Nucl - Phys.B v.669 PDF

Nuclear Physics B 669 (2003) 3–56
www.elsevier.com/locate/npe
An orientifold with fluxes and branes via T-duality

Marcus Berg a , Michael Haack a , Boris Körs b,c
a Dipartimento di Fisica, Università di Roma, Tor Vergata, 00133 Rome, Italy
b Spinoza Institute and Institute for Theoretical Physics, Utrecht University, Utrecht, The Netherlands
c Center for Theoretical Physics, Laboratory for Nuclear Science, and Department of Physics,
Massachusetts Institute of Technology, Cambridge, MA, 02139, USA
Received 12 June 2003; accepted 15 July 2003
Abstract
String compactifications with non-Abelian gauge fields localized on D-branes, with background
NSNS and RR 3-form fluxes, and with non-trivial warp factors, can naturally exist within T-dual
versions of type I string theory. We develop a systematic procedure to construct the effective bosonic
Lagrangian of type I T-dualized along a six-torus, including the coupling to gauge multiplets on
D3-branes and the modifications due to 3-form fluxes. Looking for solutions to the ten-dimensional
equations of motion, we find warped products of Minkowski space and Ricci-flat internal manifolds.
Once the warp factor is neglected, the resulting no-scale scalar potential of the effective four-
dimensional theory combines those known for 3-form fluxes and for internal Yang–Mills fields
and stabilizes many of the moduli. We perform an explicit comparison of our expressions to those
obtained from N = 4 gauged supergravity and find agreement. We also comment on the possibility
to include D9-branes with world-volume gauge fluxes in the background with 3-form fluxes.
 2003 Published by Elsevier B.V.
PACS: 11.25.-w; 11.25.Mj; 04.65.+e
1. Introduction
Orientifolds provide a natural framework for string compactifications that can accom-
modate space–time filling D-branes, internal fluxes for the various tensor field strengths
and non-trivial warp factors at the same time.1 Since these are the main ingredients in
many of the recent phenomenological investigations concerning low string scale models
E-mail addresses: berg@roma2.infn.it (M. Berg), haack@roma2.infn.it (M. Haack), kors@lns.mit.edu

(B. Körs).
1 See [1] for a comprehensive introduction to orientifolds.
0550-3213/$ – see front matter  2003 Published by Elsevier B.V.

doi:10.1016/j.nuclphysb.2003.07.008
4 M. Berg et al. / Nuclear Physics B 669 (2003) 3–56
[2,3], moduli stabilization through flux-induced potentials [4–48], Randall–Sundrum-like

warped compactifications [49,50] or any kind of brane world scenarios, they are of cen-
tral interest among the classes of string compactifications relevant for four-dimensional
particle physics. They evade the powerful no-go theorems prohibiting warped compacti-
fications with fluxes to four-dimensional Minkowski space [51–53] within the context of
Kaluza–Klein reduction of supergravity, since in contrast to traditional supergravity theo-
ries they contain objects of negative energy density like orientifold planes. However, type I
models, or their T-dual descriptions usually called type I , are notoriously difficult to treat:
no explicit description of the effective action of type I models has been given.2 The main
purpose of the present paper is to remedy this latter point and provide the effective bosonic
action for a certain theory T-dual to type I string theory, including the coupling of the su-
pergravity fields to the non-Abelian gauge theory sector localized on D3-branes, as well
as the relevant modifications due to background 3-form fluxes. More precisely, we con-
struct the type I model with 3-form fluxes, which in the absence of 3-form fluxes is dual
to type I via six T-dualities. We consider this the simplest version of an orientifold with 3-
form fluxes and D-branes and a starting point for phenomenologically more sophisticated
constructions.
The motivation to combine models with D-branes and orientifold planes (O-planes)
with background 3-form fluxes comes from the fact that the former provide interesting
non-Abelian gauge fields, potentially with chiral charged matter,3 while the latter add to
the scalar potential of the effective theory, such that at least some scalar fields get massive
and decouple. This removes some of the vacuum degeneracy, a necessary step on the way
to realistic string models. The total scalar potential gets a contribution not only from the
ten-dimensional kinetic term of the type IIB 3-form, projected to type I and including
a Chern–Simons (CS) correction, but also from the Dirac–Born–Infeld (DBI) action of
the Yang–Mills (YM) gauge fields supported by the D-branes. Both the 3-form flux and
the gauge fields add positive energy densities and the constraints that follow from the
equations of motion require this energy density to cancel against the negative contribution
of orientifold planes in a Minkowski vacuum.
Within supergravity, scalar potentials can be generated by gauging isometries of the
scalar manifold (for a recent review see [58]). A systematic procedure can be applied
to derive the form of the effective action. The effective actions obtained from string
compactifications with fluxes have so far been understood as corresponding to gaugings of
a particular type, namely Abelian gaugings of Peccei–Quinn (PQ) isometries, i.e., gaugings
of axionic shifts. The N = 4 gauged supergravity describing the type I model at hand
was studied in a series of papers [34,35,38,47,59–61]. Here we derive the bosonic part
of the effective action by explicitly performing the six T-dualities of type I and make
contact with the formulas of [47], that contains the most explicit formulation of the model,
2 But for some information on N = 1 effective four-dimensional type I supergravity in a T-duality invariant
formulation see [54].
3 For the most part of this paper we only consider D3-branes at smooth points of the background, which do
not lead to chiral spectra. Generalizations of the present model would need to include higher-dimensional branes
with non-trivial gauge fluxes, singular geometries or with point-like intersections to support chiral fermionic zero
modes (see, e.g., [55–57]), a possibility which we also comment on in Section 5.
M. Berg et al. / Nuclear Physics B 669 (2003) 3–56 5
thus elucidating the correspondence between the field variables used in the supergravity
literature and those naturally arising in string theory. A deeper understanding of the
relation between string compactifications in non-trivial backgrounds and the gaugings of
the relevant supergravity is certainly worthwhile.
It is important to note that for our present purposes T-duality always manifests itself
as a set of field redefinitions in the action. We are not going to transform a given vacuum
of type I, but instead we transform the action for the dynamical degrees of freedom of
type I string theory into a formulation in terms of the type I fields after six T-dualities. In
performing the six T-dualities we have to assume the internal space has six isometries (i.e.,
it is a six-torus). However, after performing the transformations one can reinterpret the
resulting four-dimensional action as coming from a non-covariant ten-dimensional action
of the type I theory. In the ten-dimensional Lagrangian, the closed string fields are allowed
to depend also on all the internal coordinates, while the open string fields only depend on
world-volume coordinates (compare the discussion in [62]).
In deriving the full ten-dimensional action and the correct equations of motion, we have
to make use of an important difference between type I theory and its T-dual version type I .
This difference permits us, for the type I theory, to include complex 3-form fluxes and
other non-dynamical backgrounds even though their fluctuations are projected out. This
comes about by noting that the world sheet parity Ω which is “divided out” in getting from
type IIB to type I string theory is mapped to ΩΘ(−1)FL under T-duality. Here Θ is the
reflection of all internal directions x i and FL the left-moving space–time fermion number
operator. A field of type IIB that was odd under Ω has to vanish at any point in type I
theory, whereas a field odd (even) under ΩΘ(−1)FL is only required to be anti-symmetric
(symmetric) under x i → −x i in type I theory. The zero-mode fluctuations of the bosonic
fields are constant on a torus and therefore have to be even under ΩΘ(−1)FL . On the
other hand, one can keep all kinds of background terms, which naively do not appear in the
action for the fluctuations obtained from the type I action by T-duality. Thus, even though
the fluctuations of the internal components of the two anti-symmetric 2-tensors of IIB are
projected out, one can include a background flux for the two 3-form field strengths (which
are even under ΩΘ(−1)FL , such that their corresponding potentials are odd), generating
a potential for the closed string moduli. This is the reason why one is interested in the
T-dual version of type I in the first place. In addition also the odd component of the field
strength F5 with one or five internal indices can take on a background value as long as it
is anti-symmetric under x i → −x i . This is important to find explicit solutions to the ten-
dimensional equations of motion as a non-trivial profile for C4 is needed for D3-branes or
O3-planes to exist.
In order to derive the action of type I including the non-dynamical background terms
in the sense explained above, we combine two different strategies. First we employ six
T-dualities of the type I action. This produces an effective action equivalent to the type I
action and therefore including all the consistent couplings of bulk supergravity and non-
Abelian gauge fields. However, as we already mentioned, this requires the fields to be
independent of the internal coordinates. In particular we cannot include any torsion for the
metric and therefore it is impossible to directly derive the effects of NSNS 3-form flux
in the T-dual theory this way.4 We then add in modifications due to fluxes for the NSNS
3-form and F5 . The fact that this adding in by hand is not arbitrary and can be performed
in a well-defined and systematic manner rests on the twofold definition of type I :
Type IIB
T-duality of type I ←→ . (1)
ΩΘ(−1)FL
This means the closed string sector of the theory can be deduced from the fact that it is
given by projecting the effective action of type IIB by the T-dual world sheet parity. This
second definition does not give any prescription how to incorporate the gauge fields, but
combined with the knowledge of the T-dual action of type I, it provides enough additional
information to include the closed string background fields. Thus, the T-duality of type I
and the T-dual projection of type IIB are complementary regarding terms involving the
open string fields and the non-dynamical background fields such as the NSNS 3-form flux,
respectively. The resulting action then serves as a four-dimensional effective action for a
comparison to the Lagrangian of the gauged N = 4 supergravity as mentioned already.
Reinterpreted as ten-dimensional, we can use it to study more general vacua than torus
compactifications as well.
The solutions for the ten-dimensional equations of motion generalize the situation
without gauge fields considered, e.g., in [21,51]. As in that case, one can find an explicit
solution in the form of a warped product of four-dimensional Minkowski space and a
Ricci-flat internal metric involving also a non-trivial profile for the components of F5
that respect four-dimensional Poincaré invariance. The appearance of the warp factor has
several implications. On the one hand, warp factors have been argued [21,36,49] to be
able to generate exponential hierarchies between the effective energy scales at different
locations on the internal space, such that gauge fields localized on different D-branes may
experience suppression or enhancement of gravitational and gauge-theoretical effects. On
the other hand, the appearance of the warp factor implies that our actual starting point,
a direct product R4 × T6 , no longer solves the ten-dimensional equations of motion. Thus,
in principle, the warp factor should be taken into account in a dimensional reduction.
However, in the large volume limit the warp factor scales to a constant away from singular
sources, and it has been argued that one may completely neglect it in this limit [21]. At
the classical level, the overall volume is a free parameter, as is manifest from the no-scale
structure of the effective potential (see [71,72] for its phenomenological implications).
Thus one can choose an arbitrarily large value and consider the direct-product ansatz as
an approximate solution. However, as soon as quantum corrections to the effective four-
dimensional action are taken into account, this line of reasoning should be modified, as they
seem to spoil the no-scale structure [32,43]. The same is true if one modifies the model by
including higher-dimensional branes [45,46]. These are in principle capable of fixing the
overall volume, but in any attempt to do so one has to make sure that this is done at a
value sufficiently large that one can still neglect the warp factor, if the four-dimensional
effective action derived via dimensional reduction is to remain reliable. In this context we
4 T-duality in the presence of NSNS flux and the corresponding non-trivial metric configurations have been
discussed in [63–66] (see also [42,48,67–70] for analogous works on the non-Kähler vacua of the heterotic string).
However, there T-duality is not performed along all the internal coordinates.
also include some comments on the perspectives of fixing the overall volume by adding
world-volume gauge fluxes on calibrated D9-branes. It appears that a fixing of the volume
at the string scale would be difficult to avoid, which could invalidate the effective approach.
The paper is organized as follows. In Section 2 we develop a systematic procedure to
apply T-duality to type I string theory on the level of the effective Lagrangian including
the coupling to gauge fields, and show how to add modifications due to 3-form fluxes in
the background. In Section 3 we discuss the effective potential that arises through this
procedure in some detail, its relation to the formulas known from gauged supergravity,
and the truncation to N = 1 supersymmetry. In Section 4 the equations of motions are
derived and various forms of constraints produced. We also discuss briefly how the four-
dimensional effective action of the previous sections can be justified by the large volume
scaling argument. In Section 5 we finally add a couple of comments on the addition of
higher-dimensional D-branes subject to certain calibration conditions. Appendices A–C
collect some technical material and additions to the main body of the paper.
Before getting started, let us mention the following caveat. Unfortunately, there is no
standard definition of the Hodge-star in the literature about the class of models under
consideration. Our definition, given in Appendix A.1, differs from the one used, e.g.,
in [21,27,73] but coincides with the one used in [34,35,38,47]. Thus, what is called an
imaginary self-dual (ISD) 3-form flux in [21,27,73] is imaginary anti-self-dual (IASD) in
our conventions and vice versa.
2. Construction of the action via T-duality
In this section we perform six T-dualities of type I theory. This operation defines the
effective action of type I on the dual six-torus. The main objective, then, is to find the
modifications in the action when complex 3-form fluxes are added to the background,
following the philosophy outlined in the introduction. We first go through the procedure
ignoring the coupling to the non-Abelian vector fields of the open string sector, and only
then reconsider the full model.
Our starting point is type I supergravity, which, including the coupling to non-Abelian
vectors of the gauge group SO(32), is described by the action

1 √ 1
SI = 2 d 10 x −g e−2Φ R + 4∂µ Φ∂ µ Φ − |F 3 |2
2κ10 2

1 √
− 2 d 10 x −g e−Φ tr |F |2 , (2)
2g10
where F = 12 FMN
a
T a dx M ∧ dx N is given by
FMN
a
= 2∂[M AaN] + f abc AbM AcN (3)
and we use the definition
2
κ10
3 = dC2 −
F ω3 (4)
2
g10
with ω3 the gauge CS 3-form

2
ω3 = tr A ∧ dA − i A ∧ A ∧ A . (5)
3
In general one defines the Chern–Simons forms ω2j −1 by
dω2j −1 = tr F j . (6)
Except for all the terms involving the vector fields, the type I action (2) is obtained by
quotienting world sheet parity Ω out of the type IIB action5

1 √ 1
SIIB = 2 d 10x −g e−2Φ R + 4∂µ Φ∂ µ Φ − |H3 |2
2κ10 2

1 1 1
− |F1 |2 + |F3 |2 + |F5 |2 − 2 C4 ∧ dB2 ∧ dC2 , (7)
2 2 4κ10
with
F3 = dC2 + C0 dB2 ,
1 1
F5 = dC4 + C2 ∧ dB2 − B2 ∧ dC2 (8)
2 2
and H3 = dB2 , F1 = dC0 . After modding out Ω, only the RR 2-form enters (2) via
F3 = dC2 , while the NSNS 2-form B2 is projected out. Similarly, the T-dual version is
a truncation of the type IIB theory (modding out the T-dual Ω-projection (1)) coupled to
vectors. The duality operation then replaces the degrees of freedom {gI J , C2 , Φ} of type I
by those of type I , which are
1 graviton: gµν ,
12 gauge bosons: Biµ , Ciµ ,
38 scalars: gij , Cij kl , τ = C0 + ieΦ . (9)
In terms of N = 4 supersymmetry, these make up a spin-2 and six Abelian vector
multiplets. In addition, there are vector multiplets with bosonic field content (Aaµ , Aai )
in the adjoint of the gauge group, a subgroup of SO(32). These fields are referred to as
open string fields and arise from the presence of D3-branes, the T-dual images of the D9-
branes of type I. The O9-plane of type I theory splits into 64 O3-planes, but in the presence
of fluxes the number ND3 of D3-branes needed to cancel their charge is no longer fixed.
Rather, turning on 3-form flux modifies the tadpole condition to [21]
1
Nflux + ND3 = 16, (10)
2
thus effectively replacing some of the D3-branes by flux. The precise form of Nflux will be
given later, cf. (91).
5 The issue of the self-duality constraint on F will be dealt with later.

5
2.1. T-duality rules
For some of the conventions used in the following we refer the reader to the appendix.
Using the standard Kaluza–Klein (KK) ansatz for the metric

2
ds10 = gI J dx I dx J = gµν dx µ dx ν + Gij dy i + Aiµ dx µ dy j + Ajν dx ν , (11)
the transformations of the NSNS sector can be deduced from, e.g., [74]. The formulas for
the replacements according to dualizing all six circles simplify for type I to
Gij → Gij , gµν → gµν , (12)
and
gµk → Gki Bµi . (13)
This last operation (13) amounts to replacing the KK vectors by Aiµ → Bµi , since
gµk = Gik Aiµ . Finally, the dilaton transforms according to
e2Φ
e2Φ → . (14)
det(Gij )
Note that we do not distinguish the fields of type I or its T-dual from their ancestor fields
in type IIB, since we identify the effective action that is obtained by projection from type
IIB with that of type I (respectively its T-dual) when the open string vector fields are set to
zero.
The transformation properties of the RR fields on a higher-dimensional torus can be
found, e.g., in [63,75,76]. We will stick to the formalism of [75] here, which leads to the
same results as the one of [63,76]. It was shown there that the type IIB equations of motion
for the action (7) can alternatively be derived from an action that is manifestly invariant
under T-duality, where the second line of (7), the kinetic terms for the RR forms plus the
CS term, is replaced by6
1
5
SRR+CS → − 2
|F2q−1 |2 , (15)
8κ10 q=1
and Fp is now defined more generally as

5
4
F2q−1 = e−B2 ∧ dD2q . (16)
q=1 q=0
It is important that no CS term needs to be included explicitly in (15). It would appear
automatically if one dualizes the forms of high degree, see [75] for more details. The Dp
transform in the spinor representation of O(6, 6, Z) [63,75,76]. To complete the definitions
note
1 1
D0 = C0 , D2 = C2 + C0 B2 , D4 = C4 + B2 ∧ C2 + C0 B22 , (17)
2 2
6 A similar democratic version of type II supergravity was discussed in [77].

and
∗F1 = F9 , ∗F3 = −F7 , ∗F5 = F5 , (18)
the latter to be imposed after deriving the equations of motions from the action. One can
actually give a more explicit definition for the field strengths of higher degree and we will
use this later in (89), see, e.g., [77].
We are now interested only in the particular element of the whole O(6, 6, Z) symmetry
group that corresponds to (12). This element is given by7
(−1)n(n−1)/2
Dµ1 ...µq i1 ...in → /̂i1 ...in in+1 ...i6 Dµ1 ...µq in+1 ...i6 , (19)
(6 − n)!
where the hat on /̂ means that here the epsilon-symbol takes values /̂ 1...6 = /̂1...6 = 1 and
the indices are contracted with Kronecker-deltas instead of the metric. To write (19) in
terms of tensor quantities, one has to factor out the determinant G = det(Gij ) from the
internal metric to cancel the factor G appearing in the contractions, that is
√
(−1)n(n−1)/2 G i1 ...i6
Dµ1 ...µq i1 ...in → / Dµ1 ...µq in+1 ...i6 . (20)
(6 − n)!
√ √
Now /1...6 = G, / 1...6 = 1/ G and indices are contracted with Gij . The field strengths
Fp can be transformed in a similar vein, by simply replacing Dp with dDp in (20) [63].
Thus the components of the RR 3-form field strength have the following transformation
properties8
√
{0,3} G ij klmn {0,3}
(dC2 )ij k → − / (dD2 )lmn ,
3!
√
{1,2} G ij klmn {1,4}
(dC2 )µij → − / (dD4 )µklmn ,
4!
√
{2,1} G ij klmn {2,5}
(dC2 )µνi → + / (dD6 )µνj klmn ,
5!
√
{3,0} G ij klmn {3,6}
(dC2 )µνρ → + / (dD8 )µνρij klmn . (21)
6!
It is evident that the T-dual of the type I RR 2-form also includes RR forms of degree six
and eight, which we later dualize to forms of lower degree. Finally, according to [62] the
internal components of the non-Abelian vectors are mapped via
Aai → Aai , (22)
which is easily applied to F and ω3 . It is important to note that the Aai are actually the
independent dual open string moduli, in other words, the dual scalars Aai depend implicitly
on the metric. The same is then true, e.g., for Fij and ωij k .
7 Note that the overall sign is arbitrary. The derivation of the transformation rules from the rules given in [75]
is performed in the appendix (but see also [76]).
8 Once more, we would like to refer to the appendix for the definitions of the notations used here. The upper
indices {p, q} refer to the bi-degree as forms on the non-compact and internal part of the ten-dimensional space–
time.
2.2. T-duality without vectors
We now apply the rules assembled in the previous section, first to the pure supergravity
part of the type I bosonic action. Our starting point is (2) with the vectors erased,

1 √ 1
SI [A = 0] = 2
d 10 x −g e−2Φ R + 4∂µ Φ∂ µ Φ − |F3 |2 , (23)
2κ10 2
which we split into
SI [A = 0] = SNSNS + SRR . (24)
Let us consider the NSNS part first. Using the rules of the last section and the usual form
of the NSNS action on a torus (see, e.g., (2.17) of [78]), it is straightforward to verify that
it is mapped as

1 √ −2Φ 1 ij µν
SNSNS → 2 d x −g e
10
R + 4∂µ Φ∂ Φ − G Hiµν Hj ,
µ
(25)
2κ10 4
where we defined
Hj νσ = 2∂[ν Bσ ]j . (26)
Note that after T-duality the metric in (9) has no off-diagonal components and that the
original Kaluza–Klein vectors have been mapped to components of the NSNS B-field (26).
Let us now consider the RR part. There is a helpful trick to make the appearance of the
KK vectors explicit [75]. Using (11) one can rewrite the kinetic terms for the RR forms as
|Fp |2 = |Fp |2Gij ,gµν (27)
by redefining
Fp = Fp |dy i →dy i −Aiµ dx µ . (28)
On the left-hand side of (27) the contractions are performed with the full metric, whereas
on the right-hand side, the off-diagonal part is omitted and absorbed into a redefinition
of the field strength. In other words, defining new forms as in (28) one can perform all
contractions using the internal or external components of the metric only. In the dual theory
Fp = Fp and we can omit the prime. Using (21), we obtain
√
{n,3−n} G i1 ...i6 {n,3+n}
(F3 )µ1 ...µn i1 ...i3−n → (−1) (3−n)(2−n)/2
/ (F3+2n )µ1 ...µn i4−n ...i6 . (29)
(3 + n)!
The right-hand side exactly takes the form of (16) subject to the projection in the T-dual
model. This is because the KK vectors of the metric, that appear in the definition (28) of
F3 , are mapped to those of the NSNS B-field exactly in a way required by (16). Applying
(29) in (23) maps the RR kinetic term according to

1 √
− 2 d 10 x −g |F3 |2
4κ10

1 {3,6} {3,6} {2,5} {2,5}
→ 2 F9 ∧ ∗F9 + F7 ∧ ∗F7
4κ10
{1,4} {1,4} {0,3} {0,3}
+ F5 ∧ ∗F5 + F3 ∧ ∗F3 , (30)
where the Hodge star ∗ is with respect to the ten-dimensional T-dual metric. In applying
the T-duality rules of Section 2.1 along all six internal directions, we have to assume that
none of the fields depend on the internal coordinates. This is in particular true for the
metric components (11). Thus we do not consider any non-trivial spin connection for the
metric before T-duality. From the results of [63–66] this implies that we are not able to
produce any non-trivial NSNS flux directly via T-duality and in (30) we implicitly assume
(dB2 ){0,3} = 0.
In principle (30) gives all the terms in the T-dual theory that come from the RR 3-form
{0,3}
field strength of type I, including a purely internal part F3 . In order to make contact
with the standard form that would be obtained from type IIB by the T-dual projection,
which involves only field strengths of degree five and lower, we next remove the RR forms
of unconventionally high degree from the action in favor of their dual forms. According to
the standard procedure (see, e.g., [79]) this would amount to imposing the Bianchi identity
{3,6} {2,5} {3,6}
for F9 and F7 , respectively, via Lagrange multipliers and then integrating out F9
{2,5}
and F7 , leaving the Lagrange multipliers as the dual degrees of freedom. However,
{0,3}
in the presence of 3-form flux F3 this method does not seem to be applicable in a
9
straightforward way. We therefore follow a different strategy, first setting also the RR 3-
{3,6} {2,5}
form flux in (30) to zero. For this case we perform the dualization of F9 and F7 and
only afterwards infer the effects of non-vanishing flux in the effective action.
{3,6}
So let us for the moment consider (30) without the 3-form flux. In order to dualize F9
we impose its Bianchi identity by adding a Lagrange multiplier term

1 {1,0} {3,6} {1,1} {2,5} 1 {1,1} 2 {1,4}
δS = − 2 C0 d F9 + B2 ∧ (dD6 ) − B2 ∧ (dD4 ) .
2κ10 2
(31)
We called the Lagrange multiplier C0 , anticipating that it will be identified with the RR
scalar in a moment. This becomes clear by inspection of its kinetic term (34) below and by
comparison with the (truncated) type IIB action. A partial integration in the first term of
9 The problem derives from the fact that if one naively imposes the Bianchi identities of F {3,6} and F {2,5} one
9 7
{4,3}
generates terms involving different components of the RR field strengths as appearing in (30), like, e.g., F7 .
Their proper treatment requires one to use a democratic version of type I supergravity, involving C2 and C6 on
{4,3}
the same footing before T-duality, cf. Appendix C. After T-duality this would also include kinetic terms for F7
etc., which make it possible to properly dualize these components.
(31) leads to

1 {3,6} 1 {2,5}
δS = 2
(dC0 ){1,0} ∧ F9 − 2
C0 (dB2 ){2,1} ∧ F7 . (32)
2κ10 2κ10
{3,6}
Now integrating out F9 through
{3,6}
∗F9 = (dC0 ){1,0} (33)
and plugging this back into the action produces a kinetic term for C0 :

1 1 10 √
{1,0} 2
(dC 0 ) {1,0}
∧ ∗(dC0 ) {1,0}
= − d x −g F . (34)
2 2 1
4κ10 4κ10
{2,5}
If we next want to integrate out F7 we impose its Bianchi identity by adding a Lagrange
multiplier term

1 {1,1} {2,5} {1,1} {1,4}
δS = 2 C2 ∧ d {1,0} F7 + B2 ∧ dD4 . (35)
2κ10
Again we anticipate that the Lagrange multiplier is identified with the RR 2-form, as can
{1,1}
be read off from (38). Then it is also clear that the only relevant component of C2 is C2 .
Performing a partial integration for the first term of (35) leads to

1 {2,5}
δS = − 2 (dC2 ){2,1} ∧ F7
2κ10

1
+ 2 (C2 ){1,1} ∧ (dB2 ){2,1} ∧ (dC4 ){1,4} . (36)
2κ10
{2,5}
Together with (32), integrating out F7 now gives
{2,5}
∗F7 = −(dC2 ){2,1} − C0 (dB2 ){2,1} . (37)
Inserting this into the action we obtain the kinetic term

1
2
(dC2 ){2,1} + C0 (dB2 ){2,1} ∧ ∗ (dC2 ){2,1} + C0 (dB2 ){2,1}
4κ10

1 √ {2,1} 2
=− 2 d 10 x −g F3 , (38)
4κ10
and a Chern–Simons term

1 {0,4}
SCS = 2 (dC2 ){1,1} ∧ (dB2 ){2,1} ∧ C4 . (39)
2κ10
In this way, rewriting the T-dual of the type I action in terms of potentials with degree up to
four produces the correct CS term that one expects from truncating type IIB, even though
type I does not possess such a term on its own. In all, this action is exactly what one would
get from a reduction of type IIB subject to imposing the self-duality constraint on F5 , as
we verify in Appendix B.
Let us now study what changes should occur due to inclusion of 3-form flux. As was
explained in the introduction, we make use of the fact that type I theory with all non-
Abelian vector fields set to zero is a truncation of type IIB by modding out the T-dual
projection (1). It is obvious from the last term of (30) that a scalar potential appears in the
presence of flux. From (30) we would infer a term10

1 √ {0,3} 2
− 2 d 10 x −g F3 , (40)
4κ10
{0,3}
where, as mentioned below (30), only a F3 = (dC2 ){0,3} term arises directly from T-
duality. However, comparing to the type IIB action truncated by (1), the expression (40)
for the potential receives an additional contribution from the NSNS sector of the type IIB
action, i.e., from the term

1 √ {0,3} 2
− 2 d 10 x −g e−2Φ H3 , (41)
4κ10
and another modification due to the fact that the RR 3-form field strength actually
appears in the combination (8). Then the total potential can be expressed via the complex
combination
G3 = dC2 + τ dB2 = F3 + ie−Φ H3 , (42)
where τ = C0 + ie−Φ , in the form

1 √ {0,3} 2
Spot = − 2
d 10 x −g G3 . (43)
4κ10
Coupling the theory to the D3-branes will, besides other modifications, give rise to further
contributions to the potential, coming from the world-volume scalars and from the tension
of the branes. From the point of view of the truncated type IIB theory, (43) is the obvious
part of the potential, while from the point of view of type I it is the other way around
{0,2}
and (43) cannot be derived directly via T-duality. Since the internal components C2
{0,2} {0,3}
and B2 are projected out of the spectrum, G3 is not a flux for any (dynamical)
field strength in type I , but just some anti-symmetric background parameter. It is then
not obvious that a potential term (43) can appear as part of a consistent modification of the
Lagrangian that allows supersymmetry to be preserved on the level of the action. It can be
deduced from a comparison with type IIB, but the systematic approach to determine such
potentials is through gauged supergravity [34,38,47].
{0,4}
Furthermore, the kinetic term for the scalars C4 , i.e., the penultimate term in (30),
is modified in the presence of fluxes. Again, via T-duality one can only infer a correction
{1,4} {1,1}
F5 = (dC4 ){1,4} − 12 B2 ∧ (dC2 ){0,3} , but comparison with type IIB (8) shows that
actually the combination
{1,4} 1 {1,1} 1 {1,1}
F5 = (dC4 ){1,4} − B2 ∧ (dC2 ){0,3} + C2 ∧ (dB2 ){0,3} (44)
2 2
10 Calling this term a potential is slightly imprecise, as (40) is still a term in the T-dual ten-dimensional action.
What we mean is of course that this term leads to a potential in the effective theory—after dimensional reduction.
{3,6} {2,5}
appears. Finally, when dualizing F9 and F7 one would in principle have to adapt
their Bianchi identities in (31) and (35). Fortunately, this would just influence the resulting
{1,1} {1,1}
Chern–Simons term and not the kinetic terms for C0 , C2 and B2 . Hence, the kinetic
terms in the RR sector after T-duality are

1 10 √
2 2
Skin = − 2 d x −g (dC0 ){1,0} + (dC2 ){2,1} + C0 (dB2 ){2,1}
4κ10
2
1 {1,1} 1 {1,1}

+ (dC4 ) {1,4}
− B2 ∧ (dC2 ) {0,3}
+ C2 ∧ (dB2 ) {0,3}
2 2

1 √ {1,0} 2 {2,1} 2 {1,4} 2
=− 2 d 10 x −g F1 + F3 + F5 . (45)
4κ10
{1,1} {1,1}
We see that under a gauge transformation of the Kaluza–Klein vectors C2 and B2 , the
{0,4}
scalars C4 have to transform in order to render their kinetic term gauge invariant. This
complication can be turned to our advantage in that it reveals the necessary modification
of the CS terms as in [59]. The Chern–Simons term must be modified to

{0,4}
2
2κ10 SCS = (dC2 ){2,1} ∧ (dB2 ){2,1} ∧ C4

1 {1,1} {1,1}
− C2 ∧ (dB2 ){2,1} ∧ B2 ∧ (dC2 ){0,3}
2

1 {1,1} {1,1}
− B2 ∧ (dC2 ){2,1} ∧ C2 ∧ (dB2 ){0,3} . (46)
2
In summary, the RR part of the action (without vectors) is of the form
SRR = Spot + Skin + SCS , (47)
where Spot , Skin and SCS are given by (43), (45) and (46).
To illustrate the notation let us write out the covariant derivative for the axions
{1,1}
descending from C4 . Since the only dynamical components of B2 and C2 are B2 and
{1,1}
C2 , the last term in (45) is proportional to the square of
∂µ (C4 )ij kl − 2(B2 )µ[i (dC2 )j kl] + 2(C2 )µ[i (dB2 )j kl] . (48)
In order to make contact with the standard conventions in supergravity, as used in [28,34,
47] in particular, let us introduce a different parameterization of the axionic scalars
1 1
∂µ (C4 )ij kl = √ /ij klmn ∂µ β mn ,
2 G
1
(B2 )µ[i (dC2 )j kl] = − /ij klmn (5dC2 )mnp (B2 )µp , (49)
8
and similarly for the third term, with 5 denoting the six-dimensional internal Hodge
operator. This leads to a covariant derivative
1√ 1√
Dµ β ij = ∂µ β ij + G (5dC2 )ij k (B2 )µk − G (5dB2 )ij k (C2 )µk (50)
2 2
√ √
for β ij . Note that the combinations G (5dC2 )ij k and G (5dB2 )ij k are actually constant,
independent of the metric, for constant background fluxes (dC2 )ij k and (dB2 )ij k . They
correspond to the flux parameters fαΛΣΓ of [34], while the β ij serve as axionic moduli
also independent of the metric. Later we will come back to the precise relation between
our notation and the one used in [34,47]. In the light of (21), the reparameterization (49)
just reintroduces the T-dual variables, which reflects the fact that the dependence of mass
parameters on the radii of the torus was found to be inverted in [34].
2.3. Coupling to vector fields
In the previous section we have performed the T-duality of the type I action without
vector fields, reorganized the RR forms to be able to compare to the truncated type IIB
action, and then deduced the modification due to 3-form fluxes. We now want to add in
the CS correction appearing in (4) and the Yang–Mills action for the vectors. Thus, our
starting point is now (2). In order to T-dualize (2) we need to make the appearance of the
KK vectors in the terms involving F and ω3 explicit. Let us take F first. On a torus we
have
Fµν
a
= 2∂[µ Aaν] + f abc Abµ Acν ,
Fµi
a
= Dµ Aai = ∂µ Aai + f abc Abµ Aci ,
Fija = f abc Abi Acj . (51)
Our notation does not distinguish scalars Ai from vector fields Aµ and we do not use a
background gauge flux fija . This is related to the fact that we intend to stick to D3-branes
after the T-duality. We shall come back to this point later. As in (28) we introduce the
components of F as
F µν = Fµν
a j
a
+ 2Ai[µ Fν]i
a
+ Fija Ai[µ Aν] ,
F µi = Fµi
a a
+ Fija Ajµ ,
F ij = Fija ,
a
(52)
where FMNa
is defined as in (3). Using new vector fields, invariant under KK gauge
transformations
Ãaµ = Aaµ − Aai Aiµ , (53)

it is straightforward to verify the relations
µ Aai ,
F µi = D
a
µν
F µν = F
a a
+ 2∂[µ Aiν] Aai , (54)
where we used
µ Aa = ∂µ Aa + f abc Ãbµ Ac ,
D i i i
µν
F a
= 2∂[µ Ãaν] + f abc Ãbµ Ãcν . (55)
Thus, using (22), the kinetic term for the vectors is mapped as follows under T-duality:

√
d 10 x −g e−Φ tr |F |2

√
→ d 10 x −g4 e−Φ Gij g µν D µ Aai D
ν Aaj
1 a a
µρ + Hiµρ Aai F
+ g µν g ρσ F νσ + Hj νσ Aaj
2
1
+ Gij Gkl f abc f ade Abi Ack Adj Ael , (56)
2
where now
Ãaµ = Aaµ − Aai Bµi , (57)
and Hiµν is given by (26). The term in the last line of (56) represents a contribution to the
potential for the open string fields and can also be written as tr|F {0,2} |2 up to a constant.
Notice that (56) is separately invariant under the gauge transformations
Bµi → Bµi + ∂µ /i , (58)
with Ãaµ and Aai inert, and
Ãaµ → Ãaµ + ∂µ / a + f abc Ãbµ / c , Aai → Aai + f abc Abi / c , (59)

with Bµi inert.
Let us now turn to the mapping of ω3 under the duality transformation, shifting to ω3
as before. It is given by
{0,3} 1 ij klmn {0,3}
(ω3 )ij k → / (5ω3 )lmn ,
3!
{1,2} 1 {1,4} 1 {1,1} {0,3}
(ω3 )µij → / ij klmn (5ω3 )µklmn − (B2 )µk (5ω3 )lmn ,
4! 3!

1 {2,1} 11 {2,5} 1 {1,1} {1,4}
(ω ) → / ij klmn (5ω3 )µνj klmn − (B2 )[µ|j | (5ω3 )ν]klmn
2 3 µνi 2 5! 4!

11 {1,1} {1,1} {0,3}
+ (B2 )[µ|j | (B2 )ν]k (5ω3 )lmn ,
2 3!

1 {3,0} 1 1 {3,6} 11 {1,1} {2,5}
(ω3 )µνρ → / ij klmn (5ω3 )µνρij klmn − (B2 )[µ|i| (5ω3 )νρ]j klmn
3! 3! 6! 2 5!
11 {1,1} {1,1} {1,4}
+ (B2 )[µ|i| (B2 )ν|j | (5ω3 )ρ]klmn
2 4!
1 {1,1} {1,1} {1,1} {0,3}
− (B2 )[µ|i| (B2 )ν|j | (B2 )ρ]k (5ω3 )lmn . (60)
(3!)2
Since 5 denotes the six-dimensional internal Hodge operation, 5ω3 is a formal sum of
forms of degree 3, 5, 7 and 9:
5ω3 = (5ω3 ){0,3} + (5ω3 ){1,4} + (5ω3 ){2,5} + (5ω3 ){3,6} . (61)
Together with (21) we can then write

√
3 ){p,3−p} (−1)p(p−1)/2 G i1 ...i3−p j1 ...j3+p {p,p+3}
(F −
→ / F3+2p ,
µ1 ...µp i1 ...i3−p
(p + 3)! µ1 ...µp j1 ...j3+p
(62)
where we have defined

{p,p+3}

p
{q,q+3}
{p,p+3}
F = e −B
∧ dD + γ (−1) q(q−1)/2
5 ω3 (63)
3+2p
q=0
with the abbreviation

2
κ10
γ= 2
√ . (64)
g10 G
Using this rule, (30) now becomes

1 √
− 2 d 10 x −g |F 3 |2
4κ10

1 {3,6}
→ 2
F ∧ ∗F {3,6} + F
{2,5} ∧ ∗F {2,5}
9 9 7 7
4κ10

+F {1,4} ∧ ∗F {1,4} + F {0,3} ∧ ∗F {0,3} . (65)
5 5 3 3
The potential term |F {0,3} |2 is now modified due to the open string scalars. For vanishing
3
3-form flux and taken together with the potential term of (56), this is the potential known
also for the heterotic string [80–82].
To eliminate the RR forms of high degree, we now follow the same procedure as before
and assume vanishing 3-form flux while dualizing F {3,6} and F {2,5} . In order to dualize
9 7
{3,6}
F 9 , we impose its Bianchi identity by adding the Lagrange multiplier term (anticipating
that the Lagrange multiplier will be identified with the RR scalar as in (31))

1 {1,0} {3,6} {1,1} 1 {1,1} 2
δS = − 2 C0 d F9 + B2 ∧ (dD6 ){2,5} − B2 ∧ (dD4 ){1,4}
2κ10 2

{1,1} 1 {1,1} 2
+ γ (5ω3 ){3,6} − B2 ∧ (5ω3 ){2,5} − B2 ∧ (5ω3 ){1,4}
2

1 {1,1} 3
+ B2 ∧ (5ω3 ){0,3}
3!

1 {3,6} − 1 {2,5}
= 2 (dC0 ){1,0} ∧ F 9 2
C 0 (dB2 ){2,1} ∧ F 7
2κ10 2κ10
{1,1}
+ d {1,0} γ (5ω3 ){3,6} − B2 ∧ d {1,0} γ (5ω3 ){2,5}
1 {1,1} 2
− B2 ∧ d {1,0} γ (5ω3 ){1,4}
2
1 {1,1} 3
+ B2 ∧ d {1,0} γ (5ω3 ){0,3} . (66)
3!
Integrating out F {3,6} leads to the same result as in (34) and the only difference, as
9
compared to the case without vectors, appears in the structure of the Chern–Simons terms.
{2,5} we add
To replace F7

1 {1,1} {1,0} {2,5} {1,1} {1,4}
δS = 2 C2 ∧d F7 + B2 ∧ dC4
2κ10

{1,1} 1 {1,1} 2
+ γ −(5ω3 ){2,5} − B2 ∧ (5ω3 ){1,4} + B2 ∧ (5ω3 ){0,3} . (67)
2
Again, integrating out F {2,5} leads to the old result (38). Further, apart from the kinetic
7
and potential terms given in (34), (38) and the second line of (65), we obtain the following
Chern–Simons terms

1
SCS = 2 γ (dC2 ){2,1} − C0 (dB2 ){2,1}
2κ10

{1,1} 1 {1,1} 2
∧ (5ω3 ){2,5} + B2 ∧ (5ω3 ){1,4} − B2 ∧ (5ω3 ){0,3}
2

{1,1}
+ (dC0 ){1,0} ∧ (5ω3 ){3,6} − B2 ∧ (5ω3 ){2,5}

1 {1,1} 2 {1,4} 1 {1,1} 3 {0,3}
− − B2 ∧ (5ω3 ) + B ∧ (5ω3 )
2 3! 2

1 {0,4}
+ 2 (dC2 ){2,1} ∧ (dB2 ){2,1} ∧ C4 . (68)
2κ10
Using (64), the definition
Fj µν = 2∂[µ Cν]j , (69)

and the expressions
(ω3 )µνρ = 6Aa[µ ∂ν Aaρ] + 2f abc Aaµ Abν Acρ ,

(ω3 )µν i = 2Aa[µ ∂ν] Aai + 2Aai ∂[µ Aaν] + 2f abc Aaµ Abν Aci ,
(ω3 )µ ij = −2Aa[i ∂µ Aaj ] + 2f abc Aaµ Abi Acj ,
(ω3 )ij k = 2f abc Aai Abj Ack , (70)
one finds
1 1
(ω3 )µν i + B[µ|j | (ω3 )ν] ij − B[µ|j | Bν]k (ω3 )ij k
2 2

= Aai F µν + Aai Aaj ∂[µ Bν]j − 2∂ Aai Ãa ,
[µ ν]
(ω3 )µνρ − 3B[µ|i| (ω3 )νρ] i + 3B[µ|i| Bν|j | (ω3 )ρ] ij − B[µ|i| Bν|j | Bρ]k (ω3 )ij k
= (ω̃3 )µνρ + 6Aai Ãa[µ ∂ν Bρ]i , (71)
where we also used the definitions (55) and (57). Now one can express the Chern–Simons
terms as

1 √
SCS = − 2 d 10 x −g γ / µνρσ
4κ10

1 a a ρσ 1
× C0 Fµν Fρσ − (Fj µν − C0 Hj µν ) Aaj F a
+ Aaj Aai Hiρσ
2 2

1 {0,4}
+ 2 (dC2 ){2,1} ∧ (dB2 ){2,1} ∧ C4 . (72)
2κ10
As will be pointed out in Section 2.6, this expression matches the results obtained from
supergravity (up to some numerical factors).
Let us again see which changes occur in the presence of 3-form flux. As in the case
{0,4}
without vectors, (65) shows that the kinetic term for the scalars C4 changes and
additional terms in the potential appear. The covariant derivative of the axions (C4 )ij kl ,
which was (48) when setting all open string fields to zero, now reads
∂µ (C4 )ij kl − 2(B2 )µ[i (dC2 + 2γ 5 ω3 )j kl] + 2(C2 )µ[i (dB2 )j kl] + γ (5ω3 )µij kl .
(73)
Introducing alternative variables as in (50) one can rewrite this up to an overall factor as
1√ 1√
Dµ β ij = ∂µ β ij + G (5dC2 )ij k (B2 )µk − G (5dB2 )ij k (C2 )µk
√ 2 2

+ γ G (ω3 )µ ij − (ω3 )ij k Bµk . (74)
Using (70), (55) and (57) we see that
µ Aaj ] .
(ω3 )µ ij − (ω3 )ij k Bµk = −2Aa[i D (75)
Thus (74) can alternatively be written as
1√ 1√
Dµ β ij = ∂µ β ij + G (5dC2 )ij k (B2 )µk − G (5dB2 )ij k (C2 )µk
√ 2 2
− 2 G γ Aa[i D µ Aaj ]. (76)
Now (76) is invariant under the gauge transformation (59), whereas invariance under (58)
requires β ij to transform according to
1√
β ij → β ij − G (5dC2 )ij k /k . (77)
2
Similarly, under
(C2 )µk → (C2 )µk + ∂µ /k (78)
β ijhas to transform by
1√
β ij → β ij + G (5dB2 )ij k /k . (79)
2
In the framework of gauged supergravity this implies that the same translational isometries
of the RR scalars are gauged as in the case without vectors [34,47]. Demanding invariance
of the action under gauge-transformations of the Kaluza–Klein vectors requires that the
Chern–Simons term in the last line of (72) is replaced by (46), analogously to the case
without vectors. Apart from this modification we do not expect any further changes to
(72) due to the fluxes, because the open string Chern–Simons terms do not involve the
axions (C4 )ij kl and are already by themselves invariant under gauge-transformations of
the Kaluza–Klein vectors.11
On the other hand, the 3-form potential does receive additional contributions as argued
above in the case without vectors. Comparison with type IIB theory implies that there is
{0,3}
an additional potential term of the form |H3 |2 and that instead of (dC2 + γ 5 ω3 ){0,3} it
is the full (dD2 + γ 5 ω3 ){0,3} that should enter into |F {0,3} |2 . The two terms can again be
3
combined to give
{0,3} 2
(F3 + γ 5 ω3 ){0,3} 2 + e−2Φ H {0,3} 2 = (G3 + γ 5 ω3 ){0,3} 2 ≡
G3 . (80)
3
Actually, there is a loop-hole in our strategy to combine the T-dual data and the
truncated type IIB action in this case. The expression (80) contains the cross-term
2γ C0 (dB2 )ij k (5ω3 )ij k . This term contains NSNS 3-form flux and open-string fields at
the same time and thus vanishes in both limits. Furthermore, its presence or absence is not
restricted by any symmetry argument. This is different from the corresponding cross-term
in the kinetic term for β ij , which is required by KK gauge invariance, cf. (76). Also the
shift-symmetry of C0 is broken by the presence of the D3-branes and cannot help to fix
the coefficient of γ C0 (dB2 )ij k (5ω3 )ij k . Nevertheless, we believe that (80) is the correct
combination, because it leads to the same potential as found in the supergravity approach
[38,47], as we will show in a moment.
Putting all the pieces together, the bosonic action derived by T-duality of the type I
string contains four different parts
S = SEH + Skin + Spot + SCS . (81)

In the string frame, they are given in turn by

1 √
SEH = 2 d 10x −g e−2Φ R,
2κ10

1 √ 1
d 10 x −g e−2Φ 4∂µ Φ∂ µ Φ − Gij Hiµν Hj
µν
Skin = 2
2κ10 4

1 √ {1,0} 2 {2,1} 2 {1,4} 2
− 2 d 10 x −g F1 + F3 + F5
4κ10

1 √
− 2 d 10 x −g γ e−Φ Gij g µν Dµ Aai D
ν Aaj
2κ10

1 µν ρσ a
ai a

+ g g Fµρ + Hiµρ A Fνσ + Hj νσ A aj
, (82)
2
11 Ultimately, this is justified by comparison to the gauged supergravity of [47] and, as mentioned already, our
results match the expressions given in [47] without further modifications.

1 √ {0,3} 2 2
Spot = − 2
d 10 x −g G3 + 2γ e−Φ trF {0,2} , (83)
4κ10

1 √
SCS = − 2 d 10 x −g γ / µνρσ
4κ10

1 a a aj a 1 aj ai
× C0 Fµν Fρσ − (Fj µν − C0 Hj µν ) A Fρσ + A A Hiρσ
2 2

1 {0,4}
+ 2 (dC2 ){2,1} ∧ (dB){2,1} ∧ C4
2κ10

1 {1,1} {1,1}
− 2 C2 ∧ (dB2 ){2,1} ∧ B2 ∧ (dC2 ){0,3}
4κ10

1 {1,1} {1,1}
− 2 B2 ∧ (dC2 ){2,1} ∧ C2 ∧ (dB2 ){0,3} . (84)
4κ10
2.4. Additional modifications from the non-Abelian DBI
To identify candidate couplings in the effective action that involve 3-form flux and open
string fields at the same time, one can study the non-Abelian D-brane world-volume action
in the formulation given in [62]. It turns out that only the DBI part contains a term that has
to be added to the effective action derived via T-duality. The CS part in principle has the
potential to modify the tadpole condition, but we will argue that actually it does not.
From the expansion of the DBI action for a D3-brane in the presence of NSNS 3-form
flux one deduces a correction to the effective action [62] that in our conventions reads
i 1 1
√ e−Φ tr Ai Aj Ak Hij k = − 2 γ e−Φ ω3 Hij k .
ij k
2
(85)
3g10 G 2κ10 3!
This term represents an extra contribution to the naive Abelian term for the tension of a
D3-brane in the presence of NSNS 3-form flux.
By analyzing the non-Abelian CS action of the D3-branes one may have expected a
modification of the relevant component of the equation of motion for C4 , since one finds a
{4,0}
further coupling linear in C4 , i.e.,

{4,0} √
tr C4 iA iA B2 ∼ d 4 x −g4 / µνρσ (C4 )µνρσ Hij k tr Ai Aj Ak , (86)
using the symbolic notation of [62], i.e., iA iA B2 = Aj Ai Bij . This term could potentially
modify the tadpole cancellation condition for D3-brane charge. However, one has to take
into account that the CS action involves RR-fields of all degrees and (86) is accompanied
by another contribution, the direct analogue of the term that lead to dielectric D0-branes in
[62],

√
tr P[iA iA C6 ] ∼ d 4 x −g4 / µνρσ (dC6 )µνρσ ij k tr Ai Aj Ak , (87)
where P denotes the (non-Abelian) pull-back. The component of dC6 that occurs here
is related to the 3-form flux via duality. To make this more precise, one has to use the
democratic version of the ten-dimensional type I action (already mentioned in footnote 9),
that involves C6 on the same footing as C2 even before applying the T-duality, cf.
{4,3}
Appendix C. This formulation also involves a kinetic term for the field strength F7
after T-duality. In general, F7 is given by

1
F7 = dC6 + H3 ∧ C4 − C2 ∧ B2 , (88)
2
where we made use of the general formula (see, e.g., [77]12)

5
5
5
F2q−1 = 2q−2 + H3 ∧
dC 2q−4 ,
C (89)
q=1 q=1 q=2
and took into account that our C4 differs from C 4 used in [77] by C
4 = C4 − 1 C2 ∧ B2 ,
2
whereas all other Cp coincide, i.e., Cp = Cp for p = 4. In (86) the difference between C
4
and C4 does not matter because (B2 )µν and (C2 )µν are projected out and their backgrounds
vanish in the vacuum. Now we notice that the two terms (86) and (87) nicely combine into

√
d 4 x −g4 / µνρσ (F7 )µνρσ ij k tr Ai Aj Ak + · · · , (90)
{2,2}
where the dots stand for further terms, involving also C4 , which is dualized in favor of
{0,4}
C4 in the end. We do not want to go into further details here, but just observe that upon
{4,3}
dualizing F7 in a similar vein as done in Appendix C, the term (86) disappears from
{4,0}
the action and there is no additional contribution to the tadpole condition for C4 after
eliminating the superfluous degrees of freedom. This means that in the tadpole condition
(10) the contribution of the fluxes to the effective 3-brane charge is

1 {0,3} {0,3}
Nflux = 2 F3 ∧ H3 , (91)
2κ10 µ3
T6
exactly as in the case without the open string fields, because the CS correction from the
non-Abelian D-brane action drops out. Here µ3 denotes the 3-brane string frame tension
which for general p-branes is given by13
1 −(p+1)/2
µp = √ 2π 4π 2 α . (92)
2
Moreover, notice that in the self-dual action the kinetic term for C2 appears with a factor
1/2 as compared to the usual type I action. Thus also, e.g., the 3-form potential after
{4,3}
T-duality would first come with the same additional factor. It is only after dualizing F7
that the full 3-form potential, as given in (83), is obtained.
12 Note that our NSNS B-field differs by a sign from the one used there.
√
13 Note that the tension includes a factor 1/ 2 as compared to the corresponding formula of type IIB, as is
required in type I and its T-duals, see [83] for instance.
The fact that the tadpole condition is not changed as compared to the case without open
string fields is important for the form of the effective four-dimensional potential. First
note that this tadpole constraint descends from the equation of motion of (C4 )µνρσ in the
ten-dimensional theory and does not arise as a dynamical equation in four dimensions.
Instead it needs to be imposed as a constraint that determines the number of D3-branes.
The dual theory constructed in the previous section is only consistent in the presence of 3-
form fluxes if the fluxes do not contribute any 3-brane charges, i.e., one necessarily has
Nflux = 0. This is because T-dualizing pure type I naturally leads to a theory with 16
D3-branes14 that already fully cancel the charge (and tension) of the O3-planes. A non-
vanishing Nflux would then lead to a surplus of 3-brane charge. This will be discussed in
more detail in Section 4.1, but let us go slightly ahead of things and already mention that
in general the potential receives a further extra contribution from the tension of the O3-
planes and D3-branes, when Nflux = 0. To accommodate this, the first term of (83) can be
rewritten using the splitting of G {0,3} under 5 into imaginary self-dual and anti-self-dual
3
components, i.e., 5G ISD , 5G
= iG ISD IASD = −i G IASD .15 One then verifies, similarly to
3 3 3 3
[21], that the 3-form flux potential term combines with the extra non-Abelian correction to
the brane tension into

{0,3} 2 4i −Φ i j k

G3 − γ e tr A A A Hij k dvol
3
ISD 2 {0,3} {0,3}

= 2 G3 dvol + 2e−Φ F3 ∧ H3 (93)
√
with dvol = G d 6 x. Due to the unmodified tadpole condition we see that the second term
on the right-hand side of (93) is canceled by the tension of the localized objects.
Note that the absence of the CS correction to the tadpole condition is indispensable
for producing a positive definite scalar potential as is required by matching the results of
gauged supergravity [38,47].16 In particular, couplings that could drive a dielectric Myers’
effect, which would lead to non-commutative brane solutions, now appear as the cross
terms in |G ISD |2 . At a global minimum of the potential, when G3 is IASD, they therefore
3
cancel out [27,38].
Let us make a further comment here. The vector couplings in the kinetic terms (82)
are asymmetric in that Chern–Simons corrections do not occur in the kinetic terms for C0 ,
{1,1} {1,1}
C2 and B2 . This is due to the fact that we did not include the Chern–Simons term

−µ9 dC2 ∧ ω7 (94)
14 There is a well-known ambiguity of how to count the individual branes. In a microscopic description of the
T-dual model, there are 64 O3-planes located at the fixed points of Θ, such that local charge cancellation as in
type I is only achieved in the effective action if all sources are completely smeared out. The number 16 simply
refers to the rank of the gauge group.
15 Let us remind the reader that we are using a different definition of the Hodge star than the one used, e.g., in
[21,27,73].
16 Moreover, this form of the potential is necessary to be consistent with the constraints that derive from the
equations of motion, as discussed in Section 4.
as part of

µ9 Cp ∧ ch(F ) (95)
p
for the D9-branes in our starting point type I action (2). Under T-duality such a term would
be mapped according to
(dC2 ){q,3−q} ∧ (ω7 ){4−q,q+3}
(−1)q(q−1)/2
→ − √ (dD2+2q ){q,3+q} ∧ (5ω7 ){4−q,3−q} , (96)
G
leading, among other things, to terms involving (dD8 ){3,6} and (dD6 ){2,5} together with the
appropriate components of 5ω7 . Clearly this would modify the dualizing process described
above, leading to Chern–Simons corrections involving 5ω7 in the kinetic terms for C0 ,
{1,1} {1,1}
C2 and B2 and to new Chern–Simons terms similar in nature to (68) but lengthier.
In that case the kinetic terms appearing in (82) involve the field strengths
{1,4} = (dD4 + γ 5 ω3 ){1,4} − B {1,1} ∧ (dD2 + γ 5 ω3 ){0,3} ,
F5
{2,1} = (dD2 − 2γ 5 ω7 ){2,1} − B {1,1} ∧ (dD0 + 2γ 5 ω7 ){1,0} ,
F3
{1,0} = (dD0 + 2γ 5 ω7 ){1,0} ,
F (97)
1
{2,1} {1,0}
instead of F3 and F1 . In this symmetrized version, the coupling of the truncated type
IIB action to the gauge fields, at least for the kinetic terms, can be summarized by adding to
the dDp the appropriate component of γ 5 ω3 or 2γ 5 ω7 depending on the bi-degree of the
form. However, as (94) is a higher order correction in the open string fields and we restrict
ourselves to the lowest order of the DBI action, we preferred not to include the corrections
due to ω7 in (81).
2.5. The action in the Einstein frame
Finally, in order to make contact to the supergravity literature, we have to transform (81)
into the Einstein frame. Transforming to the ten-dimensional Einstein-frame by rescaling
the metric with the string coupling, gI J → eΦ/2 gI J , leads to

10 √ 1 1
2κ10 S = d x −g R − ∂µ Φ∂ µ Φ − e−Φ Gij Hiµν Hj
2 µν
2 4
1 1 µν µν
− e2Φ ∂µ C0 ∂ µ C0 − eΦ Gij (Fiµν + C0 Hiµν ) Fj + C0 Hj
2 4
1 −1 µ Aai D
µ Aaj
− G Gik Gj l Dµ β ij D µ β kl − γ Gij D
4
1 a aµν
− γ e−Φ F µν + Hiµν Aai F µν
+ Hj Aaj
2
1 {0,3} 2 4i −Φ i j k
− eΦ G3 − γ e tr A A A Hij k
2 3

1
− γ eΦ Gij Gkl f abc f ade Abi Ack Adj Ael
2

1 √ √
− d 10 x −g4 / µνρσ γ G C0 F µν
a a
Fρσ
4

aj a 1 aj ai
− 2(Fj µν − C0 Hj µν ) A Fρσ + A A Hiρσ − β ij Fiµν Hjρσ
2

√ √
− Ciµ Bj ν Hkρσ G (5dC2 ) − Biµ Cj ν Fkρσ G (5dB2 )
ij k ij k
, (98)
where we introduced the variables β ij and used (76). This form of the action will be
relevant in Section 4 when we allow the bulk fields to vary over the whole internal space,
and the open-string fields over the world-volume of the D-branes, looking for solutions to
the ten-dimensional equations of motion.
Dimensional reduction of (98) and a Weyl rescaling gµν → G−1/2 gµν to go to the four-
dimensional Einstein-frame produces the following effective action17

√
2κ4 S = d 4 x −g4
2

1 ik j l µ 1 1 ij Hiµν H µν
R− G G ∂µ Gij ∂ Gkl − ∂µ Φ∂ µ Φ − e−Φ G j
4 2 4
1 1 ij µν µν
− e2Φ ∂µ C0 ∂ µ C0 − eΦ G (Fiµν + C0 Hiµν ) Fj + C0 Hj
2 4
1 µ Aai D
ij D µ Aaj
− Gik Gj l Dµ β D β − γ̃ G
ij µ kl
4
1 a aµν
− γ̃ e−Φ F µν + Hiµν Aai F µν
+ Hj Aaj
2

√ ISD 2 1 Φ
− ( G )−1 eΦ G3 − γ̃ e G ij G
kl f abc f ade Abi Ack Adj Ael
2

1 √
− d 4 x −g4 / µνρσ γ̃ C0 F µν
a a
Fρσ
4

aj a 1 aj ai
− 2(Fj µν − C0 Hj µν ) A Fρσ + A A Hiρσ − β ij Fiµν Hjρσ
2

√ √
− Ciµ Bj ν Hkρσ G (5dC2 ) − Biµ Cj ν Fkρσ G (5dB2 )
ij k ij k
. (99)
Here we have already taken into account the tension of the localized objects to get the form
of the potential (sixth line) and introduced the following notation
√ κ2
ij = √1 Gij ,
G γ̃ = γ G = 10 . (100)
2
g10
G
17 We take the background volume to be (2π α 1/2 )6 and therefore have κ 2 = κ 2 (4π 2 α )−3 .
4 10
The action is now in the form that should allow for a direct comparison to the gauged
supergravity theory that captures the effective dynamics of type I strings with background
3-form fluxes.
2.6. Comparison to gauged supergravity
In this section, we would like to make explicit the comparison of our results in (99) to
those found in the gauged supergravity approach, by explaining the translation of notation,
parameters and fields. We view this as strong independent confirmation that the effective
theory, obtained by modifying the T-dual action in the manner described above to capture
the effects of 3-form fluxes in type I, is a sensible approximation of the string dynamics in
the supergravity regime. To what extent it is approximate will be subject of Section 4.
The expressions we are going to compare are the covariant derivative of the axionic
scalars, already discussed in (76), and the gauge kinetic and Chern–Simons part (84) of
the action (99). Finding basic agreement (up to two factors of 2 and a sign, see below)
with the supergravity results of [47], see also [34,35,38] for partial results, we conclude
that the effective models are identical. A third object of great interest is, of course, the
scalar potential, to which the entire Section 3 will be devoted. The expression given in
Eqs. (4.97), (4.98) and (5.132) of [47] for the covariant derivative of the axion is18
Dµ B ΛΣ = ∂µ B ΛΣ + fαΛΣΓ AαµΓ − a a[Λ ∇µ a aΣ] , (101)
where the fαΛΣΓ , α = 1, 2, are the numerical parameters for the 3-form fluxes, AαµΓ the
Abelian KK vector fields and the a aΛ are scalars charged under the non-Abelian gauge
group. Finally, ∇µ denotes their gauge covariant derivative. In view of (76) this suggests
the following mapping of fields
B ΛΣ ↔ β ij , A1µΛ ↔ (B2 )µi , A2µΛ ↔ (C2 )µi , a aΛ ↔ −Aai , (102)
and flux parameters
1√ 1√
f1ΛΣΓ ↔ G (5dC2 )ij k , f2ΛΣΓ ↔ − G (5dB2 )ij k , (103)
2 2
where we have set γ̃ = 1 throughout. The signs are going to become clearer below. The
non-Abelian vectors should map to the Ãaµ of (57). Using this map leads to agreement
between our result (76) and (101) (up to a factor of 2 for the last term).
From the gauge kinetic and CS part of (99) we can read off the θ -angles and coupling
constants of the different gauge fields, which are already present in the ungauged theory
and which we define as the coefficients in front of Fµν F µν or / µνρσ Fµν Fρσ , respectively.
Fµν now stands for any kind of gauge field strength, Fµνa ,F
iµν or Hiµν . We put them into
a matrix labeled by (a, α, i) and read off
1
θ ab = − C0 δ ab , θ i1a = Aai , θ i2a = −C0 Aai , θ i1j 1 = 0,
2
18 We are using gauge group indices a, b, . . . instead of i, j, . . . and A, B, . . . instead of I, J, . . . to avoid

confusing them with our space–time indices.
1 1 1
θ i2j 2 = − C0 Aai Aaj , θ i1j 2 = θ i2j 1 = β ij + Aai Aaj (104)
2 4 4
and
ab −2 i1a −2 i2a
g −2 = −e−Φ δ ab ,
g = 0, g = −2e−Φ Aai ,
−2 i1j 1 1 ij −2 i2j 2 1 ij
g = − eΦ G , g = − e−Φ + eΦ C02 G − Aai Aaj e−Φ ,
2 2
−2 i1j 2 −2 i2j 1 1
g = g = − C0 eΦ G ij , (105)
2
leaving a factor 1/(4κ42) in front of the action. The corresponding supergravity results can
be found in (5.130) together with (3.90)19 of [47] in the form of a matrix N of complex
coupling constants. Using

± 1 i
Fµν = Fµν ± /ρσ µν F ρσ
, (106)
2 2
and suppressing indices one has
1
+ +µν
−i N Fµν F Fµν
−N − −µν
F = Im(N )Fµν F µν + Re(N )Fµν Fρσ / µνρσ .
2
Thus the θ -angles and couplings are given by20
ab i −2 iαa
N ab ↔ 2θ ab + i g −2 , N iαa ↔ θ iαa + g ,
2
iαjβ
N iαjβ ↔ 2θ iαjβ + i g −2 . (107)
The expressions one finds for the entries of N are identical to those in (104) and (105)
upon identifying
1 ij
C ↔ C0 , ϕ ↔ −Φ, g ΛΣ ↔ G , (108)
2
in addition to (102), up to the overall factor of 1/(4κ42), the sign in the last line of (105)
and a factor of 2 in the β ij -dependent term in (104). To complete the translation of fields,
the Lα in the SU(1, 1)/U (1) coset are translated into our SL(2, R)/U (1) scalars C0 , Φ by
i i
L1 ↔ − √ eΦ/2 , L2 ↔ √ eΦ/2 C0 + ie−Φ (109)
2 2
with L1 = L2 , L2 = −L1 .
19 Beware a misprint in an older version, where in the first line of (3.91) the last term should contain a factor
of i instead of i/2.
20 Note the extra factor of 2 in the second line of (5.130) in [47].
3. The potential
From the sixth line of (99) we can read off the effective four-dimensional potential
eΦ
ISD 2 γ̃ eΦ abc ade bi ck dj el
Veff = √ (G 3 + γ 5 ω3 ) + Gij Gkl f f A A A A . (110)
2κ42 G 4κ42
Before discussing the implications of this potential let us compare it to the expression
derived with the formalism of N = 4 gauged supergravity. In the conventions of [47] the
scalar potential was written21
1 2 1 2
VSUGRA = F ABC− + C ABC− + L2 f abc q bA q cB , (111)
2 4
where F is the 3-form flux
F ABC = Lα fαABC = Lα fαΛΣΓ EΛ

A B C
EΣ EΓ , (112)
the upper index F − referring to the ISD part, C ABC is the CS correction
C ABC = L2 f abc q aA q bB q cC = L2 f abc a aΛ a bΣ a cΓ EΛ

A B C
EΣ EΓ , (113)
and indices are pulled back with the vielbein A
EΛ which is related to the metric (100) via
(108) and
gΛΣ = δAB EΛ
A B
EΣ . (114)
The metric moduli together with the axions from B ΛΣ and the a aΛ parametrize the scalar
manifold
SO(6, 6 + N)
. (115)
SO(6) × SO(6 + N)
The identification of the two expressions is then accomplished by applying (102), (103)
and (108), where one has to take care that the extra signs from 55 = −1 and (109) cancel
out. We find agreement VSUGRA = Veff after setting κ42 = γ̃ = 1 in (110). Strictly speaking,
this choice of parameters is allowed only within supergravity. String theory relates the two
in a way that is not consistent with setting them equal.
An important phenomenological feature of (110) is that it involves the dilaton not just as
a prefactor, so one can hope to stabilize the string coupling. Concentrating on Minkowski
vacua we require
2
(G3 + γ 5 ω3 )ISD 2 = 0, trF {0,2} = 0. (116)
As tr |F {0,2} |2 = 0 also implies vanishing of ω3 , the relevant term for fixing Φ is given by
|GISD
3 | = 0, [38]. The condition for G3 to be IASD can be written
2
5F3 − e−Φ H3 = 0. (117)
21 We absorb factors of p! into our definition of the norm (A.1).

Rewriting this via (103) one finds a relation
f1ΛΣΓ − C0 f2ΛΣΓ + e−Φ (5f2 )ΛΣΓ = 0, (118)

where 5f2 is defined with respect to the metric gΛΣ . Setting C0 = 0 for simplicity, we see
that
1
f1ΛΣΓ + e−Φ det(gΛΣ ) g ΛΛ g ΣΣ g Γ Γ /̂Λ Σ Γ ∆ΠΩ f2∆ΠΩ = 0, (119)
3!
where /̂ is just again the anti-symmetric symbol with value ±1. It appears that in general
the dilaton becomes a function of the metric moduli that also depends on the choice
of fluxes. It was argued in [47] that the 3-form fluxes allowed by the gauging, i.e.,
consistent with a supersymmetric Lagrangian, can be put into a form such that, in complex
coordinates, all components fαmm̄n and fαmm̄n̄ vanish. We will see in Section 3.2 that this
ensures a fixing of the dilaton at least in vacua in which supersymmetry has been broken
to N = 1 via a super-Higgs-mechanism.
3.1. The role of the superpotential
In compactifications of type IIB theory on Calabi–Yau spaces with fluxes, the scalar
potential that descends from the kinetic term of G3 , after breaking supersymmetry to
N = 1 either by orientifolding [21] or by taking a certain decompactification limit as in
[5,13], can be expressed in terms of the superpotential

Wflux = G3 ∧ Ω 3 , (120)
M6
proposed by independent arguments in [11,12], with Ω3 the holomorphic 3-form. In that

case, the scalar potential only depends on the complex structure moduli of the Calabi–Yau
and the dilaton. As is well known, in the heterotic or type I string the superpotential gets
a further contribution involving the scalars descending from the ten-dimensional vectors
[80–82]

Whet/I = ω3 ∧ Ω3 . (121)
M6
As in the N = 1 case, also for N = 4 the potential descends from the ten-dimensional
ISD |2 plus tr|F {0,2} |2 , up to constants.
kinetic terms and can thus be written in the form of |G 3
Therefore, one might wonder whether it is still possible to express the potential in terms of
a “superpotential”. In view of the effective scalar potential (110) we have found in type I
with extra 3-form flux, it seems more appropriate to consider the “superpotential”

WI = (G3 + γ 5 ω3 ) ∧ Ω3 . (122)
M6
However, in the toroidal compactification the potential also depends on the Kähler form,
which reflects the fact that the moduli space of a torus does not split into a direct product
of complex structure and Kähler moduli as on a Calabi–Yau. On a Calabi–Yau manifold,

the ISD 3-forms are the (1, 2)- and (3, 0)-forms, see, e.g., [21]. Moreover, all (1, 2)-forms
are primitive [84]. On the torus, on the other hand, there are non-primitive (1, 2)-forms
of the form J ∧ d z̄m where J is the Kähler form, and these are also IASD. Moreover,
there are further ISD (2, 1)-forms of the form J ∧ dzm [27] which all enter into the scalar
potential (110). These facts make it very cumbersome to express the potential in terms of
the superpotential (122).22
Still, the superpotential (120) has been used in the literature to encode the conditions
for unbroken supersymmetry in a toroidal (or K3×T2 ) background with 3-form flux [27,
41].23 There it was shown that demanding supersymmetry is strong enough to fix the
period matrix of the torus completely, which implies fixing its complex structure. The
conditions for supersymmetry are stronger than demanding extremality of the potential,
but are equivalent to the extremality conditions of the superpotential (120). As only ISD 3-
form flux enters the potential, there are possibilities to turn on fluxes without generating a
vacuum energy. The IASD 3-forms consist in G3 being a (0, 3)-, or a primitive (2, 1)-form
or of the type J ∧ d z̄m . On the other hand, in order to preserve supersymmetry the flux
has to be a primitive (2, 1)-form [73,85] for at least one complex structure. The number
of unbroken supersymmetries depends on the number of complex structures for which this
condition is fulfilled [27,28]. Moreover, it is obvious from the formulas of the gravitino
masses given in [34] that turning on any non-trivial 3-form flux breaks supersymmetry at
least to N = 3. Since the coupling to the open string fields is manifest in the supervariation
of the gravitinos, dilatinos and gauginos by replacing F ABC with F ABC + C ABC as used in
(111) [47], it is to be expected that the supersymmetry conditions in the coupled system are
also captured by the modified superpotential (122) (in addition to the “D-term” appearing
in the second contribution to the potential (111)), although the scalar potential (110) cannot
be expressed through WI in an obvious way.
3.2. Truncating to N = 1
Regarding the discussion of the superpotential (122) in the previous section, it appears
interesting to study the breaking of supersymmetry in the present model from N = 4 to
smaller numbers of supercharges, especially to N = 1, in which case the potential should
be expressible in terms of a superpotential (and a D-term). This has been done to some
extent in the framework of supergravity [34,35,38,86,87] and in this section we would
simply like to make contact to [35].
By turning on more and more components of the 3-form flux one can successively break
supersymmetry from N = 4 to N = 1, as has been described in [34,35]. In the super-
Higgs effect, the decoupling of massive modes is restricted by the requirement that they
22 In order to follow the calculation in the Calabi–Yau case [5,13,21,24], one would have to know what should
replace the formula DZα Ω3 = χα , that gives a basis for the (2, 1)-forms χα in terms of covariant derivatives
of Ω3 with respect to the complex structure moduli Z α . For the torus, one would need a corresponding formula
giving a basis of only the primitive (2, 1)-forms. However, the split of the (2, 1)-forms into primitive and non-
primitive ones depends on the moduli.
23 The case of K3×T2 has also been analyzed from the point of view of gauged supergravity in [44].
must fill massive representations of the surviving supersymmetry algebra. If one breaks
supersymmetry from N = 4 → 3 → 2 → 1 successively, 6, 10 or all 12 (Kaluza–Klein)
vector fields get massive via their Stückelberg couplings (74). This is precisely as required
by a successive decoupling of massive 3/2 multiplets under N = 3, 2, 1 supersymmetry,
which eat just the 6, 4 and 2 vectors, respectively, those that disappear from the massless
sector. In this successive supersymmetry breaking the matter content of (9), a spin-2 and
six spin-1 multiplets of N = 4, leads to a spin-2 and three chiral multiplets of N = 1 [35].
Therefore, 3 gravitini, all 12 vectors, 18 out of 21 scalars G ij , 12 out of 15 axions β ij and
τ get masses with suitable degeneracies to fill massive N = 1 multiplets. The remaining
three complex scalars are given by the diagonal components G ,G ,G of the now
11̄ 22̄ 33̄
Hermitian metric and the appropriate components of β ij . Together with the open string
moduli they parametrize the remnant scalar manifold

SU(1, 1) SO(6, 6 + N) SU(1, 1 + N) 3
× → . (123)
U (1) SO(6) × SO(6 + N) U (1)
In this particular situation one can convince oneself that the condition (117) is really
sufficient to fix the dilaton. The square root of the determinant of the metric is the product
G
G
11̄ 22̄ G33̄ of the real parts of the only remaining three moduli, which drop out from (119)
if the fluxes are restricted as explained below (119). Thus (117) becomes independent of
the metric and Φ is fixed to some rational value once the fluxes F3 and H3 are subjected to
Dirac quantization. A natural value for the string coupling would appear to be of order one,
but moderately small numbers are also easily accessible. In [38] it was suspected that in
the process of integrating out the dilaton, the first term of the potential (110) vanishes, such
that the potential of the effective N = 1 theory used in [35] is solely given by the second
term with a constant value for the dilaton. We were not able to perform this integrating out
explicitly, but assume in the following that the N = 1 potential is only given by the second
term of (110).
It is well known in the literature how to obtain the N = 1 potential from a dimensional
reduction of the kinetic term for the vector fields [80,81]. From (110) (and setting
2πα 1/2 = 1) we extract
2
4g10 ik G
VN =1 = eΦ G j l F aij F akl = eΦ G
ik G
j l f abc f ade Abi Acj Adk Ael , (124)
and it is understood that the dilaton and all metric moduli except the diagonal ones are
frozen in the N = 1 vacuum. The potential (124) can be split into an F-term and a D-term
by identifying the metric components with the (real parts of the) Kähler coordinates tm in
the following way
tm + t¯m̄ − C am C mm̄ )−1 .
a m̄ = (G (125)
Here we have introduced the complex fields
1
C am = √ Aa(2m−1) + iAa(2m) . (126)
2
Using the Bianchi identity (which implies F a ∧ F a = 0), one can rearrange terms to arrive
at

ik G
G mn̄ G
j l F aij F akl = 2G op̄ F amo F a n̄p̄ + F amn̄ F aop̄ . (127)
The (2, 0) part can be written24

3

∂W ∂ W
G op̄ F amo F a n̄p̄ = eK
mn̄ G a m̄
tm + t¯m̄ − C am C , (128)
bm
∂C ∂ C bm̄
m=1
with superpotential
1 abc am bn co
W= f C C C /̂mno , (129)
3!
where the hat on the epsilon symbol was explained below (19), and Kähler potential

3

K=− a m̄ .
log tm + t¯m̄ − C am C (130)
m=1
ᾱ W
That the potential-term (128) is of the usual form eK (K α ᾱ Dα W D − 3|W |2 ), where α
m m α ᾱ
denotes all chiral fields, t and C , and K the inverse of the Kähler metric, was shown
in [35]. The fact that the −3|W |2 drops out in (128) is due to the no-scale structure of
the potential. The choice of complex coordinates implies a diagonal period matrix, and the
holomorphic 3-form can be written Ω3 = dz1 ∧ dz2 ∧ dz3 , so that (129) can be expressed
as in (121).25 In a similar way one can identify the (1, 1) component of (127) with an
N = 1 D-term
3 2
1

Gmn̄ Gop̄ F F
amn̄ aop̄
= bcd cm d m̄
f C C . (131)
em̄
t + t¯m̄ − C em C
m=1 m
Interestingly, adding a world-volume gauge flux by hand, i.e., adding a constant f amn̄ to
cm̄ , would lead to a Fayet–Iliopoulos term in (131).
F amn̄ = f abc C bm C
4. Vacua with fluxes and warped metric
In the previous sections we have employed T-duality to transform the effective action
obtained by compactifying type I strings on a torus and neglecting all dependence of the
fields on internal directions. The additional terms in the effective action with NSNS 3-
form flux on the internal space were added in via comparison with type IIB. However,
constant fields will in general no longer solve the ten-dimensional equations of motion
once fluxes are turned on. In this section we shall look for solutions to these equations,
based on the action (98), with more general background configurations and non-trivial
profiles admitting four-dimensional Minkowski vacua. As explained in the introduction, in
order to derive the equations of motion we allow for a dependence of the bulk fields on
the internal coordinates and further modify the action (98) by introducing a background
24 Note the misprint in formula (12) of [35].

25 Note that one could have defined the variables C am of (126) alternatively as C am = √1 (Aa(2m−1) −
2
iAa(2m) ). In that case, the superpotential (129) would have been of the form (122), with vanishing G3 .
for the 5-form field strength commensurate with the T-dual world sheet parity projection
(1) of type IIB. Moreover, we take into account that the D3-branes and O3-planes T-dual
to the D9-branes and O9-planes are actually localized objects. We consider the modified
equations of motion as describing the coupling of the bulk fields to the tension of the branes
(respectively O-planes) and their world-volume fields. We find that the gauge fields only
cause minor corrections to the known solutions for a background with vanishing world-
volume fields [21,73]. For these solutions it has been argued that constant internal fields
may still be considered approximate solutions in the large volume limit where the warp
factor may be taken to be approximately constant, thus a posteriori justifying the effective
action (99)—at least in this limit.
4.1. Generalized ansatz and modified action
Our starting point is the ten-dimensional action (98) in which we now allow the bulk
fields to depend on the internal coordinates. Furthermore, we have to implement some
modifications, that we discuss in the following.
As was already explained, since the modified world sheet projection ΩΘ(−1)FL
does not project out fields locally, but only demands symmetry or anti-symmetry of the
background, we may also allow non-trivial backgrounds for non-dynamical fields, as long
{0,3}
as they respect Poincaré invariance. This refers to the 3-form fluxes G3 , which are
not field strengths of any dynamical potentials but do survive the projection, and also
{0,5} {4,1}
to the components F5 and F5 of the 5-form. The latter is also subject to the self-
duality constraint, such that both 5-form components can be parametrized through a single
function,26
1 1
(F5 )µ1 ...µ4 i = √ /µ1 ...µ4 ∂i α, (F5 )i1 ...i5 = √ /i1 ...i5 j (∂k α)g j k , (132)
g4 g4
{4,0}
where α = α(x i ) is anti-symmetric under Θ, since Ω(−1)FL leaves C4 invariant. In the
equations of motion, this background enters in the same way as in type IIB. For the metric
we use the general warped product ansatz
−1 b
2
ds10 = gI J dx I dx J = ∆ x k ĝµν dx µ dx ν + ∆ x k ĝij dx i dx j . (133)
The factor in front of the internal part is purely conventional, but if one intends to identify
ĝij with a Ricci-flat or even constant metric eventually, it can be helpful to make the
warping explicit. In general, one may also want to consider a non-trivial internal profile
{0,1}
for the axion F1 , as was done in [88] to analyze the effects of 7-branes, but we shall not
do so here. Furthermore, in looking for a solution to the coupled system of equations we
{2,1} {1,0}
restrict our ansatz to the case of vanishing field strengths F3 and F1 .
Second, the open string fields should be allowed to vary only over the world-volume
of the D-branes present in the background. As long as all fields were constant on the
internal space, there was no distinction between fields localized on some brane and fields
26 Note that due to our different definition of the Hodge star the internal components have the opposite sign as
in [73].
propagating throughout the bulk. T-duality of type I string theory produced (56), where the
kinetic terms for the gauge fields only involve the determinant of the four-dimensional
metric, as for gauge fields localized on a D3-brane, but there is no delta-function to
localize the fields. Thus, (56) refers to “smeared out” D3-branes. In the same way, there
was no localization of CS interactions involving open string fields. Further, once we
have introduced localization, the tensions of D-branes and O-planes do not cancel locally
anymore, and we need to make the tension terms explicit. O3-planes and D3-branes appear
naturally in the toroidal model, since they are just the images of the O9-planes and D9-
branes of type I under T-duality. In fact, one could also try to introduce higher-dimensional
branes, say D7-branes, and have non-trivial gauge fluxes on their world-volume, which
induces effective D3-brane charges. We refrain from doing so here, and come back to this
option later. One of the main reasons is that adding world-volume gauge fluxes would
require us to take higher terms in the DBI effective action into account, which would
restrict us to Abelian gauge fields, since the non-Abelian DBI action is not known to higher
order.
From expanding the DBI action to second order in the gauge field strength one infers the
relative factor between the tension term and the terms of the world-volume action already
present in (98) to be 12 (2πα )2 . Setting 2πα = 1 we thus use for the D3-brane action27

√ i i j k

SD3 = −µ3 d x −g4 10
sm − tr A A A Hij k δ 6 (xm − x̄m )
3
branes

1 µ Aai D
ν Aaj
+ sm δ 6 (xm − x̄m ) gij g µν D
2
branes
1 a a
+ e−Φ g µν g ρσ Fµρ + Hiµρ Aai F νσ + Hj νσ Aaj
2

1 Φ
+ e gij gkl f abc f ade Abi Ack Adj Ael . (134)
2
The additional term from the non-Abelian correction to the DBI action has also been
included above. Just to keep track of the overall sign we have introduced coefficients
sm = ±1 for the tension of the branes. In order to avoid kinetic terms with the wrong
sign for the world-volume fields, one would choose sm = 1. In the background, the only
non-vanishing terms are the tension in the first line and the scalar potential tr|F {0,2} |2 in
the last. The sum over branes now also implies that the gauge fields are labeled individually
for any single stack of branes, but we do not want to encumber the notation with another
index. Since we have made the D-brane tension explicit we also have to add the O3-plane
tension

64
√
SO3 = −µ3 Q3 d 10 x −g4 δ 6 (xk − x̄k ) (135)
k=1
27 Note that this is consistent with (99) as 1 (2π α )2 µ = γ̃ (2κ 2 )−1 , where κ is given in footnote 17 and we
√ 2 3 4 4
2 = 2 (2π )7 α 3 , cf. [89].
used (92) and g10
for the 64 planes localized at the fixed points x̄k of Θ and Q3 = −1/4. Here we assume
that the O3-planes have standard negative tension and negative charge. The total action for
the open string sector including the O3-planes is then given by
Sop = SD3 + SO3 . (136)

In the same fashion, the CS interactions with open string fields have to be localized. For
instance, when we write |F 3 |2 , we now understand

3 |2 dvol = |dD2 |2 dvol +
|F qm (2π)6 δ 6 (xm − x̄m )
branes

× 2γ (5ω3 ) ∧ ∗dD2 + γ 2 | 5 ω3 |2 dvol , (137)
√
with dvol = d 10 x −g. The qm = ±1 are parameters for the RR charge of the D3-branes
and distinguish branes from anti-branes and enter in all the topological terms. Since the CS
forms 5ω3 will mostly only appear inside F p in the following, we will not make this kind
of localization explicit, to keep the notation compact.
4.2. Equations of motion and constraints
The most interesting equations of motion to consider are the two sets of Einstein
equations for internal and external indices, the equation for the dilaton, and the charge
conservation constraint coming from the equation of motion for C4 . Einstein’s equations
with the general warped ansatz (133) are abbreviated, using the notation of (A.10) (see
Appendices A.2 and A.3),

Rµν = R µν + 1 ĝµν ∆1−3b ĝ ij ∇
i ∆2(b−1)∂j ln(∆) = κ10
2
Sµν ,
2
Rij = κ10
2
Sij . (138)
The explicit dependence of Rij on the warp factor is given in the appendix in (A.8).
Furthermore, as in the appendix, ∇ i is the covariant derivative of ĝij .
The first set of equations allows us to determine the warp factor, which then has to
be checked with the second set. We specialize to maximally symmetric four-dimensional
µν = m2 where m2 is zero, positive or negative for Minkowski, de
space–times, i.e., ĝ µν R 4 4
Sitter or anti-de Sitter space, respectively. Then, upon taking the trace, one arrives at

i ∆2(b−1)∂j ln(∆)
ĝ ij ∇
∆−2 {0,3} 2 Φ 1 3b−1 2 ∆2+2b
=− G3 ĝ
e − ∆ m4 − (∂i α)(∂j α)ĝ ij
4 2 2ĝ4
2 µ Q
∆−2 κ10
64
3 3
− δ 6 (xk − x̄k )
ĝ6 k=1

∆ κ10 µ3
−2 2
i
− sm − tr Ai Aj Ak Hij k δ 6 (xm − x̄m )
ĝ6 3
branes

{0,2} 2
+e ∆Φ −2b
tr F s δ (xm − x̄m ) ,
6
(139)
ĝ m
branes
where ĝ4 denotes the absolute value of the determinant of ĝµν and the subscript
{0,3} |2 implies that the contractions are to be performed with ĝ ij . Furthermore,
on |G 3 ĝ
tr|F {0,2} |2ĝ = 12 ĝ ij ĝ kl f abc f ade Abi Ack Adj Ael is given by ∆4 21 ĝij ĝkl f abc f ade Abi Ack Adj Ael ,
when expressed through the metric independent fields Aai . The integral of the right-
hand side over the internal space is now forced to vanish. The D-brane tension and the
contributions of the YM sector add up with the 3-form flux, while the O3-plane tensions
serve as positive contributions on the right-hand side, which may be employed to balance
the negative contributions from the fluxes.28 The cosmological constant and the correction
term of the form tr(Ai Aj Ak )Hij k can add to either the positive or negative contributions.
Upon specializing to b = 1 the second set of equations becomes tractable. Combining
the two sets of Einstein equations one can eliminate the term with the Laplacian acting on
the warp factor and obtains
∆2 µν 1 ij
Rµν ĝ + Rij ĝ
4 6
1 ij ∆4 ij κ2
= ĝ (∂i ln ∆)(∂j ln ∆) − ĝ (∂i α)(∂j α) − 10 Tij − Tij5-flux ĝ ij . (140)
3 12ĝ4 12
=R
An obvious solution with R = 0 consists in α = ±∆−2 and (T − T 5-flux )g ij = 0.
µν ij ij ij
3 from (140) since
This can be satisfied by tr |F |2 = 0. There is, however, no constraint on G
Tij3-flux g ij = 0,
On the other hand, combining (139) with the equation of motion for the external
components of C4 leads to a further constraint. The latter has been discussed in some
detail in Section 2.4, where it was pointed out that no corrections due to the non-Abelian
CS action occur. It reads
{0,3} {0,3}

64
(dF5 ){0,6} = F3 ∧ H3 + 2κ10
2
µ3 qm πm + 2κ10
2
µ3 Q3 πk , (141)
branes k=1
and explicitly we have

1 ij 4 ij
ĝ ∇i ∂j α + ĝ (∂i ∆)(∂j α)
ĝ4 ∆
1
=− Fij k Hlmn / ij klmn
(3!)2∆

2κ102 µ
3 64
− qm δ (xm − x̄m ) + Q3
6
δ (xk − x̄k ) .
6
(142)
ĝ6 ∆4 branes k=1
28 The sign conventions are such that negative contributions on the right-hand side of (139) are related to
positive energy density.
The πm and πk stand for the delta-function-like 6-forms with support at the location of
3 in (139) and by adding ∆2 times
the 3-branes and 3-planes. Using the splitting (93) of G
(139) and − 2 ∆ times (142), one can derive the powerful constraint
1 4

∆4 ij α −2
− ĝ ∇i ∂j + ∆
2 ĝ4
6
∆ ij α −2 α −2
= − ĝ ∂i + ∆ ∂j + ∆
2 ĝ4 ĝ4
1 ISD 2 κ10 µ3 ∆
2 −2 2
− eΦ G3 ĝ − eΦ trF {0,2} ĝ sm δ 6 (xm − x̄m )
2 ĝ6 branes
2 µ
κ10 3 1
− (sm − qm )δ 6 (xm − x̄m ) − ∆4 m24 . (143)
ĝ6 branes 2
This generalizes the result of [21] by including the world-volume fields of the D3-branes
and allowing for m24 = 0. Integrating the total derivative, a number of restrictions follow.
The four-dimensional cosmological constant has to satisfy m24 0, excluding de Sitter
space within the chosen ansatz. In addition, sm = qm = 1 follows, so there are no anti-
branes. For vanishing cosmological constant, i.e., ĝµν = ηµν , all terms on the right-hand
side are negative semi-definite and therefore have to vanish individually. Thus,
ISD 2 2
α = −∆−2 ,
G3 = trF {0,2} = 0. (144)
These conditions are equivalent to asking for a global minimum of the effective potential
(110) obtained by neglecting the warp factor
ISD
Veff = 0 ⇐⇒ G = 0 and tr |F |2 = 0 . (145)
3
Since this potential is non-negative, global minima are precisely the Minkowski vacua.
These conclusions are clearly not valid forfour-dimensional anti-de Sitter space, which
seems to demand a choice for α such that α/ ĝ4 is independent of the external coordinates.
Using the formulas from Appendices A.2 and A.3 it is possible to show that the
conditions (144) ensure that the full set of Einstein equations are satisfied if
i ∂j ∆2
ĝ ij ∇

1 Φ 2 2κ10 2 µ
3 i i j k

= − e |G3 |ĝ − sm − tr A A A Hij k δ 6 (xm − x̄m )
2 ĝ6 3
branes

64
+ Q3 δ 6 (xk − x̄k ) . (146)
k=1
3 , as was
To show this one has to use the fact that Tij3-flux = 0 for purely IASD (or ISD) G
noticed in [73]. Imposing (144), the equation determining the warp factor (146) is actually
equivalent to (142).
The other equations of motions are not too difficult to find. The equation for Φ is
∆3 √
√ ∂I g I J −g ∂J Φ
−g
1 (0,3)2 (0,3)2
= eΦ F3 ĝ
− e−Φ H3 ĝ
2
κ 2 µ3 Φ {0,2} 2
+ 10 e trF sm δ 6 (xm − x̄m ).
ĝ
(147)
ĝ6 ∆2 branes
This equation imposes a vanishing of the dilaton tadpole. By four-dimensional Poincaré
invariance the right-hand side of the equation has to integrate to zero. An IASD 3-form
3 |2 = e−2Φ |H3 |2 and therefore the YM potential must vanish, tr|F |2 = 0. Thus,
fulfills |F
the solutions of the constraint coming from the Einstein equations are compatible with the
vanishing of the dilaton tadpole.
To summarize the logic: asking for a Minkowski vacuum leads to (139) with m24 = 0,
combining with (141) implies (144) and is compatible with R ij = 0 at constant dilaton. We
are then dealing with a warped product of a Calabi–Yau or torus and Minkowski space,
where the warp factor is determined via (146), which is equivalent to (142). The situation
described above is a variant of the known no-go theorems of [51–53]. Several escape routes
are at least in principle known: in an anti-de Sitter vacuum, a positive term would appear on
the right-hand side of (139) that could balance the vacuum energy of the potential terms. If
higher curvature corrections to the action were considered the semi-definiteness could also
be spoiled and de Sitter solutions may exist as well. Perturbative [32] or non-perturbative
[43] corrections to the effective potential also appear to modify the conclusions drawn
above. Finally, one may also want to allow an explicitly time-dependent background [90].
4.3. Large volume scaling limit: separation of mass scales
The solution given by the conditions (144) is too simple to fix all the metric moduli;
e.g., it always leaves the overall volume of the internal space as a free parameter. We will
later come back to the possibility to fix also this by adding world-volume gauge fluxes on
higher-dimensional branes. For the moment, one can observe that, like in [21], the relations
that determine the solution are all invariant under constant rescaling of the metric. This is in
accord with the form of the effective potential despite the fact that both terms in (110) scale
differently under gij → tgij . The reason is that for a Minkowski vacuum both terms have
to vanish separately. This agrees with the results of [34,47] where the no-scale structure of
the gauged four-dimensional N = 4 supergravity theory, that is the effective description of
the present scenario, was demonstrated explicitly.
While the scale-independence of (144) is unfortunate in that it does not lead to a fixing
of the volume modulus, it has been argued to allow for a limit of parametrically large
volume where the warping becomes insignificant [21,28]. In this limit, one can perform a
standard dimensional reduction of the action to four dimensions. Then (110) really takes
the role of an effective potential of a theory obtained by expanding around a given solution
to the equations of motion. This is justified by inspection of (142) (or alternatively of
(139)): the metric factors on the left-hand side of (142) scale like t −1 , on the right-hand
side like t −3 . As Fij k and Hij k are just constants, the warp factor itself has to go as 1 + t −2
at large t. This argument is, of course, only valid away from the positions of the branes
and planes. Ignoring the contributions from these regions one can then set ∆ to a constant
in the large t limit and derive the effective potential (110) by dimensional reduction. In
this situation, it appears challenging to find ways to stabilize the volume modulus in the
effective theory that is only valid at very large volume. Given a correction to (110) that
fixes the volume, one would need to make sure that this is done at large enough values that
the effects of the warping are still negligible. In principle, it would of course be much nicer
if one could do a reduction without using the scaling argument by explicitly including the
warp factor, similar to [36].
The scaling argument can also be used for branes of higher dimensions. In order to
neglect the warp factor at large t the contributions of all relevant types of matter on the
right-hand side of the Einstein equations have to fall off faster than t −1 . The tension of an
arbitrary Dp-brane (in Einstein-frame) leads to
p − 7 µp e(p−3)Φ/4
Sµν
Dp
= √ gµν δ 9−p (xm − x̄m ). (148)
8 g⊥
Here g⊥ indicates the determinant of the metric restricted to the normal bundle of the
brane. If the metric is constant this is the transverse volume. Sµν then scales like t (p−9)/2
and p < 7 is required for the scaling argument to work.
The validity of the approach in addition requires a separation of the flux-induced masses
and the masses of KK modes that have been neglected throughout. It has been argued in
[9,27] that the ratio of masses induced by the 3-form flux and the KK masses scales like

(m3-flux : mKK ) = R −3 : R −1 , (149)
where R denotes the dimensionless (average) radius of the background torus. Thus also
from here we see that we need a large volume, this time to ensure a decoupling of the KK
states. The same reasoning is still true in the case at hand since the additional second term
in the potential (110) does not introduce any new mass terms for the geometric moduli. This
is due to the fact that F {0,2} vanishes in the background, independently of the geometric
moduli. Thus the masses only depend on the 3-form flux, as is also apparent from the mass
formulas in [47]. This would be different if one considered higher-dimensional branes
with internal world-volume fluxes, to which we come back in the next section. In that case
one would expect an additional mass scale from the world-volume fluxes, similar to the
situation in the heterotic string discussed in [9] and an intermediate scale appears, below
the KK but still above the 3-form flux scale:

(m3-flux : m2-flux : mKK) = R −3 : R −2 : R −1 . (150)
5. Generalization of the open string sector
In this final section we want to go beyond the well-defined framework of the model
obtained from type I by T-duality, and allow higher-dimensional D-branes of even
dimension and with internal world-volume gauge fluxes. This means we use
Fija = fija + f abc Abi Acj , (151)
instead of (51). The constant flux parameters fij = fija T a
take values in the Cartan
subalgebra of the gauge group. They characterize a non-trivial gauge bundle on the
world-volume. It is well known that the open string boundary conditions with constant
fluxes change from Dirichlet to mixed Dirichlet–Neumann conditions. This implies that
performing two T-dualities on a Dp-brane wrapping a two-dimensional torus with flux
f12 = 0 turns it into another Dp-brane of identical dimension, but with the flux inverted.
Therefore, introducing such gauge fluxes into the original type I string theory, assuming
a factorization of the background into T6 = (T2 )3 for simplicity, could have taken us to
a T-dual theory with D3-, D5-, D7- or D9-branes (depending on the number of internal
directions with world-volume flux) wrapping 0, 2, 4 or all directions of the T6 after the
duality, and with non-trivial gauge fluxes on their world-volume. At the same time, the
O9-planes would still map to O3-planes. The D5-, D7- and D9-brane charges then have to
cancel among the branes, so that some of them need to carry negative charge as well and
behave as anti-branes. This leads to the conclusion that there are no supersymmetric vacua
with non-trivial world-volume flux in a flat background on a torus, which is in accord with
the known fact that the toroidal type I compactification with world-volume gauge fluxes
on D9-branes, that has been of some phenomenological interest recently, does not have a
supersymmetric ground state [91]. To achieve a better behaved model one may consider
orbifold compactifications which also have O7-planes in addition to the O3-planes. In
this case supersymmetric configurations of D9-branes with world-volume fluxes do exist
[92–94].29 Here we content ourselves with studying a few general properties of D9-brane
flux vacua to obtain a rough impression of the effects of warping and perspectives for
moduli stabilization.
In the presence of internal gauge fluxes, we have to use the DBI effective action
including terms with higher powers of F , since in the presence of world-volume fluxes the
higher powers in F contribute to leading order in a derivative expansion of the effective
action. Furthermore, some of the CS corrections in ω7 are no longer higher order and need
to be incorporated into the action as in (97). On the other hand, the full DBI effective action
can only reliably be used for an Abelian gauge group. Therefore we now pass to a U (1)N
gauge group by turning on Higgs fields. In this way we lose the possibility to produce chiral
matter in various representations, which is one of the main motivations to be interested in
this variant of the model [95–97]. However, as said above, here we are mainly interested in
the prospects for moduli stabilization. To simplify the notation we also restrict ourselves
to D9-branes, using

√
SDBI = −µ9 d 10 x −g4 e3Φ/2 det gij + Fij e−Φ/2
R4 ×M6

√
=− d 4x −g4 VDBI (152)
R4
29 This type of models, combined with 3-form fluxes, has been considered recently in [45,46].
as effective world-volume action. Considering D5- or D7-branes should pose no further

difficulties and would lead to similar conclusions.
5.1. Supersymmetry and κ-symmetry
In this section we discuss the conditions for preserving world-volume supersymmetry

on D9-branes in the simultaneous presence of 3-form fluxes and O3-planes in the
background. In their absence they are given by the calibration conditions found in [98].
The background bulk fields are not only required to satisfy the conditions (144) in
order to solve the equations of motion, but we also impose GIASD 3 to be of type (2, 1)
and primitive [73,85] to preserve supersymmetry. One then finds that in this background
there is a Killing spinor that can be expressed as a covariantly constant spinor rescaled by
a scalar function only [73],
/ = ∆−1/4 η. (153)
The background is 1/2 BPS, since η also satisfies a chirality projection. In general,
world-volume supersymmetry of D-branes in non-trivial backgrounds is preserved if the
supervariations can be absorbed by κ-symmetry variations. This is translated into the
projection
Γ / = /, (154)
where Γ = Γ (P[g], F ) is the κ-projector, an operator depending on the pull-back of the
(warped) metric and the gauge field strength F , and / is a Killing spinor of the background
bulk theory. Explicitly, the κ-symmetry projector is defined by
Γ = e−a/2(Γ0...p ⊗ iσ2 )ea/2 (155)

with
1 i j AB 1 Ai j AB
a = Yij EA EB Γ ⊗ σ3 = ∆Yij E EB Γ ⊗ σ3 ,
2 2
Yij = “arctan”(Fij ), iA E
ĝij = δAB E jB , (156)
where “arctan” is explained in [99], to which we refer the reader for further information
on the notation and techniques. The details do not play a prominent role in our discussion
here. The important point for us is the existence of a Killing spinor of the type (153) in
a background of supersymmetric 3-form fluxes and O3-planes, a rescaled (covariantly)
constant Killing spinor. Neglecting the backreaction of the D9-branes themselves, one can
use this Killing spinor to derive analogous calibration conditions as found in [98] by just
modifying the background metric via inclusion of the warp factor. A simple example with
a flat background ĝij = δij has been studied in [100].
5.2. Calibrated branes
We are now ready to apply the results of [98] to find the supersymmetry conditions
for D9-branes with world-volume gauge fluxes in a given background of O3-planes and
3-form fluxes. It is then required that the purely holomorphic components of F vanish and
that
1
eiθ det(gij + Fij ) dx1 ∧ · · · ∧ dx6 = (J + iF )3 ,
3!

1 1
cos(θ ) J ∧J ∧F − F ∧F ∧F
2! 3!

1 1
= sin(θ ) J ∧J ∧J − F ∧F ∧J (157)
3! 2!
hold for some phase θ on all the D9-branes. J = ∆Jˆ now stands for a rescaling of the
Kähler form Jˆ of the Calabi–Yau. For non-trivial ∆ this J is not closed. The background
O3-planes (and any possible O7-planes) are calibrated with an angle θ = 3π/2. If one
neglects the warp factor, i.e., uses ∆ = 1, an overall world-volume supersymmetry can
be preserved by adopting the same angle for all the D9-branes as well. It is important to
note that fixing θ , induces an implicit dependence of the Kähler moduli on F . The scalar
potential for the Kähler moduli in the Einstein frame reads
−iθ
e
−Φ/2 3
VDBI ∼ det gij + Fij e −Φ/2 = Re J + iF e . (158)
3!
M6 M6
One further needs to impose the charge conservation constraint, which now includes the
open string CS interactions µ9 C4 ∧ ch(F ), such that (141) is modified to

64
µ9
dF5 = F3 ∧ H3 + 2κ10
2
µ3 Q3 πk + 2κ10
2
F ∧ F ∧ F,
3!
k=1 branes
which leads to the condition

1
F ∧ F ∧ F = 32 − Nflux . (159)
3(4π 2 α )3
branes M
6
Similarly, the couplings of RR 8-form potentials in the open string CS action leads to
a tadpole constraint involving F , which states that the sums of the effective RR 7-brane
charge has to be zero. As already said above, in order to fulfill (157) for all branes with the
same angle θ one would have to depart from the flat torus example or to include O7-planes
via an orbifold construction as in [45,46]. Then (157) together with GISD 3 = 0 ensures the
vanishing of the total potential energy and (157) degenerates into
J ∧ J ∧ J = 3J ∧ F ∧ F (160)
for any individual stack of branes. This constraint ensures the balancing of the 9-brane and
5-brane tensions, while 3- and 7-brane charges and tensions are unconstrained and only
cancel upon adding the O-planes. It fixes the overall √ radius in an obvious manner. The
natural scale for its value is however of the order of α , with a constant of proportionality
given by a combination of flux numbers. As one should maintain the splitting of scales
(150) it would be required to have hierarchically large flux quantum numbers, which by
(159) would demand some “integer fine tuning” to cancel the tadpoles. Furthermore, once
the volume modulus gets fixed at a finite value, the warp factor is not trivial anymore.
Writing J = ∆Jˆ for closed Jˆ one sees that (160) cannot be fulfilled anymore. It would
actually be equivalent to Jˆ3 = ∆−2 Jˆ ∧ F 2 , where the left-hand side is closed and the
right-hand side not.
This probably does not mean that the volume modulus cannot be stabilized in this way,
since the qualitative stabilization mechanism due to the balancing of the 9-brane and 5-
brane tensions should still take place. On the other hand, it seems that the non-trivial warp
factor leads to a breaking of supersymmetry, once the volume is fixed. A similar result
has been found in [100]. However, for a complete analysis one would also have to take
into account the backreaction of the background towards the presence of the D9-branes
and possible O7-planes. This would probably lead to a significant modification of the
calibration condition (157) but one should expect the above results on the stabilization
of the volume to be qualitatively true also in the full story [45,46].
Acknowledgements
It is a great pleasure to acknowledge stimulating discussions and helpful explanations

provided to us by Katrin and Melanie Becker, Massimo Bianchi, Fawad Hassan,
Shamit Kachru, Jurg Käppeli, Jan Louis, Gianfranco Pradisi, Augusto Sagnotti, Henning
Samtleben, Michael Schulz, Mario Trigiante and Marco Zagermann. During the work B.K.
enjoyed the hospitality of the Università di Roma “Tor Vergata”, while M.B. and M.H.
performed part of this work at the University of Utrecht and the Spinoza Institute. M.H.
would also like to thank the theory group at CERN for hospitality during the final stages
of the work. We further like to thank the organizers of the RTN network conferences
Superstring Theory at Cambridge 2002, where this work was initiated, and those of the
quantum structure of spacetime and the geometric nature of fundamental interactions at
Torino 2003. The work of B.K. was supported by the German Science Foundation (DFG)
and in part by funds provided by the US Department of Energy (DOE) under cooperative
research agreement #DF-FC02-94ER40818. The work of M.B. and M.H. was supported
in part by INFN, by the E.C. RTN programs HPRN-CT-2000-00122 and HPRN-CT-2000-
00148, by the INTAS contract 99-1-590, by the MURST-COFIN contract 2001-025492
and by the NATO contract PST.CLG.978785.
Appendix A. Technicalities
A.1. Conventions and notation
Tangent-frame indices are generally denoted I, J, . . . = 0, . . . , 9, which decompose

into i, j, . . . = 4, . . . , 9 and µ, ν, . . . = 0, . . . , 3. Once we complexify coordinates, we use
m, n, . . . . We use the following standard conventions for differential forms: for Ωp ∈
p
T ∗ M10 , M10 = R4 × T6 , write
1
Ωp = ΩI ...I dx I1 ∧ · · · ∧ dx Ip ,
p! 1 p
1
I1 ...Ip ,
|Ωp |2 = ΩI1 ...Ip Ω (A.1)
p!
{n,p−n}
where complex forms obviously refer to the complexified cotangent bundle, for Ωp ∈
n ∗ 4 p−n ∗ 6
T R × T T analogously
{n,p−n} 1
Ωp = Ωµ ...µ i ...i
n!(p − n)! 1 n 1 p−n
× dx µ1 ∧ · · · ∧ dx µn ∧ dx i1 ∧ · · · ∧ dx ip−n ,
{n,p−n} 2 1
Ω p = Ωµ ...µ i ...i Ωµ1 ...µn i1 ...ip−n . (A.2)
n!(p − n)! 1 n 1 p−n
Ten-dimensional Hodge duality is defined by
1
∗Ωp = / I1 ...Ip Ip+1 ...I10 ΩI1 ...Ip dx Ip+1 ∧ · · · ∧ dx I10 , (A.3)
p!(10 − p)!
and the six-dimensional Hodge star is given by
{n,6−p+n} 1
(5Ωp )µ1 ...µn j1 ...j6−p+n = / i1 ...ip−n j1 ...j6−p+n Ωµ1 ...µn i1 ...ip−n . (A.4)
(p − n)!
We also define derivative operators
{n,p−n} {n,p−n} {n+1,p−n}
d {1,0} Ωp = dΩp ,
{n,p−n} {n,p−n} {n,p+1−n}
d {0,1} Ωp = dΩp . (A.5)
For the totally anti-symmetric tensor / we use the convention
√ 1
/1...D = ± gD , / 1...D = √ , (A.6)
gD
with gD being the (absolute value of the) determinant of the metric and the sign depends
on the signature of the metric.
A.2. Einstein equations with warp factors
In this appendix we give the results for the Ricci tensor of a general warped metric
ansatz (133). With that ansatz the Christoffel symbols are found
= Γ νρ
µ
µ
Γνρ µ
, Γij = Γµj
i
= 0,
1 1
= ĝµν ∆−1−b ĝ ij ∂j ln(∆),
µ
Γνi = − δνµ ∂i ln(∆), i
Γµν
2 2
b i
Γj k = Γ j k + δj ∂k ln(∆) + δk ∂j ln(∆) − ĝj k ĝ il ∂l ln(∆) .
i i i
(A.7)
2
This leads to

Rµν = R µν + 1 ĝµν ∆−1−b ĝ ij ∇ i ∂j ln(∆) − 2(1 − b)ĝ ij ∂i ln(∆)∂j ln(∆)
2

=R µν + 1 ĝµν ∆1−3b ĝ ij ∇ i ∆2(b−1)∂j ln(∆) ,
2

Rij = Rij + 2(1 − b)∇ i ∂j ln(∆) + b2 − 2b − 1 ∂i ln(∆) ∂j ln(∆)
b k ∂l ln(∆),
+ b(1 − b)ĝij ĝ kl ∂k ln(∆)∂l ln(∆) − ĝij ĝ kl ∇ (A.8)
2
where ∇ i involves the Christoffel symbols Γ i . We define the energy–momentum tensor
jk
for some action S[g] by
2 δS
TI J = − √ , (A.9)
−g δg I J
T denoting its trace TI J g I J , and the Einstein equations are then written
1 1
2
RI J = SI J , SI J = TI J − gI J T . (A.10)
κ10 8
In a flat Minkowski vacuum (R µν = 0) the trace of the space–time components then always
takes the form
1 ij 2(b−1) 1
2
ĝ ∇i ∆ ∂j ln(∆) = ∆3b−2 Tµν g µν − Tij g ij . (A.11)
κ10 4
From this follows the famous constraint [51] that the right-hand side, with the given warped
ansatz and without higher derivative terms in the action, has to integrate to zero. Thus,
whenever it is positive or negative definite, it has to vanish, which then implies the absence
of fluxes and warp factors.
A.3. Energy–momentum tensors
In this section we give the explicit forms of the energy–momentum tensors of the various
background contributions referring to the ten-dimensional action (98) in the Einstein-
frame. The only Poincaré invariant background form fields we turn on are the internal
components of the 3-form, i.e., the fluxes G {0,3} , and the components F {4,1} and F {0,5} (i.e.,
3 5 5
we do not consider any background for the RR axion that would be needed in discussing
D7-brane solutions, cf. [88]). The 5-form ansatz has to be self-dual and was given in (132).
It satisfies F5 = (1 + ∗) dα ∧ dx 0 ∧ · · · ∧ dx 3 , such that we have in particular |F5 |2 = 0.
For these fluxes one finds the following contributions to the energy–momentum tensor
∆−3b Φ {0,3} 2
3-flux
Tµν =− 2
e gµν
G3 ĝ ,
4κ10
1 {0,3} 2 {0,3} {0,3}
Tij3-flux = − 2 eΦ gij ∆−3b
G3 ĝ − G
3 (i|kl|
G
3 j )mn
g km g ln ,
4κ10

5-flux 1 1
Tµν = 2
(F5 )µρσ τ i (F5 )ναβγj g ρα g σβ g τ γ g ij
4κ10 3!
1
=− g (∂ α)(∂j α)g ij ,
2 g µν i
4κ10 4

5-flux 1 1
Tij = 2 KO LP MQ NR
(F5 )iKLMN (F5 )j OP QR g g g g
4κ10 4!

1 1
= 2 gij (∂k α)(∂l α)g kl − (∂i α)(∂j α) . (A.12)
2κ10 g4 2
Note Tij3-flux g ij = TI5J-flux g I J = 0. If one specializes to b = 1, α = −∆−2 and ĝµν = ηµν

one can rewrite (A.8) into [73]
1
Rµν − κ10
2
Tµν5-flux i ∂j ∆2 ,
= ηµν ∆−4 ĝ ij ∇
4

2 5-flux
Rij − κ10 Tij =R ij − 1 ĝij ∆−2 ĝ kl ∇
k ∂l ∆2 . (A.13)
4
This is the starting point for the self-dual 3-brane solution. In the background the only
allowed components of the open string gauge fields are F {0,2} . From (134) we then get for
the open string sector
op
TI J = TIten
J + TI J ,
YM
µ3 i i j k

Tµν
ten
= − √ gµν sm − tr A A A Hij k δ 6 (xm − x̄m ),
g6 3
branes
Tijten = 0,
µ3 ∆−2b 2
Tµν
YM
=− √ gµν eΦ trF {0,2} ĝ sm δ 6 (xm − x̄m ),
2 g6
branes
µ3
TijYM = √ eΦ g kl tr(Fik Fj l )sm δ 6 (xm − x̄m ), (A.14)
g6
branes
where we added the correction from the non-Abelian DBI action to the tension. For
orientifold planes we write
µ3 Q3 64
Tµν
O3
= − √ gµν δ 6 (xk − x̄k ), TijO3 = 0. (A.15)
g6
k=1
One then also has TIYM

J g
I J = 0.
A.4. T-duality of RR forms
In this section we specialize the rules for dualizing RR forms given in [75] to the case
relevant in this paper. The essential step is to replace a RR p-form Ωp with a state

{q,p−q} 1 1 {q,p−q}
Ωp → Ωµ1 ...µq i1 ...ip−q dx µ1 ∧ · · · ∧ dx µq ψ i1 † · · · ψ ip−q † |0",
(p − q)! q!
(A.16)
where ψ i† and ψi are raising and lowering operators

ψ i† , ψj = δji , {ψi , ψj } = 0 = ψ i† , ψ j † , (A.17)
acting on a vacuum with ψi |0" = 0 and #0|0" = 1. The elements of the duality group,
Λ ∈ O(6, 6, R), then act on the space of states (A.16) via operators , whose action on
(ψ † , ψ) = (ψ i† , ψj )i,j =1,...,6 is given by

A B A B
Λ= : ψ † −1 , ψ−1 = ψ † , ψ . (A.18)
C D C D
At the same time, the internal components of the metric (in our case of vanishing internal
NSNS B-field) transform according to
G → (AG + B)(CG + D)−1 . (A.19)

We are interested in just one particular element that corresponds to inverting the radii of all
d = 6 circles. This is in fact not a completely well specified operation, since the signs that
can appear in the mapping of the RR forms depend on the order in which the circles are
dualized. In view of (A.19) we choose the element Λ with A = D = 0 and B = C = 16 ,
where 16 denotes the six-dimensional unity matrix. This element is given by30
= − −
1 · · · 6 , −
i = ψ − ψi ,
i†
(A.20)
which satisfies 2 = −1 and
ψi −1 = ψ i† , ψ i† −1 = ψi . (A.21)

The action of on a given RR form (A.16) can then be obtained from
ψ i1 † · · · ψ in † |0" = ψi1 · · · ψin |0"

(−1)n(n−1)/2
= /̂i1 ...i6 ψ in+1 † · · · ψ i6 † |0", (A.22)
(6 − n)!
where /̂ has been introduced below (19).
30 In terms of the notation used in [75] this corresponds to = C− .

Appendix B. Kinetic and CS terms for the scalars (C4 )ij kl
The kinetic terms for the scalars (C4 )ij kl given in (45) (however without 3-form flux)
and the Chern–Simons terms of (39) are also derivable directly by a reduction of the
truncated type IIB action. In this section we calculate them in a similar fashion as is done,
e.g., in [29,79]. We start with the pseudoaction of type IIB string theory without imposing
the self-duality of the 5-form field strength, reduce it to four dimensions and afterwards
impose the self-duality by adding a suitable Lagrange multiplier. The relevant part of the
ten-dimensional action is

1 1
S= 2 F5 ∧ ∗F5 − 2 C4 ∧ dB2 ∧ dC2 + · · · . (B.1)
8κ10 4κ10
The field strength F5 is defined as in (8), i.e., after the truncation to the T-dual type I theory
1 {1,1} 1 {1,1}
F5 = (dC4 ){1,4} + (dC4 ){3,2} − B2 ∧ (dC2 ){2,1} + C2 ∧ (dB2 ){2,1} . (B.2)
2 2
This corresponds to expanding C4 as
{2,2} {0,4}
C4 = C4 + C4 . (B.3)
The self-duality condition for F5 gives

1 {1,1} 1 {1,1}
(dC4 ){1,4} = ∗ (dC4 ){3,2} − B2 ∧ (dC2 ){2,1} + C2 ∧ (dB2 ){2,1} . (B.4)
2 2
Plugging (B.2) and (B.3) into (B.1) we obtain

1 {3,2} 1 {1,1} {2,1} 1 {1,1} {2,1}
S= 2 (dC4 ) − B2 ∧ (dC2 ) + C2 ∧ (dB2 )
8κ10 2 2

{3,2} 1 {1,1} {2,1} 1 {1,1} {2,1}
∧ ∗ (dC4 ) − B2 ∧ (dC2 ) + C2 ∧ (dB2 )
2 2

+ (dC4 ){1,4} ∧ ∗(dC4 ){1,4}

1 {0,4}
− 2 C4 ∧ (dB2 ){2,1} ∧ (dC2 ){2,1} . (B.5)
4κ10
In order to implement the self-duality condition (B.4) we add the Lagrange multiplier
{0,4}
Č4

1 {3,2} 1 {1,1} {2,1} 1 {1,1} {2,1}
δS = − 2 F5 + B2 ∧ (dC2 ) − C2 ∧ (dB2 )
4κ10 2 2
∧ (d Č4 ){1,4} . (B.6)
{0,4}
The equation of motion for Č4 implies
{3,2} 1 {1,1} 1 {1,1}

F5 = (dC4 ){3,2} − B2 ∧ (dC2 ){2,1} + C2 ∧ (dB2 ){2,1} . (B.7)
2 2
{3,2}
On the other hand, the equation of motion for F5 in combination with the self-duality
condition (B.4) leads to the identification of (dC4 ){1,4} and (d Č4 ){1,4}
(d Č4 ){1,4} = (dC4 ){1,4} . (B.8)

Using this in S + δS gives (after a partial integration)

1
S= 2 (dC4 ){1,4} ∧ ∗(dC4 ){1,4}
4κ10

1 {0,4}
+ 2 (dC2 ){2,1} ∧ (dB2 ){2,1} ∧ C4 , (B.9)
2κ10
{0,4}
which obviously coincides with (39) and the kinetic term for C4 in (45) in the absence
of fluxes.
Let us also mention here that it is not straightforward to modify this procedure to include
also the 3-form fluxes. If one just naively plugs the ansatz with 3-form fluxes into the action
(B.1), one has to replace
1 {1,1} 1 {1,1}
(dC4 ){1,4} → (dC4 ){1,4} − B2 ∧ (dC2 ){0,3} + C2 ∧ (dB2 ){0,3} (B.10)
2 2
in (B.2), (B.4) and (B.5). In addition one gets a further Chern–Simons term

1 {2,2}
− 2 C4 ∧ (dB2 ){2,1} ∧ (dC2 ){0,3} − (dC2 ){2,1} ∧ (dB2 ){0,3} . (B.11)
4κ10
One immediately runs into trouble now when one considers the equation of motion for
C {2,2} . It is given by
{3,2}
d {1,0} ∗ F5 = −(dB2 ){2,1} ∧ (dC2 ){0,3} + (dC2 ){2,1} ∧ (dB2 ){0,3} , (B.12)
which, however, is not consistent with the self-duality condition (B.4) after performing the
substitution (B.10). This would require a factor 1/2 on the right side of (B.12), as
{1,4} 1 1
d {1,0} F5 = − (dB2 ){2,1} ∧ (dC2 ){0,3} + (dC2 ){2,1} ∧ (dB2 ){0,3} . (B.13)
2 2
This formula is consistent with the {2, 4}-component of the Bianchi identity
(dF5 ){2,4} = −(dB2 ){2,1} ∧ (dC2 ){0,3} + (dC2 ){2,1} ∧ (dB2 ){0,3} , (B.14)
{1,4}
as the latter actually gets two contributions. Only one is from d {1,0} F5 . The other one
{2,3}
comes from d {0,1} F5 and is identical to the first one.
Appendix C. The self-dual type I action
Due to the fact that the conventional formulation of the CS action of D-branes involves
RR potentials of all (even) degrees (in type IIB) [62], it might seem more appropriate
to start with a formulation of type I supergravity that is democratic in the appearance of
the RR fields C2 and C6 . Its closed string part could be obtained by quotienting world-
sheet parity out of the democratic version of type IIB, whose RR sector was given in
(15), and which is also much better adapted to make the invariance of the type IIB action
under T-duality explicit. We did not follow this approach because the introduction of the
superfluous degrees of freedom is only needed in a discussion of the correct treatment
of the CS-terms appearing in the D3-brane action, cf. Section 2.4. For the derivation of
the T-dual action performed in chapter 2 it is actually not necessary to work with this
democratic formulation. Nevertheless, we give it here for completeness and because of its
relevance for the discussion in Section 2.4.
The self-dual type I action which contains C2 and C6 democratically is given by31
Sself-dual

1 √ 1
= 2 d 10 x −g e−2Φ R + 4∂µ Φ∂ µ Φ − |F 3 |2 − 1 |F
7 |2
2κ10 4 4

1 √ 1
− 2 d 10 x −g e−Φ tr |F |2 − 2 dC2 ∧ ω7
2g10 4g10
2
1 κ10
− 2 dC6 ∧ ω3 + 4 ω7 ∧ ω3 . (C.1)
4g10 4g10
We have included terms dCp ∧ ω9−p stemming from the Chern–Simons action Cp ∧ ch(F )
3 we then define
of the 9-branes. In analogy to F
2
κ10
7 = dC6 −
F ω7 . (C.2)
2
g10
The normalization of the D9-brane CS terms is exactly one half of the ordinary. In addition
to this action the duality condition
∗F 7
3 = −F (C.3)
has to be imposed after deriving the equations of motion. This procedure is justified
primarily by checking that the equations of motion and Bianchi identities that result from
the self-dual action become identical to the standard equations derived from (2) plus the
D9-brane interaction dC2 ∧ ω7 , after imposing the constraint (C.3). This requirement lead
us to introduce the slightly exotic interaction ω7 ∧ ω3 .32
Let us also check that the action (C.1) really reproduces the old action (2) plus D-brane
Chern–Simons term after the duality constraint (C.3) has been eliminated. Thus we add the
Lagrange multiplier term
2
1 1 κ10
δS = − 2 d Č2 ∧ dC6 = − 2
d Č2 ∧ F7 + 2 ω7 . (C.4)
4κ10 4κ10 g10
31 Actually, as already mentioned in the main text, in the presence of the ω -terms it is not consistent to just
7
keep the quadratic term |F |2 in an expansion of the DBI action. It is understood that further terms would have to
be added in a rigorous treatment.
32 Such terms are, however, familiar from anomaly cancellation, see, e.g., formula (A.14) in [101].
7 is just
The equation of motion for F
2
κ10
7 = −F̌3 = −d Č2 +
∗F ω3 , (C.5)
2
g10
i.e., F̌3 has the same dependence on Č2 as F3 has on C2 . Comparing with (C.3) and
using ∗∗ = 1 for a form of odd degree in an even-dimensional space–time of Lorentzian
signature, we see that
3 .
F̌3 = F (C.6)
7 leads to the original action
Using this and (C.5) to eliminate F

1 √ 1
SI = 2 d 10 x −g e−2Φ R + 4∂µ Φ∂ µ Φ − |F 3 |2
2κ10 2

1 √ 1
− 2 d 10 x −g e−Φ tr |F |2 − 2 dC2 ∧ ω7 (C.7)
2g10 2g10
involving only the RR 2-form C2 including the properly normalized Chern–Simons action
of a D9-brane.
On the other hand, one may want to integrate out C2 in place of C6 . Instead of (C.4) we
add

1 1 κ2
δS = − 2 dC2 ∧ d Č6 = − 2 3 + 10 ω3 ∧ d Č6 .
F (C.8)
4κ10 4κ10 2
g10
3 is given by
The equation of motion of F
2
κ10
3 = d Č6 +
∗F ω7 . (C.9)
2
g10
Comparing this with (C.3) now leads to the relation
d Č6 = −dC6 . (C.10)

3 and get
With the help of (C.10) and (C.9) we can eliminate F

1 √ 1
SI = 2 d 10 x −g e−2Φ R + 4∂µ Φ∂ µ Φ − |F 7 |2
2κ10 2

1 √ 1
− 2 d 10 x −g e−Φ tr |F |2 − 2 F7 ∧ ω3 . (C.11)
2g10 2g10
In particular there is now an extra Chern–Simons term of the form

1
2
ω3 ∧ ω7 . (C.12)
2g10
References
[1] C. Angelantonj, A. Sagnotti, Open strings, Phys. Rep. 371 (2002) 1, hep-th/0204089.
[2] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, The hierarchy problem and new dimensions at a millimeter,
Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[3] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, New dimensions at a millimeter to a Fermi
and superstrings at a TeV, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[4] J. Polchinski, A. Strominger, New vacua for type II string theory, Phys. Lett. B 388 (1996) 736, hep-
th/9510227.
[5] J. Michelson, Compactifications of type IIB strings to four dimensions with non-trivial classical potential,
Nucl. Phys. B 495 (1997) 127, hep-th/9610151.
[6] A. Lukas, B.A. Ovrut, D. Waldram, Stabilizing dilaton and moduli vacua in string and M-theory cosmology,
Nucl. Phys. B 509 (1998) 169, hep-th/9611204.
[7] W. Lerche, Fayet–Iliopoulos potentials from four-folds, JHEP 9711 (1997) 004, hep-th/9709146.
[8] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Heterotic M-theory in five dimensions, Nucl. Phys. B 552
(1999) 246, hep-th/9806051.
[9] N. Kaloper, R.C. Myers, The o(dd) story of massive supergravity, JHEP 9905 (1999) 010, hep-th/9901045.
[10] P. Kaste, On the twisted chiral potential in 2d and the analogue of rigid special geometry for 4-folds,
JHEP 9906 (1999) 021, hep-th/9904218.
[11] S. Gukov, C. Vafa, E. Witten, CFT’s from Calabi–Yau four-folds, Nucl. Phys. B 584 (2000) 69, hep-
th/9906070;
S. Gukov, C. Vafa, E. Witten, Nucl. Phys. B 608 (2001) 477, Erratum.
[12] S. Gukov, Solitons, superpotentials and calibrations, Nucl. Phys. B 574 (2000) 169, hep-th/9911011.
[13] T.R. Taylor, C. Vafa, RR flux on Calabi–Yau and partial supersymmetry breaking, Phys. Lett. B 474 (2000)
130, hep-th/9912152.
[14] K. Behrndt, S. Gukov, Domain walls and superpotentials from M-theory on Calabi–Yau three-folds, Nucl.
Phys. B 580 (2000) 225, hep-th/0001082.
[15] P. Mayr, On supersymmetry breaking in string theory and its realization in brane worlds, Nucl. Phys. B 593
(2001) 99, hep-th/0003198.
[16] B.R. Greene, K. Schalm, G. Shiu, Warped compactifications in M- and F-theory, Nucl. Phys. B 584 (2000)
480, hep-th/0004103.
[17] M. Haack, J. Louis, M. Marquart, Type IIA and heterotic string vacua in D = 2, Nucl. Phys. B 598 (2001)
30, hep-th/0011075.
[18] G. Curio, A. Krause, Four-flux and warped heterotic M-theory compactifications, Nucl. Phys. B 602 (2001)
172, hep-th/0012152.
[19] G. Curio, A. Klemm, D. Lüst, S. Theisen, On the vacuum structure of type II string compactifications on
Calabi–Yau spaces with H-fluxes, Nucl. Phys. B 609 (2001) 3, hep-th/0012213.
[20] M. Haack, J. Louis, M-theory compactified on Calabi–Yau fourfolds with background flux, Phys. Lett. B 507
(2001) 296, hep-th/0103068.
[21] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66
(2002) 106006, hep-th/0105097.
[22] G. Curio, A. Klemm, B. Körs, D. Lüst, Fluxes in heterotic and type II string compactifications, Nucl. Phys.
B 620 (2002) 237, hep-th/0106155.
[23] K. Becker, M. Becker, Supersymmetry breaking, M-theory and fluxes, JHEP 0107 (2001) 038, hep-
th/0107044.
[24] G. Dall’Agata, Type IIB supergravity compactified on a Calabi–Yau manifold with H-fluxes, JHEP 0111
(2001) 005, hep-th/0107264.
[25] G. Curio, A. Krause, G-fluxes and non-perturbative stabilization of heterotic M-theory, Nucl. Phys. B 643
(2002) 131, hep-th/0108220.
[26] J. Louis, A. Micu, Heterotic string theory with background fluxes, Nucl. Phys. B 626 (2002) 26, hep-
th/0110187.
[27] S. Kachru, M.B. Schulz, S. Trivedi, Moduli stabilization from fluxes in a simple IIB orientifold, hep-
th/0201028.
[28] A.R. Frey, J. Polchinski, N = 3 warped compactifications, Phys. Rev. D 65 (2002) 126009, hep-th/0201029.
[29] J. Louis, A. Micu, Type II theories compactified on Calabi–Yau threefolds in the presence of background
fluxes, Nucl. Phys. B 635 (2002) 395, hep-th/0202168.
[30] C. Beasley, E. Witten, A note on fluxes and superpotentials in M-theory compactifications on manifolds of
G(2) holonomy, JHEP 0207 (2002) 046, hep-th/0203061.
[31] S. Gukov, M. Haack, IIA string theory on Calabi–Yau fourfolds with background fluxes, Nucl. Phys. B 639
(2002) 95, hep-th/0203267.
[32] K. Becker, M. Becker, M. Haack, J. Louis, Supersymmetry breaking and α -corrections to flux induced
potentials, JHEP 0206 (2002) 060, hep-th/0204254.
[33] R. Argurio, V.L. Campos, G. Ferretti, R. Heise, Freezing of moduli with fluxes in three dimensions, Nucl.
Phys. B 640 (2002) 351, hep-th/0205295.
[34] R. D’Auria, S. Ferrara, S. Vaula, N = 4 gauged supergravity and a IIB orientifold with fluxes, New J. Phys. 4
(2002) 71, hep-th/0206241.
[35] S. Ferrara, M. Porrati, N = 1 no-scale supergravity from IIB orientifolds, Phys. Lett. B 545 (2002) 411,
hep-th/0207135.
[36] O. DeWolfe, S.B. Giddings, Scales and hierarchies in warped compactifications and brane worlds, Phys.
Rev. D 67 (2003) 066008, hep-th/0208123.
[37] M. Becker, D. Constantin, A note on flux induced superpotentials in string theory, hep-th/0210131.
[38] R. D’Auria, S. Ferrara, M.A. Lledo, S. Vaula, No-scale N = 4 supergravity coupled to Yang–Mills: the
scalar potential and super-Higgs effect, Phys. Lett. B 557 (2003) 278, hep-th/0211027.
[39] M. Berg, M. Haack, H. Samtleben, Calabi–Yau fourfolds with flux and supersymmetry breaking, hep-
th/0212255.
[40] B.S. Acharya, A moduli fixing mechanism in M-theory, hep-th/0212294.
[41] P.K. Tripathy, S.P. Trivedi, Compactification with flux on K3 and tori, hep-th/0301139.
[42] K. Becker, M. Becker, K. Dasgupta, P.S. Green, Compactifications of heterotic theory on non-Kähler
complex manifolds. I, JHEP 0304 (2003) 007, hep-th/0301161.
[43] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, De Sitter vacua in string theory, hep-th/0301240.
[44] L. Andrianopoli, R. D’Auria, S. Ferrara, M.A. Lledo, 4D gauged supergravity analysis of type IIB vacua on
K3 × T2 /Z2 , hep-th/0302174.
[45] R. Blumenhagen, D. Lüst, T.R. Taylor, Moduli stabilization in chiral type IIB orientifold models with fluxes,
hep-th/0303016.
[46] J.F. Cascales, A.M. Uranga, Chiral 4d N = 1 string vacua with D-branes and NSNS and RR fluxes, hep-
th/0303024.
[47] R. D’Auria, S. Ferrara, F. Gargiulo, M. Trigiante, S. Vaula, N = 4 supergravity Lagrangian for type IIB on
T6 /Z2 in presence of fluxes and D3-branes, hep-th/0303049.
[48] K. Becker, M. Becker, K. Dasgupta, S. Prokushkin, Properties of heterotic vacua from superpotentials, hep-
th/0304001.
[49] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev. Lett. 83 (1999)
3370, hep-ph/9905221.
[50] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690, hep-
th/9906064.
[51] B. de Wit, D.J. Smit, N.D. Hari Dass, Residual supersymmetry of compactified D = 10 supergravity, Nucl.
Phys. B 283 (1987) 165.
[52] J.M. Maldacena, C. Nunez, Supergravity description of field theories on curved manifolds and a no-go
theorem, Int. J. Mod. Phys. A 16 (2001) 822, hep-th/0007018.
[53] S. Ivanov, G. Papadopoulos, A no-go theorem for string warped compactifications, Phys. Lett. B 497 (2001)
309, hep-th/0008232.
[54] L.E. Ibanez, C. Munoz, S. Rigolin, Aspects of type I string phenomenology, Nucl. Phys. B 553 (1999) 43,
hep-ph/9812397.
[55] M. Berkooz, M.R. Douglas, R.G. Leigh, Branes intersecting at angles, Nucl. Phys. B 480 (1996) 265, hep-
th/9606139.
[56] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Y.S. Stanev, Chiral asymmetry in four-dimensional
open-string vacua, Phys. Lett. B 385 (1996) 96, hep-th/9606169.
[57] G. Aldazabal, A. Font, L.E. Ibanez, G. Violero, D = 4, N = 1, type IIB orientifolds, Nucl. Phys. B 536
(1998) 29, hep-th/9804026.
[58] B. de Wit, Supergravity, hep-th/0212245.

