is well known that at the fixed
points approximations can be made, such that one can solve (TBA) analytically and hence
also obtain analytic expressions for (2.3) at these points (r = 0 is one of them). Likewise we
expect to be able to solve (2.2) for these values and compute T µ µ analytically. Following
now essentially the argumentation of [4,10], we need to make only two assumptions:
(i) The logarithmic derivative of the scattering matrix in (TBA) admits an expansion of
the form
(s)
ϕij (θ ) = − ϕij e−s|θ| . (2.6)
s
The first coefficient in (2.6) is used to define a function ρij by extracting from ϕij(1) the
masses
ϕij(1) = ρij mi mj . (2.7)

The function ρij is specific to the particular theory.
(ii) One assumes that
ε̂i (θ ) − εi eθ for θ 0 (2.8)

where the εi are the pseudo-energies of the constant TBA equation and the ε̂i (θ ) are
quantities in the r-independent TBA-equation
ϕij ∗ L̂j (θ ) = −ε̂i (θ ) + mi eθ (2.9)

obtained from (TBA) by the shift θ → θ + ln(r/2), εi (θ, r) → ε̂i (θ ). This assumption
is usually difficult to justify a priori, but is sustained in hindsight by meaningful results
or supported by numerical data.
For θ → −∞ one can now derive with (2.8) the equation

1 θ (1) j
ϕij ∗ L̂j (θ ) = −εi +e ϕij T+ + O eηθ (2.10)
2π
where η 2. Comparing then (2.9) and (2.10) for the parity invariant case, it follows
directly with (2.8) that
1 (1) j
mi = ϕ T . (2.11)
2π ij
Finally we deduce the expression for the vacuum expectation value for the energy–
momentum tensor with (2.8) and (2.4) to
µ
T µ = 2π ρij−1 . (2.12)
i,j
This quantity is of course sensitive to above mentioned renormalization issues and possibly
exhibits the distinction between the different regimes quoted. Furthermore, one has the
possibility of comparison, as there are various other methods to obtain the vacuum
energies, such as the truncated conformal space approach [11] or a matching between the
high-energy behaviour of the scattering matrix with a Feynman diagramatic analysis [12].
Let us briefly comment on the different regimes:
0 < ∆ < 1/2: As mentioned in the introduction, in this regime the individual integrals in
the expansion (1.2) are UV and IR convergent term by term when formulated on
the cylinder. From general arguments one finds for the behaviour in (1.3) that
E0 = 0 and fn (r) = r 2n(1−∆) [13]. From (1.3) and (2.4) follows also that we can
identify T µ µ |r=0 = −π 2 /3E0 . Thermodynamically this term can be seen as the
infinite volume energy and field theoretically this corresponds to the sum of all
infrared substractions, which achieve the convergence of the sums (1.3) for large
r.
1/2 < ∆ < 1: Now the individual integrals in the expansion (1.2) are still IR convergent,
but cease to be UV convergent. We may still deduce Eq. (2.12) from (2.4) as
this just requires the validity of the assumptions (i)–(ii). Less obvious is if one
can still make the identification T µ µ |r=0 = −π 2 /3E0 . If in the series (1.3) the
coefficients En vanish for all n for which ∆ > (2n − 2)/2n then we may still
take E0 = 0 and the identification of T µ µ |r=0 with the bulk term still holds
in the same manner as in the previous case. Unfortunately, at present explicit
examples for the series in this regime are not available. In any case, now the field
theoretic interpretation of this term changes. Since we require in this case UV

counterterms to make the individual integrals finite, the E0 -term corresponds now
to the sum of these UV counterterms and all infrared substractions, which achieve
the convergence of the sums (1.3) for large r. Indeed, for the concrete models
studied below this becomes visible in a change of sign in the transition from the
regime ∆ < 1/2 to ∆ > 1/2.
∆ < 0: Now the individual integrals in the expansion (1.3) are still UV convergent, but
cease to be IR convergent even on a cylinder. General arguments now yield for
the behaviour in (1.3) that E0 = 0, E0 = 0 and fn (r) = (α + ln(r))−n with α being
some constant [14–18]. One still finds that T µ µ |r=0 = −π 2 /3E0 , e.g., [12], but
now the interpretation is less obvious as some counterterms also accumulate in
the E0 -term.
∆ = 1/2: In this case one usually finds free fermions in the model and E0 = 0, E0 = 0,
fn (r) = r n . Now the vacuum energy is divergent, see, e.g., [10] for an analytical
expression.
We will investigate some concrete theories.
3. 0 < ∆ < 1/2, minimal affine Toda field theories
These theories have been studied before [10,19], nonetheless, we recall them here as
they are easy examples which illustrate the working of the above formulae and we shall
also point out some novel features. We recall first that minimal affine Toda field theories
can be realized as perturbations of the coset conformal field theories g1 ⊗ g1 /g2 , with gk
being a simply laced Kac–Moody algebra of rank and level k [20,21]. The corresponding
Virasoro central charges c and conformal dimension of the perturbing operator ∆ are
2 2
c= and ∆ = , (3.1)
2+h 2+h
respectively. Apart from h = 2, i.e., the free fermion with g = A1 , we always have for
the Coxeter number h > 2 and are therefore in the stated regime 0 < ∆ < 1/2. The
renormalization issues are handled most easily in this case and the vacuum energies
are computable with the above arguments. With regard to assumption (i), we recall the
expansion of the TBA-kernel for these theories [10,22,23]
sπ
ϕij (θ ) = −4 cot xi (s)xj (s)e−s|θ| , (3.2)
h
s∈E
with E = {s + nh}, s being an exponent of g, n ∈ N0 and xi (s) are the left eigenvectors
of the Cartan matrix. In particular, we have xi (1) = mi /m, with m being an overall mass
scale, which is needed for the assumption (ii) to hold. Having therefore the quantity ϕij(1)
in the form (2.7), we deduce immediately with the help of (2.12)
µ π π
T µ = m2 tan . (3.3)
2 h
Obviously apart from the free fermion with h = 2, when T µ µ → ∞, we have T µ µ > 0.
(1)
This result agrees with a similar formula obtained in [10] in terms of the coefficients ϕ11
without explicit evaluation and “1” referring to the lightest particle. More concrete case-
by-case studies were carried out in [19] for perturbations of gl ⊗ gk /gk+l -coset CFT’s
(see formulae (3.14) therein). When using the overall mass scale to perform suitable
normalizations the formula for k = l = 1 in there can be brought into the universal formula
(3.3), which is not obvious at first sight. The formulae in [19] are expressed in terms of a
mass scale M whose relation with our m varies for every theory as
π
A : M = m sin ,
−1
√
D : M = m/ 2,

3 π
E6 : M =m sin ,
2 12

π 2π
E7 : M = m sin sin ,
18 9

π π
E8 : M = m 2 sin sin . (3.4)
30 5
The advantage of our formulation relies on the fact that the masses are normalized with
respect to the same general mass scale m for all simply laced Lie algebras, which allows
for the very compact and generic expression (3.3). Alternatively these results were also
confirmed in [24].
4. 0 < ∆ < 1, gk -homogeneous sine-Gordon models
Let us now consider a theory which is more interesting with regard to the above
mentioned problematic, namely the gk -HSG model [6,7], with g being a simple Lie algebra
of rank and level k. These models can be viewed as perturbed Wess–Zumino–Novikov–
Witten (WZNW) [25] coset-models

m2
SHSG = SWZNW + 2
x )−1 Λ− g(
d 2 x Λ+ , g( x) . (4.1)
πβ
Here , denotes the Killing form of g and g(
x ) a group valued bosonic scalar field. Λ±
are arbitrary semi-simple elements of the Cartan subalgebra associated with the maximal
Abelian torus h ⊂ g, which have to be chosen not to be orthogonal to any root of g. The
parameters m and β are the bare mass scale and the coupling constant, respectively. The
Virasoro central charge of the coset model and the dimension of the perturbing operator
are computed to
kh − h∨ h∨
c= and ∆ = , (4.2)
k + h∨ k + h∨
with (h∨ ) h being the (dual) Coxeter number of g. We note that now the constraint ∆ < 1/2
does not automatically hold for each level and the above mentioned complications could
arise for some theories in this series when changing from k > h∨ to k < h∨ . Up to now no
indication for a different behaviour of the theories in this two different regimes have been
found in the literature. We treat the simply laced and non-simply laced cases separately.
4.1. Simply laced HSG-models
As in the original formulation of these models, the algebra g is assumed to be simply

laced. Since for this case the expansion of the kernel ϕ does not appear in the literature,
we will start with the scattering matrix, which was found originally in [26] (see [27] for an
integral representation). We cast the matrix describing the scattering between the particle
of type A = (a, ã) and B = (b, b̃), with 1 ã, b̃ ; 1 a, b < k into the form

dt sinh(at/k) sinh[(k − b)t/k] −it (θ+σ )/π
ã b̃
Sab (θ ) = ηab
ã b̃
exp K̃ãb̃ (t) e ã b̃ . (4.3)
t sinh(t/k) sinh t
Here ηab ã b̃
= exp[iπεãb̃ (2 − IAk−1 )−1
āb ] are constant phase factors not relevant for our
analysis, K̃ã b̃ (t) = 2δãb̃ cosh t/k − Iã b̃ with I being the incidence matrix of g and the
σ ’s are the resonance parameters, which indicate the presence of unstable particles in these
models. In order to evaluate the expansion for ϕ, we can employ the residue theorem for a
contour along the real axis closing up in the positive half of the complex plane encircling
all poles on the imaginary axis in the upper half plane. Noting that in (4.3) t = iπn are
simple poles, except for t = iπnk which constitute double poles for n ∈ N, we deduce for
σã b̃ = 0
∞

1 sinh(at/k) sinh[(k − b)t/k] −it θ/π
ã b̃
ϕab (θ ) = 2πi Res − K̃ã b̃ (t) e
t =iπs π sinh(t/k) sinh t
s=1;s =nk
(4.4)
∞
sin(aπs/k) sin(bπs/k) −s|θ|
= −2 K̃ã b̃ (iπs) e . (4.5)
sin(πs/k)
s=1;s =nk
The desired coefficient (2.7) follows from this directly to

(1)
ϕ = 2K̃ãb̃ (iπ)ma mb /m2 sin(π/k) (4.6)
ã b̃
where mãa = ma mã with ma = sin aπ/k being the masses of Ak−1 -affine Toda field theory
and mã are free mass scales. We choose them here to be all equal mã = m ∀ã. Finally
we derive from this a closed expression for the vacuum expectation value for the trace of
energy–momentum tensor

−1
T µ µ = πm2 sin(π/k) K̃ (iπ) ãb̃ . (4.7)
ã,b̃=1
We are not aware of a generic formulation for K̃ −1 (iπ) and analyze therefore the
expression (4.7) below in more detail case-by-case. We can summarize our findings as
for k > h ≡ ∆ < 1/2,
µ >0
T µ → ∞ for k = h ≡ ∆ = 1/2, (4.8)
<0 for k < h ≡ ∆ > 1/2.
In many cases we can attribute the divergence for ∆ = 1/2 to the presence of free fermions.
The change of sign when going from ∆ < 1/2 to ∆ > 1/2 reflects the fact that besides the
IR counterterms, which achieve the convergence of the sums (1.2) for large r, needed in
both cases in the latter we also require UV counterterms to make the individual integrals
in (1.2) finite.
4.1.1. (A )k -HSG model