[59] V.A. Tsokur, Y.M. Zinovev, Spontaneous supersymmetry breaking in N = 4 supergravity with matter, Phys.
At. Nucl. 59 (1996) 2192, Yad. Fiz. 59 (12) (1996) 2277, hep-th/9411104.
[60] L. Andrianopoli, R. D’Auria, S. Ferrara, M.A. Lledo, Duality and spontaneously broken supergravity in flat
backgrounds, Nucl. Phys. B 640 (2002) 63, hep-th/0204145.
[61] L. Andrianopoli, R. D’Auria, S. Ferrara, M.A. Lledo, Gauged extended supergravity without cosmological
constant: no-scale structure and supersymmetry breaking, hep-th/0212141.
[62] R.C. Myers, Dielectric-branes, JHEP 9912 (1999) 022, hep-th/9910053.
[63] S.F. Hassan, SO(d, d) transformations of Ramond–Ramond fields and space–time spinors, Nucl. Phys. B 583
(2000) 431, hep-th/9912236.
[64] A. Dabholkar, C. Hull, Duality twists, orbifolds, and fluxes, hep-th/0210209.
[65] S. Gurrieri, J. Louis, A. Micu, D. Waldram, Mirror symmetry in generalized Calabi–Yau compactifications,
Nucl. Phys. B 654 (2003) 61, hep-th/0211102.
[66] S. Kachru, M.B. Schulz, P.K. Tripathy, S.P. Trivedi, New supersymmetric string compactifications, hep-
th/0211182.
[67] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[68] K. Dasgupta, G. Rajesh, S. Sethi, M-theory, orientifolds and G-flux, JHEP 9908 (1999) 023, hep-th/9908088.
[69] K. Becker, K. Dasgupta, Heterotic strings with torsion, JHEP 0211 (2002) 006, hep-th/0209077.
[70] G.L. Cardoso, G. Curio, G. Dall’Agata, D. Lüst, P. Manousselis, G. Zoupanos, Non-Kähler string
backgrounds and their five torsion classes, Nucl. Phys. B 652 (2003) 5, hep-th/0211118.
[71] A.M. Uranga, D-branes, fluxes and chirality, JHEP 0204 (2002) 016, hep-th/0201221.
[72] J.R. Ellis, A.B. Lahanas, D.V. Nanopoulos, K. Tamvakis, No-scale supersymmetric standard model, Phys.
Lett. B 134 (1984) 429.
[73] M. Grana, J. Polchinski, Supersymmetric three-form flux perturbations on AdS(5), Phys. Rev. D 63 (2001)
026001, hep-th/0009211.
[74] A. Giveon, M. Porrati, E. Rabinovici, Target space duality in string theory, Phys. Rep. 244 (1994) 77, hep-
th/9401139.
[75] M. Fukuma, T. Oota, H. Tanaka, Comments on T-dualities of Ramond–Ramond potentials on tori, Prog.
Theor. Phys. 103 (2000) 425, hep-th/9907132.
[76] S.F. Hassan, R. Minasian, D-brane couplings, RR fields and Clifford multiplication, hep-th/0008149.
[77] E. Bergshoeff, R. Kallosh, T. Ortin, D. Roest, A. Van Proeyen, New formulations of D = 10 supersymmetry
and D8–O8 domain walls, Class. Quantum Grav. 18 (2001) 3359, hep-th/0103233.
[78] J. Maharana, J.H. Schwarz, Noncompact symmetries in string theory, Nucl. Phys. B 390 (1993) 3, hep-
th/9207016.
[79] P. Meessen, T. Ortin, An Sl(2, Z) multiplet of nine-dimensional type II supergravity theories, Nucl. Phys.
B 541 (1999) 195, hep-th/9806120.
[80] E. Witten, Dimensional reduction of superstring models, Phys. Lett. B 155 (1985) 151.
[81] S. Ferrara, C. Kounnas, M. Porrati, General dimensional reduction of ten-dimensional supergravity and
superstring, Phys. Lett. B 181 (1986) 263.
[82] E. Witten, New issues in manifolds of SU(3) holonomy, Nucl. Phys. B 268 (1986) 79.
[83] C.P. Bachas, Lectures on D-branes, hep-th/9806199.
[84] R.O. Wells, Differential Analysis on Complex Manifolds, Springer, New York, 1980.
[85] S.S. Gubser, Supersymmetry and F-theory realization of the deformed conifold with three-form flux, hep-
th/0010010.
[86] L. Andrianopoli, R. D’Auria, S. Ferrara, Supersymmetry reduction of N -extended supergravities in four
dimensions, JHEP 0203 (2002) 025, hep-th/0110277.
[87] L. Andrianopoli, R. D’Auria, S. Ferrara, M.A. Lledo, Super-Higgs effect in extended supergravity, Nucl.
Phys. B 640 (2002) 46, hep-th/0202116.
[88] M. Grana, J. Polchinski, Gauge/gravity duals with holomorphic dilaton, Phys. Rev. D 65 (2002) 126005,
hep-th/0106014.
[89] J. Polchinski, String Theory, Vols. 1, 2, Cambridge Univ. Press, Cambridge, 1998.
[90] P.K. Townsend, M.N. Wohlfarth, Accelerating cosmologies from compactification, hep-th/0303097.
[91] R. Blumenhagen, B. Körs, D. Lüst, T. Ott, The standard model from stable intersecting brane world
orbifolds, Nucl. Phys. B 616 (2001) 3, hep-th/0107138.
[92] R. Blumenhagen, B. Körs, D. Lüst, Type I strings with F- and B-flux, JHEP 0102 (2001) 030, hep-
th/0012156.
[93] M. Cvetic, G. Shiu, A.M. Uranga, Chiral four-dimensional N = 1 supersymmetric type IIA orientifolds from
intersecting D6-branes, Nucl. Phys. B 615 (2001) 3, hep-th/0107166.
[94] R. Blumenhagen, L. Görlich, T. Ott, Supersymmetric intersecting branes on the type IIA T 6 /Z(4)
orientifold, JHEP 0301 (2003) 021, hep-th/0211059.
[95] R. Blumenhagen, L. Görlich, B. Körs, D. Lüst, Noncommutative compactifications of type I strings on tori
with magnetic background flux, JHEP 0010 (2000) 006, hep-th/0007024;
R. Blumenhagen, L. Görlich, B. Körs, D. Lüst, Magnetic flux in toroidal type I compactification, Fortschr.
Phys. 49 (2001) 591, hep-th/0010198.
[96] C. Angelantonj, I. Antoniadis, E. Dudas, A. Sagnotti, Type-I strings on magnetised orbifolds and brane
transmutation, Phys. Lett. B 489 (2000) 223, hep-th/0007090;
C. Angelantonj, A. Sagnotti, Type-I vacua and brane transmutation, hep-th/0010279.
[97] G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, Intersecting brane worlds, JHEP 0102
(2001) 047, hep-ph/0011132;
G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, D = 4 chiral string compactifications from
intersecting branes, J. Math. Phys. 42 (2001) 3103, hep-th/0011073.
[98] M. Marino, R. Minasian, G.W. Moore, A. Strominger, Nonlinear instantons from supersymmetric p-branes,
JHEP 0001 (2000) 005, hep-th/9911206.
[99] E. Bergshoeff, R. Kallosh, T. Ortin, G. Papadopoulos, κ-symmetry, supersymmetry and intersecting branes,
Nucl. Phys. B 502 (1997) 149, hep-th/9705040.
[100] K. Dasgupta, C. Herdeiro, S. Hirano, R. Kallosh, D3/D7 inflationary model and M-theory, Phys. Rev. D 65
(2002) 126002, hep-th/0203019.
[101] M.J. Duff, R. Minasian, Putting string/string duality to the test, Nucl. Phys. B 436 (1995) 507, hep-
th/9406198.
Can branes travel beyond CTC horizon in Gödel

universe? ✩
Yasuaki Hikida a , Soo-Jong Rey a,b
a School of Physics & BK-21 Physics Division, Seoul National University, Seoul 151-747, South Korea
b School of Natural Sciences, Institute for Advanced Study, Einstein Drive, Princeton, NJ 08540, USA
Abstract
Gödel universe in M-theory is a supersymmetric and homogeneous background with rotation
and four-form magnetic flux. It is known that, as seen in inertial frame of comoving observer, all
geodesics with zero orbital angular momentum orbit inside ‘surface of light velocity’ (CTC horizon).
To learn if other probes can travel beyond the CTC horizon, we study dynamics of M-graviton and, in
particular, M2-brane, whose motion is affected by Lorentz force exerted by the four-form magnetic
flux and by nonzero orbital angular momentum. Classically, we find that both probes gyrate closed
orbits, but diameter and center of gyration depends on sign and magnitude of probe’s energy, charge
and orbital angular momentum. For M2-brane, orbits in general travel outside the CTC horizon.
Quantum-mechanically, we find that wave function and excitation energy levels are all self-similar.
We draw analogy of probe’s dynamics with Landau problem for charged particle in magnetic field.
 2003 Elsevier B.V. All rights reserved.
PACS: 11.25.-w
For us believing physicists, the distinction between the past,

the present, and the future is only an illusion
Albert Einstein
✩
Work supported in part by the KOSEF Interdisciplinary Research Grant 98-07-02-07-01-5 and the KOSEF
Leading Scientist Grant.
E-mail addresses: hikida@phya.snu.ac.kr (Y. Hikida), sjrey@gravity.snu.ac.kr (S.-J. Rey).
0550-3213/$ – see front matter  2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2003.07.010
58 Y. Hikida, S.-J. Rey / Nuclear Physics B 669 (2003) 57–77
1. Introduction
The Gödel universe [1] is a homogeneous spacetime with pressureless matter and
negative cosmological constant, featured by rotation relative to the local inertial frame
associated with each comoving observer, and, as a result of it, no “absolute” time function,1
the latter meaning that the spacetime does not admit a foliation by globally space-like
hyper-surfaces. Because of these difficulties, Gödel universe has brought up many puzzling
issues in Einstein’s general relativity: the spacetime displays closed null curves (CNCs)
and closed time-like curves (CTCs), Cauchy problem is ill-defined, and, for quantum field
theories defined on it, no obvious notion of unitary Hamiltonian evolution exists. These
issues are all concerned with the notion of chronology, on which Hawking has put forward
so-called “chronology protection conjecture” (CPC) [2].
String theory and M-theory admits the Gödel universe as a supersymmetric solution
[3,4], where, quite importantly, requisite 4-form field strengths are also turned on. It is in
this setting that one might hope to gain a better understanding of the above conceptual
puzzles and shed light on an eventual resolution. With such motivations, recently, several
works has revisited the issue of chronology and (stringy version of) CPC [5–9]. Among
them, Boyda et al. [9] claimed that the chronology is well definable once the Gödel universe
is prescribed with the macroscopic holography [10,11]. In effect, their proposal argues for
keeping (a part of) the causal region associated with a comoving observer and replace the
rest (which includes the CTC region) by an observer-dependent holographic screen. Most
recently, Drukker et al. [12] have studied dynamics of BPS supertube (cylindrical D2-brane
supported by fundamental string and D0-brane charges) in Type IIA Gödel universe, and
claimed that the supertube develops an instability in the CTC region despite being a BPS
configuration.
To draw some insight concerning the CTCs in the Gödel universe, in this work, we
introduce a probe M-graviton and M2-brane and study their dynamics. Geodesics in the
Gödel universe was studied previously [9,13]. There, it was noted that null and time-like
geodesic motions trace gyration orbits whose diameter is set by the energy, and most
notably, none of these geodesics never get into the CTC region. An interpretation based
on the result, which seems to be implicit in the holographic proposal of [9], may be that
the CTC region, though it exists, is never reachable by a stationary comoving observer.
An immediate question is whether the feature that geodesic orbits never get into the
CTC region holds also for all other available probes in M-theory. To answer this question,
in this work, we study classical and quantum dynamics of a probe M-graviton and, in
particular, M2-brane in the Gödel universe. The reason why M2-brane might serve as
a useful probe for the question posed is because of the following intuitive argument. In
addition to the rotating spacetime metric, the Gödel universe in M-theory is characterized
by constant 4-form magnetic field strengths, whose strength is set by the strength of the
rotation. M2-brane is an electrically charged object coupled minimally to the 4-form field,
so it would experience the 4-form Lorentz force in addition to the geodesic force acted
upon by the Gödel metric. This means that, under suitable circumstances (which we will
1 Technically, a “time function” is a smooth, differentiable scalar function t (x) such that (∂ t)(∂ m t) > 0
m
everywhere in spacetime.
Y. Hikida, S.-J. Rey / Nuclear Physics B 669 (2003) 57–77 59
spell out explicitly later), the Lorentz force might well cancel the geodesic force. In that
case, M2-brane’s orbit can be larger than the null geodesics, and would eventually be able
to travel into the region where comoving observers perceive CTCs present.
We find it useful to compare the situation with the well-known Landau problem:
dynamics of a charged particle on a homogeneous 2-space (such as sphere S2 , plane R2 ,
and pseudosphere H2 ) under a uniform magnetic field. Take, for definiteness, the BPS
particle m = |q| on a plane R2 . On R2 , geodesics are straight lines. The magnetic field
exerts Lorentz force to the BPS particle, causing it to undergo gyration around the Larmor
orbit, whose radius of curvature is set by initial velocity v as

c
RL = |v|. (1.1)
B
Quantum mechanically, the particle dynamics is describable as that of two-dimensional
simple harmonic oscillator, whose energy spectrum is given in terms of nonnegative
quantum number n as

1 B
EL = Ω L n + , where ΩL = . (1.2)
2 c
As R2 is a homogeneous space, the gyration takes place isomorphically everywhere and its
center can be brought to, say, the center O by Killing transformations. Dynamics on S2 or
H2 is further corrected by curvature-dependent contribution EL = ± 2R1 2 (n + 12 )2 to the
energy spectrum Eq. (1.2), where R is the radius of curvature, but otherwise qualitatively
the same. In particular, gyration orbit is closed for strong magnetic field and its center can
be brought to the center O by Killing transformations.
This paper is organized as follows. In Section 2, we investigate classical dynamics of a
probe brane in one of the simpler Gödel universes in M-theory, G5 × X6 . We study both
M-graviton and M2-brane probes, and show that M2-brane orbit can extend beyond the
CTC horizon. In Section 3, we study quantum dynamics of these probe branes. We find
that the energy spectrum is discrete and is strikingly reminiscent of that of the Landau
problem, Eq. (1.2). Section 4 is devoted to discussion of various points worthy of further
investigation.
2. Classical dynamics
2.1. Set-up
The Gödel universe in M-theory is a family of classical vacua of the form G2n+1 ×
X10−2n [4], where the index n = 1, . . . , 5 refers to the number of rotating planes. The
simplest situation in which multiple planes are rotating simultaneously is G5 × X6 , and we
shall focus our consideration mainly to this case in this work.2 The metric of G5 × X6 is
2 The simplest Gödel universe, G ×X , involving a single rotating plane preserves 8 supersymmetries only, as
3 8
contrasted to those involving multiple rotating planes, which preserve 16 + 4 = 20 supersymmetries (n = 2, 3, 4)
or 16 + 2 = 18 supersymmetries (n = 5).
given by

2
dsM = −(dt + µω)2 + ds 2 R4 + ds 2 (X6 ),
G4 ≡ dC3 = 2µJ ∧ K, (2.1)
where ω is the twist one-form on the spatial slice in G5 , while J, K are Kähler two-
R4
forms on R4 and a six-dimensional ‘internal’ manifold X6 , respectively. Locally, the one-
form ω is related to J as J = dω. The constant µ is a parameter for the simultaneous
rotation of R2 ⊕ R2 ⊂ R4 as well as for the 4-form magnetic flux G4 ≡ dC3 . It can be
set to unity by rescaling proper distance on R4 by 1/|µ|, but, for the consideration of
dimensional analysis and (pseudo)symmetry transformation, we find it useful to retain it
explicitly.
We introduce a probe M2- or M2-brane, which is coupled minimally to the 3-form
potential C3 , and examine classical dynamics of it in the background Eq. (2.1). Classical
dynamics of the probe brane is governed by the Born–Infeld action

S = −T2 − det X g − Q X∗ C3 ,
∗ (2.2)
Σ3 Σ3
where X∗ G, X∗ C3 are the pull-backs of the metric and the 3-form potential in Eq. (2.1) on
the worldvolume of M2-brane. Noether charge of the M2-brane is denoted as Q, measured
in unit of the M2-brane tension T2 . Thus, for a BPS M2- or M2-brane, Q = T2 or Q = −T2 ,
respectively.
For definiteness, consider a M2-brane wrapped on a two-cycle Σ in X6 .3 Rigid
dynamics of the M2-brane is describable by dimensionally reducing Eq. (2.1) to five-
dimensional Gödel universe G5 threaded with the magnetic flux G2 ≡ Σ G4 :

ds52 = −(dt + µω)2 + ds 2 R4 ,
√
G2 ≡ dC1 = 2 3 µJ, (2.3)
and the M2-brane reduces to a charged point-particle of mass m = T3 vol(Σ) and electric
charge q = Qvol(Σ), coupled minimally to the ‘vector’ potential C1 . From Eq. (2.2),
the effective worldline action of the 5-dimensional charge particle is given in local
reparametrization invariant way as

1 −1 q
S = dτ e gmn (x)ẋ m ẋ n − em2 − √ Cm (x)ẋ m (i = 0, 1, . . . , 4), (2.4)
2 2 3
where ẋ m ≡ dx m /dτ . We solve the equation of motion for x m (τ ):
D ẋ m ė m q
− ẋ + e √ Gm
2 n ẋ = 0,
n
Dτ e 2 3
where D/Dτ refers to the covariant derivative with respect to the Christoffel connection
m of the metric Eq. (2.3). Equation of motion for the worldline einbein e(τ ) is
Γnp

e(τ ) = m−1 −gmn (x)ẋ m ẋ n . (2.5)
3 The analysis of [4] indicates that this is a supersymmetric configuration.

The worldline action Eq. (2.4) is invariant under local reparametrization invariance, τ →
f (τ ) and e → e/|f˙(τ )|, and we fix it by choosing the gauge e = +1, so that τ is the affine
parameter of the worldline. The gauge-fixed equations of motion for x m (τ ) are
D 2 ẋ m q
+ √ Gm
2 n ẋ = 0,
n
(2.6)
Dτ 2 3
subject to the constraint Eq. (2.5).
To proceed further, we find it convenient to express the metric Eq. (2.3) in the bipolar
coordinates (see, e.g., Eq. (3.7) in [9]) as:

ds52 = −dt 2 − 2µ r12 dφ1 + r22 dφ2 dt − 2µ2 r12 r22 dφ1 dφ2

+ r12 1 − µ2 r12 dφ12 + r22 1 − µ2 r22 dφ22 + dr12 + dr22 , (2.7)
and the one-form connection C1 in the Landau gauge as:
√
G2 ≡ dC1 = 2 3 µ d r12 ∧ dφ1 + r22 ∧ dφ2 . (2.8)
We refer the origin of the coordinate system as O. Notice that the bipolar coordinates
Eq. (2.7) displays
explicitly that the signature of φ1,2 line element flips sign across
the hyper-surface r12 + r22 = 1/|µ|. Thus, at the hyper-surface, paths around φ1 or φ2
are CNCs, and, beyond the hyper-surface, they are CTCs. As such, we will refer these
codimension-one hyper-surfaces located at
1
RCTC = (2.9)
|µ|
as “CTC horizon”. We emphasize again that the “CTC horizon” is a notion set forth
specifically by each comoving observer, whom we have put conveniently at the origin O.
As Gödel universe is a homogeneous space, comoving observers placed at any location on
G5 are all equivalent and can be always brought to the origin O.
Using explicit form of the Christoffel connections as recollected in the appendix, we
find the equations of motion Eq. (2.6) are given by
t¨ + 2µ2 r1 ṙ1 t˙ + 2µ3 r13 φ̇1 ṙ1 + 2µ3 r1 r22 φ̇2 ṙ1 + 2qµ2 r1 ṙ1 + (1 ↔ 2) = 0, (2.10)

r̈1 + 2µr1 φ̇1 t˙ − r1 1 − 2µ2 r12 φ̇12 + 2µ2 r1 r22 φ̇1 φ̇2 + 2qµr1 φ̇1 = 0, (2.11)

2µ 1 2µ2 r22 ṙ1
φ̈1 − ˙
ṙ1 t + 2 − µ r1 φ̇1 ṙ1 −
2
φ̇2 ṙ1 − 2qµ = 0, (2.12)
r1 r1 r1 r1
and similar ones with (1 ↔ 2) for the latter two equations of motion.
The Gödel universe Eq. (2.3) or, equivalently, Eqs. (2.7), (2.8) exhibit the following
symmetries:
Pµ : ω → −ω and µ → −µ, (2.13)

PT: t → −t, µω → −µω and q → −q. (2.14)
Notice that ω → −ω in Eq. (2.1) amounts in bipolar coordinates to parity inversion,
φ1,2 → −φ1,2 , in each R2 -plane. Thus, Eq. (2.13) acts as parity inversion accompanied
by inversion of the rigid rotation, while Eq. (2.14) does as simultaneous action of
the time-reversal, parity-inversion, and charge-conjugation. We will find later these
(pseudo)symmetries serve useful for understanding classical and quantum dynamics of
the probe objects.
As the metric and the magnetic field strength on G5 are functions of r1 , r2 only, −∂t ,
∂φ1 , ∂φ2 are Killing vectors. Accordingly, the canonical momenta conjugate to −t, φ1 and
φ2 are conserved, first integrals of motion. They are

E = t˙ + µ r12 φ̇1 + r22 φ̇2 ,

L1 = r12 φ̇1 − µ2 r12 r12 φ̇1 + r22 φ̇2 − µr12 (t˙ + q),

L2 = r22 φ̇2 − µ2 r22 r12 φ̇1 + r22 φ̇2 − µr22 (t˙ + q). (2.15)
Inverting these relations,
L1
φ̇1 = + µ(E + q), (2.16)
r12
L2
φ̇2 = 2 + µ(E + q), (2.17)
r2

t˙ = E − µ2 r12 + r22 (E + q) − µ(L1 + L2 ). (2.18)
In these equations, the first terms are standard. The second term in Eq. (2.18) reflect frame
dragging rotation of the Gödel universe. The last term in Eq. (2.18), depending on L1 , L2 ,
is a sort of ‘spin–orbit’ coupling between frame-dragging rotation and probe’s orbital
angular momenta.
Recall that, using homogeneity of G5 , any location can be brought to the origin O
(r1 = r2 = 0 in the bipolar coordinates), Eq. (2.7) by a sequence of Killing transformations.
Thus, we shall be considering a comoving observer located at the origin O, and let the
observer perform experiments for exploring causal structure of the Gödel universe by
sending off the probe objects available such as M-graviton or M2-branes. We will study
first the probes with zero angular momenta, and then those with nonzero angular momenta.
In both cases, wherever relevant, we draw close analogy with the well-known results of the
Landau problem Eqs. (1.1), (1.2).
2.2. Probes with zero angular momenta
We will first consider L1 = L2 = 0 case. In this case, there is no centrifugal barrier,

and the probe brane passes through the origin, where the comoving observer is located. We
then obtain the radial equations of motion as exerting harmonic oscillations isomorphically
on each R2 -plane:
r̈1 + µ2 (E + q)2 r1 = 0 and r̈2 + µ2 (E + q)2 r2 = 0

subject to the gauge-fixing constraint, coupling the motion on the two R2 -planes,
2
ṙ1 + (E + q)2 µ2 r12 + ṙ22 + (E + q)2 µ2 r22 = E 2 − m2 . (2.19)
With the initial condition r1 (0) = r2 (0) = 0 that the probe starts from comoving observer’s
location O, the solution is
φ1 (τ ) = µ(E + q)τ + φ1(0) ,

(0)
φ2 (τ ) = µ(E + q)τ + φ2 ,
√
E 2 − m2
r1 (τ ) = sin |µ(E + q)|τ cos θ,
|µ(E + q)|
√
E 2 − m2
r2 (τ ) = sin |µ(E + q)|τ sin θ,
|µ(E + q)|

1 E 2 − m2 1 E 2 − m2
t (τ ) = E − τ+ sin 2|µ(E + q)|τ + t (0) . (2.20)
2 E+q 4µ (E + q) 2
Here, θ (0 θ π/2) parametrizes projection of the motion to each R2 -plane,

(0) (0)
φ1 , φ2 , t (0) are parameters fixed by initial condition, and the conserved energy E,
Eq. (2.15), ranges over E m or E −m (to yield physically meaningful solution
satisfying Eq. (2.19)). In our convention, E, m and q carries mass dimension 1, and
BPS conditions set inequalities |E| m |q|. We thus see that probe’s trajectory
is unidirectional: angular velocities φ̇1 (τ ), φ̇2 (τ ) revolve the same direction as the
background Gödel universe Eq. (2.7) for sgn (µ(E + q)) > 0 and opposite direction for
sgn (µ(E + q)) < 0, respectively.4 Under PT-conjugation of Eq. (2.14), the antiprobe
follows the same trajectory but in opposite directions.
The motion is reminiscent of the Landau problem mentioned earlier. As evident from
Eq. (2.20), probe’s trajectory traces a circular orbit on each R2 -plane, as depicted in Fig. 1.
One may describe the orbit with respect to the center of the orbit C (instead of comoving
observer’s location O). The gyration diameter D = 2R, as depicted in Fig. 1, is set by the
conserved energy
√
E 2 − m2
D = 2R = RCTC , (2.21)
|E + q|
but is constant otherwise. Using trigonometry, gyration velocity v around C (as measured
in affine time) is also determined as

v = R · 2φ̇1 = R · 2φ̇2 = sgn µ(E + q) E 2 − m2 . (2.22)
2.2.1. M-graviton
For a neutral probe, q = 0, corresponding to Kaluza–Klein modes of the M-theory
graviton, Eq. (2.20) reduces to the geodesics studied in [9,13].
• For the massless mode, m = 0, the geodesics start out radially from the origin O, and
sweeps out a circular gyration orbit, touching the CTC horizon RCTC associated with
4 This feature is also reflected in the quantum dynamics of the probes. See Section 3.
comoving observer located at O. The

Fig. 1. Trajectory of the probe with zero angular momenta, passing through
probe traces the Larmor orbit, whose radius is R and gyration speed is E 2 − m2 .
the comoving observer at O. That is,
r1 (τ ), r2 (τ ) RCTC ,
as depicted as the leftmost circles in Fig. 2. To reach the CTC horizon, φ1 , φ2 should
sweep out π/2, so it takes affine time τ = (π/2|µE|). Consequently, a complete
revolution around the orbit takes τ = (π/|µE|) in affine time, and hence from
Eq. (2.20), t = (π/2|µ|) in comoving observer’s time. Notice that the gyration orbit
is independent of graviton’s energy E. One can see this already from Eq. (2.19)—
both the harmonic frequency in the left-hand side and the constant of motion in the
right-hand side are proportional to E 2 , and can be rescaled away.
• For the massive modes, the geodesics extend radially only up to the distance

m2
r1 , r2 RCTC 1 − < RCTC
E2
viz. the projected orbit has a radius smaller than the CTC horizon RCTC (recall from
Eq. (2.9) that the causal region as perceived by the comoving observer is separated
from the CTC region by the CTC horizon r1,2 = RCTC ). It follows that geodesics
of M-theory graviton in the Gödel universe lies entirely within the causal region
[9,13].5 Otherwise, the geodesics behave essentially the same as the massless ones.
In particular, it takes exactly the same amount of affine and comoving time for the
gyration to undergo a complete revolution.
5 Taking up this observation, in [9], it was proposed to cut off the region outside the CTC horizon, paste it
with flat spacetime, and place the holographic screen inside the causal region, where the area of the light-sheet as
prescribed in [10,11] is maximized.
Fig. 2. Trajectories of various probes. The inner circle refers to the CTC horizon as seen in the inertial frame
of comoving observer at the center, outside which closed time-like curves are present. Geodesics of M-graviton,
shown in the left for three different angular initial conditions, are confined inside the CTC horizon. Trajectory of
M2-branes with E > 0 are shown in the right for q > 0 and q < 0, respectively. Notice that q > 0 orbit is always
inside the CTC horizon, while q < 0 orbit extends beyond the CTC horizon.
2.2.2. M2-branes
For q = 0, corresponding to black M2-brane, the trajectory Eq. (2.20) entails several
new features. Because of the background magnetic field, the M2-branes, being an
electrically charged probe, experiences the Lorentz force in addition to the force exerted by
the Gödel universe rotation. Recalling that the rotational force is unidirectional, depending
on the sign of M2-brane charge, the Lorentz force may act to add up to or to cancel off the
rotational force. It clearly points to the possibility that the comoving observer can arrange
a probe brane that travels beyond the CTC horizon and comes back. Let us analyze how
this may happen.
Eq. (2.20) indicates that the net effect depends crucially on the relative sign between the
energy E and the charge q. Consider an M2-brane with E > m, orbiting through comoving
observer’s location O. Diameter D of the orbit is given by Eq. (2.21), so let us examine
how D depends on E and q. We are particularly interested in whether D could be larger
than RCTC for appropriate values of E and q. We see from Eq. (2.20) that, for BPS M2-
brane with positive charge q = m,

Dsub = RCTC (E − m)/(E + m) < RCTC for sgn(Eq) > 0,
so the orbit gyrates entirely inside the CTC horizon of the comoving observer O. We
will refer the trajectory as “subgyration orbit” of diameter Dsub . For BPS M2-brane with
negative charge q = −m,

Dsuper = RCTC (E + m)/(E − m) > RCTC for sgn(Eq) < 0,
so, as observed by the comoving observer O, the orbit passes through the CTC horizon
and gyrate into the region where CTCs are present! This is in stark contrast to the situation
of M-graviton geodesics studied in the previous section. We will refer the trajectory as
“supergyration orbit” of diameter Dsuper . The two types of behavior are depicted as the
rightmost two orbits in Fig. 2.
Information concerning M2-brane’s dynamics for E < −m are obtainable by applying
the PT conjugation Eq. (2.14). One finds that the two types of behavior are interchanged:
with E < −m, BPS M2-brane with positive charge q = m traces “supergyration orbit” and
extends beyond the CTC horizon, while M2-brane with negative charge q = −m traces
“subgyration orbit” and remains inside the CTC horizon.
Let us draw further comparison between “subgyration orbits” and “supergyration
orbits”. We consider the positive-energy branch E > m only, as the negative-energy branch
E < −m is obtainable by PT conjugation of the positive-energy branch.
• Diameter of the subgyration orbit is a monotonic increasing function of E, ranging

over 0 < Dsub < RCTC . In contrast, diameter of the supergyration orbit is a monotonic
decreasing function of E, ranging over RCTC < Dsuper < ∞. Interestingly, in both
cases, orbits at extreme high-energy limit accumulate to the CTC horizon, viz.
the ‘velocity of light surface’, and the gyration velocity becomes infinite. In other
words, in so far as dynamics of M-theory probes is concerned, the CTC horizon
defines a universal infinite-momentum light-front frame, as gyration orbits of both
the M-graviton and M2-brane asymptote all to it as the ‘velocity of light’ surface. This
suggests a viable microscopic holography of the Gödel universe in M-theory [14].
• From t-velocity Eq. (2.18), we see that, for L1,2 = 0 under consideration, the
comoving observer at O would draw the standard interpretation of forward time-
flow for E > m and backward time-flow for E < −m. Far away from O, however,
Eq. (2.18) indicates that the time-flow would be seen reversed. Using the result
Eq. (2.20), one finds that the reversal takes place at a distance rr :

E
rr = RCTC .
E+q
Comparing this with the diameter D of the gyration orbit Eq. (2.21), we find that
M2-branes whose energy and charge have same sign would never reverse the time-
flow, but M2-branes whose energy and charge have opposite sign would do so. At
√ of time-flow reversal, angular position φ1,2 of the probe M2-brane is at
the moment
sin φr = E/(E + |q|).
2.3. Probes with nonzero orbital angular momenta
So far, we assumed that the probe carries no angular momentum. In case the angular
momenta L1 , L2 are nonzero, new intriguing features arise.
• Probe’s orbits all migrate off the location of the comoving observer, and, if the orbital
angular momentum meets a suitable condition, penetrate into the CTC region. This is
readily seen as follows. Substitute Eqs. (2.16), (2.17) and (2.18) into (2.5) with e = 1.
We then obtain

(ṙ1 )2 + (ṙ2 )2 = E 2 − m2 − Veff (r1 , r2 ),
where
2 2
L1 L2
Veff (r1 , r2 ) = + µr1 (E + q) + + µr2 (E + q) . (2.23)
r1 r2
The gyration orbits now range over Veff (r1 , r2 ) (E 2 − m2 ), so

2
L1
+ µr1 (E + q) E 2 − m2 cos2 θ,
r1
2
L2
+ µr2 (E + q) E 2 − m2 sin2 θ.
r2
Consider the positive-energy branch, E > m.6 We see from the above inequalities
that not all angular momenta are physically allowed. Those with L1 , L2 < 0 are
always possible, but those with L1 , L2 > 0 cannot be arbitrarily large. Rather, angular
momenta are bounded by
1
−∞ < (L1 , L2 ) < Lmax (cos θ, sin θ ), where Lmax ≡ RCTC (E − q). (2.24)
4
In this case, the classical turning points are

(±) D L1
r1 = cos θ ± cos2 θ − ,
2 Lmax

(±) D L2
r2 = sin θ ± sin θ − 2
. (2.25)
2 Lmax
• Intuitively, one expects that orbital angular momentum affects the orbits similar to
the Lorentz force. The actual effect is rather interesting. For L1 , L2 > 0, the gyration
center remains unchanged from that for zero angular momentum, viz. located when
projected on each R2 -plane at distance 12 D cos θ and 12 D sin θ , respectively. On the
otherhand, the diameter of eachorbit is affected. On each R2 -plane, the diameter
is D cos2 θ − (L1 /Lmax ) and D sin2 θ − (L2 /Lmax ), respectively. So, the gyration
orbits shrink once the orbital angular momentum is applied to the probe same direction
as the rotation of the Gödel universe background. At and above the critical angular
momenta Eq. (2.24), the orbits cease to exist. For L1 , L2 < 0, the effects are reversed:
the orbit diameter remains unchanged from that for zero orbital angular momenta, viz.
projected on each R2 -plane, it is D cos θ and D sin θ , respectively, but the gyration
center is shifted further away from the comoving observer O. So, once the orbital
angular momentum is applied to the probe the opposite direction as the rotation of
the Gödel universe background, the orbit slides off from the hand of the comoving
observer. In particular, as perceived by the comoving observer at O, the orbit passes
through the CTC horizon and travels the CTC region.
• The time-flow t˙ is now modified further by the ‘spin–orbit coupling’, the last term in
Eq. (2.18), and can be reverted, similar to the situation with the Lorentz force. This
is evident for L1 , L2 < 0: if magnitude of L1 , L2 is large enough, gyration center
migrates outside the CTC horizon, causing the second term in Eq. (2.18) to dominate
6 Again, negative-energy branch, E < −m, is deducible by applying PT-conjugation, under which L
1,2 →
−L1,2 as well, to the following consideration.
over the other terms. For L1 , L2 > 0, however, the coupling cannot revert the time-
flow due in part to the limit Eq. (2.24) and the fact that the orbit stays inside the CTC
horizon.
2.4. IIB string theory setup
Instead of uplifting to M-theory, equivalently, to Type IIA string theory, one can
uplift the five-dimensional Gödel universe G5 to Type IIB string theory, as originally
demonstrated in [6]. The Type IIB background is given by
2
1
2
dsIIB = gmn dx m dx n + dy + √ Am dx m + ds 2 (X4 ), (2.26)
2 3

1 1 1
H3 = dC2 = √ dA ∧ dy + √ A − √ ∗ dA. (2.27)
2 3 2 3 2 3
Here, the ten-dimensional coordinates are represented as (x m , y, za ) with m = 0, . . . , 4
and a = 1, . . . , 4. The metric gmn and the gauge connection A are those of maximally
supersymmetric, five-dimensional Gödel universe. The Hodge dual ∗ is defined with
respect to the five-dimensional metric gmn .
For the M-theory uplift, the probe that can couple to the background magnetic flux in the
five-dimensional Gödel universe was the M2-brane wrapped on two-cycle Σ2 in the six-
dimensional ‘internal’ space X6 . For the Type IIB string theory uplift, a viable probe that
can couple to the background 3-form flux Eq. (2.27) in the five-dimensional Gödel universe
is the Kaluza–Klein mode along the 6th dimension (whose coordinates are labelled as y in
Eq. (2.26)).
Consider null geodesics in 6-dimensional Gödel universe Eq. (2.26). Repeating the same
analysis as in Section 2.1, one finds the conserved, first integrals of motion:

E = t˙ + µ r12 φ̇1 + r22 φ̇2 , Py = −ẏ − µ r12 φ̇1 + r22 φ̇2 ,
L1 = r12 φ̇1 − µr12 (t˙ − ẏ), L2 = r22 φ̇2 − µr22 (t˙ − ẏ).
One then finds that solution of the null geodesics is precisely the same as Eq. (2.20)
provided the conserved charge Py is equated with the electric charge q of the wrapped
M2-brane.
3. Quantum dynamics
In the previous section, we have seen that classical motion of the probe, for both
M-graviton and M2-brane, all trace gyration orbits, much similar to the Landau problem. It
thus brings up an issue whether, quantum mechanically, probe’s excitation energy spectrum
would exhibit discrete spectrum analogous to Eq. (1.2). In this section, we will find that
the expectation is met precisely. We show that the energy spectrum is precisely the same
as that of two copies of the two-dimensional simple harmonic oscillators, each associated
with the two rotating R2 -planes. In doing so, we will pay particular attention to the energy
spectrum of the ‘supergyration orbits’, which in classical analysis encompassed outside the
CTC horizon. By comparing the energy spectrum with those for geodesics or ‘subgyration
orbit’, one would hope to learn, if any, pathologies associated with the CTCs. Our result
shows that both the energy spectrum and wave function of all probes are self-similar.
3.1. Graviton
We begin with the M-theory graviton. Taking, for simplicity, the polarization entirely
along R6 direction, the field equation of a 5-dimensional graviton Φ is given by the
massless Klein–Gordon equation in the background Eq. (2.7)
1 √
5 Φ = √ ∂m g5mn −g5 ∂n Φ = 0. (3.1)
−g5
As the Gödel metric depends only on r1 and r2 , it is possible to decompose the field Φ
into basis χ ’s via separation of variables as [15]

Φ = χ, χ ∗ , where χ = N u1 (r1 )u2 (r2 ) exp(iL1 φ1 + iL2 φ2 − iEt), (3.2)
where N is a normalization factor. The polar coordinates φ1,2 range over [0, 2π], so
the angular momenta L1 , L2 are quantized to integer units. The Klein–Gordon equation
Eq. (3.1) is then reduced to two coupled equations:

1 L1 2
−∂r21 − ∂r1 + µEr1 + − E 2 cos2 θ u1 (r1 ) = 0,
r1 r1

1 L2 2
−∂r22 − ∂r2 + µEr2 + − E 2 sin2 θ u2 (r2 ) = 0. (3.3)
r2 r2
Here, 0 θ π/2 is a parameter introduced for separation of the variables r1 , r2 .
Eq. (3.3) takes the form of Schrödinger equation of two-dimensional simple harmonic
oscillators whose natural frequency is given by |µE| and energy by {E 2 cos2 θ − 2µEL1 }
or {E 2 sin2 θ − 2µEL2 }, respectively. Normalizable solutions are given in terms of the
associated Laguerre polynomial Lαn :

|L | 1 (|L |)
u1 (r1 ) = r1 1 exp − |µE|r12 Ln1 1 |µE|r12 ,
2

|L | 1 (|L |)
u2 (r2 ) = r2 2 exp − |µE|r22 Ln2 2 |µE|r22 , (3.4)
2
where n1 , n2 are nonnegative integers related to other quantum numbers as
1 1 2
n1 = − 1 + |L1 | + E cos2 θ − 2µEL1 ,
2 4|µE|
1 1 2 2
n2 = − 1 + |L2 | + E sin θ − 2µEL2 . (3.5)
2 4|µE|
Adding the two radial quantum number relations in Eq. (3.5), we get

E 2 = 4|µE|(n1 + n2 + 1) + 2 µEL1 + |µEL1 | + 2 µEL2 + |µEL2 | .
Fig. 3. Energy spectrum of massless (left) and massive (right) M-theory gravitons. Horizontal axis refer to
different multiplicities associated with n1 , n2 , m1 , m2 . The energy gap is denoted as E0 = 4|µ|.
From the relation, we find that the quantum energy spectrum E of the M-theory graviton
in a remarkably simple analytic form:
|E| = 4|µ|N0 , (3.6)
where

1 1
N0 = n1 + n2 + 1 + |L1 | + sgn(Eµ)L1 + |L2 | + sgn(Eµ)L2 .
2 2
Thus, positive and negative branches of the energy spectrum are labelled by four non-
negative integers n1 , n2 , m1 , m2 7
(+) (+)
E (+) = +4|µ| n1 + n2 + m1 + m2 + 1 ,

E (−) = −4|µ| n1 + n2 + m(−) (−)
1 + m2 + 1 , (3.7)
(+) (−)
where m1,2 = (|L1,2 | + sgn(µ)L1,2 )/2 and m1,2 = (|L1,2 | − sgn(µ)L1,2 )/2, respectively.
Notice that the two branches Eq. (3.7) of the graviton spectrum are related each other by
the PT conjugation in Eq. (2.14). The energy spectrum and degeneracy is depicted in Fig. 3.
A few remarks are in order.
7 The spectrum Eq. (3.7) is in agreement with the supergravity modes of the closed string spectrum in Gödel
universe [4].
• For massive M-graviton with mass m, the energy spectrum is obtainable analogously
in terms of N0 in Eq. (3.6) as:

1 m2 1/2
E = ±2|µ| N0 + N0 +
(±) 2
.
4 µ2
In Fig. 3, we depict modification of the M-graviton spectrum due to nonzero mass.
Notice that the energy spectrum is governed by the same orbital angular momentum
(±) (±)
quantum numbers m1 , m2 as the massless one.
• Recall that geodesics of M-graviton are closed orbits around the CTC horizon,
and never travel through the CTCs. The spectrum Eq. (3.6) originates from Bohr–
Sommerfeld quantization of the M-graviton wave function around the classical
gyration orbit studied in Section 2, so it is the counterpart of the Landau level
spectrum Eq. (1.2). Orbital angular momentum of the M-graviton does not alter the
interpretation, as L1 , L2 are conserved quantum numbers and renders the M-graviton
wave equation completely separable. Notice that M-graviton’s radial wave function,
which determines the spatial spread-out of the M-graviton, is symmetric under reversal
of the orbital angular momenta, L1,2 → −L1,2 , while the energy spectrum Eq. (3.7) is
not.
• The n1 , n2 quantum numbers are interpretable as labelling ‘excitation’ above the BPS
ground state of orbiting M-graviton. The zero-point energy proportional to 4|µ| in
Eq. (3.7) may be attributed to the fact that the null orbit is closed on each R2 -plane.
The frequency |µ| in Eq. (3.7) is the parameter that sets the rotation and, in turn, the
CTC horizon in the Gödel universe, so it is the counterpart of the Larmor frequency
ΩL in Eq. (1.2). Apparently, quantization of the energy spectrum has nothing to do
with the presence of CNCs and CTCs in Gödel universe.
• The angular momentum quantum numbers L1 , L2 run over all integer values, both
positive and negative. Yet, the spectrum Eq. (3.7) is affected only when the orbital
angular momentum quantum numbers take positive (negative) values, respectively, for
positive (negative) energy. It implies that, for positive (negative) energy and negative
(positive) orbital angular momenta, the spectrum Eq. (3.7) is infinitely degenerate.
This is a direct reflection of orbit’s classical behavior with nonzero orbital angular
momenta. Consider a classical orbit with positive (negative) energy. As explained
in Section 2.3, if positive (negative) orbital angular momenta are applied to the
M-graviton, orbit’s diameter shrinks accordingly. It implies that, by Bohr–Sommerfeld
quantization rule, quantum-mechanical energy ought to depend on the orbital angular
momentum quantum numbers. For a fixed ‘excitation’ energy E of the M-graviton, the
orbital angular momentum quantum numbers cannot exceed the maximum
E
max(m1 , m2 ) = ,
4|µ|
and this is precisely the same upper bound Lmax as the classical counterpart Eq. (2.24)
(for q = 0). Instead, if negative (positive) orbital angular momenta are applied to the
M-graviton, orbit’s diameter remains the same. By Bohr–Sommerfeld quantization
rule, it implies that quantum-mechanical energy ought not to depend on the orbital
angular momentum quantum numbers. Rather, it renders the energy level degeneracy
infinite, reflecting the fact that orbit’s gyration center can migrate off all over each
R2 -plane once the orbit angular momenta are cranked up arbitrarily high.
3.2. M2-brane
Let us turn to the dimensionally reduced M2-brane. Interaction of the M2-brane with
√ field G2 = dC1 is facilitated by the minimal coupling prescription
the background gauge
∂m → ∂m + i(q/2 3)Cm in Eq. (3.1). In the Landau gauge Eq. (2.8),
√ √
Cφ1 (r1 ) = 2 3 µr12 and Cφ2 (r2 ) = 2 3 µr22 ,
and the charged and massive counterpart of Eqs. (3.3) become
2
1 L 1
−∂r21 − ∂r1 + (E + q)µr1 + − E 2 − m2 cos2 θ u1 (r1 ) = 0,
r1 r1

1 L2 2 2 2
−∂r2 − ∂r2 + (E + q)µr2 +
2
− E − m sin θ u2 (r2 ) = 0.
2
r2 r2
The solution is again obtainable in terms of the associated Laguerre polynomials as

|L | 1 (|L |)
u1 (r1 ) = r1 1 exp − |µ(E + q)|r12 Ln1 1 |µ(E + q)|r12 ,
2

|L | 1 (|L |)
u2 (r2 ) = r2 2 exp − |µ(E + q)|r22 Ln2 2 |µ(E + q)|r22 , (3.8)
2
with nonnegative integer quantum numbers
1 1
n1 = − 1 + |L1 | + E 2 − m2 cos2 θ − 2µL1 (E + q) ,
2 4|µ(E + q)|
1 1
n2 = − 1 + |L2 | + E 2 − m2 sin2 θ − 2µL2 (E + q) .
2 4|µ(E + q)|
It then follows that the dispersion relation is given by
E 2 − m2 = 4|µ(E + q)|N2 , (3.9)

where

1
N2 ≡ n1 + n2 + 1 + |L1 | + sgn µ(E + q) L1
2

1
+ |L2 | + sgn µ(E + q) L2 . (3.10)
2
The right-hand side of Eq. (3.9) is positive-definite, and yields the two branches of the
energy spectrum
E (+) : E > m and E (−) : E < −m. (3.11)

For BPS M2-brane with q = ±m, Eq. (3.9) is reduced to E = q ± 4|µ|N . Keeping again
track of sign of the orbital angular momentum quantum numbers, we find that the energy
spectrum is given by
(+) (+)
E (+) = q + 4|µ| n1 + n2 + m1 + m2 + 1 ,
(−) (−)
E (−) = q − 4|µ| n1 + n2 + m1 + m2 + 1 , (3.12)
(±)
subject to the condition Eq. (3.11). Here, m1,2 ≡ (|L1,2 | ± sgn(µ)L1,2 )/2 = 0, 1, 2, . . . .
Remarkably, apart from the rest mass of the M2-brane, the resulting energy spectrum is
identical to that of the massless M-graviton studied in the previous subsection. Notice that
the energy spectra, E (+) and E (−) , are mapped to each other under the PT conjugation
Eq. (2.14).
Several remarks are in order.
• As for the M-graviton, the energy spectrum Eq. (3.12) depends sharply on the relative
sign between the energy and the orbital angular momenta. With positive (negative)
orbital angular momenta of M2-brane, orbit’s diameter shrinks, so the energy spectrum
depends on the quantum numbers, m1 , m2 . For a fixed ‘excitation’ energy E of the
M2-brane, these quantum numbers cannot exceed the maximum
(E − q)
max(m1 , m2 ) = ,
4|µ|
and this is precisely the upper bound Lmax of the classical counterpart Eq. (2.24) (for
q = 0). With negative (positive) orbital angular momenta of the M2-brane, classical
orbit’s diameter remained unaffected, and the energy spectrum is independent of
m1 , m2 quantum numbers. Instead, the energy levels become infinitely degenerate,
reflecting classical result that orbit’s gyration center can migrate off all over each R2 -
plane once M2-brane’s orbit angular momenta are cranked up arbitrarily high.
• Classically, we learned in Section 2 that, unlike M-graviton, M2-brane probe can pass
through the CTC horizon and gyrate around in the region where the comoving observer
at O perceives CTCs, even for zero orbital angular momenta. Quantum mechanically,
wave function and hence probability density of the two probes Eqs. (3.4) and (3.8),
respectively, are self-similar. The two are merely related each other√by (energy-
dependent) rescaling radial coordinates by an energy-dependent factor 1 + (q/E).
Moreover, for both, the energy scale of the excitation spectrum is set by one and the
same Larmor frequency, 4|µ|. It may be that, in order for the quantum mechanics
reveal full-fledged pathologies of the CTCs present, more refined dynamical processes,
e.g., interference of multi-body wave functions or nonlinear mode–mode interactions,
should be considered.
• Recall that, in the foregoing analysis, we have tacitly assumed that the orientation of
the two-cycles M2-brane wraps on is such that the M2- and M2-branes carry electric
charge q positive and negative, respectively. Upon reversing orientation of the two-
cycles,8 the sign of the electric charge q would flip. All the foregoing analysis would
be the same except that q should be taken the opposite. Still, as E > m and E < −m,
the spectrum is isomorphic to the above.
3.3. IIB string theory setup
By taking similar steps, one can find excitation energy spectrum of IIB-graviton probe
by solving the six-dimensional wave equation, 6 Φ = 0, in the background Eq. (2.26).
Making use of the conserved quantum numbers identified in Section 2.4, we again take
separation of variables as

Φ = χ, χ ∗ , χ = N u1 (r1 )u2 (r2 ) exp(iL1 φ1 + iL2 φ2 + iPy y − iEt).
Radial parts of the wave equation are then given by

1 L1 2 2
−∂r21 − ∂r1 + (E + Py )µr1 + − E − Py2 cos2 θ u1 (r1 ) = 0,
r1 r1

1 L2 2 2 2
−∂r2 − ∂r2 − (E + Py )µr2 +
2
− E − Py sin θ u2 (r2 ) = 0.
2
r2 r2
(3.13)
One finds that they are the same as the ones for the M-graviton and the M2-brane for
Py = 0 and Py = q, respectively, so the wave function and energy spectrum are exactly the
same as those.
4. Discussion
In this paper, we have studied classical and quantum dynamics of M-graviton and M2-
brane probes in Gödel universe, with particular attention to possible effects of the CTCs
present beyond the CTC horizon of a comoving observer. Our results are summarized as
follows.
• On each rotating R2 -plane, all probes trace gyration orbits, much similar to the Landau
problem, Eq. (1.1). For each orbit, gyration center and diameter depends on probe’s
conserved quantum numbers: energy E, angular momenta L, and 3-form potential
charge q.
• For zero orbital angular momenta, orbit of M-graviton (geodesics) traces radial
distance between a comoving observer located at origin O and the CTC horizon RCTC
perceived by the comoving observer. Orbit of M2-brane traces between a comoving
observer at O and the maximal distance D (diameter of the orbit). For sgn(Eq) > 0,
D is less than RCTC , so as perceived by the comoving observer, the orbit does not
8 Orientation of the normal directions needs to be reversed concurrently so that the net orientation of X is
6
unaffected.
travel through the region where CTCs are present. For sgn(Eq) < 0, D is larger than
RCTC , and the orbit travels through the region where CTCs are present.
• Nonzero orbital angular momenta L modify probe’s orbit as well. For sgn µL > 0, the
gyration takes place at the same center as L = 0 but the gyration diameter is shrunken.
For sgn µL < 0, the gyration diameter is the same as L = 0 but the center migrates
away from the comoving observer.
• Quantum mechanically, probe’s wave functions and excitation energy spectrum are all
self-similar, independent of whether probe’s classical orbit is larger or smaller than the
CTC horizon. Interestingly, under the reversal of the orbital angular momentum, the
wave functions are invariant.
Our results indicate that further investigation is necessary for exploring full-fledged
pathology of the Gödel universe in M-theory. It is evident that naive application of quantum
field theory approximation to M-theory on Gödel universe is ill-defined. For example,
because Gödel universe is not globally hyperbolic, inner products is ill-defined. As such,
expectation value of the energy–momentum tensor Tmn and back-reaction thereof are not
computable.
An attitude one may take for the Gödel universe is that it is similar to negative-mass
Schwarzschild or extreme Reissner–Nordström black hole. Formally, the latter is a solution
with a naked time-like singularity, repelling all time-like geodesics. For both negative-mass
Schwarzschild black hole and the Gödel universe, an issue is that they are not likely to be
formed out of physical processes in initial spacetime without pathologies such as CTCs
or singularities. Supersymmetries preserved by the Gödel universe would not help much
as the negative-mass extreme Reissner–Nordström black hole as well is embeddable as
a supersymmetry preserving configuration. Our tentative attitude though is sympathetic
to [9]: until proven otherwise, the Gödel universe in M-theory deserves further inspection,
including possible observational constraints [16].
Acknowledgements
We acknowledge Eric Gimon, Rajesh Gopakumar, Carlos Herdeiro, Gary Horowitz,

Oleg Lunin, Juan Maldacena and Fumihiko Sugino, Horomitsu Takayanagi, and Tadashi
Takayanagi for enlightening discussions. This work was done while S.J.R. was a Member
at the Institute for Advanced Study. He thanks the School of Natural Sciences for
hospitality and for the grant in aid from the Fund for Natural Sciences.
Appendix A
Five-dimensional Gödel metric
In the bipolar coordinates, nonzero components of the inverse metric are

g t t = −1 + µ2 r12 + r22 , g r1 r1 = g r2 r2 = 1,
1 1
g φ1 φ1 = , g φ2 φ2 = , g φ1 t = g φ2 t = −µ.
r12 r22
Nonzero components of the Christoffel connection are
φ
Γrt1 t = µ2 r1 , Γφr11t = µr1 , Γr11t = −µ/r1 ,
r φ
Γφt1 r1 = (µr1 )3 , Γφ11φ1 = −r1 1 − 2µ2 r12 , Γφ11r1 = −µ2 r1 + 1/r1 ,
φ
Γφt2 r1 = µ3 r1 r22 , Γφr11φ2 = µ2 r1 r22 , Γφ21r1 = −µ2 r22 /r1 ,
φ
Γrt2 t = µ2 r2 , Γφr22t = µr2 , Γr22t = −µ/r2 ,
φ
Γφt2 r2 = (µr2 )3 , Γφr22φ2 = −r2 1 − 2µ2 r22 , Γφ22r2 = −µ2 r2 + 1/r2 ,
φ
Γφt1 r2 = µ3 r12 r2 , Γφr12φ2 = µ2 r12 r2 , Γφ12r2 = −µ2 r12 /r2 .
Six-dimensional Gödel universe
For the six-dimensional Gödel universe G6 in Type IIB string theory, nonzero
components of the inverse metric are

g t t = −1 + µ2 r12 + r22 , g φ1 t = g φ2 t = g yφ1 = g yφ2 = −µ,
1 1
g yt = µ2 r12 + r22 , g r1 r1 = g r2 r2 = 1, g φ1 φ1 = 2 , g φ2 φ2 = 2 ,
r1 r2

g = 1 + µ r1 + r2 ,
yy 2 2 2
and the Christoffel connection are

y y
Γrt1 t = Γr1 t = −Γr1 y = −Γrt1 y = µ2 r1 , Γφr11t = −Γφr11y = µr1 ,
φ 1 φ µ
Γφr11φ1 = −r1 , Γφ11r1 = , Γr11t = −Γrφ11y = −
r1 r1
along with those replaced with (1 ↔ 2).
References
[1] K. Gödel, An example of a new type of cosmological solutions of Einstein’s field equations of gravitation,
Rev. Mod. Phys. 21 (1949) 447.
[2] S.W. Hawking, The chronology protection conjecture, Phys. Rev. D 46 (1992) 603.
[3] J.P. Gauntlett, et al., All supersymmetric solutions of minimal supergravity in five dimensions, hep-
th/0209114.
[4] T. Harmark, T. Takayanagi, Supersymmetric Goedel universes in string theory, hep-th/0301206.
[5] G.W. Gibbons, C. Herdeiro, Supersymmetric rotating black holes and causality violations, Class. Quantum
Grav. 16 (1999) 3619, hep-th/9906098.
[6] C. Herdeiro, Spinning deformations of the D1–D5-system and a geometric resolution of closed timelike
curves, hep-th/0212002.
[7] L. Dyson, Chronology protection in string theory, hep-th/0302052.
[8] R. Biswas, E. Keski-Vakkuri, R.G. Leigh, S. Nowling, E. Sharpe, The taming of closed timelike curves,
hep-th/0304241.
[9] E.K. Boyda, S. Ganguli, P. Horava, U. Varadarajan, Holographic protection of chronology in universes of
the Goedel type, hep-th/0212087.
[10] D. Bak, S.-J. Rey, Cosmic holography, Class. Quantum Grav. 17 (2000) L83, hep-th/9902173.
[11] R. Bousso, Holography in general space–times, JHEP 9906 (1999) 028, hep-th/9906022.
[12] N. Drukker, B. Fiol, J. Simón, Gödel’s universe in a supertube shroud, hep-th/0306057.
[13] S. Chandrasekhar, J.P. Wright, The geodesics in Gödel universe, Proc. Nat. Acad. Sci. 47 (1961) 341;
E. Radu, Features of a charged rotating spacetime, Class. Quantum Grav. 5 (1998) 2743.
[14] In preparation.
[15] B. Mashhoon, Influence of gravitation on the propagation of electromagnetic radiation, Phys. Rev. D 11
(1975) 2679;
W.A. Hiscock, Scalar perturbations in the Gödel universe, Phys. Rev. D 17 (1978) 1497.
[16] E.F. Bunn, P. Ferreira, J. Silk, How anisotropic is our universe?, Phys. Rev. Lett. 77 (1996) 2883, astro-
ph/9605123.
Partition function and open/closed string duality in

type IIA string theory on a pp-wave
Hyeonjoon Shin a,c , Katsuyuki Sugiyama b , Kentaroh Yoshida c
a BK21 Physics Research Division and Institute of Basic Science, Sungkyunkwan University,
Suwon 440-746, South Korea
b Department of Physics, Kyoto University, Kyoto 606-8501, Japan
c Theory Division, High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki 305-0801, Japan
Abstract
We discuss partition functions of N = (4, 4) type IIA string theory on the pp-wave background.
This theory is shown to be modular invariant. The boundary states are constructed and possible
D-brane instantons are classified. Then we calculate cylinder amplitudes in both closed and open
string descriptions and check the open/closed string duality. Furthermore we consider general
properties of modular invariant partition functions in the case of pp-waves.
PACS: 11.25.-w; 12.60.Jv
Keywords: pp-wave; Modular invariance; Boundary state; D-brane; Open/closed string duality
1. Introduction
Recently, superstring theories on pp-waves have been very focused upon. The
maximally supersymmetric pp-wave type solution in eleven dimensions [1] has been
known for a long time while the maximally supersymmetric type IIB pp-wave solution
was found in [2] in the recent progress. It was also pointed out in [3] that this pp-wave
background is related to the AdS-geometry via the Penrose limit [4,5]. Then, the type IIB
superstring theory on the maximally supersymmetric pp-wave background was constructed
[6,7]. By using this pp-wave superstring theory, the study of AdS/CFT correspondence has
E-mail addresses: hshin@newton.skku.ac.kr, hshin@post.kek.jp (H. Shin),

sugiyama@phys.h.kyoto-u.ac.jp (K. Sugiyama), kyoshida@post.kek.jp (K. Yoshida).
doi:10.1016/j.nuclphysb.2003.07.015
H. Shin et al. / Nuclear Physics B 669 (2003) 78–102 79
greatly proceeded [8]. In particular, the AdS/CFT correspondence has been studied at the
stringy level beyond the supergravity analysis.
In the study of pp-wave backgrounds, the matrix model on the pp-wave, which was
proposed by Berenstein–Maldacena–Nastase [8], has been much studied. This matrix
model is closely related to a supermembrane theory on the pp-wave background [9,10].
We have discussed the supermembrane theory and matrix model on the pp-wave from the
several aspects [10–14]. In particular, we showed the correspondence of brane charges in
the supermembrane theory [10] and matrix model [11] in the pp-wave case as well as in
flat space [15].
A supermembrane in eleven dimensions is related to a string in ten dimensions via the
double dimensional reduction. We constructed the type IIA pp-wave background with 24
supersymmetries and string theory on this pp-wave background [16,17], which is called
N = (4, 4) type IIA string theory on the pp-wave. We discussed the classification of the
allowed D-branes [16,18] by following the work of Dabholkar and Parvizi [19]. After
these works, the covariant classification of D-branes was done in [20] where D0-branes
could be studied. In addition, the spectrum of this type IIA string theory was compared to
fluctuations of the linearized type IIA supergravity around the pp-wave background [21].
The thermodynamics of this type IIA string theory was recently studied in [22].1
On the other hand, as an important and interesting subject, the modular invariance of
string theories on pp-waves has been studied by several authors [22–28]. It has been turned
out that these theories are modular invariant in spite of mass terms in the action. In this
paper, motivated by the previous works, we will be interested in the N = (4, 4) type IIA
string theory on the pp-wave background obtained in [16,17] and study its partition
function and open/closed string duality. The pp-wave background we will consider is
not maximally supersymmetric but has 24 supersymmetries. Hence it is interesting to
study whether the modular invariance and consistency condition between open and closed
strings are satisfied or not in such less supersymmetric case. Moreover, since the number
of preserved supersymmetries is nontrivial even in the case of supersymmetric D-branes,
it is also interesting to construct boundary states in our model.2
In this paper we discuss partition functions in the N = (4, 4) type IIA string theory
on the pp-wave. We first show that our theory is modular invariant. The boundary states
are constructed and then we classify the allowed D-brane instantons. They are 1/2 BPS
states (preserving 12 supersymmetries) at the origin and 1/3 BPS ones (preserving 8
supersymmetries) away from the origin. This result is consistent with the classification
of D-branes. Then we calculate the cylinder amplitude in the closed and open string
descriptions and study the open/closed string channel duality in our theory. Finally, we
discuss general properties of modular invariant partition function and classify several
models.
This paper is organized as follows: in Section 2 we prove the modular invariance of
N = (4, 4) type IIA string theory on the pp-wave background. The Witten index of this
1 Section 2 has some overlap with the work [22]. Thermal partition function is discussed in the cases of other
pp-wave strings [23,24].
2 We note that, contrary to the present type IIA case, the boundary states in the type IIB string theory on the
pp-wave background have been relatively much studied [25,28–30].
80 H. Shin et al. / Nuclear Physics B 669 (2003) 78–102
theory is shown to be one. Section 3 is devoted to a brief review about the supersymmetries
of our theory. In Section 4 we construct boundary states and classify D-brane instantons. In
Section 5 we calculate the cylinder amplitude in the closed string description. In Section 6
the amplitude is derived in terms of open string and the open/closed string channel duality
in our theory is proven. From the channel duality, the normalization factor of the boundary
state is determined. In Section 7, based on the result of Section 2, we discuss several
properties of modular invariant partition function in some general setup. Finally, Section 8
is devoted to conclusions and discussions.
2. Modular invariance of type IIA string theory
In this section we will discuss the modular invariance of type IIA string theory in the
closed string description.
The action of our type IIA string in the light-cone gauge is given by
2π 2 2
1
8
µ
4
2 µ
8
2
Sclosed = dτ dσ ∂+ x i ∂− x i − xa − xb
4πα 3 6
0 i=1 a=1 b=5
2π
i
+ dτ dσ Ψ 1+T ∂− Ψ 1+ + Ψ 1−T ∂− Ψ 1− + Ψ 2+T ∂+ Ψ 2+
2π
0
µ 1−T T 2+ µ 2+T
+ Ψ 2−T ∂+ Ψ 2− − Ψ Π Ψ + Ψ ΠΨ 1−
3
3
µ µ
− Ψ 1+T Π T Ψ 2− + Ψ 2−T ΠΨ 1+ , (2.1)
6 6
where α is a string tension and we have set p+ = 1. The γ r ’s are 16 × 16 SO(9) gamma
matrices and we defined Π ≡ γ 123 (Π T ≡ γ 321 ). Each of the spinors Ψ i± (i = 1, 2)
has four independent components and the superscript ± represents the chirality measured
by γ 1234 : γ 1234 Ψ i± = ±1 · Ψ i± . This theory has 24 supersymmetries (8 dynamical
supersymmetries and 16 kinematical supersymmetries). The equations of motion are
described by
µ2 a
∂+ ∂− x a + x =0 (a = 1, 2, 3, 4), (2.2)
9
µ2 b
∂+ ∂− x b + x =0 (b = 5, 6, 7, 8), (2.3)
36
µ µ T 2+
∂+ Ψ 2+ + ΠΨ 1− = 0, ∂− Ψ 1− − Π Ψ = 0, (2.4)
3 3
µ µ
∂+ Ψ 2− + ΠΨ 1+ = 0, ∂− Ψ 1+ − Π T Ψ 2− = 0. (2.5)
6 6
By solving the equations of motion (2.2) and (2.3), we can obtain the mode-expansions of
bosonic variables represented by

µ 3 a µ α 1 a
x a
(τ, σ ) = x0a cos τ + α p0 sin τ +i αn φn + ᾱna φ̃n ,
3 µ 3 2 ωn
n
=0
(2.6)

µ 6 b µ α
1 b
b
x b (τ, σ ) = x0b cos τ + α p0 sin τ +i α φ + ᾱ φ̃ .
6 µ 6 2 ωn n n n n
n
=0
(2.7)
From the equations of motion (2.4) and (2.5), the mode-expansions of fermionic variables
are represented by

µ µ
Ψ 1− (τ, σ ) = Π T Ψ0 sin τ − Π T Ψ̃0 cos τ
3 3
3
+ cn i(ωn − n)Π T Ψn φn + Ψ̃n φ̃n , (2.8)
µ
n
=0

µ µ
Ψ (τ, σ ) = Ψ0 cos
2+
τ + Ψ̃0 sin τ
3 3

3
+ cn Ψn φn − i(ωn − n)Π Ψ̃n φ̃n , (2.9)
µ
n
=0

T µ µ
Ψ (τ, σ ) = Π Ψ0 sin
1+
τ − Π Ψ̃0 cos τ
6 6

6 T
+ cn i(ωn − n)Π Ψn φn + Ψ̃n φ̃n , (2.10)
µ
n
=0

µ µ
Ψ 2− (τ, σ ) = Ψ0 cos τ + Ψ̃0 sin τ
6 6
6

+ cn Ψn φn − i(ωn − n)Π Ψ̃n φ̃n . (2.11)

µ
n
=0
Here we have introduced several notations:

2 2
µ µ
ωn = sgn(n) n2 + , ωn = sgn(n) n2 + ,
3 6

φn = exp −i(ωn τ − nσ ) , φ̃n = exp −i(ωn τ + nσ ) ,

φn = exp −i(ωn τ − nσ ) , φ̃n = exp −i(ωn τ + nσ ) ,
2 −1/2 2 −1/2
3 6
cn = 1 + (ωn − n) 2
, cn = 1 + (ωn − n) 2
.
µ µ
Now we shall quantize the theory by imposing (anti)commutation relations. The commu-
tation relations for bosonic modes are given by
i j i j i j
x0 , p0 = iδ ij , ᾱm , αn = αm , ᾱn = 0 (i, j = 1, . . . , 8),
a a a a
αm , αn = ᾱm , ᾱn = ωm δm+n,0 δ aa (a, a = 1, 2, 3, 4),
b b b b
αm , αn = ᾱm , ᾱn = ωm δm+n,0 δ bb (b, b = 5, 6, 7, 8),
and the anticommutation relations for fermionic modes are written as

(Ψm )α , (Ψ̃n )Tβ = (Ψ̃m )α , (Ψn )Tβ = 0,
1
(Ψm )α , (Ψn )Tβ = (Ψ̃m )α , (Ψ̃n )Tβ = δm+n,0 δαβ ,
2
T
T

(Ψm )α , (Ψ̃n )β = (Ψ̃m )α , (Ψn )β = 0,
1
(Ψm )α , (Ψn )Tβ = (Ψ̃m )α , (Ψ̃n )Tβ = δm+n,0 δαβ .
2
Now let us introduce the annihilation and creation operators:

α 3 µ a a† α 3 µ
a0 ≡
a
p0 − i x0 ,
a
a0 ≡ p0a + i x0a
2 µ 3α 2 µ 3α
(a = 1, 2, 3, 4),
1 1 a 1 1 a
ana ≡ √ αna , ana† ≡ √ α−n , āna ≡ √ ᾱna , āna† ≡ √ ᾱ−n
ωn ωn ωn ωn
(n > 0),
for the sector with mass µ/3 and

α 6 µ α 6 µ
a0 ≡
b
p0a − i x0a , a0b† ≡ p0a + i x0a
2 µ 6α 2 µ 6α
(b = 5, 6, 7, 8),
1 1 b 1 1 b
anb = αnb , anb† = α−n , ānb = ᾱnb , ānb† = ᾱ−n
ωn ωn ωn ωn
(n > 0),
for the sector with mass µ/6, and those for fermionic variables:
S0 = Ψ0 + i Ψ̃0 , S0† = Ψ0 − i Ψ̃0 ,
S0 = Ψ0 + i Ψ̃0 , S0 † = Ψ0 − i Ψ̃0 ,
√ √ √ √
Sn = 2 Ψn , Sn† = 2 Ψ−n , S̃n = 2 Ψ̃n , S̃n† = 2 Ψ̃−n (n > 0),
√ √ √ √
Sn = 2 Ψn , S n = 2 Ψ−n S̃n = 2 Ψ̃n , S̃n † = 2 Ψ̃−n

†
, (n > 0).
Then the commutation relations are rewritten as
a a † a a †
am , an = δ aa δm,n , ām , ān = δ aa δm,n (a, a = 1, 2, 3, 4),
b b † b b †
am , an = δ bb δm,n , ām , ān = δ bb δm,n (b, b = 5, 6, 7, 8),
and the anticommutation relations are given by

(Sm )α , Sn† β = δαβ δm,n , (S̃m )α , S̃n† β = δαβ δm,n ,
†
(Sm )α , S n β = δαβ δm,n , (S̃m )α , S̃n † β = δαβ δm,n .
By using the above (anti)commutation relations, we can represent the Hamiltonian and
momentum in terms of creation and annihilation operators as follows:
∞
∞

H= (ωn Nn + ωn Nn ), P= (nNn + nNn ), (2.12)
n=−∞ n=−∞
where Nn and Nn are defined by

4
8
N0 ≡ a0b† a0b + S 0 S0 ,
†
N0 ≡ a0a†a0a + S0† S0 ,
a=1 b=5

4
8
Nn ≡ anb† anb + S n Sn
†
Nn ≡ ana† ana + Sn† Sn , (n > 0),
a=1 b=5

4
8
†
N−n ≡ āna†āna + S̃n† S̃n ,
N−n ≡ ānb† ānb + S˜ n S̃n (n > 0).
a=1 b=5
Now we shall introduce the Casimir energy defined by

+∞
1 2 1
∆(ν; a) ≡ ν + (n + a)2 − dk ν 2 + k 2
2 2
n∈Z −∞
∞
1 2 s− π 2 ν 2
=− cos(2πa') ds e−' s
2π 2
'1 0
∞
1 ν2
=− ds e− s θ3 (iπs, a) − 1 . (2.13)
2
0
Then we can express the vacuum energy for eight bosons as

4
1 8
1
EB0 = ωn + ωn ∼= 4 ∆(µ/3; 0) + ∆(µ/6; 0) , (2.14)
2 2
a=1 n∈Z b=5 n∈Z
and that for fermions as

1 1 ∼
EF = −4
0
ωn + ωn = −4 ∆(µ/3; 0) + ∆(µ/6; 0) , (2.15)
2 2
n∈Z n∈Z
where the symbol ∼ = means the equality after the regularization of zero-point energies and
the factor 4 appears since each fermion considered here has four independent components.
Here let us evaluate the toroidal partition function:

Z = Tr (−1)F e−2πτ2 H +2πiτ1 P , (2.16)
where τ1 and τ2 are modular parameters and F is the fermion number operator.
First we will consider a partition function for one boson with mass ν. The number
operator, Hamiltonian and momentum are given by
N0 = a0† a0 , Nn = an† an , N−n = ān†ān (n > 0),
∞
∞

H = ω0 N0 + (ωn Nn + ωn N−n ), P= (nNn − nN−n ),
n=1 n=1
and so we can obtain the partition function:
−1/2
Z = Θ(0,0) (τ, τ̄ ; ν) , (2.17)
where we have introduced the ‘massive’ theta function defined by
√
Θ(a,b) (τ, τ̄ ; ν) ≡ e4πτ2 ∆(ν;a) 1 − e−2πτ2 ν 2 +(n+a)2 +2πτ1 (n+a)+2πib 2 . (2.18)
n∈Z
Next we consider a partition function for single component fermion. The number operator,
Hamiltonian and momentum are given by
N0 = S0† S0 ,Nn = Sn† Sn , N−n = S̃n† S̃n (n > 0),

H = ω0 N0 + (ωn Nn + ωn N−n ), P = (nNn − nN−n ),
n>0 n>0
and hence we can obtain the partition function:
+1/2
Z = Θ(0,0) (τ, τ̄ ; ν) . (2.19)
If we recall the field contents of our model:
a
µ
x (a = 1, 2, 3, 4), (S, S̃) (mass 3 sector),
bosons b fermions (2.20)
x (b = 5, 6, 7, 8), (S , S̃ ) (mass µ
6 sector),
the bosonic and fermionic partition functions ZB and ZF are given by
−2 −2
ZB = Θ(0,0) (τ, τ̄ ; µ/3) Θ(0,0)(τ, τ̄ ; µ/6) ,
+2 +2
ZF = Θ(0,0) (τ, τ̄ ; µ/3) Θ(0,0) (τ, τ̄ ; µ/6) ,
and hence the total partition function is given by
Z = ZB · ZF = 1. (2.21)
Thus we have shown that our theory is modular invariant at the one-loop level since the
total partition function Z of (2.21) is independent of the modular parameters τ and τ̄ . As
will be shown in more detail with some generality in Section 7, this result implies that
the Witten index is one as in the cases of other string theories on pp-waves. It should be
remarked that our type IIA string theory is modular invariant in the sector with mass µ/3
and in that with µ/6, respectively.
3. Supersymmetries of type IIA string theory
In this section we will briefly review about N = (4, 4) supersymmetries of the type
IIA string theory on the pp-wave background [17,18], according to which the world-sheet
variables are arranged into two supermultiplets, (x a , Ψ 1− , Ψ 2+ ) and (x b , Ψ 1+ , Ψ 2− ).
Then we will rewrite the supercharges in terms of modes in order to construct the boundary
states in Section 4.
This type IIA string theory on the pp-wave background has 24 supersymmetries,
among which 8 are dynamical and 16 are kinematical. The dynamical supersymmetry
transformation laws for the multiplet (x a , Ψ 1− , Ψ 2+ ) with mass µ/3 are given by

δx a = 2iα Ψ 1−T γ a / 1+ + Ψ 2+T γ a / 2− ,
µ
δΨ 1− = ∂+ x a γ a / 1+ + x a γ 4 γ a / 2− ,
3
µ a 4 a 1+
δΨ = ∂− x γ / − x γ γ / ,
2+ a a 2−
(3.1)
3
and those for the multiplet (x b , Ψ 1+ , Ψ 2− ) with mass µ/6, are written as

δx b = 2iα Ψ 1+T γ b / 2− + Ψ 2−T γ b / 1+ ,
µ
δΨ 1+ = ∂− x b γ b / 2− + x b γ 4 γ b / 1+ ,
6
µ b 4 b 2−
δΨ = ∂+ x γ / − x γ γ / ,
2− b b 1+
(3.2)
6
where the constant spinors / 1+ and / 2− satisfy the following chirality conditions:
γ 9 / 1+ = +/ 1+ , γ 1234/ 1+ = +/ 1+ ,
γ 9 / 2− = −/ 2− , γ 1234/ 2− = −/ 2− , (3.3)
respectively. As for the kinematical supersymmetry, the transformation laws are given by
δ̃x a = δ̃x b = 0 and

µ µ
δ̃Ψ 1− = cos τ /̃ 1− − sin τ γ 123/̃ 2+ ,
3 3

µ µ
δ̃Ψ = cos
2+
τ /̃ − sin
2+
τ γ 123/̃ 1− , (3.4)
3 3

µ µ
δ̃Ψ 1+ = cos τ /̃ 1+ − sin τ γ 123/̃ 2− ,
6 6

µ µ
δ̃Ψ 2− = cos τ /̃ 2− − sin τ γ 123/̃ 1+ , (3.5)
6 6
where the constant spinors /̃ 1+ , /̃ 1− , /̃ 2+ and /̃ 2− satisfy the chirality conditions in terms
of γ 9 and γ 1234 in the same way as the dynamical supersymmetry case.
By the use of Noether’s theorem, we can construct the associated supercharges. Firstly
the dynamical supercharges are obtained as
/ T Q(µ/3) = / 1+T Q1+ + / 2−T Q2− , / T Q(µ/6) = / 2−T Q 2− + / 1+T Q 1+ .

(3.6)
Here the Q1+ and Q2− are the quantities defined as
2π
i µ
Q 1+
≡− dσ ∂+ x a γ a Ψ 1− − x a γ a γ 4 Ψ 2+ , (3.7)
2π 3
0
2π
i µ
Q 2−
≡− dσ ∂− x a γ a Ψ 2+ + x a γ a γ 4 Ψ 1− , (3.8)
2π 3
0
and, for the Q 2− and Q 1+ ,

2π
2− i µ b b 4 2−
Q ≡− dσ ∂+ x γ Ψ − x γ γ Ψ
b b 1+
, (3.9)
2π 6
0
2π
i µ
Q 1+ ≡ − dσ ∂− x b γ b Ψ 2− + x b γ b γ 4 Ψ 1+ . (3.10)
2π 6
0
Secondly, the kinematical supercharges are obtained as
Q̃(µ/3) ≡ /̃ 1−T Q̃1− + /̃ 2+T Q̃2+ , Q̃(µ/6) ≡ /̃ 1+T Q̃1+ + /̃ 2−T Q̃2− , (3.11)
where the Q̃1− and Q̃2+ are defined by

2π
i µ µ
Q̃1−
≡ dσ cos τ Ψ + sin
1− 123 2+
τ γ Ψ , (3.12)
2π 3 3
0
2π
i µ µ
Q̃2+ ≡ dσ cos τ Ψ 2+ + sin τ γ 123Ψ 1− , (3.13)
2π 3 3
0
and, for the Q̃1− and Q̃2+ ,

2π
i µ µ
Q̃1+
≡ dσ cos τ Ψ 1+ + sin τ γ 123Ψ 2− , (3.14)
2π 6 6
0
2π
i µ µ
Q̃2−
≡ dσ cos τ Ψ + sin
2− 123 1+
τ γ Ψ . (3.15)
2π 6 6
0
Now we can rewrite the above supercharges in terms of creation and annihilation
operators by inserting the mode-expansions of bosonic and fermionic degrees of freedom.

√ −1 1+ µ a† a T
α Q = c0 a0 γ Π S0 − a0a γ a Π T S0†
3
∞
√
−i cn ωn āna† γ a S̃n + āna γ a S̃n†
n=1
∞
1 µ 1 a† a T
+ √ an γ Π Sn − ana γ a Π T Sn† , (3.16)
23 cn ωn
n=1

√ −1 2− µ a† a
α Q = − ic0 a γ S0 + a0a γ a S0†
3 0
∞
√
−i cn ωn ana† γ a Sn + ana γ a Sn†
n=1
∞
1µ 1 a a
+ √ ān γ Π S̃n† − āna†γ a Π S̃n , (3.17)
23 cn ωn
n=1
and

√ −1 2− µ b† a T
α Q = c0 a0 γ Π S0 − a0b γ a Π T S0 †
6
∞

−i cn ωn ānb† γ b S̃n + ānb γ b S̃n †
n=1
∞
1 µ 1 b† b T
+ an γ Π Sn − anb γ b Π T Sn † , (3.18)
26
c ωn
n=1 n

√ −1 1+ µ b† b
α Q = − ic0 a0 γ S0 + a0b γ b S0 †
6
∞

−i cn ωn anb† γ b Sn + anb γ b Sn †
n=1
∞
1µ 1 b b
+ ān γ Π S̃n † − ānb γ b Π S̃n . (3.19)
26
cn ωn
n=1
On the other hand, the kinematical supersymmetries are rewritten as

Π i
Q̃1− = iΠ Ψ̃0 = S0 − S0† , Q̃2+ = iΨ0 = S0 + S0† , (3.20)
2 2
Π i
Q̃1+ = iΠ Ψ̃0 = S0 − S0 † , Q̃2− = iΨ0 = S0 + S0 † . (3.21)
2 2
In the next section, the above expressions of supercharges will be used for constructing
the fermionic boundary states.
4. Boundary states of type IIA string theory
In this section we will construct the boundary states of type IIA string theory on the
pp-wave background with 24 supersymmetries. To begin with, the bosonic boundary states
will be constructed. Next, we construct the fermionic boundary states and classify the
allowed D-brane instantons in our theory. The resulting boundary states will be used for
the calculation of amplitude in the closed string description.
4.1. Bosonic boundary states
Here we will consider the bosonic part of boundary states in the Type IIA string theory.
The bosonic coordinates (x a , x b ) (a = 1, . . . , 4, b = 5, . . . , 8) are classified into (x ā , x b̄ )
(for the Neumann condition) and (x a , x b ) (for the Dirichlet condition).
The definition of bosonic boundary state |B is given by the following boundary
conditions:

∂τ x ā τ =0 |B = 0, ∂τ x b̄ τ =0 |B = 0 (Neumann), (4.1)
a a b b
x0 − q0 τ =0 |B = 0, x0 − q0 τ =0 |B = 0 (Dirichlet). (4.2)
The conditions (4.1) can be rewritten as
ā ā
a0 + a0ā† |B = 0, an + ānā† |B = 0 (n > 0), (4.3)
b̄ b̄
a0 + a0b̄† |B = 0, an + ānb̄† |B = 0 (n > 0), (4.4)
and lead to the bosonic boundary state |B for Neumann directions described by
∞

1 ā† ā† 1 b̄† b̄† ā† ā† b̄† b̄†
|B = exp − a0 a0 − a0 a0 − an ān + an ān |0,
2 2
ā b̄ n=1 ā b̄
(4.5)
where |0 is the bosonic Fock vacuum state annihilated by the operators, ana,b and āna,b .
On the other hand, the second conditions (4.2) can be rewritten as

a a† 2µ 1/2 a a a†
a0 − a0 + i
q 0 |B = 0, an − ān |B = 0 (n > 0), (4.6)
3α

b b† µ 1/2 b b b†
a0 − a0 + i
q0 |B = 0, an − ān |B = 0 (n > 0). (4.7)
3α
With these boundary conditions, we can construct the boundary states for Dirichlet
directions described by

1 a† 2µ 1/2 a 2 1 b† µ 1/2 b 2
|B = exp a0 − i q0 + a0 − i q0
2 a 3α 2 3α
b
∞
a† a† b† b†
× exp an ān + an ān |0. (4.8)
n=1 a b
Here we shall introduce a diagonal matrix Mij = diag(±1, · · · , ±1) with eight components
where +1 is assigned for Dirichlet directions and −1 is assigned for Neumann ones. If we
a b
will set as q0 = q0 = 0, the bosonic boundary state can be rewritten as
|B = |Bµ/3 ⊗ |Bµ/6 ,
∞
1 a† a † †
|Bµ/3 ≡ exp Maa a0 a0 exp Maa an ān |0,
a† a
2
n=1
∞
1 b† b † †
|Bµ/6 ≡ exp Mbb a0 a0 exp Mbb an ān |0.
b† b
(4.9)
2
n=1
Thus the bosonic boundary state is the product of two sectors with mass µ/3 and that with
µ/6, and it has the SO(4) × SO(4) symmetry. We will study the fermionic boundary states
in the next subsection.
4.2. Fermionic boundary states
We will now consider the fermionic part of boundary states in our case. The fermionic
boundary states are defined by
2− 2−
Qα − iηMαβ Qβ1+ |B = 0,
(µ/3) (µ/6)
Qα − iηMαβ Q1+ β |B = 0, (4.10)
2+ (µ/3) 1− 2− (µ/6) 1+
Q̃α + iηM̂αβ Q̃β |B = 0, Q̃ + iηM̂αβ Q̃β |B = 0, (4.11)
(µ/3) (µ/6) (µ/3) (µ/6)
where matrices Mαβ , Mαβ , M̂αβ and M̂αβ satisfy the following relations:
T
Mαβ M T βγ = δαγ , M αβ Mβγ = δαγ , (4.12)
T
M̂αβ M̂ T βγ = δαγ , M̂ αβ M̂βγ = δαγ . (4.13)
The definition of the fermionic boundary states (4.11) leads to the conditions written in
terms of the zero-modes:3

(Ψ0 )α + iηM̂αβ (Π Ψ̃0 )β |B = 0.
(µ/3) (µ/6)
(Ψ0 )α + iηM̂αβ (Π Ψ̃0 )β |B = 0,
(4.14)
These conditions suggest us to take the following ansatz:
(µ/3) † (µ/6) †
(Sn )α + iηM̂αβ S̃n β |B = 0, (Sn )α + iηM̂αβ S̃n β |B = 0. (4.15)
The above equations can be easily solved and the boundary state is given by
∞
(µ/3) † † (µ/6) † †
|B = exp −iηM̂αβ Sn α S̃n β − iηM̂αβ Sn α S̃n β |B0 , (4.16)
n=1
3 If we redefine the Ψ̃ as Π Ψ̃ → Ψ̃ , then we obtain the usual expressions for zero-mode conditions. The
0 0 0
effect from the redefinition of Ψ̃0 is absorbed into the definition of the creation and annihilation operators without
the modification of anticommutation relations, and so we have no trouble for our discussion.
where |B0 is the fermionic vacuum state yet to be determined. By the way, this boundary
state is by definition the state satisfying (4.10) and (4.11). If we now act the conditions
(4.15) on (4.10), we have three types of conditions that lead us to determine the structure
of the matrices M (µ/3) , M (µ/6) , M̂ (µ/3) and M̂ (µ/6) . Firstly, we obtain the conditions

M a a γ a = −M (µ/3)γ a M̂ (µ/3)T , M b b γ b = −M (µ/6) γ b M̂ (µ/6)T , (4.17)
which are similar to those arising in flat space [29]. The second type of conditions, which
appears only in the pp-wave case, is

M aa γ a Π = −M (µ/3) γ a Π M̂ (µ/3) , M bb γ b Π = −M (µ/6) γ b Π M̂ (µ/6). (4.18)
Finally, the third type of conditions we get comes from the zero-mode parts:
a† a
a0 γ + ηM (µ/3)γ a Π S0 + M aa a0a † γ a − ηM (µ/3)γ a Π S0† |B = 0, (4.19)
b† b
a0 γ + ηM (µ/6)γ b Π S0 + M bb a0b † γ b − ηM (µ/6)γ b Π S0† |B = 0. (4.20)
By the use of the definition of fermionic vacuum S0 |B0 = 0 and the identities a0a =

M aa a0a † and a0b = M bb a0b † , we can rewrite (4.19) and (4.20) as

µ a
p0a δ aa − M aa + i x0a δ aa + M aa γ − ηM (µ/3)γ a Π S0† |B0 = 0,
3α
(4.21)
µ †
p0b δ bb − M bb + i x0b δ bb + M bb γ b − ηM (µ/6) γ b Π S0 |B0 = 0.
6α
(4.22)
Using the conditions (4.21) and (4.22), we can read off supersymmetries preserved by
D-brane instantons in terms of their positions. In the case that all position coordinates
of a D-brane instanton q r for the Dirichlet directions equals zero, the D-brane instanton
has 12 (4 dynamical + 8 kinematical) supersymmetries (i.e., 1/2(= 12/24) BPS D-brane
instanton). If the position coordinates for the Dirichlet directions are not at the origin,
then the D-brane instanton has 8 (0 + 8) supersymmetries (i.e., 1/3(= 8/24) BPS D-brane
instanton). Thus all of the dynamical supersymmetries are broken. However the D-brane
instantons apart from the origin are supersymmetric solutions since they still have 8
kinematical supersymmetries.
As a final remark regarding the structure of the matrices, M (µ/3) , M (µ/6) , M̂ (µ/3) and
M̂ (µ/6) , we now consider the chirality condition while the above three types of conditions
are almost same as in the type IIB string case [30]. First we can easily find that both
matrices M (µ/3) and M (µ/6) contain the odd number of gamma matrices in order to
preserve the SO(8) chirality measured by γ 9 . Moreover, if we consider the chirality in
terms of the matrix R = γ 1234 , we have the following conditions basically from (4.10) and
(4.11):

R, M (µ/3) = 0, R, M (µ/6) = 0,

R, M̂ (µ/3) = 0, R, M̂ (µ/6) = 0. (4.23)
Then we obtain the following complete boundary state:

∞
(µ/3) † †
|B = exp Maa ana†āna † + Mbb anb†ānb † − iηM̂αβ Sn α S̃n β
n=1

(µ/6) † †
− iηM̂αβ Sn α S̃n β |B0 ,

1 1
|B0 = MI J |I |J − iηMαβ |α|β exp Maa a0a†a0a † + Mbb a0b† a0b † |0,
2 2
(4.24)
where the state |B0 is the product of the bosonic vacuum state, which is given by picking
up the zero-mode parts in Eq. (4.9), and the fermionic one which is the solution of (4.14).
The remaining task is to determine the matrices M and M̂ from the conditions (4.17),
(4.18), (4.21), (4.22) and (4.23). The determined structure of the matrices leads to the
classification of possible D-brane instantons. This will be done in the next subsection.
4.3. Classification of D-brane instantons
We will classify the allowed D-brane instantons by determining the matrices M and M̂.
Now let us analyze each of Dp-brane instantons.
(D0) D0-brane instantons are expressed by the following matrices:
M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6) = γ I (I = 1, 2, 3). (4.25)

The x I -direction satisfies the Neumann boundary condition and other directions
satisfy the Dirichlet boundary condition. When we consider M = γ 1 as an example,
x 1 is a Neumann direction and other directions are Dirichlet ones. If we consider the
D0-brane instanton at the origin q 2,3,4,5,6,7,8 = 0, then it is a 1/2 BPS object. If we
consider the D0-brane instanton apart from the origin, then it becomes a 1/3 BPS
object;
(D2) D2-brane instantons are given by
M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6) = γ I J 4 , (4.26)

or

M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6) = γ 4 γ bb ,
where I and J take values in 1, 2, 3, and b and b run from 5 to 8. For example, if we
take M = γ 124 then x 1,2,4 -directions satisfy the Neumann condition and others are
Dirichlet directions. When the D2-brane instanton sits at the origin q 3,5,6,7,8 = 0, it
is a 1/2 BPS object. Once it goes away from the origin, it becomes 1/3 BPS;
(D4) D4-brane instantons are described by

M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6) = γ 123 γ bb , (4.27)
Table 1
List of possible D-brane instantons in our type IIA string theory. The NN is the number of Neumann directions
NN M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6)
1 γI

3 γ I J 4 , γ 4 γ bb

5 γ 123 γ bb , γ I γ 5678
7 γ I J 4 γ 5678
or
M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6) = γ I γ 5678.
When we take M = γ 12356, the x 1,2,3,5,6-directions satisfy the Neumann boundary
condition and others are the Dirichlet directions. When the D4-brane instanton is at
the origin q 4,7,8 = 0, it is a 1/2 BPS object. If it is apart from the origin, it becomes
1/3 BPS;
(D6) D6-brane instantons are given by
M (µ/3) = M (µ/6) = M̂ (µ/3) = M̂ (µ/6) = γ I J 4 γ 5678. (4.28)
For example, the case M = γ 1245678 leads to a D6-brane instanton that preserves 12
supersymmetries (i.e., 1/2 BPS) for q 3 = 0 and 8 supersymmetries (1/3 BPS) for
q 3
= 0.
We should remark that the above classification of D-brane instantons is consistent with
that of D-branes in the open string description [18,20]. The Neumann (Dirichlet) boundary
condition in the closed string description is simply related by the Dirichlet (Neumann)
one in the open string description, and hence this identification should hold as a matter of
course. For comparison with the classification of D-branes [18,20], we shall summarize
our result in Table 1.
5. Cylinder amplitude in the closed string description
In this section we will calculate the tree amplitude in the closed string description. The
interaction energy between a pair of D-branes comes from the exchange of a closed string
between two boundary states (i.e., a cylinder diagram).
The expression of the cylinder diagram (tree diagram) in the light-cone formulation can
be expressed as
j
ADp1 ;Dp2 x + , x − , q1i , q2

dp+ dp− ip+ x − +ip− x +
= e Dp1 , −p− , −p+ , q1i
2πi

1
× Dp2 , p− , p+ , q j
+ −
p (p + H ) 2
+∞ +
+ − θ (p ) + j
= dp+ eip x +
Dp1 , −p+ , q1i e−iH x Dp2 , p+ , q2 , (5.1)
p
−∞
where H is the light-cone Hamiltonian of a closed string and the |Dp, p+ , q i represents
a boundary state of a Dp-brane instanton located at the transverse position q i with the
longitudinal momentum p+ . We note that the prescription given in [25] has been used for
obtaining the last line in the above equation. If we define the variable t by
x + = πτ = −iπt
by performing the customary Wick rotation, the amplitude is rewritten as
∞
j x+ x− j
ADp1 ;Dp2 x + , x − , q1i , q2 = dt e− πt ÃDp1 ;Dp2 t, q1i , q2 , (5.2)
0
where ÃDp1 ;Dp2 (t, q1, q2 ) is the expectation value:

j j
ÃDp1 ;Dp2 t, q1i , q2 ≡ Dp1 , −p+ , q1i e−2πt (H/2) Dp2 , p+ , q2 . (5.3)
Now the tree amplitude (5.3) will be calculated by using the boundary states constructed
before. We restrict ourselves to the case of identical Dp-brane instantons for simplicity.
The calculus consists of three parts: (1) vacuum energies, (2) nonzero-modes, and (3) zero-
modes.
Let us concentrate only on the sector with mass ν ≡ µ/3. The other sector with mass
ν ≡ µ/6 results in the same final expression only with the difference in the value of mass
parameter. When we consider the contribution of vacuum energies to the amplitude, the
evaluation of vacuum energies is identical to that of the partition function in the closed
string case:
e2∆(ν;0) (for bosons), e−2∆(ν;0) (for fermions). (5.4)

The contribution of nonzero-modes to the amplitude is also the same as that of partition
function of a closed string, and so it is readily written as
∞
∞

ωn −4
(1 − q ) (bosons), (1 − q ωn )4 (fermions), q ≡ e−2πt . (5.5)
n=1 n=1
The contribution of bosonic zero-modes can be evaluated by using the formula:

† 1 † † −1/2
0|e± 2 a0 a0 q ± 2 ma0 a0 e± 2 a0 a0 |0 = 1 − q m
1 1
. (5.6)
As a result, the factor (1 − q ν )−2 is obtained from bosonic zero-modes. The fermionic
zero-modes can be evaluated by adopting the prescription given in the appendix of [25],
and the resulting contribution is (1 − q ν )2 . Consequently, the contribution of zero-modes
is summarized as follows:
−2
1 − qν (for bosons), (1 − q ν )2 (for fermions). (5.7)
After taking account of the sector with mass µ/6, the total partition function is then
represented by
ÃDp;Dp = ÃB
Dp;Dp · ÃDp;Dp = NDp ,
F 2
(5.8)
where NDp is the normalization factor of boundary states, which is not determined yet.
This factor can be fixed by calculating the cylinder diagram in the open string channel.
This task will be done in the next section.
In the work of [25], the “conformal field theory condition”
ÃDp1 ;Dp2 (t, q1 , q2 ) = Z̃Dp1 ;Dp2 (t˜, q1 , q2 ), (5.9)
was analyzed in the case of the pp-wave background. This condition ensures the
consistency between closed and open string channels. We now turn to the calculation of
the partition function with D-branes (i.e., the cylinder diagram) in order to show that the
condition (5.9) also holds in our theory.
6. Partition function in the open string description
In this section we will discuss the one-loop amplitude of open string, and confirm the
consistency condition between open and closed string channels.
We start from the light-cone action of open string defined by
π 8
1 i 2 2
Sopen = dτ dσ ∂τ x − ∂σ x i
4πα
i=1
0
2 4 2 8

µ a 2 µ b 2
− x − x
3 6
a=1 b=5
π
i
+ dτ dσ Ψ 1+T ∂− Ψ 1+ + Ψ 1−T ∂− Ψ 1− + Ψ 2+T ∂+ Ψ 2+
2π
0
µ 1−T T 2+ µ 1+T T 2−
+ Ψ 2−T ∂+ Ψ 2− −Ψ Π Ψ − Ψ Π Ψ
3
6
µ µ
+ Ψ 2−T ΠΨ 1+ + Ψ 2+T ΠΨ 1− . (6.1)
6 3
To begin with, we shall present the mode-expansion in the case of open string. The
mode-expansion of DD string is expressed as
sinh ν(π − σ ) sinh(νσ ) √ 1 a −iωn τ
x a (τ, σ ) = q0a · + q1a · − 2α α e sin(nσ ),
sinh(πν) sinh(πν) ωn n
n
=0
(6.2)
sinh ν (π − σ ) sinh(ν σ ) √ 1
x b (τ, σ ) = q0b · + q1b · − 2α α a e−iωn τ sin(nσ ),
sinh(πν ) sinh(πν ) ωn n
n
=0
(6.3)
where the endpoints satisfy the Dirichlet conditions: x i (σ = 0) = q0i , x i (σ = π) = q1i .

The mode expansion of NN string, whose endpoints satisfy the Neumann conditions, is
written as
1 √ 1
x a (τ, σ ) = x0a cos(ντ ) + · 2α p0a sin(ντ ) + i 2α α a e−iωn τ cos(nσ ),
ν ωn n
n
=0
(6.4)
1 √ 1
x b (τ, σ ) = x0b cos(ν τ ) + · 2α p0b sin(ν τ ) + i 2α α b e−iωn τ cos(nσ ).
ν ωn n
n
=0
(6.5)
The mode-expansion of fermions is the same with that in the case of closed string, but
we have to take account of boundary conditions at σ = 0 and π described by
Ψ 1− = ΩΨ 2+ , Ψ 2+ = Ω T Ψ 1− , Ψ 1+ = ΩΨ 2− , Ψ 2− = Ω T Ψ 1+ ,
Ψ̃n = ΩΨn , Ψ̃0 = −ΠΩΨ0 , Ψ0 = ΠΩ Ψ̃0 , ΠΩΠΩ = −1,
where Ω is the gluing matrix for fermionic modes on the boundaries.

We now introduce the creation and annihilation operators given by
1 1 a a a†
ana = √ αna , ana† = √ α−n , am , an = δ a,a δm,n (m, n > 0),
ωn ωn
1 1 b b b †
anb = αnb , anb† = α−n , am , an = δ b,b δm,n (m, n > 0),
ωn ωn

α a 1 a a† α a 1
a0 =
a
p0 − iν x0 , a0 = p0 + iν x0a ,
ν 2α ν 2α
a a †
a0 , a0 = δ a,a ,

α b 1 b b† α b 1 b
a0 =
b
p − iν x , a = p + iν x ,
ν 0 2α 0 0
ν 0
2α 0
b b †
a0 , a0 = δ b,b ,
then the Hamiltonian HB for the Dirichlet directions is expressed by
1 ν a 2 a 2
HB =
· q0 + q1 cosh(πν) − 2q0a q1a
4πα sinh(πν)
1 ν b 2 b 2

+ · q + q cosh(πν ) − 2q b b
q
4πα sinh(πν ) 0 1 0 1
∞ ∞
1 a† a 1 b† b
+ ωn an an + ana†ana + ωn an an + anb† anb , (6.6)
2 2
n=1 n=1
and that for the Neumann directions is represented by

1 1 †a a
HB = ω0 a0†a a0a + a0a a0†a + ωn an an + ana an†a
2 2
n1
1 1 b† b
+ ω0 a0b†a0b + a0b a0b† + ωn an an + anb anb† . (6.7)
2 2
n1
The Hamiltonian of fermions HF is written as

∞
µ µ
HF = ωn Sn† Sn + ωn Sn † Sn − iΨ0T ΠΩΨ0 − iΨ0 T ΠΩΨ0 . (6.8)
3 6
n=1
Now we shall evaluate the Casimir energy given as follows:
∞
∞ ∞
1 ∼ 1 2 1 1
ωn = n + ν − dk k + ν =
2 2 2 − ν + ∆(ν; 0) ,
2 2 2 2
n=1 n=1 0
∞
∞ ∞
1 ∼1 1 1
ω = n2 + ν 2 − dk k2 + ν 2 = − ν + ∆(ν ; 0) ,
2 n 2 2 2
n=1 n=1 0
in terms of zero-point energy.
The zero-point energies of a single boson with the Dirichlet condition are given by

1 ∼1 1 1 ∼1 1
(D): ωn = − ν + ∆(ν; 0) , ωn = − ν + ∆(ν ; 0) ,
2 2 2 2 2 2
n1 n1
and those with the Neumann condition are expressed as

1 ∼1 1 1 ∼1 1
(N): ωn = + ν + ∆(ν; 0) , ωn = + ν + ∆(ν ; 0) .
2 2 2 2 2 2
n0 n0
As for the zero-point energies for a fermion, we have

1 ∼ 1 ∼
E 0 = −4 · ωn = ν − 2∆(ν; 0), E 0 = −4 · ωn = ν − 2∆(ν ; 0),
2 2
n1 n1
where the factor 4 in front of the summation arises since each of fermions considered
here has four independent components. The contributions of zero-point energies to the

partition function ZF are then e−2πt (ν−2∆(ν;0)) = e−2πt ν e4πt ∆(ν;0) and e−2πt (ν −2∆(ν ;0)) =

e−2πt ν e4πt ∆(ν ;0).
We note that we have to treat carefully the zero-mode part of the Hamiltonian
HF0 = −iνΨ0T ΠΩΨ0 − iν Ψ0 T ΠΩΨ0 .
The (Ψ0 )α and (Ψ0 )α have four nonvanishing components, and hence four sets of creation
and annihilation operators S1± , S2± and S1 ± , S2 ± can be constructed. Here S1,2
+ +
(S1,2 )
− −
are creation operators and S1,2 (S1,2 ) are annihilation ones. Thus the Hamiltonian can be
rewritten as
ν + − ν
HF0 = S S − S1− S1+ + S2+ S2− − S2− S2+
2 1 1 2
ν + − ν + −
+ S1 S1 − S1 − S1 + + S2 S2 − S2 − S2 + ,
2 2

and the associated energies are ± ν2 and ± ν2 . Consequently, the contributions from these
zero-mode parts are evaluated as
2 2
e2πνt 1 − e−2πνt , e2πν t 1 − e−2πν t .
We now turn to the evaluation of the total partition function of the open string
connecting two identical Dp-branes. Let us first consider the partition function for bosons:
ZB = Tr e−2πt HB , q = e−2πt .
We will consider the sector with the mass ν = µ/3. The bosonic partition function for the
(4 − p) Dirichlet conditions (i.e., 4 − p = :(DD strings)) is given by
1
(4−p)· 21 − 12 ν+∆(ν;0)

ZB(ν) = · q · f q0a , q1a ,
n1 (1 − q ωn )4−p
where the function f (q0a , q1a ) is defined by

a a −t ν a 2 a 2
f q0 , q1 ≡ exp q0 + q1 cosh(πν) − 2q0 q1 .
a a
2α sinh(πν)
The partition function for the p Neumann directions (i.e., p = :(N-N strings)) is written as
1 1 1
ZB(ν) = · q p· 2 (+ 2 ν+∆(ν;0)).
n0 (1 − q ωn )p
Thus, the bosonic partition function on the sector with mass µ/3 is represented by
(ν) −p 1 1
ZB = 1 − q ν · · f q0a , q1a · q ν(−1+ 2 p)+2∆(ν;0)
n1 (1 − q )
ω n 4
2−p −2 a a
= 2 sinh(πνt) θ(0,0)(t; ν) · f q0 , q1 ,

where we have utilized the theta-like function: θ(a,b)(t; ν) = Θ(a,b) (it; −it; ν).
Next we shall evaluate the partition function for fermions with mass µ/3. The fermionic
partition function is given by
ZF = Tr(−1)F e−2πt HF .
After the similar calculation to the bosonic case, we obtain the fermionic partition function:
∞

2 4
ZF(ν) = eπνt − e−πνt · e−2πνt e4πt ∆(ν;0) 1 − e−2πt ωn
n=1
2
= θ(0,0)(t; ν) . (6.9)
It is an easy task to include the sector with mass µ/6, and thus the total partition function
is described by

Ztot = ZB(ν) ZF(ν) · ZB(ν ) ZF(ν )
2−p1 a a 2−p2 b b
= 2 sinh(πνt) f q0 , q1 · 2 sinh(πν t) f q0 , q1 . (6.10)
a∈D b∈D
Here the numbers p1 and p2 are Neumann directions in the coordinates x a ’s and x b ’s,
respectively. The net number of Neumann directions is represented by :(Neumann) =
p1 + p2 + 2 since + −
the x and x directions are Neumann directions in the light-cone
formulation. The i∈D means the product in terms of Dirichlet directions x i ’s.
By comparing the cylinder amplitude (5.8) obtained in the last section with the resulting
partition function (6.10) in the case of q0 = q1 = 0, we can determine the normalization
factor of boundary states. The modular S-transformation of ‘massive’ theta function
(2.18) relates the parameter µ (≡ µcl ) in the closed string to that µ(≡ µop ) in the open
string through the corresponding law: µop t = µcl [25]. Hence the normalization factor of
boundary states for the Dp-brane instanton is given by
(2−p1 )/2 (2−p2 )/2
NDp = 2 sinh(πµ/3) · 2 sinh(πµ/6) (p = p1 + p2 + 1).
(6.11)
Thus, we have shown the open/closed string duality in the N = (4, 4) type IIA string theory
at the origin. It was already shown in [25] that this duality holds in the case of the type IIB
string theory on the maximally supersymmetric pp-wave background. Although we are in
a situation of less supersymmetric case, the duality still holds at the origin.
It should be noted that the open/closed string duality holds at the origin. That is, the
open/closed string duality requires that there is no dependence on transverse coordinates
since the supersymmetry conditions require that both D-brane instantons should be at
the origin, as discussed in the paper [25]. The cylinder amplitude does not have sensible
behavior once the branes are located away from the origin.
7. General properties of partition functions of closed string
We have discussed the partition function and modular invariance of the type IIA string
theory on the pp-wave background above. In this consideration there are two sectors with
masses µ/3 and µ/6, and we have found that the modular properties hold in each sector. In
this section, motivated by this fact, we will discuss general properties of partition functions
of closed string apart from the type IIA string theory considered above. We suggest that
some characteristics of string theories on pp-waves should be fixed from the requirement
of modular invariance.
In the pp-wave case, theta-like function Θ(a,b) (τ, τ̄ , ν) should appear in the closed string
partition function. It contains a mass parameter ν and has peculiar properties under modular
transformations
Θ(a,b) (τ + 1, τ̄ + 1; ν) = Θ(a,b+a) (τ, τ̄ ; ν),

1 1
Θ(a,b) − , − ; |τ |ν = Θ(b,−a) (τ, τ̄ ; ν).
τ τ̄
Notably, the mass parameter ν changes into |τ |ν under S-transformation τ → −1/τ ,

and it gives us severe constraints in constructing modular invariant partition functions. In
other words, we can make modular invariant partition functions with this clue to go upon.
Now we will study a certain class of modular-invariant partition functions on the pp-wave
background by using the modular properties of Θ(a,b) (τ, τ̄ , ν).
To simplify the problem, we put the following ansatz:
(1) There are several kinds of mass parameters ν’s;

(2) For each ν, the partition function ZB (τ, τ̄ , ν) of the boson and that of fermion
ZF (τ, τ̄ , ν) cancel. That is to say, ZB (τ, τ̄ , ν) · ZF (τ, τ̄ , ν) = 1.
We impose the ansatz (2) because the modular S-transformation changes the mass
parameter ν into another one |τ |ν and it is generally difficult to construct modular invariant
combinations of ZB ’s and ZF ’s. In order to avoid this complicated problem, we take the
simplest ansatz ZB (τ, τ̄ , ν) · ZF (τ, τ̄ , ν) = 1 here. But we should emphasize that there
might be other modular invariant combinations without our ansatz and we cannot say there
are no other possibilities. Here we will investigate such restricted cases only and compare
our results with the models proposed earlier.
Now we will classify possible models by the use of the above ansatz. In order to
realize the condition (2), the degrees of freedom of bosons must be identical with those
of fermions. When we consider a transverse D-dimensional space, the degrees of freedom
of bosons are D (i.e., :(boson) = D). On the other hand, the degrees of freedom of spinors
in D dimensions are evaluated as

0, Majorana or Weyl,
[D ]+/
:(fermion) = 2 2 , / = −1, Majorana and Weyl,
+1, otherwise.
The value of / depends on what kinds of spinors we consider. From the consideration
of dimensionality, we can understand that the matching of degrees of freedom between
bosons and fermions happens for D = 1, 2, 4 and 8 only. For each D = 1, 2, 4, 8, the
corresponding degree of freedom is 1, 2, 4, 8. That is, we have to consider four kinds of
sets containing one, two, four, and eight bosons. Different sets are distinguished from mass
parameters ν’s. Due to these mass terms, Lorentz symmetry is broken into smaller one.
Next let us classify bosonic parts based on the Lorentz symmetry.
We study superstring theories and hence the dimension of transverse space should be
eight. For massless cases, the associated Lorentz symmetry is SO(8) and there are eight
massless bosons. However, bosons have mass terms in our massive case. We set Na as
the number of sets with a(= 1, 2, 4, 8) bosons with the same mass parameter. Let νa,i
(i = 1, 2, . . . , Na ) be mass parameters for bosons and fermions in the same set. Due to
mass terms, Lorentz symmetry is broken down to smaller one
SO(1)N1 ⊗ SO(2)N2 ⊗ SO(4)N4 ⊗ SO(8)N8 ⊂ SO(8),
with N1 + 2N2 + 4N4 + 8N8 = 8, N1 , N2 , N4 , N8 ∈ Z0 .
From this constraint, we can classify possible combinations (see Table 2) with ' =
0, 1, 2, 3, 4. Our type IIA model corresponds to the second case (N1 , N2 , N4 , N8 ) =
Table 2
N1 N2 N4 N8 Symmetry
0 0 0 1 SO(8)
0 0 2 0 SO(4) × SO(4)
0 2 1 0 SO(2) × SO(2) × SO(4)
2 1 1 0 SO(1)⊗2 × SO(2) × SO(4)
8 − 2' ' 0 0 SO(1)⊗(8−2') × SO(2)⊗'
(0, 0, 2, 0) in the list and symmetry is SO(4) × SO(4). The type IIB string theory on the
maximally supersymmetric background corresponds to (N1 , N2 , N4 , N8 ) = (0, 0, 0, 1). All
of type IIB pp-wave backgrounds with the above-mentioned bosonic isometry are found
(for example see [31]) and we can construct superstring theories on these backgrounds.
Here we turn to partition functions based on our ansatz. Each set is labeled by the mass
parameter νa,i (i = 1, 2, . . . , Na ; a = 1, 2, 4, 8). The associated partition function of boson
ZB (τ, τ̄ , νa,i ) and that of fermion ZF (τ, τ̄ , νa,i ) are written in the massive closed string
case
ZB (τ, τ̄ , νa,i ) = Θ(0,0)(τ, τ̄ , νa,i ), ZF (τ, τ̄ , νa,i ) = Θ(0,0) (τ, τ̄ , νa,i )−1 .
Then we can evaluate the total partition function Z as

Na
Z= ZB (τ, τ̄ , νa,i ) · ZF (τ, τ̄ , νa,i ) = 1.
a=1,2,4,8 i=1
It is actually modular invariant and many models proposed earlier are included in our
results.
Last we explain the result Z = 1 from the point of view of energy matching. Let ε be any
energy level of states in our string system. We also introduce nB (ε), nF (ε) as the number
of bosonic states and that of fermionic states at each energy level ε, respectively. Then the
associated partition function Z is defined as

Z = Tr(−1)F e−2πτ2 H = nB (ε) − nF (ε) e−2πτ2 ε .
ε
Here F is the fermion number operator and we also take τ1 = 0 for simplicity. By
comparing our result Z = 1, we understand following relations

1 = Z = nB (ε = 0) − nF (ε = 0) e−2πτ2 ·0 + nB (ε) − nF (ε) e−2πτ2 ε ,
ε>0
nB (ε = 0) − nF (ε = 0) = 1, nB (ε) = nF (ε) (ε > 0).
It shows that number of bosonic states and that of fermionic states match at each energy
level ε > 0. Then total energy of bosonic states EB = nB (ε) · ε is equal to that of fermionic
states EF = nF (ε) · ε at each energy level ε > 0. On the other hand, there is unbalance in
number between bosonic states and fermionic states for the ε = 0 part. But the associated
total energy of bosonic states EB0 = nB (ε = 0) · 0 = 0 equals to that of fermionic states
EF0 = nF (ε = 0) · 0 = 0 in this ε = 0 sector. So the partition function Z is nothing but
the Witten index Z = Tr(−1)F in our massive case. This fact is already known in previous
papers. Collecting these considerations, we conclude that our ansatz (2) is equivalent to
a condition (Witten index) = 1. When we impose this condition (2) on Z, the resulting
partition function is one and does not vanish. It also ensures that total energies of states at
each energy level ε match between bosonic states and fermionic states.
Our ansatz satisfies sufficient conditions to construct modular invariant partition
functions. But we do not know necessary condition for this problem. We think it is
important to find some further extra constraints in order to construct consistent string
backgrounds and classify possible strings for massive cases.
8. Conclusions and discussions
We have discussed the partition function of type IIA string theory obtained from the
eleven-dimensional theory through the S 1 -compactification of a transverse direction.
The modular invariance of our type IIA string theory has been proven. This type IIA
string theory is less supersymmetric but it is modular invariant by virtue of the cancellation
between bosonic and fermionic degrees of freedom.
We have constructed the boundary states and classified the D-brane instantons in our
theory. The resulting list of the allowed D-brane instantons is consistent with that of
the allowed D-branes obtained previously in different frameworks. In addition, we have
calculated the amplitude between D-branes in the closed and open string descriptions, and
checked the channel duality in our theory. Furthermore, we have briefly discussed general
modular properties. There are many nonmaximally supersymmetric pp-wave backgrounds,
but not all of them would give ‘modular invariant’ superstring theories. Thus, we believe
that the modular invariance is an available clue to classify the ‘physical’ string theories on
pp-waves.
Acknowledgements
K.Y. thanks Makoto Sakaguchi for useful discussions and comments. The work of H.S.
was supported by Korea Research Foundation Grant KRF-2001-015-DP0082. The work
of K.S. is supported in part by the Grant-in-Aid from the Ministry of Education, Science,
Sports and Culture of Japan (No. 14740115).
References
[1] J. Kowalski-Glikman, Vacuum states in supersymmetric Kaluza–Klein theory, Phys. Lett. B 134 (1984) 194.
[2] 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.
[3] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry,
Class. Quantum Grav. 19 (2002) L87, hep-th/0201081.
[4] R. Penrose, Any spacetime has a plane wave as a limit, in: Differential Geometry and Relativity, Reidel,
Dordrecht, 1976, p. 275.
[5] R. Gueven, Plane wave limits and T-duality, Phys. Lett. B 482 (2000) 255, hep-th/0005061.
[6] R.R. Metsaev, Type IIB Green–Schwarz superstring in plane wave Ramond–Ramond background, Nucl.
Phys. B 625 (2002) 70, hep-th/0112044.
[7] 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.
[8] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-Yang–
Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[9] K. Dasgupta, M.M. Sheikh-Jabbari, M. Van Raamsdonk, Matrix perturbation theory for M-theory on a PP-
wave, JHEP 0205 (2002) 056, hep-th/0205185.
[10] K. Sugiyama, K. Yoshida, Supermembrane on the pp-wave background, Nucl. Phys. B 644 (2002) 113,
hep-th/0206070.
[11] S. Hyun, H. Shin, Branes from matrix theory in pp-wave background, Phys. Lett. B 543 (2002) 115, hep-
th/0206090.
[12] K. Sugiyama, K. Yoshida, BPS conditions of supermembrane on the pp-wave, Phys. Lett. B 546 (2002) 143,
hep-th/0206132.
[13] K. Sugiyama, K. Yoshida, Giant graviton and quantum stability in matrix model on PP-wave background,
Phys. Rev. D 66 (2002) 085022l, hep-th/0207190.
[14] N. Nakayama, K. Sugiyama, K. Yoshida, Ground state of the supermembrane on a pp-wave, Phys. Rev. D 68
(2003) 026001, hep-th/0209081.
[15] T. Banks, N. Seiberg, S.H. Shenker, Nucl. Phys. B 490 (1997) 91, hep-th/9612157.
[16] K. Sugiyama, K. Yoshida, Type IIA string and matrix string on pp-wave, Nucl. Phys. B 644 (2002) 128,
hep-th/0208029.
[17] S. Hyun, H. Shin, N = (4, 4) type IIA string theory on pp-wave background, JHEP 0210 (2002) 070, hep-
th/0208074.
[18] S. Hyun, H. Shin, Solvable N = (4, 4) type IIA string theory in plane-wave background and D-branes, Nucl.
Phys. B 654 (2003) 114, hep-th/0210158.
[19] A. Dabholkar, S. Parvizi, Dp-branes in pp-wave background, Nucl. Phys. B 641 (2002) 223, hep-
th/0203231.
[20] S. Hyun, J. Park, H. Shin, Covariant description of D-branes in IIA plane-wave background, Phys. Lett.
B 559 (2003) 80, hep-th/0212343.
[21] O.K. Kwon, H. Shin, Type IIA supergravity excitations in plane-wave background, hep-th/0303153.
[22] S. Hyun, J.D. Park, S.H. Yi, Thermodynamic behavior of IIA string theory on a pp-wave, hep-th/0304239.
[23] Y. Sugawara, Thermal amplitudes in DLCQ superstrings on pp-waves, Nucl. Phys. B 650 (2003) 75, hep-
th/0209145.
[24] Y. Sugawara, Thermal partition function of superstring on compactified pp-wave, hep-th/0301035.
[25] O. Bergman, M.R. Gaberdiel, M.B. Green, D-brane interactions in type IIB plane-wave background,
JHEP 0303 (2003) 002, hep-th/0205183.
[26] M.R. Gaberdiel, M.B. Green, The D-instanton and other supersymmetric D-branes in IIB plane-wave string
theory, hep-th/0211122.
[27] T. Takayanagi, Modular invariance of strings on pp-waves with RR-flux, JHEP 0212 (2002) 022, hep-
th/0206010.
[28] M.R. Gaberdiel, M.R. Green, S.S. Nameki, A. Sinha, Oblique and curved D-branes in IIB plane-wave string
theory, hep-th/0306056.
[29] M.B. Green, M. Gutperle, Light-cone supersymmetry and D-branes, Nucl. Phys. B 476 (1996) 484, hep-
th/9604091.
[30] M. Billo, I. Pesando, Boundary states for GS superstrings in an Hpp wave background, Phys. Lett. B 536
(2002) 121, hep-th/0203028.
[31] M. Sakaguchi, IIB pp-waves with extra supersymmetries, hep-th/0306009.
The exact description of NS5-branes

in the Penrose limit
Konstadinos Sfetsos
Department of Engineering Sciences, University of Patras, 26110 Patras, Greece
Received 22 May 2003; accepted 23 July 2003
Abstract
We construct plane wave backgrounds with time-dependent profiles corresponding to Penrose
limits of NS5-branes with transverse space symmetry group broken from SO(4) to SO(2) × ZN .
We identify the corresponding exact theory as the five-dimensional logarithmic Conformal Field
Theory (CFT) arising from the contraction of the SU(2)N /U (1)×SL(2, R)−N exact CFT, times R5 .
We study several general aspects and construct the free field representation in this theory. String
propagation and spectra are also considered and explicitly solved in the light-cone gauge.
PACS: 11.25.-w; 11.25.Hf; 11.27.+d
1. Introduction
The purpose of the present paper is to provide and analyze the first example
of a logarithmic CFT arising as the exact description of certain supergravity brane
configurations in the Penrose limit [1] and to also study string propagation in this
background. Novel logarithmic CFTs arose recently in the large level N limit of coset
conformal field theories for compact groups combined with a free time-like boson [2]
and provided the exact description of various PP-waves configurations (with no brane
interpretation however). It was shown quite generally in [2] that, at the level of the chiral
algebra involving the compact parafermions [3,4] and the U (1)-current corresponding
to the free time-like boson, the Penrose limiting procedure gives rise to a Saletan-type
contraction (for mathematical details in the case of Lie groups, see [5]) and to a logarithmic
CFT theory [6,7] (for earlier work see [8]; for a review and more references see [9]).
E-mail addresses: sfetsos@des.upatras.gr, sfetsos@mail.cern.ch (K. Sfetsos).
doi:10.1016/j.nuclphysb.2003.07.017
104 K. Sfetsos / Nuclear Physics B 669 (2003) 103–127
This type of Penrose limits and contractions have already appeared in string theory
some years ago where they were used to construct plane wave solutions starting from
WZW model current algebra theories or gauged WZW model coset theories. The prototype
example is the plane wave solution of [10], corresponding to a WZW model for the
non-semisimple group H4 , which is the Penrose limit of the background corresponding
to the SU(2)N × U (1)−N current algebra theory [11] (for construction of plane waves
either directly as WZW and gauged WZW models or by contractions see [12–18]).
The interest in plane wave solutions arising in string and M-theory has been recently
revived following the construction of a maximally supersymmetric plane wave solution
of type-IIB supergravity [19], the fact that, it can be obtained as a Penrose limit of
the maximally supersymmetric vacuum solution, AdS5 × S 5 [20] and that, string theory
is exactly solvable in this plane wave background [21], which allows to extend the
AdS/CFT correspondence in a non-trivial way to include the effect of highly massive string
states [22]. In accordance with experience and general expectations, the superalgebra for
the AdS5 ×S 5 background contracts to the corresponding plane wave superalgebra, as it has
been explicitly demonstrated in [23]. These more recent developments are in accordance
with the fact that, the Penrose limiting procedure can be straightforwardly generalized to
supergravity theories [24].
The organization of this paper is as follows: in Section 2 we review the relevant
aspects of certain NS5-brane configurations which will serve as the starting point in our
constructions. In the supersymmetric case they have the distinct feature that they preserve
an SO(2) × ZN subgroup of the SO(4) symmetry group of the transverse to the branes
space R4 . We also review the non-extremal extension of this solution, representing the
most general non-extremal rotating NS5-brane solution. In Section 3 we systematically
construct plane wave backgrounds with time dependent profiles by performing several
different Penrose limits. In Section 4 we identify the exact CFT corresponding to these
plane waves as the contraction of the SU(2)N /U (1) × SL(2, R)−N theory. The result is
a one parameter family of five-dimensional logarithmic CFTs. For a particular member
of this family the theory degenerates into the four-dimensional current algebra for the
non-semisimple group H4 times a U (1) factor. In Section 5 we show that it is possible
to explicitly solve string theory in the light-cone gauge, even though our plane wave
backgrounds have time dependent profiles. We end our paper in Section 6 with concluding
remarks and directions for feature research.
2. The background
The most general solution of either type-II or heterotic string theory representing the
gravitational field and flux of N NS5-branes is given by [25]

5
4
ds 2 = dya2 − dt 2 + H dxi2 ,
a=1 i=1
Hij k = ij kl ∂l H,
e2Φ = H, (2.1)
K. Sfetsos / Nuclear Physics B 669 (2003) 103–127 105
where the coordinates ya ’s and t parametrize the world-volume of the branes and the xi ’s
the directions transverse to the branes space which is R4 . The function H is harmonic
in R4 . Demanding that asymptotically the space is flat we have in general

N−1
α
H =1+ , (2.2)
|x − xi |2
i=0
where xi denote the locations of the branes in R4 . Such configurations preserve sixteen
supercharges. A generic choice for the centers breaks the SO(4) symmetry completely. In
this paper we will consider NS5-brane configurations preserving an SO(2) × ZN subgroup
of SO(4). In particular, we will consider backgrounds corresponding to NS5-branes with
centers distributed at the circumference of a circle with radius r0 [26]. In the rest of this
section we review the relevant to this paper results of [26] where the reader is referred for
further details. The centers of the NS5-branes are located at
i
xi = r0 (0, 0, cos φi , sin φi ), φi = 2π , i = 0, 1, 2, . . . , N − 1. (2.3)
N
In our presentation we will strictly restrict to the decoupling, near horizon limit, where the
unit in (2.2) become unimportant. We also keep finite the energies of strings with their ends
attached between different centers. Then the harmonic function is computed to be
1
H = NΛN , (2.4)
(r 2 + r02 )2 − 4r02 ρ 2
where ρ 2 = x32 + x42 and r 2 = x12 + x22 + ρ 2 and

sinh Nχ
ΛN = , (2.5)
cosh Nχ − cos Nψ
with the angular variable ψ defined as (x3 , x4 ) = ρ(cos ψ, sin ψ) and the auxiliary variable
χ being given by

2
r 2 + r02 r + ρ02 2
e =
χ
+ − 1. (2.6)
2r0 ρ 2r0 ρ
When N 1 and r/r0 −1 1/N the branes are practically continuously distributed in the
circumference of the circle. Since then ΛN 1, the ZN symmetry becomes a continuous
SO(2) isometry. In the rest we will restrict to the continuum limit. It turns out that this is
enough for the construction of our plane wave solutions in the following section.
In order to exhibit the SO(2) × SO(2) symmetry explicitly we change variables as

x1 cos φ
= r0 sinh ρ cos θ ,
x2 sin φ

x3 cos ψ
= r0 cosh ρ sin θ , (2.7)
x4 sin ψ
where the range of the various variables is
π
0 r < ∞, 0θ , 0 ψ 2π, 0 φ 2π (2.8)
2
√
and also we rescale for future convenience the ya ’s and t by a factor of N . Then the
background (2.1) takes the form
1 2
5
ds10 = dya2 − dt 2 + dρ 2 + dθ 2
N
a=1
1
+ tan2 θ dψ 2 + tanh2 ρ dφ 2 ,
1 + tanh 2
ρ tan2 θ
1 1
Bφψ = ,
N 1 + tanh2 ρ tan2 θ

e−2Φ = e−2Φ0 cos2 θ cosh2 ρ + sin2 θ sinh2 ρ . (2.9)
∂
Performing a T-duality transformation with respect to the Killing vector ∂φ we can relate
this background to the one corresponding to the SL(2, R)−N /U (1) × SU(2)N /U (1) × R1,5
exact theory [26].1
The metric in (2.9) is singular at the location of the branes at the ring with coordinates
ρ = 0 and θ = π/2. The reason for this is that near the branes the continuous approximation
breaks down and one has to use the full solution (2.1) with harmonic function given by
(2.4). Then one indeed verifies that the metric is non-singular [26].
2.1. The non-extremal NS5-brane rotating solution
The background (2.9) is the extremal supersymmetric limit of the most general non-
extremal NS5-brane rotating solution constructed in [31] and further analyzed in [32]. In
the field-theory limit this solution has a metric
5
1 2 µ2 dρ 2
ds = − 1 − dt 2 + dya2 +
N ∆0
a=1
ρ 2 + a12 a22 /ρ 2 + a12 + a22 − µ2
1 2
+ dθ 2 + ρ + a12 sin2 θ dφ12 + ρ 2 + a22 cos2 θ dφ22
∆0
2
− µ dt a1 sin2 θ dφ1 + a2 cos2 θ dφ2 , (2.10)
∆0
1 For completeness we briefly mention that, an equivalence relation involves a Z orbifold of this theory.
N
This is seen by the fact that the background (2.1) (with the harmonic function H given by (2.4)) is invariant under
the discrete symmetry ψ → ψ + 2π/N . The ZN orbifold symmetry acts also on the φ coordinate and couples
the two factors in SU(2)/U (1) × SL(2, R)/U (1). This is analogous to the well-known equivalence between the
T-dual of the ZN orbifold of the SU(2)/U (1) × U (1) theory and the SU(2) WZW model [27] and [28] (for
a more recent discussion with emphasis on space–time properties). In agreement with these is the fact that the
SU(2) WZW model is part of (2.9) in the limit ρ → ∞. In a presumably equivalent way, the ZN orbifold is also
required after demanding mutual locality of supercurrents (GSO-projection) [29] (for more details on that see
also [30] and references therein). This orbifold will not play a direct role in taking the Penrose below since our
starting point will be the background (2.9) and not its T-dual.
an antisymmetric tensor
1 2 2
B= − ρ + a12 cos2 θ dφ1 ∧ dφ2 + µa2 sin2 θ dt ∧ dφ1
N ∆0