For A the Coxeter number is h = + 1. The inverse of the matrix relevant in (4.7) can
be cast in this case into a simple formula

sin(ãπ/k) sin[(h − b̃)π/k]
K̃A−1 (iπ) ã b̃
= for ã b̃. (4.9)
sin(π/k) sin(hπ/k)
Computing the sums over both entries then yields after some algebra

µ πm2 hπ π
T µ = tan − h tan . (4.10)
2 tan2 π/2k 2k 2k
Hence, the condition T µ µ > 0 becomes
hπ π
tan > h tan (4.11)
2k 2k
or equivalently, when expanding the tan,
∞ ∞
4 h 1 4 1 1
> h . (4.12)
πk (2n − 1)2 − (h/k)2 πk (2n − 1)2 − (1/k)2
n=1 n=1
It is easily seen that (4.12) holds term by term once h/k < 1, hence establishing the first
inequality in (4.8). Similar arguments show that the opposite inequality holds in the regime
h/k > 1. We comment more on the case k = h below.
4.1.2. (D )k -HSG model

For D the Coxeter number is h = 2 − 2 and by evaluating (4.7) similarly as in the
previous subsection, we find
πm2 sin πk [2 − (2 + h) cos hπ
2k ] + 2πm sin 2k
2 hπ
T µµ = π
. (4.13)
sin2 2k cos hπ
2k
The condition T µ µ > 0 is now equivalent to

π 2 hπ
sin (2 + h) − hπ
< 2 tan . (4.14)
k cos 2k 2k
Expanding the left- and right-hand side of this inequality yields by similar arguments as in
the previous subsection once more the relation (4.8).
4.1.3. (E6 )k -HSG model

For E6 the Coxeter number is h = 12 and we find
4
µ p=1 τp sin pπ/k
T µ = 2πm 2
, τ = (4, 4, 5, 3). (4.15)
2 cos 4π/k − 1
We see that the numerator is < 0 for k = 2 and > 0 for k > 2. The denominator is > 0 for
k = 2, k > 12 and < 0 for 2 < k < 12. The denominator vanishes for k = 12. Hence the
relation (4.8) holds.


µ πm2 7p=1 τp sin pπ/k
T µ = , τ = (9, 18, 20, 22, 17, 12, 7). (4.16)
cos π/k(4 cos 6π/k − 2)
We observe now that the numerator is < 0 for k < 4 and > 0 otherwise. The denominator
on the other hand is > 0 for k = 3, k > 18 and < 0 otherwise except for k = 2, 18 in which
case it is zero. Hence (4.8) holds also in this case.


µ πm2 7p=1 τp sin pπ/k
T µ = ,
cos 8π/k + cos 6π/k − cos 2π/k − 1/2
τ = (4, 8, 12, 12, 13, 10, 7, 4). (4.17)
We see that the numerator is < 0 for k < 5 and > 0 otherwise. The denominator is > 0 for
k = 2, 3, 4; k > 30 and < 0 otherwise except for k = 30 in which case it is zero. Hence
(4.8) holds also in this case.
4.2. Non-simply laced HSG-models
Now we allow the algebra g to be also non-simply laced. In this case the scattering
matrix is slightly more complicated as the symmetry between the exchange of long and
short roots is lost. It can be restored by the use of the symmetrizers tã of the incidence
matrix of g, i.e., tã Iã b̃ = tb̃ Ib̃ã , with tã = 2/αã2 and the length of long roots normalized to
αl2 = 2. In [28] the scattering between the particle of type A = (a, ã) and B = (b, b̃), with
1 ã, b̃ ; 1 a, b < kã = tã k was proposed to be described by

dt sinh(at/kã ) sinh[(1 − b/kb̃ )t] −it (θ+σ )/π
Sab (θ ) = ηab exp
ã b̃ ã b̃
K̃ (t) e ãb̃ . (4.18)
t ãb̃ sinh(t/kã b̃ ) sinh t
Here ηab ã b̃ are once more constant phase factors not relevant for our analysis. Furthermore,
one needs the quantity kãb̃ = k max(tã, tb̃ ) and the matrix K̃ with entries K̃ãb̃ (t) =
2δãb̃ cosh t/kã − Iã b̃ tb̃ / max(tã, tb̃ ). Note that in comparison with [28] we interchanged the
long and short roots, that is we have taken the t’s to be left and not the right symmetrizers
of the incidence matrix. A similar analysis as in the previous subsection leads now to the
following expansion of the TBA-kernel
∞
sin(aπs/kã ) sin(bπs/kb̃ ) −s|θ|
ã b̃
ϕab (θ ) = −2 K̃ãb̃ (iπs) e , (4.19)
sin(πs/kãb̃ )
s=1;s =nkãb̃
such that
ϕ (1) = 2K̃ãb̃ (iπ)mãa mb̃b /m2 sin(π/kã b̃ ). (4.20)

ã b̃
Here the masses are assumed to renormalize with an overall factor and are therefore
expected to be the same as in the semi-classical analysis [29], that is mãa = m sin aπ/kã .
The overall mass scales associated with each colour are once more chosen to be the same.
Thus we finally deduce

−1
T µ µ = πm2 K̂ ã b̃
, (4.21)
ã,b̃=1
where K̂ãb̃ = K̃ã b̃ (iπ) sin(π/kãb̃ ). As in the previous case, we are not aware of a generic
formulation for K̂ −1 and analyze therefore (4.21) in more detail case-by-case. Our findings
are summarized as

µ > 0 for k > h∨ ≡ ∆ < 1/2,
T µ → ∞ for k = h∨ ≡ ∆ = 1/2, (4.22)

<0 for k < h∨ ≡ ∆ < 1/2,
with similar interpretations as in (4.8). We establish (4.22) in more detail case-by-case.
4.2.1. (G2 )k -HSG model

The dual Coxeter number for G2 is h∨ = 4 and the symmetrizers are taken to be
t1 = 3, t2 = 1. With these data we compute from (4.21)
sin π/k + sin 4π/3k
T µ µ = 2πm2 . (4.23)
2 cos 4π/3k − 1
Obviously, the numerator is > 0 for k 2, whereas the denominator is < 0 for k = 2, 3
and otherwise > 0 except for k = 4 when it is zero. Evidently this agrees with (4.22).
4.2.2. (F4 )k -HSG model

The dual Coxeter number for F4 is h∨ = 9 and the symmetrizers are taken to be
t1 = t2 = 1 and t3 = t4 = 2. From (4.21) we compute

µ 2 6p=1 sin pπ/2k − sin 3π/2k
T µ = 2πm 2
. (4.24)
2 cos 3π/k − 1
The numerator is > 0 for k 2, whereas the denominator is < 0 for 2 k < 9 and
otherwise > 0 except for k = 9 when it is zero. Evidently this agrees with (4.22).
4.2.3. (B )k -HSG model

The dual Coxeter number for B is h∨ = 2 − 1 and the symmetrizers are taken to be
t1 = t2 = · · · = t−1 = 2 and t = 1. We find now for even rank
πm2
T µµ = − π
cos 2k
h∨ −1 πh∨ π (−2)/2 ∨
p=2 sin πp
2k + 2 sin 2k + 2 cos 2k

p=1 ( − 2p − 1) sin π(1+h2k −4p)
× /2 ,
1 + 2 p=1 (−1)p cos πpk
(4.25)
whereas for odd we obtain
πm2
T µµ = π
cos 2k
h∨ −1 πh∨ π (−3)/2 ∨
p=1 sin πp
2k + 2 sin 2k + 2 cos 2k

p=1 ( − 2p − 1) sin π(1+h2k −4p)
× /2 .
1 + 2 p=1 (−1)p cos πpk
(4.26)
Once more we confirm (4.22). As the details are rather involved we drop them here.
4.2.4. (C )k -HSG model

The dual Coxeter number for C is h∨ = + 1 and the symmetrizers are taken to be
t1 = t2 = · · · = t−1 = 1 and t = 2.

h∨ −1 πh∨
πm2 (i)(p − 1) sin πp
p=1 2k + 2 sin 2k

µT µ
= π /2 , for even, (4.27)
1 + 2 p=1 (−1)p cos πp
cos 2k
k
h∨ −1
µ πm2 πp πh∨
T µ = ∨ (p − 1) sin + sin , for odd. (4.28)
cos πh
2k p=2
2k 2 2k
Once more we confirm (4.22) and drop the details for the same reasons as in the previous
subsection.
5. ∆ = 1/2, gh∨ -homogeneous sine-Gordon model
The case ∆ = 1/2 is very special as then the vacuum energy diverges in the extreme
UV limit. Such type of behaviour is well known from free fermions in form of logarithmic
ultraviolet singularities, meaning that (2.4) yields T µ µ → ∞ for r → 0. Explicit analytic
formulae for the free fermion c(r)-function can be found in [10]. Indeed in many cases we
can make this connection quite explicit. It suffices to present an examples to illustrate this
point.
5.1. (A )+1 -HSG theories
Let us have a closer look at the (A )+1 -HSG theories in order to see how the fermions
arise in there. Obviously for h = k the expression (4.10) yields T µ µ → ∞. Already in
[27] it was noticed that the (A2 )3 -HSG model decomposes into four free fermions when
the resonance parameter vanishes. From the fact that the central charge (4.2) becomes in
general 2 /2 for (A )+1 -HSG models, one might suspect that they always decompose
completely into 2 free fermions for vanishing resonance parameters, such that each
fermion contributes 1/2 to the central charge. However, this is not quite the case as the
following argument shows.
In order to count the fermions, identified here simply with the amount of particles which
contribute 1/2 to the central charge, we recall the constant TBA equations, which arise
from (TBA) after some standard manipulations. For the (A )+1 -HSG models they take on
the form

N ã b̃
xaã = 1 + xbb̃ ab ã b̃
with Nab = δab δãb̃ − KA−1
ã b̃
(KA )ab . (5.1)
b,b̃=1
Solving these equations for the xaã = exp(−εaã ) yields the effective central charge as

6

xaã 2
ceff = L = , (5.2)
π 2 1 + xa ã 2
a,ã=1

with L(x) = ∞ n=1 x /n + ln x ln(1 − x)/2 denoting Rogers dilogarithm. The solutions
n 2
of (5.1) are very simple in this case

sin[π ã/(1 + )]
xaã = . (5.3)
sin[πa/(1 + )]
Therefore we have xaa = xa+1−a = 1 and since L(1/2) = π 2 /12 it follows from this that
each of the particles (a, a), (a, + 1 − a) for 1 a contributes 1/2 to the effective
central charge in (5.2). Hence in the (A )+1 -HSG models we have always 2 or 2 − 1
free fermions when is odd or even, respectively. The remaining particles can be organized
without exceptions in pairs (a, ã), (ã, a). Noting with (5.3) that obviously xaã = (xãa )−1 and
recalling the fact that L(x) + L(1 − x) = π 2 /6 explains then that the central charge has to
be an integer or a semi-integer for these models.
In general, it is less straightforward for the other algebras to identify particles which
directly contribute 1/2 to the central charge. In fact, mostly the particles occur in pairs,
triplets or higher multiplets contributing integers or semi-integer values to c.
6. ∆ < 0, affine Toda field theories
Affine Toda field theories related to simply laced and non-simply laced Lie algebras
have a quite different behaviour due to the fact that in the first case all masses renormalize
with an overall factor, which is not the case in the latter (see, e.g., [30]). As a result of this,
the strong–weak duality observed for ATFT related to simply laced algebras is broken for
those associated with non-simply laced Lie algebras. Despite the fact that there exists a
uniform formulation, we will treat them here separately as this will be more transparent.
6.1. Simply laced
ATFT are quite well studied examples of integrable models, which can be viewed in the
spirit of (1.1) which was noted first in [20,21]

1
SATFT = 2+µ
d 2 x (∂µ ϕ) ni eβ α i ·ϕ . (6.1)
2
i=0
The fixed point part of the action SCFT corresponds to the conformal Toda field theories
when the sum over the simple roots α i starts at i = 1. The µ, β are real parameters and
the ni are the Kac labels related to the negative of the highest root α 0 = − i=1 ni α i . The
Virasoro central charge of the conformal Toda field theories and the conformal dimension
of the perturbing operator V = µn0 eβ α 0 ·ϕ have been computed in [21]
4h(h + 1) 2h
c=+ and ∆ = 1 − , (6.2)
B(2 − B) 2−B
where we use the effective coupling1 0 B = 2β 2 /(β 2 + 4π) 2. Since 2h > 1 − B/2
is always true we are in the regime ∆ < 0 and expect the above mentioned complications
with regard to renormalization to arise. To establish that, we recall first [10,22,23]

sπB sπ(2 − B) sπ
ϕij (θ ) = −2 sin sin sin xi (s)xj (s)e−s|θ| , (6.3)
2h 2h h
s∈E
and deduce thereafter from (2.7) and (2.12)

πm2 sin(π/ h)
T µµ = . (6.4)
sin(πB/2h) sin[π(2 − B)/2h]
Clearly, as 0 B 2 we have T µ µ > 0. Up to an overall mass re-scaling of m → 2m,
this agrees precisely with the results in [12], which were obtained by matching the high-
energy behaviour of the scattering matrix with a Feynman diagramatic analysis.
It is very interesting to note that by an analytic continuation from real to purely complex
coupling we can also reach the regime for ∆ > 0 for (6.4) and observe similar phenomena
as for the HSG-models.2 For h = 2 this means we continue from sinh-Gordon to sine-
Gordon. Following for this case the argumentation of Destri and De Vega√[12], we relate
the sinh-Gordon coupling β to the sine-Gordon coupling β̃ via β → i β̃/ 2 according to
the standard conventions. Also we replace the breather mass scale m with the soliton mass
1 Confusion arises sometimes due to different conventions. For instance, we can relate our notations to the
√
ones used in [31]√by a simple rescaling of the fields ϕ = ϕF / 4π compensated by a scaling of the coupling
constant β = bF / 4π . In addition, one takes the effective coupling constant to be B = 2BF .
2 We are grateful to Al.B. Zamolodchikov for pointing this out to us.
scale m̃ via m2 → 4m̃2 sin2 πB/2 such that we end up with the simple formula

µ π ∆
T µ = π m̃ tan
2
, (6.5)
2 ∆−1
where ∆ = β̃ 2 /8π is the conformal dimension of the perturbing cos-term in the model.
This agrees also with [32]. Note that in the previous argument we considered sinh-Gordon
as a perturbed Liouville theory, whereas now we perturb the free theory rather than
complex Liouville. Analyzing (6.5) in more detail one observes


< 0 for 2n−2 2n−1
2n−1 < ∆ < 2n ,


µ → ∞ for ∆ = 2n−12n ,
T µ (6.6)
> 0

2n−1
for 2n < ∆ < 2n+1 , 2n


=0 for ∆ = 2n+12n
,
with n ∈ N. Note that in particular for n = 1 we have as for the homogeneous sine-Gordon
model at ∆ = 1/2 a transition point at which the sign changes by passing through a
singularity. Moreover, precisely this value corresponds to the free fermion point, which in
this case is a very explicit example for the free fermion picture advocated above. However,
in this case the structure is more complicated as first of all we have an infinite number of
such points rather than just one as in the HSG-models. In addition T µ µ is not always
divergent at these points, but can also vanish. One should keep in mind here, that in fact
we obtained (6.6) simply by an analytic continuation from the sinh-Gordon expression.
Starting from first principles the TBA works quite differently for the sine-Gordon theory
due to the non-diagonal nature of the scattering matrix.
6.2. Non-simply laced
It is known, that the above mentioned complication of mass renormalization is

reconciled if one views ATFT’s in terms of dual pairs of Lie algebras. Since simply laced
Lie algebras are self-dual, this picture does not yield anything new for that case. The
dual pairs of non-simply laced Lie algebras are (G(1) (3) (1) (2) (1) (2)
2 , D4 ), (F4 , E6 ), (B , A2−1 ) and
(C(1) (2)
, D+1 ). Each algebra of these pairs allows for a description of the form (6.1) related
to each other by the strong–weak duality transformation β → 4π/β, where the untwisted
algebras relate to the weak coupling limit. The vacuum energies associated to all non-
simply laced affine Toda theories were stated in [18]. As in there no details were presented
on how they were obtained, it will be instructive to show that the procedure outlined in
Section 2, also holds in this case.
Let us first of all see what we have to expect with regard to the arguments outlined above
and compute the Virasoro central charge and the dimension of the perturbing operator.
According to [18] we have
2 = β ρ + 1 ∨
c = + 12Q with Q ρ , (6.7)
β
with (ρ ∨ ) ρ being the (dual) Weyl vector of the untwisted Lie algebra given by half the
sum of the positive (co)roots. Note that when evaluating (6.7) for the simply laced case
yields precisely (6.2), but for the non-simply laced case it differs from the expressions
in [21] by the use of ρ ∨ rather than always ρ. The conformal dimension of a spinless
x ) in the underlying CFT is ∆( 2 − a 2 )/2, such
primary field Va (x) = e(Q+ a )·ϕ( a ) = (Q
that the perturbing field µn0 V(β α 0 −Q)
(x) has conformal dimension
β 2 α 02
= β α 0 . Q
∆(β α 0 − Q) − , (6.8)
2
α 0 defined as in the previous section, that is being the negative of the highest root. It will
turn out that these dimension will always be smaller zero. We will compute the precise
values for some concrete examples below.
Unlike to the previous cases the expansion for the kernel ϕ does not appear in the
literature, we therefore start here with the scattering matrix, which can be cast into the
universal form [33,34]

dt −1 ±it θ/π
Sij (±θ > 0) = exp ∓8 sinh(ϑh t) sinh(tj ϑH t) K (t) ij e , (6.9)
t
where ϑh = (2 − B)/2h, ϑH = B/2H with h being the Coxeter number of the untwisted
algebra and H its dual Coxeter number h∨ multiplied by the twist of the second algebra.
The effective coupling is now generalized to B = 2Hβ 2/(Hβ 2 + 4πh). The ti are the
symmetrizers of the incidence matrix of the untwisted algebra ti Iij = tj Ij i , with ti = 2/ αi2
and the length of long roots normalized to α l2 = 2. Also needed in (6.9) is the inverse
of the q-deformed Cartan matrix Kij (t) = (q q̄ tj + q −1 q̄ −tj )δij − [Iij ]q̄ with deformation
parameters q = exp(tϑh ), q̄ = exp(tϑH ) and [Iij ]q̄ = (q̄ Iij − q̄ −Iij )/(q̄ − q̄ −1 ).
In order to evaluate the expansion for ϕ, we can employ once again the residue
theorem for a contour along the real axis closing up in the positive half of the complex
plane encircling
all poles on the imaginary axis in the upper half plane. Recalling that
det K(t) = s∈E 4 cosh[(t + iπs)/2h] cosh[(t − iπs)/2h], we know the positions of all
poles and it follows from the integral representation (6.9) that the TBA kernels admit a
series expansion of the form

ϕij (θ ) = 16i Res sinh(ϑh t) sinh(tj ϑH t)Ǩ(t)ij / det K(t)eit θ/π . (6.10)
t →iπs
s∈E
We do not have a closed formula for the cofactors Ǩ, but for the sake of our argument it
will be sufficient here to present some examples.
(1) (3)
6.2.1. (G2 , D4 )-ATFT
Let us first compute (6.7) and (6.8). We carry out the computations in terms of the
quantities of the untwisted algebra G(1)2 for which we have two simple roots α 1 and α 2
normalized as α 2 = 2 = 3
2 2
α1 . Furthermore, the Weyl vector, its dual and the negative of
the highest root are given by
ρ = 5
α1 + 3
α2 , ρ ∨ = 5
α1 + α 2 and α0 = −3
α1 − 2
α2 . (6.11)
These realizations allow to compute the quantities needed in (6.7) and (6.8), that is
3ρ 2 = 14, 3ρ ∨ . ρ ∨ = 26 and 3ρ . ρ ∨ = 3ρ ∨ . ρ = 8. It follows therefore

13 + 3B(B − 3) 3B + 2
c = 2 + 32 and ∆ = . (6.12)
B(2 − B) B −2
Clearly for 0 < B < 2 we have −∞ ∆ −1.
To proceed further we need the (generalized) Coxeter number for this theory, which are
h = 6 and H = 12. The symmetrizers are t1 = 3 and t2 = 1. Evaluating (6.10) and reading
off the first order coefficient we obtain
(1)
√ sin (2−B)π sin Bπ
8 ma mb
ϕab = −8 3 12
, a, b = 1, 2, (6.13)
cos 6 (1 − 4 )
π B m2
where we normalized the masses to

π B
m1 = m cos 1+ and m2 = m. (6.14)
6 4
We deduce then with (2.12)
πm2 cos π6 (1 − B4 )
T µµ = √ . (6.15)
4 3 sin (2−B)π
12 sin Bπ8
Agreement with the results in [18] is achieved by changing to the conventions used in
there. For this one needs to re-define the effective coupling to B → B = 3B/(4 + B) and
introduce a “floating” Coxeter number H = (1 − B )h + B h∨ .
(1) (2)
6.2.2. (F4 , E6 )-ATFT
For F(1)
4 we normalize the four simple roots to α 12 = α 22 = 2
α32 = 2
α42 = 2. The Weyl
vector, its dual and the negative of the highest root are in this case given by
α1 + 30
ρ = 16 α2 + 42
α3 + 22
α4 ,
ρ ∨ = ρ + 22
α4 and α 0 = −2
α1 − 3
α2 − 4
α3 − 2
α4 (6.16)
such that ρ 2 = 39, ρ ∨ . ρ ∨ = 402 and ρ . ρ ∨ = ρ ∨ . ρ = 55. With this we find

1608 + B(331B − 1388) 16 + B
c = 4 + 12 and ∆ = . (6.17)
(2 − B)B B −2
Therefore −∞ ∆ −8.
For this theory we have h = 12, H = 18, t1 = t2 = 1 and t3 = t4 = 2. The ratios between
the masses of the four particles in the theory are

m4 π B m3 π B
= 2 sin 1+ , = 1 + 2 cos 1− ,
m1 4 18 m1 6 6

m2 π B
= 2 cos 1− . (6.18)
m1 12 6
We choose the normalization such that m1 = m and obtain from (6.10)
(1)
√ sin (2−B)π sin Bπ
18 ma mb
ϕab = −8 3 24
, a, b = 1, 2, 3, 4. (6.19)
cos 4 1 − 18
π B m2
Therefore with (2.12) we get 894/4 = 447