+ µa1 cos2 θ µ dt ∧ dφ2 , (2.11)
and a dilaton
e−2Φ = ∆0 , (2.12)
∆0 = ρ 2 + a12 cos2 θ + a22 sin2 θ (2.13)
and we have denoted by a1 , a2 the angular parameters and by µ the non-extremal

parameter. In was shown in [31] that for µ = 0 this background becomes the multicenter
continuous distribution background in (2.9) with r02 = |a12 − a22 |. Moreover, it can be
obtained by an O(3, 3) duality transformation on the background corresponding to the
SU(2) × SL(2, R)/U (1) × R 5 exact CFT. Thermodynamic properties of this solution can
be found in [31,32].
3. The Penrose limit
In this section we will construct plane waves by taking Penrose limits on the NS5
backgrounds (2.9) and (2.10).
In general, plane waves arise as classical solutions to theories of gravity and share the
attractive feature that they depart from the flat space solution in the most controllable
possible manner. This is essentially due to the existence of a covariantly constant null
Killing vector, which, as it turns out, guarantees that curvature effects are kept to a
minimum [33–39]. Plane waves, are relatively simple solutions which is reflected in the
simplicity of the equations of motions that they obey. For a background, with only NS–NS
fields turned on, of the form
ds 2 = 2 du dv + Fij (u)xi xj du2 + dxi2 ,

B = Bij (u) dx i ∧ dx j , Φ = Φ(u), (3.1)
the equations for 1-loop conformal invariance follow if the condition
1 2 d 2φ dBij
− Tr(F ) + Tr S + 2 2 = 0, Sij = , (3.2)
4 du du
is satisfied. It can be readily checked that this is indeed the case for all plane waves we
construct below. The conditions for conformal invariance at 1-loop are sufficient to prove
conformal invariance to all orders in perturbation theory [34,35].
3.1. Plane waves from continuously distributed NS5-branes
We start with the Penrose limit for the background (2.9). According to the general
prescription Penrose had suggested, we are supposed to magnify the space–time around a
null geodesic. There are three natural such geodesics dictated essentially by symmetry. The
first geodesic goes through the center of the ring corresponding to θ = 0 in our coordinate
parametrization. The second geodesic corresponds to choosing θ = π/2, therefore laying
on the plane of the ring which hits perpendicularly. Finally, the third geodesic corresponds
to ρ = 0 and hence also lays on the plane of the ring. We recall that the background (2.9)
has a singularity at ρ = 0 and θ = π/2, corresponding to a breakdown of the continuous
approximation of the NS5-brane distribution. Hence, the first geodesic never gets near the
singularity of the background and will give rise to a completely non-singular plane wave.
In contrast, the second and third geodesics hit the singularity and give rise to singular plane
wave backgrounds. Nevertheless, the origin of this singularity is well understood as arising
from the breakdown of the continuous approximation at the location of the branes. It is
remarkable that, in some sense, this is captured by the solution giving rise to a seemingly
well define string spectrum in the light-cone gauge, as we will see in Section 5.
3.1.1. Geodesic at θ = 0
We first consider the geodesic with θ = 0 that goes through the center of the ring and
has constant values for the space-like world-volume directions ya . In this case we have the
effective 3-dim metric
ds32 = −dt 2 + dρ 2 + tanh2 ρ dφ 2 . (3.3)

We search for solutions to geodesic equations with ρ, t, φ being functions of the proper
time τ . Then, energy and U (1)-charge conservation imply that
t˙ = E, tanh2 ρ φ̇ = l. (3.4)
Using these conservation laws and the nullness of the geodesic we find that
ρ̇ 2 = E 2 − l 2 coth2 ρ. (3.5)
A real solution of this obviously exists, provided that the dimensionless ratio J = l/E
obeys the condition |J | 1.
Having such a null geodesic we define a new variable u to be essentially the proper
time τ . Specifically, we choose u = Eτ and therefore the meaningful parameter will be J .
The other two variables parametrizing the directions normal to the null geodesic in the
(t, ρ, φ) space will be denoted by v and x. The appropriate change of variables in terms of
exact differentials is explicitly given by

dρ = 1 − J 2 coth2 ρ du,
√
dt = du − dv/N + J dx/ N ,
√
dφ = J coth2 ρ du + dx/ N. (3.6)
We note that the above change of variables requires that

1 1+J
ρ ln , |J | 1. (3.7)
2 1−J
Also let
z √
θ=√ , ya → ya / N (3.8)
N
and define dz2 + z2 dψ 2 = dz2 , so that z2 = z2 . The relation between the variables ρ and
u is
1
cosh ρ(u) = √ cosh 1 − J 2 u. (3.9)
1 − J2
Then the background in the limit N → ∞ takes the form

5

ds 2 = 2 du dv + dxi2 + dz2 + 1 − J 2 tanh2 1 − J 2 u dx 2 − J 2 z2 du2 ,
i=1
B12 = 2J u,

e−2Φ = e−2Φ0 cosh2 1 − J 2 u. (3.10)
Going to the Brinkman coordinates2 we find

5
1 − J2
ds 2 = 2 du dv + dxi2 + dx 2 + dz2 − 2 √ x 2
+ J
2 2
z du2 ,
i=1 cosh 1 − J u
2 2
B12 = 2J u,

e−2Φ = e−2Φ0 cosh2 1 − J 2 u. (3.13)
The parameter J takes values in the interval [−1, 1]. At the end points of this interval we
recognize two known cases. For J = 0 the non-trivial part of our background is the three-
dimensional plane wave obtained in [2,14] as the Penrose limit of the background for the
SU(2, R)N /U (1) × U (1)−N exact CFT. At the endpoints with |J | = 1 we obtain as a non-
trivial part of our background the four-dimensional plane wave of [10] constructed as a
WZW model based on the semi-simple group H4 . In our context, this background arises as
the Penrose limit of the background for the SU(2)N × U (1)−N exact CFT [11]. For general
values of J we have a non-trivial five-dimensional theory which, as we will show, it can be
obtained as a contraction of the SU(2)N × SL(2, R)−N /U (1) exact CFT.
2 If the metric has the block-diagonal form

ds 2 = 2 du dv + gi (u) dxi2 , (3.11)
i
we obtain a metric in the Brinkman form (3.1) with Fij = Fi (u)δij after performing the transformation

1 ġi 1 ġi2 1 d ġi
u → u, v→v+ , Fi = + . (3.12)
4 gi 4 g2 2 du gi
i i
3.1.2. Geodesic at ρ = 0
In this case as well the geodesic lays on the plane of the ring and can be considered in
parallel to the previous one. Skipping some of the details, we first perform the change of
variables

dθ = 1 − J 2 cot2 θ du,
√
dt = du − dv/N + J dx/ N ,
√
dψ = J cot2 θ du + dx/ N. (3.14)
Then we let
z √
ρ=√ , ya → ya / N (3.15)
N
and define dz2 + z2 dφ 2 = dz2 , so that z2 = z2 . The explicit relation between the variables
u and θ is
1
cos θ = √ cos 1 + J 2 u. (3.16)
1+ J2
Then the background in the limit N → ∞ takes the form

5

ds 2 = 2 du dv + dxi2 + dx 2 + dz2 + 1 + J 2 tan2 1 + J 2 u dx 2 − J 2 z2 du2 ,
i=1
B12 = 2J u,

e−2Φ = e−2Φ0 cos2 1 + J 2 u. (3.17)
In Brinkman coordinates it reads

5
1+ J2
ds 2 = 2 du dv + dxi2 + dz2 + 2 √ x 2 − J 2 z2 du2 ,
i=1 cos2 1 + J 2 u
B12 = 2J u,

e−2Φ = e−2Φ0 cos2 1 + J 2 u. (3.18)
We also note that (3.17) and (3.18) can be obtained from (3.10) and (3.13) by analytically
continuing as J → √ iJ , u → iu, v → −iv and x → ix. As we have mentioned,
the singularity at 1 + J 2 u = π/2 originates from the breakdown of the continuous
approximation near the location of the NS5-branes.
The plane wave, corresponding to the remaining geodesic with θ = π/2 will not be
presented in this subsection since it is a particular case of the plane waves constructed in
Section 3.2.
3.2. PP-wave limit of the non-extremal rotating NS5-brane
It turns out that, if we start with the most general rotating NS5-brane solution in the
field theory limit (2.10)–(2.12), it is enough to consider the null geodesic corresponding to
θ = π/2. For instance, null geodesics with θ = 0 and θ = π/2 are related by the symmetry
transformation of the background under θ → π/2 − θ , followed by the interchange of a1

with a2 and φ1 with φ2 . The 3-dim effective metric is then
ds32 = −Gt t dt 2 + Gρρ dρ 2 + Gφ1 φ1 dφ12 − 2Gt φ1 dt dφ1 , (3.19)
where the various functions are given by
µ2
Gt t = 1 − , G−1
ρρ = ρ + a1 a2 /ρ + a1 + a2 − µ ,
2 2 2 2 2 2 2
ρ2 + a22
ρ 2 + a12 a1
Gφ1 φ1 = , Gt φ1 = µ . (3.20)
ρ2 + a22 ρ2 + a22
The energy and U (1)-change conservation laws give
Gt t t˙ + Gt φ1 φ̇1 = 1, Gφ1 φ1 φ̇1 − Gt φ1 t˙ = J, (3.21)
with solution
Gφ1 φ1 − J Gt φ1 (ρ 2 + a22 )(ρ 2 + a12 − J µa1 )
t˙ = = ,
G2t φ1 + Gt t Gφ1 φ1 ρ 4 + (a12 + a22 − µ2 )ρ 2 + a12 a22
Gt φ1 + J Gt t (ρ 2 + a22 )(µa1 + J (ρ 2 + a22 − µ2 ))
φ̇1 = = . (3.22)
G2t φ1 + Gt t Gφ1 φ1 ρ 4 + (a12 + a22 − µ2 )ρ 2 + a12a22
In the above we understand that all of the expressions when are given in terms of general
functions Gt t etc. hold in general. Then, the nullness of the geodesic leads to
Gφ1 φ1 − 2J Gt φ1 − J 2 Gt t
ρ̇ 2 =
Gρρ (G2t φ1 + Gt t Gφ1 φ1 )

α22
= 1 + 2 (J µ − a1 )2 − J 2 a22 + 1 − J 2 ρ 2 . (3.23)
ρ
We then perform the change of variables
dρ = ρ̇ du,
√
dt = t˙ du − dv/N + J dx/ N ,
√
dφ1 = φ̇1 du + dx/ N , (3.24)
where ρ̇, t˙ and φ̇1 denote the functions in (3.23) and (3.22) and also let
π z √
θ= −√ , xi → xi / N . (3.25)
2 N
Then, in the limit N → ∞ we obtain the background
5
(a1 − J µ)2 − a22
ds = 2 du dv +
2
dya + dz + 1 − J +
2 2 2
dx 2 − J 2 z2 du2 ,
ρ 2 + a2
a=1 2
B12 = −2J u,
e−2Φ = ρ 2 + a22, (3.26)
where for the antisymmetric tensor we have used the gauge freedom in its definition.
Let us first note that if the relation between the various parameters (a1 − J µ)2 = a22
holds, then we have the case of the four-dimensional plane wave of [10] times R6 . Also, if
we shift a1 → a1 +J µ the effect of non-extremality disappears from the above background
as well as from the differential Eq. (3.23). Solving the latter we obtain a metric of the form
(3.1) with diagonal Fij . In all cases F1 = F2 = −J 2 , whereas the function Fx and the
dilaton depend on the range of various parameters. We have three different cases (we use
the shifted a1 below):
a12 > a22 and J 2 < 1: then

Gxx = 1 − J 2 coth2 1 − J 2 u, e−2Φ = e−2Φ0 sinh2 1 − J 2 u. (3.27)
a12 < a22 and J 2 < 1: then

Gxx = 1 − J 2 tanh2 1 − J 2 u, e−2Φ = e−2Φ0 cosh2 1 − J 2 u. (3.28)
a12 > a22 and J 2 > 1: then

Gxx = J 2 − 1 cot2 J 2 − 1 u, e−2Φ = e−2Φ0 sin2 J 2 − 1 u. (3.29)
All cases are related by analytic continuations of the various parameters and variables.
Passing to the Brinkman coordinates we obtain a metric of the form

5

ds 2 = 2 du dv + dya2 + dx 2 + dz2 + Fx x 2 − J 2 z2 du2 , (3.30)
a=1
with
1− J2
Fx = 2 √ ,
sinh2 1 − J 2 u
1 − J2
Fx = −2 √ ,
cosh2 1 − J 2 u
J2 − 1
Fx = 2 2 √ , (3.31)
sin J 2 − 1 u
corresponding to the three cases above, respectively. Note that the last two cases correspond
to plane waves identical to the ones we obtained in Section 3.1 (up to a trivial redefinition
of u in the trigonometric case). In addition, it can be checked that the first case is identical
to the plane wave corresponding to the geodesic with θ = π/2 in the background (2.9).
We started with the non-supersymmetric, non-extremal rotating NS5-background
(2.10)–(2.12) and obtained after the Penrose limit was taken, the same plane waves as those
we obtained from the supersymmetric extremal background (2.9) in the Penrose limit. The
physical reason for this is related to the fact that global information about a space–time is
lost after the Penrose limit is taken, since essentially we focus and expand the space–time
seen by a particular null geodesic. Hence, there is no notion of a horizon in the plane wave
geometry and of the Hawking temperature associated with it. This is related to the apparent
loss of holography in the Penrose limit (for discussion and some work in these directions
see, for instance, [40,41]). We believe that holography is present, but encoded in a string
theoretical way beyond the supergravity approximation. It can be explicitly checked, using
results of [31,32], that the Hawking temperature for the metric (2.10) goes to zero in the
Penrose limit. Hence, states in the theory whose mass degeneracy is broken due to thermal
effects, acquire again the same mass, resulting into a restoration of supersymmetry. Having
an exact CFT description in our case must be very important in working out the details of
such a scenario.
4. The relation to exact CFT
In this section we relate the plane waves (3.13) and (3.18) to backgrounds corresponding
to exact CFTs through a Penrose limit. We also construct the corresponding symmetry
algebra, via the associated with the Penrose limit, contraction.
We start with the backgrounds corresponding to the SU(2)N × SL(2, R)N /U (1) × R5
exact CFT
5
ds 2 = N dθ 2 + dφ 2 + dψ 2 + 2 cos θ dφ dψ + N −dt 2 + tanh2 t dω2 + dxi2 ,
i=1
Bφψ = N cos θ,
e−2φ = e−2Φ0 cosh2 t. (4.1)
Consider the change of variables
v
θ= + 2J u, t = 1 − J 2 u,
2J N

1 1 N/N
φ = √ (x1 + x2 ), ψ = √ (x1 − x2 ), ω= x (4.2)
2 N 2 N N + N
and take the limit
N
N, N → ∞, J2 = = finite. (4.3)
N + N
We obtain

5
ds 2 = 2 du dv + dxi2 + dx 2 + cos2 J u dx12 + sin2 J u dx22
i=1

+ 1−J 2
tanh2 1 − J 2 u,
B12 = cos 2J u,

e−2Φ = e−2Φ0 cosh2 1 − J 2 u. (4.4)
Passing to the Brinkman coordinates we obtain (3.13). We also note that, had we started
with the background for SU(2)N /U (1) × SL(2, R)−N × R5 exact CFT
5
ds 2 = N −dt + dφ 2 + dψ 2 + 2 cosh t dφ dψ + N dθ 2 + tan2 θ dω2 + dxi2 ,
i=1
Bφψ = N cosh t,
e−2φ = e−2Φ0 cos2 θ, (4.5)
we would have obtained in a Penrose limit, similar to (4.2), (4.3), the plane wave (3.18).
In order to construct the symmetry algebra for our planes waves (3.13) and (3.18) we
should start with the symmetry algebras of the original backgrounds before the Penrose
limit was taken. We do that explicitly for the plane wave in (3.18) and start with the
symmetry algebra corresponding to the background (4.5). The R5 factor corresponds to
five free currents and therefore will be ignored in our discussion. For the WZW model
factor SL(2, R)−N we have a current algebra theory generated by the three currents J0 , J±
of conformal dimension ∆ = 1 which obey the operator product expansions (OPEs)
N/2
J0 (z)J0 (w) = − + regular,
(z − w)2
±J± (w)
J0 (z)J± (w) = + regular, (4.6)
z−w
2J0 (w) N
J+ (z)J− (w) = − + + regular. (4.7)
z−w (z − w)2
The symmetry algebra corresponding to the SU(2)N /U (1) factor is the parafermionic
algebra of [3]. This algebra is generated by chiral operators ψl , l = 0, 1, . . . , N − 1,
with the complex conjugation acting as ψ † = ψN −l . Their conformal dimension is
∆l = l(N − l)/N and hence it differs from the classical result ∆clas
l = l (for finite l).
This will have very important consequences for the nature of the exact CFT that follow
from the contraction in the N, N → ∞ limit. Here we are particularly interested for the
two parafermionic operators ψ1 and its complex conjugate ψ1† , with the lowest conformal
dimension ∆1 = 1 − 1/N . Their OPEs are

2(1 − 1/N )
ψ1 (z)ψ1 (w) = ψ2 (w) + O(z − w) ,
(z − w) 2/N

† 1 2
ψ1 (z)ψ1 (w) = 1 + 1 + (z − w) Tpar. (w) + O(z − w) , (4.8)
2 2
(z − w)2∆1 N
where Tpar. is the Virasoro stress energy tensor of the parafermionic theory.
4.1. The contraction
The quantum OPEs that we have exhibited, have their classical counterparts in terms
of Poisson brackets. These are obeyed by the classical versions of the currents J0 , J± and
of the parafermions ψ1 and ψ1† which are realized in terms of the space–time variables
t, φ, ψ, θ and ω. Hence, the Penrose limit for the background (3.18) will give rise to
a limiting procedure for the corresponding symmetry generating classical objects. This
is a quite straightforward, but lengthly, procedure and, as it turns out, it gives rise to a
contraction of the Poisson bracket algebra that the symmetry generators obey. At this more
abstract level the result can be taken over and be applied to the quantum case, where OPEs
replace the classical Poisson brackets. For the special case of the plane wave obtained as
a Penrose limit of the background for SU(2)N /U (1) × U (1)−N exact CFT this procedure
was carried out in full detail in [2]. Here we skip all details and only quote the end result
dealing directly with the quantum case. We define new operators Φ, Ψ, P , P± as
1/2 i 1 J
1− J2 Φ= √ ψ1 − ψ1† − √ J0 ,
2 N N 1− J2
√

2 −1/2 i N J
1−J Ψ= ψ1 + ψ1† + √ J0 ,
2 1 − J2
ψ1 + ψ1†
P= ,
√2
P± = N J± . (4.9)
In the limit (4.3), all fields are primaries of the total Virasoro stress tensor with conformal
dimension equal to one, except for Ψ for which we find
Ψ (w) − 12 (1 − J 2 )Ψ (w) ∂Ψ (w)
T (z)Ψ (w) = + + regular. (4.10)
(z − w)2 z−w
Hence, Ψ is not a primary field unless J 2 = 1. In addition, we find that
1
Ψ (z)Ψ (w) = , Φ(z)Φ(w) = regular. (4.11)
(z − w)2
These are the defining relations of a logarithmic CFT [6]. The essential reason that a
logarithmic structure for the CFT arose is that the original theory contains as basic chiral
operators the lowest parafermions with conformal dimension 1 − 1/N as well as the
SL(2, R)−N currents with dimension exactly 1 for all N . Hence, although classically all
operators have dimension 1, quantum mechanically this dimension is protected only in the
case of the SL(2, R)−N currents. The remnant of this, in the limit N , N → ∞, is the fact
that Ψ is not a primary field. This phenomenon did not occur in the similar contraction of
current algebra theories [12,13] since the combinations of currents that one forms in order
to take the limit N → ∞, have conformal dimension 1 for all N .
The algebra of the logarithmic partners (Φ, Ψ ) in (4.11) is enhanced by the presence of
non-trivial OPEs corresponding to the other fields in the theory. We find
1 Φ(w)
P+ (z)P− (w) = +J + regular,
(z − w) 2 z−w
P± (w)
Ψ (z)P± (w) = ±J + regular,
z−w

Ψ (z)P (w) = − 1 − J 2 ln(z − w)(P Φ)(w) + regular,
1/2 1 − J2
P (w)P (w) = + ln(z − w)Φ 2 (w) + regular,
(z − w)2 2

ln(z − w) 2
Ψ (z)Ψ (w) = 1 − J 2 + 2 ln(z − w) P (w)
(z − w)2
(1 − J 2 )2 2
+ ln (z − w)Φ 2 (w) + regular. (4.12)
2
For J = 0 the operators P± become Abelian currents and decouple from Φ, Ψ and P . The
later obey the three-dimensional logarithmic CFT found in [2]. For |J | = 1 the operator P
becomes an Abelian current and decouples from Φ, Ψ and P± which, then, obey the
current algebra based on the non-semisimple group H4 [10]. In this case we also note
that Ψ is a primary field. This limiting behavior is in agreement with the behavior we have
already noted for the background (3.13).
A quite straightforward generalization of our results to higher-dimensional cases
follows by replacing the coset factor SU(2)N /U (1) with any compact coset CFT. For the
particular case of SO(D + 1)N /SO(D), this is immediate using the results of the second
referene in [2].
4.2. Free fields
In order to reproduce the operator algebra (4.12) using free fields, we utilize the free
field realizations of the coset model SU(2)N /U (1) and of the SL(2, R)−N current algebra
at finite N, N which are known [42]. We start with the former model and introduce two
real free bosons,

φi (z)φj (w) = −δij ln(z − w), i, j = 1, 2. (4.13)
One may represent the elementary parafermion currents as3
1 √
ψ1 = √ − 1 + 2/N ∂φ1 + i∂φ2 e+ 2/N φ2 ,
2
1 √
ψ1† = √ 1 + 2/N ∂φ1 + i∂φ2 e− 2/N φ2 . (4.14)
2
For the free field realization of the current algebra for SL(2, R)−N we need a time-like free
boson φ0 and two space-like free bosons ϕi , i = 1, 2. They obey

ϕi (z)ϕj (w) = −δij ln(z − w), i, j = 1, 2,

φ0 (z)φ0 (w) = ln(z − w). (4.15)
Then the currents are given by4

N √
J± = ∓ 1 − 2/N ∂ϕ1 + i∂ϕ2 e∓i 2/N (φ0 −ϕ2 ) ,
2

N
J0 = i ∂φ0 . (4.16)
2
3 All expressions appearing in the rest of this paper to involve products of free bosons and their derivatives,
are understood as being properly normal ordered.
4 These are of course nothing but the free field realization of the non-compact parafermions [43] dressed with
the extra boson φ0 .
The stress-energy tensor of the entire five-dimensional theory is

1 1 i
T (z) = − (∂φ1 )2 − (∂φ2 )2 − √ ∂ 2 φ1
2 2 2(N + 2)
1 1 1 1
+ (∂φ0 )2 − (∂ϕ1 )2 − (∂ϕ2 )2 + √ ∂ 2 ϕ1 (4.17)
2 2 2 2(N − 2)
and corresponds to a central charge c = 3N /(N + 2) − 1 + 3N/(N − 2) theory, as it
should be. Let us now consider the scalar field redefinition

2 1/2 1
2 −1/2 N
1−J φ+ = √ (φ0 + φ1 ), 1−J φ− = (φ0 − φ1 ),
2N 2
1
φ2 = φ, ϕ± = √ (∓ϕ1 + iϕ2 ). (4.18)
2
Then, the new set of scalars obey

φ+ (z)φ− (w) = ϕ+ (z)ϕ− (w) = − φ(z)φ(w) = ln(z − w) (4.19)
and have zero correlators otherwise. In order to take the N, N → ∞ limit, the expansion
√
N 1
ψ1 = − ∂φ+ + √ (i∂φ − φ∂φ+ )
2 2

1 1 1 2 1
+√ ∂φ− + iφ∂φ − φ + 1 ∂φ+ + O , (4.20)

N 2 2 N
and its conjugate one for ψ1† are useful. After some algebra we find the following free field
representation for the basic operators of our five-dimensional logarithmic CFT
Φ = −i∂φ+ ,
P± = ∂ϕ± e∓iJ φ+ ,
1
P = √ i∂φ − 1 − J 2 φ∂φ+ ,
2
i
Ψ = i∂φ− − 1 − J 2 φ 2 + 1 ∂φ+ − 1 − J 2 φ∂φ. (4.21)
2
The Virasoro stress energy tensor expressed in terms of free fields becomes
1 i
T (z) = ∂φ+ ∂φ− − (∂φ)2 + ∂ϕ+ ∂ϕ− − 1 − J 2 ∂ 2 φ+ (4.22)
2 2
and corresponds to a central change c = 5 theory. It is straightforward to check that these
obey the OPE in (4.12). We also note that for J = 0 we obtain the free field representation
of the three-dimensional logarithmic CFT of [2] and for |J | = 1 that of the current algebra
for the non-semisimple group H4 found in [44].
Using (4.9), the known correlation functions for the currents and of the parafermions
and then taking the contraction limit, we may compute correlation functions for the
basic operators of our theory Φ, Ψ, P and P± . Alternatively, we may use the free field
representation we have derived. These computations will not be presented here, but the
interested reader may get an idea of their form from the corresponding correlators in [2],
where the parameter J = 0.
5. Strings in light-cone gauge
The purpose of this section is to explicitly solve for the string spectrum in the light-
cone gauge. This is possible even though our plane wave backgrounds have u-dependent
profiles. String theory on plane wave backgrounds with only NS–NS fields turned on and
time dependent profiles of the form 1/u2 have been solved in the light-cone gauge in [45,
46]. For extensive discussions of string on other NS–NS backgrounds the reader is referred
to [47,48].
5.1. General formalism
Let us first review the light-cone approach to string propagation in plane backgrounds
(see, for instance, [34,35]) suitably fitted for our purposes. We will consider bosonic
backgrounds of the form (3.1) with diagonal Fij = Fi δij and zero antisymmetric tensor.
This contains the case with J = 0 for our backgrounds (3.13) and (3.18). The extension to
cases with J = 0 is easily done and will be briefly mentioned at the end of Section 5.2. We
will consider the variable u to be non-compact and let, as usual, u = P τ . The light-cone
action for the transverse coordinates is

1
S= dτ dσ ∂τ Xi ∂τ Xi − ∂σ Xi ∂σ Xi + P 2 Fi Xi . (5.1)
4π
The classical equations of motion are
δXi : − ∂τ2 Xi + ∂σ2 Xi + P 2 Fi Xi = 0 (5.2)

and
δS 1
Πi = = ∂τ Xi , (5.3)
δ Ẋi 2π
denotes the conjugate to Xi momenta. Using these the Hamiltonian takes the form
2π
1
H= dσ ∂τ Xi ∂τ Xi + ∂σ Xi ∂σ Xi − P 2 Fi Xi . (5.4)
4π
0
We expand the string coordinates in Fourier modes as
1
Xi (σ, τ ) = X0i (τ ) + i einσ ani Xni (τ ) − ã−n
i
Xni ∗ (τ )
n
n=0
1
= X0i (τ ) + i Xni (τ ) ani einσ + ãni e−inσ
n
n=0
∞
1 i
= X0i (τ ) + i Xn (τ ) ani einσ + ãni e−inσ
n
n=1
1 i i −inσ
− X−n (τ ) a−n e + ã−n
i
einσ , (5.5)
n
∗
where we have used a ∗ in = a−n i
and Xni = X−n i
, which follows from the reality condition
for the Xi ’s. It will be convenient to identify a complex basis for the zero mode solution as
∗ ∗
X0i (τ ) = a i χ i (τ ) + a i χ i (τ ). (5.6)
Then using (5.2) we find that the amplitudes obey the harmonic oscillator equations with
τ -dependent frequencies
2 2
Ẍni + ωni Xni = 0, χ̈ i + ω0i χ i = 0,
i 2
ωn = n2 − P 2 Fi , (5.7)
where dots will denote ordinary derivatives with respect to τ . It is particularly useful to
think of (5.7) as a Schrödinger equation
−Ẍni + Vi Xni = n2 Xni , (5.8)
where the potential is given by Vi = P 2 Fi .
For each i, the differential equation (5.7) has
two independent solutions. Their Wroskian will be normalized as
∗
Ẋni Xni − Xni Ẋni∗ = in, n = 0,
i
χ χ̇ i∗
− χ̇ χ
i i∗
= i. (5.9)
In a canonical quantization, one promotes Xi , Π i , ani , ãni and a i , a i∗ to operators and starts
from the basic commutators
i

X (σ, τ ), Xj (σ , τ ) = Π i (σ, τ ), Π j (σ , τ ) = 0,
i
X (σ, τ ), Π j (σ , τ ) = iδ ij δ(σ − σ ). (5.10)

These give rise to
i j

an , am = nδ ij δn+m , a i , a j † = δ ij , (5.11)
provided that the normalization (5.9) is chosen. Replacing the expansion (5.5) into (5.4)
we obtain for the Hamiltonian
H = H0 + Hstring, (5.12)
where H0 is the part of the Hamiltonian corresponding to the zero mode
1 i i i 2 i i
H0 = P P + ω0 X0 X0
2 0 0
1 2
1 2
2
= χ̇ i χ̇ i + ω0i χ i χ i a 2 + χ̇ i∗ χ̇ i∗ + ω0i χ i∗ χ i∗ a †
2 2
i i∗ i 2 i i∗
i† i 1
+ χ̇ χ̇ + ω0 χ χ a a + . (5.13)
2
The relation to position and momentum operators is
† ∗
X0i = a i (τ )χ i + a i (τ ) χ i ,
†
P0i = a i (τ )χ̇ i + a i (τ ) χ̇ i∗ . (5.14)
Since in the Schrödinger picture X0i and P0i are time-independent operators we have that
da i ∂a i
= + i H0 , a i = 0, (5.15)
dτ ∂τ
which can be proved by using (5.14) and the equation of motion for χ i . We have denoted
by Hstring the part of the Hamiltonian containing the string oscillation modes
∞
1 1 i i 2 i i
Hstring = 2
Ẋn Ẋ−n + ωni Xni X−n
i
an a−n + ãni ã−n
i
2 n
n=1
i i i 2 i i i i i 2 i i i
− Ẋn Ẋn + ωn Xn Xn an ãn − Ẋ−n Ẋ−ni

+ ωni X−n i
X−n a−n ã−n .
(5.16)
In the rest of this subsection we suppress for simplicity the index i.
It is convenient to write χ = reiφ and Xn = rn eiφn and then consider the equations for
the amplitudes r, rn and the phases φ and φn . Using (5.7) and (5.9) we find that
1 1
r̈ + ω02 r − = 0, φ̇ = − , (5.17)
4r 3 2r 2
as well as
n2 n
r̈n + ωn2 rn − = 0, φ̇n = . (5.18)
4rn3 2r 2
Let us also note that the stringy-part of the Hamiltonian is not diagonal, but it can be
diagonalized by means of the transformation

An fn f−n an
= ∗ (5.19)
Ã†n −f−n −fn∗ ã−n
and its conjugation, where

1 1 n
fn = √ (Ẋn + iωn Xn ) = √ e iφn
ṙn + i ωn rn + 2 . (5.20)
n 2ωn n 2ωn 2rn
Then we have the commutation rules

An , A†m = δn−m , Ãn , Ã†m = δn−m (5.21)

and zero otherwise. Then the expansion (5.5) becomes
∞
1 inσ
X(σ, τ ) = X0 (τ ) + i √ e An − Ã†n + e−inσ A†n − Ãn . (5.22)

n=1
2ωn
In addition, (5.16) assumes the diagonal form
∞
1 †
Hstring = ωn An An + Ã†n Ãn . (5.23)
2
n=1
The previous discussion was purely bosonic and in the supersymmetric case should
be supplemented by the inclusion of world-sheet fermions. In addition, we have the level
matching conditions arising from the expression for the light-cone variable v in terms
of the transverse coordinates Xi . As we are interested in some rather generic features
for string propagation in our backgrounds such discussions will be omitted in this paper.
We also note that in the case of√constant frequency ωn2 = n2 we have, according to our
normalizations, that Xn = einτ / 2 and we recover familiar results for strings propagation
in flat Minkowski space–time in the light-cone gauge.
5.1.1. The particle-like spectrum

Let us discuss in some detail the particle spectrum corresponding to the zero-mode
Hamiltonian H0 . The problem at hand is to determine a complete set of solutions to the
time-dependent Schrödinger equation
n (x, τ ) = 0,
(i∂τ − H0 )Ψ (5.24)
where H0 is given in (5.13). This Hamiltonian is that for a quantum oscillator with
time dependent frequency ω0 (τ ). It is possible to solve (5.24) for any ω0 (τ ).5 Using the
oscillator operators a and a † we can define, at a given τ (and i), a Fock space of states as
usual
1 n
a|0, τ = 0, |n, τ = √ a † |0, τ . (5.25)
n!
In configuration space the eigenstates Ψn (x, τ ) = x|n, τ can be constructed in terms
of Hermite polynomials. Using the representation of a and a † as first order differential
operators we find that the normalized to one zero-mode eigenstate satisfies
∗
a|0, τ = 0 ⇒ χ ∂x − i χ̇ ∗ x Ψ0 = 0 ⇒
√ −1/2 i χ̇ ∗ x 2
Ψ0 (x, τ ) = 2π |χ| e 2χ∗ . (5.26)
Then using (5.25) and the defining differential relation for Hermite Polynomials, we find
that in configuration space6

√ −1/2 χ n/2 i χ̇ ∗ x 2 x
Ψn (x, τ ) = 2π 2 2 |χ|
n n
e 2χ∗ Hn √
χ∗ 2 |χ|

√ −1/2 (i ṙ
− 1
)x 2 x
= 2π 2 n!r
n inφ
e e 2r 4r 2 Hn √ , (5.27)
2r
where in the last step we have used (5.17). The advantage of this space is that we can write
down immediately states |n̄, τ that obey the operator equation (5.24), simply as
n (x, τ ) = eiθ(τ ) Ψn (x, τ ),
Ψ (5.28)
where the phase θ (τ ) is independent of the quantum number n and is determined by
integrating the equation
θ̇ = −ṙ 2 − ω02 r 2 . (5.29)
5 For the description of the method see [49] and references therein. We also note that the method has been
used recently in the present context in [45,50].
6 We correct below an apparent typo in Eq. (A8) of [49].
√
We note that in the case of constant frequency we have: χ = e−iω0 τ / 2ω0 and θ =
− 12 ω0 τ . Then the total time dependence in (5.28) is in a phase factor which reads
−ω0 (n + 12 )τ , as it should be.
5.2. Application for our plane wave backgrounds
We now apply the general formalism for our backgrounds (3.13) and (3.18). For
the modes corresponding to the spatial brane directions ya we get the expected
eigenfrequencies ωan 2 = n2 . For the eigenfrequencies corresponding to the transverse
coordinates zi , i = 1, 2 the general formalism we have developed is not adequate since we

assumed zero antisymmetric tensor field. However, it is easy to see that we get a system of
two coupled differential equations
d 2 zin 2
2
+ n + P 2 J 2 zin = 2iP J nij zj n , (5.30)
dτ
where the term on the right-hand side is due to the non-vanishing antisymmetric tensor.
We easily see that the combinations z±n = z1n ± iz2n diagonalize the system. The
corresponding eigenfrequencies are
2
ω±n = (n ± P J )2 (5.31)
and coincide with those found in [51] for the light-cone treatment of strings in the plane
wave background of [10] (for a recent discussion that takes into account the world-sheet
fermions see [52]).
In the rest of this subsection we set the parameter J = 0 and concentrate on the three-
dimensional plane wave, arising from the non-trivial parts of (3.13) and (3.18).
For the modes corresponding to the coordinate x we get a Schrödinger equation as in
(5.8) with potential that depends on the specific plane wave background. In particular, in
the case of the background (3.13) we have
P2
V = −2 . (5.32)
cosh2 (P τ )
This belongs to a class of reflectionless potentials which can support a finite number
of bounds states. The reason for this is that this potential and the corresponding
eigenvalue problem is related to the problem with constant potential within the context
of supersymmetric quantum mechanics [53]. In our case there is exactly one bound state.
However, it is not admissible to the perturbative string spectrum since it corresponds to
setting n equal to the imaginary unit and that destroys the periodicity in the spatial string
world-sheet variable σ . The explicit solution for the states is (we follow the normalization
of (5.9))
1 in − P tanh aτ inτ
xn (τ ) = √ e , n = 0, − ∞ < τ < +∞, (5.33)
2 in + P
which clearly exhibits the reflectionless behavior of the potential since the wavefunctions
at τ → −∞ and at τ → +∞ differ only by a phase factor. Hence in our plane-wave
background we dot not have string-mode creation. For the string part of the Hamiltonian
we get
∞
1
Hstring = Γn (an a−n + ãn ã−n ) + ∆n an ãn + ∆∗n a−n ã−n , (5.34)
2
n=1
where
P2 1 P4 1
Γn = 1 + − ,
n2 cosh2 P τ 2n2 (P 2 + n2 ) cosh4 P τ
1 P 3 e2inτ P cosh 2P τ − in sinh 2P τ
∆n = . (5.35)
2 n2 cosh4 P τ (P + in)2
For τ → ±∞ this part of the Hamiltonian becomes diagonal and takes the same form as
in the case of strings in flat space, i.e., Hstring(±∞) = 12 ∞ n=1 (an a−n + ãn ã−n ).
For the zero mode we have the, properly normalized, solution
√ √
χ(τ ) = (2 2 P )−1/2 (P τ tanh P τ − 1) + i( 2 P )−1/2 tanh P τ, (5.36)
from which we deduce the expressions for the amplitude r(τ ) and the phase φ(τ )

1 1
r2 = √ (P τ tanh P τ − 1)2 + tanh2 P τ ,
2P 2
√
2 tanh P τ
tan φ = . (5.37)
P τ tanh P τ − 1
This Hamiltonian has of course the form (5.13) where the coefficients of the quadratic
operators are quite complicated to written down explicitly. We note that the constants of
integration in (5.36) have been √ chosen so that the Hamiltonian at τ = 0 is diagonal and
has the form H0 (τ = 0) = 2 P (a † a + 12 ). With this choice the solution at τ → ±∞
represents outgoing freely moving particle states.
Of particular interest is the amplitude to find the particle at a state Ψ n (x, τ ) having
started at τ → −∞, where the potential vanishes, as a plane wave, i.e., Ψ pin (x, −∞) =
√1 e ipx , where p is the momentum. The matrix element describing this is given by
2π
in
n (τ )
Ψ Ψp (−∞)
∞
√ −1/2 −i(nφ+θ) ipx−(i 2rṙ + 12 )x 2 x
= 2π 2π 2 n!r n
e dx e 4r Hn √ . (5.38)
2r
−∞
The integral is computed using the generating function for Hermite polynomials. We find
in
n (τ )
Ψ Ψp (−∞)
1/4
2 r −i(nφ+θ) i 2p222r 3 ṙ 2r ṙ + i n/2 − 1 ξp2
= e e 4r ṙ +1 e 2 Hn (ξp ), (5.39)
π 2n n! 2r ṙ − i
where ξp is a time-dependent function
√
2 rp
ξp (τ ) = √ . (5.40)
4r 2 ṙ 2 + 1
Therefore the corresponding amplitude is

1/2
in 2 2 r −ξp2 2
Ψ n (τ )
Ψp (−∞) = e Hn (ξp ). (5.41)
π 2n n!
This expression is completely general and holds for all time-dependent frequencies that go
to zero at τ → −∞. In our case we may find the explicit τ -dependence of the solution
by computing ξp using (5.37). The result is not particularly simple and will not be written
down here.
We end this section by briefly discussing the case of strings propagating in the
background (3.18), in the light-cone gauge, which has some distinct features. In this case
the modes corresponding to the coordinate x obey a Schrödinger equation with potential
P2
V =2 , 0 P τ π, (5.42)
sin2 (P τ )
where we have shifted P τ by π/2. Similarly to (5.32), the Schrödinger problem with this
potential is related, via supersymmetric quantum mechanics, to the same problem with
constant potential in the same finite interval for τ . It turns out that the wavefunctions that
vanish at τ = 0 and τ = π/P are

xn,m = (m + 2)P cos (m + 2)P τ − P cot P τ sin (m + 2)P τ ,
m = 0, 1, . . . , (5.43)
provided that n2 = (m + 2)2 P 2 .
Hence the light-cone momentum P is quantized in order
for a non-trivial solution to exist and in particular it should be a rational number. Given
such a number, for each value m there is only one string mode n, given by the above
relation, that can be excited. We also note the absence of the zero mode corresponding to
n = 0. The phenomenon of the quantization of the light-cone parameter P in the present
context was found before in plane waves constructed from continuously distributed D3-
branes [54]. Clearly, restricting to the finite interval τ ∈ (0, π/P ) is related to the fact the
singularities of the metric in (3.18), where the light-cone gauge breaks down. We expect
that a covariant quantization of the string will shed light on this issue.
6. Concluding remarks
We constructed plane wave backgrounds corresponding to Penrose limits of NS5-

branes. The latter have a transverse space symmetry group SO(2) × ZN ∈ SO(4) and give
rise to time-dependent profiles for the plane wave solutions. We identify the corresponding
exact theory as the five-dimensional logarithmic CFT arising from the contraction of
the SU(2)N /U (1) × SL(2, R)−N exact CFT, times R5 . We constructed a free field
representation for this theory and studied explicitly string propagation and spectra in the
light-cone gauge. In view of the delicate issues concerning the validity of a uniform choice
for the light cone gauge and states with light-cone momentum P = 0, it will be desirable
to also develop the covariant approach and construct string scattering amplitudes. Work
towards this direction has been undertaken for the plane wave of [10], corresponding to
|J | = 1 and a current algebra in our class of models, in [44] and more recently in [55].
It will be desirable to develop a similar approach at least for the case with J = 0,
corresponding to a purely logarithmic CFT.
One of our motivations in considering the Penrose limits of brane distributions was
to extend the considerations of [22] for the sector of N = 4 SYM with high spin and
dimension operators, to cases away from conformality as the starting point. In particular,
one way to break the conformal symmetry is by means of the six scalars in the theory
acquiring vacuum expectation values (vev’s). We expect that in such cases the BMN
operators are the same, but they have to act on a different vacuum. On the supergravity
side the centers of the D3-branes are distributed in the six-dimensional transverse to the
branes space, according to the distribution of vev’s on the gauge theory side, resulting
into a deformation of the AdS5 × S 5 space. A particular such case was considered in [54],
where plane wave solutions were obtained via Penrose limits on the solution representing
D3-branes uniformly distributed on a disc. In this case it was possible to obtain, in the limit
of very large vev’s, the perturbative string spectra in the light-cone gauge. However, there
were problems related to the singularity arising due to the breakdown of the continuous
approximation, similar to the case of the background (3.18) we have discussed in the
present paper. However, if we start with D3-branes distributed on a circle of radius r0 ,
instead of a disc, it is possible to obtain a plane wave which is completely non-singular by
choosing, as in the case of (2.9), a geodesic that goes through the center of the ring. The
result is quite simple and here we just mention the case with zero angular parameter. This
is purely geometrical since, as it turns out, the metric is given by

8
2
ds10 = 2 du dv + dxi2
i=1
du2
+ 3 x12 + x22 − x32 − x42 − x52 + x62 + x72 + x82 , (6.1)
(1 + u2 )2
and the self-dual five-form is zero! This solution may well serve as a starting point for
investigating the Coulomb branch of N = 4 SYM in a particular sector but beyond the
supergravity limit.
Acknowledgements
I would like to acknowledge the financial support provided through the European
Community’s Human Potential Programme under contracts HPRN-CT-2000-00122 “Su-
perstring Theory” and HPRN-CT-2000-00131 “Quantum Structure of Space–Time”, the
support by the Greek State Scholarships Foundation under the contract IKYDA-2001/22
“Quantum Fields and Strings”, as well as NATO support by a Collaborative Linkage Grant
under the contract PST.CLG.978785 “Algebraic and Geometrical Aspects of Conformal
Field Theories and Superstrings”.
References
[1] R. Penrose, Any space–time has a wave as a limit, in: Differential Geometry and Relativity, Reidel,
Dordrecht, 1976.
[2] I. Bakas, K. Sfetsos, Nucl. Phys. B 639 (2002) 223, hep-th/0205006;
K. Sfetsos, Phys. Lett. B 543 (2002) 73, hep-th/0206091.
[3] A.B. Zamolodchikov, V.A. Fateev, Sov. Phys. JETP 62 (1985) 215.
[4] K. Bardacki, M.J. Crescimanno, E. Rabinovici, Nucl. Phys. B 344 (1990) 344.
[5] R. Gilmore, Lie Groups, Lie Algebras and some of their Applications, Wiley, 1974.
[6] V. Gurarie, Nucl. Phys. B 410 (1993) 535, hep-th/9303160.
[7] A. Bilal, I. Kogan, Nucl. Phys. B 449 (1995) 569, hep-th/9503209;
J.-S. Caux, I. Kogan, A. Tsvelik, Nucl. Phys. B 466 (1996) 444, hep-th/9511134;
I.I. Kogan, N.E. Mavromatos, Phys. Lett. B 375 (1996) 111, hep-th/9512210;
A.M. Ghezelbash, M. Khorrami, A. Aghamohammadi, Int. J. Mod. Phys. A 14 (1999) 2581, hep-th/9807034;
I. Kogan, A. Nichols, JHEP 0201 (2002) 029, hep-th/0112008;
N.E. Mavromatos, R.J. Szabo, JHEP 0301 (2003) 041, hep-th/0207273.
[8] J.L. Cardy, J. Phys. A 25 (1992) L201, hep-th/9111026;
H. Saleur, Nucl. Phys. B 382 (1992) 486, hep-th/9111007;
L. Rozansky, H. Saleur, Nucl. Phys. B 376 (1992) 461.
[9] M. Flohr, Bits and pieces in logarithmic conformal field theory, hep-th/0111228.
[10] C.R. Nappi, E. Witten, Phys. Rev. Lett. 71 (1993) 3751, hep-th/9310112.
[11] K. Sfetsos, Phys. Lett. B 324 (1994) 335, hep-th/9311010.
[12] D.I. Olive, E. Rabinovici, A. Schwimmer, Phys. Lett. B 321 (1994) 361, hep-th/9311081.
[13] K. Sfetsos, Phys. Rev. D 50 (1994) 2784, hep-th/9402031.
[14] K. Sfetsos, A.A. Tseytlin, Nucl. Phys. B 427 (1994) 245, hep-th/9404063.
[15] N. Mohammedi, Phys. Lett. B 325 (1994) 371, hep-th/9312182.
[16] J.M. Figueroa-O’Farrill, S. Stanciu, Phys. Lett. B 327 (1994) 40, hep-th/9402035.
[17] A. Kehagias, P. Meesen, Phys. Lett. B 331 (1994) 77, hep-th/9403041.
[18] I. Antoniadis, N.A. Obers, Nucl. Phys. B 423 (1994) 639, hep-th/9403191.
[19] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, JHEP 0201 (2002) 047, hep-th/0110242.
[20] M. Blau, J. Figueroa-O’Farrill, G. Papadopoulos, Class. Quantum Grav. 19 (2002) 4753, hep-th/0202111.
[21] R.R. Metsaev, Nucl. Phys. B 625 (2002) 70, hep-th/0112044.
[22] D. Berenstein, J. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
[23] M. Hatsuda, K. Kamimura, M. Sakaguchi, Nucl. Phys. B 632 (2002) 114, hep-th/0202190.
[24] R. Gueven, Phys. Lett. B 482 (2000) 255, hep-th/0005061.
[25] G.T. Horowitz, A. Strominger, Nucl. Phys. B 360 (1991) 197.
[26] K. Sfetsos, JHEP 9901 (1999) 015, hep-th/9811167.
[27] S.K. Yang, Phys. Lett. B 209 (1988) 242.
[28] A. Giveon, E. Kiritsis, Nucl. Phys. B 411 (1994) 487, hep-th/9303016.
[29] A. Giveon, D. Kutasov, O. Pelc, JHEP 9910 (1999) 035, hep-th/9907178.
[30] S. Murthy, Notes on non-critical superstrings in various dimensions, hep-th/0305197.
[31] K. Sfetsos, Fortschr. Phys. 48 (2000) 199, hep-th/9903201.
[32] T. Harmark, N.A. Obers, JHEP 0001 (2000) 008, hep-th/9910036.
[33] R. Güven, Phys. Lett. B 191 (1987) 275.
[34] D. Amati, C. Klimcik, Phys. Lett. B 210 (1988) 92.
[35] G.T. Horowitz, A.R. Steif, Phys. Rev. D 42 (1990) 1950.
[36] H.J. de Vega, N. Sanchez, Nucl. Phys. B 317 (1989) 706;
H.J. de Vega, N. Sanchez, Phys. Rev. Lett. 65 (1990) 1517.
[37] J. Garriga, E. Verdaguer, Phys. Rev. D 43 (1991) 391;
R.E. Rudd, Nucl. Phys. B 352 (1991) 489;
R.E. Rudd, Nucl. Phys. B 427 (1994) 81, hep-th/9402106;
C. Duval, G.W. Gibbons, P.A. Horvathy, Phys. Rev. D 43 (1991) 3907;
C. Duval, Z. Horvathy, P.A. Horvathy, Phys. Lett. B 313 (1993) 10.
[38] A.A. Tseytlin, Nucl. Phys. B 390 (1993) 153, hep-th/9209023;

A.A. Tseytlin, Phys. Rev. D 47 (1993) 3421, hep-th/9211061.
[39] E.A. Bergshoef, R. Kallosh, T. Ortin, Phys. Rev. D 47 (1993) 5444, hep-th/9212030.
[40] D. Marolf, L.A. Pando Zayas, JHEP 0301 (2003) 076, hep-th/0210309;
V.E. Hubeny, M. Rangamani, JHEP 0211 (2002) 021, hep-th/0210234;
V.E. Hubeny, M. Rangamani, JHEP 0212 (2002) 043, hep-th/0211195.
[41] E. Kiritsis, B. Pioline, JHEP 0208 (2002) 048, hep-th/0204004.
[42] A. Gerasimov, A. Marshakov, A. Morozov, Nucl. Phys. B 328 (1989) 664;
D. Nemeschansky, Phys. Lett. B 224 (1989) 121;
J. Distler, Z. Qiu, Nucl. Phys. B 336 (1990) 533;
O. Hernàndez, Phys. Lett. B 233 (1989) 355.
[43] J.D. Lykken, Nucl. Phys. B 313 (1989) 473.
[44] E. Kiritsis, C. Kounnas, Phys. Lett. B 320 (1994) 264;
E. Kiritsis, C. Kounnas, Phys. Lett. B 325 (1994) 536, hep-th/9310202, Addendum.
[45] G. Papadopoulos, J.G. Russo, A.A. Tseytlin, Class. Quantum Grav. 20 (2003) 969, hep-th/0211289.
[46] M. Blau, M. O’Loughlin, G. Papadopoulos, A.A. Tseytlin, Solvable models of strings in homogeneous plane
wave backgrounds, hep-th/0304198.
[47] Y. Hikida, Y. Sugawara, JHEP 0206 (2002) 037, hep-th/0205200;
Y. Hikida, Superstrings on NSNS pp-waves and their CFT duals, hep-th/0303222.
[48] D. Sadri, M.M. Sheikh-Jabbari, String theory on parallelizable pp-waves, hep-th/0304169.
[49] S.P. Kim, C.H. Lee, Phys. Rev. D 62 (2000) 125020, hep-ph/0005224.
[50] E.G. Gimon, L.A. Pando Zayas, J. Sonnenschein, Penrose limits and RG flows, hep-th/0206033.
[51] P. Forgacs, P.A. Horvathy, Z. Horvath, L. Palla, Heavy Ion Phys. 1 (1995) 65, hep-th/9503222.
[52] P. Matlock, R. Parthasarathy, K.S. Viswanathan, Y. Yang, NS5-brane and little string duality in the pp-wave
limit, hep-th/0305028.
[53] E. Witten, Nucl. Phys. B 188 (1981) 513;
F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 267, hep-th/9405029.
[54] A. Brandhuber, K. Sfetsos, JHEP 0212 (2002) 050, hep-th/0212056.
[55] G. D’Appollonio, E. Kiritsis, String interactions in gravitational wave backgrounds, hep-th/0305081.
Electroweak symmetry breaking and fermion masses

from extra dimensions
Claudio A. Scrucca a , Marco Serone b , Luca Silvestrini c
a Theoretical Physics Division, CERN, CH-1211 Geneva 23, Switzerland
b ISAS-SISSA and INFN, Via Beirut 2-4, I-34013 Trieste, Italy
c INFN, Sezione di Roma, Dipartimento di Fisica, Univiversità di Roma “La Sapienza”,
P.le Aldo Moro 2, I-00185 Rome, Italy
Abstract
We study higher-dimensional non-supersymmetric orbifold models where the Higgs field is
identified with some internal component of a gauge field. We address two important and related
issues that constitute severe obstacles towards model building within this type of constructions: the
possibilities of achieving satisfactory Yukawa couplings and Higgs potentials. We consider models
where matter fermions are localized at the orbifold fixed-points and couple to additional heavy
fermions in the bulk. When integrated out, the latter induce tree-level non-local Yukawa interactions
and a quantum contribution to the Higgs potential that we explicitly evaluate and analyse. The
general features of these highly constrained models are illustrated through a minimal but potentially
realistic five-dimensional example. Finally, we discuss possible cures for the persisting difficulties
in achieving acceptable top and Higgs masses. In particular, we consider in some detail the effects
induced in these models by adding localized kinetic terms for gauge fields.
PACS: 11.10.Kk; 11.25.Mj; 12.60.-i
1. Introduction
The Standard Model (SM) of electroweak and strong interactions is extremely

successful in reproducing all the available experimental data up to currently accessible
energies. However, the problem of stabilizing the electroweak scale against quadratically
divergent radiative corrections to the Higgs mass suggests the presence of new physics
E-mail address: serone@sissa.it (M. Serone).
doi:10.1016/j.nuclphysb.2003.07.013
C.A. Scrucca et al. / Nuclear Physics B 669 (2003) 128–158 129
at scales not much larger than the Z mass. While supersymmetric extensions of the SM
technically solve the hierarchy problem and suggest the unification of gauge couplings, the
ever-rising lower bounds on supersymmetric particles have recently stimulated studies of
alternative methods of supersymmetry and electroweak breaking based on the presence of
extra dimensions at the TeV scale [1].
Higher-dimensional models open up the interesting possibility that the Higgs boson
arises from the internal component of a gauge field, with a dynamics protected by higher-
dimensional gauge invariance and controlled in a predictive way by a higher-dimensional
effective theory. Such a possibility, already advocated several years ago [2], has recently
received renewed interest [3–8]. In the case of toroidal internal spaces, the effective
potential turns out to be free of quadratic divergences, and even in phenomenologically
more viable orbifold models [9], power divergences arising through operators localized
at the fixed-points are strongly constrained [10]. Exploiting these facts, one can construct
interesting orbifold models in which electroweak interactions are unified in a larger group,
which is broken through the orbifold projection. Gauge bosons and Higgs fields arise
respectively from the four-dimensional and internal components of the higher-dimensional
gauge bosons. Matter fields can be introduced either as bulk fields in representations of the
unified group G, or as boundary fields localized at the fixed-points where this is broken
to a subgroup H . The construction of realistic models of this type is, however, a difficult
task, because Higgs interactions are severely constrained by the higher-dimensional gauge
invariance. In particular, achieving satisfactory flavor structures and electroweak symmetry
breaking at the same time represents the main problem in this class of theories.
Flavor symmetry breaking can be achieved essentially in two ways, depending on
whether standard matter fields are introduced in the bulk or at the fixed-points of the
orbifold, i.e., the boundaries. In the case of bulk matter fields, standard Yukawa couplings
can originate only from higher-dimensional gauge couplings, which are flavor-symmetric
and completely determined by the group representation. A possible way of obtaining
couplings with a non-trivial flavor structure consists in switching on mass terms with
odd coefficients and introducing mixings between the bulk matter fields and additional
boundary fields in such a way that unwanted light fields eventually decouple [8]. In the
case of boundary matter fields, standard Yukawa couplings cannot be directly introduced,
because they would violate the higher-dimensional symmetry that protects the potential
and therefore induce quadratic divergences, just as in the SM [10]. The only invariant
interactions that can be used are non-local interactions involving Wilson lines [5]; an
interesting possibility to generate this kind of interactions was suggested in [7] and consists
in introducing mixings between the matter fields and additional heavy fermions in the
bulk, which are then integrated out. These two flavor symmetry breaking mechanisms are
very similar from the microscopic point of view, and rely basically on a mixing of bulk
and boundary fields. Interestingly, they also provide a natural explanation for the large
hierarchy of fermion masses, since they are exponentially sensitive on some parameters of
the microscopic theory.
Electroweak symmetry breaking occurs radiatively in this class of models, and is
equivalent to a Wilson line symmetry breaking [11,12] (or Scherk–Schwarz twist [13]),
which reduces the rank of H (as required for electroweak symmetry breaking) if the Wilson
line and the embedding of the orbifold twist do not commute. Since the tree-level Higgs
130 C.A. Scrucca et al. / Nuclear Physics B 669 (2003) 128–158
potential is strongly constrained by gauge invariance, any theory of this type is potentially
very predictive and constrained, even though it is only an effective theory. The generic
predictive power of the theory depends on the parameter Mc /Λ, which controls the effect
of non-renormalizable operators at the compactification scale Mc . Λ is the physical cut-
off scale, whose actual value can be roughly estimated by using naive dimensional analysis
(NDA) [14,15]. There are, however, special quantities, such as the ratio of Higgs and gauge
boson masses, that turn out to be quite insensitive to Λ, because of the non-local nature of
the electroweak symmetry breaking in the internal dimensions.
The construction of phenomenologically viable models requires a detailed and com-
bined analysis of the flavor and electroweak symmetry breaking mechanisms, since the
two are closely connected. Such an analysis has never been performed so far, and the aim
of this work is to fill this gap with an investigation of the possibility of achieving real-
istic effective Yukawa couplings and at the same time a satisfactory Higgs potential. We
focus for simplicity on five-dimensional (5D) models, where the tree-level Higgs poten-
tial is actually vanishing and electroweak symmetry breaking is thus entirely governed, at
sufficiently weak coupling, by the one-loop Higgs effective potential. The contribution to
the latter arising from gauge fields or bulk fermions in fundamental or adjoint representa-
tions is easily computed and well known [12,16,17]. In the presence of a flavor symmetry
breaking mechanism of the types described above, however, the computation of the mat-
ter contribution to the effective potential is more involved, since in both cases there are
simultaneously bulk fermions and boundary fermions mixing among each other. Putting
matter fields in the bulk as suggested in [8], the situation is further worsened by the need
of odd mass terms that distort the wave functions of the bulk fields on their own. For this
reason, we consider here the case suggested in [7], with matter fields located at the fixed-
points.
The basic construction that we study consists of a pair of left-handed and right-
handed matter fermions located at possibly different fixed-points and a heavy bulk fermion
with quantum numbers allowing couplings to both boundary fermions, in such a way
that non-local Wilson line effective interactions between the two can be generated. We
explicitly compute the tree-level effective action that is induced for the matter fermions
by integrating out the massive bulk fermions. The result consists of a mass term, a wave-
function correction, and an infinite series of higher-derivative interactions. All these terms
are directly proportional to the bulk-to-boundary couplings and exponentially sensitive to
the bulk masses. At energies much below the bulk mass, higher-derivative terms can be
neglected and after rescaling the wave-function one finds a physical mass that is bounded
in size. At energies much above the bulk mass, on the contrary, the induced interactions get
exponentially suppressed by their derivative dependence. We also analyse the full one-loop
effective potential for the Higgs field induced by charged bulk fields. Its dependence on the
bulk-to-boundary couplings can be reinterpreted in terms of diagrams involving boundary
fields and the momentum-dependent effective vertex between Higgs and boundary fields
that is induced at tree level by the bulk fermion. The soft behavior of this effective vertex
at high momenta ensures that the loop integral is free of any divergence. The full one-
loop effective potential for the Higgs fields is therefore finite, as expected on symmetry
grounds.1 Moreover, it turns out to have non-trivial symmetry-breaking minima. This

shows that the flavor and electroweak symmetry breaking mechanisms that we consider
are indeed compatible.
The above building blocks can be easily used to construct models that incorporate
all SM fermions and Yukawa couplings and offer therefore a realistic arena for more
quantitative investigations. The only universal quantity to play with is the radius R, which
can be considered as a parameter. In phenomenologically viable models, it must be small
enough to satisfy the current experimental bounds (see, e.g., [18] and references therein),
and the vacuum expectation value (VEV) α of the Wilson line phase must therefore be
small as well in order to match the mass of the W bosons, given by mW = α/R. For
the minimal version of the model, with the simplest possible choice of gauge group and
representations, we find that the induced effective potential V (α) has minima
√ at α ∼ 0.2,
leading to too-low values for 1/R. The Higgs mass mH = (g4 R/2) V (α), which is
further suppressed by a gauge loop factor, tends consequently to be too low as well. Finally,
the top mass mtop cannot be adjusted to a high-enough value. A possible way of improving
the situation is to add to the model new heavy bulk fermions that do not couple to the SM
matter fields but transform in large representations of the gauge group, in such a way as to
modify V (α) and obtain minima for much lower α. For representations of rank r 1, one
can obtain minima at values as low as α ∼ 1/r, with a V (α) at the minimum that rapidly
grows with r. This helps considerably in increasing 1/R and mH , and to some extent mtop .
However, fields with high rank r induce electroweak quantum corrections that are enhanced
by large group-theoretical factors, thereby lowering the cut-off and the predictive power of
the theory.
A simple and natural generalization of our minimal set-up consists in introducing
additional localized kinetic terms for gauge fields [19].2 These have the effect of increasing
the bulk gauge coupling and to distort in a non-trivial way the wave functions and mass
spectrum of the Kaluza–Klein (KK) modes of the gauge bosons. They also modify the
gauge contribution to the effective potential, which we explicitly compute for arbitrary
values of the localized couplings. It turns out that for sufficiently large values of the latter,
where the analysis simplifies, it is possible to get values of α leading to acceptable values of
1/R; at the same time, mH is significantly increased and mtop can be reproduced. We show,
however, that the simultaneous presence of localized gauge kinetic terms and a Wilson-line
symmetry breaking leads in general to an unwanted deformation of the electroweak sector
of the theory. The main point is that the gauge interactions are no longer universal and the
masses of the lowest KK gauge bosons are deformed. This implies, among other things, that
the tree-level value of the ρ parameter departs from 1 and hence imposes generically very
severe bounds on the size of localized gauge kinetic terms. These could strongly constrain
1 This should be contrasted to [7], where a Wilson line Yukawa coupling was introduced as a fundamental
coupling in the tree-level Lagrangian and found to induce divergences in the one-loop Higgs potential. The
distinction between our situation and that of [7] lies in the locality of the fundamental Lagrangian. If the latter is
non-local, then quantum divergences are allowed to appear even in non-local operators (such as the Higgs mass
in these models), since all of them must now be introduced in the theory from the beginning as independent
couplings and can be used as counterterms.
2 We thank R. Rattazzi for suggesting this possibility to us.
model building even if they are only radiatively generated, although it is not excluded that
large values can be tolerated in special models with new symmetries. Finally, the scale at
which electroweak interactions become strong is again lowered, so that the theory tends to
become less predictive.
The paper is organized as follows. In Section 2 we present a basic model that captures
all the essential features we want to exploit. In Section 3, the non-local Yukawa couplings
between boundary fermions and the Higgs field are computed. In Section 4 we compute the
one-loop effective potential for the Higgs field. In Section 5 we construct a prototype 5D
model and discuss its main properties. In Section 6, the effects of localized gauge kinetic
terms are studied. Finally, Section 7 is devoted to some general conclusions, and some
technical details are collected in Appendix A.
2. The basic construction
The simplest framework allowing an implementation of the Higgs field as the internal
component of a gauge field is a 5D gauge theory with gauge group G = SU(3) on an
S 1 /Z2 orbifold [20]. The Z2 orbifold projection is embedded in the gauge group through
the matrix

−1 0 0
P = eiπλ3 = 0 −1 0 , (1)
0 0 1
where λa are the standard SU(3) Gell-Mann matrices, normalized as Tr λa λb = 2δab , so
that AM = AaM λa /2. The group G is broken in 4D to the commutant H = SU(2) × U (1) of
the projection P . The massless 4D fields are the gauge bosons Aaµ in the adjoint of H and a
charged scalar doublet, arising from the internal components Aa5 of the gauge field. A VEV

√ an 3additional spontaneous symmetry breaking to E = U (1), generated by
for A5 induces
Âµ = (Aµ + 3 Aµ )/2. Using the residual H symmetry, it is always possible to align
A5
8
along the λ7 component, corresponding to the down component of the doublet, and take
2α a7
Aa5 = δ . (2)
g5 R
Identifying H with the electroweak gauge group, E with the electromagnetic group, and
the zero-mode of A5 with the Higgs field H , the above construction gives a description
of electroweak symmetry breaking. In this minimal version, the weak mixing angle θW
turns out to be too large, θW = π/3, but this problem can be solved by starting with a
different unified group G, as we will see. Taking into account the normalization of the
zero-mode (see Appendix A), (2) corresponds to a VEV for the neutral component of the
Higgs doublet H equal to 2α/(g4 R), with
g5
g4 = √ . (3)
2πR
It is well known from [12,17] that a VEV for A5 induces a Wilson line which is
equivalent to a Scherk–Schwarz twist, the two situations being related through a non-
periodic gauge transformation. The twist matrix T (α) satisfies the consistency condition
T P T = P [21]; for the choice (2), it is given by

1 0 0
T (α) = e 2iπαλ7
= 0 cos 2πα sin 2πα . (4)
0 − sin 2πα cos 2πα
The orbifold projection represents an explicit symmetry breaking of G to H , the masses of
the fields in G/H being of order 1/R, whereas the Scherk–Schwarz twist corresponds to a
spontaneous symmetry breaking of H to E, the masses of the fields in H /E being of order
α/R. This situation is equivalent to an S 1 /(Z2 × Z2 ) orbifold with two non-commuting
projections P and P (α) = T (α)P = T (α/2)P T (−α/2), and radius 2R. From this point
of view, there are two perfectly symmetric projections, the first breaking G to H with fixed-
points y = 0, 2πR, and the second G to H with fixed-points y = πR, 3πR. Together, they
break G to E = H ∩ H in a non-local way: the two subgroups H and H are isomorphic,
but their embeddings in G form an angle α. The most general field content consists of bulk
fields in representations of G, and boundary fields in representations of the subgroup H or
H surviving at the fixed-point where they are localized (see, e.g., [22]).
In the following, we will adopt the perspective of the S 1 /Z2 orbifold with a Wilson
line breaking, and work on the fundamental interval y ∈ [0, πR]. The brane fields of
our basic construction consist of a left-handed fermion doublet QL = (uL , dL )T and two
right-handed fermion singlets uR and dR of H = SU(2) × U (1); these fields can also be
described by their charge conjugates QcR = (dRc , −ucR )T , dLc and −ucL . We will assume
that the doublet and the two singlet fields are located respectively at positions y1 and y2 ,
equal to either 0 or πR. The bulk fields are, in addition to the gauge fields AM , one pair
of fermions Ψa and Ψ a with opposite Z2 parities for each type of matter field a = u, d; we
take these in the symmetric and the fundamental representations of G = SU(3) for a = u
and a = d, respectively.
The parity assignments for the bulk fermions allow for bulk mass terms Ma mixing Ψa
and Ψ a , as well as boundary couplings ea with mass dimension 1/2 mixing the bulk
1,2
fermion Ψa to the boundary fermion a. Denoting the doublet and singlet components
arising from the decomposition of the bulk fermion Ψa under G → H (see Appendix A)
by ψa and χa respectively, the complete Lagrangian for matter fields is then given by the
following expression:3

Lmat = a iD iD
/ 5 Ψa + Ψ a + Ψ a + h.c.
a Ma Ψ
Ψ a / 5Ψ
a

+ δ(y − y1 ) QL iD/ 4 QL + e1d QL ψd + eu Q c
1 R ψu + h.c.

/ 4 dR + ūR iD
+ δ(y − y2 ) d̄R iD / 4 uR + e2d d̄R χd + e2u ūcL χu + h.c. .
All bulk fermion modes are massive
and, neglecting the bulk-to-boundary couplings, their
mass spectrum is given by Ma,n = ± m2n + Ma2 , where mn = n/R. After the spontaneous
symmetry breaking induced by (2), a new basis has to be defined for the bulk fermion
modes in which they have diagonal mass terms, with a shift in the KK masses mn →
/
4 5 /
3 D and D denote the 4D and 5D covariant derivatives. Defining the Hermitian matrix γ = iγ , they are
5 4
/ 5 = D/ 4 + iγ5 D5 .
related by D
mn (α). The procedure is outlined in Appendix A. The new fields Ψ (i) and Ψ (i) defining this
(i)
basis are given by Eqs. (A.5) and (A.11), and have KK masses mn = (n + q (i) α)/R, with
q (i) being an integer charge, defined by Eqs. (A.8) and (A.14); i = 1, 2 for the fundamental
representation and i = 1, 2, 3, 4 for the symmetric one. Similarly, a new basis has to be
defined for the gauge-field modes to diagonalize their mass terms. The new fields A(i) M
and their KK masses are defined as in Eqs. (A.16) and (A.19), where the field Ψ ± are
respectively identified with the gauge field components Aµ and A5 . More precisely, the
complex gauge field A(1) (1) = 1 (Ψ +(1) ) is identified with the W boson, the
µ with charge q 0
(2) +(2) (0) +(3)
real field Aµ with q (2) = 2 (Ψ0 ) with the Z boson and the neutral field Aµ (Ψ0 )
−(4)
with the photon. Similarly, the scalar field A(0)
5 (Ψ0 ) is identified with the component
of the Higgs field H that is left over after the spontaneous symmetry breaking. Using the
fact that sec θW = 2, the masses of the W and Z fields can be written as
α α
mW = , mZ = sec θW . (5)
R R
The Higgs mass is radiatively induced after the electroweak symmetry breaking and
depends on the second derivative of the potential evaluated at the minimum:
g4 R
mH = V (α). (6)
2
In the following, it will be convenient to take the size πR of the orbifold as reference
length scale and use it to define dimensionless quantities. √ In particular, it will be useful
to introduce the parameters λa = πRMa and /ia = πR/2 eia , and the integer δ =
(πR)−1 |y1 − y2 | parametrizing the distance between the location of left- and right-handed
fields.
3. Induced couplings for the boundary fermions
The heavy bulk fields couple to boundary fermions, and can therefore induce gauge-
invariant non-local couplings mixing matter fermions localized at the fixed-points and the
field A5 through Wilson lines [7]. Since the VEV (2) for A5 mixes the modes of different
(i) defined
components of the bulk fermions, it is convenient to use the fields Ψ (i) and Ψ
(i)
before and to group them into a two-vector (Ψ , Ψ ) . In this notation, the kinetic term
(i) T
of the nth KK mode of each component of the bulk fermions is encoded in the following
two-by-two matrix in momentum space:

/ − mn
p M
KΨ = . (7)
M p/ + mn
The tree-level propagator is obtained by inverting this matrix, and is given by

i / + mn
p −M
∆Ψ = 2 . (8)
p − m2n − M 2 −M / − mn
p
The couplings between bulk and boundary fields can be easily rewritten in terms of
the new fields as well; the new doublet and singlet components under the decomposition
G → H can be easily read off from Eqs. (A.6) and (A.12). The only bulk field that
couples both to the left-handed and right-handed boundary fields, and can therefore induce
mass terms, is always Ψ (2) . The other components of the bulk fermions couple only
separately to left-handed and right-handed fields; they can therefore induce only wave-
function corrections. For each bulk-to-boundary coupling, there is a wave-function factor
ξi,n for the nth mode, which is equal to 1 if yi = 0 and (−1)n if yi = πR. The interaction
Lagrangian reads
∞
1
/2u (2)
Lint = /1u ξ1,n d̄Rc Ψu,n

(1)
− ūcR Ψu,n
(2)
− √ ξ2,n ūcL Ψu,n + ηn Ψu,n
(4)
πR n=−∞ 2

(1) (2)
(2)
+ /1 ξ1,n ηn ūL Ψd,n + d̄L Ψd,n + /2 ξ2,n d̄R Ψd,n + h.c. ,
d d
(9)
√
where the factor ηn is defined to be 1 for n = 0 and 1/ 2 for n = 0. The bulk
fermions can be disentangled from the boundary fermions by completing the squares
in the kinetic operator and performing an appropriate redefinition of the bulk fields.
This generates corrections to the kinetic operators K u and K d for the boundary
fields u = uL + uR and d = dL + dR , originally given just by p / , which correspond
diagrammatically
to the exchange of bulk fermions. These contributions have the structure
Ψ Ψ
n (eξn PL,R )∆n (eξn PR,L ), where PL,R are chiral projectors and ∆n is given by the
first entry of (8) with mn as in Eqs. (A.8) and (A.14). To compute the infinite sums, it
is convenient to go to Euclidean
space, and define the dimensionless momentum variables
x = πRp and x a = πR p2 + Ma2 , as well as the basic functions fδ given by
∞
1
f0 (x, α) = = coth(x + iπα), (10)
n=−∞
x + iπ(n + α)
∞
(−1)n
f1 (x, α) = = sinh−1 (x + iπα). (11)
n=−∞
x + iπ(n + α)
The functions fδ (x, α) are related to the propagation of bulk fields between two fixed-
points separated by a distance δπR. This is particularly clear from their Taylor expansion,
which takes the simple form
∞

fδ (x, α) = e−|2k+δ|(x+iπα). (12)
k=−∞
After a straightforward computation, the total effective actions for the boundary fields
are found to be given by the following expressions:

/ d2
/ u2
K =p
u
/ Re 1 + 1d PL f0 x d , 0 + 2 u PR f0 x u , 0
x 2x
u2 u u
/ / u2 i /1 /2
+ 1u PL + 2 u PR f0 (x u , 2α) + Im √ fδ (x u , 2α) , (13)
x 2x πR 2
d2
/1 /2d2 d
/1u2 u
K =p
d
/ Re 1 + PL + d PR f0 x , α + u PL f0 x , α
x2 x x
i

+ Im /1d /2d fδ x d , α . (14)
πR
From the above expressions, we see that an infinite set of non-renormalizable interactions
involving the Higgs field are generated together with the standard Yukawas. The new
interactions have small effects in physical processes as long as the radius R is small
enough, but they all contribute comparably to spontaneous symmetry breaking, whose
effect is not only to induce masses mu and md mixing dL with dR and uL with uR , but also
to generate different wave-function corrections Z1,2 u and Z d for u
1,2 L,R and dL,R . Moreover,
all these quantities are momentum-dependent.
To be precise, the boundary fields have an effect on the spectrum of the bulk fields as
well, which in general cannot be neglected, and the mass eigenstates are mixtures of bulk
and boundary states. To find the exact spectrum of fermions, one would have to diagonalize
the full kinetic operator for the entangled bulk and boundary fermions. Assuming however
that the physical mass induced for the boundary fields is much smaller than the masses
of the bulk fields, one can neglect the distortion on the spectrum of the latter and use the
free kinetic operator (7) for bulk fermions and the results (13) and (14) for the boundary
fields. In this approximation, the spectrum of bulk fields is unchanged and the mass of the
boundary fields is obtained by looking at the zeros (13) and (14). Notice that the momentum
dependence in the latter can be safely neglected within the adopted approximation. The
a
masses ma and the wave-functions Z1,2 for left and right components then reduce to α-
dependent parameters given by
/u/u
mu = √1 2 Im fδ λu , 2α , (15)
2πR
/d /d
md = 1 2 Im fδ λd , α , (16)
πR
/ d2
/ u2
/ u2
Ziu = 1 + δi1 1d Re f0 λd , 0 + δi2 2 u Re f0 λu , 0 + δ i u Re f0 λu , 2α , (17)

λ 2λ 2 λ
i2
/ d2
/ u2
Zid = 1 + i d Re f0 λd , α + δi1 1u Re f0 λu , α . (18)

λ λ
Finally, the physical masses after symmetry breaking are obtained by rescaling the fields
to canonically normalize them; they are given by
ma
maphys = . (19)
Z1a Z2a
The arguments of [23] suggest that the effective actions induced for the boundary fields
should be non-local from the 5D point of view and generated when
A5 acquires the VEV
(2) from operators involving Wilson lines

nπR
Wn (A5 ) = P exp ig5 dy A5 . (20)
0
Indeed, the kth term in the series expansion (12) corresponds to the components of
W2|n| (
A5 ) = T (|n|α) with a winding number n = 2k + δ; this connects the two
boundaries through a path that winds k times around the internal circle, with total length
|n|πR. Since the boundary fields break explicitly G to H , the effective action will be only
H -invariant and involve the various components of the decomposition of (20). However,
it is nevertheless convenient to embed the boundary fields Q and q into G representations
of the same type as those of the bulk fields, completing them with vanishing components.
This description of the boundary fields is similar to the one that would emerge for the
boundary values of bulk fields with ‘+’ and ‘−’ parities. More precisely, the boundary
fields Q and q can be embedded into fundamental representations QF and qF or symmetric
representations QS and qS of SU(3) as follows:
 
uL 0 0 0 dRc
1
QF = dL , dF = 0 , QcS = √  0 0 −ucR  ,
0 dR 2
dRc −ucR 0

0 0 0
ucS = 0 0 0 . (21)
0 0 −ucL
In this notation, the couplings between boundary and bulk fermions can be obtained simply
by taking traces of products of Eqs. (21) with Eqs. (A.6) and (A.12).
Using the embeddings (21), one can easily verify that the leading part of the effective
∂Q +
action in a derivative expansion is given by the sum of the original action L0 = Qi/
∂ u + d̄i/
ūi/ ∂ d and the following non-local interactions LF and LS induced by the bulk
fermions in the fundamental and symmetric representations respectively (going back to
Minkowski space):
d2
d /1 /2d2
LF = e−|2k|λ F W|2k| i/
Q ∂ QF + d̄ W
F |2k| i/
∂ d F
λd λd
k
/1d /2d
+ e−|2k+δ|λ
d
F W|2k+δ| dF + h.c.],
[Q (22)
πR
k

/1u2 T c /2u2
e−|2k|λ
u
LS = Tr W Q
|2k| S W|2k| i/
∂ Qc
S + T
Tr W|2k| ūcS W|2k| i/
∂ ucS
λu λu
k
/1u /2u T
+ e−|2k+δ|λ
u
c W|2k+δ| uc + h.c. .
Tr W|2k+δ| Q (23)
S S
πR
k
We conclude that it is possible to induce generic Yukawa couplings for boundary fields
through non-local operators involving Wilson lines that connect the two boundaries and
wind around the orbifold an arbitrary number of times. The resulting physical mass maphys
of the boundary fields is exponentially sensitive to the parameter λa governing the bulk
masses. Notice also that, because of the wave-function rescaling, the value of maphys given
by (19) cannot be made arbitrarily large by increasing the values of the boundary couplings
/ia . Indeed, for /ia 1, maphys quickly saturates to a value depending only on ratios of these
parameters but not on their overall size.
4. One-loop effective potential for the Higgs
The field A5 couples only to the gauge fields and to the two bulk fermions. Its radiatively
induced potential thus depends only indirectly on the boundary couplings through diagrams
in which the virtual bulk fermions temporarily switch to a virtual boundary fermion. The
total potential is therefore the sum of a universal gauge contribution and a parameter-
dependent contribution coming from the fermions.
The contribution of the fermions is obtained by summing up all possible one-loop
diagrams of bulk fermions dressed by an arbitrary number of external A5 lines and
insertions of boundary couplings. Since bulk interactions conserve the KK momentum,
whereas boundary interactions do not, it is convenient to separately resum diagrams with
no insertion of boundary interactions and diagrams with an arbitrary but non-zero number
of these. The first piece corresponds to the contribution of bulk fields in the absence of
boundary couplings, with kinetic operator (7), whereas the second can be reinterpreted
as the contribution of boundary fermions propagating with the effective kinetic operators
(13) and (14) induced by the insertions of mixings with the bulk field. This decomposition
corresponds precisely to the shift performed in the last section to disentangle bulk and
boundary fermions. In this case, however, this leads to an exact result, because both bulk
and boundary fields are integrated out.
The bulk fields Ψa(i) couple to the gauge field A5 in a diagonal way, through the
shift induced in their KK mass by the minimal coupling. For each mode one has m(i) =
(n + q (i) α)/R, where q (i) is the charge of the mode as specified by Eqs. (A.8) and (A.14).
In total, there are only three kinds of modes with non-vanishing charges, two with q = 1
and one with q = 2. The contribution to the effective action from a pair of such modes for
the Ψ (i) and Ψ (i) fields with given charge q reads Γ Ψ (qα) = − ln det[K Ψ (qα)], where
K is the Euclidean continuation of (7). The determinant of K Ψ as a two-by-two matrix
Ψ
yields p2 + m2n + M 2 . The determinant in the space of KK modes then reduces to an

infinite product over these factors, which yields an irrelevant α-independent divergence
plus a finite α-dependent function. Finally, the determinant over spinor indices yields a
trivial factor of 4. For a pair of modes with charge q, one then finds
∞
−1
−2
VΨa (qα) = dx x 3 lnf1 x a , qα
2π 6 R 4
0
∞
3 1 + 2kλa + 4/3k 2λa2 −2kλa
= e cos(2qkπα). (24)
8π 6 R 4 k5
k=1
The boundary fields a = u, d couple to A5 only through the non-local Wilson line
effective interactions induced by the bulk fermions, and their kinetic operators K a (α) in
Euclidean space are given by (13) and (14). Their contributions to the effective action read
Γ a (α) = − ln det[K a (α)]. To evaluate this expression, we start by using the standard trick
of rewriting it in terms of the scalar quantity K a K aT . The determinant over spinor indices
then yields just a factor of 4, and dropping irrelevant constant terms one finds:
2
−1
∞ /1d2 d
/2u2 u
Vu (α) = dx x 3
ln Re 1 + δ i1 f0 x , 0 + δ i2 f0 x , 0
4π 6 R 4 xd 2x u
0 i=1
2 u2
/iu2 u
/i u
+ δ u f0 x , 2α + Im δ fδ x , α ,
2 i2 x 2 i2 x
i=1
(25)
∞ 2
−1 / d2
/ u2
Vd (α) = 6 4
dx x 3 ln Re 1 + i d f0 x d , α + δi1 1u f0 x u , α
4π R x x
0 i=1

2
/ d2
+ Im i fδ x d , α . (26)
x
i=1
The full contribution of bulk and boundary fermions to the one-loop effective potential
is finally given by
Vf (α) = VΨu (α) + VΨu (2α) + VΨd (α) + Vu (α) + Vd (α). (27)
As expected, the result is finite, thanks to the exponentially soft UV behavior of the
functions fδ . Notice also that the argument of the logarithm in the boundary contributions
can be rewritten as the determinant of a two-by-two matrix given by the identity δij ,
plus a matrix ∆ij encoding interactions between the fixed-points located at yi and yj and
involving the function fδij with δij = (πR)−1 |yi − yj |.
It is worth noting that the result (27) contains indirect information on the exact spectrum
of the bulk and boundary fermions, which allows in fact to probe the accuracy of (19).
The information about the spectrum of eigenvalues ξn = πRmn can be extracted by
comparing our result (27) with the definition of the effective action as a trace over all
the mass eigenstates of the logarithm of their free kinetic operators. Turning the sum
into a product inside the logarithm,
one would then obtain in this approach a logarithm
argument proportional to n (x 2 + ξn2 ), which has zeros at x = iξn . This means that the
exact spectrum of eigenvalues can be obtained by setting the total logarithm argument
in (27) to zero. One can verify that the boundary contribution has poles exactly where
the bulk contribution has double zeros corresponding to the original tower of degenerate
bulk modes. Half of the original zeros remain and correspond to the combination of bulk
fields whose modes are not perturbed. The other orthogonal combination has a deformed
mass spectrum, which is determined, together with the masses of the boundary fields, by
solving the transcendental equation arising from the argument of the boundary contribution
alone. The latter can be solved numerically, or analytically for the lightest modes ξ0a , under
the assumption that ξ0a 1. In that limit, the momentum dependence can be completely
neglected and the logarithm argument reduces to Z1 (α)Z2 (α)x 2 + (πRma )2 , which leads
to (19). This expression is thus valid as long as ξ0a 1. If the latter condition is not
satisfied, the mixing between bulk and boundary modes can have a significant effect on
the mass of the lowest-lying state, which has to be computed by numerically solving the
exact transcendental equation defining the spectrum.
The contributions to the effective action from gauge and ghost fields are easily computed
[12,16]. Going again to a diagonal basis, two modes with q = 1 and one with q = 2 are
found. Each contributes
∞ ∞

3 −2 9 1
VgA (qα) = dx x 3 lnf1 (x, qα) = − cos(2qkπα),
16π 6 R 4 6
64π R 4 k5
0 k=1
(28)
which gives, in total:
Vg (α) = 2VgA (α) + VgA (2α). (29)

The total one-loop effective potential is given by V = Vf + Vg . It satisfies the symmetry
property V (1 ± α) = V (α). As expected, the bulk-to-boundary couplings /ia deform the
potential in a non-trivial way; it is thus interesting to analyse the possible minima one can
get in this case. Notice first that standard 5D bulk fermions in fundamental or symmetric
representations, combined with gauge bosons, can lead to non-trivial minima for the
values α = 0.5 and α ∼ 0.3, respectively. In our situation, however, lower values can be
obtained, thanks to the effect of the boundary interactions. This is easily understood by
noticing that for α 1 both Va and Vg increase with α, whereas VΨa decreases as α
increases; the boundary contribution therefore tends to shift the minimum of V to lower
values of α. Furthermore, both fermion contributions are very sensitive to λa and decrease
exponentially with λa , whereas the dependence on the brane-to-bulk couplings /ia of the
boundary fermion potential is mild. At fixed /ia , the dependence of V on the λa ’s, which
we assume to be equal to some common value λ for simplicity, is as follows. For λ = 0,
VΨa dominates and we get α ∼ 0.3, roughly the same value as in the case of decoupled
5D massless fermions. As λ is increased, VΨa and Va decrease and the minimum moves to
lower values of α, down to α ∼ 0.2, the precise value depending on the /ia couplings. When
λ further increases, Vg eventually dominates and the only minimum that is left is the trivial
one at α = 0. We have performed a numerical study of V to determine the lowest values of
α that can be achieved in this setting. We were able to find minima for α ∼ 0.1–0.2 for a
wide range of the parameters λa and /ia (see Fig. 1).
Let us conclude with a comment on divergences. The above computation proves that
no divergences are induced for the Higgs mass at the one-loop level, neither on the
bulk nor on the boundary. There is actually a residual gauge invariance at the fixed-
points, Aa5 → Aa5 + ∂5 ξ a , with a ∈ G/H and ξ a the corresponding gauge parameters,
which forbids any local boundary mass term for A75,0 [10], and this shift symmetry is
not broken by the localized couplings in (5).This implies that no direct divergence can
appear in the potential at any order in perturbation theory. There will however be linearly
divergent wave-function corrections for the gauge and Higgs fields, which can substantially
influence the physical potential. It should be noticed, however, that these effects are G-
symmetric, and G-violating quantities that are only H -symmetric are therefore completely
insensitive to them and finite. Particularly important examples of such quantities are
the ratios mW /mH or mZ /mH , since both the gauge fields W, Z and the Higgs field
H are gauge fields from the higher-dimensional point of view, and hence receive a
common wave-function correction. At leading order, these ratios depend only on α and
Fig. 1. Different contributions to the effective potential (in arbitrary units): the bulk and boundary fermion
contributions (upper left) and the full potential (upper right) for λ = 1.57, /1 = 3.1, /2 = 0.7 and δ = 0; the
bulk and boundary fermion contributions (lower left) and the full potential (lower right) for λ = 1.83, /1 = 6.4,
/2 = 6.1 and δ = 1.
therefore represent predictable quantities for the model. They can in principle be influenced
by non-renormalizable operators, so that the accuracy of their leading-order values are
controlled by the effective theory expansion parameter (ΛR)−1 . However, the very peculiar
symmetries constraining A5 do not allow for any higher-dimensional operator that could
have a relevant effect at tree- or one loop-level.
5. A prototype 5D model
Having shown how a satisfactory mechanism for electroweak and flavor symmetry
breaking can be achieved, we now turn to the construction of a simple prototype model
in five dimensions. The first concern in model building is to introduce heavy bulk fermions
for each pair of left- and right-handed SM fermions, so as to obtain all the required Yukawa
couplings for the SM matter (including neutrinos) through the mechanism explained in
Section 3. The strong-interaction sector is completely factorized and SU(3)c therefore does
not matter. The true constraint comes from the electroweak-interaction sector and is fixed
by the SU(2)L × U (1)Y quantum numbers of the SM fields; the Higgs scalar H is a 21/2 ,
the quarks QL , uR , dR are in the 21/6 , 12/3 , 1−1/3 , and the leptons LL , lR , νR in the 2−1/2 ,
1−1 ,10 .
There are many possibilities for gauge groups unifying electroweak and Higgs
interactions, but we will stick to the basic structure in which a SU(3) group is broken to
SU(2) × U (1) through a Z2 orbifold projection. The decomposition of the simplest SU(3)
representations under this projection is as follows: the adjoint representation decomposes
as 8 = 30 ⊕ 21/2 ⊕ 2̄−1/2 ⊕ 10 , the fundamental as 3 = 21/6 ⊕ 1−1/3 , the symmetric as 6 =
31/3 ⊕2−1/6 ⊕1−2/3 , and finally the rank-three symmetric as 10 = 41/2 ⊕30 ⊕2−1/2 ⊕1−1 .
The simplest possibility is to take G = SU(3)w and H = SU(2)L × U (1)Y . In this
case, bulk fermions in the fundamental, symmetric, rank-three symmetric and adjoint
representations would have the right charges to couple respectively to the down quarks,
conjugate up quarks, charged leptons and conjugate neutrinos. However, this minimal
choice of the gauge group would lead to too high a weak mixing angle: sin2 θW = 3/4.
A possible cure to this problem consists in adjusting θW by introducing different localized
kinetic terms for SU(2)L and U (1)Y gauge bosons at the fixed-points of the orbifold. The
distortion they cause introduces however other problems, as we will discuss in the next
section. In addition, the computation of the Higgs potential is complicated by the presence
of non-universal localized gauge couplings.
An alternative way of tuning θW to a reasonable value is to add an extra overall U (1)
factor, with coupling constant g , which remains unaffected by the orbifold projection,
and identify the U (1)Y hypercharge as the sum of the U (1) and U (1) charges after the
orbifold projection SU(3)w × U (1) → SU(2)L × U (1) × U (1) . In this way the gauge
field AY associated to the hypercharge and its orthonormal combination AX are
√ √
g A8 + 3 gA 3 gA8 − g A
AY = , AX = , (30)
3g 2 + g 2 3g 2 + g 2
√
implying that gY = 3 gg / 3g 2 + g 2 and thus
gY2 3
sin2 θW = = . (31)
g2 + gY2 4 + 3g 2 /g 2
By appropriately choosing the ratio g/g we can then restore the correct value of sin2 θW .
After electroweak symmetry breaking, the following mass term for the gauge fields is
induced from their gauge kinetic term:
2
α 1 2
Lm = |W |2 + Z − 3 − 4 sin2
θ A
W X . (32)
R 2 cos2 θW
In order for the model to be realistic, the gauge boson AX must of course acquire a large
mass by some mechanism. As we shall discuss below, its associated U (1)X symmetry is
actually anomalous, and is therefore naturally expected to be spontaneously broken at the
cut-off scale, with a mass term
1
Lm Λ2 (AX )2 . (33)
2
An important consequence of (33) is that the mixing mass term arising in (32) between
Z and AX has a negligible effect. The corresponding distortion of the ρ parameter can be
quantified by integrating out the heavy AX gauge boson: this leaves a correction of order
(mZ /Λ)2 , which is safely small since Λ is experimentally constrained to be above a few
TeV.
Bulk fermions in fundamental and symmetric representations allow couplings to all the
matter fermions, since their U (1) charge can be tuned to achieve the required hypercharge;
the way in which the hypercharge of the SM field is distributed as sum of U (1) and U (1)
charges is then completely fixed. In this model, one can thus implement the construction
described in Section 3, without any additional complication. It turns out that four bulk
fermions Ψa with a = d, u, l, ν (plus their mirrors Ψ a with opposite parities) do the job,
the quantum numbers of bulk and brane fields with respect to SU(3)c × SU(3)w × U (1)
and SU(3)c × SU(2)L × U (1) × U (1) being as follows:
Ψd : (3, 3)+
0, couples to QL : (3, 2)1/6,0 and dR : (3, 1)−1/3,0,
Ψu : (3̄, 6)−
0, couples to QcR : (3̄, 2)−1/6,0 and ucL : (3̄, 1)−2/3,0,
+
Ψl : (1, 3)−2/3, couples to LL : (1, 2)1/6,−2/3 and lR : (1, 1)−1/3,−2/3,
Ψν : (1, 6)−2/3 , couples to LcR : (1, 2)−1/6,2/3 and νLc : (1, 1)−2/3,2/3. (34)
To achieve the most general flavor structure, we introduce three generations I = 1, 2, 3 of
the above bulk fields. These can have arbitrary kinetic matrices (Ma )I J in flavor space,
possibly different for a = d, u, l, ν. The couplings of the bulk fermions to the three
generations of SM boundary fields can involve generic matrices (/ia )I J in flavor space.
However, these can be made proportional to the identity through a rotation of the bulk
fields, whose only additional consequence will be to change the kinetic matrices (Ma )I J .
Without loss of generality, we can therefore set the couplings to flavor-blind constants,
parametrized by dimensionless coefficients /1a and /2a for left-handed and right-handed
fields.
5.1. Mass matrices
The computation of the induced masses for the SM matter fields proceeds exactly as in
Section 3, the only novelty being the non-trivial flavor structure of the bulk masses. The
latter can be written as Ma = Ea† MaD Fa , where Ea , Fa are unitary matrices and MaD is
diagonal, and the kinetic term of the bulk fields can thus be diagonalized in flavor space
by redefining Ψa = Ea Ψa and Ψ a = Fa Ψa . By doing so, one gets a diagonal parameter
λ = πRMa for bulk fields, but the couplings between boundary fields and the bulk
a D
fields Ψa become non-diagonal and involve the matrices Ea . The right-handed part of
the boundary fields a, which couple only to the corresponding bulk fermion Ψa , can be
diagonalized by redefining aR = Ua aR with Ud,l = Ed,l and Uu,ν = Eu,ν ∗ . For the left-
handed fields, instead, this cannot be done, since each of them couples to two different
bulk fields; this will be a first source of non-trivial mixing in the mass matrices. At this
point, all the couplings are diagonal, except those mixing the boundary fields QL , LL , QcR
and LcR to the bulk fields Ψd , Ψl , Ψu and Ψν , which involve the matrices Ud , Ul , Uu and
Uν .
The presence of the matrices Ua affects the results of Section 3 in the following way.
The contribution to the wave function Z1a of the left-handed field aL from the bulk field
a (Ψ ) = U † Z a (Ψ )Ub , and the new total wave function

Ψb , call it Z1a (Ψb ), is changed to Z
1 b b 1 b
a = b Z
Z a (Ψ ) is not diagonal. The wave-function corrections Z a for the right-handed
1 1 b 2
fields aR are instead unchanged and diagonal: Z a = Z a . Finally, the mass ma induced for
2 2
the boundary fields is changed to m̃a = ma Ua . In order to determine the new physical mass,
one has to diagonalize the kinetic term of the left-handed fields. This is achieved through
a unitary transformation; writing Z a = Va Z Da Va† , with Z Da diagonal, one can redefine
1 1 1
the left-handed fields as aL = Va† aL , to obtain a kinetic term for boundary fields which
is diagonal in flavor space. The wave functions are now all diagonal and given by Z1Da
and Z2a , whereas the mass matrices are all non-diagonal and given by ma Ua Va . Rescaling
finally the wave-function factors, one finds the following physical mass matrices:

maII (Va Ua )I J
maphys IJ
= (no sum on I, J ). (35)
(Z2a )I I (Z1Da )J J
5.2. Anomalies
We now briefly comment on the issue of anomalies, paying attention to their distribution
over the internal space. Since the bulk fermions are strictly vector-like, the only anomalies
that can arise come from the SM fermions living at the fixed-points, and depend on how
these are distributed among the two different fixed-points.
In the case where all SM fermions are located at the same fixed-point, all anomalies
that do not involve the extra U (1)X gauge field vanish, thanks to the usual cancellations
arising for the SM spectrum of fermions. We are then left with localized mixed anomalies
involving the U (1)X gauge field, which can be cured by means of a 4D version of
the Green–Schwarz mechanism (GS) [24]. One introduces a neutral 4D axion at y = 0,
transforming non-homogeneously under the U (1)X symmetry, with non-invariant 4D
Wess–Zumino couplings compensating for the one-loop anomaly. In this way all mixed
SU(3)c × SU(2)L × U (1)Y × U (1)X gauge and gravitational anomalies are canceled and
the axion is eaten by the U (1)X gauge boson, which becomes massive and decouples.
For other distributions of the SM fermions such that all the SM anomalies are still
canceled locally, a GS mechanism is again sufficient to cancel all remaining anomalies
involving the U (1)X gauge field. However, two neutral axions are now needed, one at y = 0
and one at y = πR, with non-invariant 4D Wess–Zumino couplings. One combination of
axions is again eaten by the U (1)X gauge boson, but the other combination remains as a
physical massless axion in the low-energy spectrum.
For a completely generic distribution of matter, for which not even the SM anomalies
are locally canceled, the situation is more complicated. In order to locally cancel the
SM anomalies, one has to introduce a bulk Chern–Simons term with jumping coefficient
[25–27], which can be naturally generated by integrating out certain massive states [28].
Since the hypercharge is embedded into a non-Abelian group in the bulk, this is however
not sufficient to let all of the anomalies flow on a single fixed-point. One then has
to introduce also two neutral axions, one at each fixed-point, to locally cancel all the
remaining anomalies involving U (1)X . As before, the U (1)X gauge boson gets a mass,
but a combination of axions remains massless.
5.3. Quantitative analysis
This simple 5D orbifold model we constructed has all the qualitative features to
represent a possible interesting extension of the SM, where the electroweak scale is
stabilized without supersymmetry and the hierarchy of fermion masses is explained by
the non-local origin of the Yukawa couplings. As mentioned at the end of Section 4,
the quantities mH , mW and mZ , and especially their ratios, can be reliably computed,
having a mild dependence on the cut-off Λ. In order to get a better understanding of
the range of validity of our model as an effective field theory, it is however necessary
to know the magnitude of Λ. A reasonable estimate is obtained by defining Λ as the
energy scale where the basic higher-dimensional gauge interactions become strong. NDA
then yields Λ ∼ (lD /gD 2 )1/(D−4) , where g is the higher-dimensional gauge coupling
D
and lD = (4π) Γ (D/2) in D space–time dimensions. The predictive power of the
D/2
effective theory at the compactification scale Mc is therefore governed by the 4D effective

(D−4)/2
coupling g4 = gD Mc , since the small parameter controlling corrections due to non-
renormalizable operators is given by Mc /Λ ∼ (g42 / lD )1/(D−4) , and can thus be lowered by
effects that tend to increase the effective gauge coupling.4
√ is given by l5 = 24π
In our 5D model, the loop factor 3 and the 5D and 4D gauge
couplings are related by g4 = g5 / 2πR, so that Λ ∼ (12π )/g42 . Considering respectively

2
the strong and the weak interactions, this would give roughly Λc ∼ 10/R and Λw ∼
100/R. This means that Λ can be identified with Λc and the theory is indeed reasonably
predictive. In particular, the universal wave-function corrections for the H , Z and W fields
2 / l )Λ, that is Λ /Λ , and therefore represent small corrections
are proportional to (g5,w 5 c w
that can be neglected. We can then go a step further and ask whether this minimal 5D
model is also quantitatively a phenomenologically viable model. As we will see, this not
quite the case, because the values predicted by the model for 1/R, mH and mtop are too
low.
The crucial parameter that sets the scale of the model is α, whose value is determined by
minimizing the full effective potential V (α), as in Section 4. Generically, the most relevant
fermionic contribution to the potential is induced by the bulk fermion in the symmetric of
SU(3)w that gives mass to the top quark, as expected. Neglecting the effect of the other
bulk and boundary fermions, we have numerically analyzed the form of V (α) as a function
top
of λtop and of the bulk-to-boundary couplings /i . As discussed in the previous section,
the lowest non-trivial value for α that we get is α 0.16 for δ = 0 and α 0.12 for
δ = 1 (see Fig. 1), which by means of (5) implies 1/R ∼ 500 GeV. This is in conflict
with experimental bounds for models such as ours, with localized interactions that do not
conserve KK momentum, which require roughly 1/R ∼ few TeV [18]. The Higgs mass
computed using (6) is also too low, at most mH ∼ 30 GeV. Finally, the top Yukawa coupling
top
arising from (15)–(18) turns out to be too small for any value of /i and λtop , giving
a bound mtop 65 GeV. We actually have also evaluated the top mass by numerically
solving the exact transcendental equation arising from the effective potential, as discussed
4 Of course Λ is always larger than M , since at lower energies the theory returns 4D, and the above estimate
c
for Λ is accurate only as long as Λ Mc .
in Section 4, but the deviations from the approximate relations (15)–(18) turn out to be
very small and negligible.
It is interesting to notice that all these problems could be alleviated if V (α) had minima
at lower α. For α ∼ 0.01, for instance, 1/R would be well above the current experimental
bounds and mH would increase up to more than 100 GeV. The top mass slightly increases,
but still mtop 110 GeV. This is not yet completely satisfactory, but it goes in the right
direction. One should also remember that these predictions, in particular the top mass,
could be affected by large corrections if the cut-off of the model is low. One should also
remind that all our analysis is based on minimizing the one-loop effective potential V (α),
and it is not easy to estimate how much higher-loop contributions to V (α) can alter the
actual value of α. The latter, as we have seen, is the crucial parameter in these models, and
it would be extremely interesting to find a mechanism that gives rise to a potential with
such low values of α without spoiling the nice features of the model.
5.4. Possible extensions
As anticipated in the introduction, a possible way to lower α is to consider 5D massive

bulk fermions in large SU(3)w representations. For example one can take a completely
symmetric representation of large rank r, with dimension d(r) = (r + 1)(r + 2)/2. This
contains components of charge q = k, where k is an integer ranging from 0 to r. The
multiplicities of the charged states with k = 0 are found to be Nk = 1 + [(r − k)/2], where
[· · ·] denotes the integer part. This information allows us to compute the contribution of
this field to the effective potential by summing up the contributions of all the charged
components computed with Eq. (24). We find that if the rank r 1 and the parameter λ
controlling the mass is not too large, we can have minima for α ∼ 1/r, and furthermore
the second derivative V at the minimum grows very quickly with r, leading indeed to
a substantial improvement of the situation. With r = 6, λ ∼ 2.2 and δ = 1, we obtain
α ∼ 0.13, corresponding to 1/R ∼ 0.6 TeV and mH ∼ 104 GeV, while for r = 8, λ ∼ 3.5
and δ = 0, we get α ∼ 0.096, corresponding to 1/R ∼ 0.8 TeV and mH ∼ 112 GeV (see
Fig. 2).
It should however be noted that matter fields in large representations of the gauge group
will induce electroweak quantum corrections that are enhanced by large group-theoretical
factors T (r). The scale at which the weak coupling becomes non-perturbative is therefore
lowered: Λw → Λw /T (r). It is difficult to give a precise quantitative estimate of T (r),
because it is not universal. To get an order of magnitude, one can use the Dynkin index
of the representation, which in our case is found to be T (r) = r(r + 1)(r + 2)(r + 3)/48
[29]. This shows that Λw rapidly decreases as r increases too much. When Λw becomes
comparable with Λc , the wave-function corrections to the physical masses get out of
control, and only ratios of these masses can be predicted, as long as Λw does not get
too close to 1/R. We believe that values up to r ∼ 10 could be reasonable.
Fig. 2. The full effective potential (in arbitrary units) in the presence of high-rank bulk fermions. Left: r = 8,
λ = 3.47, /1 = /2 = 9 and δ = 0, resulting in mH = 112 GeV and 1/R = 830 GeV. Right: r = 6, λ = 2.23,
/1 = 7, /2 = 1 and δ = 1, resulting in mH = 104 GeV and 1/R = 600 GeV.
6. Localized gauge kinetic terms
In the orbifold models we consider, gauge fields have certainly a bulk kinetic term, but
no symmetry forbids the occurrence of additional localized kinetic terms. It is therefore
interesting to consider the general case in which all of these are present with arbitrary
coefficients, and study the consequences on model building. This kind of situation was first
considered in [30], in the context of non-compact higher-dimensional theories, and more
recently in [19] (see also [31]) for compact orbifolds.5 We will consider for simplicity H -
universal localized kinetic terms for 4D gauge fields only, described by two couplings l1,2
with mass dimension −1. Localized terms involving A5 are allowed, but their presence
considerably complicates the analysis, and we therefore discard them. Denoting 5D and
4D indices with M, N and µ, ν respectively, the Lagrangian for the SU(3)w gauge fields
is then given by
1
Lg = − Tr FMN F MN − l1 δ(y) + l2 δ(y − πR) Tr Fµν F µν . (36)

2
As usual, it will be convenient to introduce dimensionless parameters relating the
coefficients of the localized kinetic terms to the length of the orbifold: ci = (πR)−1 li . For
simplicity, we do not add localized terms associated with the U (1) gauge field. It actually
turns out that their main effect would be a simple rescaling of the coupling constant g and
so we do not lose in generality by discarding them.
6.1. Spectrum
(i)
The wave functions and KK spectrum for the 4D gauge fields Aµ are distorted by
the localized couplings appearing in (36). The effect of the non-vanishing VEV α is
most conveniently taken into account by adopting the point of view of twisted boundary
5 Localized gauge kinetic terms do also naturally occur at tree level in certain string theory models; see, e.g.,
[32].
conditions, in which each component with definite charge q (in a diagonal basis) satisfies
fn (y +2πR, qα) = e2iπqα fn (y, qα). The differential equation defining the wave functions
fn (y, qα) of the mode with mass mn (qα) is found by proceeding as in [19]. It reads
! "
∂y2 + m2n (qα) 1 + 2πRc1 δ(y) + 2πRc2 δ(y − πR) fn (y; qα) = 0. (37)
The general solution of this equation in the interval [−πR, πR] has the form
#
cos(mn y) + βn− sin(mn y), y ∈ [−πR, 0],
fn (y; qα) = Nn (qα) (38)
cos(mn y) − βn+ sin(mn y), y ∈ [0, πR].
$ 2πR
The constant Nn is a normalization factor defined in such a way that 0 |fn |2 dy = 1.
The parameters βn± are fixed by the twisted boundary conditions and the discontinuity at
y = 0, and read
βn± = e±iπqα sec(πqα)(πRmn )c1 ∓ i tan(πqα) cot(πRmn ). (39)
Finally, the spectrum is determined by the discontinuity at y = πR, which enforces the
following transcendental equation for the dimensionless eigenvalues ξn = πRmn :

2 1 − c1 c2 ξn2 sin2 ξn + (c1 + c2 )ξn sin 2ξn − 2 sin2 (πqα) = 0. (40)
For α = 0, Eqs. (38)–(40) reduce to the equations derived in [19]. For ci = 0, they also
correctly reduce to the case of twisted gauge bosons; in particular, the solution of Eq. (40)
is then ξn = π(n + qα).
The deformation induced on the mass spectrum is one of the most important effects
of localized kinetic terms. The mass of the lightest modes can be determined by solving
the transcendental
√ equation (40) in the limit ξ0 1. One finds in this way ξ0 ∼
sin(πqα)/ 1 + c1 + c2 , and self-consistency of the assumption ξ0 1 requires that
α 1 and/or ci 1. This expression must be compared with the value ξ0 = πqα of
the standard case ci = 0, and shows that the relation between masses of light gauge boson
modes with different charges q is distorted if ci = 0. The masses of heavy KK modes
(n 1) are deformed as well, and tend to become lighter; one finds that ξn → (n − 1) if
c1 ∼ c2 whereas ξn → (n − 1/2) if c1 c2 or vice versa.
The deformation induced on the wave functions of the KK modes is also particularly
relevant, because it affects the concept of 4D effective gauge coupling constant, which
is no longer universal. More precisely, one can define an effective gauge coupling
g4,q,n (y) between matter fields localized at y = 0, πR and the nth KK mode of the
gauge bosons with charge q; defining the quantity Zn (qα) = 1 + 2πRc1 |fn (0; qα)|2 +
2πRc2 |fn (πR; qα)|2 , this is found to be
|fn (y; qα)|

g4,q,n (y) = g5 √ . (41)
Zn (qα)
Similarly, one can define a coupling g4,qi ,ni between three gauge bosons with KK modes
ni and charges qi , which is given by6

2πR
3
|fni (y; qi α)|
g4,qi ,ni = g5 dy 1 + 2πRc1 δ(y) + 2πRc2 δ(y − πR) . (42)
i=1
Z ni (q i α)
0
The above equations describe an important distortion of the gauge coupling. In particular,
the strength of the coupling depends on the type of modes and their location. Contrary to
what happens for α = 0 [19], Eqs. (41) and (42) represent a distortion also for the zero-
mode couplings, because the wave function f0 becomes non-constant and q-dependent for
α = 0.
6.2. Contribution to the effective potential
The contributions of gauge fields to the Higgs effective potential are modified by the
localized kinetic terms as well, and must be recomputed. It is convenient to use the
background gauge-fixing condition D M AM = ∂M AM + ig5 [ĀM , AM ] = 0, where ĀM =
7
δM5 (α/g5 R)λ is the background field. This gauge-fixing is not affected by the localized
boundary terms and the ghost fields η therefore have only a bulk kinetic term. After gauge-
fixing, the bulk and ghost kinetic operators are diagonal and proportional to D P , but
P D

the boundary kinetic operator involves the transverse projector D Dρ η − D D
ρ µν µ ν . As
usual, a change of basis is required to diagonalize the couplings to A5 , and there are two
modes with charge q = 1 and one mode with charge q = 2, both for ghost and gauge fields.
After KK decomposition, the kinetic operator of the ghosts and the internal component of
the gauge fields have a simple diagonal form, whereas the one of the four-dimensional
components of the gauge fields is deformed in a non-trivial way. They are given by
A ,gh
Kmn5 = δm,n p2 + m2n , (43)

Aµ 2

Kmn,µν = δm,n ηµν p + mn + c1 + (−1)

2 m+n
c2 ηµν p − pµ pν .
2
(44)
The contribution to the effective action of each type of mode with fixed charge q is given
by Γ (qα) = 1/2 ln det[K Aµ K A5 (K gh )−2 (qα)]. The determinant over the KK and vector
indices can be explicitly performed. For K A5 and K gh this is trivial; for K Aµ , the KK part
can be done by considering a finite-dimensional truncation and recursively increasing the
dimensionality, and the vector part just produces a factor of 3 in the exponent coming from
the trace of the transverse projector. The result is finally
AM
K
Det
(K gh )2
2 3

p2 2 2
p (−1)n
2 3
= p + mn
2
1 + ci − ci . (45)
n n
p2 + m2n n
p2 + m2n
i=1 i=1
6 In (42), the integral over y should be defined in the interval [−/, 2π R − /] to correctly normalize the
boundary contributions.
As for the fermions, the contribution of each mode to the potential naturally splits into the
standard bulk part and a boundary part encoding the effects of the localized interactions.
The bulk part VgA is given by Eq. (28). The boundary corrections can instead be written as
Vgci (qα)
∞
3
2

2

= dx x 3 ln Re 1 + ci xf0 (x, qα) − Re ci xf1 (x, qα) . (46)
16π 6 R 4
0 i=1 i=1
The total contribution of gauge fields to the one-loop effective potential is finally
obtained by summing up the contributions of the three charged modes; it is given by
Vg (α) = 2VgA (α) + VgA (2α) + 2Vgci (α) + Vgci (2α). (47)
Again, the result is finite, thanks to the exponentially soft UV behavior of the functions fδ .
Notice, moreover, that for ci 1, the α-dependence in the boundary contribution tends to
exactly cancel the α-dependence in the bulk contribution (see Fig. 3). This is most easily
seen by putting
(28) and (46) together and simplifying the argument of the logarithm, which
becomes i (cosh x + ci x sinh x) − cos2 (πα). Note that this expression is proportional to
Eq. (40) after the analytic continuation x → iξn , and has therefore the same zeros. This
constitutes a non-trivial consistency check of our result (47). Indeed, the latter could have
been computed as the effective action of the new eigenstates, which would have led to
a logarithm with an argument proportional to n (x 2 + ξn2 ), with zeros at x = iξn . In
this approach, however, performing the product over the KK modes is non-trivial (see
for instance [33]). It is also interesting to observe that, as in the case of the fermions, the
boundary contribution to the effective potential can be rewritten in terms of the determinant
of a two-by-two matrix encoding the fixed-point-to-fixed-point propagation. The localized
gauge kinetic terms (36) do not break the shift symmetry Aa5 → Aa5 + ∂5 ξ a , and thus no
Fig. 3. Gauge contribution to the effective potential (in arbitrary units) in the presence of localized gauge kinetic
terms, with c2 = 0 and increasing values of c1 .
direct divergence is expected in the gauge contribution to the potential at any order in
perturbation theory.
6.3. Effects on model building
The deformations of the mass spectrum, gauge coupling constants and induced effective
potential that we have described in the last two subsections have important consequences
on model building. The analysis for generic values of α and ci is however quite involved,
and we therefore consider only the special situation in which c1 1 and c2 1, for
which a substantial simplification occurs. In this limit, the zero-mode wave-function f0
reduces to a c-independent linear profile. Eqs. (41) and (42) then yield g4,q,0 (0) g4 ,
g4,q,0 (πR) g4 cos(πqα) and g4,q,0 g4 , with
g5
g4 = √ √ . (48)
c 2πR
For non-zero modes, a stronger suppression factor is found for the couplings at y = 0,
whereas those at y = πR remain finite in the limit of large c. As in the case of [19], this
phenomenon tends to suppress four-fermion operators induced from boundary fields at
y = 0 by the exchange of heavy KK modes of the gauge bosons, in spite of the fact that
these are now lighter. Correspondingly, the bounds on R become milder. The W and Z
gauge bosons, that arise from states with charge 1 and 2, respectively, now have masses
equal to7
sin(πα) sin(2πα)
mW √ , mZ √ sec θW . (49)
π cR 2π cR
Finally, the expression for the Higgs mass is affected as well, because the relation between
the 5D and 4D couplings is modified according to (48), and one finds:
g4 R √
mH c V (α). (50)
2
We see that a sizable factor c results in further improvements. The W mass is lowered,
so that the experimental bound on R can be satisfied with higher values of α and becomes
therefore less restrictive. Taking this the other way around, large values of 1/R are easier
to achieve. The problem of obtaining a reasonable value for mtop is basically solved for
high enough values of c. As a rough estimate, one would need c ∼ 100 if both the left- and
right-handed components of the top field are at the same fixed-point, and c ∼ 10 if they are
at different ones. The Higgs mass is also enhanced, and gets higher at fixed VEV α. All
the problems of our original 5D model, namely the too low values for 1/R, mH and mtop ,
can thus be solved by adding large localized kinetic terms. However, even in the limit of
very large c, an unwanted distortion remains. As can be seen from (49), the ρ parameter
7 Here we are neglecting the mixing between the Z and the anomalous A boson that results in a further
X
deformation in mZ .
does not depend on c, and it is given by

m2W
ρ= sec2 (πα) = 1. (51)
m2Z cos2 θW
Moreover, the effective gauge couplings of possible boundary fields located at πR are
deformed by a factor cos(πα) for charged interactions mediated by the W and cos(2πα)
for the neutral ones mediated by the Z. Once again, a phenomenologically acceptable
situation therefore seems to require very low values of α, which in turn is dynamically
determined by the radiatively generated Higgs effective potential.
As already mentioned, the contribution of the gauge fields to the effective potential
is also strongly deformed by localized kinetic terms. For c 1, its α-dependence gets
suppressed and the total effective potential is dominated by the fermion contributions. In
this case, the good features that we found at the end of Section 4, namely the possibility
of obtaining values for α lower than the usual ones, thanks to the boundary couplings,
is ruined and one typically gets back values close to α ∼ 0.3. This clearly goes against
what is needed to exploit the good features associated with localized gauge couplings. In
particular, α ∼ 0.3 gives an unacceptable value for ρ. Owing to the enhancement of the
bulk electroweak coupling, the scale where the latter becomes non-perturbative is lowered,
Λw → Λw /c, as for the case of high rank representations discussed in Section 5.4. This
represents another limitation to an increasing of c, but values up to c ∼ 15 appear to be
reasonable. Notice that for c ∼ 15 and α ∼ 0.3, one would get 1/R ∼ 1 TeV, which turns
out to be compatible with electroweak precision tests, thanks to the fact that the couplings
between matter and KK modes of the gauge fields are suppressed.
Summarizing, we see that the presence of localized gauge kinetic terms can drastically
improve the situation, but only if they are accompanied by low values of α, which allow
some control on the unwanted deformations that these localized terms necessarily produce.
Unfortunately, these low values for α do not appear to be generated in minimal situations.
It would therefore be again of great help to have at our disposal some mechanism that
provides an additional contribution to the potential that could lower the VEV α, without
distorting the electroweak symmetry breaking. As already mentioned in Section 5.4, one
possibility consists in introducing extra bulk fermions in large representations of SU(3)w .
Although rather unusual, such an additional large-rank heavy fermion would lead to an
optimal situation when combined with localized gauge kinetic terms.
It would be interesting to study what happens for generic values of the ci ’s, because it is
not excluded that all the deformations induced by these terms could conspire, in particular
situations, to yield a phenomenologically viable model. On the other hand, it should be
recalled that localized terms, even if not introduced at tree level, are radiatively generated in
the theory and thus a proper study of their effects is necessary to draw definite conclusions
on model building in this context.
Let us conclude this section by noting that the above considerations are valid as long
as one introduces localized gauge kinetic terms for Aµ only. As already said, this is not
a necessary restriction and a localized term involving the 5D field strength FMN could be
considered. The analysis of this case is complicated by the presence of derivatives along
the internal directions and the computation of the effective potential seems much more
involved. The KK spectrum and wave functions of the 5D gauge fields could be quite
different, in particular for the zero-mode sector, and it is not excluded that this case could
be phenomenologically interesting.
7. Outlook
We have studied in detail various aspects of orbifold models with unification of gauge
and Higgs fields, ordinary matter localized at fixed-points, and additional heavy fermions in
the bulk. We have also analyzed the effect of having large localized gauge kinetic terms in
these models. Electroweak symmetry breaking occurs at the quantum level through a rank-
reducing Wilson-line symmetry breaking and is transmitted to matter at the boundaries
by the massive bulk fermions. The main advantage of this mechanism is that the flavor
structure of the SM can be achieved in an elegant way, without spoiling the stability of the
Higgs potential.
We have presented a simple prototype example in 5D, based on the above structure
and the gauge group SU(3)c × SU(3)w × U (1) . For its minimal version, we find that
1/R, mH and mtop turn out to be too low, but acceptable values can be obtained, with a
moderate tuning of the parameters, by adding extra heavy bulk fermions and/or localized
gauge kinetic terms. By doing so, however, the predictive power of the model is lowered.
Most importantly, we have seen that in the presence of localized gauge kinetic terms the
electroweak sector of the theory is distorted in a non-universal and unwanted way.
At this stage, the prototype 5D models that we presented cannot be considered neither
as viable nor as ruled out. To this purpose, we think that a more careful phenomenological
analysis in needed, which should take systematically into account the effect of localized
kinetic terms for bulk fields and possible extra massive bulk fermions. On the other hand,
the general structure that we have illustrated can be applied to similar constructions in more
than five dimensions as well. The main new feature is the presence of a tree-level quartic
potential for the Higgs fields arising from the decomposition of the higher-dimensional
gauge kinetic term. The electroweak symmetry breaking still occurs radiatively, but the
presence of the tree-level term can help in achieving a larger Higgs mass. In particular, 6D
models represent the minimal version of this possibility, with two Higgs doublets [6,7]. We
plan to extend our analysis to this kind of models in a future work.
Acknowledgements
We would like to thank G. Giudice, B. Mele, R. Rattazzi, A. Romanino, M. Salvatori,

A. Strumia and F. Zwirner for many fruitful discussions and suggestions. We also thank
C. Csáki, C. Grojean and H. Murayama for e-mail correspondence. This research work
was partly supported by the European Community through a Marie Curie fellowship and
the RTN Program “Across the Energy Frontier”, contract HPRN-CT-2000-00148. L.S.
acknowledges CERN, and C.A.S. and M.S. the University of Rome “La Sapienza”, where
part of this work was done.
Appendix A. Mode decomposition
The mode decomposition of fields in various representations of the gauge group G in

the presence of a projection P and twist T can be easily obtained as follows. If we denote
R (y) a field multiplet transforming in the representation R of G, we have
by Ψ
R (−y) = ηΨ R(P )Ψ
Ψ R (y), R (y + 2πR) = R(T )Ψ
Ψ R (y), (A.1)
where R(P ) and R(T ) denote respectively the embedding of the projection and twist in
the gauge group in the corresponding representation, and ηΨ = ±1. In a basis in which P
is diagonal, the first relation in (A.1) is easily solved in terms of single-valued fields ΨR .
One simply gets an expansion in cosines or sines for the various components, according to
the eigenvalue of the projection matrix P . The second relation in (A.1) is then satisfied by
taking

ΨR (y) = R Ω(y) ΨR (y). (A.2)
In Eq. (A.2), Ω(y) = exp(iαa τ a y/R) when T is expressed as T = exp(2iπαa τ a ), with τ a
the generators of the Lie algebra of the group G. The field ΨR (y) automatically solves also
the first relation in (A.1) because R[Ω(−y)] = R[Ω −1 (y)] and T P T = P by consistency.
Since (A.2) is simply a non-single valued gauge transformation, we can alternatively
work with the untwisted fields ΨR only. In this gauge, the effect of the twist is encoded
in the VEV for A5 induced by the gauge transformation: A5 = (−i/g)Ω †(y)∂5 Ω(y) =
αa τ a /(gR). In this case, the problem of finding the mode decomposition of the field ΨR
is reduced to the choice of a basis in which its coupling with
A5 is diagonal.
In the following, we will adopt this second point of view and consider the decomposition
of the untwisted fields ΨR , for the case in which R is the fundamental, symmetric and
adjoint representation of G = SU(3), with P and T taken as in (1) √ and (4). It will be
convenient to introduce the factor ηn defined to be 1 for n = 0 and 1/ 2 for n = 0, as well
as basic wave functions of even and odd modes:
1 ny 1 ny
fn+ (y) = √ cos , fn− (y) = √ sin . (A.3)
2πR R 2πR R
We will mainly focus on matter fermions, but our results easily generalize to other fields.
We will denote by Ψ± the left- and right-handed components, which satisfy the first
equation in (A.1) with ηΨ = ± respectively.
A.1. Fundamental
For the fundamental representation we have simply R(P ) = P and R(T ) = T in (A.1).
It is convenient for later purposes to express Ψ± as a sum over all integer modes, both
positive and negative; this is done by defining the negative modes of a given component
as the reflection of the positive modes: Ψ−n = ±Ψn , depending on the parity of the
component. In this way, the mode expansion for the untwisted fields Ψ± is given by
 ∓ 
∞ fn (y)ψn±u

Ψ± (y) = ηn  fn∓ (y)ψn±d  , (A.4)
n=−∞ ±fn± (y)χn±
where we denoted by ψ u , ψ d the up and down components of the SU(2) × U (1) doublet,
and by χ the SU(2) × U (1) singlet.
The basis in which the coupling of the three components of the triplet to the VEV of A5
is diagonal is reached by defining the following new fields:

Ψn±(1) = ψn±u , Ψn±(2) = ηn ψn±d + χn± . (A.5)
Notice that all the modes of Ψn±(2) are physical and correspond to orthogonal combinations
of the physical modes ψn±d and χn± . In this new basis, the wave function is rewritten as
 ±(1) 
∞
ηn fn∓ (y)Ψn
 ±(2) 
Ψ± (y) =  fn∓ (y)Ψn . (A.6)
±(2)
±fn± (y)Ψn
n=−∞
The action of the twist is now diagonal, and amounts to shifting n → n + α in the
coefficients of Ψn±(2) , which describes both the down component of the doublet and the
(i) (i)
singlet. The 4D kinetic Lagrangian for the field Ψ (i) = Ψ+ + Ψ− , defined as L4D =
$ 2πR
0 L5D , is easily computed and reads
∞
∞

(1) (2)
L4D = n(1) i/
Ψ ∂ 4 − m(1)
n Ψ n + n(2) i/
Ψ ∂ 4 − m(2)
n Ψn , (A.7)
n=0 n=−∞
where
n n+α
n =
m(1) , n =
m(2) . (A.8)
R R
A.2. Symmetric
For the symmetric representation, we have R(P ) = P ⊗ P T and R(T ) = T ⊗ T T in

(A.1). Doubling again the modes for convenience, the untwisted wave functions read in
this case:
 √ ± 
∞ ± 2 fn (y)φn±a ±fn± (y)φn±c fn∓ (y)ψn±u
ηn  √ 
Ψ± (y) = √  ±fn± (y)φn±c ± 2 fn± (y)φn±b fn∓ (y)ψn±d  ,
2 √
n=−∞
fn∓ (y)ψn±u fn∓ (y)ψn±d ± 2 fn± (y)χn±
(A.9)
where, as before, we denoted the upper and lower components of the SU(2) × U (1) doublet
by ψ u and ψ d , the singlet by χ and the three components of the triplet by φ a , φ b and φ c .
The diagonal basis is defined by the new fields

χ ± − φ ±b
Ψn±(1) = ηn ψn±u − φn±c , Ψn±(2) = ηn ψn±d + n √ n , (A.10)
2
χ ± + φ ±b
Ψn±(3) = φn±a , Ψn±(4) = n √ n , (A.11)
2
±(1) ±(2)
where all the modes in Ψn and Ψn are now physical. In this way, calling for short
Ψn±(2±4) = Ψn±(2) ± ηn Ψn±(4) , one has
 √ 
∞
± 2 ηn fn± (y)Ψn±(3) ∓fn± (y)Ψn±(1) fn∓ (y)Ψn±(1)
1  ±(1) 
Ψ± (y) = √  ∓fn± (y)Ψn ∓fn± (y)Ψn±(2−4) fn∓ (y)Ψn±(2)  .
2 n=−∞ ±(1) ±(2) ±(2+4)
fn∓ (y)Ψn fn∓ (y)Ψn ±fn± (y)Ψn
(A.12)
From the above expression, we see that the field Ψ ±(2) again appears both in the down
component of the doublet and the singlet. The action of the twist amounts to shifting
n → n + α in the coefficients of Ψn±(1) and n → n + 2α in the coefficients of Ψn±(2) .
The 4D kinetic Lagrangian for the new fields is
∞
∞
(i) (i)
L4D = n(i) i/
Ψ ∂ 4 − m(i)
n Ψ n + n(i) i/
Ψ ∂ 4 − m(i)
n Ψn , (A.13)
n=−∞ i=1,2 n=0 i=3,4
where
n+α n + 2α n n
n =
m(1) , n =
m(2) , n =
m(3) , n =
m(4) . (A.14)
R R R R
A.3. Adjoint
For the adjoint representation, we have R(P ) = P ⊗ P † and R(T ) = T ⊗ T † in (A.1).