2
πm2 cos π4 (1 − 18
B
)
T µµ = √ (2−B)π
. (6.20)
4 3 sin 24 sin Bπ 18
We can match with the formulae in [18] by B → B = 4B/(6 + B), H̃ = 3(4 − B ),

m1 → m1 , m2 → m3 , m3 → m4 and m4 → m2 .
6.2.3. (B(1) (2)

2 , A3 )-ATFT
(1) (2)
Let us now present the simplest example of the family (B , A2−1 ). In general, we
(1)
choose for the algebra B the normalizations α i2 = 2 for i = 1, . . . , − 1 and α 2 = 1.
Then we have
2ρ = 3
α1 + 4
α2 , 2ρ ∨ = 3
α1 + 8
α2 and α 0 = −
α1 − 2
α2 , (6.21)
from which we compute 12ρ 2 = 30, ρ ∨ . ρ ∨ = ρ 2 + 72 and ρ . ρ ∨ = ρ 2 + 4. Therefore

447 + 24B(4B − 17) B +4
c=2+8 and ∆ = .
B(2 − B) B −2
Hence −∞ ∆ −2.
For this theory we have h = 4, H = 6, t1 = 1 and t2 = 2. The masses satisfy

m1 π B
= 2 sin 1+ , (6.22)
m2 4 6
and we choose m1 = m. Evaluating (6.10) we obtain now
(1) sin (2−B)π sin Bπ

6 ma mb
ϕab = −8 8
, a, b = 1, 2, (6.23)
sin 4 (1 + 4 )
π B m2
and therefore with (2.12)
πm2 sin π4 (1 + B4 )
T µµ = . (6.24)
4 sin (2−B)π
8 sin Bπ
6
Defining once more B → B = 4B/(6 + B) and H = 4 − B we find agreement with

(1) (2)
[18]. The previous results also hold for the (C2 , D3 )-theory by exchanging the roles
of particles 1 and 2, since the Dynkin diagrams of B(1) (1)
2 and C2 are identical up to the
exchange of the short and the long root.
These examples are sufficient to support the validity of the approach outlined in
Section 2.
7. Conclusions
We used the thermodynamic Bethe ansatz to compute vacuum energies T µ µ for

various types of perturbed conformal field theories. Despite the fact, that the models
considered exhibit different general behaviours, the assumption (i)–(ii), needed for the
validity of the approximations in the TBA, hold in all cases.
The general behaviour of T µ µ is shown to be sensitive to IR- and UV-counterterms,
whose presence can be characterized by the conformal scaling dimension ∆ of the
perturbing operator. In the regime 0 < ∆ < 1/2, realized by minimal ATFT and gk -
HSG models for k > h∨ , the quantity T µ µ can be identified with the IR-counterterms
needed to compensate the divergencies in the perturbative series expansion (1.2), when
viewed on a cylinder. In contrast, in the regime 1/2 < ∆ < 1, realized by gk -HSG models
for k < h∨ , the quantity T µ µ can be associated to the sum of the aforementioned IR
counterterms and UV counterterms needed to guarantee the finiteness of the individual
integrals in the expansion. In the models studied here these additional counterterms, when
passing from ∆ < 1/2 to ∆ > 1/2 show up in a change of sign in T µ µ . It would be
extremely interesting to verify this assertion by some explicit perturbative computations
for the HSG-models. For the regime ∆ < 0, realized here by the ATFT (simply laced as
well as non-simply laced) T µ µ constitutes a mixture of several types of counterterms,
less obvious to disentangle. The divergence of T µ µ at ∆ = 1/2 can be attributed to the
occurrence of free fermions, for which such type of behaviour is well known from explicit
analytical expressions. However, we were not able to identify the free fermions in all gh∨ -
HSG models, which can be viewed as perturbed CFT’s with ∆ = 1/2. This needs further
investigations.
Acknowledgements
This work is supported by the EU network “EUCLID, Integrable models and applica-
tions: from strings to condensed matter”, HPRN-CT-2002-00325 and was partly carried
out at the Freie Universität Berlin where it was supported by the Sonderforschungsbereich
SFB288. We are grateful to Al.B. Zamolodchikov for very useful discussions.
References
[1] A.B. Zamolodchikov, Integrals of motion and S-matrix of the (scaled) T = Tc Ising model with magnetic
field, Int. J. Mod. Phys. A 4 (1989) 4235.
[2] F. David, Cancellations of infrared divergencies in the two-dimensional nonlinear sigma models, Commun.
Math. Phys. 81 (1981) 149.
[3] M.C. Bergere, F. David, Nonanalyticity of the perturbative expansion for superrenormalizable massless field
theories, Ann. Phys. 142 (1982) 416.
[4] A.B. Zamolodchikov, Thermodynamic Bethe ansatz in relativistic models: scaling 3-state Potts and Lee–
Yang models, Nucl. Phys. B 342 (1990) 695–720.
[5] J.L. Cardy, The central charge and universal combinations of amplitudes in two-dimensional theories away
from criticality, Phys. Rev. Lett. 60 (1988) 2709.
[6] Q.-H. Park, Deformed coset models from gauged WZW actions, Phys. Lett. B 328 (1994) 329–336.
[7] C.R. Fernandez-Pousa, M.V. Gallas, T.J. Hollowood, J.L. Miramontes, The symmetric space and homoge-
neous sine-Gordon theories, Nucl. Phys. B 484 (1997) 609–630.
[8] A.V. Mikhailov, M.A. Olshanetsky, A.M. Perelomov, Two dimensional generalized Toda lattice, Commun.
Math. Phys. 79 (1981) 473.
[9] D.I. Olive, N. Turok, Local conserved densities and zero curvature conditions for Toda lattice field theories,
Nucl. Phys. B 257 (1985) 277.
[10] T.R. Klassen, E. Melzer, The thermodynamics of purely elastic scattering theories and conformal
perturbation theory, Nucl. Phys. B 350 (1991) 635–689.
[11] V.P. Yurov, A.B. Zamolodchikov, Truncated conformal space approach to scaling Yang–Lee model, Int. J.
Mod. Phys. A 5 (1990) 3221–3246.
[12] C. Destri, H.J. de Vega, New exact results in affine Toda field theories: free energy and wave function
renormalizations, Nucl. Phys. B 358 (1991) 251–294.
[13] A.B. Zamolodchikov, On the thermodynamic Bethe ansatz equations for reflectionless ADE scattering
theories, Phys. Lett. B 253 (1991) 391–394.
[14] A.B. Zamolodchikov, Resonance factorized scattering and roaming trajectories, ENS-LPS-335.
[15] M. Martins, Exact resonance A–D–E S-matrices and their renormalization group trajectories, Nucl. Phys.
B 394 (1993) 339–355.
[16] A. Fring, C. Korff, B.J. Schulz, The ultraviolet behaviour of integrable quantum field theories, affine Toda
field theory, Nucl. Phys. B 549 (1999) 579–612.
[17] C. Ahn, V.A. Fateev, C. Kim, C. Rim, B. Yang, Reflection amplitudes of ADE Toda theories and
thermodynamic Bethe ansatz, Nucl. Phys. B 565 (2000) 611–628.
[18] C. Ahn, P. Baseilhac, V.A. Fateev, C. Kim, C. Rim, Reflection amplitudes in non-simply laced Toda theories
and thermodynamic Bethe ansatz, Phys. Lett. B 481 (2000) 114–124.
[19] V.A. Fateev, The exact relations between the coupling constants and the masses of particles for the integrable
perturbed conformal field theories, Phys. Lett. B 324 (1994) 45–51.
[20] T. Eguchi, S.-K. Yang, Deformations of conformal field theories and soliton equations, Phys. Lett. B 224
(1989) 373.
[21] T.J. Hollowood, P. Mansfield, Rational conformal field theories at, and away from criticality as Toda field
theories, Phys. Lett. B 226 (1989) 73.
[22] M. Niedermeier, The quantum spectrum of the conserved charges in affine Toda theories, Nucl. Phys. B 424
(1994) 184–218.
[23] A. Fring, Braid relations in affine Toda field theory, Int. J. Mod. Phys. A 11 (1996) 1337–1352.
[24] C. Korff, K.A. Seaton, Universal amplitude ratios and Coxeter geometry in the dilute A model, Nucl. Phys.
B 636 (2002) 435–464.
[25] E. Witten, Nonabelian bosonization in two dimensions, Commun. Math. Phys. 92 (1984) 455–472.
[26] J.L. Miramontes, C.R. Fernandez-Pousa, Integrable quantum field theories with unstable particles, Phys.
Lett. B 472 (2000) 392–401.
[27] O.A. Castro-Alvaredo, A. Fring, C. Korff, J.L. Miramontes, Thermodynamic Bethe ansatz of the
homogeneous sine-Gordon models, Nucl. Phys. B 575 (2000) 535–560.
[28] C. Korff, Colours associated to non simply-laced Lie algebras and exact S-matrices, Phys. Lett. B 501 (2001)
289–296.
[29] C.R. Fernandez-Pousa, J.L. Miramontes, Semi-classical spectrum of the homogeneous sine-Gordon theories,
Nucl. Phys. B 518 (1998) 745–769.
[30] H.W. Braden, E. Corrigan, P.E. Dorey, R. Sasaki, Affine Toda field theory and exact S-matrices, Nucl. Phys.
B 338 (1990) 689–746.
[31] V.A. Fateev, Normalization factors, reflection amplitudes and integrable systems, Prog. Math. Phys. 23
(2002) 145.
[32] A.B. Zamolodchikov, Mass scale in the sine-Gordon model and its reductions, Int. J. Mod. Phys. A 10 (1995)
1125–1150.
[33] T. Oota, q-deformed Coxeter element in non-simply laced affine Toda field theories, Nucl. Phys. B 504
(1997) 738–752.
[34] A. Fring, C. Korff, B.J. Schulz, On the universal representation of the scattering matrix of affine Toda field
theory, Nucl. Phys. B 567 (2000) 409–453.
Excitonic instability in layered degenerate