The decomposition of untwisted fields reads
∞
ηn
Ψ± (y) = √
n=−∞ 2
 ±f ± (y) Z ± + √1 χ ±

±fn± (y)Yn± fn∓ (y)ψn±u 

n n3 n
 ±
†

× 
±
±fn (y) Yn ∓fn
± (y) Z ± − √1 χ ±
n 3 n f n
∓ (y)ψ ±d 
n ,

fn∓ (y) ψn±u fn∓ (y) ψn±d ± √ fn± (y)χn±

† † 2
3
(A.15)
where, as before, we denoted the upper and lower complex components of the SU(2) ×
U (1) doublet by ψ u and ψ d , the singlet by χ and the three components of the triplet by Z
and Y, Y † . The diagonal basis is defined by
√
±(1)
± ±u

±(2) ±d Zn± − 3 χn±
Ψn = ηn Yn − ψn , Ψn = ηn Re ψn + ,
2
√ ±
3 Zn + χn±
Ψn±(3) = , Ψn±(4) = Im ψn±d , (A.16)
2
±(1) ±(2)
where all modes of Ψn and Ψn are physical and the first is a complex field. Defining
for short Ψn±(2±3) = Ψn±(2) ± √
ηn
Ψn±(3) and Ψn±(2±4) = Ψn±(2) ± iηn Ψn±(4) , we obtain, in
3
the new basis:

 
± √2 ηn fn± (y)Ψn±(3) ±fn± (y)Ψn±(1) fn∓ (y)Ψn±(1)
∞
3
 
1
Ψ± (y) = √  ±f ± (y) Ψ ±(1)
† ∓fn± (y)Ψn±(2+3) fn∓ (y)Ψn±(2+4) 
2 n=−∞  n n .
∓
±(1)
† ±(2+4)
† ±(2−3)
fn (y) Ψn fn∓ (y) Ψn ∓fn± (y)Ψn
(A.17)
The action of the twist amounts to shifting n → n + α in the coefficients of Ψn±(1) and
n → n + 2α in the coefficients of Ψn±(2) . The 4D Lagrangian for the new fields is
∞
∞

† (1) (2)
L4D = n(1)
Ψ ∂ 4 − m n Ψn +
i/ (1) n(2) i/
Ψ ∂ 4 − m(2)
n Ψn
n=−∞ n=−∞
∞
+ n
Ψ (i)
∂ 4 − mn Ψn(i) ,
i/ (i)
(A.18)
n=0 i=3,4
where
n+α n + 2α n n
n =
m(1) , n =
m(2) , n =
m(3) , n =
m(4) . (A.19)
R R R R
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;

I. Antoniadis, C. Munoz, M. Quiros, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309;
A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489;
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410.
[2] N.S. Manton, Nucl. Phys. B 158 (1979) 141;
D.B. Fairlie, Phys. Lett. B 82 (1979) 97;
D.B. Fairlie, J. Phys. G 5 (1979) L55;
P. Forgacs, N.S. Manton, Commun. Math. Phys. 72 (1980) 15;
S. Randjbar-Daemi, A. Salam, J. Strathdee, Nucl. Phys. B 214 (1983) 491;
N.V. Krasnikov, Phys. Lett. B 273 (1991) 246.
[3] H. Hatanaka, T. Inami, C. Lim, Mod. Phys. Lett. A 13 (1998) 2601, hep-th/9805067.
[4] G.R. Dvali, S. Randjbar-Daemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021, hep-ph/0102307.
[5] L.J. Hall, Y. Nomura, D.R. Smith, Nucl. Phys. B 639 (2002) 307, hep-ph/0107331.
[6] I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20, hep-th/0108005.
[7] C. Csaki, C. Grojean, H. Murayama, hep-ph/0210133.
[8] G. Burdman, Y. Nomura, hep-ph/0210257.
[9] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678.
[10] G. von Gersdorff, N. Irges, M. Quiros, hep-ph/0206029;
G. von Gersdorff, N. Irges, M. Quiros, Nucl. Phys. B 635 (2002) 127, hep-th/0204223;
G. von Gersdorff, N. Irges, M. Quiros, Phys. Lett. B 551 (2003) 351, hep-ph/0210134.
[11] E. Witten, Nucl. Phys. B 258 (1985) 75;
L.E. Ibáñez, H.P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25.
[12] Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Phys. Lett. B 129 (1983) 193;
Y. Hosotani, Ann. Phys. 190 (1989) 233.
[13] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;

J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
[14] S. Weinberg, Physica A 96 (1979) 327;
A. Manohar, H. Georgi, Nucl. Phys. B 234 (1984) 189.
[15] Z. Chacko, M.A. Luty, E. Ponton, JHEP 0007 (2000) 036, hep-ph/9909248.
[16] M. Kubo, C.S. Lim, H. Yamashita, hep-ph/0111327.
[17] N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, hep-ph/0212035.
[18] A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030, hep-ph/9911252.
[19] M. Carena, T.M. Tait, C.E. Wagner, Acta Phys. Pol. B 33 (2002) 2355, hep-ph/0207056.
[20] I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69, hep-th/9310151.
[21] C. Kounnas, M. Porrati, Nucl. Phys. B 310 (1988) 355;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75.
[22] L.J. Hall, Y. Nomura, Phys. Rev. D 64 (2001) 055003, hep-ph/0103125;
A. Hebecker, J. March-Russell, Nucl. Phys. B 613 (2001) 3, hep-ph/0106166.
[23] A. Masiero, C.A. Scrucca, M. Serone, L. Silvestrini, Phys. Rev. Lett. 87 (2001) 251601, hep-ph/0107201.
[24] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117;
M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589.
[25] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 516 (2001) 395, hep-th/0103135;
C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169, hep-th/0110073.
[26] R. Barbieri, R. Contino, P. Creminelli, R. Rattazzi, C.A. Scrucca, Phys. Rev. D 66 (2002) 024025, hep-
th/0203039;
L. Pilo, A. Riotto, Phys. Lett. B 546 (2002) 135, hep-th/0202144.
[27] H.D. Kim, J.E. Kim, H.M. Lee, JHEP 0206 (2002) 048, hep-th/0204132.
[28] C.A. Scrucca, M. Serone, M. Trapletti, Nucl. Phys. B 635 (2002) 33, hep-th/0203190.
[29] R. Slanski, Phys. Rep. B 79 (1981) 1.
[30] G.R. Dvali, G. Gabadadze, M.A. Shifman, Phys. Lett. B 497 (2001) 271, hep-th/0010071.
[31] F. del Aguila, M. Perez-Victoria, J. Santiago, hep-th/0302023.
[32] I. Antoniadis, C. Bachas, E. Dudas, Nucl. Phys. B 560 (1999) 93, hep-th/9906039.
[33] E. Ponton, E. Poppitz, JHEP 0106 (2001) 019, hep-ph/0105021.
On the tensionless limit of bosonic strings, infinite

symmetries and higher spins
Giulio Bonelli
Physique Theorique et Mathematique, Universite Libre de Bruxelles, Campus Plaine C.P. 231,
B-1050 Bruxelles, Belgium
Abstract
In the tensionless limit of string theory on flat background all the massive tower of states gets
squeezed to a common zero mass level and the free theory is described by an infinite amount of
massless free fields with arbitrary integer high spin. We notice that in this situation the very notion of
critical dimension gets lost, the apparency of infinite global symmetries takes place, and the closed
tensionless string can be realized as a constrained subsystem of the open one in a natural way.
Moreover, we study the tensionless limit of the Witten’s cubic sting field theory and find that the
theory in such a limit can be represented as an infinite set of free arbitrary higher spin excitations
plus an interacting sector involving their zero-modes only.
PACS: 11.25.-w; 11.25.Sq; 11.30.Ly
1. Introduction
As far as our knowledge in quantum field theory is concerned, field theories of massless
particles enjoy the richest symmetry structures, gauge symmetries. Moreover, massless
particle theories exhibit much nicer quantum properties than massive ones.
It is, therefore, interesting to study seriously the analogous for string theories in order
to understand if richer symmetry structures appear in the limit in which the string tension
vanishes. A first encouraging and inspiring result in such a direction is contained in the
paper by Gross [1].
E-mail address: gbonelli@ulb.ac.be (G. Bonelli).
doi:10.1016/j.nuclphysb.2003.07.002
160 G. Bonelli / Nuclear Physics B 669 (2003) 159–172
The definition of the tensionless limit that we take as our starting point is the one
in which all the massive tower of string states gets squeezed to a common zero point
energy. This corresponds to the limit α → ∞ keeping fixed the typical oscillators variables
describing the string harmonics structure. This already presents the tensionless theory
within a much wider and symmetric picture where infinite global symmetries could appear
mixing all the oscillator modes. Their actual realization is one of the results obtained in
this Letter.
Let us notice that the approach that we follow is different and most probably
inequivalent to the one initiated by Schild about null strings in [20] where the tensionless
limit was considered at fixed σ -model field coordinates rather than at fixed oscillator
variables.
A much inspiring approach to the problem comes from Section 1.5 of [2], where a
possible link between a general approach to free higher spin theories and the tensionless
limit of free open strings was sketched. In this paper we will make that picture more precise
and we will extend it to the interacting case. It will turn out that the resulting formalism
naturally fits with the tensionless limit of string field theory suggesting a general scheme
to extend the model of [2] to the interacting level.
One of the results that we will obtain is that in the tensionless regime the stringy
notion of critical dimension gets lost, since the nihilpotency of the BRST charge is
established for flat backgrounds of any dimension. This means that one can consistently
formulate tensionless string theories in any dimension, their (usual) tensile deformation
being consistent at the critical value of space–time dimension only. As a countercheck,
notice that in the tensionless limit, in fact, the manifest SO(dim −2) multiplet structure of
the light-cone states are no more forced to add up and combine into representations of the
massive little group SO(dim −1) for all the higher massive levels in the string tower.
Within the approach that we develop for the tensionless limit of string theory, we
will be able to identify the U ∗ (gO ) invariance algebra of the free theory which is
the real anti-Hermitian restriction of the universal covering of the Lie algebra gO =
w(1, dim −1) ⊕ sp(∞), where w(1, dim −1) is the Lie algebra of the Weyl group
generated by iso(1, dim −1) and space–time dilatation.1 The elements in gO are given
bilinears in the string creator/annihilator operators. These transformations act mixing and
rearranging the level and the spin structure of the massless higher spin excitations of the
theory.
Moreover, we realize the tensionless limit of the free closed bosonic string as a
constrained sub-system of the open one, the constrain being the level matching condition.
We work out the form of the gauge symmetries of the theory and single out an infinite
amount of global symmetries.
Finally, we calculate the tensionless limit of the three-string vertex of the cubic open
string field theory and find (not surprisingly), that, once the string field is expanded in
ordinary field components, the interaction takes place among the infinite set of field zero-
modes only. This agrees with general expectations from higher spin field theory on flat
space–time.
1 It is an open issue if this could be extended to the full conformal algebra or not. See Section 2.2.
G. Bonelli / Nuclear Physics B 669 (2003) 159–172 161
2. Free tensionless strings
Let us devote this first section to a close plan discussion of the tensionless limit of
free bosonic strings. For later convenience we start from the free open strings with free
boundary conditions in Minkowski flat space.
2.1. Open tensionless free strings
Notice that some of the following starting arguments concerning the open strings are
already in nuce contained in [2] which we take the freedom to expand and complete for
sake of clarity.2 Moreover, the relation between the tensionless limit of string theory and
higher spin theories has already been noticed in [4].
The content of such a system is given by the string center of mass variables (xµ , pµ )
∗ ), with n > 0. They satisfy the usual canonical
and the infinite set of oscillators (anµ , anµ
commutation relations (CCRs)
∗

xµ , pν = iδµν , anµ , amν = ηµν δnm (1)
and the other commutators are vanishing. The string coordinate and its conjugate
momentum are (σ ∈ [0, π])

2α
∗
Xµ (σ ) = xµ + anµ + anµ cos(nσ ),
n
n>0

pµ 1 n ∗

P (σ ) =
µ
+
anµ − anµ cos(nσ ).
π iπ 2α
n>0
In terms of the oscillator variables the Virasoro constraints take the form
√ 1
n−1
∗
Ln = −i 2nα p · an + m(m + n) am · an+m − m(n − m) am · an−m
2
m>0 m=1
(2)
for n > 0,

L0 = α p · p + na ∗ · an and L−n = L∗n (3)
n>0
(where the dot indicates space–time scalar product).

The tensionless
√ limit appears straightforwardly. In fact, just by rescaling L0 → L0 /α

and Ln → iLn / 2|n|α (where n = 0), the Virasoro algebra in the α → ∞ limit gets
contracted to the much simpler algebra
[L0 , Ln ] = 0, [Ln , Lm ] = δn+m L0 (4)
2 Some related ideas already predate the Henneaux–Teitelboim approach, as, for example, [3], but without any
clear connection with string theory and a correct BRST analysis of the spectrum.
despite any previous central extension which scales away. This implies that for the
tensionless limit there is no notion of critical dimension (as a requirement that the central
extension of the constrain algebra should vanish).3 Notice that the algebra (4) is satisfied
by the leading orders in α of the Virasoro generators, that is L0 = p · p, Ln = p · an and
L−n = p · an∗ .
Introducing the relative anti-commuting ghosts4 cn , cn∗ and c0 as well as the anti-ghosts
b’s (normalized by [c0 , b0 ]+ = 1, [cm , bn∗ ]+ = δmn and [cn∗ , bm ]+ = δnm ), we can write the
BRST operator which implement the above constrain algebra, namely
1
Q = c0 L0 + cn L∗n + cn∗ Ln − 2cn∗ cn b0 , Q2 = 0 (5)
2
n>0
which squares to zero if (4) is satisfied.

For the tensionless open string we have
1
QO = c0 p 2 + cn p · an∗ + cn∗ p · an − 2cn∗ cn b0 . (6)
2
n>0
This satisfies
Q2O = 0
for any space–time dimension (therefore, proving the absence of a critical dimension for
the tensionless open string theory). Hence it follows that the conformal anomaly, whose
vanishing typically determines the critical dimension, is a specific property of the tensile
strings. Indeed, one can deform the tensionless constrains with quadratic terms in the
oscillators and recover the tensile string with its proper anomaly cocycle. It is an interesting
open point to check whether or not such a quadratic deformation is unique (once the closure
of the constrain algebra is enforced) and to study if other deformations are meaningful.
The relevant ghost number operator is
1
GO = (c0 b0 − b0 c0 ) + cn∗ bn − bn∗ cn ,
2
n>0
which is anti-Hermitian (in contrast with Q which is Hermitian). With this convention, the
Fock vacuum has ghost number −1/2.
Corresponding to the above structure we can define a free string field theory where the
µ
states are built over the Fock vacuum |0 annihilated by the oscillators an (as well as the
corresponding ghosts cn , bn and b0 ). The free action is
1 1
SO = ψO |QO |ψO , where GO |ψO = − |ψO (7)
2 2
3 Actually, this should be checked for the full algebra including the ghost contributions, but the core result is
all here. See in the following for the full treatment.
4 There is an armless √n rescaling factor with respect to the usual ghost fields which are used in the tensile
string theory, due to the rescaling of the constrains. Moreover there is some α rescalings whose precise form can
be obtained by comparison with the ordinary tensile string BRST charge.
and can be expanded in the field components. Extending the scheme developed in [2] for
finite sets of oscillators (and adapting the reality condition of the string field to our case),
the action (7) can be shown to consist of a superposition of free massless multi-tensors
whose indices symmetry structure is determined by the expansion of the string field in
string creation oscillators.
2.2. Symmetries of the open tensionless free strings
As it is well known, the action (7) admits the gauge symmetries δ|ΨO = QO |Λ, where
|Λ has ghost number −3/2. Moreover, there are also global symmetries given by the real
unitary operators preserving the ghost number and the BRST charge QO . These have to be
unitary to preserve the measure in the string field space and real to preserve the reality of the
string field. Therefore, the related generators are the zero ghost number real anti-Hermitian
operators commuting with QO . The action on the string field is δ|ΨO = Γ |ΨO where
Γ = Γ ∗ = −Γ † , [GO , Γ ] = 0 and [QO , Γ ] = 0.
These operators are generically characterized by an expansion of the form
[m1 ···ml ][n1 ···nl ]
Γ = Γ(l) cm1 · · · cml bn1 · · · bnl (8)
l0
where Γ(0) is the pure oscillator part. The problem of classifying the global symmetries of
the tensionless free open string amounts to solve in general for Γ of the form (8) such that
[Q0 , Γ ] = 0 and then to enforce the real anti-hermiticity constrains Γ ∗ = Γ = −Γ ∗ . Here
we will single out an infinite set of solutions of the above conditions and will not pretend
to solve in full generality such a problem.
We proceed by studying first the problem of finding the Γ s such that the expansion (8)
stops at l = 1. These will form naturally a Lie algebra in the product form g = [Space–
Time] ⊕ [Internal]. Once g will be determined, we can consider the universal covering
algebra U(g) of g formed by all the words in the generators of g (modulo the commutation
relations in g). The elements of U(g) still commute with QO and admit an expansion of
the form (8). Therefore, the real anti-Hermitian elements in U(g) will fit the definition of
generators of global symmetries. We will denote by U ∗ (g) the resulting restriction.
Therefore, we now calculate the Lie algebra g for the tensionless free open bosonic
string. In order to do it, let us notice that the requirement that the expansion (8) stops
at l = 1 means that the commutator of Γ(0) with the constrains is given by a linear
combination with numerical coefficients of the constrains themselves. Since the constrains
are at most linear in the oscillators, then Γ(0) has to be (at most) quadratic in the oscillators.
As far as the space–time symmetries are concerned, notice that the Poincaré generators
are left as they are by the tensionless limit, since they do not depend on α from the very
beginning, i.e., we have
1 i ∗ ∗

Pµ = pµ and Mµν = (xµ pν − xν pµ ) − anµ anν − anν anµ (9)
2 2
n>0
(such that [Mµν , QO ] = 0). Moreover there is an additional space–time dilatation genera-
tor5
1
∗ ∗
D = [x · p + p · x] + i[b0, c0 ] + i bm cm − cm bm , [QO , D] = 0
2
m>0
which completes a space–time Weyl group.

The reader might doubt whether the complete conformal group could be the full space–
time symmetry of the model or not. This could be verified or by building the generators
of the conformal boosts or by implementing the space–time conformal inversion as a
symmetry of the BRST charge QO . We let this point as an open issue.
As far as the internal symmetry part is concerned, we find that the constrain Lie algebra
is acted on by an sp(∞) algebra. It is easy to see that the BRST charge QO commutes
with the bilinears in the oscillators
∗ ∗
l(mn) = am · an − cm bn − cn bm , l(mn) = am · an∗ + cm
∗ ∗
bn + cn∗ bm
∗
,
∗ ∗ ∗ dim −2
hmn = am · a n + cm bn + bm cn + δmn (10)
2
(where “dim” is the space–time dimension).
The above generators close to form the following algebra
[l(mn) , l(pq) ] = 0, [l(mn) , hpq ] = δnp l(mq) + δmp l(nq) ,

[hmn , hpq ] = δnp hmq − δmq hpn ,
∗ ∗
∗ ∗ ∗
l(mn) , l(pq) = 0, l(mn) , hpq = −δnq l(mp) − δmq l(np) ,
∗

l(mn) , l(pq) = hqm δnp + hpm δnq + hqn δmp + hpn δmq , (11)
which is sp(∞).
So we find that gO = [w(1, dim −1)] ⊕ [sp(∞)]—where w(1, dim −1) is the algebra
of the Weyl group generated by iso(1, dim −1) and the dilatation D—and we conclude
that the elements in U ∗ (gO ) are global symmetry generators. It would be interesting to
study if the kind of algebras we are obtaining are related to the Borcherds ones [5].
An important open point has to do with the full determination of the free theory global
symmetry algebra. It would be interesting to check if the set of elements that we singled
out is all the global symmetry algebra or just a subalgebra. With respect to this issue, let
us notice that an infinite subset of global symmetry generators are given by the BRST
exact ones, namely the ones such that Γe = [QO , Γ−1 ]+ , with Γ−1 an arbitrary real anti-
Hermitian operator of ghost number −1. These transformations have to be considered
as trivial and therefore the problem of calculating the global symmetries reduces to the
solution of a BRST cohomology at zero ghost number. This is well defined since, if
[Γ, QO ] = 0, then Γ Γe = Γ [QO , Γ−1 ]+ = [QO , Γ Γ−1 ]+ is exact too and therefore exact
generators form a (two side) ideal.
5 This choice of the dilatation generator implements the vanishing scaling dimension of the string oscillator
variables as well as the proper scaling dimensions of the ghosts.
2.3. Closed tensionless free strings
Another property which appears in the tensionless limit is a clear embedding of the
closed string Hilbert space as a BRST invariant subspace in the open string unrestricted
one. This can be realized as follows. Let us write the closed string coordinates and momenta
by collecting the left/right moving oscillators in a common set split in odd and even, that
is (now σ ∈ [0, 2π])

α
X (σ ) = x +
µ µ
a2n e−inσ + a2n+1 einσ + a2n∗ inσ ∗
e + a2n+1 e−inσ ,
2n
n>0

pµ 1 n
Pµ (σ ) = + −ia2n e−inσ − ia2n+1 einσ
2π 2π 2α
n>0
∗ inσ ∗

+ ia2n e + ia2n+1 e−inσ ,
where the above modes satisfy the same CCRs (1) as above. Calculating the Virasoro
constrains and performing a scaling similar to the one that we already performed in the
open string case, we find that the left over constrains for the tensionless closed string are

p2 = 0, p · an = 0, p · an∗ = 0 and (−1)n nan∗ · an = 0,
n>0
the last one being the level matching condition.
The BRST charge implementing the contracted closed string Virasoro algebra is (we
add here a further ghost/anti-ghost pair c0 and b0 for the level matching condition with
[c0 , b0 ]+ = 1)
1
QC = c0 p 2 + cn p · an∗ + cn∗ p · an − 2cn∗ cn b0
2
n>0

+ c0 n(−1)n an∗ an + bn∗ cn + cn∗ bn (12)
n>0
which satisfies
Q2C = 0
in any dimension. Notice that, with respect to Eq. (6), identifying the set of canonical
coordinates with the open string one, we can rewrite

QC = QO + c0 L where L = n(−1)n an∗ an + bn∗ cn + cn∗ bn
n>0
is the (ghost extended) level matching constrain which satisfies
[QO , L]− = 0.
Therefore, one can interpret the closed tensionless string Hilbert space as an invariant
subspace, singled out by the condition L = 0 on the physical states, of the open tensionless
string one.
Let us notice that we can write an action for the free tensionless closed string field
theory, that is
1
SC = ψC |QC |ψC
2
which can be written as a constrained open tensionless one. By expanding |ψC = |ψO +
|φ and substituting Q = Q + c L, we calculate the Berezin integral over c in the
cO O C O 0 0
scalar product and find
1
SC = ψO |L|ψO − φO |QO |ψO .
2
The equations of motion corresponding to SC are in fact
QO |ψO = 0 and L|ψO = QO |φO

which correctly implement the constrain L = 0 in the BRST quantization scheme.
Let us point out that in fact such a construction embeds the free closed tensionless
string in the big Hilbert space of the open one. Once the ghost number constrain GC |ψC =
− 12 |ψC is imposed, where in our notation GC = 12 (c0 b0 − b0 c0 ) + GO , then this becomes
on the components
GO |ψO = 0 and GO |φO = −|φO

which projects orthogonally to the GO |ψO = − 12 |ψO condition.
The theory still admits gauge and global symmetries. The gauge symmetries act as
δ|ψO = QO |ΛO and δ|φO = L|ΛO − QO |ΛO

where |ΛO and |ΛO have ghost number −1 and −2, respectively. As far as the infinite
global symmetry (11) is concerned, let us point out that it is partly broken by the level
matching constrain and therefore only the subalgebra commuting with the shifted level
matching combination n>0 n(−1)n hnn is present in the free tensionless closed string.
This can be easily calculated to be generated by the elements
∗ ∗
Kmn = l(2m+1,2n), Kmn = l(2m+1,2n) ,
Imn = h2m+1,2n+1 , Jmn = h2m,2n (13)
whose commutation relations can be worked out directly from (11) by simply substitut-
ing (13).
3. Interacting tensionless strings
As we can see, in the tensionless limit the string coordinate field Xµ (σ ) gets undone
since it blows up. From the interacting theory point of view this implies the uncontrolled
string fragmentation which was observed in such a regime in [6]. In few words, the
tensionless string is unable to stabilize dynamically. The string evolves as a incoherent
set of massless particles where the string profile is just constrained to be orthogonal to the
c.m. momentum. Such a picture is intrinsically unstable under interaction, as the results in
[6] demonstrate, and the string profile gets undone.
It is therefore compelling to turn to an alternative picture to represent the string
interaction in the tensionless limit. The natural alternative picture is the string field
interacting theory [7]. In the following we study the tensionless limit of Witten’s cubic
string field theory.
3.1. The tensionless limit of cubic string field theory
The interacting action at generic value of α can be written as
1 go
SO = ΨO |Q|ΨO + V3 |W |ΨO ⊗ |ΨO ⊗ |ΨO , (14)
2 3
where Q is the full BRST charge and V3 | the three strings vertex
|V3 W

=N dp(1) dp(2) dp(3) δ p(1) + p(2) + p(3)

1 √ r,s (r)∗ (s)
× exp − am · an(s)∗ + 2 α
r,s (r)∗
Vmn Vm0 am · p
2
1r,s3 m,n>0 m>0

(1) (2) (3)
+ α V r,s p(r) · p(s) p ⊗ p ⊗ p , (15)
00
r,s
where the Neumann coefficients Vmn are computable numbers (see, e.g., [8]) and N a
normalization constant.
To obtain a picture of the three string vertex in which the tensionless limit can be
taken it is useful to pass from the c.m. momentum to the c.m. position
representation.
This amounts to insert a c.m. completition 1c.m. = s=1,2,3 |x (s) dx (s)x (s) | and to
perform the Fourier transform by integrating over the momenta. This can be done by first
Fourier transforming the δ-function imposing momentum conservation so that the resulting
integrals are Gaussians. The resulting expression reads
|V3 W

1
= N exp − am · an(s)∗
r,s (r)∗
Vmn
2
1r,s3 m,n>0

3D

(3) (1)

1 2π 2 1
× dx dx dx x
(1) (2)
⊗ x (2) ⊗ x (3)
dx D
√
(2π) α detV00

1 √ √
× exp α a (s)∗ V sr − ix (r) − ix V −1 rq α a (t )∗ V qt − ix (q) − ix .
2α n n0 00 m 0m
We can perform the zero tension limit to the above expression which gives
|V3 W O

1 mn
= NO dx exp − V am · an(s)∗
r,s (r)∗
2
1r,s3 m,n>0

dp(s) δ p(s) p(s) .
s=1,2,3

mn
where V r,s r,s
= Vmn −
r,q −1 q,t t,s
1q,t 3 Vm0 (V√
00 ) V0n and we have included some factors in
3D/2
00 (α )
the normalization constant as N = NO detV(2π) D .
As we see, in the tensionless limit the above three string vertex reduces to the zero
momentum sector the interacting part of the theory and causes the ∗-product to project on
the zero mode sector the string fields. This implies that the ∗-product in this limit does not
admit an identity.
Therefore, in the tensionless limit the propagating degrees of freedom remain free and
stay decoupled from the zero momentum sector where all the interaction occurs within the
oscillators. As far as the interacting internal oscillator degrees of freedom are concerned,
the tensionless limit therefore reduces to a zero-dimensional interacting model very much
similar to the matrix models which are conjectured to be at the basis of an M-theory
description.
In formulas, we have therefore
1 go
S = ΨO |QO |ΨO + V3 |W O |ΨO ⊗3 . (16)
2 3
In order to specify in a better way how the tensionless limit of the vertex looks, let us
rephrase the above results in a more abstract context. As it is clear from the results obtained
above, the three string vertex in the tensionless limit satisfies the tensionless limit of the
string overlap conditions, that is
(r)
pµ + pµ(r+1) |V3 W O = 0 and
1 (r+1)
∗(r) ∗(r+1)
√ anµ (r)
+ anµ + (−1)n anµ + anµ cos(nσ )|V3 W O = 0
n
n0
(where σ ∈ [0, π/2]) and is, therefore, determined, up to a normalization, by these overlap
conditions, BRST invariance, reality condition, associativity of the induced ∗-product and
cyclic symmetry. Notice that the very reason for the projection on the zero center of mass
momentum states in the interaction term is the separation of oscillators and momenta in
the overlap conditions for the three (and higher) string vertex.
As far as the ghost oscillators factor is concerned, because of the relative α rescaling
in the ghost sector,
√ this undergoes a treatment similar to the one of the matter sector.
Moreover, the n factors that we use to rescale the ghost modes enter the definition of
the ghost part of the vertex. In our notation the ghost part of the three string vertex is then

∗(r) r,s ∗(s)
exp − cm Vmn bm (17)
mn>0 1r,s3
(t )
as far as the ghost oscillators are concerned and the ghost zero-mode overall factor t c0 .
(t )
With the above ghost competition, the vertex satisfies the condition 1t 3 QO ×
|V3 W O = 0 and has the correct GO ghost number assignment (namely, 1/2 for each entry)
and satisfies the proper tensionless ghost overlap conditions.
Let us comment that the projection on the momentum zero-modes of the ∗-product in
the tensionless limit can be obtained also from the expanded for of the string field action. A
rough idea about this can be inferred, e.g., from the expansion of the tachyon √ contribution
3 −α ln(4/3 3 )∂ 2
to the cubic term of the action (see [9]), that is φ̃ , where φ̃ = e φ. The
α → ∞ limit
of the
above is well defined if the non constant part of φ scales away and
therefore φ̃ 3 → φ03 where φ0 is the φ zero-mode. At the same time, the tachyon mass
parameter lifts to zero.
The result we obtained is in agreement with the general results of higher spin theories
[10] stating that on flat backgrounds there should not exist a consistent interaction scheme
for higher spin fields.
Notwithstanding the non-dynamical form of the interaction terms which we calculated
in the last section, it would nonetheless make sense to check how much of the infinite
symmetry of the free theory is preserved by the interaction term. It seems that this problem
should be studied in the proper oscillator κ-basis where the ∗-product structure gets
diagonalized to a flat Moyal product along the lines suggested in [11]. From that point
of view it is likely that an infinite invariance group of symmetries of the flat symplectic
structure arising in the tensionless case is preserved.
In the spirit of the previous discussion, it is possible in principle to extend the string field
method for the interactions to the higher spin models considered in [2] (the generic case
with arbitrary bosonic and fermionic roots of the Hamiltonian) and to strings on consistent
curved backgrounds.
The specific form of the interacting theory action, i.e., with a proper single three field
vertex, is indicated by strong symmetry arguments that apply more in general. From a
minimal point of view, let us assume the action to be given by
1 g
S = Ψ |Q|Ψ + V3 ||Ψ ⊗3 , (18)
2 3
where V3 | is a general three field vertex (i.e., an element of the third tensorial product
dual Hilbert space) which we take cyclic symmetric and compatible with a given choice
R of reality condition6 on the field |Ψ (which makes Q Hermitian). It is clear that if the
∗-product determined by V3 | and R is associative and if Q acts as a graded differential
with respect to ∗, then the action (18) is invariant under the gauge transformations

|Ψ → |Ψ + |δΨ where |δΨ = Q|Λ − g |Ψ ∗ Λ − |Λ ∗ Ψ . (19)
Conversely, suppose we ask for a symmetry principle which extends the free theory gauge
invariance |δΨ = Q|Λ to an interacting one. Then, assuming (19), then it is a symmetry
6 In OSFT this is the usual Ra µ R = (−1)n a µ —since the conjugation ∗ is the composition of the Hermitian
n n
conjugation and BPZ conjugation—(and similarly on the other variables) and the string zero-modes invariant, but
in general it is a possible choice up to equivalences.
of (18) if the above conditions are satisfied. Notice that this is independent on the specific
nature of the vertex V3 |. By this we mean that a possible overlap condition scheme
which could be posed to single out a given vertex, has not to be understood as necessary
conditions, but rather should instead fit in a spectral scheme classifying the possible graded
differential ring structures allowed by the Hilbert space of fields.
As we see, the scheme which underlies the introduction of interactions in string field
theory along with a deformation of the free theory gauge invariance naturally extends to the
higher spin models studied in [2] and, once it is extended to anti-de Sitter spaces, resembles
very much the approach in [12]. It would be interesting to translate such an approach in the
language of [13] too.
4. Conclusions and open questions
Relations with superstring theories Most of the above considerations extend also
to superstrings. Actually, one of the motivating issues which started the program of
tensionless strings is also the challenge of understanding the M-theory prevision about little
string theory [14]. It seems that here we have some of the ingredients to formulate a string
field theory for closed tensionless superstrings with a U (N) symmetry in six dimensions,
which we would like then to compare with the expected properties of little string theory,
i.e., with the tensionless closed string theory which should describe the microscopic world-
brane dynamics the M5-branes bound states. There is still a basic point to understand in
this direction, namely if and how the string coupling constant is fixed to some “self-dual”
value and by which actual mechanism. This issue points directly to the problem of a correct
formulation of the interacting tensionless closed (super)string field theory which we believe
should develop within the scheme proposed in the previous sections.
Another issue to understand which comes naturally out of the above results (once
we presume a similar behavior to hold for type II A superstrings) is whether the
picture we obtained for string theory in the tensionless regime can be understood as a
decompactification of the matrix picture of M-theory of [15]. Notice that in such a limit it is
no more true that D-particles decouple because of infinitely massive in the weak coupling
regime, the limit being ls → 0 at gs fixed. This corresponds to the eleven-dimensional
2/3
(wrapped membrane) tensionless limit lp → 0 along with R11 / lp = gs fixed.
In principle one can consider other scalings in parallel to the tensionless limit. For
example, one could consider OSFT with a constant background B-field and then perform
a double scaling limit. This could give other interesting possibilities, although with broken
Poincaré invariance and non-commutative space–time geometry.
Another feature of the tensionless limit is the lift of the tachyon states to the common
massless level. This implies the stability of these string field theories and shows that such a
property is proper to the tensile deformation. It could be interesting to understand if some
tachyonic/unstable string sector can condense in a tensionless string field theory regime.
With this respect, it would be interesting to calculate the tensionless limits of open strings
with different Dirichlet boundary conditions. Since it introduces a further length scale,
this could also be seen as a toy model to test the effects of space–time curvature on the
tensionless string.
More general issues It is of clear importance a target space interpretation of the whole
tensionless string theory. In particular, the possibility of characterizing the tensionless
string in curved backgrounds has to be studied as well as the relation between the
microscopic symmetries of the theory and the properties of the target space–time. All this
should be encoded in the very structure of the BRST operator Q and the three string vertex
(or the ∗-product) in a clear geometric way. This extension should find wide application in
gauge/string dualities once anti-de Sitter backgrounds are concerned [16].
A further open point has to do with a proper geometric interpretation of the zero tension
limit from the world-sheet geometry point of view. Actually, the contraction of the Virasoro
algebra to the tensionless algebra, which is possible in general and not only in the free
field realization, awaits a geometric conceptual interpretation which should replace the
two-dimensional conformal symmetry in such a limit. The tensionless contraction of the
Virasoro algebra is actually well defined in general and should be at the heart of the
tensionless limit of string theory on any consistent background. Notice that with respect to
(5), we have Q2 = 0 for any realization of the tensionless string algebra (4).
It is not clear if the infinite symmetries that we have found in the tensionless limit might
somehow survive in the tensile regime. If this would be the case, which a priori seems not
given, this could be a consequence of extra properties of the Neumann coefficients such
as the ones obtained in [17]. Actually this might have a counterpart in the construction in
[18].
The point of checking the global symmetries of the cubic interaction term remains an
open one and cannot be avoided in a complete analysis of the theory.
It seems to the author that a lot of other basic questions regarding the tensionless string
theory and the meaning of its huge symmetry remain to be issued and answered. We hope
that this note has driven the reader’s attention to some expected and unexpected properties
of tensionless strings.
Note added
Recently the interesting paper [19] appeard and its approach to the tensionless limit
of string theory on flat space–time is similar in principle to the one considered in [2] and
here. Let us notice also that the approach that we follow is in principle different to the “null
strings” one initiated by Schild in [20] where the tensionless limit was considered at fixed
σ -model field coordinates rather than at fixed oscillator variables.
Acknowledgements
I would like to thank G. Barnich, M. Bertolini, L. Bonora, N. Boulanger, M. Henneaux

and A. Sagnotti for stimulating discussions and encouragement. This work is supported
by the Marie Curie fellowship contract HPMF-CT-2002-0185. Work supported in part by
the “Actions de Recherche Concertées” of the “Direction de la Recherche Scientifique—
Communauté Française de Belgique”, by a “Pôle d’Attraction Interuniversitaire” (Bel-
gium), by IISN-Belgium (convention 4.4505.86) and by the European Commission RTN

programme HPRN-CT-00131, in which G.B. is associated to K.U. Leuven.
References
[1] D.J. Gross, High-energy symmetries of string theory, Phys. Rev. Lett. 60 (1988) 1229.
[2] M. Henneaux, C. Teitelboim, First and second quantized point particles of any spin, in: C. Teitelboim,
J. Zanelli (Eds.), Proceedings, Quantum Mechanics of Fundamental Systems 2, Santiago 1987, Plenum,
New York, 1987, pp. 113–152.
[3] S. Ouvry, J. Stern, Phys. Lett. B 177 (1986) 335.
[4] D. Francia, A. Sagnotti, On the geometry of higher-spin gauge fields, Class. Quantum Grav. 20 (2003) S473,
hep-th/0212185.
[5] R.E. Borcherds, Generalized Kac–Moody algebras, J. Algebra 115 (1988) 501;
R.E. Borcherds, Central extensions of generalized Kac–Moody algebras, J. Algebra 140 (1991) 330.
[6] D.J. Gross, P.F. Mende, Phys. Lett. B 197 (1987) 129;
D.J. Gross, P.F. Mende, Nucl. Phys. B 303 (1988) 407.
[7] E. Witten, Nucl. Phys. B 268 (1986) 253.
[8] D.J. Gross, A. Jevicki, Nucl. Phys. B 283 (1987) 1.
[9] V.A. Kostelecky, S. Samuel, Phys. Lett. B 207 (1988) 169;
V.A. Kostelecky, S. Samuel, Nucl. Phys. B 336 (1990) 263.
[10] F.A. Berends, G.J. Burgers, H. van Dam, Z. Phys. C 24 (1984) 247;
F.A. Berends, G.J. Burgers, H. van Dam, Nucl. Phys. B 260 (1985) 295;
M.A. Vasiliev, Progresses in higher spin gauge theories, hep-th/0104246.
[11] C.S. Chu, P.M. Ho, F.L. Lin, JHEP 0209 (2002) 003, hep-th/0205218.
[12] M.A. Vasiliev, Higher spin symmetries, star-product and relativistic equations in AdS space, hep-
th/0002183.
[13] C. Fronsdal, Phys. Rev. D 18 (1978) 3624;
M.A. Vasiliev, Yad. Fiz. 32 (1980) 855;
M.A. Vasiliev, Fortschr. Phys. 35 (1987) 741;
D. Francia, A. Sagnotti, Phys. Lett. B 543 (2002) 303, hep-th/0207002;
X. Bekaert, N. Boulanger, Phys. Lett. B 561 (2003) 183, hep-th/0301243.
[14] A. Strominger, Phys. Lett. B 383 (1996) 44, hep-th/9512059.
[15] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hep-th/9610043.
[16] I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 550 (2002) 213, hep-th/0210114.
[17] A. Boyarsky, O. Ruchayskiy, JHEP 0303 (2003) 027, hep-th/0211010;
L. Bonora, A.S. Sorin, Phys. Lett. B 553 (2003) 317, hep-th/0211283.
[18] M.R. Gaberdiel, P.C. West, JHEP 0208 (2002) 049, hep-th/0207032.
[19] U. Lindstrom, M. Zabzine, Tensionless strings, WZW models at critical level and massless higher spin fields,
hep-th/0305098.
[20] A. Schild, Classical null strings, Phys. Rev. D 16 (1977) 1722.
Non-perturbative renormalization of the static

axial current in quenched QCD
ALPHA Collaboration
Jochen Heitger a , Martin Kurth b , Rainer Sommer c
a Westfälische Wilhelms-Universität Münster, Institut für Theoretische Physik,
Wilhelm-Klemm-Strasse 9, D-48149 Münster, Germany
b University of Southampton, Department of Physics and Astronomy,
Highfield, Southampton SO17 1BJ, United Kingdom
c Deutsches Elektronen-Synchrotron DESY, Zeuthen, Platanenallee 6, D-15738 Zeuthen, Germany
Received 27 February 2003; accepted 23 June 2003
Abstract
We non-perturbatively calculate the scale dependence of the static axial current in the Schrödinger
functional scheme by means of a recursive finite-size scaling technique, taking the continuum limit
in each step. The bare current in the O(a) improved theory as well as in the original Wilson
regularization is thus connected to the renormalization group invariant one. The latter may then
be related to the current at the B-scale defined such that its matrix elements differ from the physical
(QCD) ones by O(1/M). At present, a (probably small) perturbative uncertainty enters in this step. As
an application, we renormalize existing unimproved data on FBbare and extrapolate to the continuum
limit. We also study an interesting function h(d/L, u) derived from the Schrödinger functional
amplitude describing the propagation of a static quark–antiquark pair.
PACS: 11.15.Ha; 12.15.Hh; 12.38.Bx; 12.38.Gc; 12.39.Hg; 13.20.He; 14.65.Fy
Keywords: Lattice QCD; Heavy quark effective theory; Static approximation; Non-perturbative renormalization;
Schrödinger functional; Renormalization group invariant
E-mail address: heitger@uni-muenster.de (J. Heitger).
doi:10.1016/S0550-3213(03)00552-2
174 ALPHA Collaboration / Nuclear Physics B 669 (2003) 173–206
1. Introduction
Since the decay constant FB , governing the leptonic decay of a B-meson, is an

essential element in the quantitative analysis of the unitarity triangle, many lattice QCD
investigations have worked towards its determination. However, with its large mass, the
b-quark still escapes a direct numerical treatment [1] and effective theories have to be
used to separate the large mass scale from the low-energy bound-state dynamics. (As an
exception to this rule, it has recently been demonstrated that also finite-volume methods
on lattices with a large number of points represent a possible alternative [2,3].)
The most natural and theoretically appealing effective theory is the static approxima-
tion [4]. It describes the large-mass limit of the theory and is the starting point for a 1/m-
expansion, the heavy quark effective theory. Yet the problems of this framework have been
twofold in the past. (i) The renormalization properties of the static theory are different, i.e.,
the renormalization constant ZA stat of the axial current in (Astat ) = Z stat (µ)ψ̄ γ γ ψ stat
R 0 A d 0 5 b
becomes scale (µ) dependent, thereby entailing an additional uncertainty, and (ii) Monte
Carlo computations of the matrix element itself are difficult. For these reasons, after a sig-
nificant initial effort [5–12], the computation of FB in the static approximation has received
little attention in the recent past.
In the present work, we solve (i) by computing the renormalization factor that relates
the bare operator in lattice regularization to the renormalization group invariant (RGI )
stat
operator. Denoting its matrix element by ΦRGI , one then has a relation
√
FB mB = CPS (Mb /ΛMS )ΦRGI
stat
+ O(1/Mb ) (1.1)
with a function CPS of the renormalization group invariant b-quark mass Mb in units of
the Λ-parameter. It is scale independent, but in practice it is obtained perturbatively and
an uncertainty due to perturbation theory remains, see Section 5. The important task of
a lattice computation of the B-meson decay constant in the static approximation is to
stat
compute ΦRGI .
stat from the bare matrix element follows the one used by
Our strategy to arrive at ΦRGI
the ALPHA Collaboration for the renormalization group invariant quark mass [13]. In
this approach, an intermediate finite-volume renormalization scheme is used to follow the
scale evolution non-perturbatively to high energies (O(100 GeV)), where then perturbation
theory can safely be used to connect to the renormalization group invariants. For a more
detailed explanation of the overall strategy we refer to Ref. [14] and for a preliminary
report on our work to Refs. [15,16].
The present paper is organized as follows. In Section 2 we discuss our intermediate
renormalization scheme, formulated in the Schrödinger functional (SF). It is based on
the renormalization condition introduced in [17], but a modification has been necessary
to achieve good statistical precision. Section 3 contains the numerical determination of
the scale dependence of the current in the SF scheme which is independent of the lattice
formulation. We also relate the current renormalized at some proper low scale to the RGI
current. Section 4 gives our results for the lattice formulation dependent values of the
Z-factor at this low scale. In Section 5 we then discuss Eq. (1.1) and explain in detail how
our results are to be used. As an example we obtain FBstat
s
from published numbers of the
ALPHA Collaboration / Nuclear Physics B 669 (2003) 173–206 175
bare matrix element. We finish with a brief discussion of the results in Section 6. Details
of the numerical and perturbative calculations are described in Appendices A–C.
2. Intermediate renormalization scheme
In this section we introduce our intermediate renormalization scheme. For reasons to be

explained below, it differs from the one originally introduced in Ref. [17]. The perturbative
calculations of Ref. [17] are extended to the new scheme in Appendix B.
We choose a mass-independent renormalization scheme, leading to simple renormaliza-
tion group equations. The scheme is defined using the Schrödinger functional (SF) [18,19],

i.e., the QCD partition function Z = T ×L3 D[A, ψ̄, ψ] e−S[A,ψ̄,ψ] on a T × L3 cylinder in
Euclidean space, where periodic boundary conditions in the spatial directions of length L
and Dirichlet boundary conditions at times x0 = 0, T are imposed on the gluon and quark
fields.1 Their exact form can be found in Ref. [21]. Moreover, we set T = L throughout,
and the renormalization scale µ is identified with the inverse box extension, 1/L. Such a
finite-volume renormalization scheme is chosen, since it allows to study the scale depen-
dence in the continuum limit for a large range of µ [13,22–24]. We can then relate the
quantities renormalized at some low scale µ to the RGI quantities. Reviews of the strat-
egy are found in [14,25,26], and for a more detailed description the reader should consult
Ref. [13] which we will follow quite closely.
As detailed in [17], we consider the SF with vanishing boundary gauge fields and
θ = 0.5. These settings are identical to those used for the quark mass renormalization
in [13] and were motivated by meeting the criteria of good statistical precision of the
Monte Carlo results, well-behaved perturbative expansions of the renormalization group
functions and minimization of lattice artifacts [27]. Static quarks are included as discussed
in Ref. [17], and we use the notation of that reference. Throughout most of this paper, we
formally stay in the framework of continuum QCD; some notation and basic formulae of
the lattice regularized theory, in which the following expressions receive a precise meaning,
are collected in Appendix A.
In contrast to the relativistic current, there is no axial Ward identity which protects the
renormalized static-light axial vector current,
stat
AR 0 (x) = ZA stat
ψ̄l (x)γ0 γ5 ψh (x), (2.1)
from a scale dependence. Its scale evolution is governed by the renormalization group
equation
∂Φ
µ = γ (ḡ)Φ, (2.2)
∂µ
where

Φ ≡ Φ stat =
f| Astat
R 0 |i (2.3)
1 The spatial boundary conditions of the quark fields are only periodic up to a global phase θ [20], an additional
‘kinematical’ parameter.
is an arbitrary matrix element of the renormalized static current. The renormalization group
function γ , the anomalous dimension, has a perturbative expansion
ḡ→0
γ (ḡ) ∼ −ḡ 2 γ0 + γ1 ḡ 2 + γ2 ḡ 4 + O ḡ 6 (2.4)
with a universal leading coefficient [28,29],
1
γ0 = − , (2.5)
4π 2
and γ1 , γ2 , . . . depending on the chosen renormalization scheme.
Non-perturbatively, one computes the change of Φ under finite changes of the
renormalization scale. For a scale factor of two, the induced change defines the step scaling
function,
σAstat (u) = Φ(µ/2)/Φ(µ) = ZA

stat stat
(2L)/ZA (L), (2.6)
whose argument u ≡ ḡ 2 (L)
is taken to be exactly the coupling defined in [24], and as
always in the SF we have µ = 1/L.
2.1. The old scheme
In Ref. [17], a normalization condition was formulated in terms of suitable correlation

functions defined in the SF. It reads
stat
ZA (L)X(L) = X(0) (L) at vanishing quark mass, (2.7)
with
fAstat (L/2)
X(L) = (2.8)
f1stat
and X(0) (L) the tree-level value of X(L). Here, fAstat is a correlation function between
a static-light pseudoscalar boundary source and Astat stat denotes the
0 in the bulk, and f1
correlator between two such boundary sources at x0 = 0 and x0 = T :

1
fAstat (x0 ) = − d3 y d3 z Astat
0 (x)ζ̄h (y)γ5 ζl (z) , (2.9)
2

1
f1stat = − 6 d3 u d3 v d3 y d3 z ζ̄l (u)γ5 ζh (v)ζ̄h (y)γ5 ζl (z) . (2.10)
2L
(For the proper definition of the ‘boundary quark and antiquark fields’ ζ, ζ̄ we refer to
Refs. [17,21].) The two correlators are schematically depicted in Fig. 1, and their explicit
form on the lattice is given in Eqs. (A.10) and (A.13) of Appendix A.1. In the ratio
(2.8) both the multiplicative renormalization of the boundary quark fields and the mass
counterterm of the static field cancel.
We now have to point out a drawback of this scheme that only becomes evident, when
it is implemented numerically. Namely, the lattice step scaling function ΣAstat (u, a/L)
(cf. Eq. (A.20) for its definition), computed by means of Monte Carlo simulations, has
Fig. 1. Sketch of the correlation functions fAstat (x0 ) (left) and f1stat (right). The single and double lines represent
light (i.e., relativistic) and static quark propagators, respectively.
Fig. 2. Comparison of the numerical precision of the lattice step scaling function computed in the scheme of
Ref. [17] with the new one introduced in Section 2.3, which will be used in the rest of this paper. (The symbols
are slightly displaced for better readability.) The statistics of both computations is the same.
large statistical errors at u ≈ 1.5 and larger. In particular, these errors grow with L/a.
This can be inferred from the results tabulated in Appendix A.2 and is illustrated for three
representative coupling values in Fig. 2. For L/a = 12, 16 (which amounts to also calculate
ZAstat
for 2L/a = 24, 32), already around u ≈ 1.5 a precise determination of ΣAstat (u, a/L)
with a reasonable computational effort becomes impossible. The reason for this lies in the
boundary correlator f1stat being part of the renormalization condition Eq. (2.7): it contains
the static quark propagating over a distance T = L. Thus f1stat falls roughly like e−e1 g0 T /a
2
with e1 = 12π1
2 × 19.95 [30] the leading coefficient of the self-energy of a static quark. On
the other hand, the statistical errors fall much more slowly, leaving us with an exponential
degradation of the signal-to-noise ratio.
To circumvent this problem, we now introduce a slightly modified renormalization
scheme. (Therefore, for the rest of the paper, the scheme of Ref. [17]—if mentioned at
all—will only be referred to by labeling the corresponding quantities with an additional
‘old’, e.g., ZAstat → Z stat .) The general idea is to replace f stat containing a static and a
A,old 1
light quark by two boundary-to-boundary correlation functions. One of them contains a
light quark–antiquark pair, the other a static quark–antiquark pair. Both can be computed
with small statistical errors, the latter because the variance reduction method of Ref. [31]
can be applied. Since the static-static boundary correlator has not been studied before, we
discuss it in some detail.
2.2. The static-static boundary correlator f1hh
We define

1
f1hh (x3 ) ≡ − dx1 dx2 d3 y d3 z ζ̄h (x)γ5 ζh (0)ζ̄h (y)γ5 ζh (z) , (2.11)
2L2
represented graphically in the left part of Fig. 3. After integrating out the static quark fields,
f1hh becomes the (trace of the) product of a timelike Wilson line and the complex conjugate
one from boundary to boundary (see Eq. (B.1)). They are separated by x in space. This
quantity is integrated over x1 , x2 but retains a dependence on x3 . In the following we take
d ≡ |x3 | as its argument, where the periodicity of the system in the space directions restricts
it to 0 d L/2.
Upon renormalization the correlation function f1hh becomes
hh
f1 R (d) = e−2δm×L (Zh )4 f1hh (d),
where Zh is the wave function renormalization constant of a static boundary quark field
and δm the linearly divergent static mass counterterm. Therefore, to study the properties
of f1hh further, we form the finite ratio
(f1hh )R (d) f1hh (d)
h(d/L, u) ≡ = at ḡ 2 (L) = u. (2.12)
(f1hh )R (L/2) f1hh (L/2)
Considered on the lattice, it has a continuum limit. As outlined in Appendix B.1, the one-
loop coefficient h(1) (d/L) of the perturbative expansion
h(d/L, u) = 1 + u h(1)(d/L) + u2 h(2) (d/L) + · · · (2.13)
is given by

2 1 d 2
h(1) (d/L) = − . (2.14)
3 2 L
Remarkably, this form holds exactly on the lattice without any a/L-dependence. Some
insight why this is so is presented in the appendix as well. At two-loop accuracy, we do not
expect exact a-independence any more, but still one may hope that the favorable kinematics
keep lattice artifacts small.
Fig. 3. The correlation functions f1hh (x3 ) (left) and f1 (right). The notation is the same as in Fig. 1.
Fig. 4. Non-perturbative d-dependence of h(d/L, u) = f1hh (d)/f1hh (L/2) at two (low and high) values of the
coupling, compared to one-loop perturbation theory. As data points corresponding to various lattice resolutions
are included, the figure also reflects the weak cutoff dependence of this ratio.
The one-loop expression is compared to results from our non-perturbative computation

of f1hh for two representative values of the coupling in Fig. 4. The figure contains non-
perturbative results for L/a ∈ {12, 16, 20, 24, 32} but at the level of our statistical errors,
which are about 1% and smaller, no lattice artifacts of the ratio h can be seen.
For low d, the non-perturbative data for h are well described by c × (L/d) + c , where
the constant c grows with u = ḡ 2 . Hence the correlation function contains a non-integrable
short-distance singularity, which is the reason why we will not integrate over d in the
following. It is easy to see that this singularity is absent up to and including the order u2 ,
but in higher-order terms in perturbation theory such a singularity may appear.
2.3. The new renormalization scheme
Choosing d at its maximum value to further keep discretization errors at a minimum,

we specify our (non-perturbative) renormalization scheme by
stat
ZA (L)Ξ (L) = Ξ (0) (L) at vanishing quark mass, (2.15)
with
fAstat (L/2)
Ξ (L) = . (2.16)
[f1 f1hh (L/2)]1/4
Here, f1 is the correlator between two light-quark pseudoscalar boundary sources,

1
f1 = − 6 d3 u d3 v d3 y d3 z ζ̄1 (u)γ5 ζ2 (v)ζ̄2 (y)γ5 ζ1 (z) , (2.17)
2L
depicted in the right part of Fig. 3. The form of f1 and f1hh on the lattice is given in
Eqs. (A.12) and (A.14) of Appendix A.1. As before, the combination of these correlators
in the denominator of (2.16) is such that the boundary field renormalizations and the mass
counterterm drop out and no other scale but L appears.
For θ = 0.5 the perturbative calculation summarized in Appendix B now yields
1
γ1SF = 0.10(2) − 0.0477(13) Nf , (2.18)
(4π)2
which differs only little from the one in the old scheme [17].
Note that O(a) improvement [21,32] can be applied and is an important ingredient
in practice to reduce the cutoff effects in the numerical simulations (see Appendix A).
Returning to Fig. 2, one observes that the statistical errors of the lattice results are indeed
much smaller in the new scheme.
3. Non-perturbative running and renormalization group invariant
In this section we present our quenched results on the evolution of Φ(µ) over more than
two orders of magnitude in µ. To this end we consider the evolution of ZA stat under repeated
changes of the scale (i.e., the box size L) by a factor of two at fixed bare parameters.
Starting at some initial low-energy value (i.e., some large L = Lmax ), one thereby climbs
up the energy scale by repeated application of the inverse of the step scaling function until
the perturbative domain at high energies (i.e., small 2−k Lmax ) is reached, where finally
the associated (scale and scheme independent) renormalization group invariant may be
extracted. As in the previous section we keep the discussion in the continuum theory here;
the underlying lattice calculations are described in Appendix A.
3.1. Step scaling function
stat
The evolution of ZA from size L to 2L is given by its step scaling function, σAstat (u),
which has already been introduced in Eq. (2.6), but where it is understood that ZA stat is
defined in the new renormalization scheme according to Eq. (2.15).

As detailed in Appendix A.2, the sets of lattice parameters (L/a, β, κ), which in
practice are required to non-perturbatively compute σAstat (u), can be taken over from the
quark mass renormalization [13]. The available coupling values u allow to trace the scale
dependence of ZA stat
up to L = 2Lmax , where the scale Lmax is implicitly defined through
ḡ 2 (Lmax ) = 3.48. (3.1)

The sequence

uk = ḡ 2 2−k Lmax , k = 0, . . . , 8, (3.2)
is known from Ref. [13], and thus the corresponding sequence
stat (2−k+1 L
ZA max ) Φ(2k−1 /Lmax )
vk ≡ = , v0 = 1, (3.3)
ZA stat
(2Lmax ) Φ((2Lmax )−1 )
Table 1
Results for the continuum step scaling function σAstat (u)
u σAstat (u)
1.0989 0.9796(20)
1.3293 0.9746(25)
1.4300 0.9719(26)
1.5553 0.9668(27)
1.6950 0.9727(28)
1.8811 0.9632(28)
2.1000 0.9589(35)
2.4484 0.9432(27)
2.7700 0.9423(41)
3.4800 0.9154(41)
stat and its continuum limit extrapolations for some selected values of u.
Fig. 5. Lattice step scaling function ΣA
is simply given by
vk
v0 = 1, vk+1 = , (3.4)
σAstat (uk )
once the function σAstat (u) is available in the corresponding range of u.
The calculation of the lattice step scaling function and its subsequent continuum
extrapolation yields the pairs u and σAstat(u) listed in Table 1. An impression of the quality
of the continuum extrapolation is gained from Fig. 5, but for a more detailed account of
the lattice simulations and data analysis we refer to Appendix A.2. An interpolating fit
of σAstat (u) is shown in Fig. 6. The leading coefficients (s0 and s1 , see Appendix C.1) of
the interpolating polynomial are fixed to the perturbative predictions, Eqs. (B.14). This fit
Fig. 6. Continuum step scaling function σAstat (u) and its polynomial fit.
is then inserted into the aforementioned recursion, and propagating all errors through the
recursion we obtain
stat
ZA stat
(2Lmax )/ZA (L) = 0.7551(47) at L = 2−6 Lmax , (3.5)
with the value of the coupling at this box size being ḡ 2 (2−6 Lmax ) = 1.053(12) [13]. Let us
emphasize once more that L = 2Lmax and L = 2−6 Lmax represent low- and high-energy
scales, respectively, which in this way have been connected non-perturbatively. (Our data
actually allow to go up to L = 2−8 Lmax .)
3.2. RGI matrix elements of the static axial current
We now proceed to relate the renormalized matrix element
Φ(µ) = ZA
stat
(L)Φbare (g0 ), µ = 1/L, (3.6)
at L = 2Lmax to the renormalization group invariant one defined by2

ḡ(µ)
−γ0 /2b0 γ (g) γ0
ΦRGI = Φ(µ) 2b0ḡ (µ) 2
exp − dg − , (3.7)
β(g) b0 g
0
2 In a loose notation, we take sometimes L and sometimes µ = 1/L as the argument of ḡ.
Fig. 7. Numerically computed values of the running matrix element of the static axial current in the SF scheme
compared to perturbation theory. The dotted and solid lines are obtained from Eq. (3.7) using the 1/2- and
2/3-loop expressions for the γ - and β-functions, respectively, as well as ΛLmax = 0.211 from Ref. [13].
with the universal leading-order coefficients b0 = 11/(4π)2 and γ0 = −1/(4π 2 ) of the β-

and γ -functions, respectively. Casting this equation in the form
ḡ(µ)

ΦRGI ZAstat
(1/µ) −γ0 /2b0 γ (g) γ0
= stat 2
2b0 ḡ (µ) exp − dg − ,
Φ((2Lmax )−1 ) ZA (2Lmax ) β(g) b0 g
0
with µ = 26 /Lmax , we see that the first factor is known from Eq. (3.5), while in the second
one only couplings ḡ 2 1.05 contribute and it can safely be evaluated by perturbation
theory. Still, for the perturbative error to be negligible, γ has to be known to two-
loop accuracy and β to three-loop. Upon inserting ḡ 2 (26 /Lmax ) = 1.053 and numerical
integration of the second factor we find
Φ(µ)/ΦRGI = 1.088(8) at µ = (2Lmax )−1 (3.8)

in the SF scheme. Entirely consistent numbers, with slightly larger errors, are obtained
for Φ(µ)/ΦRGI if one switches to perturbation theory at µ = 27 /Lmax or µ = 28 /Lmax
instead.
In Fig. 7 we compare the numerically computed running with the corresponding curves
in perturbation theory. While good agreement with the perturbative approximation is seen
at high scales, a growing difference of up to 5% becomes visible when µ is lowered to
µ ≈ 2.5 Λ.
Below it will be more convenient to specify the scale µ in Eq. (3.8) in terms of r0 [33]
instead of Lmax . Taking also the small error contribution from the uncertainty of Lmax in
units of r0 , Lmax /r0 = 0.718(16) [34], into account, the final result for the regularization
independent part Φ(µ)/ΦRGI of the total renormalization factor is

Φ(µ)/ΦRGI = 1.088(10) at µ = (1.436 r0)−1 . (3.9)
Note that this result refers to the continuum limit so that the error on Φ(µ)/ΦRGI of about
0.9% should only be added in quadrature to the proper matrix element under study after its
continuum extrapolation.
stat
4. ZA at low scale and total renormalization factor
We still need to relate (Astat R )0 (µ), renormalized at some appropriate scale µ, to the
bare lattice operator. This amounts to computing ZA stat at the low-energy matching scale
L = 2Lmax = 1.436 r0, which is briefly explained in Appendix C.2. Since in this step
stat
the bare operator is involved, ZA does depend—in contrast to the result of the previous
section—on the choice of action. We have considered three different cases. The first two
are the non-perturbatively O(a) improved action of Ref. [35], with cA stat = − 1 g 2 (= one-
4π 0
loop) and separately with cA = 0. Their combination will in the future allow to study the
stat
influence of cA stat on the continuum extrapolations of renormalized matrix elements. The
third choice is the unimproved Wilson action which is of interest, because so far the best
computations of the bare matrix element did not use improvement [10,11].
The numerical results for ZA stat are shown in Fig. 8. For later use they are represented by
interpolating polynomials,

ZAstat
(g0 , L/a)L=1.436 r = zi (β − 6)i , (4.1)
0
i0
stat (g , L/a)|
Fig. 8. Numerical results for ZA 0 L=1.436 r0 together with their interpolating polynomials.
Table 2
Coefficients of the interpolating polynomials, Eqs. (4.1) and (4.3). Uncertainties are discussed in the text
csw , cA stat
cA Applicability i zi fi
non-perturbative −g02 /(4π ) 6.0 β 6.5 0 0.6944 0.6382
[35] 1 −0.0946 −0.0869
2 0.1239 0.1139
non-perturbative 0 6.0 β 6.5 0 0.6750 0.6204
1 −0.0838 −0.0771
2 0.1200 0.1103
0, 0 0 5.7 β 6.5 0 0.6090 0.5598
1 −0.1186 −0.1090
2 0.3438 0.3160
3 −0.2950 −0.2711
with coefficients zi as listed in Table 2. The statistical uncertainty to be taken into account
when using this formula is about 0.4%.
The total renormalization factor
The total renormalization factor to directly translate any bare matrix element Φbare (g0 )
of Astat
0 into the RGI matrix element, ΦRGI , can be written as

ΦRGI
ZRGI (g0 ) = × ZA stat
(g0 , L/a)L=1.436 r . (4.2)
Φ(µ) µ=(1.436 r0)−1 0
We combine Eq. (4.1) with Eq. (3.9) and represent the total Z-factor by further inter-
polating polynomials,

ZRGI (g0 ) = fi (β − 6)i , (4.3)
i0
whose coefficients are also found in Table 2. These parametrizations of ZRGI are to be
used with an uncertainty of about 0.4%3 at each β-value and an additional error of 0.9%
(from ΦRGI /Φ(µ)), which remains to be added in quadrature after performing a continuum
extrapolation.
5. Matrix elements at finite values of the quark mass
In order to use results from the static theory, one still has to relate its renormalization
group invariant matrix elements to those in QCD at finite values of the quark mass, m.
This step may also be seen as a translation to another scheme, defined by the condition
that matrix elements in the static effective theory renormalized in this scheme and at scale
3 Only in the case c = 0 the error to be associated with the formulae for Z stat and Z
sw A RGI grows to 0.5% at
β ≈ 6.3 and 0.8% at β ≈ 6.5.
µ = m are the same as those in QCD up to 1/m-corrections. This scheme is therefore

denoted as the ‘matching scheme’ [14]. Below, we will specify precisely which quark mass
m is to be taken.
5.1. Conversion to the matching scheme
Let us write the relations for the special case of the matrix element of the axial current
between the vacuum and the heavy-light pseudoscalar,
ΦRGI = ZRGI
PS|Astat
0 |0. (5.1)
We then have
√ PS (m̄)ΦRGI + O(1/m̄),
FPS mPS = C (5.2)
PS (µ) is given
where m̄ is the MS quark mass at renormalization scale m̄.4 The function C
by
ḡ(µ)

γ0 /2b0 γ (g) γ0

CPS (µ) = 2b0ḡ (µ)
2
exp dg − , (5.3)
β(g) b0 g
0
with ḡ(µ) the MS running coupling and γ the anomalous dimension in the matching
scheme. The latter is known to two loops [30,38–40] with γ0 being the same as before
and
b0
γ1 ≡ γ1match = γ1MS − , (5.4)
3π 2

1 127
γ1 = −
MS
+ 28ζ(2) − 5Nf . (5.5)
576π 4 2
PS (µ) is plotted in the upper part of Fig. 9, where for the numerical
For illustration, C
evaluation the β-function is always taken at four-loop precision [41], while to estimate
the perturbative uncertainty we show the result for the one-loop and the two-loop
approximation of γ .
Eq. (5.3) can be rewritten in a form displaying explicitly that also this step is not
restricted to perturbation theory. In terms of the renormalization group invariant quark
mass, M, we have
√
FPS mPS = CPS (M/ΛMS )ΦRGI + O(1/M), (5.6)
where now only renormalization group invariants enter. To evaluate CPS (M/ΛMS ) in
PS , Eq. (5.3), by
perturbative approximation, one changes the argument of the function C
4 Note that in [36] a similar equation with the pole mass instead of the MS mass is written. At the two-loop
order, which will be used below, this does formally not make any difference. However, the pole mass does not
have a well-behaved perturbative expansion [37], and we therefore prefer a short-distance mass such as the MS
mass.
Fig. 9. Conversion factor to the matching scheme, which translates the RGI matrix element to the one at finite
mass.
inserting

ḡ(µ)
−d0 /2b0 τ (g) d0
M/m̄(µ) = 2b0ḡ (µ) 2
exp − dg − (5.7)
β(g) b0 g
0
together with the condition m̄(mMS ) = mMS , where τ (ḡ) denotes as usual the renormaliza-
tion group function of the renormalized (running) quark mass with universal leading-order
coefficient d0 = 8/(4π)2 . A numerical evaluation (with the four-loop τ -function [42,43] in
Eq. (5.7)) is shown in the lower part of Fig. 9. Eq. (5.6) is not only the cleanest form from
a theoretical point of view but it is also practical, because the relation between bare quark
masses in the O(a) improved lattice theory and the renormalization group invariant mass
M is known non-perturbatively in the quenched approximation [44,45].
For later convenience we represent CPS in terms of the variable x ≡ 1/ ln(M/ΛMS ) in a
functional form motivated by Eq. (5.3),

CPS = x γ0 /2b0 1 − 0.065 x + 0.048 x 2 , x = 1/ ln(M/ΛMS ) 0.52, (5.8)
with b0 = 11/(4π)2 and γ0 = −1/(4π 2). It describes the result for the two-loop
approximation of the γ -function within less than 0.01%. Of course in this step a
perturbative error is involved, which is difficult to estimate. Assuming a geometric growth
Table 3
Matrix elements of Astat
0 in units of r0 for the light quark mass equal to the strange quark mass. Bare matrix
elements come from Ref. [10], with the exception of β = 6.0 which is taken from Ref. [11]. The scale r0 /a [33]
is used as determined in Ref. [34], and its uncertainty is already included here
β = 6/g02 5.7 5.9 6.0 6.1 6.3
r0 /a 2.93(1) 4.48(2) 5.37(2) 6.32(3) 8.49(4)
3/2
r0 Φbare 4.75(25) 4.09(21) 3.94(13) 3.79(36) 4.00(29)
3/2
r0 ΦRGI 2.99(16) 2.35(12) 2.21(7) 2.09(20) 2.20(16)
of the coefficients of the γ -function, we find that the γ2 -term would cause a change
by around 1% at m̄ = mb,MS and by 2.5% at m̄ = 1.2 GeV. Thus one may attribute a
2–4% error due to the perturbative approximation, which could be much reduced by a
computation of the three-loop anomalous dimension.
In principle, CPS may be computed also non-perturbatively following the strategy
outlined in Ref. [14]. It is then defined only up to 1/M-terms, consistent with Eq. (5.6).
5.2. Application: first non-perturbative renormalization of FBstat

s
We now take bare matrix elements of Astat0 for unimproved Wilson fermions from the
literature to obtain an estimate for FBs in the static approximation. This exercise serves
mainly to illustrate how to use our results.
5.2.1. The matrix elements are needed at a fixed value of the light quark mass. To
avoid issues in the extrapolation to very light quarks, we here consider only FBs . To
fix the strange quark mass, we use that the sum of the light quark masses is to a good
approximation proportional to the squared (light-light) pseudoscalar masses, m2PS (l, l) [44],
and interpolate the data for the decay constant of [10,11] as a function of m2PS (l, l)r02 to
m2PS (s, s)r02 = (2m2K − m2π )r02 = 2m2K r02 /(1 + ml /ms ) = 3.0233. (To arrive at the latter, we
employed m2K r02 = 1.5736 [44] and ms /ml = 24.4 from chiral perturbation theory [46].)
3/2
The resulting dimensionless numbers r0 Φbare are listed in Table 3.
5.2.2. We renormalize by multiplying with Eq. (4.3), using the fi from Table 2
(csw = cA = cA stat
= 0), take into account a 0.4–0.5% error from the non-universal part
of the Z-factor at each value of g0 and find the last line in Table 3. Assuming the leading
linear behavior in the lattice spacing to dominate for a/r0 < 1/4, we extrapolate to the
continuum limit as shown in Fig. 10. Adding to the extrapolation error in quadrature also
the 0.9% error contribution of ZRGI , which is independent of g0 , yields
3/2
r0 ΦRGI = 1.93(34) at a = 0. (5.9)
5.2.3. Finally, inserting Mb r0 = 17.6(5) [14,47] and ΛMS r0 = 0.602(48) [13], one
gets via the formula in Eq. (5.8)
CPS (Mb /ΛMS ) = 1.23(3), (5.10)
Fig. 10. Continuum extrapolation of the non-perturbatively renormalized matrix element of Astat
0 based on the
unimproved Wilson data for FBbare from Refs. [10,11].
where a 2% error for the perturbative approximation is assumed. With the experimental
spin-averaged B-meson mass mB = mBs = 5.4 GeV, we then obtain from Eq. (5.6):
FBstat
s 0
r = 0.64(11), (5.11)
FBstat
s
= 253(45) MeV for r0 = 0.5 fm. (5.12)
The result contains all errors apart from the uncertainty owing to the quenched approxima-
tion. Evidently, the continuum extrapolation may be done much better, once O(a) improved
results with sufficient precision and small lattice spacings are available.
6. Discussion
We have performed the scale dependent renormalization of Astat 0 by constructing

a non-perturbative renormalization group in the Schrödinger functional scheme, and
agreement with perturbation theory at large scales was demonstrated. The renormalization
factors needed to extract the associated RG invariant are computed with good numerical
accuracy, which is a crucial prerequisite for a controlled determination of FB in the static
limit. In Ref. [14] it was shown that the renormalization factors obtained in this way
differ appreciably from earlier estimates [10] based on tadpole-improved perturbation
theory [48]. Hence their non-perturbative computation is important.
We have not emphasized this so far, but our computation provides the scale dependence
of all static-light bilinears
OΓ (x) = ψ̄l (x)Γ ψh (x). (6.1)

They are renormalized by
(Oγk )R (x) = ZA
stat
Oγk ,
stat Oγ0 ,
(Oγ0 )R (x) = ZZ stat Oγk γ5 ,
(Oγk γ5 )R (x) = ZZ (6.2)
A A
with a scale independent renormalization Z. This pattern is due to the heavy quark spin
symmetry which is exact on the lattice, and due to the chiral symmetry of the continuum
theory. The latter means that the relative renormalization Z may be fixed by imposing a
suitable chiral Ward identity [49] and is thus scale independent.
Returning to the case of most interest, FBstat , our continuum extrapolation in Section 5.2
that uses unimproved data for the bare matrix elements from the literature and also
quite large lattice spacings leaves much room for improvement of the present result,
FBstat
s
= 253(45) MeV. Apart from the obvious step of obtaining O(a) improved bare matrix
elements at small lattice spacings and extrapolating to the continuum, it will be necessary
to estimate the O(1/M) correction. There are two possible roads towards this goal.
An elegant and clean way is to compute the 1/M-corrections directly as perturbations
to the static effective theory. Again, the main problem here is renormalization. Indeed, this
is a severe one, since mixing between operators of different dimensions has to be taken
into account. This will require much more theoretical and numerical effort; but a possible
strategy exists [14,47].
In the mean time, one may also compare the prediction from the static approximation
to what one obtains at M ≈ Mc and also to the results obtained directly at M = Mb , most
notably the ones of Refs. [2,3]. As emphasized in Refs. [50–52] this should be done in
the continuum limit, since O(a) errors get enhanced when the quark masses increase.
At the charm quark mass these are sizable but can be extrapolated away, at least in the
quenched approximation [53]. The comparison between finite-mass decay constants and
√
FBstat is most conveniently done by comparing FPS mPS /CPS (M/ΛMS ). Unfortunately, at
present the error of the static result is still too large to draw any strong conclusions about
1/M-corrections in FB :
3/2
r0 ΦRGI = 1.93(34), static: Eq. (5.9), (6.3)
√
3/2 FBs mBs
r0 = 1.46(23), using FBs = 192(30) MeV [2,3], (6.4)
CPS (Mb /ΛMS )
√
3/2 FDs mDs
r0 = 1.29(5), using FDs = 252(9) MeV [54]. (6.5)
CPS (Mc /ΛMS )
Still, the difference of Eqs. (6.3) and (6.5) shows that there are significant 1/M-corrections
in the charm mass region.
As a more technical remark we point out that the function h(d/L, u), Eq. (2.12), shows
very small a-effects in the quenched approximation and may be worth studying to verify
improvement with dynamical fermions.
Acknowledgements
We thank DESY for allocating computer time on the APE-Quadrics computers at DESY
Zeuthen to this project. This work is also supported in part by the EU IHP Network on
Hadron Phenomenology from Lattice QCD under grant HPRN-CT-2000-00145 and by the
DFG Sonderforschungsbereich SFB/TR 9.
Appendix A. Computation of the lattice step scaling function
This appendix describes some details of the numerical simulations on the lattice and
the subsequent calculations that we have performed in order to determine the step scaling
stat
function for ZA . At the beginning we also recall a few basic definitions and formulae,
which are specific to the inclusion of static quarks and the correlation functions that are
considered in the framework of the Schrödinger functional (SF). As the impact of the static
quarks on O(a) improvement of the static-light sector has extensively been discussed in
Ref. [17], the reader might consult this reference for further details and any unexplained
notation.
A.1. Definitions
A.1.1. Lattice action

The total lattice action is given by the sum
S[U, ψ̄l , ψl , ψ̄h , ψh ] = SG [U ] + SF [U, ψ̄l , ψl ] + Sh [U, ψ̄h , ψh ], (A.1)
where SG and SF are the standard pure gauge and O(a) improved Wilson actions for
relativistic (light) quarks, see Eqs. (A.22)–(A.26) of Ref. [17], respectively, and Sh denotes
the lattice action for the heavy quark:

Sh [U, ψ̄h , ψh ] = a 4 ψ̄h (x)∇0∗ ψh (x). (A.2)
x
The fields ψh and ψ̄h of the static effective theory are constrained in such a way (namely
P+ ψh = ψh and ψ̄h P+ = ψ̄h ) that one is left with just two degrees of freedom per space-
time point [4] and only the (time component of the) backward lattice derivative, ∇µ∗ , enters
in the action (A.2). Hence, static quarks propagate only forward in time, which also reflects
in the form of the associated quark propagator,
Sh (x, y) = U (x − a 0̂, 0)−1 U (x − 2a 0̂, 0)−1 · · · U (y, 0)−1
× θ (x0 − y0 )δ(x − y)P+ ,
1
P± = (1 ± γ0 ), (A.3)
2
being just a straight timelike Wilson line.
To impose SF boundary conditions, Eqs. (A.1) and (A.2) are supplemented by
ψl (x) = 0 if x0 < 0 or x0 > L,
ψh (x) = 0 if x0 < 0 or x0 L (A.4)
and
P− ψl (x)|x0 =0 = P+ ψl (x)|x0 =L = 0, (A.5)

while in the pure gauge part the spatial plaquettes at x0 = 0 and x0 = L receive a non-trivial
(and coupling dependent) weight, see Eq. (A.29) of Ref. [17]. In the discretization of the
SF as described in [18,21] we choose zero boundary gauge fields throughout, C = C = 0,
which translates into the boundary conditions U (x, k)|x0 =0 = U (x, k)|x0 =L = 1 for the
lattice gauge field. Similarly, the light and static quark fields at x0 = 0, L are fixed to
appropriate (space dependent) boundary functions; the corresponding boundary conditions
are collected in Eqs. (3.2), (3.3) and (3.5) of Ref. [17] and not repeated here.
A.1.2. SF correlation functions

Observables are then defined as usual through a path integral involving the total action
S. In this work we focus on SF correlation functions that are constructed from the O(a)
improved static-light axial current
stat
AI 0 (x) = Astat
0 (x) + acA δA0 (x),
stat stat
(A.6)
0 (x) = ψ̄l (x)γ0 γ5 ψh (x),

Astat (A.7)
1 ←
− −
←
0 (x) = ψ̄l (x)γj γ5
δAstat ∇ j + ∇ ∗j ψh (x). (A.8)
2
stat
Unless it is indicated differently, the improvement coefficient cA is set to its one-loop
perturbative value,
1 2
stat
cA =− g , (A.9)
4π 0
computed in Refs. [55,56]. On the lattice, in terms of the boundary quark fields ζ, . . . , ζ̄ ,
the correlation functions of these fields, as well as the various types of pseudoscalar
correlators from one boundary to the other that are needed in addition, read explicitly:
1
fAstat (x0 ) = −a 6 Astat
0 (x) ζ̄h (y)γ5 ζl (z) , (A.10)
y,z
2
1
stat
fδA (x0 ) = −a 6 δAstat
0 (x) ζ̄h (y)γ5 ζl (z) , (A.11)
y,z
2
a 12 1
f1 = − 6
ζ̄i (u)γ5 ζj (v)ζ̄j (y)γ5 ζi (z) , (A.12)
L u,v,y,z 2
a 12 1
f1stat = − 6
ζ̄l (u)γ5 ζh (v)ζ̄h (y)γ5 ζl (z) , (A.13)
L u,v,y,z 2
a8 1
f1hh (x3 ) = − ζ̄h (x)γ5 ζh (0)ζ̄h (y)γ5 ζh (z) . (A.14)
L2 x1 ,x2 ,y,z
2
Table 4
The ratio ΞI for θ = 0.5 at tree-level
L/a ΞI (0, L/a)
6 −1.5964837518021
8 −1.5996643156321
10 −1.6011462370857
12 −1.6019540566018
16 −1.6027594410020
20 −1.6031330222299
24 −1.6033361949926
32 −1.6035384024722
Moreover we introduce the two ratios

fAstat (L/2) + acAstat f stat (L/2)
XI (g0 , L/a) = δA
, (A.15)
stat
f1
fAstat (L/2) + acA
stat stat
fδA (L/2)
ΞI (g0 , L/a) = , (A.16)
[f1 f1hh (L/2)]1/4
which are constructed such that the (unknown) wave function renormalization factors of
the boundary quark fields as well as the (linearly divergent) mass counterterm δm cancel
out and only the static current remains subject to renormalization.
A.1.3. Renormalization
In Ref. [17] the renormalization constant ZA stat ≡ Z stat entering the O(a) improved
A,SF
static axial current renormalized in the SF scheme,
stat
AR 0 = ZA stat
1 + bA
stat
amq Astat
I 0
, (A.17)
was defined in terms of the ratio Eq. (A.15) by imposing the renormalization condition
(with m0 , mq and mc as defined in [17])
stat
ZA,old (g0 , L/a) XI (g0 , L/a) = XI (0, L/a) at m0 = mc . (A.18)
stat naturally runs with the scale µ = 1/L. In the present context it will be referred
Thus, ZA,old
to as the ‘old’ scheme, whereas the so-called ‘new’ scheme based on Eq. (A.16) is specified
by
stat
ZA (g0 , L/a) ΞI (g0 , L/a) = ΞI (0, L/a), m0 = mc , L = 1/µ. (A.19)
For θ = 0.5, which is chosen in our simulations, the relevant values of the tree-
level normalization constant ΞI (0, L/a) (or ΞI(0) (a/L) in the notation of Appendix B
summarizing the perturbative calculations) are collected in Table 4. As an aside we remark
that ΞI (0, L/a) = XI (0, L/a) holds.
The critical quark mass is always understood to be defined from the non-perturbatively
O(a) improved PCAC mass in the light quark sector as in Ref. [13] (i.e., for θ = 0 and
T = L, evaluating the associated combination of correlation functions at x0 = T /2).
A.1.4. Lattice step scaling function

The lattice step scaling function of the static axial current is defined through
stat
ZA (g0 , 2L/a)
ΣAstat (u, a/L) = stat at ḡ 2 (L) = u, m0 = mc . (A.20)
ZA (g0 , L/a)
The additional condition m0 = mc from above, referring to lattice size L/a, defines
the critical hopping parameter value, κ = κc . Moreover, enforcing ḡ 2 (L) to take some
prescribed value u fixes the bare coupling value g02 = 6/β to be used for given L/a. In
this way ΣAstat becomes a function of the renormalized coupling u, up to cutoff effects, and
approaches its continuum limit as a/L → 0 for fixed u.
A.2. Simulation details and results
As emphasized before, our quenched lattice simulation and the data analysis are
analogous to Ref. [13], except that the boundary coefficient ct is set to its two-loop
perturbative value [57]:
2-loop
ct = 1 − 0.089 g02 − 0.030 g04. (A.21)
The boundary O(a) improvement terms involving quark fields have to be multiplied with
a coefficient c̃t , which is known to one-loop from [58], viz.
1-loop
c̃t = 1 − 0.018 g02. (A.22)
Of course, owing to a priori unknown precision to which perturbation theory approximates
these coefficients, linear lattice spacing errors are not suppressed completely, and we will
come back to this issue later. As for the other contributing O(a) improvement coefficients,
we used the non-perturbative values for csw and cA of [35] for the relativistic fermions and
stat
the one-loop estimate (A.9) for cA in the static-light axial current.
stat stat
The renormalization constants ZA (g0 , L/a) and ZA (g0 , 2L/a) in Eq. (A.20) have
been evaluated from the correlation functions in Eqs. (A.10)–(A.14), which were computed
in a numerical simulation with θ = 0.5. (The latter parameter specifies the boundary
conditions of the quark fields, see, e.g., [20].) These simulations were performed on
the APE-100 parallel computers with 128 to 512 nodes, employing for the updating
of the gauge fields the same hybrid-overrelaxation algorithm as in [13,24] with two
overrelaxation sweeps per heatbath sweep within a full iteration. This mix of updating was
found to be close to optimal in [59]. Since the computation of SF correlators has already
been detailed in Ref. [35] and Appendix A.2.2 of [13], we just mention that we differ
from them only by using the implementation [60] of the SSOR-preconditioned BiCGStab
inverter [61] to solve the lattice Dirac equation.
The computation of f1hh (d), where d = |x3 | and a d L/2, amounts to evaluate
Eq. (B.1). In order to improve the statistical precision of f1hh , the links building up the
observable are evaluated by a 10-hit multi-hit procedure [31], where each hit consists
of a Cabibbo–Marinari heatbath update in three SU(2)-subgroups of SU(3). Translation
invariance is fully exploited.
In order to keep autocorrelations small, the measurements of the correlation functions
were always separated by L/(2a) update iterations (and, respectively, five iterations in
the case of f1hh ). Statistical errors stem from a standard jackknife analysis, where we
have checked explicitly for the statistical independence of the data by averaging them
into bins of a few consecutive measurements beforehand. The total number of measure-
ments itself was such that the statistical error of ΣAstat was dominated by the uncertainty in
ZAstat (g , 2L/a). In general, the uncertainties in the coupling and in the value of κ would
0 c
have to be propagated into the error of ΣAstat as well. But as the former can be estimated
to be much smaller than the statistical error of ΣAstat and ΣAstat is found to depend rather
weakly on the bare (light) quark mass, we neglected both contributions in the final error
estimate.
In Tables 5–7 we list our results on the step scaling functions of the static axial current5
and—since they are available from our computations as well—of the pseudoscalar density
defined as in [13]. The values of β and the critical hopping parameter κ = κc to be
simulated were taken over from Ref. [13] without changes, which means to stay with ct
and c̃t to one-loop accuracy in realizing the conditions ḡ 2 (L) = u and m0 = mc . Note
once more, however, that here, in contrast to [13], for the corresponding renormalization
constants themselves—particularly when comparing the results for ZP and ΣP quoted in
that previous work with those of the present one—the two-loop formula for ct , Eq. (A.21),
has been used.
A.3. Continuum extrapolation of ΣAstat
For fixed coupling u the step scaling function defined in Eq. (A.20) has a continuum
limit, σAstat (u). Neglecting for the moment the uncertainties on the correct values of ct , c̃t
stat
and cA , we expect the leading-order cutoff effects to be quadratic in the lattice spacing,

ΣAstat (u, a/L) = σAstat (u) + O a 2 /L2 , (A.23)
since O(a) improvement is employed. Based on this ansatz, Fig. 5 in Section 3 illustrates
the continuum extrapolation of ΣAstat for a representative subset of our available coupling
values u = ḡ 2 (L). The coarsest lattices (with L/a = 6) have been omitted from the fits as
a safeguard against higher order cutoff effects. For the remaining a/L 1/8, the one-loop
cutoff effects are quite small, see Fig. 12.
Although these extrapolations are entirely compatible with an approach to the
continuum limit quadratic in a/L, we also have investigated extrapolations linear in a/L.
These as well yield reasonable fits with consistent results and even comparable total
χ 2 /dof (when summing up the χ 2 ’s belonging to the individual fits at the ten u-values)
so that the form of the lattice spacing dependence cannot be decided on the basis of the
data. Therefore, we have studied the influence of the imperfect (i.e., only perturbative)
knowledge of some of the improvement coefficients in more detail.
Since the usage of the two-loop approximation (A.21) for ct in the calculation of the
correlation functions (and thereby also in the step scaling function ΣAstat ) should cancel the
main contributions from the related boundary terms, we only address its effect originating
5 Here we do not tabulate the results on the static-static boundary correlator f hh separately, but the numbers
1
can be obtained from the authors upon request.
Table 5
Results for the step scaling function ΣP
ḡ 2 (L) β κ L/a ZP (g0 , L/a) ZP (g0 , 2L/a) ΣP (u, a/L)
1.0989 9.5030 0.131514 6 0.8190(10) 0.7846(9) 0.9581(16)
9.7500 0.131312 8 0.8107(9) 0.7786(12) 0.9604(18)
10.0577 0.131079 12 0.8012(6) 0.7702(11) 0.9613(16)
10.3419 0.130876 16 0.7956(8) 0.7625(15) 0.9584(21)
1.3293 8.6129 0.132380 6 0.7930(10) 0.7493(10) 0.9449(17)
8.8500 0.132140 8 0.7827(11) 0.7402(11) 0.9456(19)
9.1859 0.131814 12 0.7737(7) 0.7353(12) 0.9503(17)
9.4381 0.131589 16 0.7666(11) 0.7274(18) 0.9488(27)
1.4300 8.5598 0.132453 8 0.7702(10) 0.7273(13) 0.9443(21)
8.9003 0.132095 12 0.7610(6) 0.7223(14) 0.9491(21)
9.1415 0.131855 16 0.7555(7) 0.7123(20) 0.9428(28)
1.5553 7.9993 0.133118 6 0.7666(7) 0.7165(16) 0.9346(23)
8.2500 0.132821 8 0.7590(8) 0.7134(13) 0.9399(20)
8.5985 0.132427 12 0.7473(10) 0.7035(13) 0.9414(21)
8.8323 0.132169 16 0.7421(9) 0.6976(19) 0.9401(29)
1.6950 7.9741 0.133179 8 0.7442(11) 0.6939(15) 0.9325(24)
8.3218 0.132756 12 0.7341(7) 0.6862(15) 0.9348(22)
8.5479 0.132485 16 0.7277(13) 0.6805(18) 0.9352(30)
1.8811 7.4082 0.133961 6 0.7348(9) 0.6764(6) 0.9205(14)
7.6547 0.133632 8 0.7258(7) 0.6691(15) 0.9219(23)
7.9993 0.133159 12 0.7173(5) 0.6632(8) 0.9245(12)
8.2415 0.132847 16 0.7117(13) 0.6604(20) 0.9279(33)
2.1000 7.3632 0.134088 8 0.7088(13) 0.6433(16) 0.9076(28)
7.6985 0.133599 12 0.6971(8) 0.6385(24) 0.9160(36)
7.9560 0.133229 16 0.6919(12) 0.6303(17) 0.9110(29)
2.4484 6.7807 0.134994 6 0.6845(10) 0.6110(12) 0.8925(21)
7.0197 0.134639 8 0.6784(8) 0.6061(19) 0.8933(30)
7.2025 0.134380 10 0.6733(8) 0.6021(12) 0.8943(21)
7.3551 0.134141 12 0.6722(11) 0.6012(12) 0.8944(24)
7.6101 0.133729 16 0.6661(5) 0.5962(10) 0.8950(17)
2.7700 6.5512 0.135327 6 0.6619(10) 0.5758(20) 0.8699(33)
6.7860 0.135056 8 0.6541(13) 0.5751(17) 0.8792(31)
6.9720 0.134770 10 0.6505(8) 0.5717(17) 0.8788(28)
7.1190 0.134513 12 0.6482(9) 0.5705(10) 0.8802(19)
7.3686 0.134114 16 0.6442(16) 0.5668(16) 0.8798(33)
3.4800 6.2204 0.135470 6 0.6173(8) 0.5067(11) 0.8208(21)
6.4527 0.135543 8 0.6133(8) 0.5101(21) 0.8316(35)
6.6350 0.135340 10 0.6112(11) 0.5078(19) 0.8307(35)
6.7750 0.135121 12 0.6076(7) 0.5061(14) 0.8329(24)
7.0203 0.134707 16 0.6063(7) 0.5097(11) 0.8406(21)
from the fixing of the renormalized coupling, the values of which were taken over from
Ref. [13] with ct still set to one-loop. Changing ct from one- to two-loop also in this
step then requires to adjust the bare coupling and the value of the critical quark mass
accordingly before the simulations for ΣAstat can be performed. We have done this analysis
Table 6
stat (in the ‘new’ scheme)
Results for the step scaling function ΣA
ḡ 2 (L) β κ L/a stat (g , L/a)
ZA stat (g , 2L/a)
ZA stat (u, a/L)
ΣA
0 0
1.0989 9.5030 0.131514 6 0.8926(7) 0.8698(8) 0.9745(12)
9.7500 0.131312 8 0.8860(7) 0.8668(11) 0.9782(14)
10.0577 0.131079 12 0.8800(6) 0.8630(11) 0.9806(14)
10.3419 0.130876 16 0.8786(7) 0.8586(16) 0.9773(20)
1.3293 8.6129 0.132380 6 0.8733(8) 0.8458(9) 0.9686(13)
8.8500 0.132140 8 0.8677(9) 0.8418(13) 0.9701(18)
9.1859 0.131814 12 0.8635(7) 0.8401(13) 0.9729(17)
9.4381 0.131589 16 0.8593(9) 0.8361(19) 0.9731(25)
1.4300 8.5598 0.132453 8 0.8593(7) 0.8328(13) 0.9692(17)
8.9003 0.132095 12 0.8545(7) 0.8314(14) 0.9731(18)
9.1415 0.131855 16 0.8516(6) 0.8238(23) 0.9674(28)
1.5553 7.9993 0.133118 6 0.8572(6) 0.8246(13) 0.9619(16)
8.2500 0.132821 8 0.8517(6) 0.8248(13) 0.9684(17)
8.5985 0.132427 12 0.8459(8) 0.8190(14) 0.9683(20)
8.8323 0.132169 16 0.8425(9) 0.8141(20) 0.9662(26)
1.6950 7.9741 0.133179 8 0.8414(9) 0.8069(18) 0.9590(24)
8.3218 0.132756 12 0.8359(8) 0.8074(13) 0.9659(19)
8.5479 0.132485 16 0.8329(13) 0.8081(19) 0.9703(28)
1.8811 7.4082 0.133961 6 0.8362(7) 0.7939(7) 0.9495(11)
7.6547 0.133632 8 0.8290(6) 0.7903(17) 0.9533(21)
7.9993 0.133159 12 0.8247(7) 0.7907(12) 0.9588(16)
8.2415 0.132847 16 0.8221(13) 0.7898(23) 0.9607(32)
2.1000 7.3632 0.134088 8 0.8193(10) 0.7732(20) 0.9436(26)
7.6985 0.133599 12 0.8117(9) 0.7756(25) 0.9555(32)
7.9560 0.133229 16 0.8081(12) 0.7704(21) 0.9533(29)
2.4484 6.7807 0.134994 6 0.8035(8) 0.7420(12) 0.9235(18)
7.0197 0.134639 8 0.7945(7) 0.7444(19) 0.9370(26)
7.2025 0.134380 10 0.7936(9) 0.7431(17) 0.9363(24)
7.3551 0.134141 12 0.7930(10) 0.7465(17) 0.9414(24)
7.6101 0.133729 16 0.7907(8) 0.7444(16) 0.9415(22)
2.7700 6.5512 0.135327 6 0.7886(9) 0.7133(19) 0.9045(26)
6.7860 0.135056 8 0.7791(12) 0.7187(24) 0.9225(35)
6.9720 0.134770 10 0.7786(9) 0.7223(23) 0.9276(31)
7.1190 0.134513 12 0.7740(11) 0.7220(17) 0.9329(25)
7.3686 0.134114 16 0.7755(16) 0.7281(30) 0.9388(43)
3.4800 6.2204 0.135470 6 0.7587(10) 0.6562(39) 0.8649(53)
6.4527 0.135543 8 0.7496(11) 0.6624(29) 0.8837(41)
6.6350 0.135340 10 0.7477(11) 0.6658(31) 0.8906(43)
6.7750 0.135121 12 0.7451(11) 0.6696(22) 0.8988(32)
7.0203 0.134707 16 0.7470(10) 0.6792(24) 0.9092(34)
for the largest fixed coupling, u = 3.48, where the uncertainty in ct is largest and thus
its effect most pronounced. At L/a = 6 the resulting change in ΣAstat turns out to lie
clearly inside the statistical errors, and this effect will even get smaller for decreasing
a/L. On the other hand, if we just compute ΣAstat with the one-loop value of ct as in the
Table 7
stat (in the ‘old’ scheme)
Results for the step scaling function ΣA,old
ḡ 2 (L) β κ L/a stat (g , L/a)

ZA,old stat (g , 2L/a)
ZA,old stat (u, a/L)
ΣA,old
0 0
1.0989 9.5030 0.131514 6 0.8903(8) 0.8676(8) 0.9745(12)
9.7500 0.131312 8 0.8846(7) 0.8635(11) 0.9762(14)
10.0577 0.131079 12 0.8791(5) 0.8628(14) 0.9814(17)
10.3419 0.130876 16 0.8773(9) 0.8601(21) 0.9804(26)
1.3293 8.6129 0.132380 6 0.8721(9) 0.8430(10) 0.9666(16)
8.8500 0.132140 8 0.8664(10) 0.8428(14) 0.9728(19)
9.1859 0.131814 12 0.8616(6) 0.8351(19) 0.9693(23)
9.4381 0.131589 16 0.8580(10) 0.8370(48) 0.9755(57)
1.4300 8.5598 0.132453 8 0.8577(8) 0.8325(14) 0.9706(19)
8.9003 0.132095 12 0.8535(6) 0.8332(21) 0.9761(26)
9.1415 0.131855 16 0.8488(6) 0.8172(55) 0.9628(65)
1.5553 7.9993 0.133118 6 0.8552(7) 0.8243(19) 0.9639(23)
8.2500 0.132821 8 0.8497(6) 0.8224(17) 0.9678(22)
8.5985 0.132427 12 0.8441(8) 0.8198(26) 0.9712(32)
8.8323 0.132169 16 0.8410(12) 0.8203(57) 0.9754(69)
1.6950 7.9741 0.133179 8 0.8411(11) 0.8079(20) 0.9606(26)
8.3218 0.132756 12 0.8340(8) 0.8154(32) 0.9777(39)
8.5479 0.132485 16 0.8328(16) 0.8076(87) 0.970(11)
1.8811 7.4082 0.133961 6 0.8336(7) 0.7904(7) 0.9482(12)
7.6547 0.133632 8 0.8269(7) 0.7921(21) 0.9578(27)
7.9993 0.133159 12 0.8232(5) 0.7863(20) 0.9551(25)
8.2415 0.132847 16 0.8210(18) 0.798(12) 0.972(15)
2.1000 7.3632 0.134088 8 0.8172(13) 0.7778(24) 0.9518(33)
7.6985 0.133599 12 0.8097(8) 0.7757(67) 0.9579(83)
7.9560 0.133229 16 0.8091(21) 0.786(12) 0.971(14)
2.4484 6.7807 0.134994 6 0.8016(9) 0.7416(17) 0.9252(23)
7.0197 0.134639 8 0.7941(9) 0.7399(37) 0.9317(48)
7.2025 0.134380 10 0.7932(9) 0.7422(37) 0.9357(48)
7.3551 0.134141 12 0.7901(13) 0.7382(54) 0.9344(70)
7.6101 0.133729 16 0.7870(9) 0.756(13) 0.961(17)
2.7700 6.5512 0.135327 6 0.7863(10) 0.7132(34) 0.9070(45)
6.7860 0.135056 8 0.7804(15) 0.7169(45) 0.9185(60)
6.9720 0.134770 10 0.7759(10) 0.7121(55) 0.9177(72)
7.1190 0.134513 12 0.7739(12) 0.7152(72) 0.9241(95)
7.3686 0.134114 16 0.7732(30) 0.681(34) 0.880(43)
3.4800 6.2204 0.135470 6 0.7573(9) 0.6558(24) 0.8659(34)
6.4527 0.135543 8 0.7501(9) 0.6560(77) 0.874(10)
6.6350 0.135340 10 0.7476(15) 0.661(11) 0.885(15)
6.7750 0.135121 12 0.7430(10) 0.659(16) 0.886(21)
7.0203 0.134707 16 0.7474(13) 0.642(46) 0.859(61)
computation of the coupling, we found the results, now for u = 2.77 and L/a = 6, 8, to
be indistinguishable within errors, too. We conclude that any small uncertainty present
in ct beyond the available two-loop estimate is numerically unimportant for the cutoff
dependence of ΣAstat .
Regarding the O(a) improvement coefficient c̃t , we followed the same line as in [13]
and assessed its influence on our results by artificially replacing the one-loop coefficient
in the expression (A.22) by ten times its value. I.e., we set c̃t to c̃t = 1 − 0.180 g02 in some
additional simulations at u = 3.48, and the outcome is that the corresponding estimates on
ΣAstat for L/a = 6 still differ by around 1.5%, while for L/a = 8 they already agree within
their statistical errors. As this difference drops further for growing L/a and/or smaller
couplings, a possible imperfection of c̃t does not affect the results on ΣAstat either.
Finally, we also checked for the influence of the O(a) improvement coefficient in the
static-light axial current, cAstat
, by analyzing our data with cA stat
= 0 instead of the one-loop
value (A.9). Whereas the related change in ZA stat is of the order of a few percent and hence
still substantial, it largely cancels in the ratio of Eq. (A.20) so that this effect is no more
significant for ΣAstat given its statistical errors.
All in all these findings demonstrate that at the level of our precision linear a-effects
in the data on ΣAstat are negligible, and extrapolations using (a/L)2 -terms as the dominant
scaling violation are justified indeed.
Appendix B. Perturbation theory
This appendix provides a few details on the perturbative computations, which were
required to obtain the one-loop expression for h(d/L, u), Eq. (2.12), the two-loop
anomalous dimension and the one-loop estimates of the discretization errors of the step
scaling function ΣAstat . Note that here we restrict ourselves to the case of the modified (or
‘new’) scheme introduced via Eq. (2.15) in this paper, because the perturbation theory of
the original scheme defined through Eq. (2.7) has been extensively discussed already in
Ref. [17] where also more details on the different steps involved can be found.
The correlation functions fAstat , fδA
stat
and f1 are expanded in powers of the coupling g02
as explained in [17] and [62], and the analogous expansion of f1hh is explained below.
B.1. The correlation function f1hh
After integrating out the static quark fields, the correlation function f1hh can be written
as
a2
f1hh (x3 ) = 2
tr U (x, 0)U (x + a 0̂, 0) · · · U x + (L − a)0̂, 0
L x ,x
1 2
−1 −1
× U (L − a)0̂, 0 U (L − 2a)0̂, 0 · · · U (0, 0)−1 x , (B.1)
0 =0
where the trace is taken over color indices only.

Writing U (x, µ) = exp{g0 aqµ (x)}, with the gluon field qµ (x) = qµa (x)T a , where T a
are the anti-Hermitian generators of the gauge group, the function f1hh can be expanded in
the bare coupling,

f1hh (d) = 3 + g02 f1hh,(1)(d) + O g04 , d = |x3 |. (B.2)
Fig. 11. One-loop diagrams contributing to f1hh . The two dots on the left are at x0 = 0, the dots on the right are
at x0 = L.
hh,(1)
Determining the one-loop coefficient f1 amounts to calculating and summing the
diagrams shown in Fig. 11 using the gluon propagator Dµν (x, y) (with SF boundary
conditions) given in Ref. [58].
For comparison with the non-perturbative results shown in Fig. 4, Section 2.2, we
consider the one-loop coefficient h(1) (d/L) of Eq. (2.13), which reads
1 hh,(1)
h(1) (d/L) = f1 (d) − f1hh,(1) (L/2) , (B.3)
3
and which we can obtain analytically. (For the other quantities considered in this appendix,
the diagrams are calculated numerically.) Only the last of the diagrams in Fig. 11
contributes, and we thus can write
h(1) (d/L)
4a 6
= 4
D00 (x, y)| x3=d, y3=0 − D00 (x, y)|x3=L/2, y3=0
3L x ,y x ,y x ,y
0 0 1 1 2 2
4a 2
= eip3 d − eip3 L/2 d00 x0 , y0 ; (0, 0, p3) , (B.4)
3L3 p3 x0 ,y0
with the momentum-space gluon propagator d00 (x0 , y0 ; p) defined in Ref. [58]. The p3 = 0
term does not contribute to the sum, and using the explicit form of d00 (x0 , y0 ; p) for p = 0,
one can show that
L
a2 d00 x0 , y0 ; (0, 0, p3) = 2 (B.5)
x0 ,y0 p̂3

with p̂3 = a2 sin ap2 3 . Thus we see that h(1) (d/L) is just given in terms of the one-
dimensional scalar propagator on a periodic lattice with length L, which has no lattice
artifacts. This eventually leads to the result quoted in Eq. (2.14). The absence of any lattice
spacing dependence is a consequence of the special kinematics, namely the summation
over x1 , x2 , but will of course not be exact in higher orders of perturbation theory.
B.2. Anomalous dimensions
In order to precisely connect to the RGI current, it is important to obtain the anomalous
dimension of the static-light axial current in the SF scheme at two-loop order. The
anomalous dimension is expanded as

γ (ḡ) = −ḡ 2 γ0 + γ1SF ḡ 2 + O ḡ 4 , (B.6)
with γ0 = −1/(4π 2) and the two-loop anomalous dimension in the SF scheme, γ1SF .
With the perturbative expansions for the various correlation functions, the ratio ΞI of
Eq. (A.16) can be written as a series

ΞI (g0 , a/L) = ΞI(0) (a/L) + g02 ΞI(1) (a/L) + O g04 . (B.7)
(0)
stat
Accordingly, this allows us to expand the SF renormalization constant ZA = ΞI /ΞI
(see Eq. (A.19)) as
stat,(0) stat,(1)
ZAstat
= ZA + g02 ZA + O g04 . (B.8)
With the one-loop relation between the bare lattice current and the renormalized static
axial current in the MS scheme, the anomalous dimension in the MS scheme, Eq. (5.5) [38–
40], can be converted into the SF scheme. The renormalization constant relating the SF
scheme and the MS scheme is obtained from the relation between the SF scheme and the
bare lattice current, Eq. (B.8), the connection of the bare lattice current and a ‘matching
scheme’ [63,64] and the relation between the latter and the MS scheme [30]. Here the
matching scheme is defined by the requirement that the renormalized static-light axial
current at scale µ = mh equals the relativistic axial current with a heavy quark mass
mh up to terms of O(1/mh ), and the current in the relativistic theory is normalized by
current algebra (imposing the chiral Ward identities).6 Following the steps in Ref. [17],
this analysis finally yields
1
θ = 0.0: γ1SF = 2
0.22(2) − 0.0552(13)Nf , (B.9)
(4π)
1
θ = 0.5: γ1SF = 2
0.10(2) − 0.0477(13)Nf , (B.10)
(4π)
1
θ = 1.0: γ1SF = −0.08(2) − 0.0365(13)Nf . (B.11)
(4π)2
B.3. Discretization errors
The one-loop expansion at hand is also helpful to study discretization errors in the step
scaling function. Using Eq. (B.8), the step scaling function at lattice spacing a is expanded
as
stat,(1)
ΣAstat (u, a/L) = 1 + uΣA (a/L) + O u2 , (B.12)
and its continuum limit σAstat (u) as
stat,(1) stat,(2)
σAstat (u) = 1 + uσA + u2 σA + O u3 , (B.13)
6 Since the wording in Ref. [17] is not completely clear on this, we point out that Astat in that reference refers
MS
to what we call the MS scheme here as well as in Ref. [17].
Fig. 12. Discretization errors of the step scaling function at one-loop level. The continuum extrapolation of the
non-perturbative results uses (a/L)2 < 0.02.
with
stat,(1)
σA = ln(2)γ0,
stat,(2) 1 2
σA = ln (2)γ02 + ln2 (2)b0γ0 + ln(2)γ1 . (B.14)
2
As a measure for the discretization errors, we define
ΣAstat (u, a/L) − σAstat (u)

δ(u, a/L) = (B.15)
σAstat(u)
with a perturbative expansion

δ(u, a/L) = δ (1) (a/L)u + O u2 . (B.16)
The one-loop coefficient δ (1) versus the lattice spacing squared is shown in Fig. 12 for
different values of θ . In the range of lattice spacings where our non-perturbative calculation
is performed, the discretization errors at one-loop level are smaller than 1% × u, giving
rise to the hope that also the non-perturbative discretization errors are reasonably small.
A welcome feature of δ (1) at θ = 0.5 is that it is entirely dominated by the leading a 2 /L2 -
term in the a-expansion.
Appendix C. Continuum step scaling function and matching at L = 1.436 r0
In this last appendix we briefly discuss our parametrization (i.e., the interpolating fit)
of the continuum step scaling function and some details on the calculation of ZA stat at the
matching scale 2Lmax = 1.436 r0.
C.1. Fits and error determination in the scale evolution
As described in Appendix A.3, the continuum step scaling function σAstat (u) has been
obtained by extrapolating the lattice data on ΣAstat (u, a/L) at fixed u to the continuum limit.
The next step is now to solve the recursion specified through Eqs. (3.1)–(3.4). In practice
this is done by first representing the results for σAstat (u) in Table 1 by a fit and then solving
the recursion, Eq. (3.4), with σAstat (u) given by the fit function.
Guided by the analysis for the step scaling function of the pseudoscalar density σP in
Ref. [13] and the perturbative expansion discussed in Appendix B.3,
σAstat (u) = 1 + s0 u + s1 u2 + s2 u3 + · · · + sn un+1 (C.1)

is chosen as fit ansatz. The two non-trivial leading terms are restricted by perturbation
theory,
s0 = σAstat,(1) , s1 = σAstat,(2) , (C.2)

cf. Eqs. (B.14). Up to three additional free fit parameters were allowed for. All of these
fits represent the data in Table 1 well, and we decided to quote the two-parameter fit (the
curve of which is shown in Fig. 6) as the final result for the functional form of σAstat . To
check that the polynomial fits are stable, we also investigated fits where only s0 or even no
coefficient at all is constrained to its perturbative value. This leads to consistent results for
σAstat (u); particularly the latter fit then reproduces the perturbative prediction for s0 .
Having chosen a definite expression for σAstat (u), the solution of the associated recursion
is unique. Since the errors on the step scaling function stem from different simulation runs
and are hence uncorrelated, the errors on the fit parameters in the polynomial (C.1) for
σAstat (u) and those on the vk ’s calculated from it can be estimated straightforwardly by the
standard error propagation rules. Finally, by increasing the number of free fit parameters
(while fixing s0 , s1 to perturbation theory) as mentioned above, we convinced ourselves
that the systematic error induced by the choice of fit functions is well under control: in fact,
we then observed the expected pattern of finding slightly different errors but compatible
results at comparably good overall fit quality.
stat
C.2. Calculation of ZA at the low-energy matching scale
The total renormalization factor ZRGI introduced in Section 4 involves the value of the
stat stat
renormalization constant ZA at our particular matching point: ZA (g0 , L/a)|L=1.436 r0 .
As the latter connects a bare matrix element of the static-light axial current to the one
stat
renormalized in the SF scheme, this amounts to calculate ZA for a range of bare couplings
commonly used in simulations in physically large volumes.
Table 8
stat (g , L/a) at fixed scale L = 2L
Results for ZA 0 max = 1.436 r0 with ct being set to either one- or two-loop. The
critical κ-values κ = κc [65] are the same as used for Table C.1 of Ref. [13]
2-loop 1-loop
L/a β = 6/g02 ct ct
κ stat
ZA κ stat
ZA
8 6.0219 0.13508 0.6926(15) 0.13504 0.6932(10)
10 6.1628 0.13565 0.6810(17) 0.13564 0.6824(16)
12 6.2885 0.13575 0.6786(18) 0.13574 0.6795(16)
16 6.4956 0.13559 0.6777(17) 0.13558 0.6763(18)
Table 9
stat (g , L/a) at fixed scale L = 2L stat
Results for ZA 0 max = 1.436 r0 for cA = 0 and unimproved Wilson fermions
stat
(i.e., csw = 0 and thus, cA = cA = 0 too), where ct was kept at its two-loop value
L/a β = 6/g02 stat = 0
cA csw = 0
κ stat
ZA κ stat
ZA
4 5.6791 – – 0.15268 0.6923(13)
6 5.8636 – – 0.15451 0.6315(16)
8 6.0219 0.13508 0.6736(14) 0.15341 0.6075(13)
10 6.1628 0.13565 0.6633(17) 0.15202 0.5964(18)
12 6.2885 0.13575 0.6621(17) 0.15078 0.5971(32)
16 6.4956 0.13559 0.6627(17) 0.14887 0.5991(52)
stat
To extract ZA we exploit the fact that the required pairs (L/a, β) that match the
condition L/a = 1.436 r0/a have already been determined for the relevant β-range in
Appendix C of [13] by utilizing the known parametrization of ln(a/r0 ) in terms of β
from Ref. [34]. We thus could take over these pairs and computed ZA stat for θ = 0.5 from
the renormalization condition (A.19) at the corresponding values κ = κc of the critical

hopping parameter [65]. The results for ZA stat (g , L/a)|
0 L=1.436 r0 using the one- and two-
loop expressions for ct , cf. Eq. (A.21), are given in Table 8. The difference originating from
the two perturbative approximations for ct is completely covered by the statistical errors
2-loop
so that we again consider ct to already account for the dominant part of the boundary
cutoff effects in the gauge sector. Similarly to the discussion in Appendix A.3, the influence
of the boundary improvement coefficient c̃t in the fermionic sector can also be neglected
stat
at the level of our precision. The parametrization of the results for ZA (g0 , L/a)|L=1.436 r0
2-loop
by a polynomial fit in (β − 6), with ct stat
and cA from one-loop perturbation theory, is
quoted in Section 4, where the coefficients in the first block of Table 2 are to be combined
with Eq. (4.1). The smooth dependence of ZA stat on β in the studied region of bare couplings
suggests that this representation can also be slightly extended down to β = 6.0 (even
though we could not directly simulate that point for the same reason as in case of ZP [13]),
and we therefore regard it as a reliable representation of our data over the whole range
6.0 β 6.5.
As in Appendix A.3, we also set cA stat
= 0 instead of one-loop in the analysis of the
stat
data on ZA (g0 , L/a)|L=1.436 r0 , and the results are listed in the middle part of Table 9.
In contrast to the step scaling function, which did not change appreciably under this
stat stat
replacement of the value for cA , we observe an effect of about 3% in ZA at the low-
energy matching scale L = 2Lmax = 1.436 r0.
Furthermore, we addressed the case of unimproved Wilson fermions by also setting
csw = 0 in the relativistic fermion action and, after having computed the needed estimates
of the critical hopping parameter for this situation, carried out the additional runs to
determine the renormalization constant. In this case the pairs (L/a, β) were extended to
lower values of β in order to be able to make contact with the β-region that is typically
employed in simulations to calculate the bare matrix element defining the B-meson decay
stat
constant, as, e.g., those in Refs. [10,11]. The resulting estimates on ZA are shown
in the right part of Table 9, and the corresponding polynomial representations for both
aforementioned cases are as well found via Eq. (4.1) together with the two lower blocks of
Table 2.
We conclude this discussion with the general remark that the uncertainties of the
entering critical κ-values (of 1–2 and 2–4 on the last decimal place in the case of
csw = non-perturbative and csw = 0, respectively) do not affect the Z-factors significantly.
References
[1] N. Yamada, in: Lattice 2002, Proceedings of XXth International Symposium on Lattice Field Theory, Nucl.
Phys. B (Proc. Suppl.) 119 (2003), hep-lat/0210035.
[2] M. Guagnelli, et al., Phys. Lett. B 546 (2002) 237, hep-lat/0206023.
[3] M. Guagnelli, et al., in: Lattice 2002, Proceedings of XXth International Symposium on Lattice Field
Theory, Nucl. Phys. B (Proc. Suppl.) 119 (2003), hep-lat/0209113.
[4] E. Eichten, B. Hill, Phys. Lett. B 234 (1990) 511.
[5] C.R. Allton, et al., Nucl. Phys. B 349 (1991) 598.
[6] C. Alexandrou, et al., Phys. Lett. B 256 (1991) 60.
[7] APE Collaboration, C.R. Allton, et al., Phys. Lett. B 326 (1994) 295, hep-ph/9402343.
[8] C. Alexandrou, et al., Nucl. Phys. B 414 (1994) 815, hep-lat/9211042.
[9] C.W. Bernard, J.N. Labrenz, A. Soni, Phys. Rev. D 49 (1994) 2536, hep-lat/9306009.
[10] A. Duncan, et al., Phys. Rev. D 51 (1995) 5101, hep-lat/9407025.
[11] T. Draper, C. McNeile, Nucl. Phys. B (Proc. Suppl.) 34 (1994) 453, hep-lat/9401013.
[12] UKQCD Collaboration, A.K. Ewing, et al., Phys. Rev. D 54 (1996) 3526, hep-lat/9508030.
[13] ALPHA Collaboration, S. Capitani, et al., Nucl. Phys. B 544 (1999) 669, hep-lat/9810063.
[14] R. Sommer, in: Lattice 2002, Proceedings of XXth International Symposium on Lattice Field Theory, Nucl.
Phys. B (Proc. Suppl.) 119 (2003), hep-lat/0209162.
[15] J. Heitger, Applications of non-perturbative renormalization, Contribution to 30th International Conference
on High-Energy Physics (ICHEP 2000), Osaka, Japan, 27 July–2 August 2000, hep-ph/0010050.
[16] ALPHA Collaboration, J. Heitger, M. Kurth, R. Sommer, in: Lattice 2002, Proceedings of XXth
International Symposium on Lattice Field Theory, Nucl. Phys. B (Proc. Suppl.) 119 (2003), hep-lat/0209078.
[17] ALPHA Collaboration, M. Kurth, R. Sommer, Nucl. Phys. B 597 (2001) 488, hep-lat/0007002.
[18] M. Lüscher, et al., Nucl. Phys. B 384 (1992) 168, hep-lat/9207009.
[19] S. Sint, Nucl. Phys. B 421 (1994) 135, hep-lat/9312079.
[20] S. Sint, R. Sommer, Nucl. Phys. B 465 (1996) 71, hep-lat/9508012.
[21] ALPHA Collaboration, M. Lüscher, et al., Nucl. Phys. B 478 (1996) 365, hep-lat/9605038.
[22] M. Lüscher, P. Weisz, U. Wolff, Nucl. Phys. B 359 (1991) 221.
[25] R. Sommer, Non-perturbative renormalization of QCD, Lectures given at 36th Internationale Univer-
sitätswochen für Kernphysik und Teilchenphysik, Schladming, Austria, 1–8 March 1997, hep-ph/9711243.
[26] M. Lüscher, Advanced lattice QCD, Lectures given at Les Houches Summer School in Theoretical Physics,
Probing the Standard Model of Particle Interactions, Les Houches, France, 28 July–5 September 1997, hep-
ph/9802029.
[27] ALPHA Collaboration, S. Sint, P. Weisz, Nucl. Phys. B 545 (1999) 529, hep-lat/9808013.
[28] M.A. Shifman, M.B. Voloshin, Sov. J. Nucl. Phys. 45 (1987) 292.
[29] H.D. Politzer, M.B. Wise, Phys. Lett. B 206 (1988) 681.
[30] E. Eichten, B. Hill, Phys. Lett. B 240 (1990) 193.
[31] G. Parisi, R. Petronzio, F. Rapuano, Phys. Lett. B 128 (1983) 418.
[32] K. Symanzik, Nucl. Phys. B 226 (1983) 187.
[33] R. Sommer, Nucl. Phys. B 411 (1994) 839, hep-lat/9310022.
[34] ALPHA Collaboration, M. Guagnelli, R. Sommer, H. Wittig, Nucl. Phys. B 535 (1998) 389, hep-
lat/9806005.
[35] ALPHA Collaboration, M. Lüscher, et al., Nucl. Phys. B 491 (1997) 323, hep-lat/9609035.
[36] M. Neubert, Phys. Rep. 245 (1994) 259, hep-ph/9306320.
[37] M. Beneke, Phys. Rep. 317 (1999) 1, hep-ph/9807443.
[38] X. Ji, M.J. Musolf, Phys. Lett. B 257 (1991) 409.
[39] D.J. Broadhurst, A.G. Grozin, Phys. Lett. B 267 (1991) 105, hep-ph/9908362.
[40] V. Gimenez, Nucl. Phys. B 375 (1992) 582.
[41] T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379, hep-ph/9701390.
[42] K.G. Chetyrkin, Phys. Lett. B 404 (1997) 161, hep-ph/9703278.
[43] J.A.M. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 405 (1997) 327, hep-ph/9703284.
[44] ALPHA, UKQCD Collaborations, J. Garden, et al., Nucl. Phys. B 571 (2000) 237, hep-lat/9906013.
[45] ALPHA Collaboration, M. Guagnelli, et al., Nucl. Phys. B 595 (2001) 44, hep-lat/0009021.
[46] H. Leutwyler, Principles of chiral perturbation theory, Lectures given at Hadrons 94 Workshop, Gramado,
Brazil, 10–14 April 1994, hep-ph/9406283.
[47] ALPHA Collaboration, J. Heitger, R. Sommer, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 358, hep-
lat/0110016.
[48] G.P. Lepage, P.B. Mackenzie, Phys. Rev. D 48 (1993) 2250, hep-lat/9209022.
[49] S. Hashimoto, T. Ishikawa, T. Onogi, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 352.
[50] C. Alexandrou, et al., Z. Phys. C 62 (1994) 659, hep-lat/9312051.
[51] R. Sommer, Phys. Rep. 275 (1996) 1, hep-lat/9401037.
[52] H. Wittig, Heavy Quarks on the lattice: status and perspectives, Lectures given at International School of
Physics Enrico Fermi, Varenna, Italy, 8–18 July 1997, hep-lat/9710088.
[53] ALPHA Collaboration, J. Rolf, S. Sint, JHEP 12 (2002) 007, hep-ph/0209255.
[54] ALPHA Collaboration, A. Jüttner, J. Rolf, Phys. Lett. B 560 (2003) 59–63, hep-lat/0302016.
[55] C. Morningstar, J. Shigemitsu, Phys. Rev. D 57 (1998) 6741, hep-lat/9712016.
[56] K.I. Ishikawa, T. Onogi, N. Yamada, Nucl. Phys. B (Proc. Suppl.) 83–84 (2000) 301, hep-lat/9909159.
[57] ALPHA Collaboration, A. Bode, P. Weisz, U. Wolff, Nucl. Phys. B 576 (2000) 517, hep-lat/9911018.
[58] M. Lüscher, P. Weisz, Nucl. Phys. B 479 (1996) 429, hep-lat/9606016.
[59] B. Gehrmann, PhD thesis, 2002, hep-lat/0207016.
[60] ALPHA Collaboration, M. Guagnelli, J. Heitger, Comput. Phys. Commun. 130 (2000) 12, hep-lat/9910024.
[61] S. Fischer, et al., Comput. Phys. Commun. 98 (1996) 20, hep-lat/9602019.
[62] ALPHA Collaboration, S. Sint, P. Weisz, Nucl. Phys. B 502 (1997) 251, hep-lat/9704001.
[63] ALPHA Collaboration, M. Kurth, R. Sommer, Nucl. Phys. B 623 (2002) 271, hep-lat/0108018.
[64] A. Borrelli, C. Pittori, Nucl. Phys. B 385 (1992) 502.
[65] H. Wittig, private communication.
Twisted determinants on higher genus

Riemann surfaces
Rodolfo Russo a , Stefano Sciuto b
a Laboratoire de Physique Théorique de l’Ecole Normale Supérieure, 24 rue Lhomond,
F-75231 Paris cedex 05, France
b Dipartimento di Fisica Teorica, Università di Torino; INFN, Sezione di Torino, Via P. Giuria 1,
I-10125 Torino, Italy
Abstract
We study the Dirac and the Laplacian operators on orientable Riemann surfaces of arbitrary genus
g. In particular we compute their determinants with twisted boundary conditions along the b-cycles.
All the ingredients of the final results (including the normalizations) are explicitly written in terms
of the Schottky parametrization of the Riemann surface. By using the bosonization equivalence, we
derive a multi-loop generalization of the well-known g = 1 product formulae for the Theta-functions.
We finally comment on the applications of these results to the perturbative theory of open charged
strings.
 2003 Published by Elsevier B.V.
PACS: 11.25.Db; 11.25.Hf
1. Introduction
The perturbative expansion of string theory has been thoroughly studied in the last thirty
years and the first works date back to the time of the dual models [1]. Our understanding of
this subject has greatly increased during the years, in particular in the eighties, with many
conceptual and technical breakthroughs (see for instance [2]). The geometrical meaning
of string amplitudes became much more clear and the strict relation of these physical
quantities with the theory of Riemann surfaces and Theta-functions made the subject even
more interesting.
E-mail addresses: rodolfo.russo@lpt.ens.fr (R. Russo), sciuto@to.infn.it (S. Sciuto).
0550-3213/$ – see front matter  2003 Published by Elsevier B.V.

doi:10.1016/j.nuclphysb.2003.07.016
208 R. Russo, S. Sciuto / Nuclear Physics B 669 (2003) 207–232
Among the basic building blocks of string amplitudes one finds the determinants of the
Laplacian and of the Dirac operator. Since these operators are present in the 2D action
describing the free string propagation, their determinants will appear in all perturbative
amplitudes, starting from the simplest one: the vacuum energy. A detailed study of these
determinants was performed in the papers [3–8].
It is particularly interesting to compare the fermionic and the bosonic results, because,
even though they have a very different structure, they are related through the bosonization
duality. Thus the equivalence between the fermionic and the bosonic determinants can be
used as a tool for giving “physicist proofs” of interesting mathematical identities [9–11]. In
fact, it is known that from general addition theorems for Theta-functions (see, for example,
Proposition 2.16 of [9]) one can derive some identities proving the equivalence between
bosonic and fermionic systems on a g-loop Riemann surface. For instance, the identity in
Corollary 2.17 of [9] is a consequence (or a proof, according to the points of view) of the
relation between a fermionic system1 (b, c) of spin (1, 0) and a chiral boson of background
charge Q = −1:
g g g
i<j =1 E(zi , zj )
i=1 σ (zi )
θ J (zi ) − J (w) − ∆ τ g = C det ωi (zj ) .
i=1 E(zi , w)
σ (w)
i=1
(1.1)
In the mathematical literature, the analysis of identities like the one above focuses on the
dependence of the various functions on the punctures zi . Thus often these equalities are
written in terms of some “constants” (like C in the above equation) that depend only on
the moduli of the surface, but not the punctures zi . However, it is also quite interesting
to give an explicit expressions of these constants, since they are closely related to the
partition functions of the bosonic and fermionic systems. A powerful technique for writing
explicitly C in terms of the moduli of the surface is the sewing technique. This is a very
old idea [1] allowing to construct higher loop amplitudes from tree diagrams: pairs of
external legs are sewn together taking the trace over all possible states with the insertion
of a propagator that geometrically identifies the neighborhoods around two punctures. The
results obtained in this way give rise to a particular parametrization of the g-loop Riemann
surface known as Schottky uniformization. At 1-loop, string theorists are very familiar with
this parametrization: in this case, the torus is just seen as the complex plane where two
points z, w are identified if z = k n w, ∀n ∈ Z. Here k is a complex parameter representing
the modulus of the torus and is related to period matrix entering in the Theta-functions
by k = exp(2πiτ ),n with Im τ > 0 ⇔ |k| < 1. At 1-loop the constant C in Eq. (1.1) is just
C= ∞ n=1 (1 − k ) 3 and is related to the partition function of a chiral scalar field. This
pattern can be generalized tohigher genus surfaces by sewing other handles to the 1-loop
∞
result and one gets C = α n=1 (1 − kαn )3 , where the exact meaning of kα and of product
over the Schottky group α will be explained later (see Eq. (2.11) and Appendix B).
Here we just want to stress that identities like Eq. (1.1) can be exploited also to rewrite
products over the Schottky group in terms of more geometrical quantities like Theta-
function, Abelian differentials, and prime form (see Appendix A).
1 See Appendix A for the definition of our conventions.