semimetals
D.V. Khveshchenko, H. Leal
Department of Physics and Astronomy, University of North Carolina, Chapel Hill, NC 27599, USA
Received 27 August 2003; accepted 18 March 2004
Abstract
We study excitonic pairing in quasi-two-dimensional degenerate semimetals such as graphite,
where the electron density of states nearly vanishes at the Fermi level and, therefore, the Coulomb
interactions remain unscreened. By focusing on the Dirac-like low-energy electronic excitations and
numerically solving a non-linear gap equation for the excitonic order parameter, we obtain a critical
value of the Coulomb coupling and establish the Kosterlitz–Thouless-like nature of the putative
semimetal-to-excitonic insulator transition which can also be identified as a formal analog of chiral
symmetry breaking in relativistic fermion theories.
The phenomenon of chiral symmetry breaking (CSB) has long been studied in the
context of the three-dimensional quantum electrodynamics (QED2+1 ) and other relativistic
fermion theories. In the case of QED2+1 , a zero temperature CSB transition was predicted
to occur for a sufficiently small (N < Nc ) number of fermion species, regardless of the
interaction strength [1], while a short-ranged attractive Higgs–Yukawa (HY) four-fermion
coupling can drive this transition for any N , provided that its strength is sufficiently
large [2].
Recently, the CSB scenario has also become a common theme of the QED2+1 - and
the HY-like descriptions of such condensed matter problems as the effect of spin [3] and
pairing [4,5] fluctuations in the pseudogap phase as well as the incipient secondary pairing
in the superconducting state [6] of underdoped cuprates and the conjectured charge density
wave instabilities in layered graphite [7,8] and dichalcogenides [9].
E-mail address: khvesh@physics.unc.edu (D.V. Khveshchenko).
doi:10.1016/j.nuclphysb.2004.03.020
324 D.V. Khveshchenko, H. Leal / Nuclear Physics B 687 [FS] (2004) 323–331
Notably, some of the main predictions of the QED2+1 -like theories of Refs. [3–5],
including the very possibility of the phase transitions in question [4,5], were based squarely
on the early estimate Nc(0) ≈ 3.24 for the critical number of fermion species [1].
Considering the fact that the physical number N = 2 of different Dirac-like nodal
quasiparticles in these effective theories is given merely by the number of components
of the electron spin, the seemingly unavoidable onset of CSB in the N = 2 QED2+1
was argued to be consistent with the inherent spin density wave instability of the d-wave
symmetrical pseudogap state.
However, the mounting analytical [10] and numerical [11] evidence indicating that in
the conventional QED2+1 the actual critical number of flavors Nc may be less than two
seems to suggest otherwise and calls for a need to further modify the effective theory of
the pseudogap phase.
The implications of the results obtained in Refs. [10,11] made some groups [5]
emphasize a potential importance of local four-fermion couplings and/or anisotropic
quasiparticle dispersion (in high-Tc cuprates, the quasiparticle velocity is strongly
dependent on the direction, v1 /v2 ∼ 10–20) whose systematic account has yet to be done.
Moreover, it was also argued in Ref. [12] that even in the anisotropic deformation of the
standard QED2+1 the critical number of fermions may not be affected by the dispersion
anisotropy at all, thus implying that the additional four-fermion couplings may indeed play
a crucial role in the theory of the pseudogap-to-antiferromagnet transition in underdoped
cuprates.
By contrast, in layered degenerate semimetals such as highly oriented pyrolytic
graphite (HOPG), a pseudo-relativistic (quasi-two-dimensional) dispersion of the low-
energy electronic excitations and a concomitant poor screening of the Coulomb interactions
gives rise to the effective theory with non-Lorentz-invariant, yet long-ranged, four-fermion
interactions [13].
Because of the non-relativistic nature of the Coulomb interactions, the latter theory
appears to be different from both the QED2+1 and the HY-model, which fact precludes one
from making use of any earlier results [1,2,10,11] pertaining solely to the relativistically
invariant systems.
Nevertheless, it was argued in Refs. [7,8] that a sufficiently strong long-ranged Coulomb
repulsion may set the stage for a novel form of excitonic instability which results in the
opening of a gap in the electronic spectrum and can manifest itself through the onset of an
insulating charge density wave. Such an excitonic ordering could then explain the apparent
disparity in the resistivity behavior (metallic vs. insulating) observed in the HOPG samples
of different electron densities [14].
In this paper, we study the possibility of a zero-field excitonic transition in HOPG and
other degenerate semimetals by numerically solving the excitonic gap (a.k.a. Schwinger–
Dyson) equation.
In the vicinity of two inequivalent (in the absence of a lattice strain, exactly degenerate
and labeled as i = 1, 2) conical K-points of the quasi-two-dimensional Brillouin zone of
graphite, the low-energy electronic excitations of the valence and the conduction bands
with a linear dispersion Ek = ±vF k where vF ≈ 2 × 106 m/s can be described as two-
component (Weyl) spinors ψiσ , of which there are N different species with the spin index
σ = 1, . . . , N [13].
D.V. Khveshchenko, H. Leal / Nuclear Physics B 687 [FS] (2004) 323–331 325
Provided that the Zeeman coupling to an external magnetic field is much weaker than the
orbital (diamagnetic) term, the number of fermion species in graphite is again N = 2, and
the dimensionless Coulomb coupling g = 2πe2 /vF with the dielectric constant ≈ 2.8
appears to be the only relevant parameter. However, in a strong in-plane field, the Zeeman
splitting between the spin-up and spin-down bands, while having no effect on the electron
orbital motion, reduces the number of fermions down to N = 1, which prompts one to treat
N as yet another (to a certain extent) adjustable parameter.
The pair of spinors ψiσ can be further combined into a single bi-spinor Ψσ =
(ψ1σ , ψ2σ ), thereby resulting in the Dirac-like representation for the kinetic energy
N

HK = ivF Ψ̄σ (γ̂1 ∇x + γ̂2 ∇y )Ψσ , (1)
σ =1 r
where Ψ̄σ = Ψσ† γ̂0 and the 4 × 4 (reducible) representation of the γ -matrices γ̂0,1,2 =
(τ3 , iτ2 , −iτ1 ) ⊗ τ3 satisfying the anticommutation relations {γ̂µ , γ̂ν } = 2 diag(1, −1, −1)
is constructed in terms of the Pauli triplet τ1,2,3 .
Accordingly, the Coulomb interaction term in the Hamiltonian takes the form

1
N
g
HC = Ψ̄σ (r)γ̂0 Ψσ (r) Ψ̄σ (r )γ̂0 Ψσ (r ). (2)
4π
|r − r |
σ,σ =1 r,r
Both Eqs. (1) and (2) remain invariant under arbitrary U (2N) rotations of the 2N -
component vector (ΨLσ , ΨRσ ) composed of the chiral (L, R) parts of the Dirac fermion
Ψσ defined as Ψ(L,R)σ = 12 (1 ± γ̂5 )Ψσ where the matrix γ̂5 = 1 ⊗ τ2 anticommutes with
any γ̂µ .
In the relativistic theories of Refs. [1,2], the standard CSB pattern U (2N) → U (N) ⊗
U (N) is signaled by the development of a singlet order parameter
N

∆s (r) = Ψ̄σ (r)Ψσ (r) ,
σ
which is the type of excitonic pairing that we focus upon below.
The self-consistent Schwinger–Dyson equation for the Dirac propagator reads (hereafter
pµ = (, p))
d 2k
Ĝ−1 (ω, p) = Ĝ−1
0 (ω, p) − T V (Ω − ω, k − p)γ̂0 Ĝ(Ω, k)γ̂0 , (3)
(2π)2
Ω
where Ĝ0 (ω, p) = γ̂µ pµ /p2

is the bare propagator of the gapless Dirac fermions and the
renormalized Coulomb interaction
g
V (Ω, q) = (4)
q + Ngχ(Ω, q)
is modified, as compared to its bare form, by to the fermion polarization χ(Ω, q).
Being primarily concerned with the possibility of a spontaneous gapping of the conical
spectrum, we search for a solution to Eq. (3) in the form Ĝ(p) = (γ̂µ pµ + ∆(p))−1 ,
while neglecting both the vertex and the fermion wave function renormalization. On the
basis of the experience gained from the relativistic theories [1,15], one might expect
that, albeit being capable of affecting such practically important details as the value of
the critical coupling gc (N), the latter do not alter the very existence of a solution (or
a lack thereof). Proceeding along the lines of the previous numerical analyses of the
finite temperature QED2+1 [15], below we focus on the momentum dependence of the
static (ω = 0) component of the order parameter. To this end, we neglect all but the
Ω = 0 harmonics of the effective interaction (4) which, as shown in Ref. [15], suffices
for determining the critical conditions for the emergent CSB order. In the case of interest,
the static approximation appears to be even better justified, since the Lorentz invariance is
broken already at zero temperature.
The static component of the finite-temperature fermion polarization can, in turn, be
approximated by the expression

1 T |q|
χ(0, q) = |q| + exp −C , (5)
8 C T
which, for C = π/16 ln 2, provides an up to a few percent accurate interpolation between
the two opposite limits: vF q T where Eq. (5) agrees with the zero-temperature result,
χ(0, q) ∼ |q|, and vF q T where it accounts for thermal screening, χ(0, 0) ∼ T [15].
Taking the sum over the Matsubara frequencies in Eq. (3), we then arrive at our gap
equation (hereafter we use the units v = h̄ = 1)
√
k2 +∆(k)2
d 2k tanh
∆(p) = V (0, k − p)∆(k) 2T
. (6)
8π 2 k2 + ∆(k)2
This form of the gap equation is more familiar in condensed matter-related applications
where the gap is routinely considered to be a function of momentum but not energy,
consistent with our use of the static approximation.
In fact, even a partial account of the momentum and/or energy dependence of the gap
function goes well beyond the customary BCS-like (momentum- and energy-independent)
solution for the gap. In the case of short-ranged repulsive interactions, a BCS-like solution
of the analog of Eq. (6) was recently discussed in conjunction with the conjectured
excitonic instability in hexaborides [16]. By contrast, in a degenerate semimetal such as
graphite the strong momentum dependence of the unscreened Coulomb interaction rules
out a constant solution (∆BCS (p) = ∆) altogether.
In Fig. 1, we present the results of our numerical solution to Eq. (6) for N = 1 and
several different temperatures. As a function of momentum, the gap ∆(p) levels off at
p ∼ ∆(0), in accord with the approximate analytical solution of Refs. [7,8] where Eq. (6)
was substituted with a differential equation complemented by the boundary conditions
d∆(p)/dp|p=0 = 0 and ∆(p) + p d∆(p)/dp|p=Λ = 0, Λ being the upper momentum
cutoff given by the maximum span of the Brillouin zone.
The functional dependence of the solution for N = 2 is similar to the N = 1 case, apart
from the overall suppression by roughly two orders of magnitude (see Fig. 2).
Fig. 1. Momentum dependence of the solution to the gap equation (6) for N = 1 and different temperatures.
Fig. 2. Solution to the gap equation for N = 2.
In the strong coupling zero-temperature (g → ∞ and T → 0) limit, the N -dependence

of the zero-momentum gap shown in Fig. 3 can be well fitted with the formula

∆(0) = 1.2vF Λ exp −1.7π/ Nc∞ /N − 1 , (7)
which manifests the Kosterlitz–Thouless nature of the excitonic transition that occurs at
T = 0 and Nc∞ ≡ Nc (g = ∞) ≈ 2.6, in agreement with the predictions made in Refs. [7,
8].
In Fig. 4, we present the N -dependence of the critical coupling gc (N). For N = 2, our
numerical result gc (N = 2) ≈ 7 differs by a factor of two from that obtained analytically
in Ref. [8], although the agreement improves for smaller N .
Fig. 3. Zero-momentum gap for different values of N at T → 0 and g → ∞. The solid line is the fit to Eq. (7).
Fig. 4. The critical coupling gc for different values of N and T = 0. The solid line is the analytical result of
Ref. [10].
Notably, the relevant values of gc are rather large, which indicates that any weak-
coupling approach would be utterly inadequate for the problem in question.
Even with the possible caveat that the gap equation tends to systematically underesti-
mate the critical strength of the repulsive interactions [10,11], the above values compare
favorably with the estimates g ∼ 5–10 obtained for the HOPG samples of Ref. [14].
For any N , the zero-momentum gap attains its maximum value ∆(0) ≈ 10−4 vF Λ at
g → ∞. Using the parameters of the HOPG band structure, we estimate the maximal
possible gap at N = 2 as ∆(0) ∼ 30 K, which turns out to be rather small compared to
the typical Fermi energy EF ∼ 200 K in HOPG [14]. The latter characterizes the actual (as
opposed to the idealized point-like) Fermi surface of graphite which represents a combined
effect of inter-layer hopping, finite doping, and/or disorder.
At this point, it remains unknown to what extent the above factors can modify Eq. (6)
derived for a clean two-dimensional sheet of undoped graphite. On the other hand, in a
layered system the propensity towards excitonic pairing is further strengthened by the
inter-layer Coulomb repulsion which, considering the large value of g, should be even
more important than the finite EF .
It is worth mentioning that the singlet excitonic order parameter is directly related to
the electron density imbalance between the A and B sublattices of the bi-partite hexag-