R. Russo, S. Sciuto / Nuclear Physics B 669 (2003) 207–232 209
At 1-loop, the possibility to pass from the formulation in terms of Schottky products to
the one with Theta-functions has been heavily exploited in string theory. The physical rea-
son is clear: the geometrical expressions in terms of Theta-functions has the advantage to
make manifest the modular properties of the string results. This is in contrast to the expres-
sions written in terms of Schottky products, where these properties are not manifest. In fact,
a modular transformation can be non-analytic in k (for instance, consider τ → −1/τ ). This
limitation is precisely the reason why the Schottky uniformization has not been much stud-
ied by mathematicians. However, the Schottky uniformization has an important physical
significance because it makes manifest the unitarity properties of string amplitudes. This
can be easily understood by remembering that this uniformization naturally arises from
the sewing procedure where the Schottky multipliers k are basically the equivalent of the
(exponential of the) Schwinger parameters in field theory. Thus the Taylor expansion in the
k’s of the string results isolates the contribution to the amplitude coming from the propa-
gation of particular states in the various handles of the surface. At 1-loop level, the relation
between k and τ is particularly simple and thus one can use also the Theta-function expres-
sions to analyze the unitarity properties of the string results. However, on a g-loop surface
there is no equivalent of the simple relation k = exp(2πiτ ) and only the expressions in
terms of Schottky parameters display the unitarity properties in a simple way.
Thus the expressions of string results à la Schottky and the one in terms of Theta-
functions capture two different but equally important features of string theory. Depending
on the question one would like to answer it is more convenient to write the amplitudes in
one or the other form. Because of this, it is quite crucial to be able to rewrite a general
Schottky product in terms of geometrical objects and vice versa. As we already said, this
step is by now standard in 1-loop computations, where the period matrix is just a complex
number which is related in a very simple way to the only Schottky multiplier k. In this
case, string theorists often make use of identities like the one in Eq. (3.3). At mathematical
level, these 1-loop identity can be proved by showing that both sides of the equation have
the same periodicity, poles and zeros. The generalization of this kind of identities to the
higher loop case is much harder. However, as we noticed before, it is precisely in the multi-
loop expressions where one really needs to have these identities to pass from a manifest
unitarity amplitude to a modular covariant one.
Thus the main purpose of this paper is to find new relations of the kind of Eq. (1.1)
that relate Theta-functions to the Schottky products contained the constant C. Following
the ideas of [14–16], we use the equivalence between bosonic and fermionic systems as
a device allowing to recast Schottky products in terms of Theta-functions. We basically
generalize previous analysis [14–16] in two directions. On one hand we consider fermions
of spin (λ, 1 − λ) and the dual bosonic system with the background charge Q = 1 − 2λ,
instead of focusing just on the simplest case λ = 1/2. On the other hand, for general
λ we also consider twisted periodicity conditions along the b-cycles. This twisting can
be equivalently thought as the effect of a flat gauge connection along the b-cycles2
on a minimally coupled fermionic system. Thus the periodicity parameters (µ ) can be
2 As usual, we call b-cycles the loops in the worldsheet along the τ direction and a-cycles the spatial loops
along σ .
naturally thought as non-geometrical parameters of the Riemann surface. In fact, in the

generalization of Eq. (1.1) arising in these cases, the constants C depend also on µ
beyond the usual dependence on the geometrical moduli. Our goal is to give an explicit
expressions in terms of Schottky series or products for all the quantities appearing in the
various identities. In this way, at least in principle, one can compute all the ingredients in
terms of the parameters of the surfaces with an arbitrary degree of precision.
One more clarification is due at this point, since it may appear unclear why in our study
b-cycles twists have a privileged status in comparison with a-cycle twists. The difference
is simply due to the Hamiltonian approach adopted here, where the twists along the time
direction can be taken into account simply by a modification of the sewing propagator,
while those along the spatial directions modifies radically the spectrum of the free theory.
In the higher genus diagrams, this difference implies that b-cycles twists can be described
à la Schottky by means of the usual representations of the projective group, while the
direct description of amplitudes with a-cycle twists seems to require a more complicated
formalism (for an explicit 1-loop example see [17] and compare Eqs. (7) and (8) therein).
Of course, once a result with b-cycles twists is known in terms of Theta-function, one
can explicitly perform a modular transformation and derive the equivalent quantity with
a-cycles twists. This strategy can be used to obtain the partition function of the charged
open bosonic string in presence of a constant external field.
The structure of the paper is the following. In Section 2 we recall the main features
of the sewing technique, focusing on the fermionic correlators. The presence of general
twisted periodicity along the b-cycles requires a modification of the sewing procedure
and it is crucial for the consistency of the results to carefully follow all the effects due
to the presence of µ . The main result of this section is the explicit expression (2.18) of
the twisted Abelian differentials. In Section 3 we compare the fermionic correlators with
the corresponding bosonic ones and derive Eqs. (3.5), (3.6) which generalize (1.1) to the
twisted case. Finally in the conclusions we discuss some applications of our results in the
context of string theory. Then in Appendix A we define some quantities of interest in the
theory of Riemann surfaces, in Appendix B we discuss the Schottky parametrization and
in Appendix C we give some details of the sewing technique.
2. The sewing technique
The use of the sewing technique for computing multi-loop amplitudes is discussed in
detail in [12] for bosonic systems and in [13] for fermionic systems. In order to generalize
their results to the twisted case, we will follow closely these two papers and refer to them
for the explicit derivation of the results used here as starting point. In this section, we just
present the main ideas of the sewing technique in order to clarify its application to the
physical/mathematical problem discussed in this paper.
2.1. The torus
As anticipated in the introduction, the main idea of the sewing technique is to employ
a bootstrap approach and construct g-loop amplitudes starting from the tree-level results.
The first step is of course the construction of 1-loop amplitudes. This derivation is quite
natural and well-known since it is explained in Chapter VIII of [18]. In fact, tree-level
amplitudes can be written with a special choice for the puncture of the external states
putting the first vertex in z = ∞ and the last in z = 0. Thus the first state just describes an
outgoing string with
s1 | and the last state describes an ingoing string with the ket |sN .
At this point one can simply relax the on-shell condition, insert the propagator k L0 , and
identify the two states

0|V1 (∞)V2 (1) · · · VN (0)|0 =
s1 |V2 (1) · · · VN−1 |sN → Tr V2 (1) · · · VN−1 k L0 .
(2.1)
This transforms the VEV of the vertex operators, typical of tree-level amplitudes, into a
trace over the Hilbert space of a propagating string; moreover the neighborhood around
the points z = ∞ and z = 0 are identified by the projective transformation (z → k n z)
generated by the insertion of the propagator. As an example, it is useful to start with the
analysis of the vacuum 1-loop amplitude, since it contains some of the features of the
general computation and can be derived from the simple trace in (2.1) without external
states. In fact, it is not difficult to compute the torus partition function of a fermionic
system of spin (λ, 1 − λ) with trivial periodicity conditions on the a-cycle and twisted
ones along the b-cycle. In fact, in the sewing construction the b-cycles are generated by
the identification of the neighborhoods around two points (usually z = ∞ and z = 0 at 1-
loop) enforced by the propagator P (one can take simply P = k L0 ). Since we want to have
non-trivial periodicity condition along these cycles, we need to deform the propagator by
adding an dependence so that

b(z) = k λ e−2πi P−1 b(k z)P , c(z) = k 1−λ e2πi P−1 c(k z)P . (2.2)
This can be obtained simply by inserting together with the usual propagator P also a factor
of e2πij0 , where j0 is the fermionic number operator. The torus partition function can be
computed as usual just by taking the trace of twisted P = P e2πij0
∞

λ−1
Zλ = Tr k L0 e2πij0 λ = 1 − e2πi k n 1 − e−2πi k n 1 − e2πi k r .
n=λ r=1−λ
(2.3)
The structure of this result is quite clear: the first two factors come from the action of the
modes c−n and b−n with n λ, respectively, while the last product is related to the special
modes cr with r ∈ [(1 − λ), (λ − 1)] only, since the corresponding br oscillators vanish on
the vacuum. If λ is integer the result (2.3) can be rewritten as
∞

Zλ∈N = (−1)λ eiπ(2λ−1) k λ(1−λ)/22i sin π 1 − e2πi k n 1 − e−2πi k n , (2.4)
n=1
which is clearly vanishing for = 0. This fact has an important geometrical explanation.
For integer λ, there are periodic and regular solutions of the equation of motions for both
b (b ⇔ z−λ ) and c (c ⇔ zλ−1 ). In fact, the singularities in z = 0 an z = ∞ are the fixed
points (B.3) of the projective transformation P = k L0 and so are outside the fundamental
region representing the torus generated by the propagator P (see Appendix B). In other
words, both b and c have a zero-mode on the torus and thus it is natural that the partition
function without any insertion vanishes. For generic fixed points ξ and η these zero-modes
can be written as
λ 1−λ
η−ξ η−ξ
b(z) ⇔ , c(z) ⇔ . (2.5)
(z − η)(z − ξ ) (z − η)(z − ξ )
The presence of a twist along the b-cycles lifts both zero-modes, since the solutions (2.5)
do not satisfy the new boundary conditions. This explains why the partition function does
not vanish for = 0. Notice that the difference between the number of the b zero-modes
and the number of the c zero-modes is always zero for the torus topology, as it should
be. In fact the Riemann–Roch theorem ensures that this difference can depend only on the
integrals of the curvature and of the gauge field strength. As said before, the twists are
equivalent to the presence of a flat connection along the b-cycles, and thus the difference
between the number of the b and the c zero-modes cannot depend on . To be precise, for
the cases we are interested in, the Riemann–Roch theorem says
#c zero-modes − #b zero-modes = Q(g − 1), (2.6)

where, as usual, the background charge Q is related to the spin λ of the fermionic system
(b, c): Q = 1 − 2λ.
2.2. Higher genus surfaces
Turning to the study of higher genus surfaces we observe that this procedure, however,
cannot be straightforwardly generalized, because the vertex operators V in (2.1) describe
just the emission of a specific on-shell state. They depend only on the quantum numbers of
the emitted string and thus there is no easy way to identify two of them and sum over all
intermediate states. In order to this, one should use a generalization [19,20] of the vertex
operators, where also the emitted states are described by a whole Hilbert space

X
V = I
0, x0 = 0|:exp
I dz −Xv (1 − z)∂z XI (z) :,
0

v
VIbc = I
0; q = −Q|:exp dz b (1 − z)cI (z) − c (1 − z)bI (z) :.
v
(2.7)
0
Here the coordinates with the superscript v describe a propagating (virtual) string, while
those with a subscript I describe a generic emitted state in the Hilbert space HI . The
modes of the two types of fields (anti)commute among them, since they refer to completely
independent states. Let us stress that the bra-vector in Eq. (2.7) is the vacuum in the
Hilbert space of the emitted string, so that
V| is an operator in the v-Hilbert space of
the propagating string. One can think of the vertices
V| as an “off-shell” generalization of
the usual bosonic and ghost part (for Q = −3) of string vertex operators. However, here
off-shell does not have the usual meaning as in field theory. On the contrary, off-shell just
means that the external states have not been specified yet, and thus
V| can be seen as the
generator of all possible three string interaction which are obtained by saturating it with
physical (on-shell) states.
So far, in the presentation, we always made reference to the space–time interpretation
of the formalism typical of string theory. However, the same formalism can be also applied
in the context of 2-dimensional free conformal field theory. One has just to consider the
two vertices in Eq. (2.7) separately and use them to construct partition functions or Green
functions for bosonic or fermionic systems. By using the vertex
V bc | as ingredient, we can
substitute Eq. (2.1) with a new formula that is suited for the generalization of the sewing
procedure to higher genus surfaces. Following [13] it is easy to construct the generator of
the N -point Green function on the sphere (g = 0)
N
bc −1
bc
VN;0 = v
q = 0| γI VI γI |q = 0v , (2.8)
I =1
where the N Hilbert spaces labeled by I describe the external fields and are all
independent, while γI are (arbitrary) projective transformations mapping the interaction
point from z = 1 as in (2.7) to the arbitrary point zI . Remember that the vertex in (2.8)
contains the tensor product of the N vacua I
0; q = −Q|, but for notational simplicity we
write it simply as
VN;0 bc
|. Its explicit form is written in Eq. (2.26) of [13] in terms of the
λ
representation E with weight λ of the projective group. As we have seen in the previous 1-
loop example, the partition function of a fermionic system is in general vanishing because
of the presence of zero-modes. For higher genus surfaces, the Riemann–Roch (2.6) theorem
shows that this happens also for non-trivial periodicity conditions; so the simplest non-
trivial amplitude is the correlation function
N

b

Nb
(I )
Z (z1 , . . . , zN ) =
λ
b b(zI ) = V bc Nb ;gb |q = 0I ,
−λ (2.9)
I =1 (,λ) I =1
where
VNbcb ;g | is the generalization of the vertex (2.8) at genus g. Thus Zλ is the Green
function on a surface of genus g with non-trivial boundary conditions along the b-cycles
and with Nb = |Q|(g − 1) insertions of the b field. In order to construct
VNbcb ;g | within the
sewing approach, we use the generating vertex
VN;0 bc | with N = 2g + N and then identify
b
g pairs of Hilbert spaces (labeled by the index µ = 1, . . . , 2g) by means of the twisted
propagator that was used also in the torus computation (2.2). Moreover we are interested
in the “minimal” correlation function (2.9) that contains only b-fields as external states
(labeled by the index I = 1, . . . , Nb ). So we can set to zero all the b(I ) oscillators in the
computation of
VNbcb ;g | and keep only the c(I ) fields that will be saturated by the Nb b-
insertions of (2.9). Here we report just the final result for the generating vertex
VNbcb ;g |,
while the details of the construction are postponed to Appendix C
N
bc
b
V = det(1 − H ) I
q = −Q| FN ,g . (2.10)
Nb ;g b
I =1
The formal structure of this equation is quite natural. The non-zero mode contribution
is given by a (fermionic) Gaussian integral and this is why the determinant det(1 − H )
appears. As in the usual case µ = 0, also for our computation this determinant can be
written in terms of the multipliers of the Schottky group
∞

det(1 − H ) = 1 − e−2πi·Nα kαn 1 − e2πi·Nα kαn . (2.11)
α n=λ
Nα is a vector with g integer entries; the µth entry counts how many times the Schottky
generators Sµ enters in the element of the Schottky group Tα , whose multiplier is kα ; the
appearance of each Sµ is counted with the exponent sign, so that Sµ contributes 1, while
(Sµ )−1 contributes −1 to the global value of Nα . The product α is over the primary
µ
classes of the Schottky group excluding the identity and counting Tα and its inverse only
once. Eq. (2.11) is the generalization of the first product of (2.3) for an orientable Riemann
surface of genus g.
The second term of (2.10) takes into account the zero-mode contribution. In this case it
is not possible to write a simple expression that is valid for all arbitrary λ. The complication
(2µ) (I )
arises because the zero-mode bs are entangled with the external oscillators cλ . Thus it
is not possible to derive the zero-mode contribution to the g-traces before having computed
the scalar product over the Nb external Hilbert spaces present in (2.9). This technical
problem appears clearly from the formulae of Appendix C. Thus we now consider directly
the correlation functions (2.9). By inserting Eq. (2.10) into (2.9), one gets
Zλ (z1 , . . . , zNb ) ≡ det(1 − H )F (λ), (2.12)
where

Nb
Nb
(I )
F (λ)
= I
q = −Q| FNb ,g b−1 |q = 0I . (2.13)
I =1 I =1
The case |Q| = 1

Let us first focus on the case Q = −1, where the conformal weight of b(z) is λ = 1 and
the one of c(z) is zero; we consider generic values for the twist µ . As explained previously,
the non-zero mode contribution is given by (2.11) with λ = 1. On the other hand, in
Appendix C, we show that the contribution of zero-modes contained in F (see (2.13)),
which depends on g − 1 variables zI , is naturally written as a determinant
 
ζ1 (z1 ) ... ζg (z1 )
 .. .. 
 . . 
F (1) = det  , (2.14)
 ζ1 (zg−1 ) ... ζg (zg−1 ) 
e2πi1 − 1 . . . e2πig − 1
with

2πi(·Nα +µ ) 1 1
ζµ (zI ) = e − , (2.15)
α
zI − Tα Sµ (0) zI − Tα (0)
where the sum runs over the all the elements of the Schottky group. Thus from Eq. (2.10)
one obtains
∞

Zλ=1 (z1 , . . . , zg−1 ) = F (1) 1 − e−2πi·Nα kαn 1 − e2πi·Nα kαn . (2.16)
α n=1
The functional form of this result is exactly the expected one: in fact one can interpret the
second term with the Schottky product as the non-zero mode contribution to the determi-
nant of the Dirac operator and the first determinant as the zero-mode contribution to the cor-
relation function Zλ=1 , written in terms of twisted Abelian differentials. However, even if
this interpretation will be eventually the correct one, it cannot be applied to Eq. (2.16) as it
stands. First Eq. (2.15) defines g functions ζ , while, according to the Riemann–Roch (2.6)
theorem we have only g − 1 Abelian differentials. Moreover, it is annoying that the origin
of the complex plane plays a privileged rôle in (2.15). In particular this means that the ζ ’s
just introduced are not regular: in fact, when Tα is the identity or Sµ−1 , they have a pole
for zI = 0, which in general is part of the fundamental domain representing the Riemann
surface. It turns out that one can solve these problems simultaneously.
If all the twists µ are trivial, then the determinant is zero because of the last line. This
is of course due to the extra zero-mode (c(z) ⇔ const) appearing in the untwisted case
which makes the correlator (2.12) vanish. So let us suppose that at least one is non-
trivial, for instance, g = 0. With this hypothesis it is easy to simplify (2.14) by making a
linear combination of each column with the last one so to set to zero the first g − 1 entries
of the last row
 
Ω1 (z1 ) . . . Ωg−1(z1 ) ζg (z1 )
 .. .. .. 
 . . . 
F (1) = det  , (2.17)
 Ω1 (zg−1 ) . . . Ωg−1 (zg−1 ) ζg (zg−1 ) 
0 ... 0 e2πig − 1
where

e2πiµ − 1
Ωµ (zI ) = ζµ (zI ) − 2πi ζg (zI ) , µ = 1, . . . , g − 1. (2.18)
e g −1
This shows that only the g − 1 functions Ω enter in the final result (2.16) so that it is natural
to identify these Ω’s with the twisted Abelian differentials. Let us shows that they have all
the expected properties. First, the dependence on the origin disappeared and, contrary to
the original ζ ’s, the Ω’s are everywhere regular. To see this, it is useful to rewrite the
sum (2.15) in a different form,3 separating the contributions coming from the Schottky
elements of the form Tα Sµl
(µ)
1 1
ζµ (zI ) = e2πi(·Nα +µ ) −
α
zI − Tα (ηµ ) zI − Tα (ξµ )

2πi·Nα 1 1
+ 1 − e2πiµ e − , (2.19)
α
zI − Tα (0) zI − Tα (aµα )
3 See Appendix C for the derivation.

where, in the second line, aµα = ηµ if the Tα is of the form Tα = Tβ Sµl with l 1, while
aµα = ξµ otherwise. It is also important to remember that the sum in the first line does not
contain all the Schottky elements whose rightmost generator is Sµ±1 , while the second sum
is over all the elements Tα . It is now easy to see that first term in the second square bracket
of (2.19) cancels in the combination (2.18). Then one can check the periodicity properties.
It is clear that all the expressions we have written so far are periodic along the a-cycles
(z − η) → e2πi (z − η), since they are holomorphic in z. The periodicity along the b-cycles
can be deduced by looking at the Eq. (2.15). By means of the identity (B.8) one can see
that

ζµ Sν (z) dSν (z) = e2πiν ζµ (z) dz. (2.20)
Since the Ω’s are simply linear combinations of the ζ ’s, they have the same periodicity
property, which is exactly what one expected from Eq. (2.2). The fermionic result (2.16) is
multilinear in the Abelian differentials Ω’s and thus has a non-trivial periodicity along the
b-cycles (2.20)

Zλ=1 z1 , . . . , Sν (zk ), . . . , zg−1 Sν (zk ) = e2πiν Zλ=1 (z1 , . . . , zg−1 ). (2.21)
Notice that, if some µ = 0 (of course with µ = g), the corresponding Ωµ reduces
to ζµ , where the second line of (2.19) vanishes. This is the naïve generalization of the
usual untwisted Abelian differentials ωµ (zI ) (given in (B.6)) with an additional phase
exp(2πi · Nα ), which ensures the twisting of the periodicity along the cycles bν with
ν = µ. For the general case µ = 0, the twisted periodicity along bµ requires also the
second line of Eq. (2.19).
The case |Q| 2

The generalization of the result just presented to the case λ = 3/2, 2, . . . is discussed
in Appendix C. We recall here the main qualitative features. F (λ) (C.17) has the same
structure of F (1) , but the elements of this new matrix are themselves (2λ − 1) × (2λ − 1)
blocks. This is simply because the zero-modes now are related to the oscillators br , with
r ∈ [(1 − λ), (λ − 1)]. The presence of many zero-modes implies that, in the “minimal”
correlation function (2.9) there are (2λ − 1)(g − 1) variables zI and that the building blocks
of the matrix F (λ) have extra indices r, s ∈ [(1 − λ), (λ − 1)] and become ζµ,s (zν,r ). Also
the entries of the last row in F (λ) are now replaced by (2λ − 1) × (2λ − 1) matrices

E(Sµ )rs = e2πiµ E(Sµ ) − 1 rs . (2.22)
The ζµ,s (zν,r )’s suffer the same diseases discussed in the λ = 1 case. These problems are
cured in a similar way by defining (2λ − 1)(g − 1) λ-differentials Ωµ,s (zν,r ), analogous
to the ones introduced in (2.18) (see Eq. (C.22)). An important difference with the respect
to the λ = 1 case is that, for λ > 1 and g > 1, the determinant F (λ) does not vanish even
if µ = 0, ∀µ. Therefore the correlation function (2.12) is non-trivial also when all the µ
are set to zero. Notice that, if λ is half-integer, then the expression (C.22) can be used also
in absence of twists ( = 0). In fact, in this case the matrix E(Sµ )rs is always invertible.
For integer λ, on the contrary, the determinant of E(Sµ )rs is vanishing for = 0 and the
form of the λ-differentials is more complicated, as discussed in [13]. Of course one can
still express all the correlation function in terms of the ζµ,s (zν,r ) introduced in (C.18).
3. Comparison with bosonic determinants
We are now in the position to generalize Eq. (1.1). As anticipated in the introduction,
this kind of relations can be viewed as a consequence of the equivalence between fermionic
and bosonic systems. In fact, using again the sewing technique, it is possible to derive the
bosonic correlation functions corresponding to Zλ of Eq. (2.9). In the notation of [12] these
correlators are
N

b
φ
Nb

Z (z1 , . . . , zN ) =
Q
:e −φ(zI )
: = V |q = −1I , (3.1)
b b Nb ;g
I =1 (b ,Q) I =1
where the bosonic system has background charge Q = 1 − 2λ. Moreover the twisted
boundary conditions along the b-cycles are enforced on the bosonic side by the insertion
together with the propagator of the operator e2πib p0 [12]. Notice, however, that b =
− 1/2, where the additional factor of 1/2 is necessary in order to reproduce in the bosonic
language the usual (−1)F -twist of the fermionic traces (see, for instance, Appendix A
of [21], Eq. (A.2.23)).
In order to provide a simple example, let us write the 1-loop partition function of a
bosonic system with Q = 1 − 2λ with twist b (see, for instance, Eq. (3.21) of [12] with
α = 0, β = b and N1 = N2 = 0)
∞

n −1 1
Zb = 1−k θ b − Q +∆ τ . (3.2)
2
n=1
This formula has to be equal to Eq. (2.3) for b = − 1/2 and from this relation one can
derive the usual product formula for the 1-loop Theta-function. In particular, in terms of
the odd θ1 , one has
∞

θ1 (|τ ) = 2k 1/8
sin π 1 − k n 1 − e2πi k n 1 − e−2πi k n . (3.3)
n=1
The higher genus case works in a similar way. The main difference is that the partition
function without insertions now vanishes. On the fermionic side we know that this is due
to the presence of zero-modes, while on the bosonic side this is a simple consequence of
the momentum conservation modified by the background charge. Thus, in order to obtain
a non-trivial result, we have to compare the correlation functions (2.12) and (3.1). For
Q = −1 the fermionic result has been obtained in Section 4 and the bosonic one4 is (again
from Eq. (3.21) of [12])
g−1

g−1
I =1 σ (zI ) I <J E(zI , zJ )
Q=−1
Zb (z1 , . . . , zg−1 ) = ∞ θ + ∆ − J (zI τ ,
)
(1 − k n)
α

α n=1 I =1
(3.4)
4 In Appendix A we recall the definition and the main properti es of the various geometrical objects present
in (3.4).
where we have used again b µ = µ − 1/2 ∀µ = 1, . . . , g. Notice that, thanks to (A.2)

and (A.6), the argument of the Theta-function on the r.h.s. does not depend on the
basis point z0 of the Jacobi map. The explicit expressions in terms of the Schottky
parametrization of σ (zI ) (including its normalization), E(zI , zJ ) and ∆ are given in
Appendix A of [12].
The equivalence with the fermionic result implies5 (2.16) = (3.4)

g−1

g−1

C F =
(1) (1)
σ (zI ) E(zI , zJ )θ + ∆ − J (zI ) τ , (3.5)

I =1 I <J I =1
which generalizes (1.1) to the twisted case, with

∞

C(1) = 1 − kαn 1 − e−2πi·Nα kαn 1 − e2πi·Nα kαn . (3.6)
α n=1
A first check of this identity comes from the study of the periodicity properties of the
results. The periodicity (2.21) is reproduced in the bosonic side in a complicated way
by combining the various factors coming from transformation of the objects on the r.h.s.
of (3.4). However, most of these contributions cancel as in the untwisted case and the
only -dependent contribution can come from the Theta-function. From Eqs. (A.4), (A.8)
and (A.9), one gets

ZQ=−1
b
z1 , . . . , Sν (zk ), . . . , zg−1 Sν (zk ) = e2πiν ZQ=−1
b
(z1 , . . . , zg−1 ), (3.7)
in agreement with (2.21). Actually also a stronger check of (3.5) is possible. Since we know
the explicit form for the twisted Abelian differentials (2.18), we can use it together with
the expressions contained in Appendix A of [12] in order to express both sides of (3.5) in
terms of the Schottky group. At this point one can develop the results for small kµ and keep
only a finite number of terms in all the series or infinite products so to obtain in both sides
a polynomial in the kµ ’s. The coefficients of each term are (complicated) functions of the
fixed points and of the twisting parameters ’s. A highly non-trivial check is to verify that
the coefficients obtained on the l.h.s. of (3.5) agree with those obtained from the expansion
of the r.h.s. We performed this check for the case g = 2 and verified that the first four terms
in the expansion of (3.5) actually agree.
The generalization of (3.5) to the case |Q| 2 is straightforward and reads

Nb
E(z I , zJ ) 1 1
C(λ) F (λ) = I <J θ − − Q ∆ + − J (zI ) τ , (3.8)
Nb
σ (zI )Q 2 2
I =1 I =1
with
∞

C(λ) = 1 − kαn 1 − e−2πi·Nα kαn 1 − e2πi·Nα kαn . (3.9)
α n=λ
5 From now on, we endow the λ-differentials derived in the sewing approach with the appropriate factors of
dzλ .
4. Conclusions
In this paper we exploited the sewing technique to study free fermions of arbitrary
spin (and free boson with arbitrary background charge) with twisted boundary conditions
on the b-cycles. We computed various quantities of interest like determinants and twisted
differentials and gave explicit expressions for them in terms of the Schottky uniformization
of the Riemann surface. By comparing the bosonic and the fermionic results for certain
“minimal” correlation functions we also found new identities between products of
multipliers over the Schottky group and the usual multi-loop Theta-functions. Identities
of this kind are important because they make manifest the properties under modular
transformations which are hidden when the results are written in Schottky parametrization.
This is very useful, for instance, in string theory, where the modular properties of
perturbative amplitudes are crucial for the consistency of the theory. On the other hand,
important building blocks of string amplitudes, like the partition functions, are often given
in terms of Schottky product like (3.6) and thus their modular properties are blurred. It
is clearly important to rewrite these products in terms of Theta-functions. Let us mention
here an example of this problem in the context of open bosonic strings [22].
It is well known that a system of g + 1 D-branes, connected among them with tubes
representing the closed string propagation, is equivalently described by a disk with g
holes. The two descriptions make manifest different unitarity properties of the result: in
the D-brane picture the poles of the amplitude in the Schottky multipliers are related to
the propagation of some almost on-shell closed string states between D-branes, while in
the disk description the poles are related to open string states. The difference is made
more evident if one switches on a constant gauge field strength F on the D-branes (or,
in the T -dual language, if some of the D-branes in this multi-body system have a non-
vanishing constant relative velocity). In the D-brane language the result can be derived by
introducing in the construction of [23] a twist similar to the one discussed in this paper.
In this case is related to the differences between the external fields F . In the D-brane
description these twists enter only as a phase independent of the moduli, exactly as in (3.6).
By using the results of this paper one can explictly perform the modular transformation
of this expression and obtain the same quantity in the open string description, where
the surface looks like a disk with g holes. Now the phases induced by the twists do
depend on the moduli and this is strictly related to the modification induced by F on
the mass of the open strings. This very important difference is already present at 1-loop
level (i.e., in a system of two D-branes), as it is clear by the comparison between the
results of [24,25]. We have checked [22] that, as already suggested in [24], from the
open string formulation of the charged partition function one can derive the 1-loop Euler–
Heisenberg effective action for Yang–Mills theories simply by performing the field theory
limit α → 0 of the results, in the spirit of [26] (see also references therein). Of course one
obtains the pure Yang–Mills effective action if the bosonic D3-branes are used as starting
point, or the N = 4 result if one starts from the usual type IIB D3-branes. This technique
can be extended to more complicated systems of N = 2 theories with various content
of matter [27], once the corresponding string results are derived. On the contrary, the
description of [25] and its bosonic multi-loop version [23] are not directly connected to the
Yang–Mills effective action. In this case the low-energy limit captures the (super)gravity
interaction among the D-branes seen as classical gravitational solitons. Notable exceptions
to this are known, where perturbative gauge theory results are directly connected to
supergravity computations. This striking connection is usually due to the cancellation of
the contribution related to the stringy modes. In fact, if the complete string result is due
only to massless modes exchange (of the closed and the open strings respectively in the
two descriptions), then the usual open/closed string duality connects directly gauge theory
quantity to supergravity tree-diagrams [28]. Of course this does not happen for the whole
effective action even in the maximally supersymmetric models, but it is a possible feature
of only the first non-vanishing term in the small F (or small ) expansion.
Coming back to the technical construction presented in this paper, the main open
problem is to extend the sewing approach to the case where one has twists along the
a-cycles. As just argued, one can avoid this problem for certain cases by exploiting the
modular properties of the results. However, modular transformations cannot be enough
when twists are present both along the a-cycles and the b-cycles at the same time. This
seems a challenging problem, since even in the simplest situations, one looses the usual
relation between string amplitudes and the Schottky group. The main difficulty is that the
building blocks of the amplitudes have not been expressed so far in terms of representation
of the projective group, see [17].
Acknowledgements
We would like to thank C.S. Chu for collaboration on related topics and for interesting
discussions and suggestions on the results presented here. We would like to thank Andrew
McIntyre for email exchange and for pointing out to us various interesting papers in
the mathematical literature. This work is supported in part by EU RTN contract HPRN-
CT-2000-00131 and by MIUR contract 2001-025249. R.R. is supported by European
Commission Marie Curie Postdoctoral Fellowships.
Appendix A. Definitions
In this appendix, we collect the definitions of some quantities of relevant interest in the
theory of Riemann surfaces.
Let Σ be a compact Riemann surface of genus g with a complex structure defined on it.
There exists a normalized basis of holomorphic 1-forms ωµ (µ = 1, . . . , g), called Abelian
differentials, such that

1 1
ων = δµν , ων = τµν , (A.1)
2πi 2πi
aµ bµ
where aµ and bµ are a canonical basis of homology cycles with intersection matrix:
I (aµ , bν ) = δµν , I (aµ , aν ) = I (bµ , bν ) = 0. The matrix τ is called period matrix on Σ
and is a symmetric matrix with positive definite imaginary part.
Given a base point z0 ∈ Σ, one can associate to any z ∈ Σ a complex g-component

vector J (z) by means of the Jacobi map
z
1
J : z → Jµ (z) = ωµ . (A.2)
2πi
z0
The vector J is defined up to periods around aµ or bµ (A.1) and so belongs to the complex
torus J (Σ) = Cg /(Zg + τ Zg ) called Jacobi variety.
The Riemann Theta-function associated to the surface Σ is defined for Z ∈ Cg by

θ (Z|τ ) = exp{iπn · τ · n + 2πin · Z}, (A.3)
n∈Zg
g g
where n · τ · n = µ,ν=1 nµ τµν nν and n · Z = µ=1 nµ Zµ . The Theta-function has a
simple transformation law under shifts of Z on the lattice Zg + τ Zg :
θ (Z + τ · n + m|τ ) = exp[πi n · τ · n − n · Z]θ (Z|τ ). (A.4)

The Riemann vanishing theorem states that θ (Z|τ ) vanishes if and only if there exist
g − 1 points (z1 , . . . , zg−1 ) on Σ such that

g−1
Z = ∆(z0 ) − J (zρ ), (A.5)
ρ=1
where ∆(z0 ) is a constant vector in J (Σ) (called Riemann class) and the vectors J (zρ ) are
defined by (A.2). The dependence of ∆(z0 ) from the base point z0 is given by:
z
g−1
∆(z) = ∆(z0 ) − ω. (A.6)
2πi
z0
We also used two other important functions defined on Σ: E(z, y) and σ (z).
The prime form E(z, y) is completely defined by the following properties:
• it is a holomorphic differential form on Σ ⊗ Σ of weight (−1/2, 0) × (−1/2, 0);

• it is odd under the exchange z ↔ w;
• it has a simple zero for z → w with the normalization
z−w
E(z, w) −→ √ √ ; (A.7)
z→w dz dw
• it is single valued around the a-cycles, but not around the b-cycles: if z goes around
bµ the prime form is shifted in the following way
z
E(z, w) −→ − exp −πiτµµ − ωµ E(z, w). (A.8)
w
The weight (−1/2, 0) × (−1/2, 0) means that E(z, w) contains the square root of the
differentials dz, dw at the denominator. The rôle of these differentials becomes clear on
the sphere, where E(z, w) is exactly given by the r.h.s. of Eq. (A.7). In fact, the differentials
make E(z, w) invariant under inversion: z → 1/z, w → 1/w and therefore assure a regular
behaviour of the prime form at the infinity.
Finally the function σ (z) is defined up to a constant factor by the following properties
• it is a holomorphic differential form on Σ of weight g/2,

• it has no zeros,
• it is single valued around the a-cycles and transforms as

σ (z) −→ exp πi(g − 1)τµµ − 2πi(∆z0 )µ σ (z), (A.9)
when z is moved around the cycle bµ .
In the second part of this appendix we will discuss the properties of these quantities
under the modular transformations, that is under a change of the canonical homology basis.
Two homology basis (a, b) and (ã, b̃) are related by a modular transformation if there is a
2g × 2g matrix with integer entries

A B
∈ Sp(2g, Z), (A.10)
C D
such that

b̃ A B b
= . (A.11)
ã C D a
The g × g blocks of the matrix in (A.10) satisfy the constraints AB t = BAt , CD t = DC t ,
AD t − BC t = 1g , which assure the invariance of the intersection matrix. In order to
preserve the canonical normalization of the Abelian differentials (A.1), the ω̃ in the new
basis are related to the old ones by
ω̃ = ω · (Cτ + D)−1 , (A.12)

and correspondingly one gets the new period matrix τ̃

1
τ̃µν = ω̃ν = (Aτ + B)(Cτ + D)−1 µν . (A.13)
2πi
b̃µ
The transformation of the prime form under (A.11) is given, for example, in [11]
w w
1 −1
Ẽ(z, w) = exp ω · (Cτ + D) C · ω E(z, w). (A.14)
4iπ
z z
The transformation properties of the Riemann class (A.6), the Riemann Theta-function
(A.3) and the functions σ depend on the values of the diagonal elements of CD t and AB t .
Here we focus on the case where (CD t )µµ = (AB t )µµ = 2Z, ∀µ = 1, . . . , g. The Riemann
class (A.6) has a simple transformation property under these elements of the modular group
∆˜ = ∆(Cτ + D)−1 . (A.15)

The Riemann Theta-function (A.3) transforms as
−1
θ (Z̃|τ̃ ) = ξ det(Cτ + D) eiπZ·(Cτ +D) C·Z θ (Z|τ ), (A.16)
where Z̃ = Z(Cτ + D)−1 and ξ = 1 depending only on the modular
is a phase such that ξ 8
transformation. Finally the function σ undergoes the following transformation:

iπ −1
σ̃ (z) = K exp − ∆z · (Cτ + D) C · ∆z σ (z), (A.17)
g−1
where K is a normalization factor independent of z.
Appendix B. The Schottky parametrization
It is well known that one can represent the sphere through the stereographic projection
as the extended complex plane: C ∪ {∞}. This equivalence is the starting point to represent
any Riemann surface as part of the extended complex plane by means of the so-called
Schottky uniformization. Here we will focus on closed orientable surfaces and will give a
concrete realization of the intuitive idea that one can generate higher genus surfaces just by
adding handles to the sphere. As an example let us derive the Schottky parametrization of
the torus. It is sufficient to perform two operations. First one has to choose in C two points
J = −d/c and J = a/c and cut around them two non-overlapping disks C and C of the
same radius R = 1/|c|. Then one has to identify the two circles, which are the borders
of the disks, so to obtain again a closed surface (actually this identification can be done
with a twist of an arbitrary angle θ related to the phase of c). These two operations can
be summarized mathematically by introducing the loxodromic projective transformation
S ∈ SL(2, C)

a b
S= , with ad − bc = 1 and (Tr S)2 ∈ / [0, 4]. (B.1)
c d
This defining transformation is called generator, while the complete Schottky group S is
obtained by all the possible products of the S’s (just S n with n ∈ Z in the case of the torus).
The isometric circles are described by the equations:
−1/2 −1 −1/2
dS dS
= |cz + d| = 1, = |cz − a| = 1, (B.2)
dz dz
in agreement with what we have said above. Notice that S maps any point outside the
circle C into a point inside the circle C . On the contrary S −1 maps any point outside the
circle C into a point inside the circle C. Thus the fundamental region representing the
torus is simply the extended complex plane, modulo the equivalence relation induced by
the various elements of S: (C ∪ {∞})/S. The requirement of having two non-overlapping
disks ensures that the projective map S introduced in (B.1) is loxodromic. This kind of
transformation can be equivalently characterized by a couple of fixed points (η, ξ )
lim S n (z) = η, lim S −n (z) = ξ, ∀z = ξ, η (B.3)

n→∞ n→∞
and by a multiplier k
S(z) − η z − ξ
k= , with |k| < 1, (B.4)
S(z) − ξ z − η
valid ∀z = ξ, η. Notice that a generic loxodromic transformation S can be written in terms
of its fixed points and multiplier in a simple way

1 η − kξ −ξ η(1 − k)
S=√ . (B.5)
|k| |ξ − η| 1 − k kη − ξ
In general, one can apply these same ideas to build a surface of genus g: it is sufficient
to cut in the extended complex plane g pairs of non-overlapping disks and identify
the pairs of the corresponding circles. Clearly this corresponds to the insertion of g
handles on the sphere. Mathematically this construction is described by g loxodromic
projective transformations Sµ that generate freely the Schottky group Sg . As described
in Appendix C, in the sewing procedure the generators of the group naturally arise from
combinations of the local coordinates and of the propagator (see in particular Eq. (C.7)). So
a Riemann surface with g holes is represented by the extended complex plane modded out
by the equivalence relation induced by Sg . In this uniformization of the surface the a-cycles
correspond to the circle C anticlockwise oriented, while the b-cycles are represented by
lines connecting z and Sµ (z). Notice that the condition of having non-overlapping circles
ensures that each element Tα of Sg different from the identity is a loxodromic map and so
can be characterized by the fixed points (ηα , ξα ) and by the multiplier kα . In the text we
also make use of the following standard nomenclature
• A primitive element in the group is a transformation which can not be written as an

integer power of any other element;
• A conjugacy class is the set of the elements that can be related to each other by a
cyclic permutation of their constituent factors (for instance, S1 S2 and S2 S1 belong to
the same conjugacy class);
• A primary class is a conjugacy class containing only primitive elements.
All the quantities living on Riemann surfaces introduced in an axiomatic way in the
previous appendix can be explicitly written in terms of Poincaré series on the elements of
the Schottky group (see Appendix A of [12]). We report here only the Abelian differentials
for a surface of genus g
(µ) 1 1

ωµ (z) = − dz, µ = 1, . . . , g, (B.6)
α
z − Tα (ηµ ) z − Tα (ξµ )

where (µ)α means that the sum is over all the elements of the Schottky group that do
not have Sµn , n ∈ Z − {0}, as their rightmost factor and ηµ , ξµ are the fixed points of the
generator Sµ .
Finally it is very useful to notice the following identity

T (a) − T (b) T (c) − T (d) a − b c − d
= (B.7)
T (a) − T (d) T (c) − T (b) a − d c − b
valid for any projective transformation T and for any points a, b, c and d. This means
that the cross ration of (B.7) is invariant under projective transformation. For instance, this
identity can be used to rewrite the combinations that typically appear in the expressions for
the Abelian differentials

1 1 d T (z) − x T (z0 ) − y
T (z) − = log
T (z) − x T (z) − y dz T (z) − y T (z0 ) − x
1 1
= − . (B.8)
z − T (x) z − T −1 (y)
−1
For T = Sν , x = Tα (ηµ ) and y = Tα (ξµ ), Eq. (B.8) shows that ωµ is periodic when z goes
around a cycle bν (i.e., z → Sν (z)).
Appendix C. The twisted sewing
For trivial boundary conditions µ = 0, the derivation of the generating vertex

VNbcb ;g |
was discussed in great detailed in Appendix C of [13]. So, here we will refer to those
equations (indicated as (C.#) in the following) and just point out the novelties or differences
µ
introduced by the presence of the twist e2πiµ j0 in the propagators.
The starting point is the tree-level vertex (2.8) with N = 2g + Nb Hilbert spaces. We
label the external Hilbert spaces with the index I = 1, . . . , Nb and the legs to be sewn
together in order to build the g loops with the indices 2µ − 1, 2µ (µ = 1, . . . , g). To
be more precise we generate the higher genus surface, by identifying in the tree-level
vertex g pairs of Hilbert spaces: the Hilbert spaces H2µ−1 is identified with H2µ by means
µ
of the projective transformation P (xµ ) and the twist operator e2πiµ j0 . It is standard to
indicate the local coordinates of the insertions by means of the functions Vi (z) = γi (1 − z)
satisfying Vi (0) = zi , see (2.8). Ui (z) is related to the inverse of Vi (z): Ui (z) = Γ Vi−1 (z),
where Γ is the inversion: Γ (z) = 1/z. As in [13], a tilde on V2µ−1 or U2µ−1 indicates
the composition of the local coordinates with the projective transformation generated by
the propagator P : Ṽ = V P . In the following formulae we will always write explicitly the
effect of the twist induced by e2πij0 . The result of this manipulation is a generating vertex
with Nb insertions on the g-torus which generalizes6 Eq. (C.5)
N N
bc
b b
V = I
q = −Q|
−Qg |∆b exp
(I )
c eI M|−Qg , (C.1)
Nb ;g λ
I =1 I =1
6 The reader should keep in mind that in this paper we are interested only in the “minimal” correlation
function (2.9) with Nb insertions of b field; thus in the following equation, we can ignore all the external
(I ) (I )
oscillators bn ∀n and the cn for n λ + 1. In this respect our treatment is less general than the one in [13]. It
is straightforward, although even more cumbersome, to discuss the twisted vertex
VNbc ;g | in full generality.
b
where
g
• |−Qg stays for µ=1 |q = −Q(2µ) ,
• ∆b is the fermionic delta-function on the zero-modes of the internal lines already
present at tree-level (2.8)
λ−1 g

λ−1
∆b = Ers (Ṽ2µ−1 )e2πiµ bs(2µ) + e−iπλ Ers (V2µ )bs(2µ) ,
r=1−λ s=1−λ µ=1
(C.2)
• Ens (γ ) are the matrices of the representation of the projective group with weight
λ [13],
• eI summarizes the coupling between the external lines and the zero-modes of the
internal lines

g
λ−1
(2µ)
eI = e iπ(1−λ)
Eλs (UI Ṽ2µ−1 )e2πiµ bs(2µ) + e−iπλ Eλs (UI V2µ )b−s ,
µ=1 s=1−λ
(C.3)
• M is the result of the trace over the non-zero modes (n λ) of the internal lines,
that we perform using coherent states. The computation is then reduced to a fermionic
µ
Gaussian integral and the effect of the twists e2πiµ j0 is to add some phases in the
various coefficients of the Gaussian integral (C.9).
The result of the trace over the internal non-zero modes can be written in the usual form

B1
M = det(1 − H ) exp −(C1 C2 ) (1 − H )−1 , (C.4)
B2
with a slightly modified definition for the quadratic form H and for the linear parts C1 , B1
and B2 (C2 is unchanged). It is simple to see that the new H is like (C.9) with the first g
blocks of columns multiplied by e2πiν and last g blocks of rows multiplied by e−2πiµ

Enm (U2µ Ṽ2ν−1)e2πiν e−πiλ Enm (U2µ V2ν )
Hnm = −πiλ
µν
,
e Enm (Ũ2µ−1 Ṽ2ν−1 )e2πi(ν −µ ) Enm (Ũ2µ−1 V2ν )e−2πi(µ+λ)
(C.5)
with the convention Enm (U2µ V2µ ) = Enm (Ũ2µ−1 Ṽ2µ−1 ) = 0, ∀n, m and with n, m λ,
µ, ν = 1, . . . , g. The new linear terms are

N
m=e
(C1 )µ πi(1−λ)
cλ(I ) Eλm (UI Ṽ2µ−1 )e2πiµ ,
I =1

N
(I )
m = −e
(C2 )µ 2πiλ
cλ Eλm (UI V2µ ),
I =1

g λ−1
(2ν)
n =−
(B1 )µ Enr (U2µ Ṽ2ν−1 )br(2ν)e2πiν + e−πiλ Enr (U2µ Ṽ2ν )b−r ,
ν=1 r=1−λ
g
λ−1

πi(1−λ) −2πiµ
n =e
(B2 )µ e Enr (Ũ2µ−1 Ṽ2ν−1 )e2πiν br(2ν)
ν=1 r=1−λ (2ν)
+ Enr (Ũ2µ−1 V2ν )b−r . (C.6)
(I )
Again we have neglected the terms containing cn , n λ + 1 and all b(I ) which are present
in the complete expressions of (B1 ) and (B2 ).
It is important to notice that the presence of the twists along the b-cycles does not
modify the Schottky group structure present in [13]: the building blocks of the computation
are still written in terms of the E λ representation of the projective group, and the twists
appear only as a multiplicative phase of the usual E λ matrices. In particular this means that
in the calculation of M one still reconstructs the g Schottky generators through the usual
combination of local coordinates and the untwisted propagator
−1
Sµ = Ṽ2µ−1 U2µ = V2µ−1 P Γ V2µ . (C.7)
Because of this reason, it is not difficult to see, by following Appendices E and D of [12],
that det(1 − H ) is exactly given by Eq. (2.11).
In order to compute the zero-mode contribution (i.e., the expectation value in (C.1)), it
is necessary to expand the factor of (1 − H )−1 in the exponent of (C.4) in powers of H .
Following the same steps of [13] one gets

Nb
Nb
exp cλ(I )eI M = det(1 − H ) exp cλ(I )fI , (C.8)
I =1 I =1
where
∞
g
λ−1
−2iπλ
Eλm (UI Tα )Emr (Sµ )Ers (Uµ )e2πi(·Nα +µ ) b−s ,
(2µ)
fI = e
α µ=1 m=λ r,s=1−λ
(C.9)

where the sum α is extended over all the elements Tα of the Schottky group Sg . The
operator FNb ,g in Eq. (2.10) is then given by

Nb
(I )
FNb ,g =
−Qg |∆b exp cλ fI |−Qg . (C.10)
I =1
The explicit evaluation of this equation is particularly simple in the case λ = 1, where it is
directly related to the twisted Abelian differential (2.18). First, the sums and the products
on r, s = 1 −λ, . . . , λ−1 have a single term r, s = 0; moreover from the explicit form of the
representation E, see (A.1) of [13], it is easy to see that E00 (γ ) = 1 for all the projective
g (2µ)
transformations γ . Thus Eq. (C.2) simply becomes ∆b = µ=1 (e2πiµ − 1)b0 , while

the sum m E1m (UI Tα )Em0 (Sµ ) present in fI (C.9) can be treated as done in Appendix D
of [13]7
∞

1 1
E1m (UI Tα )Em0 (Sµ ) = −(Tα )−1 (zI ) −
m=1
Tα−1 (zI ) − Sµ (0) Tα−1 (zI )

1 1
=− − . (C.11)
zI − Tα Sµ (0) zI − Tα (0)
The first identity follows from Eq. (A.11) of [13] and from the simplest possible choice
of the local coordinates VI (z) = zI − z, while the second line is obtained by using (B.8).
At this point one can easily compute the scalar product in the g Hilbert spaces of the
loops (C.10) and in the g − 1 Hilbert spaces of external fields (2.13), simply by recalling
that
q = 1|b0 |q = 1 = 1. This means that one has to select a factor of b0 for each one
(I )
of the µ = 1, . . . , g Hilbert spaces Hµ and one c1 for each of external Hilbert space
H(I ) : this yields the determinant (2.14). Then a slightly non-trivial step is to show that
the constituents of this determinant can be written in the form of Eq. (2.19). The basic
idea is to single out the sum over the elements of the form Tα SµK . For sake of notational
simplicity, we use the following abbreviations x = exp(2πiµ ) and cK = 1/(z − Tα SµK (0)),
so Eq. (2.15) can be rewritten as
(µ)
K=∞
ζµ (zI ) = x x K (cK+1 − cK ). (C.12)
α K=−∞
Let us focus on the series over K in the parenthesis and relabel the summed index in order
to combine the two terms of (C.12)

N1
lim x K (cK+1 − cK )
N1 ,N2 →∞
K=−N2

N1
= lim (1 − x) cK x K−1 + x N1 cN1 +1 − x −N2 c−N2 . (C.13)
N1 ,N2 →∞
K=−N2 +1
The convergence of the series is made manifest by breaking the sum into two pieces (K > 0
and K 0) and by adding ad subtracting the limiting values
1 1
A = lim cK = , R = lim cK = . (C.14)
K→∞ z − Tα (ηµ ) K→−∞ z − Tα (ξµ )
So the parenthesis in (C.12) can be rewritten as follows

N1

{· · ·} = lim (1 − x) x K−1 (cK − A) + 1 − x N1 A + x N1 cN1 +1
N1 ,N2 →∞
K=1
7 We refer in particular to Eq. (D.2). Notice that here we present a generalization of that identity, since (C.11)
is valid for all projective transformation Tα , without the need of the sum over all the element of the Schottky

group α .

2 −1
N
−K−1
−N2
−N2
+ (1 − x) x (c−K − R) − 1 − x R−x c−N2 .
K=0
(C.15)
Thanks to (C.14) the terms proportional to x Ni cancel in the limit and one obtains
∞ ∞

−1

−K
{· · ·} = A − R − 1 − x x (cK − A) +
K
x (c−K − R) . (C.16)
K=1 K=0
When this result, together with (C.14), is used in (C.12), the first two terms give exactly the
first line of (2.19), while the second line of (2.19) is reproduced by the square parenthesis.
In the case λ = 3/2, 2, . . . the computations are more cumbersome because we have
(2λ − 1) zero-modes and then the number of insertions necessary to have a non-trivial
result is Nb = (2λ − 1)(g − 1). Similarly to what has just been done for λ = 1, one can
start from (C.10) and derive F (λ) which is the generalization of (2.13) for a generic spin
λ. It is convenient to replace the index I = 1, . . . , Nb with a double index (µ, r) with
µ = 1, . . . , g − 1 and r = 1 − λ, . . . , λ − 1. With this notation one gets
 
ζ 1 (z1 ) . . . ζ g (z1 )
 .. .. 
 . . 
F (λ) = det  , (C.17)
 ζ 1 (zg−1 ) . . . ζ g (zg−1 ) 
E(S1 ) ... E(Sg )
where E(Sµ )rs is defined in (2.22) and where each entry of (C.17) is a (2λ − 1) × (2λ − 1)
matrix [ζ µ (zν )]rs = ζµ,s (zν,r )

−2πi(·Nα −µ ) ∂y
(r+λ−1)
(Tα (z))λ (Sµ (y))1−λ Sµ (y) 2λ−1
ζµ,r (z) = e .
(r + λ − 1)! Tα (z) − Sµ (y) Tα (z)
α y=0
(C.18)
The convergence of the Poincaré series which defines ζ is assured by the factor (Tα (z))λ .
In fact, it is easy to show that Tα (z) = kα (Tα (z) − ξα )2 /(z − ξα )2 . Moreover, the multiplier
of a composite transformation Tα depends on the multipliers (and the fixed points) of the
generators present in Tα ; in particular kα contains a factor kµl , if the generator Sµ or its
inverse appear l times in the expression of Tα . Thus when the order α increases, the element
in the sum (C.18) contains high powers of some multiplier, thanks to the factor (Tα (z))λ .
Since |k| < 1 the only condition for the convergence of the series is λ > 0.
It is quite easy to check that for λ = 1 the determinant (C.17) reduces to the result in
Eq. (2.14). In fact,
Sµ (0) dTα (z)
ζµ (z) dz = e−2πi(·Nα −µ )
α
(Tα (z) − Sµ (0))Tα (z)
Tα (z) − Sµ (0)
= e−2πi(·Nα −µ ) d log
α
Tα (z)

1 1
= e2πi(·Nα +µ ) − dz, (C.19)
α
z − Tα Sµ (0) z − Tα (0)
where in the last line we used Eq. (B.8) and then relabeled Tα−1 as Tα in the sum. In
the general case λ = 3/2, 2, . . . , the analysis is qualitatively similar to the λ = 1 case
described in Section 3. On one hand the ζµ,r defined in (C.18) have the expected periodicity
properties, since they are single valued when z is moved around an a-cycle, while around
a bν -cycle they transform as
λ
ζµ,r Sν (z) dSν (z) = e2πiν ζµ,r (z) (dz)λ. (C.20)
However, the ζµ,r in (C.18) are too many to be identified with the twisted λ-differentials,
since they are (2λ − 1)g, instead of (2λ − 1)(g − 1), as expected. In fact this identification
is not possible, since the ζµ,r are not holomorphic, because of the pole singularity in z = 0.
The singular part comes only from two elements of the Poincaré series (C.18): Tα = I and
Tα = Sµ . So it is easy to extract this singularity for z → 0

λ−1
1
ζµ,r (z) ∼ − E(Sµ )sr . (C.21)
zs+λ
s=1−λ
Therefore one can build (2λ − 1)(g − 1) twisted holomorphic differentials by generalizing
Eq. (2.18)

λ−1
−1
Ωµ,r (z) = ζµ,r (z) − ζg,s (z) E(Sg ) st E(Sµ )t r , µ = 1, . . . , g − 1.
s,t =1−λ
(C.22)
In order to write the Ω’s we have of course supposed that E(Sg ) is invertible which is
surely true for any λ if g = 0. In fact one can easily diagonalize the (2l − 1) × (2λ − 1)
matrix E(Sg ) and check that its eigenvalues are
2πi r
e g kg − 1 , with r = 1 − λ, . . . , λ − 1, (C.23)
and none of them is vanishing. As for λ = 1 (2.17), one can write (C.17) in terms of these
λ-differentials
 
Ω 1 (z1 ) . . . Ω g−1 (z1 ) ζ g (z1 )
 .. .. .. 
 . . . 
F (λ) = det  , (C.24)
 Ω 1 (zg−1 ) . . . Ω g−1 (zg−1 ) ζ g (zg−1 ) 
0 ... 0 E(Sg )
and the final result for the “minimal” correlator we are interested in is given by the product
of (2.11) and (C.24): Zλ = det(1 − H )F (λ) .
For half-integer λ (λ = 3/2, 5/2, . . .), the same computation can be done also for the
case where µ = 0 ∀µ. In fact, all the matrices E(Sµ ) are still invertible also in this case.
This can be easily seen from (C.23) which tells that for half-integer r the eigenvalues of
any E(S) are non-vanishing even if = 0. In order to compare with the result of [13], one
can show that the differential Λµ (z) introduced there, see Eq. (D.12), can be derived also
starting from the ζµ defined in (C.18) setting µ = 0 ∀µ
Λµ (z) = −ζµ E(Sµ )−1 + ζg E(Sg )−1 . (C.25)

This means that for half-integer λ the twists along the b-cycles do not introduce any
qualitative change in the geometric interpretation of the final result and their effect is
just to bring some phase factor in the explicit definition of the differentials. This is to
be contrasted with the λ = 1 case where the counting of the zero-modes of the ∂¯ operator
changes in presence of non-trivial twists.
The case λ = 1/2 is special [14–16] since all the zero-modes are absent. Of course one
can still write a Poincaré series similar to (C.18)
(Tα (z))1/2(Sµ (0))1/2
e−2πi(·Nα −µ ) , (C.26)
α
Tα (z) − Sµ (0)
which is an automorphic form of weight 1/2 closely related to the Szëgo kernel. However it
is not possible to use the trick in (C.25) and eliminate the pole in z = 0 in order obtain some
non-trivial holomorphic differentials. In fact, the substitution Tα = Sµ Tβ shows that (C.26)
does not depend on µ and thus subtracting two values of µ in (C.26), as done in (C.25),
one obtains identically zero.
For integer λ 2 and vanishing twists = 0, we have to distinguish two cases. On the
torus g = 1 both b and c have a zero-mode, see Eq. (2.5). On higher genus surfaces g 2,
only b has zero-modes. Thus, for higher genus surfaces, the = 0 case is conceptually on
the same footing as the twisted case and one can build (2λ − 1)(g − 1) differentials starting
from (C.18). First it is easy to see why the zero-modes of c are absent in this case. Usually
they are constructed starting from the ζµ,r in (C.18) by sending λ → 1 − λ. Formally this
satisfies the periodicity condition, but the series does not converge since the factor of Tα (z)
has now a negative power. Since there are no zero-modes for c, the determinant (C.17) does
not vanish even if det[E(Sµ )] = 0 ∀µ when = 0. Hence the Riemann–Roch theorem for
g 2 is realized in the same way, independently of possible twists for λ = 3/2, 2, . . . .
References
[1] V. Alessandrini, D. Amati, Nuovo Cimento A 4 (1971) 793.

[2] A.M. Polyakov, Phys. Lett. B 103 (1981) 207;
D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 271 (1986) 93.
[3] V.G. Knizhnik, Phys. Lett. B 180 (1986) 247.
[4] L. Alvarez-Gaumé, G.W. Moore, C. Vafa, Commun. Math. Phys. 106 (1986) 1.
[5] E. Verlinde, H. Verlinde, Nucl. Phys. B 288 (1987) 357.
[6] L. Alvarez-Gaumé, J.B. Bost, G.W. Moore, P. Nelson, C. Vafa, Commun. Math. Phys. 112 (1987) 503.
[7] T. Eguchi, H. Ooguri, Phys. Lett. B 187 (1987) 127.
[8] O. Lechtenfeld, Phys. Lett. B 232 (1989) 193.
[9] J.D. Fay, in: Lecture Notes in Mathematics, Vol. 352, Springer, Berlin, 1973.
[10] D. Mumford, Birkhauser, 1983.
[11] J.D. Fay, Mem. Amer. Math. Soc. 464 (1992) 96.
[12] P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda, A. Sciuto, Nucl. Phys. B 322 (1989) 317.
[13] P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda, S. Sciuto, Nucl. Phys. B 333 (1990) 635.
[14] F. Pezzella, Phys. Lett. B 220 (1989) 544.
[15] A. Losev, JETP Lett. 49 (1989) 424, Pisma Zh. Eksp. Teor. Fiz. 49 (1989) 372 (in Russian).
[16] P. Di Vecchia, NORDITA-89/30-P, in: L. Brink et al. (Eds.), Physics and Mathematics of Strings, pp. 212–
230.
[17] C.S. Chu, R. Russo, S. Sciuto, Fortschr. Phys. 50 (2002) 871, hep-th/0201118.
[18] M.B. Green, J.H. Schwarz, E. Witten, in: Cambridge Monographs On Mathematical Physics, Cambridge
Univ. Press, Cambridge, 1987, p. 596.
[19] S. Sciuto, Lett. Nuovo Cimento 2 (1969) 411;
A. Della Selva, S. Saito, Lett. Nuovo Cimento 4 (1970) 689.
[20] A. Neveu, P.C. West, Nucl. Phys. B 278 (1986) 601;
P. Di Vecchia, R. Nakayama, J.L. Petersen, S. Sciuto, Nucl. Phys. B 282 (1987) 103.
[21] J. Polchinski, An Introduction to the Bosonic String, in: String Theory, Vol. 1, Cambridge Univ. Press,
Cambridge, 1998.
[22] C.-S. Chu, R. Russo, S. Sciuto, in preparation.
[23] M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Phys. Lett. B 400 (1997) 52, hep-th/9702037.
[24] C. Bachas, M. Porrati, Phys. Lett. B 296 (1992) 77, hep-th/9209032.
[25] M. Billo, P. Di Vecchia, D. Cangemi, Phys. Lett. B 400 (1997) 63, hep-th/9701190.
[26] A. Frizzo, L. Magnea, R. Russo, Nucl. Phys. B 604 (2001) 92, hep-ph/0012129.
[27] M. Billò, F. Lonegro, I. Pesando, personal communication.
[28] P. Di Vecchia, A. Liccardo, R. Marotta, F. Pezzella, hep-th/0305061.
Towards understanding structure of the monopole

clusters
M.N. Chernodub a,b , V.I. Zakharov c
a Institute of Theoretical and Experimental Physics, Moscow 117259, Russia
b Institute for Theoretical Physics, Kanazawa University, Kanazawa 920-1192, Japan
c Max-Planck Institut für Physik, Föhringer Ring 6, 80805 München, Germany
Abstract
We consider geometrical characteristics of monopole clusters of the lattice SU(2) gluodynamics.
We argue that the polymer approach to the field theory is an adequate means to describe the monopole
clusters. Both finite-size and the infinite, or percolating clusters are considered. We find out that the
percolation theory allows to reproduce the observed distribution of the finite-size clusters in their
length and radius. Geometrical characteristics of the percolating cluster reflect, in turn, the basic
properties of the ground state of a system with a gap.
1. Introduction
Explanation of the confinement in terms of the monopole condensation was proposed

as early as around the year 1974 [1]. Moreover, the idea is strongly supported by the lattice
data [2]. Nevertheless, understanding the lattice data in terms of the continuum theory still
represents a challenge. Indeed, generically one usually thinks in terms of a Higgs-type
model:

1 2
Seff = d4 x Fµν + |Dµ φ|2 + V |φ|2 , (1)
4
where φ is a scalar field with a non-zero magnetic charge, Fµν is the field strength tensor
constructed on the dual-gluon field Bµ , Dµ is the covariant derivative with respect to the
dual gluon. Finally, V (|φ|2 ) is the potential energy ensuring that φ = 0 in the vacuum.
E-mail address: maxim@heron.itep.ru (M.N. Chernodub).
doi:10.1016/j.nuclphysb.2003.07.019
234 M.N. Chernodub, V.I. Zakharov / Nuclear Physics B 669 (2003) 233–254
There exist detailed fits to the numerical lattice data within such a framework [3]. However,
relation of the “effective” fields φ, Bµ to the fundamental QCD fields remains unclear.
One of the main problems is the lack of understanding of the nature of the monopole
field configurations. In particular, there is seemingly no reason to think about the
monopoles as quasi-classical objects, for review see [4]. Under the circumstances, it
is natural to exploit the lattice measurements to accumulate first data on, so to say,
phenomenology of the monopoles. In fact, definition of the monopoles in the non-Abelian
case is not unique and we will concentrate on the monopoles defined within the maximal
Abelian projection (MAP) in SU(2) lattice gauge model.1
A gauge-variant definition of the monopoles would not be of much interest by itself.
The basic idea is that there are gauge-invariant objects behind which are detected through
the projection. This, gauge-invariant facet is manifested in gauge invariant properties of
the MAP monopoles. In particular, (see [5] and references therein) the three-dimensional
monopole density ρmon does not depend on the lattice spacing and is given in the physical
units:
3/2
ρmon = 0.65(2) σSU(2) ≈ const, (2)
where σSU(2) is the string tension.
An important remark is in order here. While discussing the monopole density one
should distinguish between finite-size clusters and the percolating cluster [6,7]. There is
a spectrum of the finite-size clusters, as a function of their length, while the percolating
cluster is in a single copy. In other words, the percolating cluster fills in the whole of the
lattice and its length is proportional to the volume of the lattice V4 :
Lperc = ρmon V4 . (3)
The observation (2) refers only to the percolating cluster.
Also, upon identification of the monopoles in the Abelian projection, one can measure
the non-Abelian action associated with these monopoles. The results [8] turned in fact
astonishing and can be explained only as a fine-tuning [9]. To explain the meaning of the
fine tuning, let us remind the reader that the probability of finding any field configuration
is a product of action- and entropy-factors. Then, it turns out that for the monopoles at the
presently available lattices both factors diverge in the ultraviolet but cancel each other to a
great extent:
Pmon = exp(−Smon ) × (entropy) ∼ exp(−c1 L/a) exp(+c2 L/a), (4)
|c1 − c2 |
∼ ΛQCD , (5)
a
where Pmon is the probability to find the monopole trajectory of the length L, a is the lattice
spacing and c1,2 could in principle depend on the a as well but are treated as constants to
describe the data on the available lattices (with smallest lattice spacing a ∼ 0.06 fm from
Ref. [8]).
1 Our results could be extended to SU(3) case as well. However, in this case we are lacking Monte Carlo data
on the continuum extrapolation of the monopole density and the monopole size which are crucial for our analysis.
Thus, in this paper we confine ourselves to the SU(2) gluodynamics where these results are available.
M.N. Chernodub, V.I. Zakharov / Nuclear Physics B 669 (2003) 233–254 235
The observation (4) implies also that the monopoles look point-like on the presently
available lattices. Indeed, let us normalize the phenomenological monopole action for the
lattice monopoles to the case of point-like monopoles. For the Dirac monopole the action
would be:
∞
1 L
S =L B2 d3 r = const · 2 , (6)
8π ae
a
where e is the electric charge and B is the magnetic field of the monopole, B 2 ∼ 1/e2 r 4 .
Thus, at presently available lattices the MAP monopoles are structure-less, or point-
like objects. This observation immediately triggers further questions since usually one
considers generic non-perturbative fluctuations as large-scale. We would not discuss such
questions now but simply stick to the phenomenological observation that at the presently
available lattices monopoles can be treated as geometrical objects, with no size. One can
advocate then [9] the use of the polymer approach to the field theory, see, e.g., Ref. [10].
Our basic idea here is to pursue a geometrical language to describe monopoles, without
any explicit use of Lagrangians like (1).
In Section 2 we consider the relation between the percolation theory and the monopole
physics. In Section 3 we use the percolation theory to evaluate the spectrum of the finite-
size monopole clusters. In Section 4 the same finite-size clusters are described in terms
of the simplest vacuum loop in the polymer representation. In Section 5 we comment on
the properties of the percolating cluster as representing the ground state of the system. In
Section 6 we discuss measurements which could clarify further the structure of the effective
theory describing the monopoles. In Section 7 we present conclusions.
It is worth mentioning that in a few cases we reproduce in some detail argumentation
well known in other fields, in particular in the condense matter. And in this sense our
presentation is not original in these cases. However, we believe that the justification of the
use and application of the condense-matter techniques to non-perturbative fluctuations of
the Yang–Mills fields is new.
2. Monopole clusters and percolation
In this section we will outline interconnections between properties of monopole clusters,

percolation theory [11] and field theory. The aim is to motivate our basic assumptions (see
Section 2.6) and to apply later the percolation theory to the monopole clusters.
2.1. Monopole condensation in compact U (1)
A two-line theory of the monopole condensation was presented first in Ref. [12] for the
case of the compact (or lattice) U (1) theory. We will review the derivation here and dwell
on its connection with percolation.
The role of the compactness of the U (1) lattice theory is to ensure that the Dirac string
does not cost any energy (for a review see, e.g., Ref. [4]). That is why in Eq. (6) we take
into account only the energy of the radial field. Nevertheless, the monopoles are infinitely
heavy in the limit a → 0 and, at first sight, this precludes condensation since the probability
to find a monopole trajectory of the length L is suppressed as

c L
exp(−S) = exp − 2 . (7)
e a
Note that the constant c depends on details of the lattice regularization but can be found
explicitly in any particular case.
What makes the condensation feasible, is an exponentially large enhancement factor due
to the entropy. Namely, a trajectory of length L for a point-like monopole can be realized
on a cubic lattice in NL = 7L/a various ways. To evaluate the NL let us notice that the
monopole occupies the center of a cube and at each step the trajectory can be continued
to an adjacent cube. In four dimensions there are 8 adjacent cubes. However, one of them
is to be excluded from the counting since the monopole world-line is non-backtracking.2
Thus the entropy factor is:

L
NL = exp (ln 7) , (8)
a
and it cancels the suppression due to the action (7) if the coupling e2 satisfies the condition
of criticality:
2
ecrit = c/ ln 7 ≈ 1, (9)
where we quote the numerical value of 2
ecrit 2
for the Wilson action and cubic lattice. At ecrit
any trajectory length L is allowed and the monopoles condense. This simple framework
2 is concerned
is verified within about one percent accuracy as far as the prediction of ecrit
[13,14].
One can say that the coupling e2 of the compact U (1) is to be fine-tuned to trigger the
phase transition.
2.2. Relation to percolation
In fact the derivation of the preceding subsection can be viewed as an application of the
percolation theory [11]. Moreover, one thinks in terms of the simplest percolation possible,
that is, uncorrelated percolation.
Indeed, monopoles are observed as trajectories on the links of the dual lattice [2].
Postulate that the probability to occupy a link is given by:

p = exp −c/e2 , (10)
compare Eq. (7) at L = a. The probability that (uncorrelated) links form a trajectory of
length L is then given by Eq. (7).
Formation of an infinitely long, or percolating cluster at a critical value p = pcrit is a
common feature of percolating systems. In our case,
pcrit = 1/7, (11)
2 If a piece of the trajectory is covered in the both directions it is not observed on the lattice. Physically, this
cancellation corresponds to the cancellation between monopole and anti-monopole.
see Eq. (9).

It might worth mentioning that in text-books one would rather find pcrit = 1/8 since
the non-backtracking feature of the monopole trajectory is specific for charged particles.
From the theoretical point of view the non-backtracking is a manifestation of the monopole
charge conservation. Another consequence of this conservation law is the closeness of the
trajectories which has not been taken into account so far. As we shall see in Section 4 the
closeness of the trajectories brings in only a pre-exponential factor and can be ignored at
the moment for this reason.
2.3. Relation to field theory
The derivation in Section 2.1 implies that the monopole condensation occurs when the
monopole action is ultraviolet divergent. On the other hand, the onset of the condensation
in the standard field theoretical language corresponds to zero mass of the magnetically
charged field φ. It is important to realize that this apparent mismatch between the two
languages is not specific for the monopoles at all. Actually, there is a general kinematic
relation (see, e.g., Ref. [15,16]) between the physical mass of a scalar field m2prop and the
mass M defined in terms of the (Euclidean) action S, M ≡ S/L:

ln 7
m2prop a ≈ Cm2
M(a) − , Cm2
= 8. (12)
a
It is m2prop that enters the propagator of a scalar particle D(p2 ),
1
Dsc (p2 ) ∼ , (13)
p2 + m2prop
not M 2 (a). The factor ln 7 in Eq. (12) is specific for the “monopole walk” (see the
preceding subsection). Eq. (12) establishes a connection between uncorrelated percolation
and propagator of a free particle (in Euclidean space),
Cm mprop a 2 ≈ ln(pcrit /p),
2 2
pcrit > p,
where we neglect, as usual, higher corrections in a.
It is also worth mentioning that the overall normalization of the propagator of a free
particle in the polymer approach reads in the leading (in the lattice units a) order as follows:

exp −M(a)Lpath = Cm 2 2
a Dsc (xi , xf ), (14)
paths
where Dsc (x, x ) is the standard field-theoretical propagator of a scalar particle with mass
(12) in the Euclidean space–time, Lpath is the length of a particular path connecting the
initial and final points xi,f . Note that the action factor for a given path, exp(−MLpath), is
similar to the action of a polymer of length L, with the chemical potential replaced by the
mass M(a).
Eq. (12) demonstrates that the propagating mass can be kept finite in the limit a → 0
only at price of fine tuning. Moreover, Eq. (12) looks exactly the same as the standard
radiative correction to the mass of a scalar particle. In more detail, the first term in the
right-hand side is actually the magnetic field energy. The ln 7 term, on the other hand,
plays the role of a counter-term and this counter-term is calculable in our case!
2.4. Percolation and the quadratic divergence in field theory
Now we come to explain an important difference between phase transitions in case of

uncorrelated percolation and in case of the monopoles. This difference might come as a
kind of surprise since the transition point can be determined within the percolation theory
with a good accuracy (see the discussion of the U (1) case above).
The difference concerns properties of a percolating system and of the monopole vacuum
in case of p > pcrit and in the confining phase, respectively, If p > pcrit the product of
probability and entropy factors reduces to exp(+|ε|L), where ε can be arbitrarily small.
Then, it is clear that the longest possible trajectory wins and the cluster fills in a finite
fraction of the whole lattice. In the percolation-theory language the statement is that the
fractal dimension at p > pcrit coincides with the space dimension D:
(Dfr )perc = D, p > pcrit , (15)

see, e.g., Ref. [17] and references therein. Note that in the language of field theory p > pcrit
corresponds to a tachyonic mass, m2prop < 0, see Eq. (12).
In case of monopoles, Eq. (15) would correspond to
(ρmon )perc ∼ a −3 . (16)

Note that in the field theoretical language Eq. (16) would be interpreted as a power-like
divergence well known in perturbation theory of charged scalar particles. Indeed, already
on dimensional considerations one would conclude that:
2
|φ| perc ∼ a(ρmon )perc ∼ a −2 (17)

(for a more careful derivation see Ref. [9]).
The behaviour (16) is in a sharp contrast with experimental data, see (2) and this is
what we mean by the difference in the behaviour of a percolating system at p > pcrit and
of the monopoles in the confining phase. In other words, the fine tuning exhibited by the
monopoles in the confining phase is a specific feature of the vacuum state of the lattice
gauge theory.
2.5. Fine tuning and Coulomb-like interaction
Let us emphasize again that in the present note we treat the fine tuning of the monopoles
as a pure observation and look for its implications. We are not in position at all to track its
origin back to the non-Abelian Lagrangian which after all determines the results of all the
lattice measurements. One of very few theoretical points which we can nevertheless add is
that a long-range, Coulomb-like force is a necessary ingredient of a fine-tuned theory. Of
course, in case of color by long distances we mean r a but r Λ−1 QCD .
To substantiate the point, let us consider the confining phase of the compact U (1). The
relative simplicity of this case is that the dynamics is explicitly known and, moreover,
determined by the value of the electric charge e which can be tuned arbitrarily “by hand”
(while in the non-Abelian case the coupling runs and is not under control in this sense).
Thus, one can introduce two scales in the confining phase of the compact U (1) in the
following way:
a
RUV = a, RIR = , % ≡ e2 − ecrit 2
> 0; % 1. (18)
%
At first sight, we are in the same situation as in case of uncorrelated percolation since we
introduced a tachyonic mass,
%
e−S ∼ e−µL , µ ∼ − ,
a
and nothing can prevent development of a quadratic divergence, see discussion above.
On the other hand it is known from the lattice measurements (see, e.g., Ref. [18] and
references therein) that the IR scale does coexist with the UV one. And, indeed, there is
the loophole in our logic that introducing the tachyonic mass would immediately result in
a quadratic divergence. This would be true if the interaction were local. However, imagine
that there are other monopoles at a distance of order RIR . Then these monopoles can modify
the Coulomb field of the monopole considered by order unit at r ∼ RIR . This, in turn, can
be interpreted as a change in the mass (due to interactions):

δM(a) ∼ % . (19)
a
This change can compensate for the “short-distance” tachyonic mass so that the particles
are kept away from each other at a distance of order RIR . Which also means that there is
no power-like UV divergences (see discussion above). This double-face interpretation of
(19) as IR and UV effects is a unique feature of the Coulomb-like interaction.
Note that we are not really giving a proof that the introduction of RIR is self-consistent.
We just argue that without long-range forces the fine tuning would be not possible at all.
In case of non-Abelian theories there are many more open questions since the monopole
action is not bounded from below (see, e.g., Ref. [4]) and the fact that the action is tuned
to the entropy is even more difficult to explain theoretically.
2.6. Fine tuning and λφ 4 models
It is worth to emphasize that the fine tuning which is observed for the lattice monopoles
is of the same generic type considered so mystifying in case of the Standard Model. There
is, however, a peculiarity in our case. Indeed, if we compare the “natural” estimate for ρmon
(16) with the data (2) we see that there are three powers of a −1 that are “tuned away”, not
two powers as in the standard field-theoretical language. And, indeed, the data (2) imply
in the “naive” a → 0 limit [9]:

lim |φ|2 ∼ ρmon a → 0,

a→0
1 ΛQCD
lim m2prop ∼ lim M(a) − (ln 7)/a ∼ → ∞,
a→0 a→0 a a

lim m2gluon ∼ g 2 |φ|2 → 0, (20)

a→0
where mgluon is the (dual) gluon mass which arises in theories of the type (1).
Eqs. (20) are, of course, not what one would expect for the standard λφ 4 theory with
spontaneous symmetry breaking. It is worth emphasizing, however, that the limit a → 0 in
Eqs. (20) should be understood with some reservations. What we mean actually in Eqs. (20)
by a → 0 limit is “the lattice spacing a as small as possible within availability on present
lattices”. The behaviour (20) can actually change in the academic limit a → 0 (i.e., at
lattice spacings much smaller than those presently available). Subsequent considerations
in the present paper would actually suggest such a possibility (see Section 5).
Although Eqs. (20) may not survive at smaller a, they do answer questions why new
point-like fluctuations (implied by the fine-tuning) do not disturb the standard β-function
of the SU(2) gluodynamics at existing lattices. Indeed, according to (20) all the scalar
degrees of freedom are actually removed from the physical spectrum if a → 0.
The properties (20) imply also that the potential energy V (|φ|2 ) scales in the limit a →
and our effective theory can well be a useful approximation to study the vacuum properties.
In particular, the monopole confining mechanism survives in the limit [9] a → 0.
2.7. Formulating the main hypothesis
After all these preliminary discussions we are set to formulate our main hypothesis.
Namely, we will assume that we can consider the point
p = pcrit (21)
as adequately describing the physics in the confining phase of the non-Abelian gauge
theory. The justification for this hypothesis is the fine tuning observed on the lattice and
discussed in length above.
There are important reservations to be made. Namely, the fine tuning does imply that
the physics is so to say “frozen” at p = pcrit as far as the UV scale is concerned. However,
as far as the dependence on the scale ΛQCD is concerned it could be different than at the
point (21). Moreover, we can give a more quantitative meaning to the scale “ΛQCD ” in the
case considered. The point is that the percolating cluster has self-crossings and the length
of the trajectory between these self-crossings can be considered as the monopole free-path
length. Direct measurements show [19] that this length (measured along the trajectory)
scales:
Lfree ≈ 1.6 fm. (22)
Thus, we can apply our hypothesis as long as
Lfree L a. (23)
In the next section we exploit the percolation theory to describe the structure of the
finite-size monopole clusters satisfying (23).
3. Finite-size clusters and percolation
3.1. Data and percolation picture
Detailed data on the structure of the monopole clusters were obtained in [7]. As was
mentioned above, there is a single percolating cluster, whose length grows with the lattice
volume, and finite-size clusters. In this section we will concentrate on the finite-size
clusters satisfying the condition (23). These clusters are characterized, first of all, by their
length. It was found that the length spectrum is described by a power law:
c4
N(L) = τ V4 , (24)
L
where
τ ≈3 (25)
for all lattice spacings and sizes tested and the coefficient c4 depends only weakly on β.
For our purposes we can neglect this dependence.
Another important characteristics of the clusters is their radius RL as function of the
length L. By the radius one understands the average distance between two cluster links:
a a2
L/a l/a
RL2 = (xi − x̄) =
2
|xi − xj |2 , (26)
L 2L2
i=1 i,j =1

where xi,j are coordinates of the links and x̄ = (1/L) xi . The measurements indicate:
√
RL ∼ const1 + const2 · L. (27)
Thus, our problem is to clarify whether the data (25) and (27) can be understood within
the percolation theory (and our main hypothesis, see Section 2.6). It is encouraging to
observe that even without any dynamical input we can conclude that the lattice data on
the finite-size monopole clusters reveal a picture typical for percolating systems. Indeed, a
generic form of the spectrum, for p < pcrit is
exp(−µL)
N(L) ∼ V4 , (28)
Lτ
where τ is the so-called Fischer exponent and µ vanishes at the critical point, p = pcrit .
The latter is easy to understand since at the point of the phase transition the correlation
length is infinite and there is no dimensional parameter left. Moreover, a power law like
(27) is common to the percolating systems,
RL ∼ L1/Dfr , (29)
and Dfr is called the fractal dimension. For the monopoles we apparently have from the
lattice measurements:
(Dfr )mon ≈ 2. (30)
Thus, the data on the finite-size monopole clusters exhibit a typical percolation picture
and the next question is whether it is possible to evaluate the exponents (25) and (30).
3.2. Hyperscaling relation
Similarity between percolation and properties of dynamical systems undergoing the

phase transition is well known [11]. However, direct evaluation of the critical exponents
requires, generally speaking, a particular dynamical input on the system considered.
Mostly, knowledge on the excitation spectrum is required.
However, there is a general relation between the critical exponents which we are going
to apply first. Note that our considerations here are not specific for the monopoles and
similar, e.g., to the analysis of Ref. [20]. Still, for the sake of orientation let us mention a
few points important for the application of the percolation theory to the monopoles clusters.
First, one derives all relations for p pcrit , without considering p > pcrit . As we
discussed in length in Section 2, uncorrelated percolation can be indeed relevant to the
monopole physics only at p pcrit . Second, all the relations of the percolation theory
follow more or less directly from the assumption (28) and it is worth emphasizing that
the spectrum (28) is very natural physics-wise. Indeed, the presence of the exponential
factor at p − pcrit < 0 is obvious from considerations of Section 2.2. Moreover at the point
p = pcrit there is no scale left and the power-like dependence of the spectrum on L is the
only dependence allowed. In this way one comes to include the factor L−τ as well.
Starting from the spectrum (28) it is quite straightforward, see, e.g., Ref. [20], to derive
the following well-known relation:
D
τ= + 1, (31)
Dfr
where d is the dimension of space–time and Dfr is in fact Dfr (p = pcrit ).
In our case,
τ ≈ 3, D ≡ 4, Dfr ≈ 2. (32)
Thus, Eq. (31) is satisfied within the error bars of the lattice measurements and this
observation is one of our main results. In view of its importance, we will later rederive
(32) and some generalizations of it in the language of field theory.
3.3. Fractal dimension
As is mentioned above, the relation between the length and radius of the cluster is
determined by the fractal dimension, see Eq. (27). The fractal dimension, in turn, is
determined by the kind of the walk.
In fact we are dealing with the monopole walk which, to the best of our knowledge, has
not been studied in detail in the literature. However, the characteristics of the monopole
walk are so to say flanked by the characteristics of the well-known random and self-
avoiding walks. Indeed, the monopole walk chooses freely one of 7 directions available
at each step. This is in common with the random walk. On the other hand, choosing the
eighth direction would result in an immediate self-crossing and this is forbidden. The latter
feature is in common with the self-avoiding walk. However, in contradistinction from the
self-avoiding walk, self-crossings are allowed for the monopole trajectories at later stages.
The observation central for this section is that in D = 4 the fractal dimension Dfr = 2
both for the random and self-avoiding walks. Therefore, we can predict Dfr = 2 for the
monopole walk as well.
In more detail, for the random walk one has in any number of space dimensions:3
(Dfr )random = 2. (33)

For the self-avoiding walk one has (the so-called Flory’s fractal dimension):
D+2
(Dfr )self-avoiding = (34)
3
which is valid for D 4 and gives exactly the same Dfr = 2 at D = 4.
Thus, the experimental value (Dfr )mon = 2 appears to be well understood theoretically.
4. Seeing free monopoles at short distances
From our discussion of the relation between the uncorrelated percolation and free-field
theory (see Section 2.3) it is clear that the results of the preceding section can be derived in
terms of Feynman graphs as well.4 Moreover, since we are considering relatively short
trajectories, see (23), the effect of the mass can be neglected. In this section we will
reproduce the results above by considering the simplest Feynman graph relevant, that is
a single vacuum loop. This approach allows us to derive also some generalizations, like
length-of-the-trajectory spectrum at finite temperature.
4.1. Polymer representation for a massless particle propagator
In this subsection we will evaluate the Fischer exponent τ (see (31)) starting from the
observation that the fluctuations of the field φ are massless on the scale a. To this end we
reproduce first the basic points of the polymer approach to the field theory of a free scalar
particle, see, e.g., Refs. [15,16,21,22].
The partition function for a closed polymer is:
∞
1 −MN
Z= d4 x e ZN (x, x), (35)
N
N=1
where M is the chemical potential and ZN (x0 , xf ) is the partition function of a polymer
broken into N segments:
N−1 N
δ(|xi − xi−1 | − a)
ZN (x0 , xf ) = 4
d xi . (36)
2π 2 a 3
i=1 i=1
3 This relation, in connection with the monopole clusters, is in fact mentioned in Ref. [7].
4 We are grateful to D. Diakonov for a discussion on this point and providing us with a reference [21].
This partition function (35) contains a summation over all atoms of the polymer weighted
by the Boltzmann factors. The δ-functions in (36) ensure that each bond in the polymer has
length a. The starting point of the polymer is x0 and the ending point is xf ≡ xN .
Note that there is a pre-exponential factor 1/N in Eq. (35). This is due to considering
closed trajectories. Indeed, the factor is introduced to compensate for the N -multiple
counting of the same closed trajectory in the partition function (35) since any atom on
this trajectory can be considered as the initial and final point. As we shall see later, our
final result crucially depends on the pre-exponential factors.
The crucial step to relate (36) to a free particle path integral is the so-called coarse-
graining. Namely, the N -sized polymer is divided into m units by√n atoms (N = mn), and
the limit is considered when both m and n are large while a and n a are small. We get

(ν+1)n−1 2
1 2 2
δ |x i − x i+1 | − a → exp − (x (ν+1)n − x νn ) 2
,
2π 2 a 3 πna 2 na 2
i=νn
(37)
where the index i, i = νn, . . . , (ν + 1)n − 1, labels the atoms in νth unit. The polymer
partition function becomes [22]:
m−1 2m
2 m
(x ν − x ν−1 ) 2
ZN (x0 , xf ) = const · d4 x exp −2
πna 2 na 2
ν=1 ν=1
m

× exp − n · aµ . (38)
ν=1
The xi ’s have been relabelled so that xν is the average value of x in at the coarser cell.
Note also that at this stage there appears the chemical potential µ related to the original
parameter M through
ln 7
µ=M − , (39)
a
as is discussed above.
Using the variables:
1 1 8µ
s = na 2 ν, l = a 2 N, m2prop = , (40)
8 8 a
one can rewrite the partition function (35) as
∞ l
dl 1 2
Z = const · Dx exp − ẋ (s) + m2prop ds . (41)
l 4 µ
0 x(0)=x(l)=x 0
Note that the mass renormalization in Eq. (40) is consistent with Eqs. (12) and (39).
After these preliminary steps we can readily derive the distribution in the length of the
trajectories in the massless case. To this end, let us rescale xµ and s in such a way, that
there is no l dependence left in the action if mprop = 0:
l s xµ
L= , s̃ = , x̃µ = √ . (42)
a l l
Then, indeed,
∞
dL
Z = const · I, (43)
L
0
where
1
1
x̃˙ µ (s̃) ds̃
2
I≡ Dx̃ exp − . (44)
4
x̃(0)=x̃(l)=x̃ 0
At first sight, Eq. (43) implies that we have a dL/L spectrum since there is no
L-dependence left otherwise. However, the actual spectrum refers to the number of loops
in a unit volume (see (24) and the discussion of it). Thus, the dL/L spectrum refers to the
volume in the x̃ units. Since the x̃ D ∼ x D L−D/2 we have in the physical-volume units:
dL
N(L) dL = cD VD , (45)
L1+D/2
where VD is the volume in D-dimensional space and cD is a constant. Although we are
interested only in the D = 4 case we kept D as variable to emphasize that we rederive in
fact the hyperscaling relation (31). Note that Dfr = 2 is implicit in our derivation and is
encoded in fact in the transformation (37).
4.2. Coulomb-like interaction
So far we considered approximation of free particles which corresponds to the

dominance of the monopole self-energy. We expect, however, that the monopoles interact
also Coulomb-like. Other, effective interactions are not ruled out either. The Coulomb-
like interaction can readily be included into the action in the polymer representation. The
corresponding extra piece in the action is given by:
l l
g2
SCoulomb = M ds1 ds2 ẋ1,µ Dµν (x1 − x2 )ẋ2,ν , (46)
2
0 0
where gM is the magnetic charge, Dµν (z) is the massless photon propagator, and the dot,
as usual, means differentiation with respect to the proper time.
The action (46) is manifestly invariant under the rescaling (42):
11
g2
SCoulomb = M ds̃1 ds̃2 x̃˙ 1,µ Dµν (x̃1 − x̃2 )x̃˙ 2,ν , (47)
2
0 0
and, therefore, the 1/L3 behaviour of the spectrum (45) should be still true for5 D = 4
upon inclusion of this interaction.6
5 The case D = 4 is special because the gauge coupling is dimensionless.

6 We do not discuss here subtleness which can arise from consideration of the coinciding points, x → x , in
1 2
Eq. (46).
4.3. Finite temperatures, finite volume lattices
Formalism of Section 4.1 can readily be generalized to the case of a non-zero

temperature. Indeed, we should rewrite now in terms of a length-of-the-trajectory
distribution the propagator of a massless particle at finite temperature.
As usual, finite temperature in Euclidean space corresponds to a compactified fourth
direction. Thus, we simply write the L-distribution corresponding to an ensemble of non-
interacting particles with masses m2 = (2πn)2 T 2 in d = 3 space–time:
1
N(L) = c3 V3 5/2 exp −(2πnT )2 La , (48)
L
n∈Z
where V3 = V4 T is the three-dimensional volume of the time-slice.
√ Note that in the limit
T → 0 the sum over the exponentials is proportional to 1/(T La) and we come back to
(45) upon proper normalization of the constant c3 ,
√
c3 = 2c4 πa,
where c4 enters Eq. (24). However, in the opposite limit, T → ∞ we effectively get three-
dimensional theory with the volume V3 and the loop distribution N(L) ∝ L−5/2 .
A remark on the range of validity of (48) is now in order. The point is that in (48) we
disregard chemical potential µ = 0. Whether this is a valid approximation depends on the
numerical values of the parameters involved, µ, T , a. In particular, if we tend a → 0 and
keep µ in physical units the approximation (48) is not valid.
Proceeding in a similar way we can derive also the spectrum on a finite-size lattice of the
volume V4 = 4µ=1 Xµ , where Xµ is the length of the space in µth direction. Assuming
the periodic boundary condition we get:
c0
4

N(L) = exp −(2πnµ /Xµ )2 La , (49)
L
µ=1 nµ ∈Z
where
c0 = 16π 2 c4 a 2 .
5. Infrared cluster and properties of the ground state
In this section we discuss geometrical properties of the percolating cluster. It was

suggested already in [15,22] that the percolating cluster corresponds to the φ-field
condensate in the classical approximation.7 However, it is only the phenomenon of the fine
tuning that makes the properties of the vacuum state non-trivial. Indeed, we have now two
coexisting scales, a and Λ−1QCD . Moreover, we will see that the properties of the percolating
cluster differ radically from the properties of the finite-size clusters.
7 Literally, a model with a tachyonic mass was introduced first in [22] and such a model would result in
ρmon ∼ a −3 .
5.1. Lattice data
We have already mentioned that the density of the monopoles in the percolating
cluster scales, see (2). Recently, further measurements on the geometrical elements of the
percolating cluster have been performed [19]. In more detail, the cluster consists of self-
crossings and segments connecting the crossings. It was found that the average distance
between the crossings measured along the trajectory scales, see Eq. (22). Violations of the
scaling in Lfree are negligible for all the lattices tested.
One can also measure the average of the shortest, or Euclidean distance, d, between
the two crossings connected by segments. In this case violations of the scaling are more
significant and, roughly, the data can be approximated as:
√ √
d σ ≈ 0.65 + a σ − 0.25 , (50)
where a √ is the lattice spacing, σ is the string tension for the SU(2) gluodynamics, and
0.15 < a σ < 0.35.
Thus, we are confronted with the problem of interpreting the data (2), (22), (50). The
first step is to reduce the set of data to a simple picture: normalized to a free-particle case
the measurements correspond to an infinite mass [19], m2prop ∼ a −2 . Which is at first sight
a shocking observation defying everything that we said so far. Postponing discussion of the
physics till the next subsection let us reiterate the reasons for such an interpretation of the
data.
For a free particle, one can measure its mass by comparing the Euclidean distance
between two points with the length along the corresponding trajectory connecting the
same points. Indeed, one obtains the distance L by differentiating the partition function
(35) with respect to the chemical potential µ defined in Eq. (39). On the other hand, the
dependence of a free particle propagator on the chemical potential enters through the factor
exp(−mpropd) where the mprop is as in Eq. (12). Differentiating with respect to µ the
propagator D(mprop , d) of a free particle of mass mprop we get therefore:
∂ dfree particle
Lfree particle = − ln D(mprop , d) ≈ , (51)
∂µ 2(mprop(a) a)
where we neglected higher corrections in a and mprop (generally speaking, mprop also
depends on a).
Two particular cases are worth mentioning. First, if m2phys is in the physical units,
m2prop ∼ Λ2QCD then we would have
d
L ∼ ,
a
which is in blatant disagreement with the data (50). Moreover, if we assume cancellation
of only the leading 1/a term in (12) we would expect:
d
L ∼ √ ,
a
which still cannot be reconciled with the data either.
Finally, if m2prop ∼ a −2 then

L ∼ d
which does agree with the data (50) for a available as far as we neglect the correction linear
in a.
To summarize, we are coming to a paradox. Indeed, even if we accepted m2prop ∼ a −2
to satisfy L ∼ d this would not help since we would not be able then to explain the
persistence of the physical scale in (22), (50).
5.2. Analogy to the Mößbauer effect
Thus, we see that the finite-size clusters and the percolating cluster exhibit different
patterns of the monopole kinematics. In the former case we could neglect the monopole
mass while in the latter case it is not possible at all. Looking for an analogy, we naturally
come to the Mößbauer effect. Indeed, one could say that the effect is the difference in
kinematics inherent to the decays of a free atom and of an atom belonging to a lattice in
its ground state. Formally, the kinematics of the decay of the atom belonging to the lattice
looks as if the atom had an infinite mass. Thus, the analogy we are going to pursue is
between monopoles belonging to the percolating cluster and atoms in the ground state of
a lattice of atoms. The reason for the analogy is that in the both cases we are dealing with
ground states of systems with a gap.
Let us first remind the reader the basic features of the Mößbauer phenomenon (for a
review see, e.g., Ref. [23]). One considers decay of an atom belonging to a lattice of atoms.
Then there exists a probability P0 that the lattice is not excited at all, i.e., no phonons are
emitted. Then the recoil momentum is transferred to the whole of the lattice and, clearly,
this corresponds to an infinite effective mass of the decaying atom.
In the language of quantum mechanics, the probability of the Mößbauer transition is
determined by the following matrix element:
2
P0 = i| exp(ipγ · Xatom )|i , (52)
where |i is the ground state of the lattice, Xatom is the position of the decaying atom
and pγ its recoil momentum. Then, there are two crucial quantities which determine the
estimate of P0 , namely the “naive” recoil energy R and the energy of the gap, ωgap . For
non relativistic kinematics,
p2γ
R= .
2Matom
A rough estimate for P0 is as follows:
P0 ∼ exp(−R/ωgap ). (53)
Thus, when the recoil energy is much larger than the gap, R ωgap P0 tends to zero and
kinematics is the same as for a free-atom decay. In this limit we would come back to the
parton-like picture.
The lattice data on the monopoles indicate that it is the opposite limit, R ωgap , which
can be relevant to the monopoles (if the analogy is correct at all). Next we will discuss
whether such a picture is consistent with the monopole properties.
5.3. Quantum propagator as Brownian motion
The Mößbauer effect concerns kinematics in the Minkowski space. On the other hand,
we are considering the monopole motions in the Euclidean space. Moreover, there are no
decays of the monopoles of course and, at first sight, there is no physical quantity analogous
to pγ . However, we will argue in this subsection that the famous Brownian picture [11] for
the quantum propagator provides a key to identify an analog to |pγ | in the monopole case.
The classical Brownian motion (see, e.g., Ref. [11]) is the motion of a particle of mass
MBr which results from exchange of momenta with particles of a medium which are not
observed and chaotic. Then the Brownian motion is a random walk with the step
9p 9τ
b∼ ,
Mbr
where 9τ is the average time between collisions.
In case of the quantum propagator one should rather think in terms of the Brownian
motion of a particle attached to a string with a non-vanishing tension. This is the origin
of the inertia, or of the mass term (for details see Section 4.1). As for a non-vanishing
9p it is now quantum in origin and can be estimated from the uncertainty principle.
Since the monopole trajectory is measured on the scale a, generically 9p ∼ a −1 and is
parametrically large if a → 0.
This simplest estimate should be corrected, however, for the fact that one can introduce
mass and kinetic energy only after coarse-graining see Section 4.1. Therefore,
1
9p ∼ √ (54)
a n
where n was introduced in Eq. (37). Note that n does not depend on a and for practical
estimates one can use
√
n ≈ 2–3.
5.4. Limit a → 0: academic and realistic
After the discussion in preceding subsections, it is clear that the analogy with the
Mößbauer effect reduces to the following substitutions:

8µ 1
Matom → mprop ≈ , |pγ | → 9p ≈ √ ,
a a n

ωgap → mgl , R → m2prop + (9p)2 − mprop , (55)
where mgl is the lowest glueball mass, mgl ≈ 1.5 GeV, mprop was discussed in detail above,
see, in particular (12), and n is introduced in (54).
Since 9p ∼ a −1 , mprop ∼ a −1/2 , mgl ∼ a 0 we have
lim R = 9p mgl .
a→0
In other words, in this limit the probability of the Mößbauer-like transition is vanishing,
existence of the gap is not important and we can think in terms of the parton model.
This conclusion could be foreseen: at short distances, i.e., in the limit a → 0 one
can neglect the effects of the binding of the monopoles. What is more surprising is that
for realistic values of a we are in fact far from the academic limit a → 0. This is due
to interplay of various numerical factors. Consider, for example, a = (3 GeV)−1 which
corresponds to smallest a available. Then:
8µ
m2prop ≈ ∼ 3 GeV2 , (9p)2 ∼ 2 GeV2 , R ∼ 0.5 GeV, (56)
a
where we used n ≈ 5 (see Eq. (54)) and the chemical potential (39) was estimated with the
help of Eq. (22) as µ ∼ (1.6 fm)−1 . According to this estimate at presently available lattices
R is indeed considerably smaller than mgl and there is no surprise that the Mößbauer-
type transition dominates. In other words, the onset of the parton model is expected when
copious excitation of glueballs is favored kinematically. Since the glueball is relatively
heavy, mgl ∼ 1.5 GeV the realistic values of 9p are far from satisfying the condition
R mgl .
Thus, we come to the following conclusion:
The recoil-free geometrical picture of the percolating cluster √ is consistent with the
quantum mechanics as far as the resolution (which is of order n a) is not too high. At
better resolution we expect to loose both point-like monopoles and the geometrical picture
itself.
Actually, in the data of Ref. [8] one can already see some hints on violations of the 1/a
behaviour of the monopole field-theoretical mass M(a). Thus, we are inclined to correlate
these first indications to appearance of the monopole structure [8] and the scaling violations
in dav , see Eq. (50). Our prediction is that these indications become clearer at smaller a.
One can roughly estimate the values of acrit at which the parton picture begins to prevail
over the recoil-less kinematics:
−1
acrit 10 GeV. (57)
Indeed, at such a, R ≈ mgl if the estimates above are taken at their face value. At such a one
can also expect that the monopoles do not look point-like. Indeed, in the field theoretical
language it is quite trivial that the dominance of the inelastic processes and appearance of
the form-factor for an “elastic” process are determined by the same dynamics.
Estimate (57) is apparently quite uncertain since copious production of glueballs has
not been studied in detail even at colliders, to say nothing about the Euclidean analog of it.
Probably, only measurements at smaller a would clarify the situation.
Still, emergence of a large mass scale (57) is remarkable. Indeed, there exist independent
indications on relevance of large mass scales to QCD phenomenology (see, e.g., Ref. [25]
and references therein). Moreover, the large mass scale emerges in pure gluonic channels.
In particular, numerical estimates for the critical mass at which the parton model receives
large corrections in the 0++ glueball channel are close to (57), Mcrit ∼ 10 GeV. The
estimate of Mcrit is pure phenomenological [25]. Now we see how such a scale could
be build up starting from Lfree ≈ (150 MeV)−1 and proceeding through the glueball mass
and other numerical factors.
6. Monopoles and glueballs
Thus, the next crucial question seems to be the coupling of the monopoles to glueballs.
In the most general way the monopole–glueball interaction can be studied by measuring
the correlation function:
K(x − y) = 0|ρ(x), ρ(y)|0, (58)

where
ρ(t) ≡ φ + (x, t)φ(x, t).

Asymptotic at large distances is sensitive to a massive excitation [24]:
lim 0|ρ(t), ρ(0)|0 = c1 + c2 e−mgl t , (59)

t →∞
which is likely to be a glueball.
As far as we approximate monopoles as point-like and confine ourselves to an effective
λφ 4 theory without gluons then correlators like (58) are given by a sum over closed
monopole loops:

∂2
K(x − y) = Z(M) , (60)
∂j1 ∂j2 j1 =j2 =0
where
M(z) = j1 δ 4 (x − z) + j2 δ 4 (y − z),
and the partition function is given by:
∞
1 dL −m2 La −V (x(τ ))
Z(M) = 4
d x0 e e exp − M x(τ ) dτ , (61)
4π 2 L3 x0
0
where · · ·x0 means averaging over all the closed paths. Further details and definitions can
be found, e.g., in the review [26]. What is important for us now is that the sum over the
paths can be calculated numerically using the lattice data since the monopole trajectories
are directly observed.
The constant in Eq. (59) is related to the ρmon :
c1 = (8ρmon a)2 , (62)

and is entirely determined by the percolating cluster. Determination of the glueball mass
would require, on the other hand, sensitivity to the quantum corrections, or to the finite-size
clusters.8
It could quite well be so that keeping the closed paths alone would not be adequate to
determine the glueball mass through (59). This would be a signal that gluonic intermediate
8 The percolating cluster alone at distances d d is representing a cluster of fractal dimension D = 4 (see
av fr
our discussion of the uncorrelated percolation in Section 2) and cannot be sensitive to the glueball mass.
states are important and the theory is not unitary without inclusion of such states. Since the
gluons are not detected directly on the lattice the pure gluonic intermediate states would
look as breaks in the monopole loops.
Thus, measurements of the correlator (58) would be very important to further
understand the structure of the effective theory of the monopole interactions.
7. Conclusions
We started with the picture according to which non-Abelian gluodynamics, when

projected onto the scalar-field theory via monopoles, corresponds to a fine tuned theory
[9]. The monopoles which we considered are defined (“detected”) through the Maximal
Abelian projection. However, the mass scales which exhibit mass hierarchy are gauge-
independent. The scales are provided by the SU(2) invariant action per unit length of the
monopole trajectory, on one hand, and by inverse “free path length” (see (23)), on the
other. The observed fine tuning suggests the use of the geometrical language to describe
the monopoles.
In this paper we confronted the polymer approach to the effective theories of the
magnetically charged fields with the lattice data on the monopole clusters.
There are two types of clusters, namely, finite-size and percolating. In the former case
we have demonstrated that the length spectrum and the dependence of the size of the
clusters on their length are well understood within the percolation theory. An alternative
language is the simplest vacuum loop in the polymer language.
There is a striking difference, however, between the finite-size clusters and the
percolating cluster. In the former case one thinks in terms of the vanishing mass of the
monopoles (on the scale of the lattice spacing a):
(m2mon )finite-size ≈ 0 · a −2 . (63)
This picture is confirmed by evaluation of the critical exponents just mentioned. On the
other hand, the straightforward interpretation of the geometry of the percolating cluster
leads to [19]:
(m2mon )percolating ∼ a −2 . (64)
We have argued that the apparent discrepancy between the effective monopole masses
extracted in the two ways is naturally resolved if one takes into account that the finite-size
clusters correspond to quantum corrections while the percolating cluster corresponds to the
ground state. The quantum corrections reveal then presence of a massless (on the scale of
a) excitation. In case of the ground state, we observe an analog of the Mößbauer effect. The
onset of a parton-like description is delayed numerically by some factors, most notably by
a relatively high glueball mass mgl where mgl plays the role of the gap. We also predict
that at much smaller a (see (57)) one should see the convergence of (64) to (63).
Thus, first applications of the polymer picture to the monopole clusters turn successful.
The most important implication is that the monopoles emerge as new point-like probes
in the Euclidean version of the SU(2) gluodynamics. However, justification for such an
approach remains at present pure phenomenological in nature and relies on the observation
of the fine tuning. Any further theoretical development is therefore to be checked by

measurements on the lattice. In particular, studies at smaller a, both theoretical and
numerical, are desirable.
Acknowledgements
We are grateful to D. Antonov, P.Yu. Boyko, F.V. Gubarev, D. Diakonov, R. Hofmann,

M. Koma, Y. Koma, K. Konishi, K. Langfeld, G. Marchesini, S. Narison, M.I. Polikarpov,
L. Stodolsky, T. Suzuki, N. Uraltsev, and P. van Baal for discussions. M.N.Ch. is supported
by the JSPS Grant No. P01023. V.I.Z. is partially supported by the grant INTAS-00-00111
and by the DFG program “From lattice to hadron phenomenology”.
References
[1] Y. Nambu, Phys. Rev. D 10 (1974) 4262;

G. ’t Hooft, in: High Energy Physics, Editrice Compositori, Bologna, 1975;
S. Mandelstam, Phys. Rep. C 23 (1976) 516.
[2] M.N. Chernodub, M.I. Polikarpov, in: Cambridge 1997, Confinement, Duality, and Nonperturbative Aspects
of QCD, p. 387 hep-th/9710205;
T. Suzuki, Prog. Theor. Phys. Suppl. 131 (1998) 633;
A. Di Giacomo, Prog. Theor. Phys. Suppl. 131 (1998) 161.
[3] T. Suzuki, H. Shiba, Phys. Lett. B 351 (1995) 519;
S. Kato, et al., Nucl. Phys. B 520 (1998) 323;
M.N. Chernodub, et al., Phys. Rev. D 62 (2000) 094506.
[4] M.N. Chernodub, F.V. Gubarev, M.I. Polikarpov, V.I. Zakharov, Yad. Fiz. 64 (2001) 615, hep-th/0007135.
[5] V.G. Bornyakov, M. Muller-Preussker, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 646.
[6] T.L. Ivanenko, A.V. Pochinsky, M.I. Polikarpov, Phys. Lett. B 252 (1990) 631;
S. Kitahara, Y. Matsubara, T. Suzuki, Prog. Theor. Phys. 93 (1995) 1.
[7] A. Hart, M. Teper, Phys. Rev. B 58 (1998) 014504;
A. Hart, M. Teper, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 497.
[8] V.G. Bornyakov, et al., Phys. Lett. B 537 (2002) 291;
V.A. Belavin, M.I. Polikarpov, A.I. Veselov, JETP Lett. 74 (2001) 453.
[9] V.I. Zakharov, Hidden mass hierarchy in QCD, hep-ph/0202040.
[10] K. Symanzik, in: R. Jost (Ed.), Proc. Int. School of Physics “Enrico Fermi”, Varenna Course XLV, Academic
Press, San Diego, CA, 1969;
C.A. De Carvalhom, S. Caracciolo, J. Frohlich, Nucl. Phys. B 215 (1983) 209;
G. Parisi, Statistical Field Theory, Addison–Wesley, Reading, MA, 1987, Chapter 16.
[11] C. Itzykson, J.-M. Drouffe, in: Statistical Field Theory, Vol. 1, Cambridge Univ. Press, Cambridge, UK,
1989;
D. Stauffer, A. Aharony, Introduction to Percolation Theory, Taylor and Francis, London, 1994;
G. Grimmett, Percolation, in: Grundlehren der mathematischen Wissenschaften, Vol. 321, Springer, Berlin,
1999.
[12] A.M. Polyakov, Phys. Lett. B 59 (1975) 82.
[13] T. Banks, R. Myerson, J.B. Kogut, Nucl. Phys. B 129 (1977) 493.
[14] H. Shiba, T. Suzuki, Phys. Lett. B 343 (1995) 315.
[15] S. Samuel, Nucl. Phys. B 154 (1979) 62.
[16] J. Ambjorn, B. Durhuus, T. Jonsson, Quantum Geometry. A Statistical Field Theory Approach, Cambridge
Univ. Press, Cambridge, UK, 1997;
J. Ambjorn, Quantization of geometry, Talk given at Les Houches Summer School, Les Houches, France,
hep-th/9411179.
[17] M. Aizenman, H. Kesten, C.M. Newman, Commun. Math. Phys. 111 (1987) 505.
[18] J. Jersak, C.B. Lang, T. Neuhaus, Phys. Rev. Lett. D 54 (1996) 6909.
[19] P.Yu. Boyko, M.I. Polikarpov, V.I. Zakharov, Geometry of percolating monopole clusters, hep-lat/0209075.
[20] A.M.J. Schakel, Phys. Rev. E 63 (2001) 026115;
S. Bund, A.M.J. Schakel, Mod. Phys. Lett. B 13 (1999) 349.
[21] A.M. Polyakov, Gauge Fields and Strings, Harwood Academic, Reading, UK, 1987, Chapter 9.
[22] M. Stone, P.R. Thomas, Phys. Rev. Lett. 41 (1978) 351.
[23] H.J. Lipkin, Ann. Phys. (USA) 9 (1960) 332;
H.J. Lipkin, Ann. Phys. (USA) 18 (1962) 182.
[24] T.L. Ivanenko, A.V. Pochinsky, M.I. Polikarpov, Phys. Lett. B 302 (1993) 458;
K. Langfeld, H. Reinhardt, Monopole–anti-monopole in MAG projected lattice gauge theory, hep-
lat/0206021.
[25] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 191 (1981) 301;
F.V. Gubarev, M.I. Polikarpov, V.I. Zakharov, Phys. Lett. B 438 (1998) 147;
K.G. Chetyrkin, S. Narison, V.I. Zakharov, Nucl. Phys. B 550 (1999) 353.
[26] Ch. Schubert, Phys. Rep. 355 (2001) 73.
Atmospheric neutrino oscillations, θ13 and neutrino

mass hierarchy
J. Bernabéu a , Sergio Palomares-Ruiz a , S.T. Petcov b,1
a Departamento de Física Teórica and IFIC, Universidad de Valencia-CSIC, 46100 Burjassot, Valencia, Spain
b Scuola Internazionale Superiore di Studi Avanzati and Istituto Nazionale di Fisica Nucleare,
I-34014 Trieste, Italy
Abstract
We derive predictions for the Nadir angle (θn ) dependence of the ratio Nµ /Ne of the rates of the
µ-like and e-like multi-GeV events measured in water-Čerenkov detectors in the case of 3-neutrino
oscillations of the atmospheric νe (ν̄e ) and νµ (ν̄µ ), driven by one neutrino mass squared difference,
| m231 | ∼ (2.5–3.0) × 10−3 eV2 m221 . This ratio is particularly sensitive to the Earth matter
effects in the atmospheric neutrino oscillations, and thus to the values of sin2 θ13 and sin2 θ23 , θ13
and θ23 being the neutrino mixing angle limited by CHOOZ and Palo Verde experiments and that
responsible for the dominant atmospheric νµ → ντ (ν̄µ → ν̄τ ) oscillations. It is also sensitive to the
type of neutrino mass spectrum which can be with normal ( m231 > 0) or with inverted ( m231 < 0)
hierarchy. We show that for sin2 θ13 0.01, sin2 θ23 0.5 and at cos θn 0.4, the Earth matter
effects modify substantially the θn -dependence of the ratio Nµ /Ne and in a way which cannot be
reproduced with sin2 θ13 = 0 and a different value of sin2 θ23 . For normal hierarchy the effects can
be as large as ∼ 25% for cos θn ∼ (0.5–0.8), can reach ∼ 35% in the Earth core bin cos θn ∼ (0.84–
1.0), and might be observable. They are typically by ∼ 10% smaller in the inverted hierarchy case.
An observation of the Earth matter effects in the Nadir angle distribution of the ratio Nµ /Ne would
clearly indicate that sin2 θ13 0.01 and sin2 θ23 0.50.
E-mail address: sergio.palomares@uv.es (S. Palomares-Ruiz).

1 Also at: Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 1784 Sofia,
Bulgaria.
doi:10.1016/j.nuclphysb.2003.07.025
256 J. Bernabéu et al. / Nuclear Physics B 669 (2003) 255–276
1. Introduction
The publication of the first results of the KamLAND experiment marks the beginning
of a new era in the studies of neutrino mixing and oscillations—the era of high precision
determination of the neutrino mixing parameters. The data obtained by the solar neutrino
experiments Homestake, Kamiokande, SAGE, GALLEX/GNO and Super-Kamiokande
(SK) [1,2] provided the first strong evidences for oscillations of flavour (electron)
neutrinos. Strong evidences for oscillations of the atmospheric νµ (ν̄µ ) neutrinos were
obtained by the Super-Kamiokande (SK) experiment [3]. The evidences for solar νe
oscillations into active neutrinos νµ,τ , were significantly reinforced during the last two
years (i) by the combined first data of the SNO experiment [4], and the SK data [2],
(ii) by the more recent SNO neutral current data [5], and (iii) by the first results of the
KamLAND [6] experiment. The KamLAND data practically establishes [6], under the
plausible assumption of CPT-invariance, the large mixing angle (LMA) MSW solution as
unique solution of the solar neutrino problem. This result brings us, after more than 30
years of research, initiated by the pioneer works of Pontecorvo [7] and the experiment
of Davis et al. [8], very close to a complete understanding of the true cause of the solar
neutrino problem.
The interpretation of the solar and atmospheric neutrino, and of the KamLAND data
in terms of neutrino oscillations requires the existence of 3-neutrino mixing in the weak
charged lepton current (see, e.g., [9]):

3
νlL = Ulj νj L . (1)
j =1
Here νlL , l = e, µ, τ , are the three left-handed flavor neutrino fields, νj L is the left-handed
field of the neutrino νj having a mass mj and U is the Pontecorvo–Maki–Nakagawa–
Sakata (PMNS) neutrino mixing matrix [10],

Ue1 Ue2 Ue3
U = Uµ1 Uµ2 Uµ3
Uτ 1 Uτ 2 Uτ 3

c12 c13 s12 c13 s13 e−iδ
= −s12 c23 − c12 s23 s13 e iδ c12 c23 − s12 s23 s13 e iδ s23 c13 , (2)
s12 s23 − c12 c23 s13 eiδ −c12 s23 − s12 c23 s13 eiδ c23 c13
where we have used a standard parametrization of U with the usual notations, sij ≡ sin θij ,
cij ≡ cos θij , and δ is the Dirac CP-violation phase.2 If we identify m221 and m231 with
the neutrino mass squared differences which drive the solar and atmospheric neutrino
oscillations, m2
= m221 > 0, m2A = m231 , the data suggest that | m231| m221 .
In this case θ12 and θ23 , represent the neutrino mixing angles responsible for the solar and
2 We have not written explicitly the two possible Majorana CP-violation phases which do not enter into the
expressions for the oscillation probabilities of interest [11] (see also [12]). We assume throughout this study
0 θ12 , θ23 , θ13 < π/2.
J. Bernabéu et al. / Nuclear Physics B 669 (2003) 255–276 257
atmospheric neutrino oscillations, θ12 = θ

, θ23 = θA , while θ13 is the angle limited by the
data from the CHOOZ and Palo Verde experiments [13,14].
Combined νe → νµ(τ ) and ν̄e → ν̄µ(τ ) oscillation analyzes of the solar neutrino and
KamLAND [6] data, performed under the assumption of CPT-invariance which we will
suppose to hold throughout this study, show [15,16] that the data favor the LMA MSW
solution with m2
∼ = (6.9–7.3) × 10−5 eV2 and tan2 θ
∼ = (0.42–0.46). A second,
statistically somewhat less favored LMA solution (LMA II) exists at [15,16] m2
∼ =
−4
1.5 × 10 eV . The atmospheric neutrino data, as is well known, is best described [3]
2
in terms of dominant νµ → ντ (ν̄µ → ν̄τ ) oscillations with | m2A| ∼ = 2.5 × 10−3 eV2 and
2 ∼
sin 2θA = 1.0. The 90% C.L. allowed intervals of values of the two-neutrino oscillation
parameters found in [3] read | m2A | ∼ = (1.9–4.0) × 10−3 eV2 and sin2 2θA ∼ = (0.89–1.0).
According to the more recent combined analysis of the data from the SK and K2K
experiments [17], one has | m2A | ∼ = (2.7 ± 0.4) × 10−3 eV2 . We will often use in our
analysis as illustrative the values | m2A | = 3.0 × 10−3 eV2 and sin2 2θA = 0.92; 1.0. Let
us note that the atmospheric neutrino and K2K data do not allow one to determine the signs
of m2A , and of cos 2θA when sin2 2θA = 1.0. This implies that in the case of 3-neutrino
mixing one can have m231 > 0 or m231 < 0. The two possibilities correspond to two
different types of neutrino mass spectrum: with normal hierarchy (NH), m1 < m2 < m3 ,
and with inverted hierarchy (IH), m3 < m1 < m2 . The fact that the sign of cos 2θA is not
determined when sin2 2θA = 1.0 implies that when, e.g., sin2 2θA ≡ sin2 2θ23 = 0.92, two
values of sin2 θ23 are possible, sin2 θ23 ∼= 0.64 or 0.36.
In what regards the mixing angle θ13 , a 3-ν oscillation analysis of the CHOOZ data [18]
led to the conclusion that for m2
10−4 eV2 , the limits on sin2 θ13 practically coincide
with those derived in the 2-ν oscillation analysis in [13]. A combined 3-ν oscillation
analysis of the solar neutrino, CHOOZ and the KamLAND data, performed under the
assumption of m2
| m2A | (see, e.g., [9,19]), showed [15] that sin2 θ13 < 0.05 at
99.73% C.L. The authors of [15] found the best-fit value of sin2 θ13 to lie in the interval
sin2 θ13 ∼= (0.00–0.01).
Getting more precise information about the value of the mixing angle θ13 , determining
the sign of m2A , or the type of the neutrino mass spectrum (with normal or inverted
hierarchy), and measuring the value of sin2 θ23 with a higher precision is of fundamental
importance for the progress in the studies of neutrino mixing.
The mixing angle θ13 , or the absolute value of the element Ue3 of the PMNS matrix,
|Ue3 | = sin θ13 , plays a very important role in the phenomenology of the 3-neutrino
oscillations. It drives the sub-dominant νµ ↔ νe (ν̄µ ↔ ν̄e ) oscillations of the atmospheric
νµ (ν̄µ ) and νe (ν̄e ) [20–22]. The value of θ13 controls also the νµ → νe , ν̄µ → ν̄e , νe → νµ
and ν̄e → ν̄µ transitions in the long baseline neutrino oscillation experiments (MINOS,
CNGS), and in the widely discussed very long baseline neutrino oscillation experiments at
neutrino factories (see, e.g., [23–26]). The magnitude of the T-violating and CP-violating
probabilities in neutrino oscillations is directly proportional to sin θ13 (see, e.g., [27–
29]). Thus, in the sub-dominant channels of interest to T- and CP-violation studies, the
corresponding asymmetries become proportional to m2
/ sin θ13 [29,30]. The value of
sin θ13 is thus of prime importance.
If the neutrinos with definite mass are Majorana particles (see, e.g., [12]), the predicted
value of the effective Majorana mass parameter in neutrinoless double β-decay depends
strongly in the case of hierarchical neutrino mass spectrum on the value of sin2 θ13 (see,
e.g., [31]).
The sign of m2A determines, for instance, which of the transitions (e.g., of atmospheric
neutrinos) νµ → νe and νe → νµ , or ν̄µ → ν̄e and ν̄e → ν̄µ , can be enhanced by the
Earth matter effects [32–34]. The predictions for the neutrino effective Majorana mass
in neutrinoless double β-decay depend critically on the type of the neutrino mass spectrum
(normal or inverted hierarchical) [31,35]. The knowledge of the value of θ13 and of the
sign of m2A = m231 is crucial for the searches for the correct theory of neutrino masses
and mixing as well.
Somewhat better limits on sin2 θ13 than the existing one can be obtained in the MINOS
experiment [36]. Various options are being currently discussed (experiments with off-axis
neutrino beams, more precise reactor antineutrino and long base-line experiments, etc.,
see, e.g., [37]) of how to improve by at least an order of magnitude, i.e., to values of
∼ 0.005 or smaller, the sensitivity to sin2 θ13 . The sign of m2A can be determined in
very long base-line neutrino oscillation experiments at neutrino factories (see, e.g., [23,
24]), and, e.g., using combined data from long base-line oscillation experiments at the
JHF facility and with off-axis neutrino beams [38]. If the neutrinos with definite mass are
Majorana particles, it can be determined by measuring the effective neutrino Majorana
mass in neutrinoless double β-decay experiments [31,35]. Under certain rather special
conditions it might be determined also in experiments with reactor ν̄e [39].
In the present article we study possibilities to obtain information on the value of sin2 θ13
and on the sign of m2A = m231 using the atmospheric neutrino data, accumulated by
the SK experiment, and more generally, that can be provided by future water-Čerenkov
detectors, like UNO and Hyper-Kamiokande. We consider 3-neutrino oscillations of the
atmospheric νµ , ν̄µ , νe and ν̄e under the condition m2
= m221 | m2A | = | m231|,
which is suggested to hold by the current solar and atmospheric neutrino data. Under
the indicated condition, the expressions for the probabilities of νµ → νe (νe → νµ ) and
ν̄µ → ν̄e (ν̄e → ν̄µ ) transitions contain sin2 θ23 as a factor, which determines their maximal
values. Depending on the sign of m231 , the Earth matter effects can resonantly enhance
either the νµ → νe and νe → νµ , or the ν̄µ → ν̄e and ν̄µ → ν̄µ transitions if sin2 θ13 = 0.
The effects of the enhancement can be substantial for sin2 θ13 0.01. They are larger in
the multi-GeV e-like and µ-like samples of events and for atmospheric neutrinos with
relatively large path length in the Earth, crossing deeply the mantle [24,25] or the mantle
and the core [20,22,40–42], i.e., for cos θn 0.4, where θn is the Nadir angle characterizing
the neutrino trajectory in the Earth.
The νµ → νe (ν̄µ → ν̄e ) and νe → νµ (ν̄e → ν̄µ ) transitions in the Earth lead to
the reduction of the rate of the multi-GeV µ-like events and to the increase of the rate
of the multi-GeV e-like events in the Super-Kamiokande (or any other water-Čerenkov)
detector with respect to the case of absence of these transitions (see, e.g., [20–22,41,42]).
Correspondingly, as observables which are sensitive to the Earth matter effects, and thus
to the value of sin2 θ13 and the sign of m231 , as well as to sin2 θ23 , we consider the
Nadir-angle distributions of the ratios Nµ3ν /Ne3ν and Ne3ν /Ne0 , where Nµ3ν and Ne3ν and
are the multi-GeV µ-like and e-like numbers of events (or event rates) in the case of 3-ν
oscillations of the atmospheric νe , ν̄e and νµ , ν̄µ , and Ne0 is the number of e-like events
in the case of absence of oscillations (sin2 θ13 = 0). The ratio of the energy and Nadir
angle integrated µ-like and e-like events, Nµ /Ne , has been measured with a relatively
high precision by the SK experiment [3]. The systematic uncertainty, in particular, in the
Nadir angle dependence of the ratio Nµ /Ne can be smaller than those on the measured
Nadir angle distributions of µ-like and e-like events, Nµ and Ne . As was pointed out in
Ref. [54], an event-by-event distinction of µ− ’s could be carried out by observing the
proton above Čerenkov threshold in quasi-elastic events. These protons events are rare,
though very distinctive when they do occur. This could also be useful for evaluating charge-
asymmetries.
We obtain predictions for the Nadir-angle distributions of Nµ3ν /Ne3ν and of Ne3ν /Ne0
both for neutrino mass spectra with normal ( m231 > 0) and inverted ( m231 < 0)
hierarchy, (Nµ3ν /Ne3ν )NH , (Nµ3ν /Ne3ν )IH , (Ne3ν /Ne0 )NH and (Nµ3ν /Ne0 )IH , and for sin2 θ23 =
0.64; 0.50; 0.36. We compare the latter with the predicted Nadir-angle distributions
(i) of the ratio Nµ /Ne for the case the 3-neutrino oscillations taking place in vacuum,
(Nµ3ν /Ne3ν )vac , and (ii) of the ratio Nµ2ν /Ne0 , where Nµ2ν is the predicted number of µ-like
events in the case of 2-neutrino νµ → ντ and ν̄µ → ν̄τ oscillations of the atmospheric
νµ and ν̄µ . Predictions for the different types of ratios indicated above of the suitably
integrated Nadir angle distributions of the µ-like and e-like multi-GeV events are also
given. Our results show, in particular, that for sin2 θ23 0.50 sin2 θ13 0.01 and m231 >
0, the effects of the Earth matter enhanced νµ → νe and νe → νµ transitions of the
atmospheric νµ and νe , might be observable with Super-Kamiokande detector. Conversely,
if the indicated effects are indeed observed in Super-Kamiokande experiment, that would
suggest in turn that sin2 θ13 0.01, sin2 θ23 0.50 and that the neutrino mass spectrum is
with normal hierarchy, m231 > 0.
Let us note finally that the Earth matter effects have been widely investigated (for a
recent detailed study see, e.g., Ref. [43] which contains also a rather complete list of
references to earlier work on the subject). However, the study of the magnitude of the
Earth matter effects in the Nadir angle distribution of the ratio of the multi-GeV µ-like and
e-like events, measured in water-Čerenkov detectors, performed here overlaps very little
with the earlier investigations.
2. 3-ν oscillations of atmospheric neutrinos in the Earth
In the present section we summarize the results on the oscillations of atmospheric

neutrinos crossing the Earth, which we use in our analysis.
2.1. Preliminary remarks
The νµ → νe (ν̄µ → ν̄e ) and νe → νµ(τ ) (ν̄e → ν̄µ(τ ) ) oscillations of atmospheric

neutrinos should exist and their effects could be observable if genuine three-flavour-
neutrino mixing takes place in vacuum, i.e., if sin2 2θ13 = 0, and if sin2 2θ13 is sufficiently
large [20] (see also, e.g., [21,22,41,42]). Under the condition | m231 | m221 , which
the neutrino mass squared differences determined by the existing atmospheric and solar
neutrino and KamLAND data satisfy, the relevant three-neutrino νµ → νe (ν̄µ → ν̄e )
and νe → νµ(τ ) (ν̄e → ν̄µ(τ ) ) transition probabilities reduce effectively to a two-neutrino
transition probability [44]. with m231 and sin2 2θ13 = 4|Ue3 |2 (1 − |Ue3 |2 ) playing the
role of the relevant two-neutrino oscillation parameters. Thus, searching for the effects
of νµ → νe (ν̄µ → ν̄e ) and νe → νµ(τ ) (ν̄e → ν̄µ(τ ) ) transitions of atmospheric neutrinos,
amplified by Earth matter effects, can provide unique information, in particular, about the
magnitude of sin2 θ13 .
As is not difficult to show using the results of [44], the 3-neutrino oscillation
probabilities of interest for atmospheric νe,µ having energy E and crossing the Earth along
a trajectory characterized by a Nadir angle θn , have the following form (see also, e.g., [22,
25]):
P3ν (νe → νe ) ∼
= 1 − P2ν , (3)
∼ P3ν (νµ → νe ) ∼
P3ν (νe → νµ ) = 2
= s23 P2ν , (4)
∼ c P2ν ,
P3ν (νe → ντ ) = 2
(5)
23

P3ν (νµ → νµ ) ∼
= − s23
1 4
P2ν − 2c23 s23 1 − Re e−iκ A2ν (ντ → ντ ) ,
2 2
(6)
P3ν (νµ → ντ ) = 1 − P3ν (νµ → νµ ) − P3ν (νµ → νe ). (7)
Here P2ν ≡ P2ν ( m231 , θ13 ; E, θn ) is the probability of two-neutrino oscillations in the
Earth which coincides in form with, e.g., the two-neutrino νe → ντ transition probability,
P2ν (νe → ντ ), but describes νe → ντ transitions, where ντ = s23 νµ + c23 ντ [44], and κ and
A2ν (ντ → ντ ) ≡ A2ν are known phase and two-neutrino transition probability amplitude.
Analytic expressions for P2ν , κ and A2ν will be given later.
Using Eqs. (3)–(7) it is not difficult to convince oneself that the fluxes of atmospheric
νe,µ of energy E, which reach the detector after crossing the Earth along a given trajectory
specified by the value of θn , Φνe,µ (E, θn ), are given by the following expressions in the
case of the three-neutrino oscillations under discussion [21,22]:
2
Φνe (E, θn ) ∼
= Φν0e 1 + s23 r − 1 P2ν , (8)
−1
Φνµ (E, θn ) ∼
= Φνµ 1 + s23 s23 r
0 4 2
− 1 P2ν

− 2c23 s23 1 − Re e−iκ A2ν (ντ → ντ ) ,
2 2
(9)
where Φν0e(µ) = Φν0e(µ) (E, θn ) is the νe(µ) flux in the absence of neutrino oscillations and
Φν0µ (E, θz )
r ≡ r(E, θn ) ≡ . (10)
Φν0e (E, θz )
The interpretation of the SK atmospheric neutrino data in terms of νµ → ντ oscillations
requires the parameter s23 2 to lie approximately in the interval (0.30–0.70), with 0.5 being
the statistically preferred value. For the predicted ratio r(E, θn ) of the atmospheric νµ and
νe fluxes for (i) the Earth core crossing and (ii) only mantle crossing neutrinos, having
trajectories for which 0.4 cos θn 1.0, one has [45–47]: r(E, θz ) ∼ = (2.0–2.5) for the
neutrinos giving contribution to the sub-GeV samples of Super-Kamiokande events, and
r(E, θn ) ∼
= (2.6–4.5) for those giving the main contribution to the multi-GeV samples. If
2 = 0.5 and r(E, θ ) ∼ 2.0, we have (s 2 r(E, θ ) − 1) ∼ 0 and the possible effects of the
s23 z = 23 z =
νµ → νe and νe → νµ(τ ) transitions on the νe and νµ fluxes, and correspondingly on the
sub-GeV e-like sample of events, would be strongly suppressed even if these transitions
were maximally enhanced by the Earth matter effects. For the multi-GeV neutrinos we have
2 r(E, θ ) − 1) 0.3 (0.9) for s 2 = 0.5 (0.7). The factor (s 2 r(E, θ ) − 1), for instance,
(s23 z 23 23 z
amplifies the effect of the νµ → νe transitions in the e-like sample for E (5–6) GeV, for
which r(E, θz ) 4 [45–47].
The same conclusions are valid for the effects of oscillations on the fluxes of, and
event rates due to, atmospheric antineutrinos: ν̄e and ν̄µ . Actually, the formulae for anti-
neutrino fluxes and oscillation probabilities are analogous to those for neutrinos: they can
be obtained formally from Eqs. (3)–(7) by replacing the neutrino related quantities—
probabilities, κ, A2ν (ντ → ντ ) and fluxes, with the corresponding quantities for an-
tineutrinos: P2ν ( m231 , θ13; E, θn ) → P̄2ν ( m231 , θ13; E, θn ), κ → κ̄, A2ν (ντ → ντ ) →
(0) (0)
A2ν (ν̄τ → ν̄τ ) ≡ Ā2ν , P3ν (νl → νl ) → P3ν (ν̄l → ν̄l ), Φνe,µ (E, θn ) → Φν̄e,µ (E, θn ) and
r(E, θn ) → r̄(E, θn ).
Eqs. (3)–(6), (8), (9) and the similar equations for antineutrinos imply that in the case
under study the effects of the νµ → νe , ν̄µ → ν̄e , and νe → νµ(τ ) , ν̄e → ν̄µ(τ ) , oscillations
(i) increase with the increase of s232 and are maximal for the largest allowed value of s 2 ,
23
(ii) should be considerably larger in the multi-GeV samples of events than in the sub-GeV
samples, (iii) in the case of the multi-GeV samples, they lead to an increase of the rate of
e-like events and to a slight decrease of the µ-like event rate. This discussion suggests that
the quantity most sensitive to the effects of the oscillations of interest should be the ratio
of the µ- and e-like multi-GeV events (or event rates), Nµ /Ne .
The magnitude of the effects we are interested in depends also on the 2-neutrino
oscillation probabilities, P2ν ( m231, θ13 ; E, θn ) and P̄2ν ( m231, θ13 ; E, θn ). In the case
of oscillations in vacuum we have P2ν ( m231, θ13 ; E, θn ) = P̄2ν ( m231 , θ13; E, θn ) ∼
sin2 2θ13 . Given the existing limits on sin2 2θ13 , the probabilities P2ν and P̄2ν cannot
be large if the oscillations take place in vacuum. However, P2ν or P̄2ν can be strongly
enhanced by the Earth matter effects. The latter depend on the Earth density profile and we
will discuss it next briefly.
2.2. The Earth model and the two-layer density approximation
As is well-known, the Earth density distribution in the existing Earth models is assumed
to be spherically symmetric3 and there are two major density structures—the core and the
mantle, and a certain number of substructures (shells or layers). The core radius and the
depth of the mantle are known with a rather good precision and these data are incorporated
in the Earth models. According to the Stacey 1977 and the more recent PREM models [48,
49], which are widely used in the calculations of the probabilities of neutrino oscillations in
the Earth, the core has a radius Rc = 3485.7 km, the Earth mantle depth is approximately
Rman = 2885.3 km, and the Earth radius is R⊕ = 6371 km. The mean values of the
3 Let us note that because of the approximate spherical symmetry of the Earth, a given neutrino trajectory
through the Earth is completely specified by its Nadir angle.
matter densities of the core and of the mantle read, respectively: ρ̄c ∼ = 11.5 g/cm3 and
∼
ρ̄man = 4.5 g/cm . 3
All the interesting features of the atmospheric neutrino oscillations in the Earth can be
understood quantitatively in the framework of the two-layer model of the Earth density
distribution [20]. The density profile of the Earth in the two-layer model is assumed to
consist of two structures—the mantle and the core, having different densities, ρman and ρc ,
and different electron fraction numbers, Yeman and Yec , none of which however vary within
a given structure. The densities ρman and ρc in the case of interest should be considered as
mean effective densities along the neutrino trajectories, and they vary somewhat with the
change of the trajectory [20]: ρman = ρ̄man and ρc = ρ̄c . In the PREM and Stacey models
one has for cos θn 0.4: ρ̄man ∼ = (4–5) g/cm3 and ρ̄c ∼= (11–12) g/cm3 . For the electron
fraction numbers in the mantle and in the core 4 one can use the standard values [50]
Yeman = 0.49 and Yec = 0.467. Numerical calculations show [51] (see also [20,27]) that,
e.g., the νe → νµ oscillation probability of interest, calculated within the two-layer model
of the Earth with ρ̄man and ρ̄c for a given neutrino trajectory determined using the
PREM (or Stacey) model, reproduces with a remarkably high precision the corresponding
probability calculated by solving numerically the relevant system of evolution equations
with the much more sophisticated Earth density profile of the PREM (or Stacey) model.
We give below the expressions for the probability P2ν (E, θn ; m231, θ13 ), for the
phase κ and for the amplitude A2ν (ντ → ντ ) in the two-layer approximation of the
Earth density distribution and in the general case of neutrinos crossing the Earth core.
The expression for P2ν (E, θn ; m231, θ13 ), as we have already indicated, coincides with
that for the probability of the two-neutrino νµ → νe (νe → νµ(τ ) ) transitions and has the
form [20]:5

P2ν E, θz ; m231, θ13
1
= [1 − cos E X ] sin2 2θm
2
1
+ [1 − cos E X ][1 − cos E X ] sin2 (2θm − 4θm ) − sin2 2θm
4
1
− [1 − cos E X ][1 − cos 2 E X ] sin2 2θm cos2 (2θm − 2θm )
4
1
+ [1 + cos E X ][1 − cos 2 E X ] sin2 2θm
4
1
+ sin E X sin 2 E X sin2 2θm cos(2θm − 2θm )
2
1
+ cos( E X − E X ) − cos( E X + E X )
4
× sin 4θm sin(2θm − 2θm ). (11)
4 The electron fraction number is given by Y = N /(N + N ) = N /(N + N ), where N , N and N

e e p n p p n e p n
are the electron, proton and neutron number densities in matter, respectively.
5 The expression for P (Eq. (11)) can be obtained from the expression for the probability P = P (ν → ν )
2ν e2 2 e
given in Eq. (7) in [20] by formally setting θ = 0 while keeping θm = 0 and θ = 0.
m
Here

2
m 31 ρ̄man(c) 2
E ( E ) = 1 − res cos2 2θ13 + sin2 2θ13, (12)
2E ρman(c)
is the difference between the energies of the two energy-eigenstate neutrinos in the mantle
(core), θm and θm are the mixing angles in matter in the mantle and in the core, respectively,
sin2 2θ13
sin2 2θm sin2 2θm = ρ̄man(c) 2
, (13)
1− res
ρman(c)
cos 2 2θ + sin2 2θ
13 13
X is half of the distance the neutrino travels in the mantle and X is the length of the path
res and ρ res are the resonance densities in the mantle and in
of the neutrino in the core, ρman c
the core, and ρ̄man and ρ̄c are the mean densities along the neutrino trajectory in the mantle
and in the core for a neutrino trajectory which is specified by a given Nadir angle θn we
have:

X = R⊕ cos θn − Rc2 − R⊕ 2 sin2 θ ,
n X = 2 Rc2 − R⊕2 sin2 θ ,
n (14)
where R⊕ = 6371 km is the Earth radius (in the PREM [49] and Stacey [48] models) and
Rc = 3485.7 km is the core radius. The neutrinos cross the Earth core on the way to the
detector for θn 33.17◦.
The resonance densities in the mantle and in the core can be obtained from the
expressions
m231 cos 2θ13
ρ res = √ mN , (15)
2E 2 GF Ye
mN being the nucleon mass, by using the specific values of Ye in the mantle and in the core.
res = ρ res because Y c = 0.467 and Y man = 0.49 [50] (see also [52]). Obviously,
We have ρman c e e
Ye ρc = Yeman ρman
c res res .
The phase κ and the probability amplitude A2ν which appear in Eq. (9) for the
flux of atmospheric νµ in the case of three-flavour neutrino mixing and hierarchy
between the neutrino mass squared differences and therefore can play important role in
the interpretation of the, e.g., Super-Kamiokande atmospheric neutrino data, have the
following form in the two-layer model of the Earth density distribution [20–22]:

1 m231 √ 1 c
κ∼= X + 2 GF X Ye ρ̄c + 2X Yeman ρ̄man − 2 E X − E X
2 2E mN
m221
− X cos 2θ12, (16)
2E
−i2 E X −i E X 2
A2ν (ντ → ντ ) = 1 + e −1 1+ e − 1 cos (θm − θm ) cos2 θm
1
+ e−i E X − 1 cos2 θm + e−i E X − 1 e−i E X − 1
2
× sin(2θm − 2θm ) sin 2θm , (17)
where6 X = X + 2X .