†
onal lattice: ∆s (p = 0, r) = σ =1,2 Ψ̄σ (r)Ψσ (r) = iσ =1,2 (δr,A ψiσ (A)ψiσ (A) −
†
δr,B ψiσ (B)ψiσ (B) ).
In a multi-layer system stacked in the staggered (A–B) configuration, the inter-layer
Coulomb repulsion favors spontaneous depletion (respectively, pile-up) of the electron
density on a sublattice formed by the carbon atoms above and below the centers
(respectively, corners) of the hexagons in adjacent layers. The resulting commensurate
charge density wave alternates between the layers, thereby keeping the electrons in the
adjacent layers as far apart as possible and nudging the system closer to the excitonic
instability.
In order to decide on the ultimate outcome of the competition between the frustrating
effect of the extended Fermi surface and the strong (yet, non-singular, unlike Eq. (2)) inter-
layer Coulomb repulsion, the present analysis has to be further refined by incorporating
the above factors through, e.g., a non-linear fermion dispersion and effective four-fermion
terms.
Although, thus far, we were only interested in the possibility of singlet pairing,
for N = 2 and in the leading Coulomb approximation (small transferred momenta) the
formation of a triplet order parameter appears to be just as likely. In fact, similar to
the case of the short-ranged repulsion discussed in Ref. [17], the excitonic ground state
possesses a degeneracy with respect to arbitrary SU(4) rotations of
the four-dimensional
complex vector composed of the singlet ∆s and the triplet ∆t = σ,σ =1,2 Ψ̄σ σσ σ Ψσ
order parameters.
This approximate degeneracy gets lifted upon including the (so far, neglected) short-
ranged Coulomb exchange interactions which cause transitions between the conduction
and valence bands [17]. Alongside the Zeeman coupling, the latter favor the triplet order
parameter, thus enforcing the Hund’s rule.
It is also conceivable that, with increasing doping, the triplet excitonic insulator gives
way to an itinerant ferromagnetic metal. The interest in this scenario which had been
previously discussed only in the case of the three-dimensional non-degenerate semimetals
was bolstered by the discovery of a weak (∼ 0.07µB per carrier), yet robust (Tc ∼ 600–
1000 K), ferromagnetism in hexaborides [16].
However, more recent experimental studies of hexaborides [18] which have indicated
the presence of a large spectral gap which seem to rule out the excitonic mechanism,
thus putting graphite in a rather unique position of the best currently known candidate for
excitonic ferromagnet. This possibility appears to be particularly intriguing in the light of
the reports of a comparably weak (∼ 0.1µB per carrier) magnetization observed in HOPG
samples at room temperatures [19] which, according to the authors of this work, is not
related to magnetic impurities.
It was also argued in Refs. [7,8] that a magnetic field normal to the layers can further
facilitate a formation of excitonic insulator which would then be reminiscent of the
phenomenon of magnetic catalysis in (2 + 1)-dimensional relativistic fermion theories
[20]. The scenario proposed in Refs. [7,8] was also motivated by the recent reports of a
magnetic field-induced insulating behavior in HOPG [21].
To summarize, in the present paper we obtained a numerical solution of the gap equation
describing the conjectured excitonic transition in layered semimetals such as graphite.
From our solution, we inferred the minimal necessary strength of the Coulomb coupling gc
and the largest possible number of fermion species Nc∞ for which this Kosterlitz–Thouless-
type transition may occur at zero temperature. Although this analysis was carried out for a
single undoped layer, we believe that our predictions for the existence of a critical Coulomb
coupling, momentum dependence of the gap function, and the nature of the excitonic
transition should remain robust upon including further complicating factors such as inter-
layer coupling, doping, and disorder.
Acknowledgements
The authors gratefully acknowledge the North Carolina Supercomputing Center for
providing access to their computing facilities. This research was supported by the NSF
Grant DMR-0071362.
References
[1] T. Appelquist, D. Nash, L.C.R. Wijewardhana, Phys. Rev. Lett. 60 (1988) 2575;
T. Appelquist, J. Terning, L.C.R. Wijewardhana, Phys. Rev. Lett. 75 (1995) 2081.
[2] B. Rosenstein, B.J. Warr, S.H. Park, Phys. Rep. 205 (1991) 59;
S. Hands, A. Kocić, J.B. Kogut, Ann. Phys. 224 (1993) 29.
[3] D.H. Kim, P.A. Lee, Ann. Phys. 272 (1999) 133;
W. Rantner, X.-G. Wen, Phys. Rev. Lett. 86 (2001) 3871;
W. Rantner, X.-G. Wen, cond-mat/0105540;
W. Rantner, X.-G. Wen, cond-mat/0201521;
J. Ye, Phys. Rev. B 65 (2002) 214505.
[4] M. Franz, Z. Tesanovic, Phys. Rev. Lett. 87 (2001) 257003;
M. Franz, Z. Tesanovic, Phys. Rev. Lett. 88 (2002) 109902, Erratum;
M. Franz, Z. Tesanovic, O. Vafek, Phys. Rev. B 66 (2002) 054535.
[5] I.F. Herbut, Phys. Rev. Lett. 88 (2002) 047006;
I.F. Herbut, Phys. Rev. B 66 (2002) 094504;
B.H. Seradjeh, I.F. Herbut, Phys. Rev. B 66 (2002) 184507.
[6] D.V. Khveshchenko, J. Paaske, Phys. Rev. Lett. 86 (2001) 4672.
[7] D.V. Khveshchenko, Phys. Rev. Lett. 87 (2001) 206401;
D.V. Khveshchenko, Phys. Rev. Lett. 87 (2001) 246802.
[8] E.V. Gorbar, V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, Phys. Rev. B 66 (2002) 045108.
[9] A.H. Castro-Neto, Phys. Rev. Lett. 86 (2001) 4382;
B. Uchoa, A.H. Castro-Neto, G.G. Cabrera, cond-mat/0307449.
[10] T. Appelquist, A.G. Cohen, M. Schmaltz, Phys. Rev. D 60 (1999) 045003.
[11] S.J. Hands, J.B. Kogut, C.G. Strouthos, Nucl. Phys. B 645 (2002) 321;
S.J. Hands, J.B. Kogut, L. Scorzato, C.G. Strouthos, Nucl. Phys. B (Proc. Suppl.) 119 (2003) 974.
[12] D. Lee, cond-mat/0212204.
[13] G. Semenoff, Phys. Rev. Lett. 53 (1984) 2449;
F.D.M. Haldane, Phys. Rev. Lett. 61 (1988) 2015;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. Lett. 69 (1992) 172;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Nucl. Phys. B 406 (1993) 771;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Nucl. Phys. B 424 (1994) 595;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. Lett. 77 (1996) 3589;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. B 59 (1999) 2474;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. B 63 (2001) 134421.
[14] H. Kempa, P. Esquinazi, Y. Kopelevich, Phys. Rev. B 65 (2002) 241101(R);
M.S. Sercheli, et al., Solid State Commun. 121 (2002) 579;
H. Kempa, et al., Solid State Commun. 125 (2003) 1.
[15] I.J.R. Aitchison, et al., Phys. Lett. B 294 (1992) 91;
N. Dorey, N. Mavromatos, Nucl. Phys. B 386 (1992) 614;
I.J.R. Aitchison, M. Klein-Kreisler, Phys. Rev. D 50 (1994) 1068;
I.J.R. Aitchison, Z. Phys. C 67 (1995) 303;
I.J.R. Aitchison, N. Mavromatos, Nucl. Phys. B (1996) 614;
G. Triantaphyllou, Phys. Rev. D 58 (1998) 065006.
[16] D.P. Young, et al., Nature 397 (1999) 412;
M.E. Zhitomirsky, T.M. Rice, V.I. Anisimov, Nature 402 (1999) 251;
L. Balentz, C.M. Varma, Phys. Rev. Lett. 84 (2000) 1246;
V. Barzykin, L.P. Gorkov, Phys. Rev. Lett. 84 (2000) 2207.
[17] L. Balents, Phys. Rev. B 62 (2000) 2346;
M.Y. Veillette, L. Balents, Phys. Rev. B 65 (2002) 014428;
E. Bascones, A.A. Burkov, A.H. MacDonald, Phys. Rev. Lett. 89 (2002) 086401.
[18] J.D. Denlinger, et al., Surf. Rev. Lett. 9 (2002) 1309;
J.D. Denlinger, et al., Phys. Rev. Lett. 89 (2002) 157601.
[19] Y. Kopelevich, et al., J. Low Temp. Phys. 119 (2000) 691;
P. Esquinazi, et al., Phys. Rev. B 66 (2002) 024429;
Y. Kopelevich, et al., cond-mat/0209442.
[20] V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, Phys. Rev. Lett. 73 (1994) 3499;
V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, Phys. Rev. D 52 (1995) 4718;
V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, Phys. Rev. D 61 (2000) 045005;
J. Alexandre, K. Farakos, G. Koutsoumbas, Phys. Rev. D 62 (2000) 105017;
J. Alexandre, K. Farakos, G. Koutsoumbas, Phys. Rev. D 63 (2001) 065015.
[21] H. Kempa, Y. Kopelevich, F. Mrowka, A. Setzer, J.H.S. Torres, R. Hohne, P. Esquinazi, Solid State
Commun. 115 (2000) 539;
H. Kempa, P. Esquinazi, Y. Kopelevich, Phys. Rev. B 65 (2002) 241101.
CUMULATIVE AUTHOR INDEX B681–B687
Abel, S.A. B682 (2004) 183 Bobisut, F. B686 (2004) 3