The expressions for P̄2ν , κ̄ and A2ν (ν̄τ → ν̄τ ) can be obtained from the corresponding
expressions for neutrinos by replacing ρman,c with (−ρman,c ) in Eqs. (12) and (13).
2.3. Oscillations in the Earth mantle
In the two-layer model, the oscillations of atmospheric neutrinos crossing only the Earth
mantle (but not the Earth core), correspond to oscillations in matter with constant density.
The relevant expressions for P2ν , κ and A2ν (ντ → ντ ) follow from Eqs. (11), (16) and (17)
by setting X = 0 and using X = R⊕ cos θn . The expressions for P2ν (P̄2ν ) and A2ν (Ā2ν )
have the standard well-known form.
In the case under study θ13 plays the role of a two-neutrino vacuum mixing angle in the
probabilities P2ν and P̄2ν . Since sin2 θ13 < 0.05, we have cos 2θ13 > 0. Consequently, the
Earth matter effects can resonantly enhance P2ν for m231 > 0 and P̄2ν if m231 < 0 [25].
Due to the difference of cross-sections for neutrinos and antineutrinos, approximately 2/3
of the total rate of the µ-like and e-like multi-GeV atmospheric neutrino events in the
SK (and in any other water-Čerenkov) detector, i.e., ∼ 2Nµ /3 and ∼ 2Ne /3, are due to
neutrinos νµ and νe , respectively, while the remaining ∼ 1/3 of the multi-GeV event rates,
i.e., ∼ Nµ /3 and ∼ Ne /3, are produced by antineutrinos ν̄µ and ν̄e . This implies that the
Earth matter effects in the multi-GeV samples of µ-like and e-like events will be larger if
m231 > 0, i.e., if the neutrino mass spectrum is with normal hierarchy, than if m231 < 0
and the spectrum is with inverted hierarchy. Thus, the ratio Nµ /Ne of the multi-GeV µ-
like and e-like event rates measured in the SK experiment is sensitive, in principle, to the
type of the neutrino mass spectrum.
Consider for definiteness the case of m231 > 0. It follows from Eqs. (8) and (9) that the
oscillation effects of interest will be maximal if P2ν ∼
= 1. The latter is possible provided (i)
the well-known resonance condition [33,34], leading to sin2 2θm ∼ = 1, is fulfilled, and (ii)
cos 2 E X ∼ = −1. Given the values of ρ̄man and Yeman , the first condition determines the
neutrino energy at which P2ν can be enhanced:
−1
Eres ∼
= 6.6 × m231 10−3 eV2 N̄eman NA cm−3 cos 2θ13 GeV, (18)
where m231[10−3 eV2 ] is the value of m231 in units of 10−3 eV2 and N̄eman [NA cm−3 ] is
the electron number density, N̄eman = Yeman ρ̄man /mN , in units of NA cm−3 , NA being the
Avogadro number. If the first condition is satisfied, the second determines the length of the
path of the neutrinos in the mantle for which one can have P2ν ∼ = 1:

2X ( E )res ∼
= 1.2π tan 2θ13N̄eman NA cm−3 2X 104 km , (19)
where X is in units of 104 km. Taking m231 = ∼ (2.1–3.3) × 10−3 eV2 , N̄ man ∼
= 2NA cm−3
e
and cos 2θ13 ∼ = 1 one finds from Eq. (18): Eres ∼ = (7–11) GeV. The width of the
resonance in E, 2δE, is determined, as is well-known, by tan 2θ13 : δE/Eres ∼ tan 2θ13 . For
sin2 θ13 ∼ (0.01–0.05), the resonance is relatively wide in the neutrino energy: δE/Eres ∼=
6 One can get the expression for the amplitude A (ν → ν ) from Eq. (1) in Ref. [20] by formally setting
2ν τ τ
and θ arbitrary, and then interchanging sin θ (sin θ ) and cos θ (cos θ ).
θ = π/2 while keeping θm m m m m m
(0.27–0.40). Eq. (19) implies that for sin2 θ13 = 0.05 (0.025) and N̄eman ∼ = 2NA cm−3 , one
can have P2ν ∼ = 1 only if 2X ∼ 8000 (10000) km.
=
It follows from the above simple analysis [25] that the Earth matter effects can amplify
P2ν significantly when the neutrinos cross only the mantle (i) for E ∼ (5–10) GeV, i.e., in
the multi-GeV range of neutrino energies, and (ii) only for sufficiently long neutrino paths
in the mantle, i.e., for cos θn 0.4. The magnitude of the matter effects in the ratio Nµ /Ne
of interest increases with increasing sin2 θ13 .
The same results, Eqs. (18) and (19), and conclusions are valid for the antineutrino
oscillation probability P̄2ν in the case of m231 < 0. As a consequence, the ideal
situation for distinguishing the type of mass hierarchy would be a detector with charge
discrimination.
2.4. Oscillations of atmospheric neutrinos crossing the Earth core
In this case P2ν , κ and A2ν are given by Eqs. (11), (16) and (17). In the discussion
which follows we will assume for concretness that m231 > 0, consider the probability P2ν
and the transitions of neutrinos. If m231 < 0, the results we will briefly discuss below will
be valid for the probability P̄2ν and the transitions of antineutrinos.
For sin2 θ13 < 0.05 and m231 > 0, we can have P2ν ∼ = 1 only due to the effect of
maximal constructive interference between the amplitudes of the νe → ντ transitions in
the Earth mantle and in the Earth core [20,40,41]. The effect differs from the MSW
one [20]. The mantle-core enhancement effect is caused by the existence (for a given
neutrino trajectory through the Earth core) of points of resonance-like total neutrino
conversion, P2ν = 1, in the corresponding space of neutrino oscillation parameters [40,
41]. The location of these points determines the regions where P2ν is large, P2ν 0.5.
These regions vary slowly with the Nadir angle, they are remarkably wide in the Nadir
angle and are rather wide in the neutrino energy [41], so that the transitions of interest
produce noticeable effects in the ratio Nµ /Ne : we have, e.g., δE/E ∼ = 0.3 for the values of
sin2 θ13 of interest.
The resonance-like total neutrino conversion due to the mantle-core enhancement effect
takes place for a given neutrino trajectory through the Earth core if the following two
conditions are satisfied [40,41]:

1 − cos 2θm
tan E X = ± ,
2 cos(2θm − 4θm )
1 cos 2θm
tan E X = ± , (20)
2 − cos 2θm cos(2θm − 4θm )
where the signs are correlated and cos 2θm cos(2θm − 4θm ) 0. As was shown in [41],
conditions (20) are fulfilled for the νµ → νe and νe → νµ(τ ) transitions of the Earth core-
crossing atmospheric neutrinos. A rather complete set of values of m231/E and sin2 2θ13
for which both conditions in (20) hold and P2ν = 1 for neutrino trajectories with Nadir
angle θn = 0◦ ; 13◦ ; 23◦; 30◦ was also found in [41].
For sin2 θ13 < 0.05, there are two sets of values of m231 and sin2 θ13 for which Eq. (20)
is fulfilled and P2ν = 1. These two solutions of Eq. (20) occur for, e.g., θn = 0◦ ; 13◦; 23◦ ,
at (1) sin2 2θ13 = 0.034; 0.039; 0.051, m231/E = 7.2; 7.0; 6.5 × 10−7 eV2 /MeV, and at
(2) sin2 2θ13 = 0.15; 0.17; 0.22, m231 /E = 4.8; 4.5; 3.8 × 10−7 eV2 /MeV (see Table 2
in Ref. [41]). The first solution corresponds to cos 2 E X = −1, cos E X = −1 and7
and sin2 (2θm − 4θm ) = 1. For m2 = 3 × 10−3 eV2 , the total neutrino conversion occurs
in the case of the first solution at E ∼= (4.2–4.7) GeV. The atmospheric νe and νµ with
these energies contribute to the multi-GeV samples of e-like and µ-like events in the SK
experiment. The values of sin2 2θ13 at which the second solution takes place are marginally
allowed. If m2 = 3 × 10−3 eV2 , one has P2ν = 1 for this solution for a given θn in the
interval 0 θn 23◦ at E lying in the interval E ∼ = (6.3–8.0) GeV.
The above discussion suggests, in particular, that the effects of the mantle-core (NOLR)
enhancement of P2ν (or P̄2ν ) in the ratios Nµ /Ne and Ne3ν /Ne0 increase rapidly with
sin2 θ13 as long as sin2 θ13 0.01, and should exhibit a rather weak dependence on sin2 θ13
for 0.01 sin2 θ13 < 0.05. If 3-neutrino oscillations of atmospheric neutrinos take place,
the magnitude of the matter effects in the multi-GeV e-like and µ-like event samples,
produced by neutrinos crossing the Earth core, should be larger than in the event samples
due to neutrinos crossing only the Earth mantle (but not the core). This is a consequence of
the fact that in the energy range of interest the atmospheric neutrino fluxes decrease rather
rapidly with energy—approximately as E −2.7 , while the neutrino interaction cross-section
rises only linearly with E, and that the maximum of P2ν (or P̄2ν ) due to the NOLR takes
place at approximately two times smaller energies than that due to the MSW effect for
neutrinos crossing only the Earth mantle (e.g., at E ∼ = (4.2–4.7) GeV and E ∼ = 10 GeV,
respectively, for m2 = 3 × 10−3 eV2 ).
3. Results
The results of our analysis are summarized graphically in Figs. 1–5. We have used in the
calculations the predictions for the Nadir angle and energy distributions of the atmospheric
neutrino fluxes given in [47]. In this analysis, we only consider deep inelastic scattering
(DIS) cross-sections and we make use of the GRV94 parton distributions given in [53].
The predicted dependences on cos θn of the ratios of the multi-GeV µ- and e-like events
(or event rates), integrated over the neutrino energy from the interval E = (2.0–10.0) GeV,
in the case (i) of two-neutrino νµ → ντ and ν̄µ → ν̄τ oscillations in vacuum and no
νe and ν̄e oscillations, Nµ2ν /Ne0 , (ii) three-neutrino oscillations in vacuum of νµ , ν̄µ , νe
and ν̄e , (Nµ3ν /Ne3ν )vac , (iii) three-neutrino oscillations of νµ , ν̄µ , νe and ν̄e in the Earth
in the cases of neutrino mass spectrum with normal hierarchy (Nµ3ν /Ne3ν )NH , and with
inverted hierarchy, (Nµ3ν /Ne3ν )IH , for sin2 θ23 = 0.64; 0.50; 0.36, sin2 2θ13 = 0.05; 0.10
and | m231 | = 3 × 10−3 eV2 are shown in Fig. 1.
As cos θn increases from 0 to ∼ 0.2, the neutrino path length in the Earth increases
from 0 to 2X = 2R⊕ cos θn ∼ = 2550 km. The νµ → ντ (ν̄µ → ν̄τ ) oscillations, which for
cos θn 0.2 proceed in the Earth essentially as in vacuum, fully develop. For | m231 | =
7 The term “neutrino oscillation length resonance” (NOLR) was used in [20] to denote the mantle–core
enhancement effect in this case.
Fig. 1. The dependence on cos θn of the ratios of the multi-GeV µ- and e-like events (or event rates), integrated
over the neutrino energy in the interval E = (2.0–10.0) GeV, in the cases (i) of two-neutrino νµ → ντ and
ν̄µ → ν̄τ oscillations in vacuum and no νe and ν̄e oscillations, Nµ2ν /Ne0 (solid lines), (ii) three-neutrino
oscillations in vacuum of νµ , ν̄µ , νe and ν̄e , (Nµ3ν /Ne3ν )vac (dash-dotted lines), (iii) three-neutrino oscillations of
νµ , ν̄µ νe and ν̄e in the Earth and neutrino mass spectrum with normal hierarchy (Nµ3ν /Ne3ν )NH (dashed lines),
or with inverted hierarchy, (Nµ3ν /Ne3ν )IH (dotted lines). The results shown are for | m231 | = 3 × 10−3 eV2 ,
sin2 θ23 = 0.36 (upper panels); 0.50 (middle panels); 0.64 (lower panels), and sin2 2θ13 = 0.05 (left panels); 0.10
(right panels).
3 × 10−3 eV2 and E = 3 GeV, for example, the maximum of the νµ → ντ (ν̄µ → ν̄τ )
oscillation probability occurs for cos θn ∼ = 0.1, or 2X ∼ = 1270 km. For cos θn 0.2
2 −3
and | m31| = 3 × 10 eV , the oscillations involving the atmospheric νe (νe → νµ,τ ,
2
νµ → νe ) and ν̄e (ν̄e → ν̄µ,τ , ν̄µ → ν̄e ) with energies in the multi-GeV range, E ∼
(2.0–10.0) GeV, are suppressed. If m231 > 0, for instance, the Earth matter effects
suppress the antineutrino oscillation probability P̄2ν , but can enhance the neutrino mixing
in matter, or sin2 2θm . However, since the neutrino path in the Earth mantle is relatively
short one has 2X E 1, and correspondingly P2ν 1. Thus, for cos θn 0.2, all four
types of ratios we consider, Nµ2ν /Ne0 , (Nµ3ν /Ne3ν )vac , (Nµ3ν /Ne3ν )NH and (Nµ3ν /Ne3ν )IH ,
practically coincide and exhibit the same dependence on cos θn : they decrease as cos θn
Fig. 2. The same as in Fig. 1 but for the ratios of the µ- and e-like events (or event rates), integrated, respectively,
over the neutrino energy in the intervals E = (4.0–10.0) GeV (left panel), and E = (2.0–100.0) GeV (right
panel), and for | m231 | = 3 × 10−3 eV2 , sin2 θ23 = 0.50 and sin2 2θ13 = 0.10.
increases from 0, reaching a minimum at cos θn ∼ = 0.1, and begin to rise as cos θn increases
further. This behavior is clearly seen in Figs. 1 and 2.
At cos θn 0.4, the Earth matter effects in the oscillations of the atmospheric νµ , ν̄µ ,
νe and ν̄e , can generate noticeable differences between Nµ2ν /Ne0 (or (Nµ3ν /Ne3ν )vac ) and
(Nµ3ν /Ne3ν )NH(IH), as well as between (Nµ3ν /Ne3ν )NH and (Nµ3ν /Ne3ν )IH . For sin2 θ23 =
0.36 and sin2 2θ13 0.05 (upper left panel in Fig. 1), we have at cos θn 0.8
(neutrinos crossing only the Earth mantle): (Nµ3ν /Ne3ν )NH ∼ = (Nµ3ν /Ne3ν )IH ∼
= Nµ2ν /Ne0 ∼
=
(Nµ /Ne )vac . For the Earth-core-crossing atmospheric neutrinos, cos θn 0.84, the
3ν 3ν
mantle-core interference effect (or NOLR) suppresses the ratios (Nµ3ν /Ne3ν )NH,IH with
respect to Nµ2ν /Ne0 (or (Nµ3ν /Ne3ν )vac ): at sin2 2θ13 = 0.10 the relative averaged difference
between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH is8 11%, while the difference between (Nµ3ν /Ne3ν )NH
and (Nµ3ν /Ne3ν )IH is rather small (upper right panel in Fig. 1).
At sin2 θ23 0.50, the differences between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH,IH become
noticeable already at cos θn 0.4. They increase with the increasing of sin2 θ23 and/or
sin2 2θ13 , and are maximal for sin2 θ23 = 0.64 and sin2 2θ13 = 0.10. The dependence
on sin2 θ23 is particularly strong. The deviations from the vacuum oscillation ratio
(Nµ3ν /Ne3ν )vac increase with the increasing of cos θn 0.4 as well; they are maximal for
the Earth-core-crossing neutrinos, cos θn 0.84.
For sin2 θ23 = 0.50 and sin2 2θ13 = 0.05; 0.10, the relative difference between Nµ2ν /Ne0
and (Nµ3ν /Ne3ν )NH in the interval cos θn ∼ = (0.5–0.84) is approximately 11%; 13%; for
sin 2θ23 = 0.64 and sin 2θ13 = 0.05; 0.10, it is 18%; 24%. The same differences are
2 2
larger in the Earth core interval cos θn ∼ = (0.84–1.0) due to the mantle-core enhancement
(NOLR) [20,40,41], reaching on average values of 24%; 27% for sin2 θ23 = 0.50 and
e NH is defined as (1 − (Nµ /Ne )NH /(Nµ /Ne )).

8 The relative difference of, e.g., N 2ν /N 0 and (N 3ν /N 3ν ) 3ν 3ν 2ν 0
µ e µ
Fig. 3. The four different ratios of the multi-GeV µ- and e-like events (or event rates), integrated over the
neutrino energy in the interval E = (2.0–10.0) GeV and over the Nadir angle in the interval corresponding to
0.40 cos θn 1.0, as functions (i) of sin2 2θ13 for | m231 | = 3 × 10−3 eV2 (left panels), and (ii) of | m231 |
for sin2 2θ13 = 0.10 (right panels): Nµ2ν /Ne0 (solid lines), (Nµ3ν /Ne3ν )vac (dash-dotted lines), (Nµ3ν /Ne3ν )NH
(dashed lines) and (Nµ3ν /Ne3ν )IH (dotted lines). The results shown are obtained for sin2 θ23 = 0.36 (upper
panels); 0.50 (middle panels); 0.64 (lower panels).
sin2 2θ13 = 0.05; 0.10, and 35%; 38% for sin2 2θ23 = 0.64 and the same two values of
sin2 2θ13 .
For sin2 θ23 = 0.50, and sin2 2θ13 = 0.05; 0.10, the relative difference between Nµ2ν /Ne0
and the ratio Nµ /Ne in the case of IH neutrino mass spectrum, (Nµ3ν /Ne3ν )IH , in the interval
cos θn ∼
= (0.50–0.84) has a mean value of approximately 6%; 8%, while if sin2 2θ23 = 0.64
it is approximately 10%; 14%. It is larger in the Earth core bin, cos θn ∼ = (0.84–1.0),
reaching approximately 25% for sin2 2θ23 = 0.64 and sin2 2θ13 = 0.10.
The magnitude of the difference between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH(IH) exhibits a
relatively strong dependence on the minimal value of the neutrino energy E from the
integration interval, Emin , and a rather mild dependence on the maximal E in the interval,
Emax . With the increase of Emin , the relative magnitude of the contributions to the energy-
Fig. 4. The ratio Ne3ν /Ne0 of e-like multi-GeV events (or event rates), for νe and ν̄e taking part in 3-neutrino
oscillations in the Earth (Ne3ν ), and νe and ν̄e not taking part in the oscillations (Ne0 ), as a function of cos θn
for sin2 2θ13 = 0.10 (solid lines); 0.05 (dashed lines), and for neutrino mass spectrum with normal hierarchy
(upper solid or dashed lines) and with inverted hierarchy (lower solid or dashed lines). The results shown are for
sin2 θ23 = 0.5 (left panel); 0.64 (right panel), and for | m231 | = 3 × 10−3 eV2 . See text for details.
integrated event rates of interest, coming from the energy interval in which the matter
effects are significant, also increases, leading to larger difference between Nµ2ν /Ne0 and
(Nµ3ν /Ne3ν )NH(IH). This is illustrated in Fig. 2, where the ratios of interest are shown as
functions of cos θn for | m231 | = 3 × 10−3 eV2 , sin2 θ23 = 0.50, sin2 θ13 = 0.10, and (i)
Emin = 4 GeV, Emax = 10 GeV (left panel), and ii) Emin = 2 GeV, Emax = 100 GeV
(right panel). Increasing Emin (Emax ) while keeping Emax (Emin ) intact would lead to
the decreasing (increasing) of the statistics in the samples of µ-like and e-like events of
interest.
As Fig. 2 illustrates, the relative differences between (i) Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH and
(ii) Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )IH , increase noticeably in the interval cos θn ∼ = (0.40–0.65)
with the increase of Emin , being almost constant for Emin = 4 GeV and reaching the values
of approximately 29 and 19%, respectively.9 For cos θn ∼ = (0.84–1.0), the differences
under discussion exhibit a characteristic oscillatory pattern. These differences have a
completely different behavior as functions of cos θn if Emax is increased to 100 GeV,
keeping Emin = 2 GeV (Fig. 2, right panel): they increase approximately linearly with
cos θn , starting from 0 at cos θn ∼ = 0.25, and having at cos θn ∼ = 0.70; 0.84 the values
(i) 17, 21% and (ii) 9, 13%, respectively.
In Fig. 3 we show the predictions for the different ratios of µ-like and e-like event
rates we consider, integrated over the neutrino energy in the interval E = (2–10) GeV and
over cos θn in the interval cos θn = (0.4–1.0), (i) as functions of sin2 2θ13 for | m231| =
3 × 10−3 eV2 (left panels), and (ii) as functions of | m231 | for sin2 2θ13 = 0.10 (right
panel). In each case the results presented are for three values of sin2 θ23 = 0.36; 0.50; 0.64.
The differences between the ratios of the integrated µ- and e-like event rates of interest,
e NH is ∼ 12%.
9 The relative difference between (N 3ν /N 3ν ) and (N 3ν /N 3ν )
µ e IH µ
Fig. 5. The dependence of the ratio Ne3ν /Ne0 (i) on sin2 θ13 for | m231 | = 3 × 10−3 eV2 (left panels), and (ii)
on | m231 | for sin2 θ13 = 0.10 (right panels), in the cases of neutrino mass spectrum with normal hierarchy
(solid lines) and inverted hierarchy (dashed lines). In this figure Ne3ν and Ne0 are the multi-GeV e-like events
(or event rates) for νe and ν̄e taking part in 3-neutrino oscillations in the Earth and νe and ν̄e not taking
part in the oscillations, respectively, integrated over the Nadir angle θn in the intervals corresponding to (a)
0.84 cos θn 1.0—neutrinos crossing the Earth core (upper panels), and (b) 0.40 cos θn 1.0 (lower
panels). The results shown are for sin2 θ23 = 0.36 (lower doubly thick solid or dashed lines); 0.50 (middle thin
solid or dashed lines); 0.64 (upper doubly thick solid or dashed lines).
i.e., (i) between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH and (ii) between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )IH ,
increase rather rapidly as sin2 2θ13 increases from 0 to sin2 2θ13 ∼ = 0.05, while the
increase is slower in the interval sin2 2θ13 ∼ = (0.05–0.10). At sin2 2θ13 ∼ = 0.05, the relative
differences between Nµ2ν /Ne0 and (i) (Nµ3ν /Ne3ν )NH and (ii) (Nµ3ν /Ne3ν )IH in the case
under discussion are respectively 13 and 8% for sin2 θ23 = 0.50; for sin2 θ23 = 0.64 they
are considerably larger, 21 and 13%. If sin2 2θ13 ∼ = 0.10, the same two differences for
sin2 θ23 = 0.50; 0.64 read 17; 28% and 10; 17%, respectively.
It follows from Fig. 3, right panel, that differences between the integrated ratios Nµ2ν /Ne0
(or (Nµ3ν /Ne3ν )vac ) and (Nµ3ν /Ne3ν )NH , and (Nµ3ν /Ne3ν )IH and (Nµ3ν /Ne3ν )NH , are maximal
for values of | m231| lying in the interval (2–3) × 10−3 eV2 , which are favored by the
current atmospheric neutrino data.
In Figs. 4 and 5 we present results just for the multi-GeV e-like event rate. The
Earth matter effects are largest in the oscillations of the atmospheric νe or ν̄e . The
dependence of the ratio Ne3ν /Ne0 , Ne3ν and Ne0 being the numbers of multi-GeV e-like
events (or event rates) predicted in the cases of 3-ν oscillations of νe , ν̄e , and of absence
of oscillations (sin2 θ13 = 0), on cos θn is shown in Fig. 4 for sin2 θ23 = 0.50; 0.64 and for
sin2 θ13 = 0.05; 0.10. The deviation of Ne3ν /Ne0 from 1 would signal that νe and ν̄e take
part in oscillations and that sin2 θ13 = 0. We can have Ne3ν /Ne0 > 1 only if a substantial
fraction of the atmospheric νµ (or ν̄µ ) oscillate into νe (ν̄e ). As Fig. 4 shows, the ratio
Ne3ν /Ne0 increases with the increasing of cos θn and can be significantly greater than 1 for
cos θn 0.4. At cos θn = 0.8, for instance, we have for sin2 θ23 = 0.50 and sin2 2θ13 =
0.05; 0.10 in the case of NH neutrino mass spectrum (Ne3ν /Ne0 )NH ∼ = 1.14; 1.19,
while for sin2 θ23 = 0.64 one finds (Ne3ν /Ne0 )NH ∼ = 1.21; 1.29. For IH neutrino mass
spectrum the ratio of interest, (Ne3ν /Ne0 )IH , is smaller and the corresponding values read:
(Ne3ν /Ne0 )IH ∼
= 1.07; 1.10 and (Ne3ν /Ne0 )IH ∼ = 1.11; 1.15, respectively. For the Earth-
core-crossing neutrinos both (Ne3ν /Ne0 )NH and (Ne3ν /Ne0 )IH are larger due to the NOLR
effect, and for sin2 θ23 = 0.64 and sin2 θ13 = 0.10 reach the values (Ne3ν /Ne0 )NH ∼ = 1.45
and (Ne3ν /Ne0 )IH ∼
= 1.26.
For given sin2 θ23 and sin2 θ13 , the maximum of the ratio of the multi-GeV e-
like event rates of interest Ne3ν and Ne0 , integrated over θn in the intervals corre-
sponding to (a) cos θn = (0.84–1.0) (Earth core bin) and (b) cos θn = (0.40–1.0), oc-
curs again for values of | m231| = (2–3) × 10−3 eV2 (Fig. 5, right panels), favored
by the existing atmospheric neutrino data. For | m231| = 3 × 10−3 eV2 , sin2 θ13 =
(0.05–0.10) and sin2 θ23 = 0.64, the two types of θn -integrated ratios can be as large
as: case (a) (Ne3ν /Ne0 )NH(IH) ∼
= 1.36–1.39 (1.20–1.23), and case (b) (Ne3ν /Ne0 )NH(IH) ∼ =
1.19–1.26 (1.11–1.15) (Fig. 5, left and right panels).
4. Conclusions
In the present article we have studied the possibility to obtain evidences for Earth matter
enhanced 3-neutrino oscillations of the atmospheric neutrinos involving, in particular, the
νe (or ν̄e ), from the analysis of the µ- and e-like multi-GeV event data accumulated by
the SK experiment, or that can be provided by future water-Čerenkov detectors. Such
evidences would give also important quantitative information on the values of sin2 θ13 and
sin2 θ23 and on the sign of m2A = m231 . We have considered 3-neutrino oscillations
of the atmospheric νµ , ν̄µ , νe and ν̄e , assuming that the inequality m2
= m221
| m2A | = | m231| holds, as is suggested by the current solar and atmospheric neutrino
data. Depending on the sign of m231 , the Earth matter effects in this case can enhance
either the νµ → νe and νe → νµ , or the ν̄µ → ν̄e and ν̄µ → ν̄µ transitions if sin2 θ13 = 0.
The effects of the enhancement can be substantial for sin2 θ13 0.01 and sin2 θ23 0.50.
They are largest in the multi-GeV e-like and µ-like samples of events and for atmospheric
neutrinos with relatively large path length in the Earth, crossing deeply the mantle or the
mantle and the core, i.e., for cos θn 0.4, where θn is the Nadir angle characterizing the
neutrino trajectory in the Earth.
As observables which are particularly sensitive to the Earth matter effects, and thus
to the values of sin2 θ13 and sin2 θ23 , and to the sign of m231 , we have considered the
Nadir-angle distributions of the ratios Nµ3ν /Ne3ν and Ne3ν /Ne0 , where Nµ3ν and Ne3ν are the
multi-GeV µ- and e-like numbers of events (or event rates) in the case of 3-ν oscillations
of the atmospheric νe , ν̄e and νµ , ν̄µ , and Ne0 is the number of e-like events in the case of
absence of oscillations (sin2 θ13 = 0). As is well-known, the ratio of the energy and Nadir
angle integrated µ- and e-like events, Nµ /Ne , has been measured with a relatively high
precision by the SK experiment [3].
We have obtained predictions for the Nadir-angle distributions of Nµ3ν /Ne3ν and of
Ne /Ne0 both for neutrino mass spectra with normal ( m231 > 0) and inverted ( m231 < 0)
3ν
hierarchy, (Nµ3ν /Ne3ν )NH , (Nµ3ν /Ne3ν )IH , (Ne3ν /Ne0 )NH and (Ne3ν /Ne0 )IH , and for sin2 θ23 =
0.64; 0.50; 0.36. We compared the latter with the predicted Nadir-angle distributions
(i) of the ratio Nµ /Ne for the case the 3-neutrino oscillations taking place in vacuum,
(Nµ3ν /Ne3ν )vac , and (ii) of the ratio Nµ2ν /Ne0 , the predicted number of µ- and e-like events
in the case of 2-neutrino νµ → ντ and ν̄µ → ν̄τ oscillations of the atmospheric νµ and
ν̄µ , and νe and ν̄e not taking part in the oscillations (sin2 θ13 = 0). The dependence of the
Nadir angle distributions of (Nµ3ν /Ne3ν )NH , (Nµ3ν /Ne3ν )IH , (Nµ3ν /Ne3ν )vac and Nµ2ν /Ne0 on
the minimal and maximal neutrino energies in the energy integration interval has also been
studied. Predictions for these four different types of ratios of the suitably integrated over
θn Nadir angle distributions of the µ- and e-like multi-GeV events were also derived.
Our results are presented graphically in Figs. 1–5. We find that for sin2 θ23 = 0.50 and
sin 2θ13 = 0.05, the relative difference between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH in the interval
2
cos θn ∼ = (0.5–0.84) is approximately 11%. If sin2 2θ23 = 0.64 and sin2 2θ13 = 0.10, the
same difference is approximately 24% (Fig. 1). The relative difference between Nµ2ν /Ne0
and (Nµ3ν /Ne3ν )NH is larger in the Earth core interval cos θn ∼ = (0.84–1.0) due to the
mantle-core enhancement (NOLR) [20,40,41], reaching on average values of 24% for
sin2 θ23 = 0.50 and sin2 2θ13 = 0.05, and of 38% for sin2 2θ23 = 0.64 and sin2 2θ13 = 0.10
(Fig. 1).
In the case of neutrino mass spectrum with inverted hierarchy, the relative difference
between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )IH in the interval cos θn ∼ = (0.50–0.84) has a mean value
of approximately 14%, for sin2 θ23 = 0.64 and sin2 2θ13 = 0.10. It reaches approximately
25% in the Earth core bin, cos θn ∼ = (0.84–1.0), for these values of sin2 2θ23 and sin2 2θ13
(Fig. 1).
The magnitude of the difference between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH(IH) exhibits
a relatively strong dependence on the minimal value of the neutrino energy E from
the integration interval, Emin , and a rather mild dependence on the maximal E in the
integration interval, Emax (Fig. 2). Increasing Emin from 2 GeV to 4 GeV and keeping
Emax = 10 GeV, leads for | m231| = 3 × 10−3 eV2 , sin2 θ23 = 0.50 and sin2 θ13 = 0.10 to
a considerably larger relative difference between Nµ2ν /Ne0 and (Nµ3ν /Ne3ν )NH(IH) in the
interval cos θn ∼ = (0.40–0.65), which reaches approximately 29% (19%) (Fig. 2). This
difference is larger also in the Earth core bin, cos θn ∼ = (0.84–1.0). We have also found
that the differences between the Nadir angle and energy integrated ratios Nµ2ν /Ne0 (or
(Nµ3ν /Ne3ν )vac ) and (Nµ3ν /Ne3ν )NH , and (Nµ3ν /Ne3ν )IH and (Nµ3ν /Ne3ν )NH , are maximal for
values of | m231| lying in the interval (2–3) × 10−3 eV2 , which are favored by the current
atmospheric neutrino data (Fig. 3). The same conclusion is valid for the integrated ratio of
e-like events Ne3ν /Ne0 (Fig. 5).
We have also shown that in the case of the 3-neutrino oscillations of the atmospheric
neutrinos considered, the ratio Ne3ν /Ne0 increases with the increasing of cos θn and can
be significantly greater than 1 for cos θn 0.4. At cos θn = 0.8, for instance, we have
for sin2 θ23 = 0.50 and sin2 2θ13 = 0.10 in the case of NH neutrino mass spectrum
(Ne3ν /Ne0 )NH ∼
= 1.19, while for sin2 θ23 = 0.64 one finds (Ne3ν /Ne0 )NH ∼
= 1.29 (Fig. 4). For
IH neutrino mass spectrum the ratio of interest, (Ne3ν /Ne0 )IH , is smaller: (Ne3ν /Ne0 )IH ∼
=
1.10 and 1.15. For the Earth-core-crossing neutrinos both (Ne3ν /Ne0 )NH and (Ne3ν /Ne0 )IH
are enhanced due to the NOLR effect, and for sin2 θ23 = 0.64 and sin2 θ13 = 0.10 reach the
values (Ne3ν /Ne0 )NH ∼
= 1.45 and (Ne3ν /Ne0 )IH ∼
= 1.26 (Fig. 4).
It follows from our results that the Earth matter effects in the Nadir-angle distribution
of the ratio of the multi-GeV µ- and e-like atmospheric neutrino events (or event rates),
measured in the Super-Kamiokande (or any future water-Čerenkov) experiment, might
be observable if the atmospheric neutrinos, including the νe and ν̄e , take part in 3-
neutrino oscillations and sin2 θ13 and sin2 θ23 are sufficiently large. The observation of
relatively large Earth matter effects (20–30%) at cos θn 0.4 would clearly indicate that
sin2 θ13 0.01, sin2 θ23 0.50, and would suggest that the neutrino mass spectrum is with
normal hierarchy, m231 > 0. However, distinguishing statistically between the neutrino
mass spectrum with normal and inverted hierarchy requires a high precision measurement
of the Nadir angle distribution of the multi-GeV ratio Nµ /Ne and is a rather challenging
task.
Acknowledgements
This work is supported in part by the Italian MIUR and INFN under the programs
“Fenome-nologia delle Interazioni Fondamentali” and “Fisica Astroparticellare”, by the
US National Science Foundation under Grant No. PHY99-07949 (S.T.P.), by the Spanish
Grant FPA2002-00612 of the MCT (J.B. and S.P.-R.) and by the Spanish MECD for a
FPU fellowship (S.P.-R.). S.T.P. would like to thank the organizers of the Program on
“Neutrinos: Data, Cosmos and the Planck Scale” at KITP, Univ. of California at Santa
Barbara, where part of the work on the present study was done, for kind hospitality, and
M. Freund for discussions at the initial stage of this work. S.P.-R. would like to thank the
Elementary Particle Physics Sector of SISSA, Trieste, Italy, for kind hospitality during part
of this study.
References
[1] B.T. Cleveland, et al., Astrophys. J. 496 (1998) 505;

Y. Fukuda, et al., Phys. Rev. Lett. 77 (1996) 1683;
V. Gavrin, Nucl. Phys. B (Proc. Suppl.) 91 (2001) 36;
W. Hampel, et al., Phys. Lett. B 447 (1999) 127;
M. Altmann, et al., Phys. Lett. B 490 (2000) 16.
[2] Super-Kamiokande Collaboration, Y. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5651;
Super-Kamiokande Collaboration, Y. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5656.
[3] Super-Kamiokande Collaboration, M. Shiozawa, talk given at the International Conference on Neutrino
Physics and Astrophysics “Neutrino’02”, Munich, Germany, May 25–30, 2002.
[4] SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 87 (2001) 071301.
[5] SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011302;
SNO Collaboration, Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011301.
[6] KamLAND Collaboration, K. Eguchi, et al., Phys. Rev. Lett. 90 (2003) 021802.
[7] B. Pontecorvo, Chalk River Lab. report PD-205, 1946;
B. Pontecorvo, Zh. Eksp. Teor. Fiz. 53 (1967) 1717.
[8] R. Davis, D.S. Harmer, K.C. Hoffman, Phys. Rev. Lett. 20 (1968) 1205;
R. Davis, D.S. Harmer, K.C. Hoffman, Acta Physica Acad. Sci. Hung. Suppl. 29 (4) (1970) 371;
R. Davis, in: A. Frenkel, G. Marx (Eds.), Proceedings of the “Neutrino’72” International Conference,
Balatonfured, Hungary, June 1972, OMKDK-TECHNOINFORM, Budapest, 1972, p. 5.
[9] S.M. Bilenky, C. Giunti, W. Grimus, Prog. Part. Nucl. Phys. 43 (1999) 1.
[10] 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.
[11] S.M. Bilenky, et al., Phys. Lett. B 94 (1980) 495;
M. Doi, et al., Phys. Lett. B 102 (1981) 323;
J. Bernabéu, P. Pascual, Nucl. Phys. B 228 (1983) 21;
P. Langacker, et al., Nucl. Phys. B 282 (1987) 589.
[12] S.M. Bilenky, S.T. Petcov, Rev. Mod. Phys. 59 (1987) 671.
[13] M. Apollonio, et al., Phys. Lett. B 466 (1999) 415.
[14] F. Boehm, et al., Phys. Rev. Lett. 84 (2000) 3764;
F. Boehm, et al., Phys. Rev. D 62 (2000) 072002.
[15] G.L. Fogli, et al., Phys. Rev. D 67 (2003) 073002.
[16] V. Barger, D. Marfatia, Phys. Lett. B 555 (2003) 144;
M. Maltoni, T. Schwetz, J.W.F. Valle, hep-ph/0212129;
A. Bandyopadhyay, et al., Phys. Lett. B 559 (2003) 121;
J.N. Bahcall, M.C. González-García, C. Peña-Garay, JHEP 0302 (2003) 009;
H. Nunokawa, W.J.C. Teves, R.Z. Funchal, hep-ph/0212202;
P. Aliani, et al., hep-ph/0212212;
P.C. de Holanda, A.Y. Smirnov, JCAP 0302 (2003) 001;
A.B. Balantekin, H. Yuksel, J. Phys. G 29 (2003) 665.
[17] G.L. Fogli, et al., Phys. Rev. D 66 (2002) 093008.
[18] S.M. Bilenky, D. Nicolo, S.T. Petcov, Phys. Lett. B 538 (2002) 77.
[19] A. De Rújula, et al., Nucl. Phys. B 168 (1980) 54.
[20] S.T. Petcov, Phys. Lett. B 434 (1998) 321;
S.T. Petcov, Phys. Lett. B 444 (1998) 584, Erratum.
[21] S.T. Petcov, Nucl. Phys. B (Proc. Suppl.) 77 (1999) 93, hep-ph/9809587;
S.T. Petcov, hep-ph/9811205;
S.T. Petcov, hep-ph/9907216.
[22] M.V. Chizhov, M. Maris, S.T. Petcov, hep-ph/9810501.
[23] A. De Rújula, M.B. Gavela, P. Hernández, Nucl. Phys. B 547 (1999) 21;
V. Barger, et al., Phys. Rev. D 62 (2000) 013004.
[24] M. Freund, et al., Nucl. Phys. B 578 (2000) 27.
[25] M.C. Bañuls, G. Barenboim, J. Bernabéu, Phys. Lett. B 513 (2001) 391.
[26] J. Bernabéu, S. Palomares-Ruiz, hep-ph/0112002;
J. Bernabéu, S. Palomares-Ruiz, Nucl. Phys. B (Proc. Suppl.) 110 (2002) 339, hep-ph/0201090.
[27] P.I. Krastev, S.T. Petcov, Phys. Lett. B 205 (1988) 84.
[28] J. Arafune, J. Sato, Phys. Rev. D 55 (1997) 1653;
J. Bernabéu, in: Proceedings WIN’99, World Scientific, Singapore, 2000, p. 227, hep-ph/9904474.
[29] K. Dick, et al., Nucl. Phys. B 562 (1999) 29.
[30] J. Bernabéu, M.C. Bañuls, Nucl. Phys. B (Proc. Suppl.) 87 (2000) 315, hep-ph/0003299.
[31] S.M. Bilenky, S. Pascoli, S.T. Petcov, Phys. Rev. D 64 (2001) 053010;
S. Pascoli, S.T. Petcov, Phys. Lett. B 544 (2002) 239;
S. Pascoli, S.T. Petcov, W. Rodejohann, Phys. Lett. B 558 (2003) 141.
[32] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369;

L. Wolfenstein, Phys. Rev. D 20 (1979) 2634.
[33] V. Barger, et al., Phys. Rev. D 22 (1980) 2718.
[34] S.P. Mikheyev, A.Yu. Smirnov, Yad. Fiz. 42 (1985) 1441 (in Russian), Sov. J. Nucl. Phys. 42 (1985) 913.
[35] S.M. Bilenky, et al., Phys. Lett. B 465 (1999) 193.
[36] MINOS Collaboration, D. Michael, talk at the International Conference on Neutrino Physics and
Astrophysics “Neutrino’02”, Munich, Germany, May 25–30, 2002.
[37] M. Spiro, Summary talk at the International Conference on Neutrino Physics and Astrophysics “Neutri-
no’02”, Munich, Germany, May 25–30, 2002.
[38] V. Barger, D. Marfatia, K. Whisnant, Phys. Lett. B 560 (2003) 75;
P. Huber, M. Lindner, W. Winter, Nucl. Phys. B 654 (2003) 3.
[39] S.T. Petcov, M. Piai, Phys. Lett. B 533 (2002) 94.
[40] M.V. Chizhov, S.T. Petcov, Phys. Rev. Lett. 83 (1999) 1096.
[41] M.V. Chizhov, S.T. Petcov, Phys. Rev. D 63 (2001) 073003.
[42] J. Bernabéu, S. Palomares-Ruiz, A. Pérez, S.T. Petcov, Phys. Lett. B 531 (2002) 90.
[43] M.C. González-García, M. Maltoni, Eur. Phys. J. C 26 (2003) 417.
[44] S.T. Petcov, Phys. Lett. B 214 (1988) 259.
[45] M. Honda, et al., Phys. Rev. D 52 (1995) 4985.
[46] V. Agraval, et al., Phys. Rev. D 53 (1996) 1314.
[47] G. Fiorentini, V.A. Naumov, F.L. Villante, Phys. Lett. B 578 (2000) 27.
[48] F.D. Stacey, Physics of the Earth, 2nd Edition, Wiley, London, 1977.
[49] A.D. Dziewonski, D.L. Anderson, Phys. Earth Planet. Inter. 25 (1981) 297.
[50] M. Maris, S.T. Petcov, Phys. Rev. D 56 (1997) 7444.
[51] M. Maris, Q.Y. Liu, S.T. Petcov, Study performed in November–December of 1996, unpublished.
[52] R. Jeanloz, Annu. Rev. Earth Planet. Sci. 18 (1990) 356;
C.J. Allègre, et al., Earth Planet. Sci. Lett. 134 (1995) 515;
W.F. McDonough, S.-S. Sun, Chem. Geol. 120 (1995) 223.
[53] M. Gluck, E. Reya, A. Vogt, Z. Phys. C 67 (1995) 433.
[54] J.F. Beacom, S. Palomares-Ruiz, Phys. Rev. D 67 (2003) 093001, hep-ph/0301060.
Popping out the Higgs boson off vacuum at Tevatron

and LHC
M. Boonekamp a,b , R. Peschanski c , C. Royon b
a CERN, CH-1211, Geneva 23, Switzerland
b CEA/DSM/DAPNIA/SPP, CE-Saclay, F-91191 Gif-sur-Yvette cedex, France
c CEA/DSM/SPhT, Unité de recherche associée au CNRS, CE-Saclay, F-91191 Gif-sur-Yvette cedex, France
Received 30 January 2003; accepted 16 June 2003
Abstract
In the prospect of diffractive Higgs production at the LHC collider, we give an extensive study
of Higgs boson, dijet, diphoton and dilepton production at hadronic colliders via diffraction at
both hadron vertices. Our model, based on non-factorizable pomeron exchange, describes well the
observed dijet rate observed at Tevatron run I. Taking the absolute normalization from data, our
predictions are given for diffractive processes at Tevatron and LHC. Stringent tests of our model and
of its parameters using data being taken now at Tevatron run II are suggested. These measurements
will also allow to discriminate between various models and finally to give precise predictions on
diffractive Higgs boson production cross-section at the LHC.
PACS: 12.38.-t; 13.85.Qk
1. Double diffractive hard production through non-factorizable vacuum exchange
The discovery of the Higgs boson is one of the main goals of searches at the present
and next hadronic colliders, the Tevatron and the LHC. The standard, non-diffractive
production mechanisms are being studied extensively. The main decay modes of a low-
mass Higgs boson are b-quark pairs and τ -leptons, which are difficult to extract from
the standard model background processes. A promising channel is the γ γ decay mode;
however, due to the small branching fraction (about 10−3 ), a luminosity of the order of
50 fb−1 is needed to establish the existence of the Higgs boson in the mass range below
E-mail address: pesch@spht.saclay.cea.fr (R. Peschanski).
doi:10.1016/S0550-3213(03)00535-2
278 M. Boonekamp et al. / Nuclear Physics B 669 (2003) 277–305
135 GeV. It is thus important to investigate other, complementary ways to produce and
detect the Higgs boson in this mass range.
One promising production mode, the exclusive double diffractive production, was
proposed some time ago in Refs. [1,2]. In this case, the Higgs boson is diffractively
produced in the central region resulting in a final state composed of the two protons
scattered at very small angles and detected in the roman pot detectors, the decay products
of the Higgs boson in the main detector, and nothing else. It is thus a very clean signal. The
kinematic constraints coming from the proton detection in the roman pot detectors allow
a very precise determination of the Higgs boson mass [3], hence improving the signal
to background ratio. Contrary to non-diffractive production, the main Higgs boson decay
modes, like b b̄ or τ τ are thus promising channels. However, the exclusive cross-sections
may be very low, and thus put strong limitations to the potentialities of double diffractive
production.
Recently, we have studied the possibility of producing the Higgs boson together with
other particles, i.e., inclusive diffractive production [4]. The expected cross-sections are
increased compared to exclusive production, and the model can be “calibrated” using the
diffractive dijet production measured by the CDF Collaboration at Tevatron run I [5]. The
experimental result, and in particular the dijet mass fraction spectrum, shows that hadronic
activity in the central region (coming from hard QCD radiation, and from soft pomeron
remnants) needs to be accounted for, in order to describe these data. We have proposed
a model for double diffractive production of heavy “objects”, based on a non-factorizable
pomeron model and able to describe the observed features of dijet production data in a
qualitative way. Normalizing our raw predictions to the CDF measurement allows to make
quantitative predictions for the Tevatron and the LHC, given specific assumptions for the
model parameters. We have also shown [6] that it is possible to reconstruct precisely the
mass of the Higgs boson if both protons in the final state can be detected with roman
pot detectors and if the pomeron remnants can be measured in the forward region with
sufficient resolution.
At the Tevatron collider, the Higgs boson production cross-section is severely limited
by the small available phase space. At the LHC, since the beam energy is much
higher, diffractive Higgs boson production has larger cross-section, and might thus be
an interesting channel, as we shall discuss further on. On the other hand, even if double
diffractive Higgs boson production at the Tevatron is probably too small by itself, it is
however possible to verify and constrain models of double diffraction at the Tevatron run II,
through the study of difermion production.
In this paper, we wish to provide an extensive study of our model, including predictions
for dijet, diphoton and dilepton production and their ratios for both the Tevatron run II
and the LHC. We will discuss possible values of its characteristic parameters, and propose
means to discriminate between the existing models at the Tevatron. In turn, this study will
help to formulate more precise predictions for double diffractive Higgs boson production
at the LHC in the near future.
The plan of our study is the following. In Section 2, we will recall and discuss the
original exclusive model of Bialas and Landshoff [1], and our extension to inclusive
production. In Section 3 predictions are given for dijet, diphoton, dilepton and Higgs
boson cross-sections. Dijet results for the Tevatron run I, using a fast simulation of
M. Boonekamp et al. / Nuclear Physics B 669 (2003) 277–305 279
detector effects, are in good agreement with available data and are used to normalize our
predictions. A discussion of pomeron remnants and their possible detection is also given.
In Section 4, ways to verify our model and determine its parameters using Tevatron data are
proposed. The possibility of distinguishing our model from the existing models of double
diffraction is discussed in Section 5. Section 6 summarizes the interest of diffractive Higgs
boson production compared to standard production. Finally, Section 7 gives conclusions
and an outlook on the promising future studies on hard diffraction production at the
Tevatron and the LHC.
2. Inclusive vs. exclusive production
2.1. Exclusive production
Let us first introduce the original model of [1] describing exclusive Higgs boson and q q̄
production in double diffractive production (noted DPE1 in the following). This process is
depicted in Fig. 1, “exclusive” case.
In [1], the diffractive mechanism is based on two-gluon exchange between the two
incoming protons. The soft pomeron is seen as a pair of gluons non-perturbatively coupled
to the proton. One of the gluons is then coupled perturbatively to the hard process (either
the Higgs boson, or the q q̄ pair, see Fig. 1), while the other one plays the role of a
soft screening of color, allowing for diffraction to occur. This soft character requires
the phenomenological introduction of a distinctive non-perturbative gluon propagator [7]
whose parameters are constrained by the description of total cross-sections using the same
formalism.
g g
The hard gluons, carrying all remaining momentum (x1 = x2 = 1), fuse to produce the
heavy object (Higgs boson and diquarks in the original model). The corresponding cross-
sections for q q̄ and Higgs boson production read:
2
s
dσqexc
q̄ (s) = C exc
q q̄ δ (2)
(v i + k i )
Mq2q̄ i=1,2
2α v 2
× d 2 vi d 2 ki dξi dηi ξi i exp −2λJ J vi2 σqexc
q̄ ,
i=1,2
2
s 2 MH
dσHexc (s) = CH 2
δ ξ1 ξ2 −
MH s

dξi 2α vi2
× d 2 vi ξi exp −2λH vi2 . (1)
1 − ξi
i=1,2
1 We keep the new standard notation DPE (i.e., double pomeron exchange) but the model we describe, together
with its extension to inclusive production, considers a non-factorizable soft pomeron exchange with one common
exchanged gluon, while a factorizable double pomeron mechanism implies two different pairs of gluons (see
Section 4).
g
Fig. 1. Double diffractive vs. standard gluon fusion production schemes. xi are the momentum fractions of the
fusing gluons, ki are the transverse 2-momenta of the outgoing difermion in the central rapidity region. The
“standard” Higgs boson production is displayed for reference. “Exclusive” and “inclusive” double diffraction are
represented in the framework of the non-factorizable pomeron model (see formulae (1), (4)–(7) and text for the
complete kinematical notations). The hatched region represents the diffractive interactions at both (anti)proton
g
vertices, while the vertical thin line is for the soft gluon exchange in the model. Note that xi ≡ 1 in the “exclusive”
case.
The variables vi and ki respectively denote the transverse momenta of the outgoing
protons and quarks, ξi are the proton fractional momentum losses (restricted to be smaller
than 10%2), and ηi are the quark rapidities. σH is the gluon-initiated Higgs boson
production cross-section while σ̄qexc
q̄ is the hard q q̄ production cross-section in the exclusive
case. Indeed, in this exclusive process submitted to the JZ = 0 constraint [8]3 (a helicity
selection rule for the production of a scalar system from the vacuum channel), the q q̄
2 This value for ξ 0.1 comes from the usual cut on rapidity gap to be 2.
3 Note that the exclusive model of formula (1) is related to the soft pomeron exchange with a low value of the
intercept, by contrast with the model of Ref. [9].
differential cross-section writes,4 apart from normalizations included in Cq q̄ :

π dσ ρ(1 − ρ) 4m2Q
q̄ ≡
σqexc = 2 2 , ρ= . (2)
24 dt mT 1 mT 2 Mq2q̄
In the model, the non-perturbative input is naturally related to the soft pomeron
trajectory taken from the standard Donnachie–Landshoff parametrization [10], namely
α(t) = 1 + + α t, with ≈ 0.08 and α ≈ 0.25 GeV−2 .
There are other parameters to be fixed, coming from the non-perturbative gluon
propagators. Phenomenological constraints are obtained from the physical values of the
total cross-sections, leading to four unknown parameters in formula (1), namely the
normalizations Cq q̄ , CH and the slopes in momentum transfer λq q̄ , λH . At this point of
our study, the slopes λq q̄ , λH are kept as in the original papers [1].5
The problem of the normalization constants Cq q̄ , CH requires more care, since the
evaluation of cross-sections is the main subject of the paper. All constants of the problem
(non-perturbative proton–gluon coupling, normalization of the non-perturbative gluon
propagators, color factors, perturbative Higgs boson and q q̄ production vertices) are
contained in the normalizations Cq q̄ and CH . The non-perturbative factors are poorly
determined by the current data, implying large uncertainties on production cross-section
predictions. In particular, even if every other parameter is fixed, the non-perturbative
proton–gluon coupling G remains undetermined and is arbitrarily fixed to G2 /4π = 1
in the original publications [1]. However, and it is an important aspect of our model
studies, the ratio CH /Cq q̄ is well defined, and independent of everything else than the color
structure of the process. This means that the ratio of Higgs boson to diquark production is
fixed by this factor. Given the expressions of CH and Cq q̄ [1], one finds
√
CH 2 GF
= , (3)
Cq q̄ 3π
where enter only in (3) the Fermi constant GF , and the ratio of color factors to produce
either a color singlet q q̄ or the scalar Higgs boson in the non-factorizable pomeron model
(see for details, the second reference of [1]). This feature will remain valid in our model
of inclusive production, suggesting that the known, large cross-section dijet production
process can be used to calibrate Higgs boson production.
Also note that, as can be seen in (2), the exclusive production rate for a given quark
flavor is proportional to its mass squared, so that light quark production is expected to be
negligible, reflecting in another way the JZ = 0 constraint [8].
2.2. Inclusive production
The inclusive mechanism is described in the third graph of Fig. 1. The idea is to take into
account that a pomeron is a composite system, made itself from quarks and gluons. In our
4 Here, as in the rest of the paper, the hard differential cross-sections σ are normalized to the usual dσ/dt
expressions through σ = 24π dσ .
dt
5 They are constrained by using (an approximate parametrization of ) the nucleon form factor as the pomeron
coupling to proton [7].
model, we thus apply the concept of pomeron structure functions to compute the inclusive
diffractive Higgs boson cross-section. The H1 measurement of the diffractive structure
function [11] and the corresponding quark and gluon densities are used for this purpose.
This implies the existence of pomeron remnants and QCD radiation, as is the case for the
proton. This assumption comes from QCD factorization of hard processes. However, and
this is also an important issue, we do not assume Regge factorization at the proton vertices,
i.e., we do not use the H1 pomeron flux factors in the proton or antiproton.
Regge factorization is known to be violated between HERA and the Tevatron. Moreover,
we want to use the same physical idea as in the exclusive model [1], namely that a non
perturbative gluon exchange describes the soft interaction between the incident particles,
as in Fig. 1. In practice, the Regge factorization breaking appears in three ways in our
model.
(i) We keep as in the original model of Ref. [1] the soft pomeron trajectory with an
intercept value of 1.08.
(ii) We normalize our predictions to the CDF run I measurements, allowing for
factorization breaking of the pomeron flux factors in the normalisation between the
HERA and hadron colliders.6
(iii) The color factor (3) derives from the non-factorizable character of the model, since
it stems from the gluon exchange between the incident hadrons. We will see later the
difference between this and the factorizable case.
The formulae for the inclusive production processes considered here follow. We have,
for dijet production,7 considering only the dominant gluon-initiated hard processes:
g g 2
x1 x2 s
J = CJ J
dσJincl δ (2) vi + ki
MJ2 J i=1,2
2
× dξi dηi d 2 vi d 2 ki ξi 2α vi exp −2vi2 λJ J
i=1,2
g g
× σJ J GP x1 , µ GP x2 , µ (4)
and for Higgs boson production:

g g 2 2
x1 x2 s MH
dσHincl = CH 2
δ ξ1 ξ2 − g g
MH x1 x2 s

g g 2 dξi 2α vi2

× GP xi , µ dxi d vi ξi exp −2vi λH .
2
(5)
1 − ξi
i=1,2
6 Indeed, recent results from a QCD fit to the diffractive structure function in H1 [12] show that the discrepancy
between the gluonic content of the pomeron at HERA and Tevatron [13] appears mainly in normalization.
7 We call “dijets” the produced quark and gluon pairs.
For the two following processes, the quark-initiated contribution cannot be ignored. We
have, for dilepton production:
q q 2
x x s
dσllincl = Cll 1 22 σq q̄→ll δ (2) vi + ki
Mll i=1,2
g 2
× QP xi , µ dξi dηi d 2 vi d 2 ki ξi 2α vi exp −2vi2 λll (6)
i=1,2
and for diphoton production:

g g 2
x1 x2 s
dσγincl
γ = Cll δ (2)
vi + k i
Mγ2 γ
i=1,2
2
× dξi dηi d 2 vi d 2 ki ξi 2α vi exp −2vi2 λγ γ
i=1,2
g g g g
× σgg→γ γ GP x1 , µ GP x2 , µ + σq q̄→γ γ QP x1 , µ QP x2 , µ . (7)
In the above, the GP (respectively QP ) are the pomeron gluon (respectively quark)
g q
densities, and xi (respectively xi ) are the pomeron’s momentum fractions carried by
the gluons (respectively quarks) involved in the hard process, see Fig. 1. We use as
parametrizations of the pomeron structure functions the fits to the diffractive HERA data
performed in [14]. The dijet cross-section8 is now (summing over quark flavors f , and now
including the contribution from gluon jets):

σJ J = σgg→q q̄ ρ f + 108σgg→gg ρ g ,
f
f f
4mT 1 mT 2 4pT 1 pT 2
ρf = , ρg = ,
MJ2 J MJ2 J

ρf ρf 9ρ f
σgg→q q̄ = f f
1− 1− ,
(mT 1 )2 (mT 2 )2 2 16

1 ρg 3
σgg→gg = 2 2 1 − (8)
pT 1 pT 2 4
to be compared with (2). The above formulae are derived using [16], and the dilepton
and diphoton cross-sections are taken from [17,18]. The expressions for σgg→q q̄ , σgg→gg ,
σq q̄→ll , and σgg→γ γ , σq q̄→γ γ in terms of the Mandelstam variables are recalled in
Appendix A.
In the inclusive case,contrary to the exclusive case, dijet production is flavour
democratic and thus the f in (8) extends over all flavors except for the too massive
top quark, due to kinematics. Note that the non perturbative parameters are kept the
same as in the exclusive case. Indeed, the expressions (1) can be recovered in the limit
8 The formulae (8) are corrected for a factor 2 error coming from a known misprint in the normalization of
σff in [15].
GP → δ(x − 1), and substituting back Eq. (2) instead of (8) for the hard cross-section,
which restricts to the JZ = 0 component of the q q̄ cross-section and reintroduces the
flavor mass hierarchy. The normalization of GP is not determined by the HERA data
(since it is mixed with the flux factors) but is fixed in the model in order to lead to the
same result for the energy–momentum sum rule as for the exclusive case. It is interesting
to note that the comparison with the observed dijet production rates will give the correct
order of magnitude for the inclusive model, while the exclusive one leads to too important
rates,9 as discussed in the next section. Note that the normalization cancels in the ratio
(CH /CJ J ), with the same value as in the exclusive case, see (3).
3. Model predictions for DPE
3.1. DPE dijets at the Tevatron run I
The CDF measurement of double diffractive dijet production [5] is used as a verification
of the validity of our approach (namely concerning the inclusive picture we consider,
and the application of structure functions measured in electron–proton collisions in our
context). Once the model validity tested, the measured cross-section will allow us to fix
(within experimental errors) the absolute normalization of the cross-sections.
In the CDF measurement, an outgoing antiproton is measured on one side of the
detector, and the DPE nature of the events is ensured by requiring a rapidity gap on the
opposite side. The selection then requires at least two jets satisfying a transverse energy
criterion. The details of the selections can be found in [5]. The measured cross-section is
43.6 ± 4.4(stat) ± 21.6(syst) nb, with large error bars.
Reproducing the experimental selections on the cross-section estimates, and keeping
fixed all parameters as in the exclusive case we obtain a raw prediction of 11.4 nb, i.e.,
a factor 3.8 smaller than the measured mean value.10 This prediction includes a fast sim-
ulation of detector resolution effects, using SHW [19]. Considering the large uncertainties,
this result is quite encouraging. Indeed, as mentioned above, the experimental errors are
yet quite large.
In order to verify the dynamics of the model, it is interesting to consider the dijet mass
fraction, defined as the ratio of the mass measured in the central detector to the missing
mass to the outgoing proton and antiproton. For exclusive events, this ratio is expected to
be about 0.8 [5] given detector inefficiencies; if inclusive events dominate, and the model is
correct, one should observe a broad distribution below one, essentially given by the product
9 One may think naively that this overestimation using parameters from soft hadronic cross-sections might
be an argument against the non-perturbative gluon model. This argument does not hold against the inclusive
predictions which gives already the correct order of magnitude.
10 At a theoretical level, it is interesting to note that a mere reduction of the non-perturbative proton–gluon
coupling (arbitrarily fixed in original papers [1]) from G2 /4π = 1 → 1/2 swallows the normalization factor.
Fig. 2. Observed dijet mass fraction (CDF run I), compared to our model prediction, using the pomeron structure
functions.
of the pomeron structure functions. Fig. 2 displays the mass fraction as measured by CDF,
and the prediction of the present model.11
Reasonable agreement is observed, suggesting that the HERA pomeron structure
functions allow for a correct description of inclusive DPE events. It is worthwhile to
emphasize the strong influence of the behavior of the (hard) gluon distribution of the
pomeron on the predicted mass fraction. In Fig. 3, we consider a “proton-like” (i.e., soft)
gluon distribution in the pomeron, leading to unsatisfactory results.
3.2. Model predictions for Tevatron and LHC: dijets, diphotons and dileptons
Assuming a global normalization as determined here and all other model parameters as
given above, one is now in situation to give predictions for Higgs boson production at the
Tevatron and the LHC. In a first stage, we assume no evolution of this normalization with
energy. We will discuss later on the influence of the various parameters on the predictions,
11 Recall that in our predictions, we only included the gluon structure functions in the pomeron (G ),
P
neglecting the quark structure function. This assumption is justified by the weakness of the quark structure
function (10%), and the large error bar on the gluon structure function of about 50% [14].
Fig. 3. Dijet mass fraction obtained using the gluon structure function from the proton.
analyze the (large) uncertainties which still affect the results and, above all, what can be
done in the future to handle them, both on experimental and theoretical grounds. For this
sake, after scaling our model to the CDF Tevatron run I predictions, we give predictions
for various DPE processes which are relevant for Tevatron run II and LHC measurements.
In Figs. 4 and 5, we give the integrated cross sections for dijets, quarks, bb̄, dileptons
and diphotons at generator level for a mass higher than the mass m given on the abscissa for
the Tevatron and the LHC. Each figure is displayed for two different mass ranges, namely
between 10 and 100 GeV, and 100 and 160 GeV.
The differential cross-sections for dijet, dilepton and diphoton production are also given
in Figs. 6 and 7. The dijet cross-section is dominated by the gluon contribution. The quark
jet contribution amounts 1% at small masses, and goes down to 0.1% at higher masses,
both at the Tevatron and the LHC. The bb̄ contribution, which is about 20% of the total
quark contribution, represents only about 0.02% of the dijet yield. The diphoton cross-
section is small (4 orders of magnitude below the bb̄ cross-section both at Tevatron and
LHC) but still measurable at Tevatron at low masses to test the model. These processes are
much smaller than dijet production (due to the weak QED coupling constant, and to the
small quark component of the pomeron which initiates these processes), but they do have
appreciable advantages which will appear in the following discussion. The dilepton cross-
section is small (same order of magnitude as the diphoton) but enhanced by the presence
of the Z pole (see Fig. 7). Numerical values for the cross-sections are also given in Table 1.
Table 1
DPE cross-section (pb) for dijets, diquarks, bb̄, diphotons and dileptons (fb). (1) At the Tevatron, |y| < 5,
m > 40 GeV; (2) At the Tevatron, |y| < 5, m > 120 GeV; (3) At the LHC, |y| < 5, m > 40 GeV; (4) At the
LHC, |y| < 5, m > 120 GeV
Process (1) (2) (3) (4)
Dijets 7.0 × 105 3.2 × 103 1.2 × 107 2.5 × 106
Diquarks 3.8 × 103 4.4 3.6 × 104 1.7 × 103
bb̄ 7.6 × 102 8.7 × 10−1 9.4 × 103 4.5 × 102
γγ 3.5 × 10−2 4.9 × 10−5 4.4 × 10−1 1.9 × 10−2
l+ l− 4.6 × 10−2 6.8 × 10−5 1.1 1.3 × 10−2
Fig. 4. Dijet, diquark, bb̄, dilepton, and diphoton cross-sections (pb) at the Tevatron. The cross-sections are given
for a mass m above the value on the abscissa, for two different mass ranges. For comparison, we also display the
Higgs boson cross-section for MHiggs = m.
Fig. 5. Dijet, diquark, bb̄, dilepton, and diphoton cross-sections (pb) at the LHC. The cross-sections are given for
a mass m above the value on the abscissa, for two different mass ranges. For comparison, we also display the
Higgs boson cross-section for MHiggs = m.
3.3. Model predictions for the Tevatron and the LHC: Higgs boson production
Our results for Higgs boson production at the Tevatron and the LHC are displayed in
Figs. 8, 9 and 10, and in Tables 2 and 3. In Fig. 8, we see that the Higgs production cross-
section at the LHC is higher by more than two orders of magnitude than at the Tevatron.
In the same figure is also displayed for comparison the cross-section for standard Higgs
production, which is more than two orders of magnitude larger. In Figs. 9 and 10, we give
the Higgs boson production cross section at the Tevatron and the LHC for the different
decay modes of the Higgs boson.
The diffractive Higgs production at the LHC is mainly interesting for the lower Higgs
boson masses. When the Higgs boson mass is heavy enough, the W W and ZZ (∗) decay
modes become dominant and the visibility of the corresponding channels are already very
Fig. 6. Differential dijet production cross-section (pb) at the Tevatron and the LHC. The transverse energy of the
central jets satisfies ET > 10 GeV, and their rapidity is limited to |y| < 4.
Fig. 7. Differential diphoton and dilepton production cross-sections (fb) at the Tevatron and the LHC. The
dilepton cross-section corresponds to a single lepton flavor. The transverse energy of the central particles satisfies
ET > 10 GeV, and their rapidity is limited to |y| < 4.
good in standard non-diffractive events; hence we do not expect the contribution from
diffractive channels to be as important there.
At lower Higgs masses, the standard non-diffractive searches are done using the γ γ
decay mode, which is loop-mediated and has very small branching fraction. In the present
Table 2
DPE Higgs production cross-section at the Tevatron (fb). (1) Total cross-section; (2) bb̄ channel; (3) τ τ channel;
(4) W + W − channel
MH (1) (2) (3) (4)
100 3.8 3.2 0.3 0.0
110 2.3 1.8 0.2 0.0
120 1.3 0.9 0.1 0.1
130 0.7 0.4 0.0 0.2
140 0.3 0.1 0.0 0.2
Fig. 8. Higgs boson production cross-section at the LHC and the Tevatron. The upper plot gives the cross-sections
at the Tevatron and the LHC for low Higgs mass to show the difference between both accelerators. The bottom
plots show the same distribution at higher Higgs masses (the standard Higgs production cross-section is shown
for comparison).
Fig. 9. Higgs boson production cross-section at the Tevatron. Various decay channels are plotted as a function of
the Higgs boson mass.
Table 3
DPE Higgs production cross-section at the LHC (fb). (1) Total cross-section; (2) bb̄ channel; (3) τ τ channel;
(4) W + W − channel
MH (1) (2) (3) (4)
100 182.3 152.1 12.4 1.5
110 172.6 138.2 11.4 6.4
120 158.5 114.3 9.6 18.1
130 147.0 85.2 7.2 37.6
140 137.7 54.3 4.6 61.6
150 127.5 26.8 2.3 83.4
160 122.5 6.2 0.5 109.0
170 115.3 1.3 0.1 110.8
180 108.9 0.8 0.1 101.4
190 103.8 0.5 0.0 81.3
200 98.1 0.3 0.0 72.5
case, the good mass resolution obtained using roman pots or microstation detectors [20]
allows to use the Higgs boson decays into b-quarks and into τ -leptons, which are the two
main decay modes when the Higgs boson is lighter than ∼ 140 GeV/c2 . We will discuss
this point in more detail later in this paper.
Fig. 10. Higgs boson production cross-section at the LHC. Various decay channels are plotted as a function of the
Higgs boson mass.
4. Parameter dependence of DPE cross-sections
In this section, we study the dependence of our predictions for dijet, diphoton, dilepton
and Higgs boson cross-sections on the different parameters of the model. If we refer to the
formulae given in the first section, the relevant parameters of our models are the momentum
transfer slopes λi (where i stands for H , J J , ll, qq, γ γ ), and and α respectively the
intercept and slope of the pomeron Regge trajectory. We will discuss the dependence on
in the next section (note that this value is however imposed in our model, but changes in
other pomeron models, as discussed later on).
The dependence on λH , λi (where i stands for ll, J J , bb̄, qq, and γ γ ) is displayed in
Tables 4 (for Tevatron) and 5 (for LHC). In our model, we assumed λH = 2 and λi = 3 (as
in the original exclusive model [1]). We now vary these values by one unity, keeping values
(assuming or not the equality of the λ’s) around those physically connected to the nucleon
form factors. We also give numbers with the pomeron slope α taken to be 0.1. We note as
expected that the cross-sections for dilepton, diphoton, dijet and bb̄ production vary by the
same factor when we vary the parameters.
The results of Tables 4 and 5 are shown for each value of the λ’s (except the reference
one on the first row) on two lines: on the first line, we give the cross-section values as they
Table 4
Numerical values of the cross-sections (fb) for different values of the parameters, at the Tevatron (the subscript
i stands for ll, bb̄, J J , γ γ ). The Higgs cross-section is for a Higgs mass of 120 GeV. All other cross-sections
are given for a dijet, dilepton, or diphoton mass greater than 10 GeV. The first lines read the direct output of
the program, whereas the second lines are rescaled to the same value of the dijet cross-section, namely the CDF
measurement
Param. σdijets σbb̄ σγ γ σll σHiggs
λH = 2, λi = 3 7.2 × 106 2.5 × 104 1.6 1.6 × 102 1.2
λH = 3, λi = 4 4.7 × 106 1.6 × 104 1.0 1.2 × 102 0.7
7.2 × 106 2.4 × 104 1.5 1.8 × 102 1.1
λH = 2, λi = 2 1.3 × 107 4.5 × 104 2.8 3.3 × 102 1.2
7.2 × 106 2.5 × 104 1.5 1.8 × 102 0.7
λH = 4, λi = 4 4.7 × 106 1.6 × 104 1.0 1.2 × 102 0.42
7.2 × 106 2.4 × 104 1.5 1.8 × 102 0.64
λH = 1, λi = 2 1.3 × 107 4.5 × 104 2.8 3.3 × 102 3.5
7.2 × 106 2.5 × 104 1.5 1.8 × 102 1.9
α = 0.1 9.8 × 106 3.4 × 104 2.2 2.5 × 102 1.8
7.2 × 106 2.5 × 104 1.6 1.8 × 102 1.3
Table 5
Numerical values of the cross-sections (fb) for different values of the parameters at the LHC (i means ll, bb̄,
J J , γ γ ). The Higgs cross-section is for a Higgs mass of 120 GeV. All other cross-sections are given for a dijet,
dilepton, or diphoton mass greater than 10 GeV. The first lines read the direct output of the program, whereas the
second lines are rescaled to the same value of the dijet cross-section, namely the CDF measurement
Param. σdijets σbb̄ σγ γ σll σHiggs
λH = 2, λi = 3 4.2 × 107 1.1 × 105 8.8 × 103 1.6 × 103 1.6 × 102
λH = 3, λi = 4 2.7 × 107 7.5 × 104 5.5 × 103 1.0 × 103 9.0 × 101
4.2 × 107 1.2 × 105 8.6 × 103 1.6 × 103 1.4 × 102
λH = 2, λi = 2 7.1 × 107 1.9 × 105 1.6 × 104 2.8 × 103 1.6 × 102
4.2 × 107 1.1 × 105 9.5 × 103 1.7 × 103 9.5 × 101
λH = 4, λi = 4 2.7 × 107 7.5 × 104 5.5 × 103 1.0 × 103 5.7 × 101
4.2 × 107 1.2 × 105 8.6 × 103 1.6 × 103 8.9 × 101
λH = 1, λi = 2 7.1 × 107 1.9 × 105 1.6 × 104 2.8 × 103 3.8 × 102
4.2 × 107 1.1 × 105 9.5 × 103 1.7 × 103 2.2 × 102
α = 0.1 6.1 × 107 1.7 × 105 1.5 × 104 2.3 × 103 2.5 × 102
4.2 × 107 1.2 × 105 1.0 × 104 1.6 × 103 1.7 × 102
come directly from the generator simulation, and on the second lines we rescale the values
to the reference dijet cross-section.12 This method allows us to determine the error on the
Higgs production cross-section due to the assumptions made on the respective values of λH
and λi . For the Tevatron, the cross-section we obtain for a Higgs boson mass of 120 GeV
is 1.2 fb, and varies between 0.6 and 1.9, which gives an incertitude of about 50% on the
Higgs boson cross-section prediction. For the LHC, the default value is 160 fb and varies
between 89 and 220, which gives also an error of about 50%.
12 We remind that we scale our cross-section predictions to the CDF run I measurement.
The other assumption in our study is to take the gluon density in the pomeron at
Q2 = 75 GeV2 , since this is the upper value of Q2 for the QCD fit. If we take a value
at a higher Q2 by using a DGLAP evolution we obtain that the gluon density can vary by
a factor 2 at Q2 = 4000 GeV2 . This has a clear effect on our predictions which can thus
vary by a factor 2.
Another assumption of our model which needs to be tested against data is that the
fraction of events with tagged protons is the same at the LHC and the Tevatron, which
in other words means that the gap survival probability does not depend much on the center
of mass energy. To verify this point will be possible only when the LHC energy range will
be available to experimentation.
To summarize, we estimate that, on the basis of our model calculations, the errors due
to the parameter dependence on our predictions for Higgs cross-section (once normalized
to dijets) are of the order of a factor 4.
5. Testing and comparing models at the Tevatron run II
Let us discuss and propose ways to verify our model and determine its parameters more
precisely using the Tevatron run II data for dijets, diphotons and dileptons. This will allow
to make improved predictions for, e.g., Higgs boson production later on at the LHC. In
the same time, this will give the possibility to differentiate, and thus compare the existing
models which give predictions for diffractive Higgs production.
5.1. Brief discussion of other models
Models attempting to describe central diffractive production from pp̄ or pp interactions

base their predictions on either an explicit color singlet exchange of two gluons (where
one may be soft, as in our model), or in terms of hadronic interactions, in which diffractive
features (rapidity gaps, leading protons) appear in relation with soft initial and/or final state
interactions.
In this section we attempt to summarize results obtained in different pictures for
the same observables, and assess how observation could allow to distinguish them.
We concentrate on recent models13 which have provided explicit numbers for various
processes of interest (namely dijet, diphoton and Higgs boson production) so that a
comparison of predictions is made possible. We therefore compare our results with the
exclusive model predictions of Ref. [21a], with the double pomeron inclusive model of
Ref. [21b] and with the soft color interaction models of Ref. [21c]. These three approaches
will be shortly described below.
Two-gluon models for exclusive production have been first proposed, (cf., see [2]), but
led in general to too large cross-sections, in conflict with the upper bound from CDF dijets.
Despite their initiatory rôle in the problem, they will not be discussed in detail here. We
13 In fact, the observable tests that we propose could be applied to other models as well, and thus have a wide
applicability.
cannot either include the original Bialas–Landshoff exclusive model [1] in this discussion,
since the model has not been extended to gluon pair production. A confrontation of the
exclusive Bialas–Landshoff dijet production with the CDF upper bound is thus not possible
at this point.
Exclusive production is evaluated in [21a] as a perturbative process involving two-
gluon exchange tamed by Sudakov suppression factors supplemented by rapidity-gap
suppression. Two gluons are extracted from the beam particles according to the usual
proton structure functions (the process can thus be interpreted as “proton-induced”) and
one of them couples to the hard central system (e.g., a high mass dijet, Higgs boson, etc.).
The infrared divergence of this process (dσ ∝ dQ2t /Q4t , Qt being the gluon transverse
momentum) is regularized by the requirement that no radiation occurs along the gluon
which would fill the rapidity gap. This leads to Sudakov-like form factors effectively
damping the low-Qt region, so that the whole process can be considered as entirely
perturbative and thus calculable. Finally a conventional rapidity gap survival factor corrects
for the soft rescattering corrections. Clues for future studies on distinguishing the exclusive
from the inclusive DPE processes are given in Section 5.2.
The model for inclusive DPE presented in [21b], and implemented in POMWIG [22]
is a direct transposition of the pomeron flux and parton densities from the ep to the pp
context, and is therefore referred to as a factorizable model.14 In particular, it uses the full
Regge parametrization of pomeron flux factors as determined on the HERA data quoted
before. The factorizable model is formally similar to ours, since both models satisfy QCD
factorization by the use of gluon structure functions in the pomeron to describe the hard
process in the central rapidity region. However, the choice of parameters and color structure
reflects a different interpretation of the nature of the pomeron, as outlined in Section 5.3.
The closely related formulation of both models will allow to discuss their differences in
terms of physical parameters.
Two models have been designed to describe diffractive physics as soft color rescattering
over a hard subprocess [21c], namely SCI (soft color interaction) and GAL (general area
law). They are implemented as a transition between the hard interaction and hadronization,
and therefore fit naturally in Monte Carlo programs such as PYTHIA [23]. In these models,
color exchanges occur in the final state, potentially stopping color flow between the
remnants of the incoming hadrons and the central system, leading to diffractive event
topologies. These final state interaction models have a different formulation from the
previous ones.15 We will restrict ourselves to recall their predictions for various DPE final
states, insisting on differences with the other models.
14 As in our model, the model is pomeron-induced. However, our model is based on a soft pomeron interaction
between the initial hadrons which do not verify Regge factorization at the incident vertices. Here the model is
based on a Regge-factorized mechanism with a pomeron at each vertex, the necessary factorization breaking
coming from the gap suppression factor.
15 Note a formal similarity with the Bialas–Landshoff mechanism with the superposition of hard and soft color
exchanges in the same probability amplitude.
5.2. Exclusive “proton-induced” model
The question of the observation of exclusive DPE events is open. As we have said,
the current data allow only to place an upper bound on the process, and this bound is
still compatible with (but not far lower from) the predictions of [21a]. The forthcoming
Tevatron run II data will give new insight on this question, by looking for diffractively
produced central dijets, or even better diphotons and dileptons since the events will be
cleaner.
A specific question concerns the separability of the exclusive process from the inclusive
background. The exclusive processes will show a high value of the dijet, dilepton or
diphoton mass fraction. In our model, these events correspond to the case when the
pomeron remnants show very little energy. It is thus worthwhile to verify whether one
is able to distinguish between these two configurations. If we assume an experimental
resolution of 15% on the measured dijet mass, one should require Mfrac > 0.85 for both the
exclusive or the inclusive events. With this cut, we obtain a cross-section from our inclusive
model of about 2 fb at LHC, which is large enough to be competitive with the exclusive
numbers given in [21a]. The Higgs boson cross-section as a function of the mass fraction
is given in Fig. 11. In fact, this shows that it will be difficult to distinguish between pure
Fig. 11. Remaining diffractive Higgs boson cross-section at the LHC, as a function of the cut on the mass fraction.
exclusive events and the tail of the inclusive events which we can call “quasi-exclusive”
events.16 It is thus important to be able to tag both kinds of events [6].
Another example of discriminating observable is given by the ratio of the diphoton and
dilepton cross-sections, as a function of the mass-fraction value. In the inclusive models,
this ratio is determined by the quark and gluon distributions inside the pomeron, and the
presence of the Z pole in the dilepton cross-section. It is illustrated in Fig. 12. The presence
of an exclusive contribution to the cross-section would result in a sharp enhancement at
Mfrac ∼ 1, since diphotons are allowed in the exclusive case, whereas dileptons are absent.
In any case, measurements of the cross-section of exclusive or quasi-exclusive events at
Tevatron, in the case of dilepton, diphoton or dijet production will provide a direct test of
the models.
Fig. 12. Diphoton to dilepton cross-section ratio, as a function of the mass fraction.
16 Let us mention another source of backgrounds to the exclusive process, coming from the experimental
difficulty of telling that a tagged forward proton is not resulting from an N ∗ decay and accompanied by other
undetected particles. Provided the decay proton is still in the required momentum range (ξ < 0.1), this will
result in an underestimation of the missing mass, i.e., an overestimation of the mass fraction, hence polluting the
selection.
5.3. Comparison of the “pomeron-induced” models at the Tevatron run II
It has already been noticed that our model presented here, and the factorizable pomeron
induced model of [21b] in practice differ in the relative normalizations of Higgs boson vs.
dijets and in the value of the parameters. In particular, we take = 0.08, see formula (4)–
(7) while the factorizable model assumes a value of the pomeron intercept = 0.2 coming
from the HERA measurements. Also, the slope in momentum transfer originates from the
nucleon form factor in our model (eventually supplemented by an additional term related to
the non-perturbative gluon propagator) whereas the factorizable pomeron induced model
takes the value from HERA measurements. It is thus interesting to vary this parameter.
Note an important difference between model predictions due to the distinctive color
reconnection pattern of the two models.√In the factorized model the Higgs boson to dijet
2 GF
normalization ratio is: (CH /CJ J )fact = 24π , with a factor 8 smaller than (3) for the non-
factorized model. The reason is that the factorizable model authorizes dijets coming from
two quarks or gluons bearing different color charges, while they are restricted to be in
an overall singlet state in the non-factorizable model due to the common gluon exchanged
between the incident nucleons (in other terms the whole of pomeron remnants is essentially
color singlet). Hence the prediction of the Higgs boson vs. dijet cross-section ratio keeps
track of this difference. Here again, insight can be gained from the comparison of diphoton
(or dilepton) and dijet cross-sections: since the first ones have the same color structure
as Higgs boson production, they can be used to infer which, of the factorizable and non-
factorizable pictures, is correct.
Fig. 13 shows the variation of the Higgs and dijet cross-sections as a function of at
the generator level, and their ratio. As can be guessed from formulae (4)–(7), an increase
of will enhance the cross-section essentially at low diffractive mass, therefore the effect
on the dijet cross-section is much larger than on the Higgs boson cross-section. When
is increased, the dijet cross-section is increased, and thus the multiplicative factor needed
to adjust the prediction to the CDF run I measurement is smaller. This explains why the
predicted Higgs cross-section is expected to decrease when increases.
Fig. 14 shows the -dependence of the differential dijet cross-section dσJ J /dMJ J ,
when compared to (i.e., divided by) a reference distribution, taken at = 0.08. A strong
dependence is found, meaning that the study of this distribution, independently from its
normalization, should allow to constrain the value of , and to obtain a more precise
prediction for the Higgs boson cross-section at the LHC.17 Similar studies can be done
with dilepton and diphoton events, which will benefit from much central mass resolution.
In Fig. 15 we give the cross-sections for dijets, bb̄, γ γ , and Higgs boson, and in Fig. 16,
the ratio of the Higgs cross-section with respect to γ γ and bb̄ as a function of for
the LHC. Namely, it will allow to get a prediction on these cross-sections, once been
determined, e.g., at the Tevatron.
These last remarks highlight once more the interest of DPE diphoton and dilepton
production. Indeed, the experimental resolution on photon, electron and muon momenta
17 The sensitivity of dσ /dM

JJ J J to will however crucially depend on the experimental determination of the
jet energy scale and on the remaining uncertainties on the pomeron structure function GP .
Fig. 13. Higgs boson and dijet production at Tevatron: raw cross-sections and their ratio for different values of ,
at MH = 120 GeV.
is much better than the jet energy resolution, since in this case the events benefit from
tracking information and/or electromagnetic calorimetry with a much better defined energy
scale. Moreover, especially in the dilepton case, the hard sub-process is essentially quark-
initiated (see formula (6)), and the knowledge of QP is much more precise than that of GP ,
thus eliminating another source of uncertainty. The measurement of will also be possible
using the diphoton and dilepton measurement and the dependence of the cross-sections as
a function the diphoton or the dilepton mass.
5.4. Predictions: summary and comparison
In Table 6, we give, where available, the cross-sections obtained by the different models.
The factorizable numbers at the LHC (second column) should contain a factor accounting
for the unknown rapidity gap survival probability (denoted SP ); this factor is around 0.1 at
the Tevatron. The numbers from the proton-induced model refer to exclusive production,
and are as such not referring to strictly the same process as the pomeron induced ones. The
various models have fundamentally different philosophies and assumptions, so that caution
should be taken not to conclude from any numerical agreement that the models themselves
agree.
Fig. 14. Differential dijet cross-sections for different values of , compared to a reference taken at = 0.08. All
cross-sections are normalized to the same value. The values = 0.08, 0.2 and 0.5 correspond, respectively, to the
soft, hard (measurement at HERA), and hard BFKL pomeron intercepts.
Table 6
DPE Higgs and diphoton production cross-section at the LHC and the Tevatron (fb) for different models.
(1) Nonfactorizable pomeron based model (present work), (2) Factorisable pomeron based model (the results
contain both the pomeron and the reggeon components. For the γ γ cross-section, we give in parenthesis the
contribution of the pomeron only); (3) proton based model (exclusive case; the results are given for a Higgs mass
of 120 GeV), (4) soft color interaction models
Process and cuts (1) (2) (3) (4)
Higgs, 115 GeV, TeV 1.7 0.029–0.092 0.03 0.00012
Higgs, 115 GeV, LHC 169 379–486 × SP 1.4 0.19
Higgs, 160 GeV, LHC 123 145 × SP 0.55 –
γ γ , Tev, ET > 12 GeV, η < 2 71 128 (27) – ∼ 20
γ γ , Tev, ET > 12 GeV, η < 1 9 8 (2) – –
γ γ , LHC, ET > 50 GeV, η < 2 1.5 – – 0.1
γ γ , LHC, ET > 120 GeV, η < 5 19 – 0.12 –
Fig. 15. Raw cross-sections vs. .
6. Advantage of diffractive vs. “standard” Higgs boson production
As is now well-known [1–3,21a], exclusive DPE events are very attractive from the
experimental point of view, since the measurement of the outgoing protons momenta,
using dedicated forward detectors (like roman pots or microchambers [20]) allows to
determine the mass of the centrally produced system with excellent precision, so that the
Higgs boson signal can be extracted from the DPE continuum background. Moreover,
the background itself is strongly suppressed due to angular momentum conservation
constraints (the “JZ = 0” rule [8], see Section 2.1). Exclusive DPE events are however
not an experimentally established process according to the data currently at our disposal.
As we have emphasized, the “quasi-exclusive processes” which are quite close in topology
from the exclusive processes are established processes and may keep an essential part of
the nice kinematical features of the exclusive production.
In general inclusive DPE events, the angular momentum constraint above does not apply
and the background is not suppressed, so that the a priori signal-to-background ratio (S/B)
is similar to that of “standard” non-diffractive events. The missing mass method will not
work as such, since the equality between the missing mass to the outgoing protons and the
mass of the central system does not hold.
However, inclusive DPE events can still be kinematically constrained in a much stronger
way than non-diffractive events. If, in addition to the forward proton detectors, the
Fig. 16. Higgs compared to dijets and diphotons as a function of .
experiments can dispose of very forward calorimetry (e.g., up to η ∼ 8, as was proposed in

[24]), one may use the measurement of the pomeron remnants to fully constrain the events
kinematics. Combined with the higher cross-sections predicted in this last case, it shows
that such forward calorimetry would be adapted and necessary. The first resolutions using
these detectors are given in [6].
The idea is to use the forward calorimeters as vetos. The central mass is reconstructed
with the proton momenta only, and events are considered only if the measured remnant
energy is small enough. In a sense, this is a way to select “quasi-exclusive” events. Fig. 3
of Ref. [6] shows the achievable resolution as a function of the remnant veto.
Let us finish this discussion with an important point concerning the interest of the τ τ
decay mode. According to Section 3, the inclusive DPE τ τ cross-section is very small
compared to the DPE dijet cross-section, and this is understood from the smallness of the
QED coupling constant and the quark component of the pomeron. Compared to this, the
Higgs boson branching fraction into τ pairs is still ∼ 10% in the mass range we consider.
It follows from the cross-sections given earlier (see Table 3, and Fig. 10) that the cross-
section ratio for τ -pair events is O(10) at for example mτ τ ∼ 130 GeV for inclusive DPE
events, whereas this mode is invisible through non-diffractive events.
Of course, the missing energy in τ decays, the presence of the Z peak in the τ -pair
cross-section, and other background processes not considered here spoil these numbers
to some extent, but we do consider that this channel needs investigation. Also note that a
pseudoscalar Higgs boson A, which appears in models with an extended Higgs sector (such
as the supersymmetric models), does not couple to gauge bosons and hence preserves a
significant branching fraction into τ -pairs up to very high masses, where the effect of the
Z peak is negligible; the same remark is true for the MSSM h boson in some regions of the
parameter space.
7. Summary
The present work is meant to serve as a reference to the non-factorizable, pomeron

induced DPE model.
• The original exclusive DPE formulation are recalled, and the inclusive expressions are
introduced, see formulae (4)–(7).
• A systematic study is performed, that sheds light on the role of the various model
parameters. It is emphasized that the forthcoming Tevatron data will allow to constrain
their values, and answer the pending question of the existence of exclusive DPE
production.
• Precise predictions for discovery physics at the LHC will then be possible; we focused
on the Higgs boson case.
• It is recalled that, assuming favorable experimental installations, the bb̄ decay mode
of the Higgs boson may well be exploited in double diffractive production.
• On the theoretical side, the H → τ τ channel is most promising, although some
experimental difficulties have to be overcome.
Acknowledgements
We wish to thank M. Albrow, B. Cox, R. Enberg, V. Khoze, R. Orava, A. de Roeck,

M. Ryskin and L. Schoeffel for useful discussions. One of us (M.B.) is grateful to the
CEA/DSM/SPhT for support.
Appendix A
The cross-sections for the hard processes that are used in our study are summarized
below, in terms of the usual Mandelstam variables. For dijet production, we take

dσgg→qq παS2 1 u2 + t 2 3 u2 + t 2
= 2 − ,
dt s 6 ut 8 s2

dσgg→gg πα 2 9 ut us st
= 2S 3− 2 − 2 − 2 . (A.1)
dt s 2 s t u
For dilepton production, we have
dσqq→ll πα 2 4 u2 + t 2
= 2 CEW . (A.2)
dt s 3 s2
Here CEW includes electromagnetic and weak factors accounting for photon and Z
exchange between the initial quark pair and the final leptons. Its expression is, for a quark
of flavor i:
s(s − MZ2 )(Le + Re )(Lq + Rq )
CEW = ei2 − ei
8xW (1 − xW )[(s − MZ2 )2 − MZ2 ΓZ2 ]
s 2 (L2e + Re2 )(L2q + Rq2 )
+ 2 (1 − x )2 [(s − M 2 )2 − M 2 Γ 2 ]
. (A.3)
64xW W Z Z Z
Above, xW is the weak mixing angle, MZ and ΓZ are the Z boson mass and width, and
L and R are the left- and right-handed chiral couplings. Finally, the diphoton production
cross-section reads:

dσqq→γ γ πα 2 4 1 u t
= 2 ei + , (A.4)
dt s 3 t u

dσgg→γ γ α 2 αS2 2 1
= e i Cloop, (A.5)
dt 8πs 2 8
i
where Cloop is a complicated expression resulting from the loop integral occurring in the
gg → γ γ process. Its detailed expression can be found in [18].
References
[1] A. Bialas, P.V. Landshoff, Phys. Lett. B 256 (1990) 540;

A. Bialas, W. Szeremeta, Phys. Lett. B 296 (1992) 191;
A. Bialas, R. Janik, Z. Phys. C 62 (1994) 487.
[2] J.D. Bjorken, Phys. Rev. D 47 (1993);
For a complete list, see also: A. Schafer, O. Nachtmann, R. Schöpf, Phys. Lett. B 249 (1990) 331;
J.-R. Cudell, O.F. Hernandez, Nucl. Phys. B 471 (1996) 471;
H.J. Lu, J. Milana, Phys. Rev. D 51 (1995) 6107;
D. Graudenz, G. Veneziano, Phys. Lett. B 365 (1996) 302;
M. Heyssler, Z. Kunszt, W.J. Stirling, Phys. Lett. B 406 (1997) 95;
E.M. Levin, hep-ph/9912403, and references therein;
V.A. Khoze, A.D. Martin, M.G. Ryskin, Eur. Phys. J. C 14 (2000) 525;
V.A. Khoze, A.D. Martin, M.G. Ryskin, Eur. Phys. J. C 19 (2001) 477, hep-ph/0006005;
V.A. Khoze, hep-ph/0105224.
[3] M.G. Albrow, A. Rostovtsev, hep-ph/0009336.
[4] M. Boonekamp, R. Peschanski, C. Royon, Phys. Rev. Lett. 87 (2001) 251806.
[5] CDF Collaboration, T. Affolder, et al., Phys. Rev. Lett. 85 (2000) 5043.
[6] M. Boonekamp, A. De Roeck, R. Peschanski, C. Royon, hep-ph/0205332;
M. Boonekamp, A. De Roeck, R. Peschanski, C. Royon, Phys. Lett. B 550 (2002) 93.
[7] P.V. Landshoff, O. Nachtmann, Z. Phys. C 35 (1987) 405.
[8] J. Pumplin, Phys. Rev. D 52 (1995) 1477;
A. Berera, J.C. Collins, Nucl. Phys. B 474 (1996) 183.
[9] See V.A. Khoze, in [2].
[10] A. Donnachie, P.V. Landshoff, Phys. Lett. B 207 (1988) 319.
[11] H1 Collaboration, C. Adloff, et al., Z. Phys. C 76 (1997) 613.
[12] P. Laycock, Talk given at 10th Intl. Workshop on Deep Inelastic Scattering (DIS 2002), Cracow, May, 2002,
Acta Phys. Pol. B 33 (2002) 3413;
F.P. Schilling, hep-ex/0209001;
F.P. Schilling, Acta Phys. Pol. B 33 (2002) 3419;
P.R. Newman, Talk at Low-X Meeting, Antwerpen, September, 2002.
[13] CDF Collaboration, T. Affolder, et al., Phys. Rev. Lett. 84 (2000) 4215.
[14] C. Royon, L. Schoeffel, J. Bartels, H. Jung, R. Peschanski, Phys. Rev. D 63 (2001) 074004.
[15] B.L. Combridge, C.J. Maxwell, Nucl. Phys. B 239 (1984) 429.
[16] B.L. Combridge, J. Kripfganz, J. Ranft, Phys. Lett. B 70 (1977) 234.
[17] E. Eichten, I. Hinchliffe, K.D. Lane, C. Quigg, Rev. Mod. Phys. 56 (1984) 579;
E. Eichten, I. Hinchliffe, K.D. Lane, C. Quigg, Rev. Mod. Phys. 58 (1986) 1065, Addendum.
[18] E.L. Berger, E. Braaten, R.D. Field, Nucl. Phys. B 239 (1984) 52.
[19] SHW, see http://www.physics.rutgers.edu/jconway/soft/shw/shw.html.
[20] R. Orava, Talk at LISHEP02, Rio de Janeiro, February, 2002.
[21] (a) V.A. Khoze, A.D. Martin, M.G. Ryskin, Eur. Phys. J. C 24 (2002) 581;
(b) B. Cox, J. Forshaw, B. Heinemann, Phys. Lett. B 540 (2002) 263;
R.B. Appleby, J.R. Forshaw, Phys. Lett. B 541 (2002) 108;
(c) R. Enberg, G. Ingelman, A. Kissavos, N. Timneanu, Phys. Rev. Lett. 89 (2002) 081801;
R. Enberg, G. Ingelman, N. Timneanu, Diffractive Higgs bosons and prompt photons at hadron colliders,
hep-ph/0210408, with references to the two versions (SCI and GAL) of the model.
[22] B.E. Cox, J.R. Forshaw, Comput. Phys. Commun. 144 (2002) 104.
[23] T. Sjostrand, P. Eden, C. Friberg, L. Lonnblad, G. Miu, S. Mrenna, E. Norrbin, Comput. Phys. Commun. 135
(2001) 238.
[24] CMS Collaboration, Technical Design Report (1997).
Newton law on the generalized singular brane

with and without 4d induced gravity
Eylee Jung, Sung-Hoon Kim, D.K. Park
Department of Physics, Kyungnam University, Masan 631-701, South Korea
Received 20 May 2003; received in revised form 30 June 2003; accepted 28 July 2003
Abstract
Newton law arising due to the gravity localized on the general singular brane embedded in AdS5
bulk is examined in the absence or presence of the 4d induced Einstein term. For the RS brane, apart
from the subleading correction, Newton potential obeys 4d- and 5d-type gravitational law at long- and
short-ranges if it were not for the induced Einstein term. The 4d induced Einstein term generates an
intermediate range at short distance, in which the 5d Newton potential 1/r 2 emerges. For Neumann
brane the long-range behavior of Newton potential is exponentially suppressed regardless of the
existence of the induced Einstein term. For Dirichlet brane the expression of Newton potential is
dependent on the renormalized coupling constant v ren . At particular value of v ren Newton potential
on Dirichlet brane exhibits a similar behavior to that on RS brane. For other values the long-range
behavior of Newton potential is exponentially suppressed as that in Neumann brane.
1. Introduction
A braneworld scenario is a physical picture which assumes that our 4d spacetime

universe is embedded in higher-dimensional world. Although higher-dimensional theory
has its own long history [1–3] in the context of Kaluza–Klein theory or not, modern
braneworld scenarios such as large extra dimensions [4,5] or warped extra dimensions
[6,7] seem to be mainly motivated from the string theories [8]. Making use of the recent
scenarios a flurry of activities has tried to examine the various physical problems such as
big bang universe [9–11], cosmological constant hierarchy [12–14], brane inflation [15],
E-mail addresses: eylee@mail.kyungnam.ac.kr (E. Jung), shoon@mail.kyungnam.ac.kr (S.-H. Kim),

dkpark@hep.kyungnam.ac.kr (D.K. Park).
doi:10.1016/j.nuclphysb.2003.07.020
E. Jung et al. / Nuclear Physics B 669 (2003) 306–324 307
and black hole physics [16–21]. In this paper we will examine Newton law arising due to
the gravity localized on the brane when general singular brane is embedded in AdS5 bulk.
The Newton law on the brane was firstly computed by RS in Ref. [7] when the bulk
spacetime is AdS5 . In this case the gravity localized on the brane yields an 4d-type 1/r and
1/r 3 subleading correction arising due to Kaluza–Klein excitation:

m1 m2 R2
V ∼G 1+ 2 , (1.1)
r r
where R is a radius of AdS5 . Under the same setup an improvement of the potential
was tried by involving the brane-bending effect [18,22] and by computing the one-loop
correction to the gravitational propagator [23], which slightly changes the sub-leading
correction by a overall multiplication factor

m1 m2 2 R2
V ∼G 1+ . (1.2)
r 3 r2
Subsequently, the gravitational propagator for the linearized fluctuation equation in
RS picture is generally computed by adopting the singular quantum mechanics (SQM)
as a calculational tool [24,25]. From the viewpoint of SQM the linearized gravitational
propagator is crucially dependent on the boundary condition (BC) that the propagator
should satisfy at the location of the brane. The physically relevant BC is not uniquely
determined in the framework of SQM due to the singular nature of the potential and the
physically consistent BCs are parametrized by a real parameter, say ξ . The physically
consistent BCs are in general introduced by a self-adjoint extension technique [26,27],
which is effectively identical to the coupling constant renormalization [28,29]. Recently,
this generalized gravitational propagators are used to compute the Newton potential on the
brane [30,31] with a particular choice of ξ as ξ = 1/2, which means the Dirichlet and
Neumann BCs are included with an equal weight.
Recently, the Newton law with different setup is also considered. Especially, much
attention is paid to the case of Minkowski bulk with an involving the 4d induced Einstein
term arising due to the quantum effect of one loop [32,33]. In the case of the flat bulk
the Newton potential becomes 4d-type 1/r at the short range, i.e., r λ, and 5d-type
1/r 2 at the long range, i.e., r λ, where λ is a ratio of 4d Planck scale with that of 5d;
λ ≡ M42 /M53 . This fact is used to explain the observed acceleration of the Universe [34].
The effect of the induced Einstein term is also examined when the 5d bulk is AdS5
spacetime [35,36]. In this setup the 5d-type Newton potential arises in the region of
λ r R. At other ranges the 4d gravitational potential is recovered. Recently, this fact
is again confirmed in the context of SQM [31].
In this paper we will examine the Newton potential using the generalized gravitational
propagators derived using SQM. Thus, the gravitational propagator contains in general
the parameter ξ , which parametrizes the physically relevant BCs. The ξ -dependence of
the Newton potential is examined throughout the paper. We will confine ourselves to the
case of single copy of the AdS5 bulk. Thus our computation is limited to the original RS
plus AdS/CFT [37,38] setup with and without the induced Einstein term. In particular,
we will examine in detail the cases of ξ = 0, ξ = 1, and ξ = 1/2. When ξ = 1, we have
a Neumann brane in which the gravity acquires mass [24,25]. It is well known that the
308 E. Jung et al. / Nuclear Physics B 669 (2003) 306–324
Newton potential generated due to the exchange of a massive graviton is exponentially

suppressed at long range. This fact will be explicitly shown in the following. At ξ = 0
we have a Dirichlet brane. In this case the non-trivial gravitational propagator can be
derived via the coupling constant renormalization [24,25]. Thus the final expression of the
Newton potential is dependent on the renormalized coupling constant. We will show in the
following that when the renormalized coupling constant has a particular valus, the Newton
potential on the Dirichlet brane is proportional to that on RS brane which corresponds to
the brane at ξ = 1/2.
The paper is organized as following. In Section 2 we will derive a formula which
shows how to compute Newton potential arising due to a localized gravity from the fixed-
energy amplitude. Using the formula Newton law in general singular brane is examined
in Section 3 without consideration of the 4d induced Einstein term when bulk is a single
copy of AdS5 spacetime. A particular attention is paid to the RS-brane (ξ = 1/2), Neumann
brane (ξ = 1) and Dirichlet brane (ξ = 0). The effect of the 4d induced Einstein term to
Newton potential is studied in Section 4. In Section 5 brief conclusion and further remark
are given. In Appendices A and B Newton laws on Neumann and Dirichlet branes are
explicitly derived in the presence of the 4d induced Einstein term.
2. Newton potential from the fixed-energy amplitude
In this section we will consider briefly how to compute Newton potential arising due
to the localized gravity on the brane from the fixed-energy amplitude1 computed in the
context of the SQM. As an example we will consider the usual RS scenario without the
induced Einstein term. However, the final result is model-independent.
Let us consider the 5d Einstein equation2
1 1 µ ν
RMN − GMN R = − 3
ΛGMN + vb Gµν δM δN δ(y) , (2.1)
2 4M
where M, Λ and vb are 5d Planck scale, 5d cosmological constant and brane tension,
respectively. As is well known it has a solution
ds 2 = e−2k|y| ηµν dx µ dx ν + dy 2 (2.2)

if thefine-tuning conditions Λ = −24M 3k 2 and vb = 24M 3 k
are satisfied.
Introducing the linearized gravitational fluctuation hµν (x, y) as following

ds 2 = e−2k|y| ηµν + hµν (x, y) dx µ dx ν + dy 2 (2.3)
µ
and ignoring the tensor structure for simplicity by choosing a gauge hµν,µ = hµ = 0, one
can derive the following linearized fluctuation equation
2k|y| (4)
e ✷ + ∂y2 − 4k 2 + 4kδ(y) hµν = 0, (2.4)
1 The definition of the fixed-energy amplitude is a Laplace transform of the Euclidean propagator.
2 Our notation is that M and N are bulk spacetime indices and, µ and ν are brane spacetime indices.
which reduces to
m2
ĤRS ψ̂(z) = ψ̂(z),
2
1 15 3
ĤRS = − ∂z2 + − kδ(z), (2.5)
2 8(|z| + R) 2 2
where R ≡ 1/k is a radius of AdS5 and
ek|y| − 1
z = !(y) ,
k
hµν (x, y) = e−k/2|y|ψ̂(y)eipx ,
m2 = −p2 . (2.6)
At this stage we should stress the fact that if our calculation involves the brane-bending
effect, Eq. (2.4) is modified into
2k|y| (4)
e ✷ + ∂y2 − 4k 2 + 4kδ(y) hµν = −Σµν δ(y), (2.7)
where Σµν is a tensor quantity and represents the brane-bending effect [22]. Thus, the
additional term plays the role of the source and recovers the tensor structure of hµν .
Although it might be possible, in principle, to compute the propagator of Eq. (2.7) from
the viewpoint of SQM with imposing a general BC on the brane after computing Σµν
explicitly and changing Eq. (2.7) as a Schrödinger-type equation, it may lead an extreme
complication. Since our interest is to examine the effect of the general boundary condition
in the Newton potential, we will confine ourselves to Eq. (2.5) without recovering the
tensor structure.
The Newton potential in the bulk is in general computed by the time-integration of the
retarded Green function U (
x2 , y2 ; x
1, y1 ; t) [22,32]:

V (
x2 , y2 ; x
1 , y1 ) ≡ dt U (
x2 , y2 ; x
1, y1 ; t). (2.8)
In terms of the time-dependent propagator, U (

x2 , y2 ; x
1 , y1 ; t) is represented as

1 d 4 p −k/2(|y1|+|y2 |)
x2 , y2 ; x
1, y1 ; t) = 3
U (
e G[y1 + R, y2 + R; t]
M (2π)3

x2 −
x1 ) −p0 t

× ei p·(
e δ p0 − |p|
. (2.9)
The factor e−k/2(|y1|+|y2 |) in Eq. (2.9) is introduced due to the redefinition of hµν (x, y)
in Eq. (2.6). The time-integration in Eq. (2.8) changes the Euclidean propagator G[y1 +
R, y2 + R; t] into the fixed-energy amplitude Ĝ[y1 + R, y2 + R; p02 /2] as following:

1 −k/2(|y1 |+|y2 |)
x2 −

x2 , y2 ; x
1 , y1 ) =
V (
3 3
e d 4 p ei p·(
x1 )
δ p0 − |p|

8π M

p02
× Ĝ y1 + R, y2 + R; . (2.10)
2
Thus the p0 -integration in Eq. (2.10) results in
1
x2 , y2 ; x
1 , y1 ) =
V (
e−k/2(|y1|+|y2 |)
8π 3 M 3

x2 −
p
2
× d 3 p
ei p·(
x1 )
Ĝ y1 + R, y2 + R; . (2.11)
2
If the brane has three spatial dimensions and is located at y = 0, the Newton potential on
the brane is expressed as

x2 − x
1 |, y1 = y2 = 0
V (r) ≡ V |
∞
1 m2
= dm m sin mr Ĝ R, R; , (2.12)
2π 2 M 3 r 2
0
where r ≡ |
x2 − x
1 |.
Eq. (2.12) enables us to compute the Newton potential generated by the localized gravity
on the brane from the fixed-energy amplitude Ĝ[R, R; m2 /2]. Since the general fixed-
energy amplitude which depends on the BC is computed in Refs. [24,25] when there is
no 4d induced Einstein term, we can compute the Newton potential by making use of
Eq. (2.12). This will be examined in the next section in detail.
3. Newton potential from fixed-energy amplitude: without 4d induced Einstein term
In this section we will compute the Newton potential arising due to the localized gravity
on the brane when there is no 4d induced Einstein term. Thus what we have to do first is to
derive the fixed-energy amplitude for the linearized gravitational fluctuation (2.5). In Refs.
[24,25] Eq. (2.5) is slightly generalized as following
Ĥ1 ψ̂(z) = E ψ̂(z),

Ĥ1 = Ĥ0 − vδ(z),
1 g
Ĥ0 = − ∂z2 + . (3.1)
2 (|z| + c)2
Of course, Ĥ1 in Eq. (3.1) coincides with ĤRS in Eq. (2.5) when c = 1/k ≡ R, g = 15/8,
and v = 3k/2.
As we commented before we require the bulk spacetime is a single copy of AdS5 . This
requirement leads us naturally to combine the usual Schulman procedure [39,40] for the
treatment of the δ-function potential with an half-line constraint. The half-line constraint
usually makes the fixed-energy amplitude to be dependent on the BC on the brane. This
procedure is explicitly addressed in Refs. [24,25]. The final expression of the fixed-energy
amplitude for Ĥ1 under these circumstance is
Ĝ1 [a, b; E] = ĜD

0√[a, b; E] √ √
ab Kγ ( 2E a)Kγ ( 2E b)
+ √
cv Kγ2 ( 2E c)
√ √
γ − 1/2 2E Kγ −1 ( 2E c) −1
× −1 + √ , (3.2)
2ξ cv 2ξ v Kγ ( 2E c)
√
where γ ≡ 1 + 8g/2 and Kγ (z) is an usual modified Bessel function. The parameter ξ
is explicitly introduced in Eq. (3.2). The real parameter ξ parametrizes the BCs for the
fixed-energy amplitude corresponding to Ĥ0 . For example, ξ = 1 (or ξ = 0) means that we
chose the Neumann (or Dirichlet) BC for the fixed-energy amplitude corresponding to Ĥ0 .
ĜD0 [a, b; E] in Eq. (3.2) is a fixed-energy amplitude for Ĥ0 when we adopt a Dirichlet BC
at the brane. The explicit form of ĜD 0 [a, b; E] is given in Refs. [24,25]. However we do
not need to know its explicit form. The only one we should know is the fact that it satisfies
the usual Dirichlet BC, i.e., ĜD 0 [a, c; E] = Ĝ0 [c, b; E] = 0.
D
The fixed-energy amplitude on the brane Ĝ1 [c, c; E] is easily computed from Eq. (3.2):
√ √
1 γ − 1/2 2E Kγ −1 ( 2E c) −1
Ĝ1 [c, c; E] = −1 + √ . (3.3)
v 2ξ cv 2ξ v Kγ ( 2E c)
Thus inserting Eq. (3.3) into Eq. (2.12) we can explicitly compute the Newton potential.
Firstly, we consider ξ = 1/2 case which corresponds to the case of usual RS brane.
Letting c = R, g = 15/8 and v = 3/2R makes the fixed-energy amplitude to be
1 [R, R; E] = ∆0 + ∆KK ,
ĜRS (3.4)
where √
1 1 K0 ( 2E R)
∆0 = , ∆KK = √ √ . (3.5)
ER 2E K1 ( 2E R)
Of course, ∆0 and ∆KK represent the zero mode and KK excitation, respectively. Inserting
Eq. (3.4) into Eq. (2.12) it is easy to show that the Newton potential on the RS brane
reduces to
VRS (r) = V0,RS (r) + ∆VRS (r), (3.6)
where
G
V0,RS (r) = ,
r
∞
G1 r K0 (u)
∆VRS (r) = du sin u , (3.7)
π r R K1 (u)
0
and 4d Newton constant is defined as G ≡ 1/2πM 3R.

It is not difficult to show that the integral in ∆V RS (r) is not well defined. Thus we need
an appropriate regularization. This is achieved by introducing the infinitesimal parameter
! as following [31,35]
G1
∆VRS (r) = J, (3.8)
π r
where
∞
r K0 (u) −!u
J ≡ lim du sin u e . (3.9)
!→0 R K1 (u)
0
The factor sin((r/R)u) in Eq. (3.8) is crucial to extract the long-range behavior and
short-range behavior of the Newton potential. Firstly, let us consider the long-range
behavior, i.e., r R. The high oscillation behavior of sin((r/R)u) in this case makes
the small u region to be a dominant contribution. Since K0 (u) ∼ − ln u and K1 (u) ∼ 1/u
at u ∼ 0, J becomes approximately in this region
∞
r
J ∼ − lim du sin u u ln ue−!u . (3.10)
!→0 R
0
Using an integration formula
∞
1 −1 p
dx xe−qx sin(px) ln x = sin 2 tan
p2 + q 2 q
0

1 p p
× 1 − γ − ln p2 + q 2 + tan−1 cot 2 tan−1 ,
2 q q
(3.11)
where γ in Eq. (3.11) is an Euler’s constant, it is simple to show J ∼ πR 2 /2r 2 .
Secondly, let us consider the short-range behavior, i.e., r R. In this case contrary to
√ u region makes a dominant contribution to J . Since at
the long-range behavior the large
this region K0 (u) ∼ K1 (u) ∼ π/(2u) e−u , J reduces to
∞
r
J∼ du sin u e−!u , (3.12)
R
0
which makes J to be R/r. Thus we can summarize the behavior of J as following:
∞
r K0 (u) −!u πR 2 /(2r 2 ), if r R,

J ≡ lim du sin u e ∼ (3.13)
!→0 R K1 (u) R/r, if r R.
0
Eq. (3.13) is confirmed numerically. Fig. 1 shows that J and πR 2 /2r 2 at long-range.
For simple numerical calculation we chose ! = 0.003. Fig. 1 indicates that J and πR 2 /2r 2
merge with each other at r ∼ 15R. Of course, if we chose smaller !, the mergence may
occur rapidly. Fig. 2 shows that J and R/r at short-range with choosing ! = 0.003. Fig. 2
also indicates that J and R/r merge with each other at r ∼ 0.05R.
Using Eq. (3.13) one can show that the Newton potential on the RS brane becomes

(G/r)(1 + R 2 /(2r 2 )), if r R,

VRS (r) ∼ (3.14)
(GR)/(πr 2 )(1 + πr/R), if r R.
Fig. 1. Plot of J and π R 2 /2r 2 with choosing ! = 0.003. This figure indicates J and π R 2 /2r 2 merge with each
other at r ∼ 15R. Thus the long-range behavior of J in Eq. (3.13) is confirmed.
Fig. 2. Plot of J and R/r with choosing ! = 0.003. This figure indicates J and R/r merge with each other at
r ∼ 0.05R. Thus the short-range behavior of J in Eq. (3.13) is confirmed.
Thus we have 5d Newton potential 1/r 2 at the short-range. Of course, the Newton potential
is 4d-type 1/r at long-range as shown in Ref. [7].
Here, we should point out that the long-range behavior of the subleading term of VRS (r)
is different from both Eqs. (1.1) and (1.2) due to 1/2 factor. In fact, the subleading term
for our approach is J /π and we computed J with adopting the simple regularization
scheme in Eq. (3.9). However, the different scheme may assign different value to J . In
this way, the coefficient of the subleading term is dependent on the regularization scheme.
It seems to be interesting to find a suitable regularization scheme which yields Eq. (1.1)
or Eq. (1.2). Since, however, our interest in this paper is only to find a global behavior of
Newton potential in general singular brane, we will not explore this issue further.
Now, we consider ξ = 1 case which means that the Neumann BC is chosen. Inserting
c = R, g = 15/8 and v = 3/2R into Eq. (3.3) one can show that the fixed-energy amplitude
on the Neumann brane is
2 K2 (mR)
m2 m K1 (mR)
N
Ĝ1 R, R; = 1 K2 (mR)
. (3.15)
2 1 − 32 mR K1 (mR)
Inserting Eq. (3.15) into (2.12) shows that the Newton potential arising due to the localized
gravity on the Neumann brane reduces to
∞ K2 (u)
1 r K1 (u)
VN (r) = 2 3 du sin u . (3.16)
π M Rr R 1 − 3 K2 (u)
2u K1 (u)
0
Since the term sin((r/R)u) makes that the small u region contributes dominantly at the
long-range, the Newton potential at this range becomes
∞
−2 r u
VN (r) ∼ 2 3 du sin u . (3.17)
π M Rr R 3 − u2
0
Since the small u should contribute to VN (r) dominantly, one can approximately change
Eq. (3.17) into
∞
−2 r u
VN (r) ∼ 2 3 du sin u (3.18)
π M Rr R 3 + u2
0
which results in
1 √
− 3 (r/R)
VN (r) ∼ − e (3.19)
πM 3 Rr
at the long-range, i.e., r R. Thus one can conclude that the gravitational force at the
Neumann brane is exponentially suppressed. It is in fact conjectured from the fact that
−1
1 [R, R; E] has a pole at mN ∼ 2.48R , which makes the graviton on the Neumann
ĜN
brane to be massive.
At the short-range one can use the asymptotic formula for the modified Bessel function,
which results in
∞ ∞
1 r −!u 3 sin((r/R)u)
VN (r) ∼ 2 3 lim du sin u e − du , (3.20)
π M Rr !→0 R 2 3/2 − u
0 0
where the infinitesimal parameter ! is introduced again for the regularization. Carrying out
the integration using the integral formula
∞
sin(ax)
dx = sin(βa) ci(βa) − cos(βa) si(βa) + π , (3.21)
β −x
0
where si(z) and ci(z) are usual sine and cosine integral, respectively, one can express VN (r)
at short-range as

1 R 3 3r 3r 3r 3r
VN (r) ∼ 2 3 − sin ci − cos si
π M Rr r 2 2R 2R 2R 2R

3r
− π cos . (3.22)
2R
Since r R, one can expand VN (r) using the following expansions of si(z) and ci(z) [41]:
∞
π (−1)k+1 z2k−1
si(z) = − + ,
2 (2k − 1)(2k − 1)!
k=1
∞
z2k
ci(z) = γ + ln z + (−1)k , (3.23)
2k(2k)!
k=1
where γ = 0.577 . . . is an Euler’s constant. Then finally VN (r) reduces to

2
G5 3r 3r
VN (r) ∼ 2 1 − ln (3.24)
r 2R 2R
at short-range, where 5d Newton constant G5 is defined as G5 = 1/π 2 M 3 . Thus, for the
Neumann brane 5d Newton potential arises at the short-range like RS brane. The different
point, however, is that VN (r) has a logarithmic sub-leading correction.
Now, finally let us consider ξ = 0 case which means we have chosen the Dirichlet BC
at the brane for the fixed-energy amplitude of Ĥ0 in Eq. (3.1). If one naively inserts ξ = 0
in Eq. (3.2), ĜD1 [a, b; E] becomes to be identical to Ĝ0 [a, b; E] because the modification
D
term in Eq. (3.2) arising due to the δ-function potential via Schulman procedure [39,40]
vanishes at ξ = 0. Even in this case, however, one can generate the non-trivial fixed-energy
amplitude through a coupling constant renormalization. In order to adopt this procedure
we should regard the coupling constant v as an unphysical and infinite bare one. Then
one can introduce a physical renormalized coupling constant v ren which is related to v by
v ren = (1/v − 2!)/2! 2, where ! is an infinitesimal parameter. Through this renormalization
procedure one can derive a non-trivial fixed-energy amplitude [24,25]:3

m2
ĜD,ren
1 a, b;
2
√
2 ab K2 (ma)K2 (mb)
= Ĝ0 [a, b; E] +
D
K1 (mR)
, (3.25)
(Rv + 3/2) + mR K (mR)
ren K22 (mR)
2
which results in

D,ren m2 2R
Ĝ1 R, R; = . (3.26)
2 (Rv + 3/2) + mR K
ren 1 (mR)
K2 (mR)
3 There is a factor 2 mistake in Eq. (26) of Ref. [24].

Inserting Eq. (3.26) into Eq. (2.12) the Newton potential arising due to the exchange of the
gravity localized on the Dirichlet brane becomes
∞
1 r 1
VD (r) = 2 3 du u sin u . (3.27)
π M Rr R (Rv ren + 3/2) + u K1 (u)
K2 (u)
0
Firstly, let us consider a case of Rv ren + 3/2 = 0. Comparing Eq. (3.26) in this case with
Eq. (3.4) the fixed-energy amplitude ĜD,ren1 [R, R; E] becomes 2ĜRS1 [R, R; E], which
implies VD (r) = 2VRS (r) which is summarized in Eq. (3.14).
Next, let us consider a case of Rv ren + 3/2 = 0. Since the small u region contributes
dominantly in the long-range, Eq. (3.27) becomes
∞
2 u sin((r/R)u)
VD (r) ∼ 2 3 du , (3.28)
π M Rr u2 + (2Rv ren + 3)
0
which reduces to
1 √
e− 2Rv +3 (r/R).
ren
VD (r) ∼ 3
(3.29)
πM Rr
The exponential suppression of VD (r) in this long-range behavior indicates that the gravity
localized on the Dirichlet brane is massive.
If r R, the same calculation makes VD (r) to be
∞
1 r
VD (r) ∼ 2 3 lim du sin u e−!u
π M Rr !→0 R
0
∞
3 sin((r/R)u)
− Rv ren + du , (3.30)
2 u + (Rv ren + 3/2)
0
where the infinitesimal parameter ! is introduced again for the regularization. Using the
integral formula
∞
sin ax
dx = f (aβ), (3.31)
x+β
0
where f (z) ≡ ci(z) sin(z) − si(z) cos(z), VD (r) becomes

1 R 3 r 3
VD (r) ∼ 2 3 − Rv + ren
f Rv +
ren
. (3.32)
π M Rr r 2 R 2
Thus the short-range behavior of VD (r) is governed by the renormalized coupling
constant. If (r/R)(Rv ren + 3/2) 1, VD (r) becomes

2R 2 (Rv ren + 3/2)−2 3 −2 12R 2
VD (r) ∼ 1 − Rv +ren
. (3.33)
π 2M 3r 4 2 r2
When deriving Eq. (3.33) we used the asymptotic expression [41]:

1 2! 4!
f (z) = 1 − 2 + 2 − ··· . (3.34)
z z z
It is interesting to note that the leading order in this case is 1/r 4 which should be a leading
term of the seven-dimensional Newton law. It is unclear at least for us why this peculiar
behavior of Newton potential arises on Dirichlet brane.
If (r/R)(Rv ren + 3/2) 1, we should use Eq. (3.23) which results in

1 π 3 r
VD (r) ∼ 2 3 2 1 − Rv +
ren
. (3.35)
π M r 2 2 R
Thus in this case VD (r) exhibits the 5d-type Newton potential.
4. Newton potential from fixed-energy amplitude: with 4d induced Einstein term
In this section we will compute the Newton potential on the brane when the 4d induced
Einstein term is involved. The remarkable feature in this case is an appearance of the
δ-function potential which has an energy- (or mass-)dependent coupling constant in the
linearized fluctuation equation [31,35]:
Ĥ2 ψ̂(z) = E ψ̂(z), Ĥ2 = Ĥ1 + λEδ(z), (4.1)
where λ ≡ M42 /M 3 and Ĥ1 is defined in Eq. (3.1). Thus, the fixed-energy amplitude for
Ĥ2 can be obtained by performing the Schulman procedure [39,40] again
Ĝ1 [a, c; E]Ĝ1[c, b; E]
Ĝ2 [a, b; E] = Ĝ1 [a, b; E] − . (4.2)
1
λE + Ĝ1 [c, c; E]
Inserting a = b = c in Eq. (4.2) yields the fixed-energy amplitude on the brane
λE Ĝ1 [c, c; E]
1
Ĝ2 [c, c; E] = . (4.3)
λE + Ĝ1 [c, c; E]
1
Thus, combining Eqs. (2.12) and (4.3) we can compute the Newton potential generated by
the exchange of the gravity localized on the brane.
Firstly, let us consider ξ = 1/2 case in which the fixed-energy amplitude reduces by
Eq. (4.3) to

m2 2 K2 (mR)
RS
Ĝ2 R, R; = . (4.4)
2 m 2K1 (mR) + λmK2 (mR)
Of course, we used c = R, g = 15/8 and v = 3/2R when deriving Eq. (4.4). Inserting
Eq. (4.4) into (2.12) the Newton potential becomes
∞
1 r K2 (u)
V2,RS (r) = 2 3 du sin u . (4.5)
π M Rr R 2K1 (u) + (λ/R)uK2 (u)
0
Firstly, let us consider the long-range behavior of V2,RS (r). Since small u region
contributes dominantly at long-range, one can approximate
1 u 1 1
K1 (u) ∼ + ln u, K2 (u) ∼ 2
− , (4.6)
u 2 u 2
which makes V2,RS (r) to be
1
V2,RS (r) ∼
π 2 M 3 R(2 + λ/R)r
∞ ∞

sin((r/R)u) 1 r −!u
× du − lim du ue sin u ln u .
u 2 + λ/R !→0 R
0 0
(4.7)
In Eq. (4.7) we introduced again the infinitesimal parameter ! for the proper regularization.
Making use of Eq. (3.11), one can show that V2,RS (r) at long-range obeys a 4d Newton
law with 1/r 3 subleading correction as following:

1 R2
V2,RS (r) ∼ 1+ . (4.8)
2πM 3 R(2 + λ/R)r (2 + λ/R)r 2
Next, let us examine the short-range behavior of V2,RS (r). Since the large u region
makes a dominant contribution to V2,RS (r), the asymptotic formulae

π −u 3 π −u 15
K1 (u) ∼ e 1+ , K2 (u) ∼ e 1+ , (4.9)
2u 8u 2u 8u
can be used. Then one can change V2,RS (r) into
1
V2,RS (r) =
π 2 M 3 λr
∞
sin((r/R)u) 15 1
× du +
u + (R/λ)(2 + 15λ/(8R)) 8 (R/λ)(2 + 15λ/(8R))
0
∞ ∞
sin((r/R)u) sin((r/R)u)
× du − du .
u u + (R/λ)(2 + 15λ/(8R))
0 0
(4.10)
Making use of the integral formula (3.31), V2,RS (r) at r R reduces to

1 2r 15r
V2,RS (r) ∼ 2 3 f +
π M λr λ 8R

15 1 π 2r 15r
+ −f + .
8 (R/λ)(2 + 15λ/(8R)) 2 λ 8R
(4.11)
If the quantum effect of the one loop is so large, i.e., λ R r, the Newton potential
V2,RS (r) becomes

1 15r 15 1 π 15r
V2,RS (r) ∼ 2 3 f + −f , (4.12)
π M λr 8R 8 15/8 + 2R/λ 2 8R
which reduces to approximately

1 4r 15r
V2,RS (r) ∼ 1 + ln . (4.13)
2πM 3 λr πλ 8R
Thus, the Newton potential recovers the 4d-type 1/r gravitational potential. If the quantum
effect of the one loop is very small compared to R, i.e., λ R, Eq. (4.11) becomes

1 2r
V2,RS (r) ∼ 2 3 f . (4.14)
π M λr λ
Thus in the region λ r R, which means the quantum effect is extremely small, the
Newton potential exhibits a 5d-type potential with 1/r 4 correction term as following

1 λ2
V2,RS (r) ∼ 1− 2 . (4.15)
2π 2 M 3 r 2 2r
When deriving Eq. (4.15) we have used the expansion (3.34). If r λ R, we should use
the expansion (3.23), which results in

1 4r 2r
V2,RS (r) ∼ 1 + ln . (4.16)
2πM 3 λr πλ λ
Thus at the extremely small-range the 4d Newton law is recovered. This result is also
confirmed in Refs. [31,35].
In Appendix A we examined the Newton potential when ξ = 1. Let us summarize the
result. The long-range behavior in this case is an exponential suppression like the picture
without the 4d induced Einstein term:

1 r 3
V2,N (r) ∼ − exp − . (4.17)
πM 3 (R + 2λ)r R 1 + 2λ/R
This indicates that the graviton localized on the Neumann brane is massive. In the region
of λ r R the Newton potential shows the 5d-type with 1/r 4 subleading correction:

1 2λ2
V2,N (r) ∼ 2 3 2 1 − 2 . (4.18)
π M r r
However, at the extremely short-range, i.e., r λ R, the potential recovers the 4d-type
Newton law:

1 2r r
V2,N (r) ∼ 1+ ln . (4.19)
2πM 3 λr πλ λ
In Appendix B we compute the Newton potential at ξ = 0. In this case the fixed-energy
amplitude as well as the Newton potential depend on the renormalized coupling constant
v ren as in the case without the 4d induced Einstein term. If Rv ren + 3/2 = 0, the final
expression of the Newton potential is related to V2,RS (r) as following:

V2,D (r) = 2V2,RS (r)λ→2λ . (4.20)
If Rv ren + 3/2 = 0, Newton potential is exponentially suppressed in the long-range as

the case in the absence of the 4d induced Einstein term as following:

1 r Rv ren + 3/2
V2,D (r) ∼ exp − . (4.21)
2πM 3 (λ + R/2)r R 1/2 + λ/R
This indicates that the graviton localized on the Dirichlet brane is massive. In the short-
range the Newton potential becomes

1 2λ2
V2,D (r) ∼ 2 3 2 1 − 2 , (4.22)
π M r r
when λ r R and

1 2r r
V2,D (r) ∼ 1 + ln , (4.23)
2πM 3 λr πλ λ
when r λ R. Thus, we have an intermediate range λ r R in which the 5d
Newton law emerges. It is interesting to note that V2,D (r) at short-range is independent
of the renormalized coupling constant v ren . Furthermore, the potential on Dirichlet brane
coincides with that on Neumann brane in the short-range.
5. Conclusion
In this paper we examined Newton law generated by the localized gravity on the general
singular brane when the bulk spacetime is a single copy of AdS5 . We used a fixed-energy
amplitudes which are obtained from the linearized gravitational fluctuation equation by
applying the technique of SQM. Since the bulk is a single copy of AdS5 , SQM naturally
makes the fixed-energy amplitude to be dependent on the BC at the location of the brane.
We examined Newton potential on RS brane (ξ = 1/2), Neumann brane (ξ = 1), and
Dirichlet brane (ξ = 0).
For RS brane the usual RS scenario gives rise to the 4d-type Newton potential at long-
range and 5d-type at short-range. However, the 4d Einstein term induced by a quantum
effect of one-loop changes the general behavior of Newton potential. In this case there is
an intermediate range λ r R, in which Newton potential is five-dimensional. At other
ranges the four-dimensional Newton law is recovered.
For Neumann brane the long-range Newton potential is exponentially suppressed
regardless of the existence of the 4d induced Einstein term. This is because the gravity
localized on the Neumann brane acquires a mass. In the short-range Newton potential
becomes five-dimensional with a logarithmic subleading correction if there is no 4d
induced Einstein term. The Einstein term in Neumann brane yields a similar intermediate
range to that in RS brane, where Newton law is five-dimensional.
For Dirichlet brane the non-trivial fixed-energy amplitude can be derived via the
coupling constant renormalization scheme. Thus, Newton potential on the Dirichlet brane
is usually dependent on the renormalized coupling constant v ren . If v ren = −3/2R, the
final expression of Newton potential is proportional to that in RS brane. If v ren = −3/2R,
the long-range behavior of Newton potential is exponentially suppressed due to the massive
gravity regardless of the 4d induced Einstein term. In short-range Newton potential exhibits
an peculiar 1/r 4 behavior if (r/R)(Rv ren +3/2) 1. However, the 4d Einstein term makes
the short-range behavior of Newton potential on Dirichlet brane to be exactly identical to
that on Neumann brane.
Recently, Newton law arising due to the localized gravity on various curved brane
is examined when bulk is dS5 or AdS5 [42]. Our SQM technique may be applied
straightforwardly to these various scenarios. The explicit result will be given elsewhere.
Acknowledgement
This work was supported by the Kyungnam University Research Fund, 2003.
Appendix A
In this appendix we will examine the Newton potential arising due to the gravity
localized on the Neumann brane when the 4d induced Einstein term exists. The fixed-
energy amplitude for ξ = 1 with c = R, g = 15/8 and v = 3/2R yields
2 K2 (mR)
m2 m K1 (mR)
ĜN
2 R, R; = K2 (mR) (A.1)
2 1 + λm − 2mR 3
K1 (mR)
and the corresponding Newton potential derived from Eq. (2.12) is
∞ K2 (u)
1 r K (u)
V2,N (r) = du sin u λ 1 3 K2 (u) . (A.2)
π 2 M 3 Rr R 1+ u−
0 R 2u K1 (u)
If two unit masses are separated at long distance compared to the AdS5 radius, i.e., r R,
the small u region contributes to V2,N (r) dominantly. Thus we can use an expansion (4.6)
for the modified Bessel function, which results in approximately
∞
2 r u
V2,N (r) ∼ − 2 3 du sin u . (A.3)
π M (R + 2λ)r R 1+2λ/R + u
3 2
0
Making use of the integral formula
∞
x sin(ax) π
dx = e−aβ , (A.4)
β +x
2 2 2
0
one can show that the gravitational force on the Neumann brane is exponentially
suppressed as following:

1 r 3
V2,N (r) ∼ − exp − . (A.5)
πM 3 (R + 2λ)r R 1 + 2λ/R
If two unit masses are separated at short distance, the large u region contributes
dominantly. Thus we can use the asymptotic formula (4.9), which results in the following
Newton potential
1
V2,N (r) ∼
π 2 M 3 λr
∞
sin((r/R)u) 15/8
× du +
u + ((15/8) + R/λ) 15/8 + R/λ
0
∞ ∞
× du − du . (A.6)
u u + (15/8 + R/λ)
0 0
Computing the integrations of Eq. (A.6) with use of Eq. (3.31), one can express V2,N (r) at
short-range as

1 15r r 15/8 π 15r r
V2,N (r) ∼ 2 3 f + + −f + . (A.7)
π M λr 8R λ 15/8 + R/λ 2 8R λ
If λ R, V2,N (r) in Eq. (A.7) becomes

1 15r r
V2,N (r) ∼ 2 3 f + . (A.8)
π M λr 8R λ
Thus using the expansion of the sine and cosine integral functions (3.23) and (3.34), the
Newton potential on the Neumann brane becomes

1 2λ2
V2,N (r) ∼ 2 3 2 1 − 2 (A.9)
π M r r
in the region of λ r R, and

1 2r r
V2,N (r) ∼ 1+ ln (A.10)
2πM 3 λr πλ λ
in the region of r λ R.
Appendix B
In this appendix we will examine the Newton potential arising due to the gravity
localized on the Dirichlet brane when the 4d induced Einstein term exists. The fixed-energy
amplitude for ξ = 0 case can be obtained from Eq. (4.3) with replacing Ĝ1 [c, c; E] by a
renormalized fixed-energy amplitude Ĝren1 [c, c; E]. Then using Eq. (3.26) one can easily
calculate the fixed-energy amplitude

m2 2R
D
Ĝ2 R, R; = (B.1)
2 (Rv + 3/2) + mR K
ren 1 (mR)
K2 (mR) + Rλm
2
which leads the Newton potential to

∞
1 u sin((r/R)u)
V2,D (r) = 2 3 du . (B.2)
π M Rr (Rv ren + 3/2) + u K 1 (u)
K2 (u) + λ/Ru
2
0
Firstly, let us examine the case of Rv ren + 3/2 = 0. Comparing Eq. (B.1) with Eq. (4.4)
one can conclude

m2 m2
D
Ĝ2 R, R; = 2Ĝ2 R, R;
RS
. (B.3)
2 2 λ→2λ
Thus, the Newton potential V2,D (r) in this case can be read from V2,RS (r) as following:

V2,D (r) = 2V2,RS (r)λ→2λ . (B.4)
Next, let us examine the case of Rv ren + 3/2 = 0. Using the expansion (4.6) one can
show V2,D (r) reduces to in the long-range, i.e., r R,
∞
1 u sin((r/R)u)
V2,D (r) ∼ 2 3 du Rv ren +3/2
, (B.5)
π M (λ + R/2)r + u2
0 1/2+λ/R
which exhibits an exponential suppression

1 r Rv ren + 3/2
V2,D (r) ∼ exp − . (B.6)
2πM 3 (λ + R/2)r R 1/2 + λ/R
In the short-range we should use the expansion (4.9) which makes V2,D (r) to be
1
V2,D (r) ∼
π 2 M 3 λr
∞
sin((r/R)u) 15/8
× du +
u + (15/8 + R/λ) 15/8 + R/λ
0
∞ ∞
× du − du . (B.7)
u u + (15/8 + R/λ)
0 0
It is interesting to note that V2,D (r) is independent of the renormalized coupling constant
v ren . Furthermore, V2,D (r) in Eq. (B.7) is exactly same with V2,N (r) in Eq. (A.6). Thus
the short-range behavior of V2,D (r) should be same with that of V2,N (r).
References
[1] H.C. Lee (Ed.), An Introduction to Kaluza–Klein Theories, World Scientific, Singapore, 1984.
[2] V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 125 (1983) 139.
[3] M. Visser, Phys. Lett. B 159 (1985) 22.
[4] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[5] L. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[6] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[7] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
[8] P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209.
[9] P. Binetruy, C. Deffayet, D. Langlois, Nucl. Phys. B 565 (2000) 269, hep-th/9905012.
[10] C. Csáki, M. Graesser, C. Kolda, J. Terning, Phys. Lett. B 462 (1999) 34, hep-ph/9906513.
[11] J.M. Cline, C. Grojean, G. Servant, Phys. Rev. Lett. 83 (1999) 4245, hep-ph/9906523.
[12] J.E. Kim, B. Kyae, H.M. Lee, Phys. Rev. Lett. 86 (2001) 4223, hep-th/0011118.
[13] S. Alexander, Y. Ling, L. Smolin, Phys. Rev. D 65 (2002) 083503, hep-th/0106097.
[14] D.K. Park, H.S. Kim, S. Tamaryan, Phys. Lett. B 535 (2002) 5, hep-th/0111081.
[15] N. Arkani-Hamed, H.C. Cheng, P. Creminelli, L. Randall, hep-th/0302034, and references therein.
[16] A. Chamblin, S.W. Hawking, H.S. Reall, Phys. Rev. D 61 (2000) 065007, hep-th/9909205.
[17] R. Emparan, G.T. Horowitz, R.C. Myers, JHEP 0001 (2000) 007, hep-th/9911043.
[18] S.B. Giddings, E. Katz, L. Randall, JHEP 0003 (2000) 023, hep-th/0002091.
[19] S.B. Giddings, S. Thomas, Phys. Rev. D 65 (2002) 056010, hep-th/0106219.
[20] S.B. Giddings, hep-th/0205027.
[21] S.B. Giddings, Gen. Relativ. Gravit. 34 (2002) 1775, hep-th/0205205.
[22] J. Garriga, T. Tanaka, Phys. Rev. Lett. 84 (2000) 2778, hep-th/9911055.
[23] M.J. Duff, J.T. Liu, Phys. Rev. Lett. 85 (2000) 2052, hep-th/0003237.
[24] D.K. Park, S. Tamaryan, Phys. Lett. B 532 (2002) 305, hep-th/0108068.
[25] D.K. Park, H.S. Kim, Nucl. Phys. B 636 (2002) 179, hep-th/0204122.
[26] A.Z. Capri, Nonrelativistic Quantum Mechanics, Benjamin–Cummings, Menlo Park, CA, 1985.
[27] M. Reed, B. Simon, Methods of Modern Mathematical Physics, Academic Press, New York, 1975.
[28] R. Jackiw, in: A. Ali, P. Hoodbhoy (Eds.), M.A. Bég Memorial Volume, World Scientific, Singapore, 1991.
[29] D.K. Park, J. Math. Phys. 36 (1995) 5453, hep-th/9405020.
[30] D.K. Park, S. Tamaryan, Phys. Lett. B 554 (2003) 92, hep-th/0212023.
[31] D.K. Park, Phys. Lett. B 562 (2003) 316, hep-th/0304056.
[32] G. Dvali, G. Gabadadze, M. Porrati, Phys. Lett. B 485 (2000) 208, hep-th/0005016.
[33] G. Dvali, G. Gabadadze, M. Kolanović, F. Nitti, Phys. Rev. D 64 (2001) 084004, hep-ph/0102216.
[34] V.A. Rubakov, hep-th/0303125.
[35] E. Kiritsis, N. Tetradis, T.N. Tomaras, JHEP 0203 (2002) 019, hep-th/0202037.
[36] M. Ito, Phys. Lett. B 554 (2003) 180, hep-th/0211268.
[37] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[38] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hep-th/9905111.
[39] B. Gaveau, L.S. Schulman, J. Phys. A 19 (1986) 1833.
[40] L.S. Schulman, in: M.C. Gutzwiller, A. Inomata, J.R. Klauder, L. Streit (Eds.), Path Integral from meV to
MeV, World Scientific, Singapore, 1986.
[41] M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions, Dover, New York, 1972.
[42] S. Nojiri, S.D. Odintsov, Phys. Lett. B 548 (2002) 215, hep-th/0209066, and references therein.
A note on α-vacua and interacting field theory

in de Sitter space
Kevin Goldstein, David A. Lowe
Department of Physics, Brown University, Providence, RI 02912, USA
Received 24 February 2003; received in revised form 8 July 2003; accepted 22 July 2003
Abstract
Using an imaginary time formalism, we propose a consistent renormalizable perturbation theory
of a scalar field in a non-trivial α vacuum in de Sitter space. Although one representation of the
effective action involves non-local interactions between anti-podal points, we argue the theory leads
to causal physics when continued to real-time, and we prove a spectral theorem for the interacting
two-point function. We construct the renormalized stress energy tensor and show this develops no
imaginary part at leading order in the interactions, consistent with stability.
PACS: 04.62.+v; 98.80.Cq
1. Introduction
A common problem in formulating quantum field theory on a curved background is

ambiguity in the choice of vacuum. In de Sitter space there is a one-parameter family of
vacua invariant under the de Sitter group, which have been dubbed the α vacua [1–4].
These vacua are perfectly self-consistent in the context of free theories. It has long been
suggested that the only physically sensible vacuum is the Euclidean (a.k.a. Bunch–Davies)
vacuum.
One reason for this choice is that the free propagators in the Euclidean vacuum exhibit
a Hadamard singularity, which matches with what is expected in the flat space limit
[5,6]. However the physical motivation for restriction to Hadamard singular propagators
is obscure in the context of interacting quantum field theory, and certainly nothing appears
to go wrong with the α-vacuum propagators at the free level. In particular, as shown in [4]
E-mail addresses: kevin@het.brown.edu (K. Goldstein), lowe@het.brown.edu (D.A. Lowe).
doi:10.1016/j.nuclphysb.2003.07.014
326 K. Goldstein, D.A. Lowe / Nuclear Physics B 669 (2003) 325–340
the commutator Green function is vanishing at spacelike separations in an α-vacuum and

in fact is independent of α.
The Green functions in a non-trivial α vacuum exhibit singular correlations between
anti-podal points. Of course since the commutator is compatible with locality this does
not lead to a causal propagation of information. However some authors have suggested
that once interactions are included the α-vacua do not lead to a sensible perturbative
expansion of Green functions. Banks et al. [7] have argued non-local counter-terms render
the effective action inconsistent. Einhorn et al. [8] have argued that α vacuum correlation
functions are non-analytic and conclude that they are physically unacceptable. Related
arguments are made in Kaloper et al. [9], who argue that because an Unruh detector is not
in thermal equilibrium in an α vacuum, thermalization will lead to decay to a Euclidean
vacuum.
This issue has direct bearing on the theory of inflation. The conventional view of
inflation places the inflaton in the Euclidean vacuum. However, as emphasized in [10–
12], the initial conditions for inflation may place the inflaton in a non-trivial α-vacuum,
see also [13] for earlier work in this direction. This has a potentially large effect on the
predictions for the CMB spectrum, see for example [10–12,14] and references therein.
Furthermore, if there is a residual value of α today, there are many interesting predictions
for other observable quantities such as cosmic rays, that we consider in more detail in [15].
In this paper we argue the α-vacua do indeed have a well-defined perturbative
expansion in an imaginary time formulation, that yields finite renormalized amplitudes
in a conventional manner. This goes a long way to refuting some of the objections raised
in [7–9], see also [16] for discussion of consistency of α-vacua.
We begin in Section 2 by reviewing the free field results of [1–4]. In particular, the α-
vacuum may be regarded as a squeezed state created by a unitary operator U acting on the
Euclidean vacuum. This idea will be central to the formalism we develop. This leads to a
generalized Wick’s theorem, which allows us to expand any free α-vacuum Green function
in terms of products of Euclidean vacuum two-point functions. In Section 3 we describe
the interacting theory in an imaginary time formalism. In particular, in interaction picture,
we show the effective Lagrangian becomes non-local when U is commuted through the
fields. We show that UV divergences in amplitudes satisfy a non-trivial factorization
relation which relates the coefficients of local counter-terms to non-local ones. Once
local counter-terms are fixed, non-local terms are completely determined, which implies
the theory is renormalizable in the conventional sense. In Section 4 we outline how to
continue the imaginary time amplitudes to real time. We carry this through in detail for the
interacting two-point function, and prove a spectral theorem in this case. One immediate
consequence is that even in the interacting theory, the expectation value in an α-vacuum of
the commutator of two fields vanishes at spacelike separations, as required for causality of
local observables. In Section 5 we use the Green function to define a renormalized stress
energy tensor, and we conclude in Section 6.
K. Goldstein, D.A. Lowe / Nuclear Physics B 669 (2003) 325–340 327
2. Free fields
Let us begin by reviewing the construction of the α-vacua [1–4]. We will set the
Hubble radius 1/H = 1 unless otherwise stated. In general a free scalar field has the mode
expansion

φ(x) = an φn (x) + an† φn∗ (x), (1)
n
†
where {φn } satisfy the Klein–Gordon equation and [an , am ] = δnm . The φn are complete
and orthonormal with respect to the Klein–Gordon product

φ1 (x), φ2 (x) = −i φ1 ∂µ φ2∗ − φ2∗ ∂µ φ1 dΣ µ (2)
Σ
with Σ a spacelike slice. In this paper we consider scalars with mass m and Rφ 2
coupling ξ . In a fixed de Sitter background we can absorb ξ into a redefinition of m, so we
drop ξ from now on.
The vacuum state is characterized by
an |Ωa
= 0. (3)
In general we can expand in another set of modes

φ(x) = bn φnα (x) + bn† φnα ∗ (x) (4)
n
related to the first by a Bogoliubov transformation. A new vacuum state is defined by
bn |Ωb
= 0. (5)
As mentioned in dS space there is a one complex-parameter family of the dS invariant
vacua, dubbed the alpha vacua, |α
. One of these, the Euclidean or Bunch–Davies
vacuum, |E
, is defined by using mode functions obtained by analytically continuing mode
functions regular on the lower half of the Euclidean sphere. The Euclidean modes can
be chosen such that φnE (x)∗ = φnE (x̄), where x̄ is the anti-pode of x, (x̄ ≡ −x). See, for
example, [3,17] for explicit expressions of these mode functions.
The modes of an arbitrary α-vacuum general are related to the Euclidean ones by a
mode number independent Bogoliubov transformation,
∗
φnα = Nα φnE + eα φnE , (6)
where Nα ≡ (1 − exp(α + α ∗ ))−1/2 and we require that Re(α) < 0. The Euclidean vacuum
corresponds to α = −∞.
In terms of creation and annihilation operators

bn = Nα an − eα∗ an† , (7)
where b and a are operators satisfying (3) with respect to the α-vacuum and Euclidean,
respectively. This transformation can be implemented using a unitary operator
bn = Uan U † , (8)
where

†2 ∗ 2
Uα ≡ exp cα a n − cα a n , (9)
n

1 Re(α)
cα ≡ exp −i Im(α) log tanh − (10)
4 2
and we use the standard Taylor expansion of the exponential to define the ordering. The
vacua are related by
|α
= Uα |E
(11)
since
bn U|E
= Uan |E
= 0. (12)
From this perspective the α-vacuum may be viewed as a squeezed state on top of the usual
Euclidean vacuum.
2.1. Generalized Wick’s theorem
The Fock space built on the α vacuum is not unitary equivalent to that of the Euclidean
vacuum in general, because the unitary transformation mixes positive and negative
frequencies. However, as far as quantum field theory in a fixed de Sitter background goes,
this unitary transformation leaves the complete set of physical observables invariant. For
our purposes, we take these observables to be finite time Green functions, from which one
may obtain S-matrix elements as described in [18,19]. All this unitary transformation does
is to mix these observables up in a non-local way, as we explain in more detail later in this
section. This is the underlying reason that the interacting α-vacuum theory is consistent.
If we were only considering the φ field on its own this would be the end of the story.
However if we wish to view φ as the inflaton, physics dictates that φ should be locally
coupled to other fields. Thus we are interested in correlators of the field φ with respect to
the α vacuum. The conjugated field Uφ(x)U † on the other hand, would yield correlators
in the α-vacuum identical to the usual Euclidean vacuum correlators of φ, but would be
coupled non-locally to other fields.
Since the unitary transformation involves modes of arbitrarily high frequency (up to
some physical cut-off) the systematics of renormalizable perturbation theory will be quite
different from the usual Euclidean vacuum perturbation theory [20–25]. It will be our goal
in the rest of this paper to elaborate on renormalizable perturbation theory in the α-vacuum.
The correlators of interest take the form of expectation values of products of fields φ
with respect to the state |α
, or equivalently as conjugated fields φ̃ ≡ U † φU with respect
to |E
α|φ(x1 )φ(x2 ) · · · φ(xn )|α

= E|U † φ(x1 )UU † φ(x2 )UU † · · · UU † φ(xn )U|E
= E|φ̃(x1 )φ̃(x2 ) · · · φ̃(xn )|E

. (13)
Now, letting γ = eα (so that |γ | < 1),
φ̃(x) = U † φ(x)U

=U †
φn (x)bn + φn (x)bn U
α α∗ †
n

= φnα (x)an + φnα∗ (x)an†
n
∗
= Nα φnE (x) + γ φnE (x̄) an + φnE (x) + γ φnE (x̄) an†
n

= Nα φ0 (x) + φ1 (x) , (14)
∗
φ0 (x) ≡ φ(x), φ1 (x) ≡ γ φnE (x̄)an + γ ∗ φnE (x̄)an† . (15)
n
If γ is real, then φ̃(x) is simply a linear combination of φ(x) and φ(x̄). For γ complex this
is not quite true, but the additional phases are simple to keep track of.
Using these relations we can express any α-vacuum correlator in terms of a sum
of Euclidean vacuum correlators, giving us a generalized Wick’s theorem. The simplest
example is

α|φ(x)φ(y)|α
= Nα2 GE (x, y) + |γ |2 GE (x̄, ȳ) + γ GE (x̄, y) + γ ∗ GE (x, ȳ)
≡ Gα (x, y), (16)
where GE (x, y) ≡ E|φ(x)φ(y)|E
is the Wightman function on the Euclidean vacuum.
It is convenient to introduce a two index notation,

Gα (x, y) = Gij (x, y), (17)
i,j =0,1
where
G00 (x, y) = Nα2 GE (x, y), G10 (x, y) = Nα2 γ GE (x̄, y),
G01 (x, y) = Nα2 γ ∗ GE (x, ȳ), G11 (x, y) = Nα2 |γ |2 GE (x̄, ȳ), (18)
which we will use later.
The Wightman function diverges when x and y are null separated. As we can see from
(16), for the α-vacua, there are additional divergences when one point is null separated with
the anti-pode of another. (See Fig. 1.) This feature has led many to consider the α-vacua
unphysical [7–9].
Any correlation function of the form (13) in the free theory can be found in terms of
products of Green’s functions by normal-ordering the creation and annihilation operators,
and retaining the fully contracted terms. For example,
α|φ(x)φ(y)φ(z)φ(w)|α
= Gα (x, y)Gα (z, w) + Gα (x, z)Gα (y, w) + Gα (x, w)Gα (y, z), (19)
Fig. 1. Feynman diagram for (16). The thick line represents the α-vacuum propagator Gα (x, y). Thin lines
represent Euclidean vacuum propagators with a factor of Nα2 , and the other factors are shown explicitly. Grey
dots denote points that appear in propagators as anti-podes. We have defined γ = eα .
where we have in mind using (16) to expand in terms of the Euclidean vacuum Green’s
functions. Note it is important the ordering of the arguments of the Green’s functions
is inherited from the ordering in the operator expression on the left-hand side. This is
because we are stating the generalized Wick’s theorem in the form of Wightman functions
rather than the usual form with time-ordered Green’s functions [26]. The expansion
of operator products in free field theory using Wightman functions actually predates
Wick’s theorem [27]. The theorem may be extended to time-ordered expectation values
by replacing the Wightman functions with time-ordered two-point functions.
To convert some diagram written in terms of the Gα ’s into one in terms of Euclidean
propagators, replace each thick line with a sum of 4 thin ones. To find the coefficient of
each term just count up the number of grey dots, noting their orientation with respect to the
arrows. This is written more compactly using the two index notation (17), with an index
i = 0, 1 appearing at the end of each propagator, and all indices summed over.
3. Interacting fields
So far all we have said is valid regardless of whether we work on de Sitter space, or
its Euclidean continuation, the four-sphere. Once we introduce interactions, however, the
choice of Lorentzian versus Euclidean signature has a major impact on the formalism
used to setup the perturbative expansion. This is familiar from finite temperature field
theory where one has an imaginary time formalism [28,29] or alternatively one can use
a formulation in terms of real time propagators at the price of doubling the number of
fields [30,31].
Describing interacting fields in curved spacetime with event horizons using Lorentzian
signature formalism is problematic. Inevitably one must deal with propagators on opposite
sides of the horizon, and this leads to ambiguities in the formulation of Feynman rules.
The same problem exists for spacetimes with cosmological horizons, as would arise if one
attempted to quantize a field in de Sitter space in the static coordinate patch. To avoid
these issues we formulate interacting field theory using the imaginary time, or Euclidean
continuation, as advocated in [32]. Eventually we have in mind defining real-time ordered
correlators which may be used to construct in–out S-matrix elements as described in
[18,19].
In the Euclidean vacuum, this problem has been much studied in the literature [20–
25,33] and corresponds to doing field theory on S 4 . Our strategy will be to use (11) to
define the interacting α-vacuum, and construct physical observables as correlators of the
field φ which couple locally to physical sources. To evaluate these observables we set up
the perturbation theory on the Euclidean sphere as a non-local field theory in terms of
φ̃ = U † φ(x)U . We define these fields as interaction picture fields, and will work with the
standard methods of canonical quantization.
The interacting part of the non-local action for φ̃ is obtained by conjugating the local
bare interaction terms written in terms of the φ fields with the operator U , or more precisely
int int
Tτ eiSnon-local (φ̃) ≡ U † Tτ eiSlocal (φ) U, (20)
where Tτ denotes imaginary time ordering. The actions are obtained by integrating the
Lagrangian density describing the interactions (which we assume to be polynomial in φ)
over the Euclidean sphere. We note that when (14) is substituted into this expression, the
anti-podal components φ1 (x) are to be ordered according to x rather than x̄, since the
ordering is to be inherited from the right-hand side of (20). The determination of any
correlation function of φ’s in an α-vacuum then reduces to a standard Euclidean vacuum
correlator computation, albeit with some terms involving fields with unconventional time
ordering. That is, we expand (20) perturbatively in the interactions, generating a sum of
correlators of φ̃ with respect to the Euclidean vacuum. These may then be evaluated using
the generalized Wick’s theorem of the previous section.
Working on the Euclidean sphere has the advantage that a wide range of sensible cut-
offs are available. For example, one can choose dimensional regularization as in [22–25],
or simply a mode cut-off corresponding to a cut-off on angular momentum on the 4-sphere.
Pauli–Villars is another option, as is point-splitting (with spherically symmetric averaging
assumed to restore the symmetries), or zeta-function regularization [19,34,35]. Little of
what we say in the present work is dependent on a particular choice of cut-off.
Let us comment further on the form of the correlators. The normalized Green functions
in the α-vacuum take the form
int
U † Tτ φ(x1 ) · · · φ(xn )eiSlocal (φ) U
G(x1 , . . . , xn ) = int
U † Tτ eiSlocal (φ) U
int
Tτ φ̃(x1 ) · · · φ̃(xn )eiSnon-local (φ̃)
= int
. (21)
Tτ eiSnon-local (φ̃)
In imaginary time we cannot take an asymptotic limit where interactions turn off, which is
important in the usual definition of the S-matrix to obtain the interacting vacuum. Instead
we will simply compute correlators with respect to the free vacuum as in (21). As usual
the denominator in (21) implies we drop disconnected diagrams when we compute Green
functions.1 Because we are not taking an LSZ-type limit, the relevant Green functions
correspond to unamputated diagrams. We discuss continuation to real-time amplitudes in
the next section.
Let us now go through an example to illustrate the renormalization of mass in λφ 3 . The
relevant Feynman diagram in position space is shown in (2). The vertex Vij k = 1 for all
1 By disconnected we mean diagrams not connected to external lines.

Fig. 2. Feynman diagram for propagator in λφ 3 . The indices ik label end-points of the propagators Gij . All
indices are to be summed over. Vertex factors carry no ik dependence.
i, j, k, so the amplitude is

A= det gµν (w) dw det gµν (z) dz
{ik =0,1}
× GFi1 i2 (x, w)GFi3 i4 (w, z)GFi5 ,i6 (w, z)GFi7 i8 (z, y), (22)
where GF is the time-ordered Green function

GF (x, y) = θ x 0 − y 0 Gα (x, y) + θ y 0 − x 0 Gα (y, x), (23)
and x 0 is the imaginary time coordinate on the sphere. UV divergences arise when w → z
or w → z̄. In these limits the propagator has the form
1
lim GF (w, z) = Nα2 1 + |γ |2 ,
w→z (w − z)2
1
lim GF (w, z) = Nα2 (γ + γ ∗ ) (24)
w→z̄ (w − z̄)2
in locally Minkowski coordinates. The UV divergent part of the amplitude is then

δm2
AUV = det gµν (z) dz U † φ(x)φ(z)2 φ(y)U , (25)
2
where δm2 is the cut-off dependent counter-term. If we adopt a simple point-splitting
regularization, this is given by

δm2 = + det gµν (w) dw Gi3 i4 (w, z)Gi5 i6 (w, z)
|w−z|ε |w−z̄|ε ik
2 2
∝ Nα4 1 + |γ |2 + γ + γ ∗ log ε. (26)
We conclude therefore that the counter-term is indeed simply a local mass counter-term
when expressed in terms of φ variables, but appears non-local when written in terms of φ̃
variables.
It is perhaps worthwhile to highlight the difference between our computation and a
similar computation of [7]. Ref. [7] assumed the basic vertex was local. However in our
formulation of the α-vacuum field theory, the vertex takes the form U † φ(z)3 U which
looks non-local when we expand this out in terms of the fields φ0 (z) and φ1 (z), since
we can view φ1 as localized at z̄. This non-locality is exactly what we need to make sense
of the non-local counter-terms encountered in [7]. When all the diagrams are included
the coefficients of the non-local counter-terms are such that they arise from the local
counter-term (1/2)δm2φ(x)2 , prior to conjugation by the U ’s. This implies the α-vacuum
perturbation theory is rendered finite by the same number of renormalization conditions as
the corresponding Euclidean vacuum theory.
4. Real-time correlators and causality
We now discuss how to continue the imaginary time Green functions to real time.
In general this procedure is rather difficult as the analytic continuation is not uniquely
defined. One encounters similar problems in the formulation of Minkowski space quantum
field theory at finite temperature [36–39]. There the analytic continuation from imaginary
time to real-time, with the extra condition that propagators be analytic in the lower
half frequency plane, computes retarded Green functions. Retarded and advanced Green
functions may be further combined to give real time-ordered Green functions. This
procedure of determining propagators by analytic continuation can be avoided by working
with the real-time thermo-field expansion of [30], where the field content is doubled. In
thermo-field theory, one also can formulate a non-perturbative path integral definition of
the theory using a non-trivial real-time integration contour. It would be interesting to see
if the α-vacuum theory could be formulated in an analogous way. We will not develop
that here, but content ourselves for the moment with the perturbative description of the
theory described in the previous section. We will use these results to obtain the analytic
continuation to real-time of the general interacting two-point function.
The general two-point function G in the interacting theory is
G(x, y) = Ω|φ(x)φ(y)|Ω
(27)
and is perturbatively defined by (21). We use φ(x) to denote Heisenberg operators in this
section. We can insert a complete set of states to obtain

G(x, y) = Ω|φ(x)|χ, n
χ, n|φ(y)|Ω
, (28)
χ,n
where |χ, n
denotes a scalar state with quantum numbers n. We now use de Sitter sym-
metry to translate φ(x) = T φ(0)T −1 , where T is a de Sitter translation. |Ω
is invariant
under this translation. Usually one would assume T −1 |χ, n
= φnE (x)|χ
, but as we have
learned, invariance under the subgroup of the de Sitter group continuously connected to
the identity, in general only implies T −1 |χ, n
= Nα(χ) (φnE (x) + eα(χ) φnE∗ (x))|χ
, where
α is now a function of the state χ , and φn (x) is the generalization of the modes to a scalar
field of general mass m(χ). This implies

G(x, y) = ρ(χ)Gα(χ) m(χ); x, y , (29)
χ
where ρ is positive semi-definite. We can choose to parameterize this instead as

∞
G(x, y) = dm dα dα ∗ ρ(m, α)Gα (m; x, y), (30)
mmin
where Gα (m; x, y) is the generalization of (18) to a free field of mass m. In [40] it

was argued mmin = 3/2 for the theory in the Euclidean vacuum. The m > 3/2 scalar
representations of the de Sitter group are known as the principal series, and only these have
a smooth limit to representations of the Poincare group as H → 0 [41]. The 0 < m < 3/2
representations are known as the complementary series. There do not appear to be any
obvious problems with quantizing
√ fields with 0 < m < 3/2. For example, the conformally
coupled free scalar (m = 2 ), is related by a conformal transformation to a massless
field in flat space. Since we are often interested in fields with m < 3/2, we include the
complementary series in our space of allowed states, so take mmin = 0.
The iε prescriptions for the propagators Gj k are defined in Appendix B, which allows
the G to be continued to a function regular on the Lorentzian section. This prescription is
fixed by imposing the boundary condition that each component of the two-point function
Gα (m; x, y) match the free Wightman propagators constructed by Mottola and Allen [3,4].
Appropriate linear combinations of G(x, y) define the real-time retarded, advanced, and
time-order propagators. We note the complete propagator G is not analytic in the lower
half t plane (see Appendix B for notation), but it is built out of terms, each of which
separately enjoys analyticity in the upper or lower half t plane.
Demonstrating causality of the interacting two-point function is now trivial. We simply
apply (30) to the commutator of two fields

φ(x), φ(y) α = G(x, y) − G(y, x)
∞

= dm dα dα ∗ ρ(m, α) Gα (m; x, y) − Gα (m; y, x)
0
∞

= dm dα dα ∗ ρ(m, α) GE (m; x, y) − GE (m; y, x) , (31)
0
which vanishes at spacelike separations of x and y. Here we have used the result of [4] that
the commutator in the free theory is independent of α.
To sum up, we have defined a continuation of the general interacting two-point function
from imaginary time to real time, using a spectral theorem and we have shown the real-
time commutator function is causal despite the apparent non-analyticity of the perturbative
expansion.
5. Stress-energy tensor
Numerous techniques for calculating Tµν

in general, and in the Euclidean vacuum of
de Sitter space in particular, are reviewed in [19,34,35] and references therein. Since [42]
considered the stress-energy tensor in the α-vacua some time ago, we mainly quote their
results. The stress-energy tensor for a scalar field is given by
1
Tµν = φ,µ φ,ν − gµν φ ,α φ,α + gµν V (φ), (32)
2
where, for simplicity, we have set the Rφ 2 coupling to 0.

In general, we can find the renormalized expectation value of Tµν for a non-interacting
(1)
scalar field from the symmetric Greens function Gxy = {φx , φy }
, as follows:
free
Tµν (x) ren

1 1 2 1 (1)
= lim ∇µ ∇ν − gµν ∇ ∇γ + m gµν
γ
G (x , x ) − Gref (x , x ) ,

x ,x →x 2 2 2
(33)
where Gref is a reference two-point function which removes the singularities in G(1) . Note
the limit, and Gref must be chosen to preserve covariance and T µν ;ν = 0.
Consistent with previous work cited above [42] found for a non-interacting Euclidean
vacuum [34],

gµν
Tµν
free
ren = − 2
m2 m2 − 2H 2
64π
2
3 3 H
× ψ − iν + ψ + iν + ln
2 2 µ2

8 359 4
+ m2 H 2 − H , (34)
3 180

where H is Hubble’s constant, ν = m2 /H 2 − 9/4 and µ is some mass renormalization
scale. Tµν
for a general α-vacuum, with a non-interacting scalar field, has been found by
[42] to be

gµν 1 + |γ |2
Tµν
ren = −
free
m2 m2 − 2H 2
64π 1 − |γ |
2 2
2
3 3 H
× ψ − iν + ψ + iν + ln
2 2 µ2

π γ +γ∗ 8 2 2 359 4
+ + m H − H . (35)
cosh πν 1 + |γ |2 3 180
The main difference between (34) and (35) is an extra factor of (1 + |γ |2 )/(1 − |γ |2 ).
The origin of this constant can be seen from the short distance limit of (16) which
gives Gα (x, x) ∼ (1 + |γ |2 )/(1 − |γ |2 )GE (x, x). The α-dependence of the short distance
singularity means that our counter terms must be α-dependent. The fact that α-dependent
counter-terms are required for a finite Tµν
was viewed as problematic in [9]. As we have
already emphasized previously these are precisely the sort of counter-terms we naturally
expect. We emphasize both (34) and (35) are proportional to gµν which is covariantly
constant, implying conservation of energy.
An important conclusion we draw from (35) is that no imaginary part appears in Tµν
(and hence the action at one-loop order). This indicates the α-vacuum is stable at this order.
We discuss the possibility of higher-order instabilities in the conclusions.
Now to calculate the Tµν (x)
for the interacting case we need to replace the free Green
function with the interacting one (and add in VI (φ)
). The spectral representation (30)
then yields a straightforward generalization of (35).
6. Conclusions
We have proposed a renormalizable perturbation theory for scalar field amplitudes in an

α vacuum using an imaginary time formulation. We have shown the theory is causal when
continued to real-time, at the level of the two-point function. It remains an interesting open
problem to use this formalism to construct the higher-order real-time correlators. Our hope
is this may be achieved by taking appropriate linear combinations of the imaginary time
amplitudes, with external legs continued to real-time, as is the case for finite temperature
field theory.
These results are of importance for the theory of inflation, because if the inflationary
phase sat in a general α vacuum, the amplitude and spectrum of cosmic microwave
background perturbations can be dramatically effected [12]. If the present day universe
is likewise asymptoting toward a universe dominated by positive cosmological constant,
the asymptotic value of α can produce observable effects today, and may be responsible
for a component of the diffuse cosmic ray flux (see [43] for a study of this effect). We plan
to develop further the phenomenology of these vacua in future work.
The formalism we have developed may also have useful generalizations to computations
in flat space in squeezed state backgrounds. See [44] for QED calculations in squeezed state
backgrounds.
Let us emphasize that our motivation for this study was to try to discover a problem
with the α-vacuum, which would lead one to conclude the Euclidean vacuum was unique.
Thus far we have not found such a problem, which raises the question whether we must
think of α as a new cosmological constant fixed by initial conditions, or whether dynamics
leads to decay to the Euclidean vacuum at late times.
One hint that it might be the latter comes from the fact that requiring an Unruh
detector see a thermal distribution of particles uniquely selects the Euclidean vacuum.
Thus demanding local equilibrium (assuming no extra chemical potentials are turned on)
selects the Euclidean vacuum. However, that raises the question whether non-trivial α
simply corresponds to a new chemical potential needed to uniquely specify the scalar
field theory.2 This then introduces a new tunable parameter into the effective field theory
description of inflation. In Section 5 we found no imaginary part in the action at the one-
loop level, indicating that at leading order the α-vacua are stable, consistent with this latter
interpretation.
In [12] we argued in more general cosmological backgrounds, α should be tied to the
cosmological constant Λ ∼ Heff 2 M2
Planck , by e
α(t ) ∼ H /M
eff cut-off . For concreteness, let us
take Mcut-off ∼ 10 GeV, around the GUT scale. We find it intriguing that bounds on αtoday
16
from diffuse cosmic ray observations [43] (eα 10−6 (Htoday/Mcut-off )/(MPlanck /Mcut-off )
in our notation) place it in within a few of orders of magnitude of the scale Htoday/Mcut-off .3
Furthermore, cosmic rays have currently been observed with energy up to 1011 GeV [46],
with no obvious upper cut-off in sight, consistent with α vacuum predictions.
2 [17] suggest α can be interpreted as a marginal deformation of the CFT in the context of dS/CFT [45].
3 In [43] it was assumed photons (and possibly other Standard Model fields) were in an analog of an α vacuum.
It is perhaps more natural to assume only the inflaton(s) are in an α vacuum, which could substantially weaken
the bounds of [43].
Acknowledgements
We thank R. Brandenberger, A. Jevicki, S. Theisen and the Harvard High Energy

theory group for helpful discussions. This research is supported in part by DOE Grant
DE-FE0291ER40688-Task A.
Appendix A. Squeezed states
We record some formulas useful in the manipulation of squeezed states.

1 2
U(ζ ) ≡ exp ζ̄ a − ζ a †2
= eA , (A.1)
2
eA ae−A = eLA a, (A.2)
where LA B ≡ [A, B]. Let ζ = ρeiφ , then
[A, a] = ζ a † = ρeiφ a † ,

A, a † = ζ ∗ a. (A.3)
This implies
L2n+1
A a = eiφ ρ 2n+1 a † ,
L2n
A a = ρ a,
2n
(A.4)
so we obtain

eA ae−A = a ρ 2n /2n! + a † eiφ ρ 2n+1 /(2n + 1)!
= ak cosh(ρ) + ak† eiφ sinh(ρ) (A.5)
and
∗
bk = Uak U † = Nα ak − eα ak† = ak cosh(ρ) + ak† eiφ sinh(ρ). (A.6)
Some other expressions that we use:

∗
eα−α = e−2iφ ⇒ φ = − Im(α), (A.7)
α+α ∗
e = tanh ρ,
2
(A.8)

1 1 + eRe(α) 1 1
ρ = tanh−1 eRe(α) = ln = ln tanh − Re(α) , (A.9)
2 1 − eRe(α) 2 2

1 −i Im(α) 1
ζ= e ln tanh − Re(α) . (A.10)
2 2
Appendix B. Some useful facts
Global coordinates
ds 2 = −dt 2 + cosh2 t dΩ 2, (B.1)
where dΩ 2 is the metric on the unit 3-sphere. We will work in units where the Hubble
radius is 1. Define Euclidean vacuum using mode functions
ψklm (x) = yk (t)Yklm (Ω), (B.2)
where k, l, m label the complete set of scalar spherical harmonics on −|l| m |l|.
S3,
The yk (t) may be expressed in terms of the hypergeometric function 2 F1 [3]. These are
regular on the Euclidean section, and may be analytically continued to functions regular
on the lower half ζ plane, ζ = i sinh t. They have a branch cut from ζ = 1 to ζ = ∞.
Define linear combination
eiπk/2 iπ/4
φklm = √ e ψklm (x) + e−iπ/4 ψkl−m (x) . (B.3)
2
This set of modes is the basis of the complete set of modes we will use. They are
orthonormal and positive norm, and satisfy
∗
φklm (x̄) = φklm (x), (B.4)
(φklm , φk l m ) = δkk δll δmm . (B.5)
By definition the Euclidean vacuum Green function satisfies

GE (x, y) = φn (x)φn∗ (y), (B.6)
n
where we have compressed the k, l, m indices into the single index n.
It is useful to define z(x, y) = X · Y where X and Y are the coordinates of points on 5d
Minkowski space, where de Sitter can be embedded as −X02 + X12 + X22 + X32 + X42 = 1.
For spacelike separations z < 1, for null separations z = 1, and for timelike separations
z > 1. We have in mind continuing z to complex values for which the relation to geodesic
distance breaks down. Note also that z(x̄, y) = −z(x, y). In terms of z, (B.6) can be written
explicitly as
Γ (3/2 + iν)Γ (3/2 − iν)
GE (x, y) = 2 F1 (3/2 + iν, 3/2 − iν, 2; (1 + z)/2).
(4π)2
(B.7)
This function has a pole at z = 1 and a branch cut extending from z = 1 along the positive
real axis. The function is analytic in the lower-half z plane. When z is real, GE (z) is real for
z < 1, and develops an imaginary part for z > 1. The sign of this imaginary part changes
as one moves across the branch cut.
To make (B.6) well-defined for time-like separations, we must specify an iε prescrip-
tion. Near the singularity z = 1, we specify this in locally Minkowski coordinates (t, x )
by [17]
1
GE (x, x ) ∼ . (B.8)
(t − t − iε)2 − |
x − x |2
In the text we introduce the Green functions Gij (x, y). These are likewise defined using
GE but the iε prescriptions are as follows (for simplicity we set x and x to 0):
G00 (x, x ) = Nα2 GE (t − iε, t ), (B.9)
G10 (x, x ) = Nα2 γ GE (−t − iε, t ),

(B.10)
G01 (x, x ) = Nα2 γ ∗ GE (−t + iε, t ), (B.11)
G11 (x, x ) = Nα2 |γ |2 GE (t + iε, t ). (B.12)
References
[1] N.A. Chernikov, E.A. Tagirov, Quantum theory of scalar fields in de Sitter space–time, Ann. Inst.
H. Poincaré Phys. Theor. A 9 (1968) 109.
[2] E.A. Tagirov, Consequences of field quantization in de Sitter type cosmological models, Ann. Phys. 76
(1973) 561–579.
[3] E. Mottola, Particle creation in de Sitter space, Phys. Rev. D 31 (1985) 754.
[4] B. Allen, Vacuum states in de Sitter space, Phys. Rev. D 32 (1985) 3136.
[5] B.S. Kay, R.M. Wald, Theorems on the uniqueness and thermal properties of stationary, nonsingular,
quasifree states on space–times with a bifurcate killing horizon, Phys. Rep. 207 (1991) 49–136.
[6] R.M. Wald, Quantum Field Theory in Curved Space–Time and Black Hole Thermodynamics, Univ. Chicago
Press, Chicago, 1994, p. 205.
[7] T. Banks, L. Mannelli, de Sitter vacua, renormalization and locality, hep-th/0209113.
[8] M.B. Einhorn, F. Larsen, Interacting quantum field theory in de Sitter vacua, hep-th/0209159.
[9] N. Kaloper, M. Kleban, A. Lawrence, S. Shenker, L. Susskind, Initial conditions for inflation, JHEP 11
(2002) 037, hep-th/0209231.
[10] U.H. Danielsson, A note on inflation and transplanckian physics, Phys. Rev. D 66 (2002) 023511, hep-
th/0203198.
[11] U.H. Danielsson, Inflation, holography and the choice of vacuum in de Sitter space, JHEP 07 (2002) 040,
hep-th/0205227.
[12] K. Goldstein, D.A. Lowe, Initial state effects on the cosmic microwave background and trans-Planckian
physics, hep-th/0208167.
[13] S. Shankaranarayanan, Is there an imprint of Planck scale physics on inflationary cosmology?, Class.
Quantum Grav. 20 (2003) 75–84, gr-qc/0203060.
[14] L. Bergstrom, U.H. Danielsson, Can map and Planck map Planck physics?, JHEP 12 (2002) 038, hep-
th/0211006.
[15] K. Goldstein, D.A. Lowe, in preparation.
[16] U.H. Danielsson, On the consistency of de Sitter vacua, JHEP 12 (2002) 025, hep-th/0210058.
[17] R. Bousso, A. Maloney, A. Strominger, Conformal vacua and entropy in de Sitter space, Phys. Rev. D 65
(2002) 104039, hep-th/0112218.
[18] B.S. Dewitt, Quantum field theory in curved space–time, Phys. Rep. 19 (1975) 295–357.
[19] N.D. Birrell, P.C.W. Davies, Quantum Fields in Curved Space, Cambridge Univ. Press, Cambridge, 1982,
p. 340.
[20] S.L. Adler, Massless, Euclidean quantum electrodynamics on the five-dimensional unit hypersphere, Phys.
Rev. D 6 (1972) 3445–3461.
[21] S.L. Adler, Massless electrodynamics on the five-dimensional unit hypersphere: an amplitude-integral
formulation, Phys. Rev. D 8 (1973) 2400–2418.
[22] I.T. Drummond, Dimensional regularization of massless theories in spherical space–time, Nucl. Phys. B 94
(1975) 115.
[23] I.T. Drummond, G.M. Shore, Conformal anomalies for interacting scalar fields in curved space–time, Phys.
Rev. D 19 (1979) 1134.
[24] I.T. Drummond, Conformally invariant amplitudes and field theory in a space–time of constant curvature,
Phys. Rev. D 19 (1979) 1123.
[25] I.T. Drummond, G.M. Shore, Dimensional regularization of massless quantum electrodynamics in spherical
space–time. 1, Ann. Phys. 117 (1979) 89.
[26] G.C. Wick, The evaluation of the collision matrix, Phys. Rev. 80 (1950) 268–272.
[27] A. Houriet, A. Kind, Classification invariante des termes de la matrices, Helv. Phys. Acta 22 (1949) 319.
[28] A.A. Abrikosov, L.P. Gorkov, I.E. Dzyaloshinski, Methods of Quantum Field Theory in Statistical Physics,
Dover, New York, 1975.
[29] J.I. Kapusta, Finite Temperature Field Theory, Cambridge Univ. Press, Cambridge, 1989.
[30] Y. Takahasi, H. Umezawa, Thermo field dynamics, Collect. Phenom. 2 (1975) 55–80.
[31] N.P. Landsman, C.G. van Weert, Real and imaginary time field theory at finite temperature and density,
Phys. Rep. 145 (1987) 141.
[32] S.W. Hawking, Interacting quantum fields around a black hole, Commun. Math. Phys. 80 (1981) 421.
[33] B.A. Harris, G.C. Joshi, A new formulation of quantum field theory on s(4), Int. J. Mod. Phys. A 9 (1994)
3245–3282.
[34] T.S. Bunch, P.C.W. Davies, Quantum field theory in de Sitter space: renormalization by point splitting, Proc.
R. Soc. London A 360 (1978) 117–134.
[35] J.S. Dowker, R. Critchley, Effective Lagrangian and energy momentum tensor in de Sitter space, Phys. Rev.
D 13 (1976) 3224.
[36] T.S. Evans, What is being calculated with thermal field theory?, hep-ph/9404262.
[37] T.S. Evans, N point finite temperature expectation values at real times, Nucl. Phys. B 374 (1992) 340–372.
[38] T.S. Evans, Spectral representation of three point functions at finite temperature, Phys. Lett. B 252 (1990)
108–112.
[39] F. Guerin, Retarded-advanced N point Green functions in thermal field theories, Nucl. Phys. B 432 (1994)
281–314, hep-ph/9306210.
[40] J. Bros, U. Moschella, J.P. Gazeau, Quantum field theory in the de Sitter universe, Phys. Rev. Lett. 73 (1994)
1746–1749.
[41] J.P. Gazeau, J. Renaud, M.V. Takook, Gupta-bleuler quantization for minimally coupled scalar fields in
de Sitter space, Class. Quantum Grav. 17 (2000) 1415–1434, gr-qc/9904023.
[42] D. Bernard, A. Folacci, Hadamard function, stress tensor and de Sitter space, Phys. Rev. D 34 (1986) 2286.
[43] A.A. Starobinsky, I.I. Tkachev, Trans-Planckian particle creation in cosmology and ultra-high energy cosmic
rays, JETP Lett. 76 (2002) 235–239, astro-ph/0207572.
[44] K. Svozil, Quantum electrodynamics in the squeezed vacuum state: Feynman rules and corrections to the
electron mass and anomalous magnetic moment, hep-ph/9402316.
[45] A. Strominger, The dS/CFT correspondence, JHEP 10 (2001) 034, hep-th/0106113.
[46] HIRES Collaboration, D.J. Bird, et al., Evidence for correlated changes in the spectrum and composition of
cosmic rays at extremely high-energies, Phys. Rev. Lett. 71 (1993) 3401–3404.
K3 decays in chiral perturbation theory ✩

Johan Bijnens a , Pere Talavera b
a Department of Theoretical Physics 2, Lund University, Sölvegatan 14A, S-223-62 Lund, Sweden
b Departament de Física i Enginyeria Nuclear, Universitat Politècnica de Catalunya,
Jordi Girona 1-3, E-08034 Barcelona, Spain
Abstract
The process K3 is calculated to two-loop order (p6 ) in chiral perturbation theory (ChPT) in the
isospin conserved case. We present expressions suitable for use with previous work in two-loop ChPT
where the order p4 parameters (Lri ) were determined from experiment. We point out that all the order
p6 parameters (Cir ) that appear in the value of f+ (0) relevant for the determination of |Vus | can be
determined from K3 measurements via the slope and the curvature of the scalar form-factor. As by
product we update the value of the CKM matrix element |Vus |.
PACS: 12.15.Hh; 13.20.Eb; 12.39.F; 14.40.Aq
1. Introduction
Weak semileptonic kaon decays to a pion and a lepton–neutrino pair (K3 ) have a
long history. The early theoretical treatments can be found in the review [1]. This decay
plays an important role in the determination of the CKM matrix-element Vus , see, e.g.,
the discussion in [2] or [3]. The theoretical basis for this determination is the paper by
Leutwyler and Roos [4]. The basis for this evaluation were radiative corrections and
estimates of one-loop chiral corrections. The full chiral perturbation theory calculation
to order p4 (see Section 2 for a short explanation) was performed by Gasser and
Leutwyler [5]. References to earlier work on the non-analytic corrections can be found
there. Recent reviews of the situation can be found in [6] or [7].
✩
Supported in part by the European Union TMR network, Contract No. HPRN-CT-2002-00311 (EURIDICE).
E-mail address: bijnens@thep.lu.se (J. Bijnens).
doi:10.1016/S0550-3213(03)00581-9
342 J. Bijnens, P. Talavera / Nuclear Physics B 669 (2003) 341–362
Since that time a lot of work has been performed in chiral perturbation theory. An update
of the calculation of [5] to order p6 is thus necessary. Partial studies are done, the double
logarithm contribution is small [8] and a possibly large role for terms with two powers of
quark masses has been argued for in Ref. [9]. One full order p6 calculation exists [10], but it
uses outdated values of the ChPT constants as well as an older version of the classification
of p6 constants. In this paper we present an independent calculation of the K3 amplitudes
to order p6 in ChPT in the isospin limit. We present numerical results with values for the
ChPT constants resulting from fits to order p6 [11–14]. Related work is the update of the
electromagnetic radiative corrections given in [15].
We present a few definitions of ChPT in Section 2 to determine our notation. Section 3
defines the form-factors used in K3 decays. Analytical results are presented explicitly in
Section 4 for the form-factors up to order p4 and for the part depending on the order p6
parameters (Cir ). The remaining parts are rather long and can be obtained from the authors
on request. Section 5 describes one of our main results, the fact that all needed p6 constants
for the value of f+ (0) can be experimentally determined from Kµ3 experiments. The value
of f+ (0) is of course needed for determinations of Vus and is of use for future precise
measurements of K → πν ν̄.
In Section 6 the presently available data set is discussed. Here we also point out that
the often used linear approximation in the form-factors can have a sizable effect on the
measured value of the slope and the value at t = 0. This effect is of similar size as the
experimental errors quoted. We present an extended discussion of the numerical results in
Section 8 after a short discussion of the inputs used in Section 7. Our final conclusions for
λ+ are in Section 8.3 and of f+ (0) in Section 8.5. We summarize our results in Section 9.
2. Some definitions
Chiral perturbation theory is the modern way to derive the predictions following from
the fact the SU(nf )L × SU(nf )R chiral symmetry in the limit of nf massless flavours in
QCD is spontaneously broken by non-perturbative QCD dynamics to the diagonal vector
subgroup SU(nf )V . It is the effective field theory method to use this property at low
energies. It takes into account the singularities associated with the Goldstone boson degrees
of freedom caused by the spontaneous breakdown of chiral symmetry and parametrizes
all the remaining freedom allowed by the chiral Ward identities in low energy constants
(LECs). The LECs are the freedom in the parts of the amplitudes that depend analytically
on the masses and momenta. The expansion is ordered in terms of momenta, quark masses
and external fields. Recent lectures introducing this area are given in Ref. [16]. We use here
the standard ChPT counting where the quark mass, scalar and pseudoscalar external fields
are counted as two powers of momenta. Vector and axial-vector external currents count as
one power of momentum. The Lagrangian can be ordered as
Leffective = L2 + L4 + L6 + · · ·

10
90
94
= L2 + Li O4i + Ci O6i + Ci O6i + · · · . (2.1)
i=1 i=1 i=91
J. Bijnens, P. Talavera / Nuclear Physics B 669 (2003) 341–362 343
Fig. 1. The tree level diagrams contributing to the K3 form-factor. A • indicates a vertex from L2 , a × a vertex
from L4 and a ⊗ a vertex from L6 . The insertion of the weak current is indicated by the wiggly line.
The index i in Li stands for the chiral power. The precise form of L2 and L4 is given below
while L6 can be found in [17]. The lowest order, O(p2 ), in the expansion corresponds to
tree level diagrams with vertices from L2 , the next-to-leading order, NLO or O(p4 ), to
one-loop diagrams with vertices from L2 or tree level diagrams with one vertex from L4
and the rest from L2 . The next-to-next-to-leading order, NNLO or O(p6 ), has two-loop
diagrams, one-loop diagrams with one vertex from L4 and tree level diagrams with one
vertex from L6 or two vertices from L4 and all other vertices from L2 . The loop diagrams
take all singularities due to the Goldstone bosons correctly into account. The singularities
are the real predictions of ChPT while the other effects from QCD are in the values of
the LECs. The diagrams, in addition to wave-function-renormalization, relevant for the
processes discussed in this paper are shown in Figs. 1, 2 and 3.
The expressions for the first two terms in the expansion of the Lagrangian are given by
(F0 refers to the pseudoscalar decay constant in the chiral limit)
F02
L2 = Dµ U † D µ U + χ † U + χU † , (2.2)
4
and
2
L4 = L1 Dµ U † D µ U + L2 Dµ U † Dν U D µ U † D ν U

+ L3 D µ U † Dµ U D ν U † Dν U + L4 D µ U † Dµ U χ † U + χU †
2
+ L5 D µ U † Dµ U χ † U + U † χ + L6 χ † U + χU †
2
+ L7 χ † U − χU † + L8 χ † U χ † U + χU † χU †
R µ
− iL9 Fµν D U D ν U † + Fµν D U D U + L10 U † Fµν
L µ † ν R
U F Lµν , (2.3)
while the next-to-next-to-leading order is a rather cumbersome expression [17]. The special
unitary matrix U contains the Goldstone boson fields
 √1 0 √1 
√ π + η π+ K+
2 6
i 2  π− π + √1 η K 0 
−1 0
√
U = exp M , M = 2 6 . (2.4)
F0 −2
K− K0 √ η
6
The formalism is the external field method of [18] with s, p, lµ and rµ matrix valued
scalar, pseudo-scalar, left-handed and right-handed vector external fields, respectively.
These show up in
χ = 2B0 (s + ip), (2.5)

Fig. 2. The one and two-loop diagrams without overlapping loop integrals contributing to the K3 form-factor.
A • indicates a vertex from L2 , a × a vertex from L4 and a ⊗ a vertex from L6 . The insertion of the weak current
is indicated by the wiggly line.
Fig. 3. The two-loop diagrams with overlapping loop integrals contributing to the K3 form-factor. A • indicates
a vertex from L2 , a × a vertex from L4 and a ⊗ a vertex from L6 . The insertion of the weak current is indicated
by the wiggly line.
in the covariant derivative
Dµ U = ∂µ U − irµ U + iU lµ , (2.6)
and in the field strength tensor

L(R)
Fµν = ∂µ l(r)ν − ∂ν l(r)µ − i l(r)µ , l(r)ν . (2.7)
For our purpose it is sufficient to set

m̂
s= m̂ , rµ = 0,
ms
 
Vud Wµ+ Vus Wµ+
g2  ∗ − 
lµ = − √  Vud Wµ  (2.8)
2 ∗ −
Vus Wµ
with g2 the weak coupling constant, related to the Fermi constant by
GF g2
√ = 22 . (2.9)
2 8mW
2.1. Renormalization scheme
We use the renormalization scheme as explained in [17] and [19]. It extends the scheme
from [18] naturally to two-loops. Notice that the work of Post and Schilcher [10,20,21]
used a slightly different scheme. The scheme employed here does not introduce the & 2
term in Eq. (39) of Ref. [10]. Subtractions are performed via

Li = (Cµ)(d−4) Γi + Li (µ) ,

1 (2) 2 (1) (L)
Ci = (Cµ) 2(d−4)
Ci (µ) − 2 Γi Λ + Γi + Γi (µ) Λ ,
r
(2.10)
F
with
1 1
ln C = − ln(4π) + Γ (1) + 1 , Λ= . (2.11)
2 16π 2 (d − 4)
(2) (1) (L)
The coefficients Γi can be found in [17,18] and the Γi , Γi Γi (µ) in [17]. We will in
the remainder always write Cir and Lri but the µ dependence is of course present.
3. The K3 form-factors: definition and O(p4 )
The decays we consider are

+
K + (p) → π 0 (p )+ (p )ν (pν ),
K3 , (3.1)
− +
0
K (p) → π (p ) (p )ν (pν ), K3
0
(3.2)
and their charge conjugate modes. stands for µ or e.
+
The matrix-element for K3 , neglecting scalar and tensor contributions, has the structure
GF + µ +
T = √ Vus Fµ (p , p), (3.3)
2
with
µ = ū(pν )γ µ (1 − γ5 )v(p ),

F + (p , p) = π 0 (p )V 4−i5 (0)K + (p) ,
µ µ
1 + 0 + 0
= √ (p + p)µ f+K π (t) + (p − p )µ f−K π (t) . (3.4)
2
To obtain the K30
matrix-element, one replaces Fµ+ by

Fµ0 (p , p) = π − (p )Vµ4−i5 (0)K 0 (p)
0π − 0π −
= (p + p)µ f+K (t) + (p − p )µ f−K (t). (3.5)
+π 0 0π −
The processes (3.1) and (3.2) thus involve the four K3 form-factors f±K (t), f±K (t)
which depend on
t = (p − p)2 = (p + pν )2 , (3.6)

the square of the four momentum transfer to the leptons.
In this paper we work in the isospin limit thus
+π 0 0π −
f± = f±Kπ = f±K = f±K . (3.7)
f+Kπ is referred to as the vector form-factor, because it specifies the P -wave projection
of the crossed channel matrix-elements 0|Vµ4−i5 (0)|K + , π 0 in. The S-wave projection is
described by the scalar form-factor
t
f0 (t) = f+ (t) + f− (t). (3.8)
m2K − m2π
Analyses of K3 data frequently assume a linear dependence

t
f+,0 (t) = f+ (0) 1 + λ+,0 2 . (3.9)
mπ +
For a discussion of the validity of this approximation see [5] and references cited therein.
We will discuss it to order p6 and in comparison with the data. At the expected future
precision it will be necessary to go beyond this approximation.
Eqs. (3.8) and (3.9) leads to a constant f− (t),
m2K − m2π
f− (t) = f− (0) = f+ (0)(λ0 − λ+ ) . (3.10)
m2π +
The form-factors f±,0 (t) are analytic functions in the complex t-plane cut along the
positive real axis. The cut starts at t = (mK + mπ )2 . In our phase convention, the form-
factors are real in the physical region
m2 t (mK − mπ )2 . (3.11)

A discussion of the kinematics in K3 decays can be found in [6].
4. Analytical results
The total result we obtain is split by chiral order.
fi (t) = fi(2) (t) + fi(4) (t) + fi(6) (t), (i = +, −, 0). (4.1)
4.1. Order p2
This has been known for a very long time and is fully determined by gauge invariance.
(2) (2) (2)
f+ (t) = f0 (t) = 1, f− (t) = 0. (4.2)
It results from the diagram in Fig. 1(a).
4.2. Order p4
This was first calculated within the ChPT framework by Gasser and Leutwyler in 1985
[5]. The result contains the non-analytic dependence in the symmetry parameters predicted
by [22]. The form of the form-factors we use (which is equivalent to the result of [5] to
order p4 ) is the one which our expressions for the p6 contribution correspond to.

Fπ2 f+(4)(t) = 2Lr9 t + 3/8 Ā m2η + Ā m2π + 2Ā m2K
2 2
− 3/2 B 22 m2K , m2η , t .
22 mπ , mK , t + B (4.3)
(4)
Fπ2 f− (t) = 4m2K − 4m2π Lr5 − 2 m2K − m2π Lr9 + 1/2Ā m2η − 5/12Ā m2π

+ 7/12Ā m2K + B m2π , m2K , t −1/12m2π − 5/12m2K + 5/12t

+B m2K , m2η , t 1/12m2π − 7/12m2K + 1/4t

+B1 m2π , m2K , t −7/12m2π + 19/12m2K − 5/12t

+B1 m2K , m2η , t −11/12m2π + 23/12m2K − 1/4t

+B21 m2π , m2K , t 3/2m2π − 3/2m2K − 5/6t

+B21 m2K , m2η , t 3/2m2π − 3/2m2K − 1/2t

+B22 m2π , m2K , t (−5/6) + B 22 m2K , m2η , t (−1/2). (4.4)
These results are obtained from the diagrams in Figs. 1(b), 2(a) and (c), together with
wave-function renormalization.
4.3. Order p6
The p6 contribution we split in several parts

(6) 1 C
fi (t) = f (t) + fiL (t) + fiB (t) + fiH (t) + fiV (t) , (i = +, −, 0). (4.5)
Fπ4 i
The split between the last three terms is not unique and depends on how the irreducible
two-loop integrals are separated from the reducible ones. The first two terms are the ones
containing the dependence on the p6 and p4 coupling constants.
The ones with dependence on Cir stem from wave-function renormalization and the
diagram of Fig. 1(c). The results are
f+C (t) ≡ R+0

Kπ
+ tR+1
Kπ
+ t 2 R+2Kπ
2 2
= −8 mπ − m2K C12 r
+ C34
r
r
+ t −4m2π 2C12 + 4C13 r
+ C64 r
+ C65
r
+ C90r

+ m2K −8C12 r
− 32C13r
− 8C63
r
− 8C64
r
− 4C90
r

+ t 2 −4C88r
+ 4C90 r
, (4.6)
and
f−C (t) ≡ R−0

Kπ
+ tR−1
Kπ
+ t 2 R−2
Kπ
2
= mK − m2π m2π + 24C12 r
− 16C13 r
+ 8C15 r
+ 16C17
r
+ 8C34 r
+ 4C64
r

+ 4C65r
+ 4C90
r
+ m2K +24C12 r
+ 32C13
r
+ 16C14
r

+ 16C15 r
+ 8C34r
+ 4C63 r
+ 8C64
r
+ 4C90
r
2
+ t4 mK − m2π −2C12 r
+ C88
r
− C90r
. (4.7)
We have followed a notation very close to the one in [14]. Notice that we have the
relations
Kπ
R−2 = 0,
+
Kπ
R+2 = RVπ 2 = RVK2 ,
1 + 0
Kπ
R+1 = RVπ 1 + RVK1 + RVK1 . (4.8)
2
The other RiM are similar combinations of the Cir but in the expansion of the electromag-
netic form-factors [14]. Notice that the last relation is really Sirlin’s relation [23] and the
second satisfies it as well.
We have not quoted the remaining formulas, but will quote below some approximate
numerical expressions. The exact formulas can be obtained from the authors on request.
Our expressions satisfy the Ademollo–Gatto theorem [24].
5. Getting the value of f+ (0)
One of the problems we face here is whether the needed Cir can be determined from
experiment. There are many of these coefficients showing up but as is obvious from
Eq. (4.6), what we need is a value for C12r + C r . It turns out that this combination can
34
actually be determined from K3 measurements. The derivation given below relies on the
fact that we need values for the p4 constants, determined to order p4 only, in the order p6
part to be correct to the accuracy that we are working. We can determine all needed Lri to
this accuracy relying only on data.
We construct the quantity
t
f˜0 (t) = f+ (t) + f− (t) + 1 − FK /Fπ
− mπ
m2K
2
t
= f0 (t) + 2 (1 − FK /Fπ ). (5.1)
mK − m2π
This has no dependence on the Lri at order p4 , only via order p6 contributions. Inspection
of the dependence on the Cir shows that
8 r 2 2 t r 2
f˜0 (t) = 1 − 4 C12 + C34
r
mK − m2π + 8 4 2C12 + C34
r
mK + m2π
Fπ Fπ
8 2 r ¯ + ∆(0).
− 4 t C12 + ∆(t) (5.2)
Fπ
We emphasize that the quantities ∆(t) ¯ and ∆(0) can in principle be calculated to order
p accuracy with knowledge of the Lri to order p4 accuracy. In practice, since a p4 fit
6
will include in the values of the Lri effects that come from the p6 loops (due to the fitting
to experimental values) we consider the p6 fits to be the relevant ones to avoid double
counting effects. Numerical results will be discussed in Section 8.
The definition in (5.1) has essentially used the Dashen–Weinstein relation [25] to
remove the Lri dependence at order p4 . It has also the effect that it removed many of
the Cir from the scalar form-factor as well. The corrections which appear in the Dashen–
Weinstein relation are included in the functions ∆(t)¯ and ∆(0), these have both order p4
6
[5,22] and order p contributions.
It is obvious from Eq. (5.2) that the needed combination of Cir can be determined from
the slope and the curvature of the scalar form-factor in K3 decays.
r can be measured from the curvature of the pion scalar form-
It seems possible that C12
factor near 0 [26]. When this calculation is complete, one can use the dispersive estimates
of the pion scalar form-factor together with only a λ0 measurement in Kµ3 to obtain the
p6 value for f+ (0). There are also some dispersive estimates for the relevant scalar form-
factor. Unfortunately, these are not in a usable form at present [27].
The discussion of Ref. [9] can be shown in this light too. The constant A3 of
generalized perturbation theory correspond to a combination of the Cir from [17]. The
precise combination is

1 r 1 r
A3 = (2B0 )2 −C34r
+ C14 + C17
r
+ C26r
+ C29
r
+ C31
r
+ C33 . (5.3)
2 6
r , so this plays the role of A here.
As we have shown it is possible to eliminate all but C34 3
r
C12 is higher order in the counting employed in [9] and was not considered there.
In Ref. [9] the additional observation was made that a relatively small change in the
ratio FK /Fπ , together with an adjustment of the constant A3 can accommodate the CKM
unitarity.
0 . The data are Clark77 [29], Donaldson74 [28], Buchanan75 [31] and
Fig. 4. The data on f0 (t)/f+ (0) from Kµ3
Birulev81 [30]. The latter reference has two distinct data sets. For comparison a linear approximation and a pole
approximation with a mass of 800 MeV are shown as well.
6. Data
0
6.1. Kµ3
The most useful data points are those where the full dependence on the kinematical
variable t = q 2 /m2π is shown. That means that the experiments that determined the value
of λ0 from the branching ratio or did not provide an actual t dependence but just fitted the
linear slope in a global way are not that useful for us, but see below.
Of the more recent experiments that quote data not from the branching ratio, the ones
that gave a plot or numbers for f0 (t) are [28–31]. Ref. [32] gives a plot but mentions that
it is not statistically significant, which inspection of the plots confirms.
There are some obvious problems with the data. E.g., the f0 (t) from [28] do not go to 1
at t = 0. We have shown these data in Fig. 4 together with a linear and a pole approximation
corresponding to a mass of 800 MeV. This shows the accuracy needed to see the curvature.
+
6.2. Kµ3
Here we have not been able to find data that show a plot of f0 (t). All experiments
are analyzed in terms of a constant form-factor as discussed in [2]. There is one more
experiment [33] that quotes measurements of λ0 not included in [2].
0 . The CPLEAR data are from Ref. [34]. The older data are for
Fig. 5. The data on f+ (t)/f+ (0) from Ke3
comparison, Buchanan75 is [31] and Birulev81 is [30]. The latter reference has two distinct data sets.
0
6.3. Ke3
Here the data are dominated by the recent high statistics CPLEAR data [34]. There
exists a very high statistics older experiment [35]. They provide plots with different data
assumptions and can thus not be easily compared at the level of f+ (t) directly. But [35]
quoted both a linear and quadratic fit to f+ (t). In order to show the relation between the
most recent and older data we have plotted the data of [30,31,34] in Fig. 5.
We have performed some simple fits of the form

t
f+ (t) = a+ 1 + λ+ 2 + c+ t 2 (6.1)
mπ +
to the CPLEAR data. The fits agree extremely well with those reported in [34] and are
given in Table 1. Notice that the fits that go beyond the linear approximation and leave
the normalization free give a significantly lower λ+ and with larger errors. It is within its
errors compatible with the linear fit, but outside the errors from the linear fit, we consider
that result from the fit with curvature to be more reliable. The shift is of similar size to that
observed in Table 1 in [35].
Table 1
The fits to the CPLEAR data of various theoretical forms of f+ . In the last four fits, which use th ChPT results,
λ+ and c+ are derived quantities. The symbol ≡ means this quantity was set to this value in the fit
Form a+ λ+ c+ Kπ
105 R+1 Kπ
103 R+2
−4

GeV GeV−2
Eq. (6.1) ≡1 0.0245 ± 0.0006 ≡0
Eq. (6.1) 1.000 ± 0.004 0.0245 ± 0.0015 ≡0
Eq. (6.1) ≡1 0.0238 ± 0.0017 0.5 ± 1.2
Eq. (6.1) 1.008 ± 0.009 0.0181 ± 0.0068 2.8 ± 2.8
Eq. (8.2) 1.008 ± 0.008 0.0180 ± 0.0067 2.7 ± 3.0 −4.3 ± 2.5 0.19 ± 0.21
Eq. (8.2) ≡1 0.0236 ± 0.0019 0.4 ± 1.2 −2.2 ± 0.7 0.02 ± 0.09
Eq. (8.2) ≡1 0.0201 ± 0.0006 3.2 −3.5 ± 0.3 ≡ 0.22
Eq. (8.2) 1.009 ± 0.004 0.0170 ± 0.0015 3.2 −4.7 ± 0.5 ≡ 0.22
Table 2
The PDG averages for λ+ and λ0 and the values from the most recent experiments. µe means that lepton
universality has been used in the measurement. The result of [37] is an update of [39] which was included in
the PDG averages. The last result [38] is preliminary
Process Ref. λ+ λ0
+
Kµ3 [2] 0.033 ± 0.010 0.004 ± 0.009
+
K3 [2]µe 0.0282 ± 0.0027 0.013 ± 0.005
0
Kµ3 [2] 0.033 ± 0.005 0.027 ± 0.006
0
K3 [2]µe 0.0300 ± 0.0020 0.030 ± 0.005
0
Ke3 [2] 0.0291 ± 0.0018 –
+
Ke3 [2] 0.0278 ± 0.0019 –
+
Ke3 [36] 0.0293 ± 0.0025 –
+
Kµ3 [33] 0.0321 ± 0.0045 0.0209 ± 0.0045
+
Ke3 [37] 0.0278 ± 0.0023 –
0
Ke3 [38] 0.02748 ± 0.00084 –
+
6.4. Ke3
There are recent high statistics experiments that show plots of f+ (t). They are [36,37].
We will discuss the data from [37] below.
6.5. λ0 and λ+
Most experiments have analyzed their data assuming linear form-factors. In Table 2 we
have quoted the PDG2002 values and the more recent experiments not included in it. We
will not use these numbers much, given the possible shifts when introducing a curvature in
the analysis.
7. Inputs
As relevant combinations we have obtained in our earlier work [14] experimental values
for RVπ 2 leading to
Kπ
R+2 = (0.22 ± 0.02) × 10−3 (7.1)
and
Lr9 = (5.93 ± 0.43) × 10−3 (7.2)
which used the estimate RVπ 1= −0.49 × 10−5 GeV . The two resonance estimates for the
2
M
RV 1 quantities done in [14,40] via naive vector-meson dominance and the chiral inspired
variety lead to basically the same estimate
Kπ
R+1 ≈ −4 × 10−5 GeV2 . (7.3)
For inputs for the other parameters we use our fits that include the latest K4 data [41].
These are the p4 fit, and fit 10 to 13 in Table 2 in [13]. This is a reasonable variation of the
various input parameters.
We use the PDG2002 mass values for all the particles involved and
Fπ = 92.4 MeV, FK /Fπ = 1.22. (7.4)
+
The amplitudes for the decays K3
are calculated with the mass of the charged kaon and
0
the neutral pion. Those for K3 with the mass of the neutral kaon and the charged pion.
The scale we use in all the coupling constants and the loop integrals is µ = mρ =
700 MeV. Almost all conclusions are done from experimental determinations of the various
parameters so the results are µ-independent.
8. Numerical results
8.1. Size of the pure loop contributions
In Fig. 6 we show the results from the pure loop diagrams for f+ (t). The different lines
are for different choices of pion, kaon and eta masses. They give some indication of the
size of quark-mass isospin breaking, but it does not include the enhanced effect discussed
in [4,5]. In Fig. 7 we show the equivalent results for f− (t).
8.2. Comparison with Refs. [10,21]
At this point we should also compare with the calculation of [10]. In that paper
numerical results are quoted in Eqs. (91)–(94). We agree, if we use input Lri and masses
the same as theirs, well with their numerical expressions for ∆loop
p6 −
f (their Eqs. (92) and
(94)) and with their numerical value for ∆loop
p6 +
f (0) (their Eq. (93)). We do not agree with
their expression for the t-dependence of ∆loop f , our numerical values differ by roughly a
p6 +
factor of two. Given the good agreement with the other values this is rather puzzling.
We have attempted to trace the possible sources of discrepancy, and the present
conclusions are: (i) recalculating with all Lri = 0 or only Lr9 = 0 does not lead to agreement.
Fig. 6. The pure loop diagram contributions to f+ (t). Shown are the p 4 and the p 6 results for several different
choices of the pion, kaon and eta masses.
If we use as input instead Lr9 = 0.0082 we get good agreement with f+ (t), but it spoils the
agreement for f− (t). (ii) We have used a different split for the reducible and irreducible
parts of the integrals and a different subtraction scheme. Comparing direct subresults is,
therefore, rather difficult. We have also compared our results for the pion electromagnetic
form-factor [14] with those in [21]. We have a small discrepancy for the real part there
but a rather large discrepancy in the imaginary part. The imaginary part as calculated
in [14] would give a phase of a few degrees in [14] while those of [21] give a phase
significantly above 10 degrees towards 500 MeV. The latter is not compatible with the
expected perturbative buildup of the phase from ChPT. (iii) The main expected source
of discrepancy is the slightly different renormalization scheme. The quantity we disagree
most strongly on has a strong cancellation for the value of the Lri used between the pure
loop part and the Lri dependent part, possibly amplifying differences in the result.
8.3. f+ (t) and comparison with the CPLEAR and KEP-PS E246 data
The numerical expression for the p6 contribution with the Cir = 0 and the Lri from fit
10 is
Fig. 7. The pure loop diagram contributions to f− (t). Shown are the p 4 and the p 6 results for several different
choices of the pion, kaon and eta masses.
1 L
4
fi (t) + fiB (t) + fiH (t) + fiV (t)
Fπ

= 0.01462 + 0.0896353t + 0.0006313t 2 + 0.3414t 3 0
K3
+
= 0.01424 + 0.0840569t + 0.0071463t 2 + 0.3493t 3 K3 (8.1)
with t expressed in GeV2 and it is valid in the range 0 t 0.13 GeV2 .
We now compare our ChPT expression at order p6 to the CPLEAR data [34]. The
latter data are normalized to one assuming a linear dependence. It is, therefore, that the
polynomial fits done in Section 6.3 added a normalization factor as well. We now perform
a fit using the inputs for the Lri from fit 10. The other choices of the Lri give essentially
similar results. So we fit the CPLEAR data to
a+ (4) (6)
f+ (t) = 1 + f+ (t) + f+ (t) . (8.2)
∆(0)
Kπ Kπ Kπ
The effect of R+0 of Eq. (4.6) goes into a+ while R+2 gives the Cir part of c+ and R+1
r
gives the Ci part of λ+ (c+ and λ+ are defined in Eq. (6.1)).
Notice that the fitted value, using the input from the pion electromagnetic form-factor
Kπ Kπ
for R+2 , gives a value for R+1 in good agreement with the naive expectation. Notice also
that the presence of a curvature does change the fitted value of the normalization by a little
less than one %. A rather important change due to the inclusion of the curvature is the
Fig. 8. The CPLEAR data together with the normalized ChPT result without the Cir , the last fit reported in Table 1
and the linear fit done by the CPLEAR Collaboration.
effect on the value of the slope. Notice that, using the ChPT expression and the curvature
as determined from the electromagnetic form-factor leads to
λ+ = 0.0170 ± 0.0015. (8.3)

This value comes from the ChPT in the following way

λ+ = 0.0283 p4 + 0.0011 loops p6 − 0.0124 Cir . (8.4)
The p6 correction is about 30%. The difference with the conclusions on λ+ of [10] is to a
large extent due to their fixing the normalization at one.
The CPLEAR data together with the normalized ChPT result without the Cir contribu-
tion, the last fit reported in Table 1 and the linear fit done by the CPLEAR Collaboration,
is shown in Fig. 8.
+
We have performed a similar exercise for the Ke3 data from the KEP-PS E246
experiment. This is shown in Fig. 9. Again, both the linear fit and the one with the curvature
fixed from the pion electromagnetic form-factor have a similar χ 2 . The difference in
normalization, relevant for |Vus | is about 0.6% and
Kπ
R+1 = −2.5 × 10−5 . (8.5)
Kπ fitted and the linear fit of the KEK-PS

Fig. 9. The KEK-PS E246 data together with the ChPT result, with R+1
Collaboration.
Kπ
The value for R+1 is somewhat different from the value determined from the CPLEAR
data, but still compatible with the resonance estimate. The value for the slope is
λ+ = 0.0214 ± 0.0018. (8.6)

Notice that in both experiments, KEP-PS E-246 and CPLEAR, we have neglected the
experimental systematic errors. A possible discrepancy can only be put in after a full
experimental analysis is performed. But again, both the normalization and slope are
changed significantly from the linear fit case.
8.4. The scalar form-factor f0 (t)
The scalar form-factor as we have shown above is important since it can be used to
determine the p6 constants needed to evaluate f+ (0). In this section we show numerical
¯
results for f0 (t) and ∆(t). We have used here the value of FK /Fπ = 1.22.
In Fig. 10 we show the function f0 (t) − f0 (0) at order p4 and for the various sets of the
Lri of Ref. [13]. The value of f0 (0) = f+ (0) is discussed in the next subsection.
As can be seen the convergence from p4 to p6 is quite good. These are the curves
labeled “fit 10 p4 ” and “fit 10”. Notice that all fits of the Lri done at order p6 (fit 10–13)
give basically identical results. The fit of the Lri at p4 (labeled “p4 fit”) deviates somewhat
Fig. 10. The form-factor f0 (t) − f0 (0) with the Cir = 0. Shown are the cases for the neutral kaon decays for
various sets of the Lri together with old current algebra result.
but this we consider an artefact as discussed above. For comparison we have shown the
part due to t (1 − FK /Fπ )/(m2K − m2π ).
A good fit over the entire phase space 0 t 0.13 (t in GeV2 ) is given by
0
¯ = −0.25763t + 0.833045t 2 + 1.25252t 3
∆(t) Ke3 ,
+
¯ = −0.260444t + 0.846124t + 1.33025t
∆(t) 2 3
Ke3 . (8.7)
The error from the values of the different sets of Lri is about 0.0013 at t = 0.13 GeV2 .
We have not attempted to do a fit to any of the data given the experimental situation
on λ0 . We would however like to point out that the predicted curvature in f0 (t) is small but
of the same order as in f+ (t). As we saw above, this curvature made a rather large change
in the measured value of λ+ . A similar effect in λ0 can thus not be excluded and should be
studied experimentally.
8.5. The value of f+ (0) and ∆(0)
The results for f+ (0) with Cir = 0 (which is equivalent to ∆(0)) are shown in Table 3.
+
The isospin breaking shown is only an estimate, we have calculated the K3 case with the
0
masses mK + and mπ 0 and K3 with mK 0 and mπ + . Further work on including isospin
Table 3
The various contributions to ∆(0) = f+ (0)|C r =0 − 1
i
0 +
K3 K3
p4 −0.02266 −0.02275
p 6 loops only 0.01130 0.01104
p 6 -Li fit 10 0.00332 0.00320
p 6 -Li fit 11 0.00375 0.00355
p 6 -Li fit 12 0.00216 0.00189
p 6 -Li fit 13 0.00539 0.00526
p 6 -Li p 4 fit 0.00891 0.00863
breaking fully to two-loop order is in progress [42]. As discussed above, we consider the
results with the Lri determined at p4 order rather extreme. We have also investigated how
∆(0) varies if we vary the Lri according to the errors and correlations determined in [13].
For fit 10, the 68% CL error gives 0.00124 and for fit 11 it gives 0.00273. Notice that this
latter set allows for a very large variation of Lr4 . We take the latter 0.00273 as a sign of
the variation with the Lri , notice that includes all the p6 fits given above. As a conservative
estimate of this error we take half of the p6 loop contribution as error and add to it the error
from the Lri . We thus obtain

∆(0) = −0.0080 ± 0.0057[loops] ± 0.0028 Lri . (8.8)
The value of f+ (0) is related to ∆(0) via
8 r 2 2
f+ (0) = 1 + ∆(0) − 4
C12 + C34
r
mK − m2π . (8.9)
Fπ
r can be made from scalar dominance of the pion scalar form-factor
A naive estimate of C12
(SMD) leading to
r
Fπ4
C12 SMD
= − ≈ −1.0 × 10−5 . (8.10)
8m4S
The other combination can in principle be estimated from λ0 via

m2 FK 8m2 r 2 d
λ0 = 2 π − 1 + 4π 2C12 + C34r ¯
mK + m2π + m2π ∆(t). (8.11)
mK − m2π Fπ Fπ dt
As an example take λ0 = 0.009 ± 0.010 leading to
r
2C12 + C34
r
= (−1.0 ± 1.7) × 10−5 . (8.12)
Putting both estimates together gives
f+ (0)|Cir ≈ 0.0 ± 0.1. (8.13)

Essentially a 1% precision on f+ (0) requires a measurement of λ0 to 0.001 (about 5%),
assuming we can determine C12r with the relevant precision from other sources [26].
The estimate of the p6 corrections given in [4] is for the analytic contribution
(proportional to (ms − m̂)2 ) and contributes to the Cir dependent part
f+ (0)|Cir ≈ −0.016 ± 0.008. (8.14)
If we use this value we get for f+ (0) the present best estimate

f+ (0) = 1 − 0.02266 p4 + 0.01130 p6 pure loops + 0.00332 p6 Lri

− 0.016 p6 [4] ± 0.0057[loops] ± 0.0028 Lri ± 0.0080 Cir [4]
= 0.9760 ± 0.0102, (8.15)
where errors have been added in quadrature. This satisfies the bound f+ (0) 1 [43] (as
quoted in [4]). Notice that this bound when combined with the other results here gives a
r + C r and thus on a combination of the slope and curvature
constraint on the values of C12 34
in the scalar form-factor. The total value and error on f+ (0) awaits an experimental
determination of the needed constants but Eq. (8.15) is our present best estimate.
The net order p6 contribution is roughly one order of magnitude smaller, and with the
same sign, as the p4 one, due to the sizable cancellation between the calculation of the
loop contributions and the estimate of the contribution from the p6 constants. We have
reevaluated |Vus | following the basic steps presented in Section 7.3 of [15] but using as
input the previous expression (8.15) obtained for the K + decay case. The result should
only be taken as preliminary since the full analysis including the isospin breaking terms at
p6 is still missing as well as an analysis of the effect of the curvature in the form-factor
on the experimental value. It explicitly contains: the e.m. corrections at one-loop order and
the next-to-next-to-leading order correction in the isospin limit.
|Vus | = 0.2175 ± 0.0029, (8.16)
this value is indeed compatible with world-average value |Vus | = 0.2196 ± 0.0023 within
errors. Notice that the theoretical errors are larger due to the uncertainty in the Cir and they
are at the same footing as the actual experimental ones. Another recent evaluation of Vus
along the same lines can be found in Ref. [44].
9. Summary and conclusions
We have performed a calculation to two-loop order of K3 decays in the isospin limit.
As far as we have been able to check, this calculation agrees analytically with the earlier
on in [10]. We agree with some of the numerical results of that work but not all.
For K3 decay measurements we have shown how the value of f+ (0) needed for the
determination of Vus can be determined from the slope and curvature of f0 (t) which can
be measured in Kµ3 . It is possible that additional information from the pion scalar form-
factor near 0 allows the measurement of the slope only to be sufficient [26].
We have presented a present best value for f+ (0) based on an estimate of the p6
constants C12r and C r . It is clear that this can be further improved after the above
34
measurements are performed. From the calculated best value of f+ (0) we can deduce
directly an improved estimated value of the CKM matrix element |Vus |.
As can be seen from our comparison with the data for f+ (t) the presence of curvature
can make a sizable impact both on the determination of the value of the form-factor at zero
and on the slope. The large change in the value λ+ we found is entirely due to this effect
and was compatible with estimates of the Cir involved.
Acknowledgements
We thank Gabriel Amorós for participation in the early stages of this project and the
CPLEAR and KEP PS E246 Collaborations for providing us with their data. This work has
been funded in part by the Swedish Research Council, the European Union TMR network,
Contract No. HPRN-CT-2002-00311 (EURIDICE). FORM 3.0 has been used extensively
in these calculations [45]. We want to thanks to F. Blanc, P. Post and J. Stern for comments
on the manuscript.
References
[1] L.M. Chounet, J.M. Gaillard, M.K. Gaillard, Phys. Rep. 4 (1972) 199.
[2] K. Hagiwara, et al., Particle Data Group Collaboration, Phys. Rev. D 66 (2002) 010001.
[3] G. Calderon, G. Lopez Castro, Phys. Rev. D 65 (2002) 073032, hep-ph/0111272.
[4] H. Leutwyler, M. Roos, Z. Phys. C 25 (1984) 91.
[5] J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 517.
[6] J. Bijnens, G. Colangelo, G. Ecker, J. Gasser, in: L. Maiani, G. Pancheri, N. Paver (Eds.), The Second
DAPHNE Physics Handbook, 1995, hep-ph/9411311.
[7] J. Bijnens, in: J. Rosner, B. Winstein (Eds.), 1999 Conference on Kaon Physics, Chicago, IL, USA, 21–26
June 1999, Chicago Univ. Press, Chicago, IL, 2001, p. 395, hep-ph/9907514.
[8] J. Bijnens, G. Colangelo, G. Ecker, Phys. Lett. B 441 (1998) 437, hep-ph/9808421.
[9] N.H. Fuchs, M. Knecht, J. Stern, Phys. Rev. D 62 (2000) 033003, hep-ph/0001188.
[10] P. Post, K. Schilcher, Eur. Phys. J. C 25 (2002) 427, hep-ph/0112352.
[11] G. Amorós, J. Bijnens, P. Talavera, Nucl. Phys. B 568 (2000) 319, hep-ph/9907264.
[12] G. Amorós, J. Bijnens, P. Talavera, Phys. Lett. B 480 (2000) 71, hep-ph/9912398;
G. Amorós, J. Bijnens, P. Talavera, Nucl. Phys. B 585 (2000) 293;
G. Amorós, J. Bijnens, P. Talavera, Nucl. Phys. B 598 (2001) 665, hep-ph/0003258, Erratum.
[13] G. Amorós, J. Bijnens, P. Talavera, Nucl. Phys. B 602 (2001) 87, hep-ph/0101127.
[14] J. Bijnens, P. Talavera, JHEP 0203 (2002) 046, hep-ph/0203049.
[15] V. Cirigliano, M. Knecht, H. Neufeld, H. Rupertsberger, P. Talavera, Eur. Phys. J. C 23 (2002) 121, hep-
ph/0110153.
[16] A. Pich, Lectures at Les Houches Summer School in Theoretical Physics, Session 68: Probing the Standard
Model of Particle Interactions, Les Houches, France, 28 July–5 September 1997, hep-ph/9806303;
G. Ecker, Lectures given at Advanced School on Quantum Chromodynamics (QCD 2000), Benasque,
Huesca, Spain, 3–6 July 2000, hep-ph/0011026;
S. Scherer, hep-ph/0210398.
[17] J. Bijnens, G. Colangelo, G. Ecker, JHEP 9902 (1999) 020, hep-ph/9902437.
[18] J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
[19] J. Bijnens, G. Colangelo, G. Ecker, J. Gasser, M.E. Sainio, Nucl. Phys. B 508 (1997) 263;
J. Bijnens, G. Colangelo, G. Ecker, J. Gasser, M.E. Sainio, Nucl. Phys. B 517 (1998) 639, hep-ph/9707291,
Erratum.
[20] P. Post, K. Schilcher, Phys. Rev. Lett. 79 (1997) 4088, hep-ph/9701422.
[21] P. Post, K. Schilcher, Nucl. Phys. B 599 (2001) 30, hep-ph/0007095.
[22] R.F. Dashen, L. Ling-Fong, H. Pagels, M. Weinstein, Phys. Rev. D 6 (1972) 834.
[23] A. Sirlin, Ann. Phys. 61 (1970) 294;
A. Sirlin, Phys. Rev. Lett. 43 (1979) 904.
[24] R.E. Behrends, A. Sirlin, Phys. Rev. Lett. 4 (1960) 186;
M. Ademollo, R. Gatto, Phys. Rev. Lett. 13 (1964) 264.
[25] R.F. Dashen, M. Weinstein, Phys. Rev. Lett. 22 (1969) 1337.
[26] J. Bijnens, P. Dhonte, Scalar form-factors at two-loop in ChPT, hep-ph/0307044.
[27] M. Jamin, J.A. Oller, A. Pich, Nucl. Phys. B 622 (2002) 279, hep-ph/0110193.
[28] G. Donaldson, et al., Phys. Rev. D 9 (1974) 2960;
G. Donaldson, Phys. Rev. Lett. 31 (1973) 337.
[29] A.R. Clark, R.C. Field, W.R. Holley, R.P. Johnson, L.T. Kerth, R.C. Sah, G. Shen, Phys. Rev. D 15 (1977)
553.
[30] V.K. Birulev, et al., Sov. J. Nucl. Phys. 31 (1980) 622, Yad. Fiz. 31 (1980) 1204 (in Russian);
V.K. Birulev, et al., Nucl. Phys. B 182 (1981) 1.
[31] C.D. Buchanan, et al., Phys. Rev. D 11 (1975) 457.
[32] D.G. Hill, et al., Nucl. Phys. B 153 (1979) 39.
[33] I.V. Ajinenko, et al., Phys. Atom. Nucl. 66 (2003) 105, Yad. Fiz. 66 (2003) 107 (in Russian), hep-
ph/0202061.
[34] A. Apostolakis, et al., CPLEAR Collaboration, Phys. Lett. B 473 (2000) 186.
[35] S. Gjesdal, et al., Nucl. Phys. B 109 (1976) 118.
[36] I.V. Ajinenko, et al., Phys. Atom. Nucl. 65 (2002) 2064, Yad. Fiz. 65 (2002) 2125 (in Russian), hep-
ex/0112023.
[37] A.S. Levchenko, et al., KEK-PS E246 Collaboration, Phys. Atom. Nucl. 65 (2002) 2232, Yad. Fiz. 65 (2002)
2294 (in Russian), hep-ex/0111048.
[38] R.J. Tesarek, KTeV Collaboration, Talk given at American Physical Society (APS) Meeting of the Division
of Particles and Fields (DPF 99), Los Angeles, CA, 5–9 January 1999, hep-ex/9903069.
[39] S. Shimizu, et al., KEK-E246 Collaboration, Phys. Lett. B 495 (2000) 33.
[40] J. Bijnens, G. Colangelo, P. Talavera, JHEP 9805 (1998) 014, hep-ph/9805389.
[41] S. Pislak, et al., BNL-E865 Collaboration, Phys. Rev. Lett. 87 (2001) 221801, hep-ex/0106071.
[42] J. Bijnens, P. Talavera, in preparation.
[43] G. Furlan, et al., Nuovo Cimento 38 (1965) 1747.
[44] V. Cirigliano, Talk given at 38th Rencontres de Moriond on Electroweak Interactions and Unified Theories,
Les Arcs, France, 15–22 March 2003, hep-ph/0305154.
[45] J.A. Vermaseren, math-ph/0010025.
A microscopical description of giant gravitons II:

the AdS5 × S 5 background
Bert Janssen a , Yolanda Lozano b , Diego Rodríguez-Gómez b,c
a Institute for Theoretical Physics, Katholieke Universiteit Leuven, Celestijnenlaan 200D,
B-3001 Leuven, Belgium
b Departamento de Física, Universidad de Oviedo, Avda. Calvo Sotelo 18, 33007 Oviedo, Spain
c Departamento de Física Teórica, Universidad Autónoma de Madrid,
Cantoblanco, 28049 Madrid, Spain
Received 26 March 2003; accepted 13 June 2003
Abstract
In this article we continue with the microscopical investigation of giant graviton configurations in
AdSm × S n spacetimes, initiated in hep-th/0207199. Using dualities and a matrix theory derivation
we propose an action that describes multiple type IIB gravitons. This action contains multipole
moment couplings to the type IIB background potentials. Using these couplings, we study, from
the microscopical point of view, the giant graviton and dual giant graviton configurations in the
AdS5 × S 5 background. In both cases the gravitons expand into a non-commutative 3-sphere, that is
defined as an S 1 -bundle over a fuzzy 2-sphere. When the number of gravitons is large we find perfect
agreement with the Abelian, macroscopical description of giant gravitons in this spacetime, given in
the literature.
1. Introduction
Giant gravitons [1–4] are stable brane configurations with non-zero angular momentum,
that are wrapped around (n − 2)- or (m − 2)-spheres in AdSm × S n spacetimes, and carry a
dipole moment with respect to the background gauge potential. They are not topologically
stable, but are at a dynamical equilibrium because the contraction due to the tension of the
brane is precisely cancelled by the expansion due to the coupling of the angular momentum
to the background flux field. It turns out that these spherical brane configurations are
E-mail addresses: bert.janssen@fys.kuleuven.ac.be (B. Janssen), yolanda@string1.ciencias.uniovi.es

(Y. Lozano), diego@fisi35.ciencias.uniovi.es (D. Rodríguez-Gómez).
doi:10.1016/S0550-3213(03)00532-7
364 B. Janssen et al. / Nuclear Physics B 669 (2003) 363–378
massless, conserve the same number of supersymmetries and carry the same quantum
numbers of a graviton. The fact that they are extended objects of finite size has lead to
the name of giant gravitons.
These configurations were first proposed in [1] as a way to satisfy the stringy exclusion
principle implied by the AdS/CFT correspondence. The spherical (n − 2)-brane expands
into the S n part of the geometry with a radius proportional to its angular momentum. Since
this radius is bounded by the radius of the S n , the configuration has associated a maximum
angular momentum. The (m − 2)-brane giant graviton configurations later found in [2,3]
expand, on the other hand, into the AdSm component of the spacetime, so that there is no
upper bound implied on their angular momentum. They are referred to in the literature
[2] as dual giant gravitons. One expects that quantum mechanically these different states
will tunnel into each other, forming a unique ground state. In [2] it is speculated that the
stringy exclusion principle may still be satisfied if once the quantum mixing is taken into
account there is no supersymmetric ground state for angular momenta beyond the exclusion
principle bound. Other giant graviton solutions in various backgrounds have been studied
elsewhere in the literature [5,6]. Giant gravitons in pp-wave backgrounds have been studied
in [7–11].
The appearance of giant gravitons as blown up massless particles hints to a connection
with other examples of expanded brane configurations, more precisely to the dielectric
effect, where multiple coinciding Dp-branes can expand into higher-dimensional D-brane
configurations. There are two complementary descriptions of this effect. Consider the case
in which the Dp-branes expand into a spherical D(p + 2)-brane. The first one is an Abelian
description, describing the spherical D(p + 2)-brane with a large number of Dp-branes
dissolved on its worldvolume [12]. The second one is a non-Abelian formulation [13–15],
describing how multiple coinciding Dp-branes expand into a D(p + 2)-brane with the
topology of a fuzzy 2-sphere [15]. Both descriptions agree in the limit where the number
N of Dp-branes is very large.
It has been suggested in the literature (see for example [7,16–18]) that it should be
possible to describe the giant gravitons of [1–4] in terms of dielectric gravitational waves.
Since massless particles, in particular gravitons, are the source terms for gravitational
waves, it is natural to expect that a dielectric effect for these gravitational waves will
provide a microscopic picture for the giant graviton configurations. By analogy with
the dielectric effect for D-branes, it is believed that in the limit when the number of
gravitons, N , is large, this non-Abelian, microscopical description should match the
Abelian, macroscopical description of [1–4], in terms of an spherical brane with angular
momentum, where microscopically the angular momentum of the Abelian description is
interpreted as the total momentum of the multiple gravitational waves, propagating in the
spherical part of the geometry.
In [18] an action describing type IIB gravitational waves was derived using matrix string
theory in a weakly curved background. This action shows the linear couplings to closed
string backgrounds, and contains the now familiar multipole couplings that give rise to the
dielectric effect [15]. From them one can construct configurations of multiple coinciding
gravitational waves expanding into higher-dimensional fuzzy surfaces. However, with a
non-Abelian action for gravitational waves only known up to linear order in the background
fields it is not possible to check whether dielectric gravitational waves really provide a
B. Janssen et al. / Nuclear Physics B 669 (2003) 363–378 365
microscopical description for the giant gravitons in AdSm × S n backgrounds, given that
the AdSm × S n spacetime cannot be taken as a linear perturbation to Minkowski.
In Ref. [19] we gave one step further in this direction, by providing a closed expression
for the worldvolume action for multiple M-theory gravitons, valid beyond the linear
approximation. A linear action for M-theory gravitons can easily be constructed by just
uplifting the linear action for multiple type IIA gravitons constructed in [18] using matrix
string theory in a weakly curved background.1 Demanding consistency with the action
for coincident D0-branes when the gravitons propagate along the eleventh direction it is
possible to extend this action beyond the linear approximation used in the matrix theory
calculation. In fact, a non-trivial check of this action is that it predicts the polarisation of the
gravitons into fuzzy surfaces, in such a way that in AdSm × S n backgrounds the energy and
radius of the dielectric configurations exactly match those of the giant graviton solutions
of [1,2].2
With a closed expression for the action for M-theory gravitational waves it is possible to
construct an action for type IIA gravitational waves valid beyond the linear approximation
of [18]. This action was presented in [19]. Given that type II waves are simply related by
a T-duality transformation, it is straightforward to construct an action for multiple type
IIB gravitational waves valid beyond the linear approximation of [18]. With this action it
is then possible to construct explicit configurations of multiple type IIB gravitons in the
AdS5 × S 5 background, which we can compare with the giant graviton [1] and the dual
giant graviton [2,3] solutions in this spacetime. This is precisely the aim of this paper.
The construction of non-Abelian giant gravitons in AdS5 × S 5 turns out to be more
involved than the eleven-dimensional non-Abelian giant graviton solutions of [19]. In this
background the gravitational waves expand into a spherical D3-brane, and hence a fuzzy 3-
sphere ansatz is needed in the non-Abelian construction, rather than the well-known fuzzy
2-sphere, in terms of SU(2) matrices. Non-commutative odd spheres have been constructed
in [20–22], as subspaces of fuzzy even spheres. It turns out that these constructions are not
applicable to our case. The main reason is that due to our construction of the action for
type IIB gravitational waves via T-duality, this action exhibits an extra isometry direction
associated to the coordinate over which the T-duality was performed. This isometry will
require a non-manifestly SO(4)-symmetric solution, which is not of the type presented in
[20–22]. Furthermore, the fact that we are dealing with gravitational waves, which have
a one-dimensional worldvolume, allows only non-Abelian couplings similar to the ones
one encounters in the case of fuzzy two-spheres. The solution to this apparent paradox
is to consider the three-sphere as a U (1)-bundle over S 2 and choose the extra isometry
direction to be the fibre coordinate. The fuzzy version of the S 3 will then consist of a
(Abelian) U (1)-fibre over a fuzzy S 2 .
1 The derivation in [18] considers matrix string theory in a weakly curved background in the Sen–Seiberg
limit, so that the strings carry spatial momentum, and then takes static gauge. In this way one arrives at an action
that describes multiple massless particles carrying momentum. In the Abelian limit, this action is related to the
perturbative action for massless particles through a Legendre transformation (we refer the reader to [18] for the
details of this construction).
2 The explicit calculation in [19] is for the giant graviton solution in AdS × S 4 and the dual giant graviton
7
solution in AdS4 × S 7 , i.e., the cases involving fuzzy 2-spheres.
The paper is organised as follows: we start in Section 2 with the construction of the
action for type IIB gravitational waves in the way explained in the previous paragraphs.
This action is valid beyond the linear order approximation taken in [18], and can therefore
be used to study the AdS5 × S 5 background considered in this paper. With this action,
we provide, in Section 3, a microscopical description of the giant graviton configuration
of [1] in AdS5 × S 5 . This solution is expected to occur, microscopically, in the form of
N gravitons polarised in a fuzzy 3-sphere contained in S 5 . Describing the S 3 as an S 1
bundle over S 2 and making non-commutative the base 2-sphere we find a giant graviton
solution which in the large N limit reproduces exactly the (Abelian) result of [1]. This is
a non-trivial check of the validity of our microscopical description and, in particular, of
our non-commutative ansatz. Further support is provided by the calculation in Section 4.
In this section we construct microscopically the dual giant graviton solution of [2,3].
Describing the 3-sphere contained in AdS5 as an S 1 bundle over S 2 and making this S 2 non-
commutative we find a dual giant graviton solution which is also in perfect agreement with
the results in [2,3] when N is large. Section 5 contains the discussion, where we comment
on the relations between our non-commutative 3-sphere solutions and other constructions
of odd non-commutative spheres previously discussed in the literature [20–22].
2. The action for multiple type IIB gravitational waves
In this section we construct an action for type IIB gravitational waves suitable for the
study of giant graviton configurations in the AdS5 × S 5 background.3 The starting point is
the action for multiple type IIA gravitational waves presented in [19]. This action is simply
the weakly coupled (type IIA) version of the action there proposed for the microscopical
study of M-theory gravitons in AdSm × S n spacetimes. The M-theory action of [19] gives
the couplings of multiple gravitational waves to M-theory background fields, in the form of
a closed expression valid beyond the weak background field limit. The extension beyond
the linear order of matrix theory can be done by demanding agreement with the action for
multiple D0-branes when the gravitons propagate along the eleventh direction. T-dualising
the action for the type IIA waves we will obtain an action for type IIB waves also valid
beyond the linear order approximation taken in [18].
The Born–Infeld action proposed in [19] to describe multiple type IIA gravitational
waves is given by
i
SWA = −T0 dτ Str k −1 −P E00 + E0i Q−1 − δ k E kj Ej 0 det Qij ,
BI
(2.1)
where

Eµν = gµν − k −2 kµ kν + k −1 eφ ik C (3) µν , (2.2)

Qij = δji + i Xi , Xk e−φ kEkj , i, j = 1, . . . , 9. (2.3)
3 We will be dealing throughout the article with only the bosonic part of the worldvolume effective actions.
This is enough for the study of the bosonic giant graviton configurations.
In this action the direction of propagation of the waves occurs as an isometry direction,
k µ being the Killing vector pointing in this direction.4 In our notation k 2 = gµν k µ k ν
(3)
and (ik C (3) )µν = k ρ Cρµν . This action is manifestly invariant under global gauge
transformations along the Killing direction: δXµ = Λk µ , since this direction is projected
out through the effective metric Gµν = gµν − k −2 kµ kν and the contraction of the 3-form
with k µ . The condition that k µ Gµν = 0 implies that the gravitational field is transversal to
the direction of propagation of the waves. For simplicity, this action has been calculated
for vanishing C (1) , B (2) and the worldvolume scalar field A.5
We now make a T-duality transformation along a transverse direction Z in order to
obtain the action for type IIB gravitational waves. This transformation was explained
in detail in Ref. [18], for the linearised action. The action for type IIB gravitational
waves contains two worldvolume scalars A and Z, being, respectively, the T-dual of the
type IIA worldvolume scalar A and the T-dual of the embedding scalar in the direction
of the dualisation. Their gauge invariant curvatures are given by F = ∂A − P [il C (2) ]
and F = ∂Z + P [il B (2) ] and together they form a doublet under the type IIB S-duality
transformations. Consistently with S-duality and the truncation imposed in the type IIA
action, we will set C (2) , B (2) , A and Z equal to zero. Also, for simplicity we take kz = 0.
This truncation is suitable for the study of gravitational waves in the AdS5 ×S 5 background,
as we show below. Hence, we obtain the following Born–Infeld action

−1
i
SWBI
= −T dτ Str k −P E + E Q−1 − δ i E kj E det Qj , (2.4)
B 0 00 0i k j 0
where now

Eµν = gµν − k −2 kµ kν − l −2 lµ lν − k −1 l −1 eφ ik il C (4) µν , (2.5)

Qij = δji + i Xi , Xk e−φ klEkj . (2.6)
Here l µ is a new Killing vector, pointing along the direction in which we performed the
T-duality transformation. It is easy to check that this action is invariant under the global
isometric transformations generated by k µ and l µ : δXµ = Λ(1) k µ + λ(2) l µ , since these
directions are projected out through the effective metric Gµν = gµν − k −2 kµ kν − l −2 lµ lν
and the double contraction of the 4-form with the two Killing vectors. It is due to the
fact that we have two isometric directions that C (4) can couple to the Born–Infeld part
of the action. This coupling plays a key role in the microscopical description of the
AdS5 × S 5 giant graviton solution [1], that we will perform in the next section. In this
action k µ still points in the direction of propagation of the waves, since under the T-duality
transformation and with the truncations above the monopole term is mapped onto itself.
The other isometric direction, Z, is however not physical, but just an artifact of the
T-duality transformation. We are therefore describing waves which are smeared in the
Z-direction. In the Abelian limit, (2.4) is still a complicated expression containing two
4 This is inferred by the analysis of the monopole term in the Chern–Simons effective action, which shows
that the momentum arises as the charge with respect to the background field k −2 kµ . For details see [18].
5 The worldvolume scalar field comes from the reduction of the eleventh scalar, and forms an invariant field
strength with the RR 1-form potential.
isometric directions. However, together with the (Abelian) Chern–Simons coupling, which
is given by [18]

CS
SW B
= T 0 dτ k −2 kµ ∂Xµ , (2.7)
this effective action can be related to the (dimensional reduction along the Z direction of
the) usual perturbative action for massless particles

NT0
S[γ ] = − dτ |γ | γ −1 ∂Xµ ∂Xν gµν (2.8)
2
by means of a Legendre transformation that restores the dependence on the direction of
propagation (see Section 2 in [18] for the details). In the non-Abelian case, however, it is
not possible to restore this dependence. First of all, this direction is not matrix-valued, so
even though we could in principle restore some explicit dependence on its time derivative,
the new terms would be Abelian. Second, it is clear that the Legendre transformation cannot
give rise to non-Abelian commutators involving this direction, so we would end up in any
case with an action with reduced transverse rotational invariance. Thus, we are constrained
to work, in the non-Abelian case, with an action with two isometries, and assume the
presence of the extra unphysical isometry.
This isometry will however play a key role in our microscopical description of giant
graviton configurations in AdS5 × S 5 . Indeed, the presence of this compact isometry
suggests the representation of the 3-sphere of the giant graviton configurations as a U (1)
bundle over S 2 , with the U (1) invariance being associated to translations along this
direction. We will see that the giant graviton configurations correspond to the polarisation
of the gravitons in fuzzy 3-spheres represented as U (1) bundles over a fuzzy S 2 .
Finally, we turn to the Chern–Simons part of the action for the type IIB waves.
This action was constructed, to linear order in the background fields, in Ref. [18] (see
expression (4.5)). In particular, the linear coupling to the 4-form RR-potential, relevant for
the construction of giant gravitons in the AdS5 × S 5 background, was shown to be given by

SWB = −i dτ STr P (iX iX )il C (4) .
CS
(2.9)
In this action the pull-backs into the worldvolume are defined in terms of gauge covariant
derivatives
DXµ = ∂Xµ − A(1)k µ − A(2) l µ = ∂Xµ − k −2 kρ ∂Xρ k µ − l −2 lρ ∂Xρ l µ (2.10)

with respect to the scaling symmetry
δXµ = Λ(1)(τ )k µ + Λ(2)(τ )l µ . (2.11)

In this way we ensure (local) invariance under the isometric transformations generated by
the two Killing vectors. Using gauge covariant pull-backs it is possible to eliminate the
pull-back of the isometric coordinates, and to reproduce the isometric couplings in the
action in a manifestly covariant way. For example, the pull-back of the reduced metric in
(2.4), Gµν ∂Xµ ∂Xν , can be written as gµν DXµ DXν , in terms of the covariant pull-backs.
The action is then given by a gauged sigma model of the type considered in [23–25].
The coupling in (2.9) plays a key role in the microscopical description of the AdS5 × S 5
dual giant graviton solution [2,3], that we perform in Section 4. It is due to the fact that
we have a second isometric direction that the C (4) electric potential of this background can
couple in the one-dimensional worldvolume effective action of the gravitational waves.
The effective action that we will use in the next sections to study the giant graviton
configurations in the AdS5 × S 5 background is then given by
SWB = SW
BI
B
+ SW
CS
B
(2.12)
BI given by (2.4) and S CS by (2.9).
with SW B WB
3. The giant graviton in AdS5 × S 5
3.1. The macroscopical description
The giant graviton solution in AdS5 × S 5 was computed in [1], by looking at stable
test brane solutions where a D3-brane with angular momentum in S 5 had expanded to a
3-sphere contained inside the S 5 . We briefly review this construction in order to compare
it, in the end, with our microscopical description.
Taking the line element for the metric on AdS5 × S 5 as ds 2 = dsAdS
2 + dsS2 , with

r2 dr 2
2
dsAdS = − 1 + 2 dt 2 + 2
+ r 2 dΩ32 , (3.1)
L 1 + Lr 2
and

dsS2 = L2 dθ 2 + cos2 θ dφ 2 + sin2 θ dΩ32 , (3.2)
the trial solution considered in [1] has θ = constant, φ = φ(τ ), where t = τ in static gauge,
and r = 0, i.e., it corresponds to a spherical D3-brane with radius L sin θ orbiting the S 5 in
the φ direction:
ds 2 = −dt 2 + L2 cos2 θ dφ 2 + L2 sin2 θ dΩ32 . (3.3)

This D3-brane carries a non-vanishing magnetic moment with respect to the RR 4-form
potential of the background, which prevents its collapse to zero size. Parametrising the
unit 3-sphere in (3.3) as

dΩ32 = dβ12 + sin2 β1 dβ22 + sin2 β2 dβ32 , (3.4)
the RR 4-form potential is given by
(4) √
Cφβ1 β2 β3 = L4 sin4 θ gβ , (3.5)
√
where gβ is the volume element on the unit 3-sphere.
Substituting this trial solution into the worldvolume action of the D3-brane, one arrives
at the following Hamiltonian:
2
Pφ N
H= 1 + tan2 θ 1 − sin2 θ . (3.6)
L Pφ
is the integer arising through the quantisation condition of the 4-form flux on S 5
Here N

N
2π 2 T3 = , (3.7)
L4
with T3 the tension of the D3-brane, and Pφ is the angular momentum carried by the brane,
which is a constant given that φ is a cyclic coordinate.
The Hamiltonian (3.6) has two stable minima, one for sin θ = 0 and another for

sin2 θ = Pφ /N. (3.8)
The value of the Hamiltonian is in both cases E = Pφ /L, i.e., both solutions represent
massless particles with angular momentum Pφ . So the first solution corresponds to a
pointlike graviton, while the second minimum corresponds to a D3-brane with radius
L(Pφ /N )1/2 , which is the giant graviton solution. The giant graviton satisfies the stringy
exclusion principle, since the condition sin2 θ 1 implies an upper bound on the angular
momentum: Pφ N .
3.2. The microscopical description
In this section we provide a description of the giant graviton solution in terms of

coincident gravitons expanding into a D3-brane, which is inside the S 5 -part of the
background geometry. We will check the correctness of this description by looking
whether, for a large number of gravitons, it is in agreement with the previous macroscopical
description in terms of a test D3-brane.
The similarity between the giant graviton construction in [1] and the Abelian description
of the dielectric (or magnetic moment) effect for Dp-branes suggests a microscopical
description of the giant graviton in terms of gravitons expanding into a D3-brane with the
topology of a fuzzy 3-sphere. In this description the expansion of the gravitons takes place
due to their interaction with the RR 4-form potential of the background. At the level of the
graviton worldvolume effective action this interaction occurs in the form of a non-Abelian
dielectric coupling.
The action that we have proposed in Section 2 to describe type IIB gravitons
contains two isometric directions, one of which is identified with the direction of
propagation, whereas the other one reflects the fact that the background on which the
gravitons propagate contains a compact direction, which is the direction of the T-duality
transformation involved in the construction. The point now is to identify these isometries
in the background (3.3). It is clear that the first isometry lays in the φ-direction. The second
isometric direction can, on the other hand, be identified when one considers the 3-sphere
as a U (1)-bundle over S 2 . The natural choice for the second isometry is the direction
associated to the U (1)-fibre.
Representing the S 3 with radius L sin θ as a submanifold of C2 with coordinates

(z0 , z1 ) satisfying z̄0 z0 + z̄1 z1 = L2 sin2 θ , the Hopf fibering, p : S 3 → S 2 , is given by a
stereographic projection of a point (z0 , z1 ) of the S 3 to a point z of the S 2 (see for example
[26]):

z = z1 /z0 when z0 = 0,
p(z0 , z1 ) = (3.9)
1/z = z0 /z1 when z1 = 0.
Points on the S 3 that differ by an overall factor λ ∈ U (1) get mapped onto the same point
z of S 2 : p(z0 , z1 ) = p(λz0 , λz1 ). In this way the Hopf map is dividing out a U (1) fibre in
the S 3 . Inversely the coordinates xi (i = 1, 2, 3) on the S 2 with x12 + x22 + x32 = R 2 can be
obtained from the Hopf map via
2 2 1
x1 = Re(z0 z̄1 ), x2 = Im(z0 z̄1 ), x3 = |z0 |2 − |z1 |2 , (3.10)
a a a
where a is an arbitrary parameter with dimension of length, that relates the radius L sin θ
of the S 3 with the radius R of the S 2 :

3
1 L4 sin4 θ
R2 = (xi )2 = 2 [z̄0 z0 + z̄1 z1 ]2 = . (3.11)
a a2
i=1
Parametrising the geometry of the S 3 in terms of Euler angles

χ1 χ1
z0 = L sin θ ei(χ3 +χ2 )/2 cos , z1 = L sin θ ei(χ3 −χ2 )/2 sin , (3.12)
2 2
where χ1 ∈ [0, π[, χ2 ∈ [0, 2π[ and χ3 ∈ [0, 4π[, we get the round metric for S 3 :
L2 sin2 θ 2
dsS23 = dχ1 + sin2 χ1 dχ22 + (dχ3 + cos χ1 dχ2 )2 . (3.13)
4
Here the angles χ1 and χ2 parametrise the S 2 -base manifold and χ3 the S 1 -fibre bundle.
Note that the metric has the necessary twist in the fibre in order to obtain the S 3 as the
global space. The coordinates xi on the S 2 are then given by
x1 = R sin χ1 cos χ2 , x2 = R sin χ1 sin χ2 , x3 = R cos χ1 , (3.14)

and we can identify the coordinate χ3 with the isometric direction in the gravitons effective
action not associated to the direction of propagation.
Using Euler angles to write the metric of the three-sphere in coordinates adapted to the
fibre structure, the background metric (3.3) and four-form gauge field are given by
ds 2 = −dt 2 + L2 cos2 θ dφ 2
L2 sin2 θ 2
+ dχ1 + sin2 χ1 dχ22 + (dχ3 + cos χ1 dχ2 )2 ,
4
(4) 1
Cφχ1 χ2 χ3 = − L4 sin4 θ sin χ1 . (3.15)
8
In order to make a non-Abelian ansatz, it is convenient to go to Cartesian coordinates
describing the S 2 -base manifold of the S 3 , keeping in mind the constraint that (x1 )2 +
(x2 )2 + (x3 )2 = R 2 . In these coordinates, the metric and the four-form RR-field become
L2 sin2 θ 2
ds 2 = −dt 2 + L2 cos2 θ dφ 2 + 2
dx1 + dx22 + dx32
4R
2
1 2 2 x3
+ L sin θ dχ3 + (x1 dx2 − x2 dx1 ) , (3.16)
4 R(x12 + x22 )
(4) L4 sin4 θ
Cχ3 φij = 5ij k x k , for i, j, k = 1, 2, 3.
8R 3
We can then make the following non-commutative ansatz for the 2-sphere that is
parametrised by x 1 , x 2 , x 3 :
R
Xi = √ J i, i = 1, 2, 3, (3.17)
N2 − 1
with the J i , forming an N × N representation of SU(2) (in our conventions [J i , J j ] =
2i5 ij k J k ). Trivially, with this choice
1 2 2 2 3 2
X + X + X = R 2 1, (3.18)
so we are dealing with a non-commutative version of the S 2 contained in (3.15). Therefore,
with this non-commutative ansatz, the 3-sphere becomes an S 1 -bundle over a fuzzy S 2 .
This situation is forced by the topology of the space in which the type IIB gravitons
propagate, having an extra compact S 1 direction. Thus, the physical picture will correspond
to the gravitons expanding into a non-commutative D3-brane, described in coordinates that
2
reflect a Sfuzzy × S 1 structure. To see that this is the right microscopical picture we have to
check that in the limit in which the number of gravitons is large we recover the description
in [1], in terms of a D3-brane with the topology of a classical, commutative, S 3 . We will
carry out this calculation in the remaining part of this section.
The action (2.12) for type IIB waves in the AdS5 × S 5 background described by (the
non-commutative version of) the coordinates (3.16), contains

L2 sin2 θ sin θ
E00 = −1, E0i = 0, Eij = δij − k
5ij k X ,
4R 2 R cos θ
L4 sin3 θ cos θ ij k L4 sin4 θ i
Qi j = δji − √ 5 Xk + √ X Xj − δji X2 (3.19)
4R N 2 − 1 4R 2 N 2 − 1
in the Born–Infeld part, while the contributions to the Chern–Simons part vanish. We then
find

1
SWB = −T0 dτ STr
L cos θ

L4 sin4 θ L8 sin6 θ cos2 θ 2 L8 sin8 θ
× 1− √ X +
2 X + 2 2
X X ,
2R 2 N 2 − 1 16R 2 (N 2 − 1) 16R 4 (N 2 − 1)
(3.20)
where we have dropped those contributions to det Q that will vanish when taking the
symmetrised average involved in the symmetrised trace prescription. In our description,
since the direction of propagation is isometric, we are effectively dealing with a static
configuration, for√which we can compute the potential as minus the Lagrangian. In the
limit L4 sin4 θ N 2 − 1 we can approximate the square root by
L4 sin4 θ L8 sin6 θ cos2 θ 2

1− √ X2 + X , (3.21)
4R 2 N 2 − 1 32R 2 (N 2 − 1)
and since X2 = R 2 1, we have for the potential

NT0 L4 sin4 θ L8 sin6 θ cos2 θ
VWB (θ ) = 1− √ + , (3.22)
L cos θ 4 N2 − 1 32(N 2 − 1)
which can be seen as the first order expansion of

NT0 L4 sin4 θ L8 sin6 θ
VWB (θ ) = 1− √ + (3.23)
L cos θ 2 N 2 − 1 16(N 2 − 1)
in the same limit above. Introducing cos θ inside the square root we have

NT0 L4 sin2 θ 2
VWB (θ ) = 1 + tan θ 1 − √
2 . (3.24)
L 4 N2 − 1
This potential has two minima, the point-like graviton at sin θ = 0, and a solution that
should correspond to the giant graviton solution at
√
4 N2 − 1
sin θ =
2
. (3.25)
L4
Both solutions have an energy E = NT0 /L, and are therefore associated to massless
particles with angular momentum Pφ = NT0 . There is an upper bound on the angular
momentum that comes from the condition sin2 θ 1, which implies that
T0
8
Pφ = NT0 L + 16. (3.26)
4
Comparing the potential (3.24), the radius (3.25) and the upper bound (3.26) with their
Abelian counterparts in the macroscopical derivation, we find that there is exact agreement
in the large N limit when

N L4
= ⇐⇒ Pφ = 8π 2 NT3 . (3.27)
Pφ 4N
Making use of the fact that in the microscopical description the total angular momentum
is quantised in terms of the tension of the gravitational waves, Pφ = NT0 , we obtain the
following relation between the tension T0 of the waves and the tension T3 of the D3-brane
T0 = 8π 2 T3 , (3.28)
which is indeed satisfied, given that the isometric transverse direction is an angular variable
with period 4π .
4. The dual giant graviton in AdS5 × S 5
4.1. The macroscopical description
The dual giant graviton solution in AdS5 × S 5 was computed in [2,3], by looking at
stable test brane solutions where the D3-brane with angular momentum in S 5 expands to
the 3-sphere contained inside the AdS5 component of the spacetime. For this giant graviton
type of solution there is no upper bound for the angular momentum, because it expands in
the non-compact part of the geometry. Let us briefly summarise the construction in these
references in order to compare it, in the end, with our microscopical description.
The trial solution in this case is taken with r = constant, φ = φ(τ ) and θ = 0, which
corresponds to a spherical D3-brane with radius r orbiting the S 5 in the φ direction:

r2
ds 2 = − 1 + 2 dt 2 + r 2 dΩ32 + L2 dφ 2 . (4.1)
L
This D3-brane carries a non-vanishing dipole moment with respect to the RR 4-form
potential, which prevents its collapse to zero size. Parametrising the unit 3-sphere in (4.1)
as

dΩ32 = dα12 + sin2 α1 dα22 + sin2 α2 dα32 , (4.2)
we have
(4) r4 √
C0α1 α2 α3
=− gα . (4.3)
L
Substituting this trial solution into the worldvolume action of the D3-brane, one arrives at
the following Hamiltonian [2,3]:
2 r 6 N r 4
1 r2 N
H= 1+ 2 Pφ +
2 − 4 , (4.4)
L L L6 L
is given by (3.7).
where N
The stable solutions correspond to r = 0, the point-like graviton, and r = L Pφ /N , the
dual giant graviton, both of which have energy E = Pφ /L, representing massless particles
with angular momentum Pφ . Contrary to the giant graviton solution of the previous section,
the dual giant graviton solution does not satisfy the stringy exclusion principle, because the
absence of an upper bound for its radius implies that Pφ is neither bounded.
4.2. The microscopical description
We now want to provide a microscopical description of the dual giant graviton solution.
This description will be in terms of coincident gravitons expanding into a D3-brane which
is now expanded in a fuzzy surface contained inside the AdS5 part of the geometry. As
in the previous section, we represent the S 3 in (4.1) as an S 1 bundle over S 2 , take Euler
angles such that
r2 2
dsS23 = dχ1 + sin2 χ1 dχ22 + (dχ3 + cos χ1 dχ2 )2 (4.5)
4
and identify the coordinate χ3 parametrising the S 1 with the compact isometric direction
coming from the T-duality construction. The isometry associated to the propagation
direction is again taken to be φ. The background metric (4.1) is then given by

r2 r2 2
ds 2 = − 1 + 2 dt 2 + dχ1 + sin2 χ1 dχ22 + (dχ3 + cos χ1 dχ2 )2 + L2 dφ 2 .
L 4
(4.6)
Taking Cartesian coordinates in the S 2 that is parametrised by χ1 and χ2
x 1 = R sin χ1 cos χ2 , x 2 = R sin χ1 sin χ2 , x 3 = R cos χ1 , (4.7)

the metric and 4-form potential take the form

r2 r2 2
ds 2 = − 1 + 2 dt 2 + L2 dφ 2 + dx 1 + dx 2
2 + dx 2
3
L 4R 2
2
r2 x3
+ dχ3 + (x 1 dx 2 − x 2 dx 1 ) ,
4 R(x12 + x22 )
(4) r4
Cχ3 0ij = − 5ij k x k (4.8)
8R 3 L
for i, j, k = 1, 2, 3. Thus, the natural non-commutative ansatz to make in this case is
R
Xi = √ J i, (4.9)
N2 − 1
with J i forming an N × N representation of SU(2). Our description of the dual giant
graviton will again be in terms of gravitons expanding into a non-commutative D3-brane
2
with topology Sfuzzy × S 1 , the validity of which should be tested by checking the agreement
with the macroscopical description for large number of gravitons.
The action (2.12) for type IIB waves in the particular background defined by (the non-
commutative version of) (4.8) contains

r2 r2
E00 = − 1 + 2 , E0i = 0, Eij = δij , (4.10)
L 4R 2
Lr 3
Qij = δji − √ 5 ij k Xk . (4.11)
4R N 2 − 1
The Born–Infeld part of the action then takes the form

T0 r2 L2 r 6
SWB = −
BI
dτ STr 1+ 2 1+ X 2 , (4.12)
L L 16R 2 (N 2 − 1)
while the Chern–Simons part is, in turn, given by

NT0 r4
SWB = −T0 dτ STr iP (iX iX )il C
CS (4)
= dτ √ . (4.13)
4 N2 − 1 L
We then have a potential

T0 r2 L2 r 6 NT0 r4
VWB (r) = STr 1+ 2 1+ X 2 − √ . (4.14)
L L 16R (N − 1)
2 2
4 N2 − 1 L
√
In the limit Lr 3 N 2 − 1 we can approximate the square root by its first order
expansion, take the symmetrised trace (which to this order will only produce a factor of N
in front of the action) and regard the remaining expression as the first order expansion of
the potential

NT0 r2 L2 r 6 NT0 r4
VWB (r) = 1+ 2 1+ − √ , (4.15)
L L 16(N − 1)
2
4 N2 − 1 L
exactly as we did in the previous section.
This potential has two minima, at r = 0, corresponding to the point-like graviton, and
at
√
4 N2 − 1
r =
2
, (4.16)
L2
which should correspond to the dual giant graviton solution. Both minima have energy E =
NT0 /L. Comparing these results with the Hamiltonian and the radius of the macroscopical
derivation, we find that, in the large N limit, there is exact agreement when the condition
(3.27) is fullfilled.
5. Discussion
In this paper we have discussed a explicit matrix action which is solved by a

non-commutative 3-sphere. This matrix model arises as an action for coincident type
IIB gravitational waves. This action is constructed using T-duality from the action for
coincident type IIA gravitational waves of [18,19], and is such that the T-duality direction
appears as a special isometric direction, on which neither the background fields nor
the pull-backs depend. The presence of this compact isometric direction suggests the
representation of the fuzzy 3-sphere solution as an S 1 bundle over a fuzzy 2-sphere
base manifold. Accordingly, our solution does not show manifest SO(4) covariance, this
invariance being broken down to SU(2) × U (1). The SO(4) invariance should, still, be
present in a non-manifest way, in the same fashion than the SO(4) covariance of the
classical 3-sphere is not explicit when the S 3 is described as an S 1 Hopf fibering. This
fuzzy 3-sphere should occur as a BPS solution of the matrix model, preserving the
same supersymmetries as the point-like graviton. We have not checked out this property
explicitly, but it is expected from the agreement with the dual macroscopical description,
in which the supersymmetry properties of these spherical configurations are demonstrated
explicitly [2,3].
Odd non-commutative spheres have been previously discussed in the literature [20–22].
We would like to mention some relations between these different constructions.

In Refs. [20–22] an SO(4)-covariant matrix realisation of the condition 4i=1 Xi2 = 1 is
found, in terms of matrices acting on some vector space. As for any fuzzy sphere with more
than two dimensions, this matrix algebra contains more representations than is necessary
to describe functions on the 3-sphere, so certain projections need to be done to eliminate
the excess of degrees of freedom. The non-commutative 3-sphere that results has manifest
SO(4) covariance, but its non-commutativity cannot be removed completely in the large N
limit. On the contrary, the 3-sphere solution that we have constructed in this paper shows
only manifest SU(2) × U (1) covariance, but approaches neatly the classical S 3 in the large
N limit, where all the non-commutativity disappears. The fuzzy 3-sphere of [20–22] can be
represented in the large N limit as a fibration of a 2-sphere over an interval.6 For finite N ,
it is expected that this representation is in terms of a discrete S 1 fibration over a 2-sphere,
where both the S 2 and the S 1 are non-commutative. The reason for this is that the algebra
of functions on S 1 is infinite-dimensional so it cannot fit inside a finite matrix algebra as
that of the fuzzy S 3 . It would be interesting to develop a careful description of this S 1
fibration structure.
The fuzzy S 3 solution with manifest SO(4) covariance is expected to play a role in a
plausible description of type IIB gravitons in terms of an action with no isometry directions
(other than the direction of propagation). We should mention at this point that the reason
why we have not succeeded in constructing such an action is a technical one, namely the
impossibility of restoring the dependence on the T-duality direction in the non-Abelian
case. Should the construction of such an action be possible, an SO(4) covariant solution
as that of [20–22] would most likely be the right ansatz for the study of giant graviton
configurations.
In fact, an explicit physical realisation of the fuzzy 3-sphere discussed above is in
the context of non-BPS type IIB D0-branes [20]. The 3-sphere arises as a solution of
the action for coincident non-BPS type IIB D0-branes [27,28] in a flat background with
constant 5-form field strength. In this construction the tachyonic field of the D0-branes
plays an essential role in describing the fuzzy S 3 as a subspace of a fuzzy S 4 . This solution
cannot however be interpreted in the large N limit as a spherical D3-brane, because its
dipole moment vanishes for large N .7 Our 3-sphere solution, on the other hand, has an
interpretation as a spherical D3-brane, since its dipole, or magnetic moment, coupling to
the 5-form field strength resembles that of a spherical D3-brane, not only at finite N but
also in the large N limit.8 A dual macroscopical description in terms of a single D3-brane
with whom to compare the microscopical description in [20] is however not possible,
given that it is not known how to dissolve non-BPS D0-branes in the worldvolume of a
BPS D3-brane.9 Our solution on the contrary arises when studying the polarisation of,
BPS, gravitational waves. For this system not only the Lagrangian is defined in a better
6 We thank Sanjaye Ramgoolam for pointing this out to us and for the discussion below for finite N .
7 As discussed in [20] this could be due to the fact that the background they consider is not a consistent
supergravity background.
8 This result seems to confirm the previous observation in [20], since our solution occurs in a consistent
supergravity background. One should take into account however that we are dealing with gravitational waves and
not with non-BPS D0-branes.
9 It is likely that considering a D3, anti-D3 system would help for this purpose.
way, since there is no uncontrolled tachyon dynamics, but, moreover, a dual description
in terms of a single D3-brane exists, which can be compared in the large N limit with the
microscopical description. The agreement between the two approaches provides in fact the
strongest support to our construction.
Acknowledgements
The authors are grateful to Jos Gheerardyn, Walter Troost and especially Sanjaye
Ramgoolam for very useful discussions. The work of B.J. has been done as a post-doctoral
fellow of the F.W.O.-Vlaanderen. B.J. was also partially supported by the F.W.O.-project
G0193.00N, by the European Commission R.T.N.-program HPRN-CT-2000-00131 and
by the Belgian Federal Office for Scientific, Technical and Cultural Affairs through the
Interuniversity Attraction Pole P5/27. The work of Y.L. has been partially supported
by CICYT grant BFM2000-0357 (Spain). D.R.-G. was supported in part by a F.P.U.
Fellowship from M.E.C. (Spain). He would like to thank the Departamento de Física
Teórica at Universidad Autónoma de Madrid for its hospitality while part of this work
was done.
References
[1] J. McGreevy, L. Susskind, N. Toumbas, JHEP 0006 (2000) 008, hep-th/0003075.