Abel, S.A. B685 (2004) 150 Bœhm, C. B683 (2004) 219
Ahn, C. B683 (2004) 177 Böhm, G. B682 (2004) 585
Alishahiha, M. B686 (2004) 53 Bolognesi, S. B686 (2004) 119
Alonso Izquierdo, A. B681 (2004) 163 Bonciani, R. B681 (2004) 261
Anderson, L.B. B686 (2004) 285 Bouchez, J. B686 (2004) 3
Anoka, O.C. B686 (2004) 135 Boyd, S. B686 (2004) 3
Anoka, O.C. B687 (2004) 3 Branco, G.C. B686 (2004) 188
Anselmi, D. B687 (2004) 124 Braun, V.M. B685 (2004) 171
Anselmi, D. B687 (2004) 143 Bueno, A. B686 (2004) 3
Aoki, H. B684 (2004) 162 Bunyatov, S. B686 (2004) 3
Arnaudon, D. B687 (2004) 257
Arnowitt, R. B682 (2004) 347 Caffo, M. B681 (2004) 230
Astier, P. B686 (2004) 3 Camilleri, L. B686 (2004) 3
Attal, R. B684 (2004) 369 Cardini, A. B686 (2004) 3
Carmelo, J.M.P. B683 (2004) 387
Autiero, D. B686 (2004) 3
Castro-Alvaredo, O. B682 (2004) 551
Auzzi, R. B686 (2004) 119
Castro-Alvaredo, O. B687 (2004) 303
Avan, J. B687 (2004) 257
Cattaneo, P.W. B686 (2004) 3
Cavasinni, V. B686 (2004) 3
Babu, K.S. B686 (2004) 135
Cervera-Villanueva, A. B686 (2004) 3
Babu, K.S. B687 (2004) 3
Challis, R.C. B686 (2004) 3
Bajnok, Z. B682 (2004) 585
Chapovsky, A.P. B686 (2004) 205
Baldisseri, A. B686 (2004) 3
Chen, H. B683 (2004) 196
Baldo-Ceolin, M. B686 (2004) 3 Choi, K. B687 (2004) 101
Banner, M. B686 (2004) 3 Chukanov, A. B686 (2004) 3
Bassompierre, G. B686 (2004) 3 Collazuol, G. B686 (2004) 3
Becker, M. B683 (2004) 67 Conforto, G. B686 (2004) 3
Beisert, N. B682 (2004) 487 Constantin, D. B683 (2004) 67
Bellucci, S. B684 (2004) 321 Conta, C. B686 (2004) 3
Beneke, M. B685 (2004) 249 Contalbrigo, M. B686 (2004) 3
Beneke, M. B686 (2004) 205 Cousins, R. B686 (2004) 3
Benslama, K. B686 (2004) 3 Crampé, N. B687 (2004) 257
Besson, N. B686 (2004) 3 Cucchieri, A. B687 (2004) 76
Bianchi, M. B685 (2004) 65 Cudell, J.R. B682 (2004) 391
Bianchi, M. B686 (2004) 261 Czyż, H. B681 (2004) 230
Bird, I. B686 (2004) 3
Bjerrum-Bohr, N.E.J. B684 (2004) 209 Dai, J. B684 (2004) 75
Bloch, J.C.R. B687 (2004) 76 Dall’Agata, G. B682 (2004) 243
Blumenfeld, B. B686 (2004) 3 Daniels, D. B686 (2004) 3
0550-3213/2004 Published by Elsevier B.V.

doi:10.1016/S0550-3213(04)00268-8
334 Nuclear Physics B 687 (2004) 333–337
Datta, A. B681 (2004) 31 Ghinculov, A. B685 (2004) 351

D’Auria, R. B682 (2004) 243 Ghosh, P.K. B681 (2004) 359
de Forcrand, P. B686 (2004) 85 Giannakis, I. B681 (2004) 120
Degaudenzi, H. B686 (2004) 3 Gibin, D. B686 (2004) 3
Delfino, G. B682 (2004) 521 Gili, V. B685 (2004) 3
Del Prete, T. B686 (2004) 3 Giudice, G.F. B685 (2004) 89
Derkachov, S. B681 (2004) 295 Gninenko, S. B686 (2004) 3
De Santo, A. B686 (2004) 3 Godley, A. B686 (2004) 3
Dignan, T. B686 (2004) 3 Gogoladze, I. B686 (2004) 135
Di Lella, L. B686 (2004) 3 Gogoladze, I. B687 (2004) 3
Djouadi, A. B681 (2004) 31 Gomez-Cadenas, J.-J. B686 (2004) 3
do Couto e Silva, E. B686 (2004) 3 González León, M.A. B681 (2004) 163
Doikou, A. B687 (2004) 257 Gosset, J. B686 (2004) 3
Dumarchez, J. B686 (2004) 3 Gößling, C. B686 (2004) 3
Dutta, B. B682 (2004) 347 Gottwald, S. B685 (2004) 171
Gouanère, M. B686 (2004) 3
Eden, B. B681 (2004) 195 Grant, A. B686 (2004) 3
Ellis, M. B686 (2004) 3 Graziani, G. B686 (2004) 3
Engelhardt, M. B685 (2004) 227 Grinza, P. B682 (2004) 521
Evslin, J. B686 (2004) 119 Grzadkowski, B. B686 (2004) 165
Grzelińska, A. B681 (2004) 230
Fayet, P. B683 (2004) 219 Guchait, M. B681 (2004) 31
Feldman, G.J. B686 (2004) 3 Guglielmi, A. B686 (2004) 3
Feldmann, Th. B685 (2004) 249
Feng, H. B683 (2004) 168
Hagner, C. B686 (2004) 3
Ferrari, R. B686 (2004) 3
Hardmeier, A. B682 (2004) 150
Ferrère, D. B686 (2004) 3
Harmark, T. B684 (2004) 183
Ferroglia, A. B681 (2004) 261
Hatsuda, M. B681 (2004) 152
Flaminio, V. B686 (2004) 3
Havare, A. B682 (2004) 457
Forde, D.A. B684 (2004) 125
Heinrich, G. B682 (2004) 265
Frampton, P.H. B687 (2004) 31
Hernando, J. B686 (2004) 3
Frappat, L. B687 (2004) 257
Ho, I.-L. B684 (2004) 281
Fraternali, M. B686 (2004) 3
Hou, H.-S. B683 (2004) 196
Fré, P. B685 (2004) 3
Hu, B. B682 (2004) 347
Frenkel, J. B685 (2004) 393
Hubbard, D. B686 (2004) 3
Freyhult, L. B681 (2004) 65
Hurst, P. B686 (2004) 3
Friedrich, R. B687 (2004) 279
Hurth, T. B685 (2004) 351
Fring, A. B682 (2004) 551
Hyett, N. B686 (2004) 3
Fring, A. B687 (2004) 303
Fukuma, M. B682 (2004) 377
Iacopini, E. B686 (2004) 3
Gaillard, J.-M. B686 (2004) 3 Imai, T. B686 (2004) 248
Gangler, E. B686 (2004) 3 Intriligator, K. B682 (2004) 45
García Fuertes, W. B681 (2004) 163 Isidori, G. B685 (2004) 351
Gardi, E. B685 (2004) 171 Iso, S. B684 (2004) 162
Gargiulo, F. B685 (2004) 3 Itoyama, H. B686 (2004) 155
Gates Jr., S.J. B683 (2004) 67 Itzhaki, N. B684 (2004) 264
Gegelia, J. B682 (2004) 367 Ivanov, E. B684 (2004) 321
Gehrmann, T. B682 (2004) 265
Gehrmann-De Ridder, A. B682 (2004) 265 Jaikumar, P. B683 (2004) 264
Geiser, A. B686 (2004) 3 Janke, W. B682 (2004) 618
Geng, C.Q. B684 (2004) 281 Johnston, D.A. B682 (2004) 618
Geppert, D. B686 (2004) 3 Joseph, C. B686 (2004) 3
Geyer, B. B684 (2004) 351 Juget, F. B686 (2004) 3
Nuclear Physics B 687 (2004) 333–337 335
Kalkkinen, J. B687 (2004) 279 Lupi, A. B686 (2004) 3

Kanno, H. B686 (2004) 155 Lyubushkin, V. B686 (2004) 3
Karakhanyan, D. B681 (2004) 295
Kawai, H. B683 (2004) 27 Ma, W.-G. B683 (2004) 196
Kenna, R. B682 (2004) 618 Marchetti, A. B686 (2004) 261
Kent, N. B686 (2004) 3 Marchionni, A. B686 (2004) 3
Khan, A. B686 (2004) 75 Martelli, F. B686 (2004) 3
Khoze, V.V. B682 (2004) 217 Martynov, E. B682 (2004) 391
Khveshchenko, D.V. B687 (2004) 323 Mastrolia, P. B681 (2004) 261
Kim, I.-W. B687 (2004) 101 Mateos Guilarte, J. B681 (2004) 163
Kirsanov, M. B686 (2004) 3 Mazumdar, A. B683 (2004) 264
Kirschner, R. B681 (2004) 295 McArthur, I.N. B683 (2004) 3
Klebanov, I.R. B684 (2004) 264 Méchain, X. B686 (2004) 3
Klimov, O. B686 (2004) 3 Meessen, P. B684 (2004) 235
Kokkonen, J. B686 (2004) 3 Mendes, T. B687 (2004) 76
Konishi, K. B686 (2004) 119 Mendiburu, J.-P. B686 (2004) 3
Konishi, Y. B682 (2004) 465 Merrell, W. B683 (2004) 67
Kono, Y. B682 (2004) 377 Meyer, J.-P. B686 (2004) 3
Körs, B. B681 (2004) 77 Mezzetto, M. B686 (2004) 3
Korunur, M. B682 (2004) 457 Miller, D.J. B681 (2004) 3
Kostov, I.K. B683 (2004) 309 Mishra, S.R. B686 (2004) 3
Kotikov, A.V. B685 (2004) 405 Misiak, M. B683 (2004) 277
Kovacs, S. B684 (2004) 3 Miwa, A. B682 (2004) 377
Kovzelev, A. B686 (2004) 3 Moorhead, G.F. B686 (2004) 3
Krasnoperov, A. B686 (2004) 3 Moortgat, F. B681 (2004) 31
Kraus, P. B682 (2004) 45 Morita, T. B683 (2004) 27
Krivonos, S. B684 (2004) 321 Mosaffa, A.E. B686 (2004) 53
Kuroki, T. B683 (2004) 27 Movshev, M. B681 (2004) 324
Kuzenko, S.M. B683 (2004) 3 Mück, A. B687 (2004) 55
Kyae, B. B683 (2004) 105 Mülsch, D. B684 (2004) 351
Mussardo, G. B687 (2004) 189
Lacaprara, S. B686 (2004) 3
Lachaud, C. B686 (2004) 3 Nagao, K. B684 (2004) 162
Lakić, B. B686 (2004) 3 Nath, P. B681 (2004) 77
Laliena, V. B683 (2004) 455 Naumov, D. B686 (2004) 3
Langfeld, K. B687 (2004) 76 Necco, S. B683 (2004) 137
Lanza, A. B686 (2004) 3 Nédélec, P. B686 (2004) 3
La Rotonda, L. B686 (2004) 3 Nefedov, Yu. B686 (2004) 3
Larsen, A.L. B686 (2004) 75 Nehme, A. B682 (2004) 289
Laveder, M. B686 (2004) 3 Nevzorov, R. B681 (2004) 3
Leal, H. B687 (2004) 323 Nguyen-Mau, C. B686 (2004) 3
Lechtenfeld, O. B684 (2004) 321 NOMAD Collaboration B686 (2004) 3
Lee, C.-A. B683 (2004) 105 Notari, A. B685 (2004) 89
Letessier-Selvon, A. B686 (2004) 3
Levy, J.-M. B686 (2004) 3 Obers, N.A. B684 (2004) 183
Li, G.-L. B687 (2004) 220 Orestano, D. B686 (2004) 3
Linch III, W.D. B683 (2004) 67 Ortín, T. B684 (2004) 235
Linssen, L. B686 (2004) 3 Ouyang, P. B684 (2004) 264
Lipatov, L.N. B685 (2004) 405 Owen, A.W. B682 (2004) 183
Liu, J.T. B681 (2004) 120
Ljubičić, A. B686 (2004) 3 Pando Zayas, L.A. B682 (2004) 3
Long, J. B686 (2004) 3 Pastore, F. B686 (2004) 3
Lukyanov, S.L. B683 (2004) 423 Peak, L.S. B686 (2004) 3
Lunghi, E. B682 (2004) 150 Penc, K. B683 (2004) 387
336 Nuclear Physics B 687 (2004) 333–337
Pennacchio, E. B686 (2004) 3 Schofield, B.W. B685 (2004) 150

Pessard, H. B686 (2004) 3 Schwarz, A. B681 (2004) 324
Petcov, S.T. B687 (2004) 31 Sconza, A. B686 (2004) 3
Petti, R. B686 (2004) 3 Serban, D. B683 (2004) 309
Phillips, J. B683 (2004) 67 Sevior, M. B686 (2004) 3
Pilaftsis, A. B687 (2004) 55 Shafi, Q. B683 (2004) 105
Pirjol, D. B682 (2004) 150 Sheikh-Jabbari, M.M. B687 (2004) 161
Placci, A. B686 (2004) 3 Shi, K.-J. B687 (2004) 220
Polesello, G. B686 (2004) 3 Shigemori, M. B682 (2004) 45
Pollmann, D. B686 (2004) 3 Shimada, H. B685 (2004) 297
Polyarush, A. B686 (2004) 3 Siegel, W. B681 (2004) 152
Ponsot, B. B683 (2004) 309 Siegel, W. B683 (2004) 168
Popov, B. B686 (2004) 3 Signer, A. B684 (2004) 125
Poulsen, C. B686 (2004) 3 Signer, A. B686 (2004) 205
Profumo, S. B681 (2004) 247 Sillou, D. B686 (2004) 3
Silva-Marcos, J.I. B686 (2004) 188
Quandt, M. B685 (2004) 227 Smith, J. B682 (2004) 421
Sogut, K. B682 (2004) 457
Ragoucy, E. B687 (2004) 257 Soler, F.J.P. B686 (2004) 3
Raidal, M. B685 (2004) 89 Sommovigo, L. B682 (2004) 243
Rastelli, L. B684 (2004) 264 Song, W.Y. B687 (2004) 101
Ravindran, V. B682 (2004) 421 Sonnenschein, J. B682 (2004) 3
Rebelo, M.N. B686 (2004) 188 Sorin, A. B685 (2004) 3
Rebuffi, L. B686 (2004) 3 Sotkov, G. B687 (2004) 189
Reinhardt, H. B685 (2004) 227 Soyez, G. B682 (2004) 391
Remiddi, E. B681 (2004) 230 Sozzi, G. B686 (2004) 3
Remiddi, E. B681 (2004) 261 Splittorff, K. B683 (2004) 467
Ren, H.-C. B681 (2004) 120 Stanev, Y.S. B685 (2004) 65
Rico, J. B686 (2004) 3 Stanishkov, M. B683 (2004) 177
Riemann, P. B686 (2004) 3 Steele, D. B686 (2004) 3
Riotto, A. B685 (2004) 89 Steinhauser, M. B683 (2004) 277
Riva, V. B687 (2004) 189 Stiegler, U. B686 (2004) 3
Roda, C. B686 (2004) 3 Stipčević, M. B686 (2004) 3
Rodejohann, W. B687 (2004) 31 Stolarczyk, Th. B686 (2004) 3
Román, J.M. B683 (2004) 387 Strumia, A. B685 (2004) 89
Rossi, G. B685 (2004) 65 Sun, Y.-B. B683 (2004) 196
Rubbia, A. B686 (2004) 3
Rückl, R. B687 (2004) 55 Takács, G. B682 (2004) 585
Rulik, K. B685 (2004) 3 Takayama, Y. B686 (2004) 248
Ryzhov, A.V. B682 (2004) 45 Tareb-Reyes, M. B686 (2004) 3
Taylor, G.N. B686 (2004) 3
Sadri, D. B687 (2004) 161 Taylor, J.C. B685 (2004) 393
Sagnotti, A. B682 (2004) 83 Tereshchenko, V. B686 (2004) 3
Sakaguchi, M. B681 (2004) 137 Toharia, M. B686 (2004) 165
Sakaguchi, M. B684 (2004) 100 Toropin, A. B686 (2004) 3
Sakai, K. B682 (2004) 465 Touchard, A.-M. B686 (2004) 3
Salvatore, F. B686 (2004) 3 Tovey, S.N. B686 (2004) 3
Sarma, D. B681 (2004) 351 Tran, M.-T. B686 (2004) 3
Sato, T. B682 (2004) 117 Trigiante, M. B685 (2004) 3
Schahmaneche, K. B686 (2004) 3 Tsesmelis, E. B686 (2004) 3
Scherer, S. B682 (2004) 367 Tsulaia, M. B682 (2004) 83
Schindler, M.R. B682 (2004) 367
Schmidt, B. B686 (2004) 3 Ulrichs, J. B686 (2004) 3
Schmidt, T. B686 (2004) 3 Uzawa, K. B683 (2004) 122
Nuclear Physics B 687 (2004) 333–337 337
Vacavant, L. B686 (2004) 3 Wu, Y.-S. B683 (2004) 363

Vafa, C. B682 (2004) 45 Wu, Y.S. B684 (2004) 75
Valdata-Nappi, M. B686 (2004) 3 Wyler, D. B682 (2004) 150
Valuev, V. B686 (2004) 3
Vaman, D. B682 (2004) 3 Yabsley, B.D. B686 (2004) 3
van der Bij, J.J. B681 (2004) 261 Yaguna, C.E. B681 (2004) 247
van Neerven, W.L. B682 (2004) 421 Yamamoto, M. B683 (2004) 177
Vannucci, F. B686 (2004) 3 Yao, Y.-P. B685 (2004) 351
Varvell, K.E. B686 (2004) 3 Yavartanoo, H. B686 (2004) 53
Vaulà, S. B682 (2004) 243 Yee, H.-U. B686 (2004) 31
Veltri, M. B686 (2004) 3 Yetkin, T. B682 (2004) 457
Verbaarschot, J.J.M. B683 (2004) 467
Yi, P. B686 (2004) 31
Vercesi, V. B686 (2004) 3
Yoshida, K. B681 (2004) 137
Vettorazzo, M. B686 (2004) 85
Yoshida, K. B683 (2004) 122
Vidal-Sitjes, G. B686 (2004) 3
Yoshida, K. B684 (2004) 100
Vieira, J.-M. B686 (2004) 3
Yue, C. B683 (2004) 48
Vinogradova, T. B686 (2004) 3
Yue, R.-H. B687 (2004) 220
Vitchev, E.S. B683 (2004) 423
Wang, W. B683 (2004) 48 Zaccone, H. B686 (2004) 3

Wang, X.-J. B683 (2004) 363 Zamolodchikov, A.B. B683 (2004) 423
Weber, F.V. B686 (2004) 3 Zanderighi, G. B686 (2004) 205
Weigert, H. B685 (2004) 321 Zei, R. B686 (2004) 3
Weisse, T. B686 (2004) 3 Zerwas, P.M. B681 (2004) 3
Wheeler, J.T. B686 (2004) 285 Zhang, R.-Y. B683 (2004) 196
Wilson, F.F. B686 (2004) 3 Zhou, P.-J. B683 (2004) 196
Winton, L.J. B686 (2004) 3 Zuber, K. B686 (2004) 3
Wu, T.H. B684 (2004) 281 Zuccon, P. B686 (2004) 3

Nucl - Phys.B v.687 PDF

Uploaded by

Copyright:

Available Formats

Nucl - Phys.B v.687 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.687 PDF

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 687 (2004) 3–30

Constraining Z from supersymmetry breaking

Received 30 January 2004; accepted 10 March 2004

PACS: 12.60.-i; 12.60.Cn; 12.60.Fr; 12.60.Jv; 13.38.Dg; 14.60.St; 14.70.Pw

E-mail addresses: anoka@okstate.edu (O.C. Anoka), babu@okstate.edu (K.S. Babu), gogoladze.1@nd.edu

U (1)x = xY − (B − L). (3)

The scalar potential (involving Hu , Hd , S+ , S− fields) of the model is given by:

We parameterize the VEVs of Hu , Hd , S+ and S− as

mh mA | cos 2ψ|. (25)

in the Lagrangian [12,13].

mνic = fνic z, (36)

4. The SUSY spectrum

4.1. Slepton masses

4.2. Squark masses

The heavy right-handed sneutrinos (ν̃ic ) split into scalar (ν̃is

5. Numerical results for the spectrum

As inputs at MZ we choose the central values (in the MS scheme) [22]

U11 U12 U21 U22 V11 V12 V21 V22

6. Z decay modes and branching ratios

U11 U12 U21 U22 V11 V12 V21 V22

The Z decay rate into a fermion–antifermion pair is then

The Z decay rate into the chargino pair is then

v11 = QHd V12

7. Other experimental signatures

7.1. Z decay and precision electroweak data

{∆u , ∆d , ∆ν } = {0.00100, 0.00171, 0.00206} (Model 1). (101)

The predicted value of MW is modified as

7.2. Z mass limit

Lagrangian for this process is

7.4. Signatures of SUSY particles

We have suggested in this paper a new class of supersymmetric Z models motivated

A.1. Anomalous dimensions

16π 2 γν c = 4fν2c − 2gx2 , (A.9)

A.2. Beta function

The trilinear soft SUSY breaking terms are given by

A.4. Gaugino masses

The soft masses of the gauginos are given by:

A.5. Soft SUSY masses

[25] T. Moroi, L. Randall, Nucl. Phys. B 570 (2000) 455, hep-ph/9906527;

On deviations from bimaximal neutrino mixing

Received 2 February 2004; accepted 16 March 2004

E-mail address: werner@sissa.it (W. Rodejohann).

2. The deviations from bimaximal mixing

3. The case of CP-conservation

(i) all λij 0.35 (“small”);

3.1. Small λij

tan2 θsol 1 − 4|Ue3 | + 8|Ue3 |2 − 8|Ue3 |3 . (9)

3.2. The case of λ23 = 1

If λ23 = 1 and λ12,13 are “small”, λ12,13 0.35, we find:

4 The requisite precision may be achieved in experiments at neutrino factories [23].

3.3. Large λij

tan2 θsol 1 − 4|Ue3 | + 8|Ue3 |2 − 16|Ue3|3 . (16)

3.4. Implied forms of the charged lepton mass matrix

4.1. General considerations

mlep = ei( R − L ) Q†L ŨL† PL† mlep PR ŨR QR .

5 Using a different representation of the unitary matrices U

It determines the magnitude of CP-violation effects in neutrino oscillations [26]. Let us

If S1 = 0 and/or S2 = 0, CP is not conserved due to the Majorana phases β and/or (α − β).

4.2. Small λij

mlep = ei(R −L ) Q†L ŨL† PL† mlep PR ŨR QR .

If S1 = 0 and/or S2 = 0, CP is not conserved due to the Majorana phases β and/or (α − β).

1 m2 1 7 · 10−5 eV2

|m | m| cos 2θsol − i8JCP |. (51)

We find that if the CP-violating phases in Pν satisfy a certain condition, |m | depends in