[2] M.T. Grisaru, R.C. Myers, Ø. Tafjord, JHEP 0008 (2000) 040, hep-th/0008015.
[3] A. Hashimoto, S. Hirano, N. Itzhaki, JHEP 0008 (2000) 051, hep-th/0008016.
[4] S.R. Das, A. Jevicki, S.D. Mathur, Phys. Rev. D 63 (2001) 044001, hep-th/0008088.
[5] J.M. Camino, A.V. Ramallo, JHEP 0109 (2001) 012, hep-th/0107142.
[6] J.M. Camino, A.V. Ramallo, Phys. Lett. B 525 (2002) 337, hep-th/0110096.
[7] D. Berenstein, J. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
[8] K. Skenderis, M. Taylor, JHEP 0206 (2002) 025, hep-th/0204054.
[9] H. Takayanagi, T. Takayanagi, JHEP 0212 (2002) 018, hep-th/0209160.
[10] N. Kim, J.-T. Yee, Supersymmetry and branes in M-theory plane-waves, hep-th/0211029.
[11] J.-T. Yee, P. Yi, JHEP 0302 (2003) 040, hep-th/0301120.
[12] R. Emparan, Phys. Lett. B 423 (1998) 71, hep-th/9711106.
[13] W. Taylor, M. van Raamsdonk, Nucl. Phys. B 558 (1999) 63, hep-th/9904095.
[14] W. Taylor, M. van Raamsdonk, Nucl. Phys. B 573 (2000) 703, hep-th/9910052.
[15] R.C. Myers, JHEP 9912 (1999) 022, hep-th/9910053.
[16] S.R. Das, S.P. Trivedi, S. Vaidya, JHEP 0010 (2000) 037, hep-th/0008203.
[17] Y. Lozano, Phys. Rev. D 64 (2001) 106011, hep-th/0012137.
[18] B. Janssen, Y. Lozano, Nucl. Phys. B 643 (2002) 399, hep-th/0205254.
[19] B. Janssen, Y. Lozano, A microscopical description of giant gravitons, hep-th/0207199, Nucl. Phys. B, in
press.
[20] Z. Guralnik, S. Ramgoolam, JHEP 0102 (2001) 032, hep-th/0101001.
[21] S. Ramgoolam, Nucl. Phys. B 610 (2001) 461, hep-th/0105006.
[22] S. Ramgoolam, JHEP 0210 (2002) 064, hep-th/0207111.
[23] E. Bergshoeff, B. Janssen, T. Ortín, Phys. Lett. B 410 (1997) 131, hep-th/9706117.
[24] E. Bergshoeff, Y. Lozano, T. Ortín, Nucl. Phys. B 518 (1998) 363, hep-th/9712115.
[25] E. Eyras, B. Janssen, Y. Lozano, Nucl. Phys. B 531 (1998) 275, hep-th/9806169.
[26] M. Nakahara, Geometry, Topology and Physics, Institute of Physics, Bristol, 1990.
[27] B. Janssen, P. Meessen, Phys. Lett. B 526 (2002) 144, hep-th/0009025.
[28] S. Mukhi, N.V. Suryanarayana, JHEP 0011 (2000) 006, hep-th/0009101.
Erratum
Erratum to: “Dynamical rearrangement of gauge

symmetry on the orbifold S 1 /Z2 ”
[Nucl. Phys. B 657 (2003) 169–213] ✩
Naoyuki Haba a , Masatomi Harada b , Yutaka Hosotani b ,
Yoshiharu Kawamura c
a Institute of Theoretical Physics, University of Tokushima, Tokushima 770-8502, Japan
b Department of Physics, Osaka University, Toyonaka, Osaka 560-0043, Japan
c Department of Physics, Shinshu University, Matsumoto, Nagano 390-8621, Japan
Received 18 June 2003
1. There is an error in the effective potential, Eq. (7.13), in the supersymmetric SU(5)
GUT model. The correct effective potential is given by

1 1
Veff (a, b) = −2C 2(1 − Nh ) f5 (a) − f5 (a + 2β) − f5 (a − 2β)
2 2

1 1
+ f5 (b) − f5 (b + 2β) − f5 (b − 2β)
2 2
+ 2f5 (a + b) − f5 (a + b + 2β) − f5 (a + b − 2β)
+ 2f5 (a − b) − f5 (a − b + 2β) − f5 (a − b − 2β)
1 1
+ f5 (2a) − f5 2(a + β) − f5 2(a − β)
2 2
1 1
+ f5 (2b) − f5 2(b + β) − f5 2(b − β)
2 2
∞
1
= −2C 5
(1 − cos 2πnβ) 2(1 − Nh )(cos πna + cos πnb)
n
n=1

+ 4 cos πna cos πnb + cos 2πna + cos 2πnb , (1)
where Nh is the number of the sets of hyper-multiplets H + H. In Eq. (7.13) of the paper
the factor (1 − Nh ) was incorrectly written as (1 − 2Nh ), and the last three terms in the first
✩
doi of original article: 10.1016/S0550-3213(03)00142-1.
E-mail address: haba@eken.phys.nagoya-u.ac.jp (N. Haba).
doi:10.1016/S0550-3213(03)00549-2
382 N. Haba et al. / Nuclear Physics B 669 (2003) 381–382
equality were misspelled as f5 (2a) in place of f5 (2b), etc. The factor 2Nh in Eqs. (7.14)
and (7.16) should be replaced by Nh accordingly.
For Nh = 1, which is of the most phenomenological interest, the two minima of
Veff (a, b) at (a, b) = (0, 0) and (1, 1) are degenerate. For Nh 2 the global minimum
of Veff is located at (a, b) = (1, 1). The table in (7.15) should be replaced by
Nh (a, b) = (0, 0) (a, b) = (1, 1)

0 Global min. Local min.
1 Degenerate global min. Degenerate global min. (2)
2∼4 Local min. Global min.
5∼ Unstable Global min.
Table 2 should be corrected accordingly. Fig. 4 is for Nh = 2, but not for Nh = 1.

The estimate of the tunneling rate from the false vacuum to the true vacuum in the SUSY
model described in the last part of Section 8 is for Nh = 2, but not for Nh = 1. For Nh = 1
the two minima are degenerate so that once the universe is settled in the color-conserving
minimum, it stays there forever.
We stress that in the SUSY model with Nh = 1, which is most interesting from the
phenomenological viewpoint, the color-conserving vacuum is legitimate choice of the
nature.
2. In the fourth line from Eq. (4.15) in page 182, “Θ † Θ → Θ † Θ = Ω2 ΘΘ † Ω2† ”

should read “Θ † Θ → Θ † Θ = Ω2 Θ † ΘΩ2† ”.
3. In the expressions for φ (+−) and φ (−+) in Eq. (A.4), the summation over n extends
from 0 to ∞.
Nuclear Physics B 669 [FS] (2003) 385–416
Thermodynamic properties of an integrable quantum

spin ladder with boundary impurities
M.T. Batchelor a,b , X.-W. Guan a,b , A. Foerster c , A.P. Tonel c ,
H.-Q. Zhou d
a Department of Theoretical Physics, Research School of Physical Sciences and Engineering,
Australian National University, Canberra, ACT 0200, Australia
b Centre for Mathematics and its Applications, School of Mathematical Sciences,
Australian National University, Canberra, ACT 0200, Australia
c Instituto de Fisica da UFRGS, Av. Bento Goncalves, 9500, Porto Alegre 91501-970, Brazil
d Centre for Mathematical Physics, School of Physical Sciences, The University of Queensland, 4072, Australia
Received 9 May 2003; received in revised form 20 June 2003; accepted 22 July 2003
Abstract
An integrable quantum spin ladder based on the SU(4) symmetry algebra with boundary defects
is studied in the framework of boundary integrability. Five nontrivial solutions of the reflection
equations lead to different boundary impurities. In each case the energy spectrum is determined
using the quantum inverse scattering method. The thermodynamic properties are investigated by
means of the thermodynamic Bethe ansatz. In particular, the susceptibility and the magnetization of
the model in the vicinity of the critical points are derived along with differing magnetic properites
for antiferromagnetic and ferromagnetic impurity couplings at the edges. The results are applicable
to the strong coupling ladder compounds, such as Cu2 (C5 H12 N2 )2 Cl4 .
PACS: 75.10.Jm; 75.30.Kz; 75.40.Cx
Keywords: Yang–Baxter equation; Reflection equations; Thermodynamic Bethe ansatz; Impurity effects;
Quantum spin ladders
E-mail address: murrayb@wintermute.anu.edu.au (M.T. Batchelor).
doi:10.1016/j.nuclphysb.2003.07.012
386 M.T. Batchelor et al. / Nuclear Physics B 669 [FS] (2003) 385–416
1. Introduction
Research on quantum spin ladders continues to attract considerable attention from

both theoretical and experimental points of view due to their relevance to a large
number of low-dimensional materials, such as particular cuprates and organic compounds
[1–6], among others. Initially, most of the theoretical results concerning ladder systems
were obtained from the standard Heisenberg ladder. Subsequently, other models with
generalized interactions have been proposed. In this context, Nersesyan and Tsvelik [7]
introduced a spin ladder model incorporating a biquadratic spin exchange interaction
term, which, when sufficiently strong, exhibits new dimerized phases [8]. Various ladder
models have been developed by an extension of the symmetry algebra [9–13]. A special
case of the Nersesyan–Tsvelik model [7] was proposed later by Wang [12]. This model,
based on the SU(4) symmetry algebra, is exactly solvable by Bethe ansatz methods
and exhibits a spin gap in the spectrum of elementary triplet excitations, a necessary
condition for superconductivity to occur under hole doping. In addition, it was recently
observed [14] that this model can be used to describe some physical properties of different
types of two-leg ladder compounds, such as Cu2 (C5 H12 N2 )2 Cl4 [3], (C5 H12 N)2 CuBr4 [4],
(5IAP)2CuBr4 · 2H2 O [5] and KCuCl3 , TlCuCl3 [6]. In the absence of a magnetic field the
model exhibits three quantum phases, while in the presence of a strong magnetic field there
is a gapped phase in the regime H < Hc1 , a fully polarized gapped phase for H > Hc2
and a Luttinger liquid magnetic phase in the regime Hc1 < H < Hc2 . This observation
suggests that the physical properties of the ladder compounds can be accessed via the
well-established knowledge of integrable systems.
On the other hand, the effect of boundary impurities and defects also plays an important
role in quasi-one-dimensional systems. An integrable SU(4) spin ladder model with
a boundary defect has been proposed and investigated recently through Bethe ansatz
methods [15]. A generalization of this method to other integrable ladders can be found
in [16]. This model, however, is just a particular case of a more general family of exactly
solvable ladder models based on the SU(4) symmetry algebra that can be constructed from
more general types of bounday conditions. Basically, by this strategy, a set of equations
to deal with the boundaries, called reflection equations (RE) are introduced [17,18]. The
solutions of these equations [19–21], referred to as K-matrices, in turn introduce boundary
interactions into the Hamiltonian of the system, in such a manner that integrability is
preserved. The boundary interaction terms in spin ladder models may be realized by
impurity doping at the ends of the ladder. Impurity doping in a spin ladder system with
a spin gap has been performed [22]. Substantial change in macroscopic properties such as
enhancement in spin correlations and magnetic susceptibilities are observed in the low
impurity concentration region. The boundary impurity doping may change the critical
behaviour at the boundaries of the ladder systems.
The purpose of this paper is to present a complete family of integrable spin ladder
systems based on the SU(4) symmetry algebra with boundary impurities in a systematic
way. An analytic analysis of the thermodynamic properties of these models is then
performed by means of the thermodynamic Bethe ansatz (TBA) method and, in particular,
the effect of these impurities on the free energy, the susceptibility and the magnetization is
M.T. Batchelor et al. / Nuclear Physics B 669 [FS] (2003) 385–416 387
discussed. So far, the results obtained provide a clear interpretation of the impurity effects
in the low temperature regime of an integrable open spin ladder system.
The paper is organized as follows. In Section 2, we present the SU(4) solution of the
YBE and solve the corresponding RE. Furthermore, we give the explicit expressions for the
Hamiltonian with different types of boundary defects. Section 3 is devoted to the derivation
of the Bethe-ansatz solution by means of the quantum inverse scattering method (QISM).
The reader more interested in the physics of the model may choose to skip Section 3.
In Section 4, the ground state properties, the quantum phase diagram and the boundary
impurity effects are studied via the TBA. A summary and discussion of our main results is
given in Section 5.
2. The integrable spin ladder model with boundary impurities
Let us begin by introducing the integrable spin ladder model based on the SU(4)
symmetry with boundary fields,
J L
Sj · Tj + H1 + HL ,
(m) (l)
H= Hleg + J⊥ (1)
γ
j =1
where the leg part consists of Heisenberg exchange and four-spin interaction terms
1
L−1
Hleg =
+ Sj · Sj +1 + Tj · Tj +1 + 4Sj · Sj +1 · Tj · Tj +1 . (2)
4
j =1
The left (right) boundary terms H1(m) (HL(l) ) depend on arbitrary parameters U± and are
given by


 −U− S1 · T1 − 4 U− , for m = 1,
1




 −U− 2 − S1 2 − T1 + 2 U− for m = 2,
1 z 1 z 1
 ,
H1 = U− S1 · T1 − 2 − S1 2 − T1 + 4 U− , for m = 3,
(m) 1 z 1 z 1 (3)

 1 1 1

 U− S1 · T1 − 2 + S1 2 + T1 + 4 U− , for m = 4,
z z



−2U− S1z T1z + 12 U− , for m = 5,


 −U+ SL · TL − 4 U+ ,
1
for l = 1,

 1 1 1
 z z
 −U+ 2 − SL 2 − TL + 2 U+ ,
 for l = 2,
HL = U+ SL · TL − 2 − SL 2 − TL + 4 U+ , for l = 3,
(l) 1 z 1 z 1 (4)

 1 1 1

 U+ SL · TL − 2 + SL 2 + TL + 4 U+ , for l = 4,
z z



−2U+ SLz TLz + 12 U+ , for l = 5.
In the above Sj and Tj are the standard spin- 21 operators acting on site j of the upper
and lower legs, respectively, J and J⊥ are the intrachain and interchain couplings (see
Fig. 1) and L is the length of the ladder. It is worth noticing that γ is a rescaling constant
which can be used to minimize the biquadratic term such that the quantum phase of the
Fig. 1. The su(4) spin ladder with boundary impurities. U± are the boundary impurity coupling constants. J and
J⊥ are the intrachain and interchain couplings.
model (1) lies in the same Haldane spin liquid phase as that of the conventional spin ladder
(see Section 4).
Notice that the boundary terms corresponding to the first case (m = 1, l = 1) in (3) and
(4) act as Heisenberg-type rung couplings, whereas in the second case (m = 2, l = 2) they
act as a z-component spin interaction with boundary magnetic fields. In the third (m = 3,
l = 3) and fourth (m = 4, l = 4) cases, they act as a combination of Heisenberg-type rung
coupling and z-component spin interaction with boundary magnetic fields. In the last case
(m = 5, l = 5) only the z-component spin interaction terms survive. The Hamiltonian (1)
thus contains five different types of boundary rung interactions at each edge of the ladder
realizing different impurity dopings. This leads to twenty five possible choices of boundary
impurities. The rung interaction
in the bulk was, as usual, introduced by the chemical
potential terms given by −J⊥ L e
j =1 j
11 in the canonical basis e ⊗ e . The rung states split
i j
into a singlet and a triplet denoted by
1
|1 = √ |↑↓ − |↓↑ , |2 = |↑↑,
2
1
|3 = √ |↑↓ + |↓↑ , |4 = |↓↓, (5)
2
respectively. The leg interaction part of the Hamiltonian (2) does not change under the
basis transformation (5). However, the bulk rung interaction part and the boundary rung
interaction terms alter with respect to the choice of the order of singlet and triplet in the
basis (5).
In order to derive this model let us begin by recalling the SU(4) R-matrix
R12 (u) = u · I + P , (6)

where P is the permutation operator with matrix elements Pαβ,γ δ = δαδ δβγ with
α, β, γ , δ = 1, 2, 3, 4. I is the identity operator. For later use, we denote the Boltzmann
weights in the R-matrix (6) as
w1 = u + 1, w2 = u, w3 = 1. (7)
The quantum R-matrix (6) satisfies the Yang–Baxter equation (YBE)
R12 (u − v)R13 (u)R23 (v) = R23 (v)R13 (u)R12 (u − v), (8)

guaranteeing the integrability of the model with periodic BC. This R-matrix enjoys the
properties
t1 t2
R12 (u)R21 (−u) = 1 − u2 , R12 (u) = R12 (u), (9)
where superscript ta denotes the transposition in the space with index a. For other types
of boundary conditions, the YBE will still account for the integrability of the bulk part of
the model, but the boundary terms have to be chosen appropriately in order to preserve the
integrability. In particular, the left and right reflection matrices, K− and K+ , respectively,
are required to satisfy the REs [18,20]
1 2
R12 (u − v) K − (u)R21 (u + v) K − (v)
2 1
=K − (v)R12 (u + v) K − (u)R21 (u − v), (10)
t1 t2 1 2
R21 (v − u) K t+1 (u)R̃12 (−u − v) K t+2 (v)
2 1
=K t+2 (v)R̃21 (−u − v) K t+1 (u)R12t1 t2
(v − u). (11)
In the above we have introduced the object R̃ which may be determined by the relations
t2 t1 t1 t2
R̃12 (−u)R21 (u) = 1, R̃21 (−u)R12 (u) = 1, (12)
and we have used the conventional notation
1 2
X ≡ X ⊗ IV2 , X ≡ IV1 ⊗ X, (13)
where IV denotes the identity operator on V and, as usual, R21 = P · R12 · P, with P being
the permutation operator. After a lengthy calculation we find the possible solutions of the
REs for the diagonal K± -matrices (see also Ref. [19])
 
K1± (u) 0 0 0
 0 K2± (u) 0 0 
K± (u) =  . (14)
0 0 K3± (u) 0
0 0 0 K4± (u)
The solutions for K− , corresponding to the left boundary are:
Case 1
K1− (u) = u + ξ− , K2− (u) = K3− (u) = K4− (u) = −u + ξ− ; (15)
Case 2
K1− (u) = K2− (u) = K3− (u) = u + ξ− , K4− (u) = −u + ξ− ; (16)
Case 3
K1− (u) = K4− (u) = −u + ξ− , K2− (u) = K3− (u) = u + ξ− ; (17)
Case 4
K1− (u) = K2− (u) = −u + ξ− , K3− (u) = K4− (u) = u + ξ− ; (18)
Case 5
K1− (u) = K3− (u) = u + ξ− , K2− (u) = K4− (u) = −u + ξ− . (19)
On the other hand, the solutions for K+ , corresponding to the right boundary, are:
Case 1
K1+ (u) = −u + ξ+ + 3,
K2+ (u) = K3+ (u) = K4+ (u) = u − ξ+ + 1; (20)
Case 2
K1+ (u) = K2+ (u) = K3+ (u) = −u − ξ+ − 1,

K4+ (u) = u − ξ+ + 3; (21)
Case 3
K1+ (u) = K4+ (u) = u − ξ+ + 2,

K2+ (u) = K3+ (u) = −u − ξ+ − 2; (22)
Case 4
K1+ (u) = K2+ (u) = u − ξ+ + 2,

K3+ (u) = K4+ (u) = −u − ξ+ − 2; (23)
Case 5
K1+ (u) = K3+ (u) = −u − ξ+ − 2,

K2+ (u) = K4+ (u) = u − ξ+ + 2. (24)
J
In the above ξ± = γ U± are free parameters related to the left − (right +) boundary coupling
(m) (l)
U− (U+ ), respectively. The boundary pairs K− (u), K+ (u), l, m = 1, . . . , 5, lead to the
(m) (l)
boundary terms H− and H+ in (3) and (4). Mathematically, they combine to give twenty-
five possible choices of boundary impurities if we put impurities at both ends. However,
they are not correlated to each other directly, so it is enough to consider one impurity at
one end in each boundary condition so as to give five independent boundary defects. We
note that a special choice of the boundary term H (1) has already been investigated [15].
However, the second boundary impurity model proposed there does not have a counterpart
in the cases considered here.
The symmetries enjoyed by the R-matrix (6) and the REs (10) and (11) constitute the
necessary ingredients for the integrability of the model with boundary impurities, due to
the fact that the double-row transfer matrix of the system

τ (u) = tr0 K+ (u)T (u)K− (u)T −1 (−u) , (25)
commutes for different values of the spectral parameter. Here T (u) denotes the mon-
odromy matrix given by
T (u) = R0,L (u)R0,L−1 (u) · · · R0,2 (u)R0,1 (u) (26)
and T −1 its inverse. The Hamiltonian (1) associated with the quantum R-matrix (6) is
related to the double-row transfer matrix (25) by

J d L
H =− ln τ (u) − J⊥ ej11 + const. (27)
2γ du u=0 j =1
Here
(0)P
d L−1
tr0 K+ (0)R0L
ln τ (u) −1 0L
=2 Hjj +1 + K− (0)K− (0) + 2 , (28)
du u=0 tr K
0 + (0)
j =1
where the prime denotes the derivative with respect to the spectral parameter. The
J
relation (27) clearly indicates the identification U± = γ ξ± between the boundary impurity
couplings U± and the free parameters ξ± of the boundary scattering matrices. So far, we
have completed the first step towards the solution of the model with boundary impurities.
Next we proceed with the diagonalization of the transfer matrix (25) by means of the open
algebraic Bethe ansatz [25,26].
3. The algebraic Bethe-ansatz approach
3.1. First-level nesting structure
In order to find the spectrum of our Hamiltonian with boundary defects, we first need
to solve the eigenvalue problem of the transfer matrix, namely τ Φ = λΦ. As usual, the
transfer matrix (25) can be written in the form

τ (λ) = tr0 K+ (u)T̃− (u) , (29)
where T̃− (u) is the double-row monodromy matrix defined by
T̃− (u) = T (u)K− (u)T −1 (−u). (30)

One can verify that T̃− (u) also satisfies the RE (10). Following the notation used in
Refs. [23–26], we label the elements of the monodromy matrix T (u) by
 
A(u) B1 (u) B2 (u) B3 (u)
 C1 (u) D11 (u) D12 (u) D13 (u) 
T (u) =  , (31)
C2 (u) D21 (u) D22 (u) D23 (u)
C3 (u) D31 (u) D32 (u) D33 (u)
and further
 
Ā(u) B̄1 (u) B̄2 (u) B̄3 (u)
 C̄1 (u) D̄11 (u) D̄12 (u) D̄13 (u) 
T −1 (−u) =  
 C̄2 (u) D̄21 (u) D̄22 (u) D̄23 (u)  , (32)
C̄3 (u) D̄31 (u) D̄32 (u) D̄33 (u)

 
B̃(u) Ã1 (u) B̃2 (u) B̃3 (u)
 C̃1 (u) D̃11 (u) D̃12 (u) D̃13 (u) 
T̃− (u) =  
 C̃2 (u) D̃21 (u) D̃22 (u) D̃23 (u)  . (33)
C̃3 (u) D̃31 (u) D̃32 (u) D̃33 (u)

According to the first level Bethe ansatz, the eigenvectors |Φ of the transfer matrix can be
written as
i1 ...iN
|Φ = B̃i1 (u1 ) · · · B̃iN (uN )|φF(1) , (34)
where the summation is taken on the repeated indicies. The coefficients with indices
in = 1, 2, 3, n = 1, . . . , N will be determined later by the second level Bethe ansatz. The
first level pseudovacuum |φ is chosen as the standard ferromagnetic state
|φ = |0L ⊗ · · · ⊗ |0i ⊗ · · · ⊗ |01 , (35)
where |0i = (1, 0, 0, 0)ti acts as a highest-weight vector. This state corresponds to the
product of the rung singlet state in the basis (5). Different choices of the order of the
basis (5) will change the eigenvalues of the Hamiltonian which facilitates the analysis
of the ground state in different regions. From the structure of the R-matrix (6), one can
deduce that the operators Bi (u) and B̄i (u) (i = 1, 2, 3) act on the reference state as creation
operators creating particles with pseudo-momenta u and −u, respectively. The operators
Ci (u) (i = 1, . . . , 3) behave as annihilation fields. Furthermore, using an invariant version
of the Yang–Baxter algebra,
2 1 1 2
−1 −1
T (−u)R12 (2u) T (u) =T (u)R12 (2u) T (−u), (36)
we obtain, apart from an overall factor Q(u) = K1− (u)K1+ (u), the eigenvalue of the
transfer matrix acting on the reference state |φ

3
+ +
τ (u)|φ = ωA (u)Ã(u) + ωa (u)D̂aa (u) |φ, (37)
a=1
where we have introduced the transformations

w3 (2u)
D̂ij (u) = D̃ij (u) − δij Ã(u) = ωi− (u)Dii (u)D̄ii (u), (38)
w1 (2u)
with i = 1, 2, 3 and
−
Ã(u) = ωA (u)A(u)Ā(u). (39)
In the above expression,

−
ωA (u) = 1, for m = 1, . . . , 5,
 (u+2)(u+ξ+)

 (u+1/2)(u+ξ++3) , for l = 1,



 (u+2)(u+ξ+) ,
+ (u+1/2)(u+ξ++1) for l = 2,
ωA (u) =
 (u+2)(−u+ξ )
(u+1/2)(−u+ξ+ −2) , for l = 3, 4,
+




 (u+2)(u+ξ+)
(u+1/2)(u+ξ++2) , for l = 5,

 u(−u+ξ− −1)
 (u+1/2)(u+ξ−) ,
 for m = 1,


 u(u+ξ − +1)
, for m = 3,
ω1− (u) = (u+1/2)(−u+ξ −)


u
, for m = 2, 4,


u+1/2

 u(−u+ξ −1)
−
(u+1/2)(u+ξ−) , for m = 5,
 u(−u+ξ −1)


−
(u+1/2)(u+ξ−) , for m = 1,



 u(u+ξ +1)
−
, for m = 3,
ω2− (u) = (u+1/2)(−u+ξ−)
 u(u+ξ− +1)
 (u+1/2)(−u+ξ , for m = 4,

 −)

 u ,
u+1/2 for m = 2, 5,
 u(−u+ξ −1)


−
(u+1/2)(u+ξ−) , for m = 1,



 u+1/2 ,
u
for m = 3,
ω3− (u) = +1)
 u(u+ξ
 (u+1/2)(−u+ξ−) , for m = 4,
−



 u(−u+ξ− −1) , for m = 2, 5,
(u+1/2)(u+ξ−)
with ω1+ (u) = K2+ (u), ω2+ (u) = K3+ (u), ω3+ (u) = K4+ (u) for l = 1, . . . , 5. In addition,
A(u)Ā(u)|0 = w1 (u)2 |0, Dii (u)D̄ii (u)|0 = w2 (u)2 |0.
We note that the operators B̃i (u), i = 1, 2, 3, constitute a three-component vector with both
positive and negative pseudo-momenta still playing the role of the creation fields acting on
the pseudovacuum state. In order to make further progress we return to the RE (10) and
derive the commutation relations
(u2 − u1 + 1)(u1 + u2 )
Ã(u1 )B̃a (u2 ) = B̃a (u2 )Ã(u1 )
(u2 − u1 )(u1 + u2 + 1)
(u1 + u2 )
− B̃a (u1 )Ã(u2 )
(u2 − u1 )(u1 + u2 + 1)
3
1 1
− B̃b (u1 )D̂ba (u2 ) + δab B̃b (u1 )A(u2 ) ,
u1 + u2 + 1 2u2 + 1
b=1
(40)
(u1 − u2 + 1)(u1 + u2 + 2)
D̂bd (u1 )B̃c (u2 ) =
(u1 − u2 )(u1 + u2 + 1)

× r (1) (u1 + u2 + 1)eb
gh r (u1 − u2 )cd B̃e (u2 )D̂gi (u1 )
(1) ih
2(u1 + 1) gb
− r (1) (2u1 + 1)id B̃g (u1 )D̂ic (u2 )
(2u1 + 1)(u1 − u2 )
4(u1 + 1)u2
+
(2u1 + 1)(2u2 + 1)(u1 + u2 + 1)
gb
× r (1)(2u1 + 1)cd B̃g (u1 )Ã(u2 ), (41)
between the diagonal fields and the creation fields. The summation convention is implied
for repeated indices. The matrix r (1) , which satisfies the Yang–Baxter equation, takes the
form
1
aa
raa = 1, a = 1, 2, 3, ab
rab = , a = b = 1, 2, 3,
u+1
u
ab
rba = , a = b = 1, 2, 3. (42)
u+1
We notice that the first term in the rhs of each of the commutation relations (40), (41)
contribute to the eigenvalues of the transfer matrix, which should be analytic functions of
the spectral parameter u. Consequently, the residues at singular points must vanish. This
yields the Bethe ansatz equations, which in turn assure the cancellation of the unwanted
terms in the eigenvalues of the transfer matrix after the whole nesting procedure. For
convenience, we make a shift in the spectral parameters, u = v − 1/2, ui = vi − 1/2,
such that the eigenvalue of the transfer matrix (25) can be obtained as
τ (v)|Φ

= Λ v, {vi } |Φ

N
(v − vi − 1)(v + vi − 1)
= WA+ (v − 1/2)WA− (v − 1/2) |Φ
(v − vi )(v + vi )
i=1

N
(v − vi + 1)(v + vi + 1)
+ Wa+ (v − 1/2)Wa− (v − 1/2) Λ(1) v, {vi } |Φ,
(v − vi )(v + vi )
i=1
(43)
provided that
WA+ (vi − 1/2)WA− (vi − 1/2)(2vi − 1)
W1+ (vi − 1/2)W1− (vi − 1/2)(2vi + 1)

N
(vi − vl + 1)(vi + vl + 1)

= Λ(1) v, {vi } . (44)
(vi − vl − 1)(vi + vl − 1)
l=1 v=vi
l=i
Here a = 1, 2, 3, and we appropriately choose

WA− (u) = 1, for m = 1, . . . , 5, (45)
 (u+2)(u+ξ+)

 , for l = 1,
 (u+1/2)(u+ξ++3)


 (u+2)(u+ξ )
+
(u+1/2)(u+ξ++1), for l = 2,
WA+ (u) = (46)
 (u+2)(−u+ξ+ )


 (u+1/2)(−u+ξ+ −2) , for l = 3, 4,

 (u+2)(u+ξ+)
(u+1/2)(u+ξ++2) , for l = 5,
u−ξ +1
+
, for l = 1,
Wa+ (u) = −u−ξ+ −3 (47)
1, for l = 2, . . . , 5,
u(−u+ξ −1)
−
, for m = 1,
Wa− (u) = (u+1/2)(u+ξ −)
(48)
u
u+1/2 , for m = 2, . . . , 5.
Λ(1) (v, {vi }) is the eigenvalue of the second level transfer matrix τ (1) related to an SU(3)
invariant open chain, i.e.,
(1) (1)
τ (1) = Tr0 K+ (v)T (1) (v)K− (v)T̄ (1) (v), (49)
where
(1) e g
T (1) v, {vi } = r12 (v + v1 )he11ag1 · · · r12
(1)
(v + vN )hNN gN−1
N
, (50)
(1)
T̄ (1) v, {vi } = r21 (v − vN )elN−1
N hN (1) e1 h1
iN · · · r21 (v − v1 )ai1 . (51)
(1)
Here we have used the standard notation r12 (v) = P · r (1) (v) where P is the standard
permutation operator, which can be represented by a 32 × 32 matrix, i.e., pαβ,γ δ =
i1 ...in
δαδ δβγ . It can be seen that the coefficients F(1) act as the multi-particle vectors for the
inhomogeneous transfer matrix (49). We remark that the coefficients W given in (45)–(48)
(1)
are chosen in order to match the choice of the transfer matrix (49) with the nested K± -
matrices,
 
K1(1)
± (v) 0 0
K± (1)
(v) =  0
(1)
K2± (v) 0 . (52)
(1)
0 0 K3± (v)
Now corresponding to the first solution (20), we have
(1) (1) (1)
K1− (v) = K2− (v) = K3− (v) = 1, (53)
(1) (1) (1)
K1+ (v) = K2+ (v) = K3+ (v) = 1. (54)
And to the second solution (21):
(1) (1) (1) −v + ξ− − 1/2
K1− (v) = K2− (v) = 1, K3− (v) = , (55)
v + ξ− − 1/2
v − ξ+ + 5/2
K1(1) (1)
+ (v) = K2+ (v) = 1, K3(1)
+ (v) = . (56)
−v − ξ+ − 1/2
And to the third solution (22):
v + ξ− + 1/2
K1(1) (1)
− (v) = K2− (v) = , K3(1)
− (v) = 1, (57)
−v + ξ− + 1/2
(1) (1) −v − ξ+ − 3/2 (1)

K1+ (v) = K2+ (v) = , K3+ (v) = 1. (58)
v − ξ+ + 3/2
And to the fourth solution (23):
(1) (1) (1) v + ξ− + 1/2

K1− (v) = 1, K2− (v) = K3− (v) = , (59)
−v + ξ− + 1/2
(1) (1) (1) −v − ξ+ − 3/2
K1+ (v) = 1, K2+ (v) = K3+ (v) = . (60)
v − ξ+ + 3/2
And to the last solution (24):
(1) (1) (1) −v + ξ− − 1/2

K2− (v) = 1, K1− (v) = K3− (v) = , (61)
v + ξ− − 1/2
v − ξ+ + 3/2
K2(1)
+ (v) = 1, K1(1) (1)
+ (v) = K3+ (v) = . (62)
−v − ξ+ − 3/2
We can show that the reflection matrices (52) with the entries (53)–(62) do satisfy the RE
(1) 1 (1) (1) 2 (1)

r12 (u − v) K − (u)r21 (u + v) K − (v)
2 (1) (1) 1 (1) (1)
=K − (v)r12 (u + v) K − (u)r21 (u − v), (63)
(1) t1 t2 1 2
r21 (v − u) K (1) t1 (1) (1) t2
+ (u)r̃12 (−u − v) K + (v)
2 (1) t (1) 1 (1) t (1) t1 t2
=K + 2
(v)r̃21 (−u − v) K + 1
(u)r12 (v − u). (64)
3.2. Second-level Bethe ansatz
In order to proceed in the nested algebraic Bethe ansatz, we have to repeat the whole
procedure presented above for the internal block of the monodromy matrix. Similarly, from
the RE (63), we can derive commutation relations,
Ã(1) (v1 )B̃a(1) (v2 )

(v1 − v2 − 1)(v1 + v2 ) (1)
= B̃ (v2 )A˜(1) (v1 )
(v1 − v2 )(v1 + v2 + 1) a
(v1 + v2 )
+ B̃ (1) (v1 )Ã(1) (v2 )
(v1 − v2 )(v1 + v2 + 1) a
2
1 (1) (1) 1 (1)
− B̃b (v1 )D̂ba (v2 ) + δab (1)
B̃ (v1 )A (v2 ) , (65)
v1 + v2 + 1 2v2 + 1 b
b=1
(1)
D̂bd (v1 )B̃c(1) (v2 )
(v1 − v2 + 1)(v1 + v2 + 2)
=
(v1 + v2 + 1)(v1 − v2 )

× r (2) (v1 + v2 + 1)eb
gh r (v1 − v2 )cd B̃e (v2 )D̂gi (v1 )
(2) ih (1)
2(v1 + 1) gb (1)
− r (2)(2v1 + 1)id B̃g(1) (u1 )D̂ic (v2 )
(2v1 + 1)(v1 − v2 )
4(v1 + 1)v2 gb
+ r (1) (2v1 + 1)cd B̃g(1) (v1 )Ã(1)(v2 ), (66)
(2v1 + 1)(2v2 + 1)(v1 + v2 + 1)
between the diagonal and the creation fields. Where again the summation convention is
implied for the repeated indices. The matrix r (2) (v) is nothing but the SU(2) invariant
R-matrix, i.e.,
1
aa
raa = 1, a = 1, 2, ab
rab = , a = b = 1, 2,
v+1
v
ab
rba = , a = b = 1, 2. (67)
v+1
If we define the second level Bethe ansatz as
(1)
Ψ = B̃l(1) (µ1 ) · · · B̃l(1) (µM )|0F(2) l1 ...lM , (68)
1 M
and make the rescalings µl → µl − 1/2 and v = ṽ − 1/2 we obtain from the commutation
relations (65) and (66) the eigenvalue Λ(1) (v, {vi }{µj }), i.e.,

τ (1) (ṽ)Ψ (1) = Λ(1) ṽ, {vi }{µj } Ψ (1)

(1) (1)
M
(ṽ − µl − 1)(ṽ + µl − 1)
= W+A (ṽ − 1/2)W−A (ṽ − 1/2)
(ṽ − µl )(ṽ + µl )
l=1
(1) (1)

N
(ṽ − vi − 1/2)(ṽ + vi − 1/2)
+ W+a (ṽ − 1/2)W−a (ṽ − 1/2)
(ṽ − vi + 1/2)(ṽ + vi + 1/2)
i=1

M
(ṽ − µl + 1)(ṽ + µl + 1) (2) (1)
× Λ ṽ, {vi }, {µl } Φ ,
(ṽ − µl )(ṽ + µl )
l=1
(69)
provided that
(1) (1)
W+A (µl − 1/2)W−A (µl − 1/2)(2µl − 1)
(1) (1)
W+1 (µl − 1/2)W−1 (µl − 1/2)(2µl + 1)

N
(µl − vi − 1/2)(µl + vi − 1/2)
=
(µl − vi + 1/2)(µl + vi + 1/2)
i=1

M
(µl − µi + 1)(µl + µi + 1)
× Λ ṽ, {vi }, {µl }
(2)
. (70)
(µl − µi − 1)(µl + µi − 1)
i=1 ṽ=µl
i=l
Here a = 1, 2, and
 v+ξ +1/2

 −v+ξ− +1/2 , for m = 3,
−
(1)
W−A (v) = 1, for m = 1, 2, 4, (71)

 −v+ξ− −1/2 , for m = 5,
v+ξ− −1/2
 v+3/2

 v+1/2 , for l = 1,



 (v+3/2)(v+ξ+−1/2)

 , for l = 2,
 (v+1/2)(v+ξ++1/2)
(1) (v+3/2)(v+ξ +1/2)
W+A (v) = (v+1/2)(−v+ξ++−3/2) , for l = 3, (72)



 (v+3/2)(−v+ξ++1/2) ,

 (v+1/2)(−v+ξ+−3/2) for l = 4,


 (v+3/2)(−v+ξ+−1/2)
(v+1/2)(v+ξ++3/2) , for l = 5,
 v

 , for m = 1,
 v+1/2 v


 for m = 2,
 (v+1/2)(v+ξ−−1/2) ,

(1)
W−1 (1)
(v) = W−2
v
(v) = (v+1/2)(−v+ξ−+1/2) , for m = 3, (73)



 v(v+ξ− +3/2)
, for m = 4,

 (v+1/2)(−v+ξ−+1/2)

 v
(v+1/2)(v+ξ−−1/2) , for m = 5,

 1, for l = 1,



 1
for l = 2,

 v+ξ+ +1/2 ,

(1) (1)
W+1 (v) = W+2 (v) =
1
−v+ξ+ −3/2 , for l = 3, (74)

 v+ξ+ +3/2

 for l = 4,

 −v+ξ+ −3/2 ,

 1
,
v+ξ+ +3/2 for l = 5.
Now Λ(2) (ṽ, {vi }, {µl }) is the eigenvalue of the third level transfer matrix τ (2) related to
an SU(2) invariant open chain, i.e.,
(2) (2)
τ (2) = Tr0 K+ (ṽ)T (2) (ṽ)K− (ṽ)T̄ (2) (ṽ), (75)
where
(2) (2) e g
T (2) ṽ, {vi }, {µl } = r12 (ṽ + µ1 )he11ag1 · · · r12 (ṽ + µM )hMM gM−1
M
, (76)
(2) (2)
T̄ (1) ṽ, {vi }, {µl } = r21 (ṽ − µM )elM−1
M hM e1 h1
iM · · · r21 (ṽ − µ1 )ai1 . (77)
3.3. Third-level Bethe ansatz
It has been shown so far that the eigenvalue problem of the transfer matrix is reduced to
the diagonalization of the isotropic Heisenberg model with K-matrices


 ṽ + ξ− − 1 0

 , for m = 2,

 0 −ṽ + ξ− − 1


(2) ṽ + ξ− 0
K− (ṽ) = , for m = 3, 5, (78)

 0 − ṽ + ξ−



 1 0

 , for m = 1, 4,
0 1


 ṽ + ξ+ 0

 , for l = 2,

 0 −ṽ + ξ+


(2) ṽ + ξ+ + 1 0
K+ (ṽ) = , for l = 3, 5, (79)

 0 − ṽ + ξ+ − 1



 1 0

 , for l = 1, 4.
0 1
Following the derivation in [18], we immediately obtain the eigenvalues of the nested
transfer matrix (75), given by
l1 ...lM l1 ...lM
τ (2) (ṽ)F (1) = Λ(2) v, {vi }, {µl }, {wq } F (1)

(2) (2)
Q
(ṽ − wl − 1)(ṽ + wl )
= W+1 (ṽ)W−1 (ṽ)
(ṽ − wl )(ṽ + wl + 1)
l=1
(2) (2)

M
(ṽ − µl )(ṽ + µl )
+ W+2 (ṽ)W−2 (ṽ)
(ṽ − µl + 1)(ṽ + µl + 1)
l=1

Q
(ṽ − wl + 1)(ṽ + wl + 2) l1 ...lM
× F (1) , (80)
(ṽ − wl )(ṽ + wl + 1)
l=1
provided that
(2)
W+1 (2)
(wl )W−1 (wl )wl
M
(wl − µj )(wl + µj )
=
(2)
W+2 (2)
(wl )W−2 (wl )(wl + 1) (wl − µj + 1)(wl + µj + 1)
j =1

Q
(wl − wm + 1)(wl + wm + 2)
× . (81)
(wl − wm − 1)(wl + wm )
m=1
m=l
Here

ṽ + ξ− − 1, for m = 2,
(2)
W−1 (ṽ) = ṽ + ξ− , for m = 3, 5, (82)
1, for m = 1, 4,
 (ṽ+1)(ṽ+ξ −1)


+
, for l = 2,
 2ṽ+1
(2) (ṽ+1)(ṽ+ξ+ )
W+1 (ṽ) = , for l = 3, 5, (83)


2ṽ+1
 ṽ+1 , for l = 1, 4,
ṽ+1/2
 −ṽ(ṽ−ξ− +2)

 , for m = 2,
 ṽ+1/2
(2) −ṽ(ṽ−ξ− +1)
W−2 (ṽ) = , for m = 3, 5, (84)


ṽ+1/2
 ṽ
ṽ+1/2 , for m = 1, 4,

−ṽ + ξ+ − 2, for l = 2,
(2)
W+2 (ṽ) = −ṽ + ξ+ − 1, for l = 3, 5, (85)
1, for l = 1, 4.
The eigenvalues (69) and (80) as well as the constraints (70), (81) on the rapidities µl and
wl have paved the way for the complete diagonalization of the transfer matrix (25). Making
a further shift on the rapidities, wl → wl − 1/2, wm → wm − 1/2 and ṽ = v + 1/2, the
eigenvalues of the transfer matrix (25) are given by

Λ v, {vi }, {µl }, {wj }

1 1
= K1− v − K1+ v −
2 2
N
+ 1 − 1 1 2L (v − vi − 1)(v + vi − 1)
× WA v − WA v − v+
2 2 2 (v − vi )(v + vi )
i=1

+ 1 − 1 (1) (1) 1 2L
+ W1 v − W1 v − W+A (v)W−A (v) v −
2 2 2
N
(v − vi + 1)(v + vi + 1) (v − ul − 12 )(v + ul − 12 )
M
×
(v − vi )(v + vi ) (v − ul + 12 )(v + ul + 12 )
i=1 l=1

+ 1 − 1 (1) (1) (2) 1
+ W2 v − W2 v − W+1 (v)W−1 (v)W+1 v +
2 2 2
2L
(2) 1 1
× W−1 v + v−
2 2

M
(v − µl + 32 )(v + µl + 32 )
Q
(v − wj )(v + wj )
×
(v − µl + 1
+ µl +
2 )(v
(v − wj + 1)(v + wl + 1)
1
2 ) j =1
l=1

+ 1 − 1 (1) (1) (2) 1
+ W3 v − W3 v − W+2 (v)W−2 (v)W+2 v +
2 2 2
2L
(2) 1 1
× W−2 v + v−
2 2

(v − wj + 2)(v + wj + 2)
Q
× . (86)
(v − wj + 1)(v + wl + 1)
j =1

The three rapidities vi , µj , wk of flavor waves satisfy the Bethe ansatz equations
(vi + 12 )2L
ζ (vi , ξ+ )ζ (vi , ξ− )
(vi − 12 )2L

N
(vi − vl + 1)(vi + vl + 1) (vi − µl − 12 )(vi + µl − 12 )
M
= , (87)
(vi − vl − 1)(vi + vl − 1) (v − µl + 12 )(vi + µl + 12 )
l=1 l=1 i
l=i

N
(µj − vi + 12 )(µj + vi + 12 )
η(µj , ξ+ )η(µj , ξ− )
i=1
(µj − vi − 12 )(µj + vi − 12 )
M
(µj − µi + 1)(µj + µi + 1) (µj − wl − 12 )(µj + wl − 12 )
Q
= , (88)
(µj − µi − 1)(µj + µi − 1) (µj − wl + 12 )(µj + wl + 12 )
i=1 l=1
i=j

M
(wk − µl + 12 )(wk + µl + 12 )
Ω(wk , ξ+ )Ω(wk , ξ− )
l=1
(wk − µl − 12 )(wk + µl − 12 )

Q
(wk − wl + 1)(wk + wl + 1)
= , (89)
(wk − wl − 1)(wk + wl − 1)
l=1
l=k
for i = 1, . . . , N , j = 1, . . . , M and k = 1, . . . , Q, respectively. Here, we have introduced

the notation

 v +ξ − 1
 i ± 21 ,
 for l = 1, m = 1,

 vi −ξ± + 2


 vi −ξ± − 12
, for l = 3, m = 3,
ζ (vi , ξ± ) = vi +ξ± + 12 (90)



 1, for l = 2, 4, m = 2, 4,



 vi +ξ± − 21 ,
1
for l = 5, m = 5,
vi −ξ± + 2
 1, for l = 1, 2, 3, m = 1, 2, 3,

 µ +ξ
 ±− 2
1
η(µj , ξ± ) =
j
, for l = 4, m = 4, (91)
µj −ξ± + 12


 µj −ξ± , for l = 5, m = 5,
µj +ξ±

 wk +ξ± − 32

 , for l = 2, m = 2,
 wk −ξ± + 32
Ω(wk , ξ± ) = w +ξ − 1
± 2 (92)


k
1 , for l = 3, 5, m = 3, 5,

 wk −ξ± + 2
1, for l = 1, 4, m = 1, 4.
These boundary factors coupled to the three degrees of freedom will result in a rich physical
scenario. From (27) and (86), we finally obtain the eigenvalues of the Hamiltonian (1) as
 J
N J
 U U
 2+ + 2− + γ − J⊥ L + i=1 γ v 2 1− 1 + J⊥ , for l = m = 1, 2, 5,
E=

i 4

 J
 − U2+ − U2− + γ − J⊥ L + N
J 1
i=1 γ 2 1 + J⊥ , for l = m = 3, 4.
vi − 4
(93)
4. Boundary impurity effects
Having diagonalised the Hamiltonian (1) by means of the algebraic Bethe ansatz, the
next step is to derive the thermodynamic Bethe ansatz equations.
4.1. Derivation of TBA
For later convenience in the analysis of the Bethe ansatz equations, we make the change
of variables: vi → −ivi , µl → −iµl , wk → −iwk and some rescalings in the boundary
parameters ξ± . The Bethe ansatz equations are now
N
vi − rvl − i vi − rµl +
M i
(vi − 2i )2L
ζ (vi , β± ) 2
= , (94)
r=±
vi − rvl + i vi − rµl − i
2 (vi + 2i )2L
l=1 l=1
l=i
M
µj − rµi − i µj − rwl +
Q i N
µj − rvi + i
η(µj , β± ) 2 2
= 1, (95)
µ − rµi + i
r=± i=1 j µ − rwl − i
µj − rvi − i
l=1 j 2 i=1 2
i=j

Q
wk − rwl − i wk − rµl +
M i
Ω(wk , β± ) 2
= 1, (96)
r=± l=1
wk − rwl + i wk − rµl − i
2
l=1
l=k
where
v +iβ
i ±
vi −iβ± , for l = 1, 3, 5, m = 1, 3, 5,
ζ (vi , β± ) = (97)
1, for l = 2, 4, m = 2, 4,
 1, for l = 1, 2, 3, m = 1, 2, 3,


 µj +iβ± , for l = 4, m = 4,
η(µj , β± ) = µj −iβ± (98)


 µj −i(β± + 2 ) , for l = 5, m = 5,
1
µj +i(β± + 12 )
 w +iβ
 k ±
 wk −iβ± , for l = 2, 5, m = 2, 5,
Ω(wk , β± ) = wk −i(β± +1) , for l = 3, m = 3, (99)

 wk +i(β± +1)
1, for l = 1, 4, m = 1, 4.
The shifts in the parameters ξ± are given by

1
β± = ξ± − , for l = m = 1, 4, 5,
2
3
β± = ξ± − , for l = m = 2,
2
1
β± = −ξ± − , for l = m = 3. (100)
2
Correspondingly, the energy spectrum is given by
N

J 1
E= − + J⊥ . (101)
γ vi2 + 1
4
i=1
Here we have dropped some constants appearing in Eq. (93), which will be used in deriving
one point correlation functions later.
From the above Bethe ansatz equations (94)–(96), it is found that in the cases l = m =
1, 3, 5, the solutions vl = ±iβ− and vr = ±iβ+ form two boundary bound sates in the
charge rapidity when β± are negative. Nevertheless, in the case l = m = 5, besides the
charge boundary bound states, the boundary bound states exist also in the spin rapidites,
i.e.,

±i β− + 12 , ±iβ− ,
µ= w =
±i β+ + 12 , ±iβ+ .
No boundary bound state exists in the remaining cases. We observe that when J⊥ >
2J
γ (1 − cos k), the reference state becomes the true ground state, i.e., the ground state
is given by a product of the singlet rung states. The minimal gap can be easily calculated
and is given by
2J
∆ = J⊥ − (1 − cos k), (102)
γ
where k = π/[1 + 1 1
4L ( β+ + 1
β− )]. It is obvious that gap remains almost unchanged in the
4J
thermodynamic limit and is almost the same as ∆ = J⊥ − γ in the periodic case because

4J
L β1± . In the regime − 12 < β± < − 12 1 − γ J⊥ , the boundary bound states are stable.
Otherwise, in the remaining regime, they become excited states. In the limit J⊥ → ∞,
all the boundary bound states are excitations. We shall see that the boundary bound states
radically affect the edge ground state properties. For Jc− = − γ ( √π −ln 3) < J⊥ < γ , the
J 4J
3
ground state consists of three branches of Luttinger liquids associated with the rapidities
v, µ and w. Here Jc− is the critical transition point from the SU(3) phase into the SU(4)
phase in the absence of a magnetic field. The triplet states can exist in the ground state.
This corresponds to a continuum of massless excitations.
The thermodynamics of the boundary fields can be derived from the Bethe ansatz
equations (94)–(96). We now focus on the analysis of the Bethe ansatz equations. As usual,
we define the functions

x + i n2
en (x) = , θn (x) = i ln en (x),
x − i n2
1 n 1 d
an (x) = ≡ θn (x), (103)
2π x 2 + n2 2π dx
4
in terms of which the Bethe ansatz equations (94)–(96) become
2L
M
N
ζ (vi , β± ) e1 (vi ) e1 (vi − rµl ) = e2 (vi − rvl ), (104)
r=± l=1 r=± l=1
l=i

N
M
Q
η(µj , β± ) e1 (µj − rvl ) = e2 (µj − rµl ) e−1 (µj − rwl ), (105)
r=± l=1 r=± l=1 l=1
l=j

M
Q
Ω(wk , β± ) e1 (wk − rµl ) = e2 (wk − rwl ). (106)
r=± l=1 r=± l=1
l=k
In order to study the thermodynamics of the model with boundary impurities we begin
by adopting the string hypothesis [27–29]. If we define v−j = −vj , µ−l = −µl and
w−k = wk , the Bethe ansatz equations (94)–(96) admit the string solutions
1
vαn1 j = vαn1 + i (n + 1 − 2j ),
2
1
µα2 j = µα2 + i (n + 1 − 2j ),
n n
2
1
wα3 j = wα3 + i (n + 1 − 2j ),
n n
2
in thermodynamic limit, where j = 1, . . . , n, αa = 1, . . . , Nn(a) and vαn1 , µnα2 and wαn3 are
the positions of the center of the strings. The number of n-strings Nn(a) satisfy the relation

P (a) = n nNn(a) . By taking the thermodynamic limit, the Bethe ansatz equations become
1 (1)
ρn(1)h = an + ρbn − Anm ∗ ρm (1)
+ anm ∗ ρm (2)
, (107)
2L m m
1 (2)
ρn(2)h = ρbn − Anm ∗ ρm
(2)
+ anm ∗ (ρm
(1)
+ ρm (3)
), (108)
2L m m
1 (3)
ρn(3)h = ρbn − Anm ∗ ρm
(3)
+ anm ∗ ρm
(2)
, (109)
2L m m
where the symbol ∗ denotes the usual convolution. Here ρn(a) (v), a = 1, 2, 3 are the
(a)h
densities of roots of the three flavors, ρn (v), a = 1, 2, 3 are the densities of holes of
(i)
the three flavors and ρbn , i = 1, 2, 3 are the contributions from boundary fields associated
with different rapidities. These boundary phase factors are given by

n
(1)
ρbn = ± l=1 an+2β± +1−2l (λ) + an2 (λ), for l = 1, 3, 5, m = 1, 3, 5, (110)
an2 (λ), for l = 2, 4, m = 2, 4,

n2 (λ) − an1 (λ),
 a
for l = 1, 2, 3, m = 1, 2, 3,

n
(2)
ρbn = ±
a n+2β ± +1−2l (λ) + a n2 (λ) − a n1 (λ), for l = 4, m = 4,

l=1
n
− ± l=1 an+2β± +2−2l (λ) + an2 (λ) − an1 (λ), for l = 5, m = 5,
(111)


n

±
l=1 a n+2β ± +1−2l (λ) + a n2 (λ), for l = 2, 5, m = 2, 5,
(3)
ρbn = − ± nl=1 an+2β± +3−2l (λ) + an2 (λ), for l = 3, m = 3, (112)

an2 (λ), for l = 1, 4, m = 1, 4.
In addition

Min(n,m)−1
Anm (λ) = δ(λ)δnm + (1 − δnm )a|n−m| (λ) + an+m (λ) + 2 a|n−m|+2l (λ),
l=1

Min(n,m)
anm (λ) = an+m+1−2l (λ).
l=1
We emphasize that the boundary potentials enter in the expression for the ground state
(a)
energy implicitly via ρb (v) in the above equations, with contributions to the densities
of the roots at the order of 1/L. In order to find the equilibrium state of the system at
fixed temperature T and external magnetic field H ( 0), we minimize the free energy
F = E − T S − H S z with respect to the densities to obtain the TBA in the form
 (1) 
ln 1 + ηn
 (2) 
 ln 1 + ηn 
(3)
ln 1 + ηn
 ln1 + 1 



− m anm
(1)
m Anm 0 ηm
Gn

 1 

= +  − m anm Anm − m anm  ∗  
ln 1 + (2)
ηm 
. (113)
T
m

0 − m anm m Anm ln 1 + (3)
1
ηm
The driving matrix Gn depends on the choice of the reference state. Explicitly, for J⊥ < 0,
J
G = column(− γ 2πan + nH, nH, −n(J⊥ + H )), giving the free energy
∞
∞ (1)
F (T , H ) Hn (λ)
= −H − T an (λ) ln 1 + e− T dλ
2L
−∞ n=1
∞
∞
T
3 (a)
Hn (λ)
− (a)
ρbn ln 1 + e− T dλ. (114)
2L
a=1 −∞ n=1
J
On the other hand, for J⊥ > 0, G = colum(− γ 2πan + n(J⊥ − H ), nH, nH ) and the free
energy is given by
∞
∞ (1)
F (T , H ) Hn (λ)
= −T an (λ) ln 1 + e− T dλ
2L
−∞ n=1
∞
∞
T
3 (a)
Hn (λ)
ρbn ln 1 + e− T
(a)
− dλ. (115)
2L
a=1 −∞ n=1
Here ηn(l) (λ) = ρ (l)h (λ)/ρ (l) (λ) ≡ exp(Hn(l)(λ)/T ), l = 1, 2, 3, with the dressed energy Hn(l)
playing the role of an excitation energy measured from the Fermi level.
Using the relations
(a) (a)
(a0 + a2 ) ∗ ln ηn(a) = a1 ∗ ln 1 + ηn+1 + ln 1 + ηn−1

1 1
− ln 1 + (a−1) − ln 1 + (a+1) , (116)
ηn ηn
another form of the TBA is given by
H
(a) ∞
(a)
Hm+1
H1 = g1 + T a2 ∗ ln 1 + e− T + T (a0 + a2 ) am ∗ ln 1 + e− T
(a) (a) 1
m=1
∞
(a−1) (a+1)
Hm Hm
−T am ∗ ln 1 + e− T + ln 1 + e− T , (117)
m=1
(a)
Hn−1
Hn(a) = gn(a) + T a1 ∗ ln 1 + e T
(a)
Hn
∞
(a)
Hm
+ T a2 ∗ ln 1 + e− T + T (a0 + a2 ) am−n ∗ ln 1 + e− T
mn
∞
(a−1) (a+1)
H H
− mT − mT
−T am−n+1 ∗ ln 1 + e + ln 1 + e , (118)
mn
for n 2. In the above a = 1, 2, 3 and Hn(0) (λ) = Hn(4) (λ) = 0 is assumed. The driving terms
are given explicitly by
J
g1(1) = − γ 2πa1 + H, (1)
gn = H,
(2)
(2)
g1 = H, gn = H, for J⊥ < 0, (119)
g1(3) = −(J⊥ + H ), gn(3) = −(J⊥ + H ),
(1) J
g1 = − γ 2πa1 + J⊥ − H, gn(1) = J⊥ − H,
(2)
g1 = H,
(2)
gn = H, for J⊥ 0. (120)
(3)
(3)
g1 = H, gn = H.
4.2. Boundary bound states and impurity effects
In the low temperature limit, the states with positive dressed energy are empty. The
(a)
zeros of the dressed energies define the fermi energies. We decompose Hn into its positive
(a) (a)+ (a)−
and negative parts, Hn = Hn + Hn . An analysis of Eqs. (117) and (118) in the limit
T → 0 reveals that for the ground state, the roots are all real corresponding to n = 1. All
(a)+
dressed energies Hn with n 2 correspond to excitations. Thus the TBA for the ground
state is, for J⊥ < 0,
(1)
H (1) = g1 − a2 ∗ H (1)− + a1 ∗ H (2)− ,

H (2) = H − a2 ∗ H (2)− + a1 ∗ H (1)− + H (3)− ,
H (3) = −H − J⊥ − a2 ∗ H (3)− + a1 ∗ H (2)− , (121)
and for J⊥ 0,
(1)
H (1) = g1 − a2 ∗ H (1)− + a1 ∗ H (2)− ,

H (2) = H − a2 ∗ H (2)− + a1 ∗ H (1)− + H (3)− ,
H (3) = H − a2 ∗ H (3)− + a1 ∗ H (2)− . (122)
In this case, the free energy is given by
!∞
F (0, H ) −H + −∞ a1 (λ)H1(1)− (λ) dλ + 1
2L fb , for J⊥ < 0,
= !∞ (1)−
(123)
2L
−∞ a1 (λ)H1 (λ) dλ + 2L
1
fb , for J⊥ 0,
where
3 Qa
(a)
fb = ρb1 (λ)H1(a)− (λ) + θ (β± + βc )Ebs , (124)
a=1−Q
a

4J
and θ (x) denotes a step-like function. Define βc = − 12 1 − γ J⊥ , then in the interval
− 12 < β < βc , θ (β) = 1, else θ (β) = 0. In the above Ebs denotes the boundary bound
state energy, given by


 J
 ± − γ −β 21+ 1 + J⊥ , for J⊥ 0,
± 4
Ebs =
J (125)

 ± −γ 1
2 1 ,
−β± + 4
for J⊥ < 0.
It is worth noticing that if β± < βc we should take the boundary bound states into account
(a)
in the boundary contributions ρb1 for the cases l = m = 1, 3, 5. The TBA (121) and
(122) provide a clear physical picture of the ground-state and in turn the thermodynamic
properties, such as the free energy, the magnetization, the susceptibility, etc. The boundary
impurities coupled to the three rapidities affect the low temperature physics at the edges in
various different ways, which we now explore.
From the TBA (122), we notice that if J⊥ > Jc+ =

4J
γ the triplet excitations are
4J
massive with energy gap ∆ = J⊥ − The rescaling γ = 4 was fixed [14] for strong
γ .
coupling compounds, e.g., Cu2 (C5 H12 N2 )2 Cl4 [3], (C5 H12 N)2 CuBr4 [4], etc. Here Jc+
is the critical point at which the quantum phase transition from the three branches of
Luttinger liquid to the dimerized U (1) phase occurs. If J⊥ > Jc+ , we can show that in
the presence of a strong magnetic field two of the triplet states (|3 and |4 in (5)) in the
bulk part will never be involved in the ground state. However, at the boundaries this is
not always true due to the presence of the boundary impurities. In a strong magnetic field
the ground-state may be considered as a condensate of SU(2) hard-core bosons. The gap
4J
is reduced by the magnetic field H , i.e., ∆ = J⊥ − γ − H . Thus the first critical field
4J
occurs at the point Hc1 where the gap is closed, i.e., gµB Hc1 = J⊥ − γ . The quantum
phase transition from a gapped phase to gapless Luttinger phase occurs. By continuing
to increase the magnetic field H over Hc1 , the triplet state |2 becomes involved in the
ground state with a finite susceptibility, also affected by the boundary impurities in the
low concentration regime. If the magnetic field is greater than the rung coupling, i.e.,
h > J⊥ , the triplet component |2 becomes the lowest level. Therefore, it is reasonable
to choose the basis order as (|2, |1, |3, |4)T . Subsequently the driving terms are given
by g (1) = −2πJ a1 − J⊥ + H , g (2) = J⊥ and g (3) = H . A second critical field Hc2
(Hc2 > Hc1 ) can be determined by the magnetization arriving at its saturation value S z = 1.
Then the reference state becomes the true physical state and the critical field Hc2 is given
by
4J
Hc2 = J⊥ + . (126)
γ
In this case, all the boundary impurities are gapfull with the ferromagnetic gap ∆ =
µB g(H − Hc2 ).
Let us now discuss the boundary impurity effects in the vicinity of the critical point Hc1 .
After a lengthy calculation, similar to that employed in [14] for the periodic case, we find
the free energy in the presence of a strong magnetic field H ,

F (0, H ) 4Q(Jc+ − Jeff ) 2Q 1
≈− 1− + fb , (127)
2L π π 2L
"
Jc+ −Jeff
where Q is the fermi point given by Q ≈ + and fb is the surface free
4Jc −5(H −Hc1 )
energy from the boundary impurities in the vicinity of Hc1 . Explicitly, for |β± | 12 , or
J
say 0 < U± γ or U± < 0, it is given by

 − 2Q(Jc+−Jeff ) 1 + 1
+ 1

, for l = m = 1, 3, 5,
π β+ β−
fb ≈ (128)
 − 2Q(Jc+−Jeff ) , for l = m = 2, 4.
π

4J
For − 12 < β± βc , or say U± Ubs = 2J /γ 1 − 1 − γ J⊥ , we have

 2Q(Jc+ −Jeff ) 3
J 1
− ±
 π 2 + β± + β± +1 − β± −1 +
1 1 1
± − γ −β 2 + 1 + J⊥ ,
± 4
fb ≈ for l = m = 1, 3, 5, (129)


 − 2Q(Jc+−Jeff ) , for l = m = 2, 4.
π
J
While for |β± | very small, or say γ < U± < Ubs , it is given by

 − 2(Jc+−Jeff )Q +
f (β ), for l = m = 1, 3, 5,
± ±
fb ≈ π
(130)
 − 2Q(Jc+−Jeff ) , for l = m = 2, 4,
π
where

8J 1 Q 2Jeff Q
f (β± ) = − arctan − 4Qβ± + arctan
πγ 1 − 4β±2 β± π β±

4Q(Jc+ − Jeff ) 1 Q
+ arctan − β ± Q . (131)
π(π + 2Q) 1 − β± 2 β±
In the above Jeff = J⊥ − H and the parameters β± are related to the boundary impurity
coupling U± by
 2U±

 , for l = m = 1, 4, 5,

2J
 γ −U±

 2U±
1
= 2J −3U , for l = m = 2, (132)
β±   γ ±


 2U
 − 2J ± , for l = m = 3.
γ +U±
2
d F (0,H )
The magnetic susceptibility follows from χ ≈ − dH 2 2L . Here, to illustrate the
boundary effects, we will focus on the discussion of the strong coupling compounds
J⊥ J with the boundary impurities in the case l = m = 1. Other regimes can be handled
in a similar way. It is very clear that the stable boundary bound states are exhibited only
in the strong ferromagnetic boundary coupling U± > Ubs . In Eq. (132), we emphasize
that the mathematical singular points do not exist, or alternatively β± = 0 does not mean
that the rhs of Eq. (132) has singular points. For instance, if U± = 2J /γ , the boundary
parameters ξ± = 1/2. Thus the phase factors in the Bethe ansatz equations (94), (95)
and (96) are equal to 1 for the case l = m = 1. In such a case, the model exihibits
special symmetry (the quantum algebra SU q (4) invariant Bethe ansatz equations) which
leads to a different expression for the boundary free energy than the above ones. For
antiferromagnetic boundary coupling U± < 0, the susceptibility is given by

3 1 1 1
χ≈ # 1+ + , (133)
π 4Jc+ (H − Hc1 ) 4L ± 2 β±
while for the strong ferromagnetic coupling U± Ubs , with U± > 0,

3 1 3 1 1 1
χ≈ # 1+ + + − . (134)
π 4Jc+ (H − Hc1 ) 4L ± 2 β± β± + 1 β± − 1
Fig. 2. The susceptibility versus magnetic H for different impurity coupling U± = U (U = 0 corresponds
to free boundary conditions). Here we consider the strong coupling compound Cu2 (C5 H12 N2 )2 Cl4 [3] with
J⊥ = 13.2K, J = 2.5K and γ = 4 with an impurity concentration 2 percent in a ladder with length L = 50.
Notice, in both cases, that the susceptibility diverges with the square root of the field in the
bulk and in the boundaries. In addition, the susceptibility at the boundaries is enhanced or
decreased by different impurity dopings. This behaviour is illustrated in Fig. 2. From the
Bethe ansatz equations, we can also calculate the magnetization in the vicinity of Hc1 ,
Q
Sz 4Q 2Q 1 z
= ρ1(1) (λ) dλ = 1− + S . (135)
2L π π 2L ± b
−Q
For antiferromagnetic boundary coupling U± < 0 this expression reduces to

2Q 2Q

1 1

Sbz ≈ 1− + , (136)
±
π π 2 β±
while for strong ferromagnetic boundary coupling U± Ubs with U± > 0,

2Q 2Q

3 1 1 1

Sbz ≈ 1− + + − . (137)
±
π π 2 β± β± + 1 β± − 1
A plot of the magnetization S z against the magnetic field for different boundary impurities
U± is given in Fig. 3. By analyzing both figures we can observe the competition between
the boundary impurities and the magnetic field in the thermodynamic properties. In
particular, we find an enhancement of the susceptibility in the weak anti- and ferromagnetic
Fig. 3. The magnetization versus magnetic field H for different impurity coupling U± = U (U = 0 corresponds
to free boundary conditions). As in Fig. 2, we consider the strong coupling compound Cu2 (C5 H12 N2 )2 Cl4 [3]
with J⊥ = 13.2K, J = 2.5K and γ = 4 with an impurity concentration 2 percent in a ladder with length L = 50.
regimes (we consider the sizes 2L = 100, the impurity concentration 2 percent). The
susceptibility and the magnetization are lifted slightly in the weak antiferromagnetic
boundary regime in the case of open boundary conditions, whereas they contribute
negatively to the bulk when U± becomes more and more negative. This is reasonable,
since negative U± energetically favours the singlet state (recall the boundary terms in the
Hamiltonian (1)), even if the magnetic field is very strong, such that the spin-1 component
of the triplet is involved in the ground state. The point is that a very negative U± can
overcome the spin-1 component of the triplet and dominate the edge rung state. In this
circumstance, the edge state is a pure singlet state and the edge magnetization (136) is
zero due to the fact that U± effectively decreases the edge magnetic field H to Hc1 such
that the fermi boundary Q = 0. This results in negative susceptibility and magnetization
contributions to the bulk. In contrast to this case, the ferromagnetic impurities lift the
J
susceptibility and the magnetization in the weak coupling regime U± < γ . When U±
becomes larger, the triplet edge state is energetically favoured so that the boundary
coupling can overcome the magnetic field to bring the three components of the triplet into
the edge states. Therefore, it causes a negative contribution to the bulk susceptibility and
magnetization. This situation is different from the case of the bulk impurities, where the
susceptibility is increased by the impurity coupling due to the forward-scattering. This fact
can be seen clearly from the one point correlation function of the ground state at the edges,
for antiferromagnetic boundary coupling and weak ferromagnetic boundary coupling, i.e.,
J
U± γ ,
3 d 3 2Q(H − Hc1 ) 4J /γ

!Sa .Ta = − + fb = − + , (138)
4 dU± 4 π (2J /γ − U± )2
and for ferromagnetic impurities in the strong coupling regime U± Pbs ,
$
1 2Q(H − Hc1 ) 4J /γ 4J /γ
!Sa .Ta = + −
4 π (2J /γ − U± ) 2 (2J /γ − 3U± )2
%
4J /γ 1
+ − . (139)
(2J /γ + U± )2 (1 − γJU± )2
In the above a = 1, L. The boundary one point correlation functions are given by
3 1
!Sa .Ta = − !NS + !NT . (140)
4 4
Here NS and NT are the probabilities of the singlet and triplet state respectively. This is
because the eigenvalue of the one point correlation function !Sa .Ta acting on the singlet
(triplet) state is − 34 ( 41 ). We have plotted the correlation function for antiferromagnetic
boundary coupling in Fig. 4.
Fig. 4. One point correlation function (138) versus antiferromagnetic boundary coupling U as a function of
magnetic field. The curve is lifted by the magnetic field, however it is decreased by the boundary impurities
which favour the singlet state. Here we consider the strong coupling compound Cu2 (C5 H12 N2 )2 Cl4 [3] with
J⊥ = 13.2 K, J = 2.5 K and γ = 4 and U± = U . The case U = 0 corresponds to the free boundary effect.
(a)
(b)
Fig. 5. One point correlation function vs ferromegnetic boundary coupling U : (a) The function (138) is lifted by
the magnetic field and weak magnetic impurity coupling U . (b) The function (139) tends to 14 as the boundary
impurity coupling becomes larger. Here we again consider the strong coupling compound Cu2 (C5 H12 N2 )2 Cl4 [3]
with J⊥ = 13.2K, J = 2.5K and γ = 4 and U± = U . The case U = 0 corresponds to free boundaries.
We see that the magnetic field always lifts the spin-1 triplet component. However,
in the case of antiferromagnetic boundary impurities and open boundaries the singlet
state is favoured as long as U± becomes more negative, the triplet moves out of the
edge state and the one point correlation function tends to − 34 . On the other hand, for
ferromagnetic coupling impurities (see Fig. 5), the correlation function increases due to
J
the ferromagnetic properties and the magnetic field in the weak coupling regime U± < γ .
However, if U± becomes larger, the three components of the triplet get involved in the
edge state, such that the correlation function tends to 14 for strong ferromagneitc impurity
coupling. This result indicates that the edge state can be a pure singlet state in the strong
antiferromagnetic boundary coupling regime whereas it turns out to be a pure triplet state in
the strong ferromagnetic boundary coupling regime. This reveals the role of antimagnetic
and ferromagnetic impurities.
On the other hand, the boundary impurities coupled to the spin degrees of freedom,
(2) (3)
namely, ρb and ρb will also affect the ground state properties nontrivially. From the
free energy (123), these impurity densities will contribute to the low energy. Considering
the case J⊥ < 0, in the absence of the magnetic field, the triplet is completely degenerate
while the fermi surface of the singlet is lifted as J⊥ becomes more negative. Certainly,
if J⊥ < Jc− = − γ ( √π − ln 3) the singlet rung state is not involved in the ground-state,
J
3
namely H (3) (0) 0, whereas two triplet fermi seas still have fermi boundaries at infinity.
Under such a configuration, the dressed energy potentials are
2πJ cosh π3 λ 2πJ sinh π3 λ
H (1) (λ) = − √ , H (2)(λ) = − √ . (141)
3 γ cosh πλ 3 γ sinh πλ
The free energy can be given by

F (0, 0) 2J 1 1
≈− ψ(1) − ψ + fb , (142)
2L 3γ 3 2L
where
∞ ∞
(1) (1) (2) (2)
fb = ρb1 (λ)H1 (λ) dλ + ρb1 (λ)H1 (λ) dλ. (143)
−∞ −∞
The first part in (142) is nothing but the standard SU(3) ground state energy of the bulk.
The remaining part is the boundary surface energy for various boundary impurities.
5. Conclusion and discussion
In summary, we have discussed in detail the algebraic Bethe-ansatz solution of an

integrable spin ladder system based on the SU(4) symmetry with boundary impurities.
Five different classes of solutions of the graded RE leading to different boundary rung
interactions in the Hamiltonian were obtained. The Bethe-ansatz equations, the eigenvalues
of the transfer matrix and the energy spectrum were given explicitly. Furthermore, the
three-level transfer matrices, characterizing the different flavour sectors separately, allowed
us to embed different impurities into the system. From the Bethe ansatz solutions (107)–
(109), we found that the boundary impurity effects characterized by ζ(vi , ξ± ), η(µj , ξ± )
and Ω(wk , ξ± ) act indeed nontrivially on the densities of roots for the three rapidities and
thus change the ground state properties, the boundary bound states as well as the low-lying
energy spectrum. In the thermodynamic limit, the spin gap remains almost unchanged.
However, the boundary susceptitblity and magnetization reveal novel magnetic properties
for strong and weak impurity couplings. In strong impurity coupling, the impurities
induced by the open boundary conditions can result in either a pure triplet or a singlet edge
state due to the nature of the pure back-scattering at the edges and the magnetic impurities.
Strictly speaking, the edge state can be a pure singlet state in a strong antiferromagnetic
boundary coupling regime whereas a triplet state with an effective magnetic moment can
exist in a strong ferromagnetic boundary coupling regime. Correspondingly, the one point
correlation function for strong antiferromagnetic boundary impurities tends to the singlet
eigenvalue −3/4, whereas for strong ferromagneitc impurity coupling it tends to the triplet
eigenvalue 1/4. This behaviour may be observed in experiments due to different boundary
magnetic moments. Although the TBA solution of the SU(4) ladder model (1) predicts
the quantum phase diagram in good agreement with experimental results for the strong
coupling compounds, the full finite temperature thermodynamic properties of the model
remain to be calculated.
Acknowledgements
A.F. and X.W.G. would like to thank J. Links, I. Roditi, R.A. Römer and Z.-J. Ying for
helpful discussions. M.T.B., X.W.G. and H.Q.Z thank the Australian Research Council for
financial support. A.F. and A.P.T. thank CNPq and FAPERGS for financial support.
References
[1] E. Dagotto, T.M. Rice, Science 271 (1996) 618;

E. Dagotto, Rep. Prog. Phys. 62 (1999) 1525.
[2] M. Azuma, Z. Hiroi, M. Takano, K. Ishida, Y. Kitaoka, Phys. Rev. Lett. 73 (1994) 3463.
[3] G. Chaboussant, et al., Phys. Rev. Lett. 79 (1997) 925;
G. Chaboussant, et al., Phys. Rev. Lett. 80 (1998) 2713;
G. Chaboussant, P.A. Crowell, L.P. Levy, O. Piovesana, A. Madouri, D. Mailly, Phys. Rev. B 55 (1997)
3046.
[4] B.C. Watson, V.N. Kotov, M.W. Meisel, Phys. Rev. Lett. 86 (2001) 5168.
[5] C.P. Landee, M.M. Turnbull, C. Galeriu, J. Giantsidis, F.M. Woodard, Phys. Rev. B 63 (2001) 100402.
[6] W. Shiramura, et al., J. Phys. Soc. Jpn. 66 (1997) 1900.
[7] A.A. Nersesyan, A.M. Tsvelik, Phys. Rev. Lett. 78 (1997) 3939.
[8] A.K. Kolezhuk, H.-J. Mikeska, Int. J. Mod. Phys. B 12 (1998) 2325;
A.K. Kolezhuk, H.-J. Mikeska, Phys. Rev. Lett. 80 (1998) 2709.
[9] P.W. Anderson, Science 235 (1987) 1196;
F.C. Zhang, T.M. Rice, Phys. Rev. B 37 (1988) 3759.
[10] H. Frahm, A. Kundu, J. Phys.: Condens. Matter 11 (1999) L557.
[11] M.T. Batchelor, M. Maslen, J. Phys. A 32 (1999) L377;
M.T. Batchelor, M. Maslen, J. Phys. A 33 (2000) 443;
M.T. Batchelor, J. de Gier, M. Maslen, J. Stat. Phys. 102 (2001) 559;

M.T. Batchelor, J. de Gier, M. Maslen, Phys. Rev. B 61 (2000) 15196;
J. de Gier, M.T. Batchelor, Phys. Rev. B 62 (2000) R3584.
[12] Y. Wang, Phys. Rev. B 60 (1999) 9236.
[13] J. Links, A. Foerster, Phys. Rev. B 62 (2000) 3845;
A. Foerster, K.E. Hibberd, J.R. Links, I. Roditi, J. Phys. A 34 (2001) L25.
[14] M.T. Batchelor, X.-W. Guan, A. Foerster, H.-Q. Zhou, New J. Phys. 5 (2003) 107.
[15] Y. Wang, P. Schlottmann, Phys. Rev. B 62 (2000) 3845.
[16] Y. Wang, P. Schlottmann, J. Appl. Phys. 89 (2001) 7335;
J. Dai, P. Schlottmann, Y. Wang, Phys. Rev. B 65 (2002) 144411.
[17] I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977.
[18] E.K. Sklyanin, J. Phys. A: Math. Gen. 21 (1988) 2375;
L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 6 (1991) 5231;
L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 7 (1992) 5657.
[19] H.J. de Vega, A. Gonzalez Ruiz, J. Phys. A: Math. Gen. 26 (1993) L519;
M.T. Batchelor, V. Frikin, A. Kuniba, Y.K. Zhou, Phys. Lett. B 376 (1996) 266.
[20] A.J. Bracken, X.Y. Ge, Y.Z. Zhang, H.Q. Zhou, Nucl. Phys. B 516 (1998) 588;
H.Q. Zhou, Phys. Rev. B 53 (1996) 5089;
H.Q. Zhou, Phys. Lett. A 228 (1997) 48.
[21] M.J. Martins, X.-W. Guan, Nucl. Phys. B 583 (2000) 721;
X.-W. Guan, J. Phys. A 33 (2000) 5391;
M. Shiroishi, M. Wadati, J. Phys. Soc. Jpn. 66 (1997) 2288.
[22] Y. Motome, N. Katoh, N. Furukawa, M. Imada, J. Phys. Soc. Jpn. 65 (1996) 1949, and references therein.
[23] A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611.
[24] M.J. Martins, P.B. Ramos, Nucl. Phys. B 522 (1998) 413.
[25] X.-W. Guan, A. Foerster, U. Grimm, R.A. Römer, M. Schreiber, Nucl. Phys. B 618 (2001) 650.
[26] A. Foerster, X.-W. Guan, J. Links, I. Roditi, H.-Q. Zhou, Nucl. Phys. B 596 (2001) 525.
[27] M. Takahashi, Prog. Theor. Phys. 46 (1971) 401;
P. Schlottmann, Phys. Rev. B 33 (1986) 4880.
[28] H.M. Babujian, Nucl. Phys. B 215 (1983) 317.
[29] C.N. Yang, C.P. Yang, Phys. Rev. B 150 (1966) 321;
C.N. Yang, C.P. Yang, Phys. Rev. B 150 (1966) 327;
N. Fukushima, Y. Kuramoto, J. Phys. Soc. Jpn. 71 (2002) 1238.
N = 2 boundary supersymmetry in integrable

models and perturbed boundary conformal field
theory
P. Baseilhac a , K. Koizumi b
a Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan
b Department of Physics, Kyoto Sangyo University, Kyoto 603-8555, Japan
Received 14 April 2003; accepted 1 August 2003
Abstract
Boundary integrable models with N = 2 supersymmetry are considered. For the simplest boundary
N = 2 superconformal minimal model with a Chebyshev bulk perturbation we show explicitly how
fermionic boundary degrees of freedom arise naturally in the boundary perturbation in order to
maintain integrability and N = 2 supersymmetry. A new boundary reflection matrix is obtained for
this model and N = 2 boundary superalgebra is studied. A factorized scattering theory is proposed
for a N = 2 supersymmetric extension of the boundary sine-Gordon model with either (i) fermionic
or (ii) bosonic and fermionic boundary degrees of freedom. Exact results are obtained for some
quantum impurity problems: the boundary scaling Lee–Yang model, a massive deformation of the
anisotropic Kondo model at the filling values g = 2/(2n + 3) and the boundary Ashkin–Teller model.
PACS: 11.10.Kk; 11.25.Hf; 12.60.Jv; 11.55.Ds; 12.40.Ee
Keywords: N = 2 supersymmetry; Massive boundary integrable field theory; Reflection equations; Reflection
matrix
1. Introduction
In superstring theory, it has recently been shown that certain backgrounds are associated
with two-dimensional integrable massive field theories on the worldsheet [1,2]. For
instance, the sine-Gordon model at its N = 2 supersymmetric point [3] corresponds to
E-mail addresses: pascal@yukawa.kyoto-u.ac.jp (P. Baseilhac), kkoizumi@cc.kyoto-su.ac.jp (K. Koizumi).
doi:10.1016/j.nuclphysb.2003.08.002
418 P. Baseilhac, K. Koizumi / Nuclear Physics B 669 [FS] (2003) 417–434
the simplest generalization of the pp-wave background [4] and the N = 2 supersymmetric
sine-Gordon model [5] was suggested as a good string background [1]. Similarly, for open
superstrings boundary integrable models with N = 2 supersymmetry arise naturally [6,7].
Exact results for such integrable field theories are then desirable.
In statistical physics, several low-dimensional systems around their critical points
possess a description in terms of two-dimensional integrable boundary quantum field
theories. Among the famous examples, one finds the Kondo model which describes the
s-wave scattering of electrons off a magnetic spin impurity. Another example corresponds
to free fermions on the half-line which describes the scattering of fermions off a
monopole (Callan–Rubakov effect). Also, minimal models and their extensions (with
N = 1, 2 supersymmetry or W -symmetry) arise in quantum impurity problems. Although
nonperturbative results for (pure) bulk perturbation can be derived using certain quantum
group restrictions of (super)sine-Gordon, Bullough–Dodd or more generally affine Toda
models, in case of an integrable boundary perturbation such a relation is still an open
problem which needs further analysis.
As explained in [8,9], a boundary perturbation of a massive integrable quantum
field theory will preserve integrability if the perturbing boundary operator possesses a
representation with respect to the underlying chiral algebra on the half-plane1 and provided
it belongs to the same representation than the bulk potential. It is important to mention that
[10] the choice of conformal boundary conditions plays a crucial role as it restricts the
representations of the boundary operators.
Among the simplest known examples, one finds the critical Ising field theory (with
central charge c = 1/2) perturbed by the energy operator in the bulk (Φ13 representation
of the Virasoro algebra) [11]. Due to the argument [8] recalled above, it follows that a
boundary operator in the Φ13 representation leads to a boundary integrable model. Indeed,
from the fusion rule of the spin operator σ σ → I + Φ13 a boundary integrable model can
be constructed as an off-critical Ising model restricted on the half-line with free conformal
boundary conditions perturbed by a boundary spin operator coupled with one fermionic
boundary degree of freedom [8].
As mentioned in [9], it would be interesting to construct a nontrivial boundary integrable
perturbation of a critical Ising field theory on the half-line perturbed by a relevant
magnetic operator in the bulk (Φ12 representation). Indeed, if one starts with standard
boundary conditions (free or fixed), one simply cannot obtain a boundary operator in
the representation of the Virasoro algebra [10]. Starting from certain boundary conditions
changing operators [10], it is however expected that one might be able to construct a well-
defined boundary operator that preserves integrability. But in this case, the underlying
boundary conformal field theory (BCFT) with these special boundary conditions has to
be identified, which remains an open question. In order to solve this problem, it has been
pointed out that introducing new degrees of freedom located at the boundary might provide
a useful tool.
1 This can be shown in first order of conformal perturbation theory. If operators are very relevant, scaling
arguments can be used to show that local conserved currents exists at all orders in conformal perturbation theory.
P. Baseilhac, K. Koizumi / Nuclear Physics B 669 [FS] (2003) 417–434 419
In both previous examples, a perturbed boundary minimal model (critical Ising field
theory) is considered for which the representations of the Virasoro algebra (labeled by the
conformal weights) are in finite number. Fusion rules then restrict drastically the number
of models that can be constructed.2 It is probably one reason why the presence of boundary
degrees of freedom seems to be sometimes necessary.
On the other hand, we may now wonder if boundary degrees of freedom occur in case
of BCFTs with a continuous series of representations like the Liouville or, more generally,
Toda field theories perturbed simultaneously by bulk and boundary operators in the same
representation. An interesting example is provided by the sine-Gordon (SG) field theory
on the half-line without [8,14] or with [15,16] degrees of freedom at the boundary. In full
generality, we define its Euclidean action as:
∞ 0 ∞
1
AbSG = dy (∂ν φ)2 − 2µ cos(β̂φ) +
dx B
dy Φpert (y) + Aboundary,
8π
−∞ −∞ −∞
(1)
where the interaction between the SG field and the boundary terms E± (y) with dimension
∼ µ1/2 reads
B
Φpert (y) = E− (y)ei β̂φ(0,y)/2 + E+ (y)e−i β̂φ(0,y)/2.
√
Here we used the notation β̂ = β/ 4π and introduced the UV mass parameter µ. The last
term only depends on the boundary degrees of freedom, and is only considered if necessary.
If there are no degrees of freedom at the boundary, the terms E± (y) ∼ const are just
c-numbers and we set Aboundary ≡ 0 in (1). This two-parameter family3 of boundary
integrable models have been studied in details since many years, for which exact results
[8,17–30] have been confirmed using various methods. This model can be considered as an
integrable perturbation of the boundary Liouville field theory as suggested in [31]. Using
vertex operators representations [31–34], it is not difficult to see in (1) that the boundary
perturbing operators are indeed in the same representation than the bulk ones (for identical
bulk and boundary background charges). Without boundary degrees of freedom, the UV
limit of the boundary SG model defined in (1) can be identified either with the Gaussian
field with Neumann boundary conditions or with the boundary Liouville field theory as
considered in [31].
For certain degrees of freedom at the boundary and generic values of the coupling β̂, the
classical Hamiltonian associated with the model (1) above is integrable [15]. At quantum
level, integrability is preserved [16] for
1 − β̂ 2 1 − β̂ 2
E+ (y) = ±2eUV cosh p̂(y) and E− (y) = ±2eUV cosh q̂(y), (2)
β̂ 2 β̂ 2
2 For boundary conformal field theories with higher symmetries such as W -minimal models [12], Z models
n
[13], . . . the same situation occurs. For a discussion about the relation between boundary Toda models with
imaginary coupling and perturbed boundary W -minimal models, see for instance [9].
3 Only boundary parameters are mentioned here. The bulk mass is sometimes considered as a third (bulk)
parameter.
with [p̂(±∞), q̂(±∞)] = α mod(4iπ) and eUV is a boundary mass parameter. Depending
on each sign in (2), the boundary quantization length α/i associated with the boundary
quantum mechanical system is fixed to
(β̂ 2 − 1) (β̂ 2 − 2)
α = i4π (+) or α = i2π (−). (3)
β̂ 2 β̂ 2
Introducing a “dynamical” extension [16] of the Cartan subalgebra (identified with the
topological charge in the SG model) and using the boundary quantum group structure
2 )), we obtained the soliton/antisoliton and
discovered in [35]4 (coideal subalgebra of Uq (sl
breathers boundary reflection matrices. In particular, for generic values of the coupling β̂,
we found that the bulk and the boundary masses are locked together.5 We refer the reader
to [16] for more details.
For generic values of the coupling β̂, this model deserves certainly some interest, but
taking some special values provides interesting new insights as we are going to show.
Indeed, for β̂ 2 = 2p/p with 0 < p < p integers the bulk SG model is known to be related
with the Φ13 bulk perturbation of minimal models. Furthermore, at the special point β̂ 2 =
4/3 (repulsive regime) it describes an integrable bulk perturbation (Chebyshev type) of
N = 2 superconformal minimal model. The SG model with boundary degrees of freedom
introduced and studied in [16] being one possible boundary integrable perturbation6 of the
bulk SG model, it is important to analyse in details these special points.
In this paper we will focus on β̂ 2 = 4/(2n + 3), n ∈ N in (1) for several reasons.
First, it will provide new examples of perturbed conformal field theories with boundary
degrees of freedom for which the factorized scattering theory is proposed. Secondly, for
generic values of the coupling β̂ in the model (1), free parameters can be introduced in
the boundary operators E± (y) through a canonical transformation of p̂(y), q̂(y). Then, it
is important to see if different boundary operators E± (y) preserving integrability can be
constructed at these special points for which the explicit dependence in terms of the free
parameters might be changed. Third, for n = 0 a massive N = 2 boundary supersymmetry
algebra appears explicitly.
The paper is organized as follows. In the next section, we exhibit an exact relation
between the simplest bulk-boundary perturbed N = 2 superconformal minimal model
and the sine-Gordon model at special point β̂ = 4/3 with fermionic boundary degrees
of freedom. It shows that, similarly to the critical Ising model in a boundary magnetic
field, boundary degrees of freedom appear naturally in this model. The boundary
supersymmetric charges are explicitly constructed from the Lagrangean and differ from
the ones proposed in [9,41]. Although their expressions are similar to the ones in [41]
(which may probably be considered as “effective” supersymmetry charges in terms of
4 This structure has been further studied in [36–38] for which new reflection matrices associated with non-
dynamical boundary have been obtained.
5 Up to a canonical transformation of the boundary degrees of freedom, the UV parameter e
UV is fixed by the
boundary non-local conserved charges algebraic structure in terms of the bulk UV mass parameter µ. Although
not shown explicitly, Warner predicted such phenomena [9].
6 There, we did not discuss integrable massless models with boundary degrees of freedom considered, for
instance, in [39,40].
the boundary structure) the main difference here is the fermionic boundary algebra that
appears associated with the fermion number operator. These supersymmetry charges
are used to construct a new boundary reflection matrix with N = 2 supersymmetry
which possesses two free parameters and is shown to satisfy the boundary Yang–Baxter
equations. The corresponding massive N = 2 boundary superalgebra is constructed and
the boundary free energy is proposed. In Section 3, a boundary scattering associated with a
N = 2 supersymmetric version of the boundary sine-Gordon model with purely fermionic
(proposed in [42]) or a mixing of fermionic and bosonic boundary degrees of freedom is
proposed. In any case, the supersymmetric part is different from the one suggested in [9,
41]. In Section 4, we present applications of our results to quantum impurity problems.
The boundary scaling Lee–Yang model is revisited in light of our results. The anisotropic
Kondo model is shown to admit a simple massive integrable extension at special points
g = 2/(2n + 3) with n ∈ N, for which scattering amplitudes are proposed. Finally, a
boundary version of the Ashkin–Teller model is briefly presented.
2. N = 2 supersymmetry in the boundary sine-Gordon model
Among the class of relevant perturbation of N = 2 superconformal minimal model with

central charge c = 3/ + 2, the ones corresponding to integrable N = 2 supersymmetric
field theories have been studied in details [43]. Three different types of relevant bulk
perturbations can be considered. Here, we focus on the model associated with a
perturbation of the form (the so-called Chebyshev perturbation):
bulk
Φpert (z, z̄) = λ̂G− 1 G ¯ + G
− 1 Φ + (z, z̄) + λ̂G + 1 Φ − (z, z̄), (4)
−2 −2k 1
−2 k−2
where λ̂ is a complex parameter that characterizes the strength of the perturbation. The
chiral primary fields Φk± possess conformal dimensions ∆k = ∆ k = k/2( + 2) and G±
−1/2
− ) are N = 2 supersymmetry generators. Actually, this perturbation is known to be
(G −1/2
the N = 2 superconformal analogue of the energy perturbation of the ordinary critical Ising
model. Following the case of the off-critical Ising model in a boundary magnetic field [8]
and in order to maintain supersymmetry (see details in [9]), it is then natural to consider
the model with the bulk perturbation (4) on the half-line and a boundary perturbation of
the form [9]:
√ √
(y) = µB e−iφ0 / ν− (y)G− 1 Φ
+ (y) + µB eiφ0 / 3 ν+ (y)G+ 1 Φ
− (y).
boundary 3 uv uv
Φpert k k (5)
− 2 − 2
Here G∓ ±
−1/2 Φ̂k (y) are boundary operators in the same N = 2 superconformal representa-
tion as the bulk operators, φ0 is a phase and µB is the boundary mass scale. It is important
to stress that the resulting model is neither expected to be integrable nor N = 2 super-
symmetric for any boundary degrees of freedom ν± uv (y). Also, both requirement must be
simultaneously consistent. This can be checked from the structure of the Lagrangean, but
also in the scattering amplitude. To proceed further, let us consider a concrete example: the
sine-Gordon model at β̂ 2 = 4/3.
In the bulk and for = 1 the simplest N = 2 superconformal minimal model possesses
a Lagrangean representation in terms of a single free boson φ̃(x, y) compactified on
the “supersymmetric” radius. In terms of its holomorphic and antiholomorphic parts

¯
it reads φ̃(x, y) = ϕ̃(z) + ϕ̃(z̄) with z = y + ix and z̄ = y − ix. If we denote the
expectation value over the Fock vacuum space of massless fields · · ·0 , then the
holomorphic/antiholomorphic components are normalized such that

ϕ̃(z)ϕ̃(w) 0 = − ln(z − w), ¯ ϕ̃(
ϕ̃(z̄) ¯ w̄) = − ln(z̄ − w̄), ¯ w̄) = 0.
ϕ̃(z)ϕ̃(
0 0
The N = 2 supersymmetric generators correspond to the stress-energy tensor which
ensures conformal invariance, the U (1) current and the two supersymmetry generators.
Their holomorphic parts read, respectively,
1 i √
T (z) = − (∂z ϕ̃)2 , J (z) = √ ∂z ϕ̃, G± (z) = exp ±i 3 ϕ̃(z) . (6)
2 3
The order parameter and its conjugate are defined as Φk=1 ± ¯
(z, z̄) = exp(± √i (ϕ̃(z) + ϕ̃(z̄))).
3
Using the definitions (6) above, it follows that the interacting terms in the bulk perturbation
(4) become

± ± ∓ 2i ¯

G 1 G 1 Φk (z, z̄) = exp ± √ ϕ̃(z) + ϕ̃(z̄) . (7)
−2 −2 3
On the half-line, one can use the method of mirror images for Neumann boundary
conditions ∂x φ̃(x, y)|x=0 = 0 i.e., one defines φ(x, y) = φ̃(x, y) + φ̃(−x, y). In terms of
its holomorphic and antiholomorphic part restricted on the half-line it gives
¯
ϕ(z) = ϕ̃(z) + ϕ̃(z) ¯ + ϕ̃(z̄),
and ϕ̄(z) = ϕ̃(z̄) (8)
from which one deduces the two-point functions in the BCFT with Neumann boundary
conditions

ϕ(z)ϕ(w) 0 = −2 ln(z − w), ϕ̄(z̄)ϕ̄(w̄) 0 = −2 ln(z̄ − w̄),

ϕ(z)ϕ̄(w̄) 0 = −2 ln(z − w̄).
In particular, from these definitions7 and using the supersymmetric generators restricted on
±
√
the half-line G (z) = exp(±i 3 ϕ(z)) the boundary perturbation becomes

i
i
± ∓
G 1 Φk (y) ≡ exp ∓ √ ϕ(z) + ϕ̄(z̄)

= exp ∓ √ φ(0, y) . (9)
−2 3 3
x=0
It follows that the action associated with the simplest N = 2 boundary superconformal
minimal model corresponds to the boundary sine-Gordon model (1) at the special point
β̂ 2 = 4/3 with the substitutions
√
µ → −λ̂ = −λ̂¯ and E± (y) → µB e±iφ0 / 3 uv
ν± (y). (10)
At the supersymmetric point, the bulk SG model is in the repulsive regime (no
breathers): the spectrum consists of one soliton and one antisoliton. In the following, they
are denoted |u(θ ) and |v(θ ) with fermion number +1/2 and −1/2, respectively, and form
x=0 = ϕ̄(z̄)|x=0 which gives ϕ(z)|x=0 = φ(0, y)/2.

7 One has ϕ(z)|
a two-dimensional supermultiplet [5,44,45]. The corresponding bulk S-matrix is known for

a long time [46]. For general values of the coupling in the SG model, four nonlocal charges
with fractional spin and one topological charge generating the Uq (sl 2 ) affine quantum
enveloping algebra exist [44]. However, at β̂ = 4/3 they become local (spin 1/2) and
2
together with the fermion number operator they generate the massive superalgebra. If θ
denotes the particle rapidity and M its mass, in the supermultiplet representation πθ they
act as
√ √
πθ (Q+ )|d(θ ) = 2M eθ/2 |u(θ ), − |u(θ ) = 2M e−θ/2 |d(θ ),
π θ (Q
√ √
πθ (Q− )|u(θ ) = 2M eθ/2 |d(θ ), + )|d(θ ) = 2M e−θ/2 |u(θ ), (11)
π θ (Q
and
1 1
F |u(θ ) = |u(θ ), F |d(θ ) = − |d(θ ). (12)
2 2
Using this representation, the massive superalgebra can be constructed [5,45,47] in terms of
the Hamiltonian H , momentum P and particle number N with eigenvalues on one-particle
states E = M cosh(θ ), P = M sinh(θ ) and 1 or 0, respectively. Using the definition of the
coproduct [5,45] which acts on multiparticle states one can check that the bulk S-matrix
commutes with these supercharges and the fermion number operator.
If we now consider the SG model on the half-line at the special value β̂ 2 = 4/3 as
defined above (1), using the results of [16] together with a scale transformation it follows
that the boundary supercharges
√
± = Q± + Q ∓ + ν̂± 2 M ir
Q ir
(−1)F with ν̂± ir
=∓ ν± (13)
k
are conserved.8 Here k denotes a real parameter. To relate the IR and UV data (using
the relation in the UV of the sine-Gordon non-local charges and the supercharges),
uv
asymptotically we have assumed the boundary conditions for the operators ν± (y) at both
“ends” of the boundary:
uv
ν+ (y = ±∞) ∼ ν+
ir uv
and ν− (y = ±∞) ∼ ν−
ir
. (14)
Similarly to the boundary SG model with boundary degrees of freedom, the asymptotic
boundary operators ν±ir commute with the massive (bulk) N = 2 superalgebra. Thus we
have
u u √
πθ Q± = −i ν̂± ir
, ± = 2M e±θ/2 ,
πθ Q
u d
d √ d

πθ Q± u = 2M e ∓θ/2
, πθ Q± = i ν̂±
ir
.
d
If we impose N = 2 boundary supersymmetry in the quantum model, the supercharges (13)
must commute with the boundary reflection matrix. Then, we are looking for the “minimal”
8 To do that, one uses the local expression of the supercharges in terms of the field and derives their
conservation in first order of conformal perturbation theory. Using scaling arguments, it can be shown that this is
true at all orders too, which allows to define them asymptotically, i.e., in the IR limit.
solution of the dynamical extension of the intertwining equation considered in [9,41]:

ν δ
± = π−θ Q
Ksusy δν (θ )πθ Q ± Ksusy ν (θ ), (15)
ζ ν ζ
where indices {δ, ν, ζ } ∈ {u, d} refer to two-dimensional supermultiplet representation

on asymptotic soliton states |u(θ ), |d(θ ). Then, using this representation the solution
Ksusy (θ ) is written as a 2 × 2 matrix with entries expressed in terms of the boundary
operators. In the non-dynamical case [8,9], the entries are just analytic functions of θ as
ir are c-numbers. Let us define
ν±
Ksusy uu (θ ) = A(θ ), Ksusy ud (θ ) = B(θ ),
Ksusy du (θ ) = D(θ ), Ksusy dd (θ ) = E(θ ). (16)
After some calculations, we find that the entries of the minimal solution Ksusy (θ ) of the
intertwining equation (15) takes the following form:

1 1 i(α + 1) ir ir
A(θ ) = √ eθ/2 ν+ ir
+ e−θ/2 ν−ir
, B(θ ) = −i sinh(θ ) + ν ,
+ − ν ,
k 2 2 k2

1 1 i(α − 1) ir ir
E(θ ) = √ eθ/2 ν− ir
+ e−θ/2 ν+ir
, D(θ ) = −i sinh(θ ) + ν ,
+ −ν ,
k 2 2 k2
(17)
where the boundary operators satisfy
ir ir ir ir 2 ir
ν± , ν+ , ν− = 0 and ν± , ν+ = 0. (18)
Here α is a real parameter. One obvious realization is ν±ir ≡ e ±iξ in which case we recover
the Ghoshal–Zamolodchikov–De Vega–Gonzalez-Ruiz minimal solution [8,48]. An other

solution to (18) is to introduce a dimensionless complex fermion at the boundary b(y),
b† (y) such that using (14), asymptotically it satisfies

{b, b} = b† , b† = 0 and b, b† = 1 with b(±∞) ≡ b. (19)
Together with the asymptotics (14) we choose the representation ν± ir = σ in (17), where
±
σ± are Pauli matrices which act on the (pure) boundary space of states. In the Lagrangean
(1) with (10), it leads to the choice
uv
ν+ (y) = b(y) uv
and ν− (y) = b† (y). (20)
We would like to stress that the minimal part (16) with (17) of the boundary reflection
matrix associated with the SG model at its N = 2 supersymmetric point possesses two free
parameters {k, α}. In full generality, one may choose to introduce a phase e±iρ in front
ir in (17). However, this phase can be removed using a gauge transformation over the
of ν±
fermionic boundary degrees of freedom.
2.1. Boundary Yang–Baxter equations and boundary reflection matrix
In the bulk, the perturbed N = 2 superconformal minimal model (the SG model at

β̂ 2 = 4/3) is massive and integrable. Integrability imposes strong constraints on the system
which implies that the S-matrix has to satisfy the quantum Yang–Baxter equations. The
entries at β̂ 2 = 4/3 read for u = −iθ :
Ssusy ±±
±± (θ ) ≡ a(θ ) = cos(u/2)Z(u), Ssusy ±∓
±∓ (θ ) ≡ b(θ ) = sin(u/2)Z(u),
Ssusy ±∓
∓± (θ ) ≡ c(θ ) = Z(u) (21)
and vanish otherwise. The factor Z(u) ensures unitarity and crossing symmetry of the S-
matrix. Its exact expression can be found in [45].
For boundary integrable field theories, the soliton/antisoliton reflection matrix is
constrained by the boundary Yang–Baxter equations (the so-called reflection equations).
In our case, it reads Rsusy (θ ) and obeys
1 2
Ssusy (θ − θ )R susy (θ )Ssusy (θ + θ )R susy (θ )
2 1
= R susy (θ )Ssusy (θ + θ )R susy(θ )Ssusy (θ − θ ), (22)
1 2
where we used the notations M = M ⊗ I , M = I ⊗ M. If N = 2 supersymmetry is
preserved, we write it in terms of the solution (16) as
Rsusy (θ ) = Ksusy (θ )Ysusy (θ ), (23)
where the scalar factor Ysusy (θ ) has to be determined using some physical assumptions (see
below). Recalling the definition of the entries (16) it leads to fourteen functional equations
[16]:

(i) a− c+ BD − B D + a− a+ [A, A ] = 0,
(ii) b− b+ [A, E ] + c− c+ [E, E ] + c− a+ (DB − D B) = 0,
(iii) c− b+ (EA − E A) + b− c+ (AA − E E) + b− a+ [B, D ] = 0,
(iv) b− b+ AD + c− c+ ED + c− a+ DA − a− a+ D A − a− c+ E D = 0,
(v) b− b+ B A + c− c+ B E + c− a+ A B − a− a+ AB − a− c+ BE = 0,
(vi) b− b+ D E + c− c+ D A + c− a+ E D − a− a+ ED − a− c+ DA = 0,
(vii) b− b+ EB + c− c+ AB + c− a+ BE − a− a+ B E − a− c+ A B = 0,
(viii) b− a+ BE + c− b+ EB + b− c+ AB − a− b+ E B = 0,
(ix) b− a+ A B + c− b+ B A + b− c+ B E − a− b+ BA = 0,
(x) b− a+ E D + c− b+ D E + b− c+ D A − a− b+ DE = 0,
(xi) b− a+ DA + c− b+ AD + b− c+ ED − a− b+ A D = 0,
where we used the shorthand notations a− = a(θ − θ ), a+ = a(θ + θ ) and similarly
for b and c as well as A = A(θ ) and A = A(θ ) and similarly for B, D and E. The
remaining three equations are obtained from (i), (ii), (iii) through the substitutions A ↔ E
and B ↔ D. Straightforward calculations show that Ksusy (θ ) given by (16) with (17)
indeed satisfies all reflection equations above.
In [8], Ghoshal and Zamolodchikov proposed the use of the “boundary unitarity” and
“boundary cross-unitarity” conditions to determine the overall factor Y (θ ) associated with
a non-dynamical boundary. In case of a dynamical boundary, it can be applied directly [16]

and gives the following equations:
• Boundary unitarity: Rsusy ac (θ )Rsusy cb (−θ ) = δba I; (24)
• Boundary cross-unitarity:

Rsusy āb (iπ/2 − θ ) = Ssusy ab
a b (2θ )Rsusy b̄ (iπ/2 + θ ),
a
(25)
where the operator I denotes the identity which acts trivially on the boundary ground state.
We refer the reader to [8] for details. Let us now introduce two meromorphic functions
susy susy
Y0 (θ ) and Y1 (θ ) such that the prefactor in (23) is written
susy susy
Ysusy (θ ) = Y0 (θ )Y1 (θ ). (26)
Using the explicit expressions (17), they are chosen such that they solve the functional
equations
susy susy
Y0 (θ )Y0 (−θ ) = 1,
susy susy
Y0 (iπ/2 − θ ) = cos(u)Z(2u)Y0 (iπ/2 + θ ),
−1
susy susy 1 1 − α 2 + 2k 2
Y1 (θ )Y1 (−θ ) = − sin (u/2) cos (u/2) − 2 sin2 (u/2) +
2 2
,
k 4k 4
susy susy
Y1 (iπ/2 − θ ) = Y1 (iπ/2 + θ ) (27)
in order to have (24) and (25). Using the results of [8] and [16] for λ = 2/β̂ 2 − 1 we finally
obtain

Y0 (θ ) = R0 (u)G0 (u)
λ=1/2
susy
(28)
where we used [8]
∞

@ 4λk − π @ 1 + 4λ(k − 1) − π
2λu 2λu
R0 (u) = [u → −u]−1
k=1
@ λ(4k − 3) − π @ 1 + λ(4k − 1) − π
2λu 2λu
and [16]
∞

@ 1 + (4k − 2)λ − 2λu
π @ (4k − 2)λ − π
2λu
−1
G0 (u) = [u → −u] .
k=1 π @ 1 + 4kλ − π
@ (4k − 4)λ − 2λu 2λu
The other part is given by

susy σ (η, u)σ (iϑ, u)
Y1 (θ ) = (29)
cos(η) cosh(ϑ)
λ=1/2
with [8]
cos x
σ (x, v) =
cos(x + λv)
∞
1
@ 12 + (2l − 1)λ + πx − λv
π @ 2 + (2l − 1)λ − π − π
x λv
−1
× 1 [v → −v] .
l=1
@ 12 + (2l − 2)λ − πx − λv
π @ 2 + 2lλ + π − π
x λv
Here η and ϑ are two real IR boundary parameters related with k and the parameter ξ by
1 1
cos(η) cosh(ϑ) = − cos ξ and cos2 (η) + cosh2 (ϑ) = 1 + 2 . (30)
k k
Using these definitions, the parameter α entering in the boundary reflection matrix (23)
with (16) and (26) becomes

α± = ± 1 − 2k 2 cos(2ξ ). (31)
ir ≡ e ±iξ , there are no values
It should be stressed that apart from the commutative case ν±
of the parameters for which the boundary reflection matrix (23) reduces to the Ghoshal–
Zamolodchikov one. Furthermore, we also checked that our solution cannot be obtained
from the Ghoshal–Zamolodchikov one using the fusion procedure suggested in [9]. Notice
that for the commutative case the exact relations between the IR boundary parameters
{η, ϑ} and the UV boundary parameters have been obtained by Al.B. Zamolodchikov [49]
and are supported by truncated conformal space analysis [50].
2.2. The massive N = 2 boundary superalgebra and boundary free energy
Let us now see in which manner the presence of boundary fermionic degrees of
freedom affects the N = 2 supersymmetry algebra. To see this, one can use the one-
particle asymptotic states |u(θ ), |d(θ ) representation (11) [45] as before. They provide a
representation for the N = 2 massive superalgebra [47]

{Q+ , Q− } = 2(H + P ), Q+ , Q− = 2(H − P ),

− = 2MN ,
Q+ , Q Q+ , Q− = 2MN ,

Q2± = Q ±2 = 0, {Q± , F } = Q ± , F = 0. (32)
Using these relations, the expressions for the boundary supercharges (13) and anticommu-
tation relations for the asymptotic boundary degrees of freedom (19), it is straightforward
to derive the relations

Q±2 = 2MN and Q − = 4 H − E λ=1/2 (−1)2F ,
+ , Q
boundary
λ=1/2 M
where Eboundary ≡ . (33)
k2
Here it is interesting to notice that the term associated with the boundary energy is nothing
but the boundary energy of the non-dynamical boundary SG model9 at special point
β̂ 2 = 4/3, given by [49] (see [50] for details)

M η ϑ 1 π 1 π 1
λ
Eboundary =− cos + cos − cos + sin − , (34)
2 cos(π/2λ) λ λ 2 2λ 2 2λ 2
9 Whereas we checked explicitly that it does not coincide with the boundary free energy calculated for general
values of the coupling, i.e., in case of boundary degrees of freedom of the form (2).
together with (30). Consequently, both terms are positive definite for any value of the
parameter k and any boundary ground state, as required. As an example, we can consider
the massless limit M → 0 with k finite. In this case, the expectation value in the second
term of (33) takes its minimal value, independently of k. Notice that this subalgebra is
different from the one given in [41]. Although the second anticommutator is identical, the
first differs by boundary contributions. Indeed, in [41] a term proportional to the operator
F 2 arises in the r.h.s. of the first anticommutator. In other words (apart from the massless
case), in order to avoid negative (or complex) expectation values of Q ±2 Bogomolnyi
bounds are required in [41] for the boundary parameters (k, ξ ). It is however not the case
here.
3. Comments about the boundary N = 2 supersymmetric sine-Gordon
In the bulk, the boundary N = 2 supersymmetric sine-Gordon model possesses local

(nonlocal) conserved charges which commute with each other [5]. Together, they generate
2 ) ⊗ Uq 2 =−1 (sl
the quantum affine algebra Uq (sl 2 ). It follows that the soliton S-matrix has
the factorized form [5]:
N=2
SSG (θ ) = SSG,β̂ 2 (θ ) ⊗ SSG,q 2 =−1 (θ ), (35)
where the bosonic part SSG,β̂ 2 (θ ) is the sine-Gordon scattering matrix. From the tensor
product structure, in the boundary case the boundary reflection matrix also takes a
factorized form:
N=2
RSG (θ ) = RSG,β̂ 2 (θ ) ⊗ Rsusy (θ ). (36)
In [42], an action for the N = 2 supersymmetric sine-Gordon model has been proposed. It
is exact in case of massless bulk and an approximation to first order in the bulk mass.
At the boundary, this action contains fermionic boundary degrees of freedom coupled
with the fermionic and bosonic fields ψ ± , ψ̄ ± , ϕ ± . Starting from this Lagrangean, we
checked explicitly that the boundary supercharges take a form similar to (13) with (14) and
(20). They are different from the ones proposed in [41], which do not contain fermionic
boundary degrees of freedom in front of the term associated with the topological charge.
Higher order corrections in the bulk mass term, if needed, would lead to the same form.
Consequently, an exact action for the N = 2 supersymmetric boundary sine-Gordon model
would lead to conserved boundary supercharges of the form (13) generating the N = 2
massive boundary superalgebra (33), instead of the ones proposed in [41]. Using previous
analysis, it follows that the supersymmetric part Rsusy (θ ) of the reflection matrix (36) is
given by (23).
Let us now turn to the bosonic part. Actually, there are two kinds of Lagrangean that
can be constructed. On one hand, with a boundary interaction of the form [42] which
only contains fermionic boundary degrees of freedom (here denoted b(y), b† (y)). In this
case, the pure bosonic boundary reflection scattering matrix RSG,β̂ 2 (θ ) contains two free
parameters and follows from the one proposed by Ghoshal–Zamolodchikov [8]. On the
other hand, it is well expected that a boundary interaction including bosonic (denoted
p(y), q(y)) and fermionic boundary degrees of freedom can be constructed as well. In
such case, the bosonic boundary reflection matrix follows from the results of [16], i.e., it
reads
RSG,β̂ 2 (θ ) = K0 (θ )Y (θ ), (37)
where Y (θ ) ensures boundary unitarity and boundary cross-unitarity symmetry. Its
“minimal” part K0 (θ ) takes the same form as in (16) with entries
1 −1 θ/β̂ 2
cosh(p) − qe−θ/β̂ cosh(q) q − q −1 ,
2
A(θ ) = ± q e
2c
1 −1 θ/β̂ 2
cosh(q) − qe−θ/β̂ cosh(p) q − q −1 ,
2
E(θ ) = ± q e
2c
1
B(θ ) = 2 −c2 q −1 e2θ/β̂ − c2 qe−2θ/β̂
2 2
2c

q − q −1 −1
+ q cosh(q) cosh(p) − q cosh(p) cosh(q) ,
q + q −1

1
D(θ ) = 2 −c2 q −1 e2θ/β̂ − c2 qe−2θ/β̂
2 2
2c

q − q −1 −1

+ −q cosh(q) cosh(p) + q cosh(p) cosh(q) . (38)
q + q −1
2
Here, the deformation parameter q = eiπ/β̂ and c = sin(π/β̂ 2 ). Depending on the sign in
(38), the boundary quantization condition is fixed to
2iπ
[p, q] = − mod(2iπ). (39)
β̂ 2
In total, the boundary reflection matrix (36) of each model contains four or two boundary
parameters (up to a canonical transformation of the boundary degrees of freedom p and
q), respectively. Probably there exists some relation between them, following the analysis
of pole structure in [41]. Details as well as further study of the second model will be
considered elsewhere.
To conclude, notice that considering the analytic continuation β̂ 2 = −b 2 would provide
the factorized scattering theory for the (second) N = 2 supersymmetric boundary sinh-
Gordon model. In this case, up to the sign, the boundary quantization condition for
boundary bosonic degrees of freedom becomes [p, q] = 2iπ/b2 . Due to the recent
conjecture [51] about the dual representation of N = 2 supersymmetric Liouville field
theory, it is worth interesting to understand whether a dual boundary action with [p, q] =
2iπb2 can be explicitly constructed.
4. Relations with quantum impurity problems
Several quantum impurity problems can be analyzed using analytical methods. For
strongly interacting systems, nonperturbative ones are crucial in order to study bosonized
versions of gapless (critical) quantum systems (for a review, see, for instance, [52]). Among
the boundary quantum field theories that can describe such systems, the boundary massless
SG model has been proposed and some exact results obtained from its nonperturbative
analysis [53]. Also, massive theories have applications in 1D impurity systems with an
excitation gap. Let us see how our model provides exact results for such systems.
4.1. The boundary scaling Lee–Yang model
The scaling limit of the Ising model with a purely (bulk) imaginary magnetic field is
described by the Lee–Yang model. Its UV limit leads to the non-unitary M2/5 minimal
conformal field theory (with central charge c = −22/5) which possesses only one primary
field with conformal dimension ∆ = −1/5. The bulk perturbation is identified with this
field. In the IR limit, there is only one specie of particles associated with the Faddeev–
Zamolodchikov creation operator “A(θ )” whose scattering is described by the bulk
S-matrix [54]

1 2
A(θ1 )A(θ2 ) = S(θ1 − θ2 )A(θ2 )A(θ1 ) with S(θ ) = − , (40)
3 3
and where we used the notations of the previous sections. In particular, the Lee–Yang
model can be thought as the SG model at coupling constant β̂ 2 = 4/5 (λ = 3/2). Then, the
bulk S-matrix factor (40) can be obtained from the first SG breather calculated in [46] after
projecting out the soliton sector.
On the half-line, there is a single relevant boundary perturbation of the BCFT with
certain conformal boundary conditions [55]. The scattering of the fundamental particle on
the boundary (with creation operator B) is described by the reflection factor R(θ ) which
satisfies
iπ
R(θ )R(−θ ) = 1, R iπ 2 − θ = S(2θ )R 2 + θ ,

R(θ ) = S(2θ )R θ + iπ3 R θ− 3 .
iπ
(41)
Without degrees of freedom at the boundary, the boundary scaling Lee–Yang model has
been studied in details in [55] from both analytical and numerical approach. In case of
boundary degrees of freedom, it is thus natural to expect some changes in the scattering
data. Taking β̂ 2 = 4/5 (λ = 3/2) in the model (1), it is not difficult to show that the
corresponding minimal part of the soliton/antisoliton reflection matrix can be written in
terms of (16) with (17) as10

K(θ )
λ=3/2 = σ3 Ksusy (3θ )σ3 where σ3 = diag(1, −1). (42)
For λ = 3/2 there is only one breather n = 1. Using the boundary bootstrap equation [17,
18], it is straightforward to obtain the first breather reflection amplitude

RB (θ ) = −R0 (u)S (1) (0, u)S (1)(π/2, u)S (1) (η, u)S (1) (iϑ, u)
λ=3/2,
(1) (1)
(43)
10 This solution is obvious as the boundary Yang–Baxter equations are invariant under the simultaneous change
of sign in front of B(θ ) and D(θ ).
where [17]
1
l 1+ l
n n−1
2 1 + 2λ
R0(n) (u) = 3 2λ
3
2λ
,
2 + 2λ
n l 2
l=1 2 + 2λ

x − 1 + n−2l−1
n−1
S (n) (x, u) = xλπ 2 2λ
, (44)
l=0 λπ + 1
2 + n−2l−1
2λ
with the shorthand notation (x) = sin(u/2 + xπ/2)/ sin(u/2 − xπ/2). However, in order
to be a solution of (41) it is necessary to restrict the values of the boundary parameters to
2η/π = ±(b/2 + 2), 2ϑ/π = ±(b/2 + 1). The final result for the reflection factor reads
−1 −1 −1
3 5 4 1+b 1 − b −1 5 + b −1 5 − b
R(θ ) = . (45)
6 6 6 6 6 6 6
Following the notations of [55], for b = 0, 1 and 2 one recovers the minimal solutions
R(2) , R(4) and R(1) of (41), respectively. In particular, in [55] the boundary condition I was
identified with the reflection factor R(1) . In case of fermionic boundary degrees of freedom
added, together with
S(θ − i(b − 2)/6)
R(iπ/2 − θ ) = R(iπ/2 + θ ), (46)
S(θ + i(b − 2)/6)
we conclude that the fixed point b = 2 of the symmetry transformation (46) is associated
with the conformal boundary condition I. Without fermionic boundary degrees of freedom,
it was not the case in [55]. This phenomena clearly needs further investigation which
however goes beyond the scope of this paper.
4.2. The anisotropic Kondo model and its massive extension
In [56] two integrable massive versions of the anisotropic spin 1/2 Kondo model have
been proposed at the reflectionless points. For general values of the coupling β̂, the model
(1) is an other one [16], although boundary degrees of freedom do not belong anymore to
suq (2). From the results of the previous section, we can now consider the Hamiltonian
0
1
HMK = dx (π(x))2 + (∂x φ(x))2 + 8πgµ cos(2φ(x))
4πg
−∞
2
− µB S+ eiφ(0) + S− e−iφ(0) for g = , (47)
2n + 3
where n ∈ N and S± are Pauli matrices. In the limit µ = 0, it corresponds to the usual
anisotropic (at zero voltage) Kondo model. Taking φ0 = 0 and ν± uv
(y) = S± (y) in (5),
the transformation φ ↔ −φ with S+ ↔ S− leaves the Hamiltonian invariant. Then, the
corresponding soliton/antisoliton reflection matrix follows from (23) setting α = 0 and
θ → (2n + 1)θ in (17), i.e., it reads

σ (η, u)σ (iϑ, u)
(n)
R (θ ) = Ksusy (2n + 1)θ α=0 R0 (u)G0 (u) . (48)

cos(η) cosh(ϑ) λ=n+1/2
Notice that the massive deformation for n = 0 in (47) preserves the known N = 2
supersymmetry of the massless Kondo model at that point [57]. Breathers boundary
reflection amplitudes are calculated directly and given by

RB (θ ) = (−1)n R0 (u)S (n) (0, u)S (n) (π/2, u)S (n) (η, u)S (n) (iϑ, u)
λ=1/2+n ,
(n) (n)
(49)
together with (44). It shows that the massive extension proposed in [56] is also integrable
at the special points g = 2/(2n + 3). If one considers a non-zero voltage, an interesting
question would be to check whether the duality g ↔ 1/g [58,59] still exists at these points
in the massive case, or not.
4.3. The boundary critical Ashkin–Teller model
In two dimensions, the Ashkin–Teller model [60] corresponds to two Ising models
coupled by a local four spin interaction. The action associated with the critical line of the
Ashkin–Teller (Z4 ) model in the bulk corresponds to a conformal field theory with central
charge c = 1 (a free massless scalar field φ [61]). This critical line√can be parametrized
by the conformal dimension ∆D = β̂ 2 /2 of the thermal operator D ≡ 2 cos(β̂φ). Then, its
D-perturbation coincides with the sine-Gordon model. The order parameters are the fields
σ , σ † and the field Σ ∼ σ 2 with conformal dimensions ∆ = 1/16 (independently of β̂
[62]) and ∆Σ = β̂ 2 /8, respectively. The field Σ which is local with respect to φ can be
realized in terms of exp(±β̂φ/2). At the special point β̂ 2 = 4/3, the Ashkin–Teller model
enjoys N = 2 supersymmetry.
For generic values of the coupling β̂, a natural boundary version of the Ashkin–Teller
model can be associated with action (1) and (2). At the special points β̂ 2 = 4/(2n + 3), it
admits a two-parameter family of boundary Lagrangean representation of the form
∞ 0
1
AbAT = dy dx (∂ν φ)2 − 2κD(x, y)
8π
−∞ −∞
∞

+ dy ν0 b(y) + ν̄0 b† (y) Σ(y) + Aboundary, (50)
−∞
where ν0 is a complex parameter and the boundary dynamics are described by
∞
d
Aboundary = i b† b dy. (51)
dy
−∞
The factorized scattering theory associated with these two phases follows from the results
of [16] and those presented here. In particular, for n = 0 this model exhibits N = 2
boundary supersymmetry defined in Section 2.
Acknowledgement
P.B. thanks P.E. Dorey, A. Furusaki, C. Mudry, H. Saleur, R. Sasaki, M. Stanishkov and
R. Tateo for discussions and R.I. Nepomechie for comments. P.B.’s work was supported by
JSPS fellowship.
References
[1] J. Maldacena, L. Maoz, JHEP 0212 (2002) 046.

[2] J.G. Russo, A.A. Tseytlin, JHEP 0209 (2002) 035.
[3] A. LeClair, C. Vafa, Nucl. Phys. B 401 (1993) 413.
[4] C. Gomez, A comment on masses, quantum affine symmetries and pp-wave backgrounds, hep-th/0211137.
[5] K. Kobayashi, T. Uematsu, Y. Yu, Nucl. Phys. B 397 (1993) 283.
[6] K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
[7] Y. Hikida, S. Yamaguchi, JHEP 0301 (2003) 072;
U. Lindstrom, M. Zabzine, JHEP 0302 (2003) 006.
[8] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
[9] N.P. Warner, Nucl. Phys. B 450 (1995) 663.
[10] J.L. Cardy, Nucl. Phys. B 240 (1984) 514;
J.L. Cardy, Nucl. Phys. B 275 (1986) 200;
J.L. Cardy, Nucl. Phys. B 324 (1989) 581;
J.L. Cardy, Phys. Lett. B 259 (1991) 274;
J.L. Cardy, D.C. Lewellen, Phys. Lett. B 259 (1991) 274.
[11] A.B. Zamolodchikov, Adv. Stud. Pure Math. 19 (1989) 641.
[12] S. Lukyanov, V.A. Fateev, Int. J. Mod. Phys. A 3 (1988) 507.
[13] V.A. Fateev, A.B. Zamolodchikov, Sov. Phys. JETP 62 (1985) 215;
V.A. Fateev, A.B. Zamolodchikov, Sov. Phys. JETP 63 (1986) 913;
V.A. Fateev, A.B. Zamolodchikov, Nucl. Phys. B 280 (1987) 644;
V.A. Fateev, A.B. Zamolodchikov, Phys. Lett. B 271 (1991) 91.
[14] E.K. Sklyanin, Funct. Anal. Appl. 21 (1987) 164;
E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
[15] P. Baseilhac, G.W. Delius, J. Phys. A 34 (2001) 8259.
[16] P. Baseilhac, K. Koizumi, Nucl. Phys. B 649 (2003) 491.
[17] S. Ghoshal, Int. J. Mod. Phys. A 9 (1994) 4801.
[18] A. Fring, R. Koberle, Int. J. Mod. Phys. A 10 (1995) 739.
[19] P. Fendley, H. Saleur, N.P. Warner, Nucl. Phys. B 430 (1994) 577.
[20] H. Saleur, S. Skorik, N.P. Warner, Nucl. Phys. B 441 (1995) 421.
[21] S. Skorik, H. Saleur, J. Phys. A 28 (1995) 6605.
[22] A. LeClair, G. Mussardo, H. Saleur, S. Skorik, Nucl. Phys. B 453 (1995) 581.
[23] V.A. Fateev, S. Lukyanov, A.B. Zamolodchikov, Al.B. Zamolodchikov, Phys. Lett. B 406 (1997) 83.
[24] P. Dorey, R. Tateo, G. Watts, Phys. Lett. B 448 (1999) 249.
[25] E. Corrigan, Int. J. Mod. Phys. A 13 (1998) 2709.
[26] E. Corrigan, G.W. Delius, J. Phys. A 32 (1999) 8601.
[27] E. Corrigan, A. Taormina, J. Phys. A 33 (2000) 8739.
[28] P. Mattsson, P. Dorey, J. Phys. A 33 (2000) 9065.
[29] Z. Bajnok, L. Palla, G. Takacs, G.Z. Toth, Nucl. Phys. B 622 (2002) 548.
[30] P. Mattsson, Integrable quantum field theories, in the bulk and with a boundary, Ph.D. Thesis, Durham,
hep-th/0111261.
[31] V.A. Fateev, A.B. Zamolodchikov, Al.B. Zamolodchikov, Boundary Liouville field theory. 1. Boundary state
and boundary two point function, hep-th/0001012.
[32] V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312;
V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 (1985) 691.
[33] J. Schulze, Nucl. Phys. B 489 (1997) 580.

[34] S. Kawai, Coulomb-gas approach for boundary conformal field theory, hep-th/0201146.
[35] L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. 13 (1998) 2747.
[36] G.W. Delius, N. MacKay, Commun. Math. Phys. 233 (2003) 173.
[37] R.I. Nepomechie, Lett. Math. Phys. 62 (2002) 83.
(1)
[38] G.W. Delius, A. George, Quantum affine reflection algebras of type dn and reflection matrices,
math.QA/0208043.
[39] V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Nucl. Phys. B 549 (1999) 529.
[40] V.V. Bazhanov, A.N. Hibberd, S.M. Khoroshkin, Nucl. Phys. B 622 (2002) 475.
[41] R.I. Nepomechie, Phys. Lett. B 516 (2001) 161.
[42] R.I. Nepomechie, Phys. Lett. B 516 (2001) 376.
[43] P. Fendley, S. Mathur, C. Vafa, N.P. Warner, Phys. Lett. B 243 (1990) 257;
P. Fendley, W. Lerche, S.D. Mathur, N.P. Warner, Nucl. Phys. B 348 (1991) 66;
P. Mathieu, M.A. Walton, Phys. Lett. B 254 (1991) 106.
[44] D. Bernard, A. LeClair, Commun. Math. Phys. 142 (1991) 99.
[45] P. Fendley, K.A. Intriligator, Nucl. Phys. B 372 (1992) 533.
[46] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1980) 253.
[47] E. Witten, D. Olive, Phys. Lett. B 78 (1978) 97.
[48] H.J. De Vega, A. González-Ruiz, J. Phys. A 27 (1994) 6129.
[49] Al.B. Zamolodchikov, unpublished.
[50] Z. Bajnok, L. Palla, G. Takacs, Nucl. Phys. B 622 (2002) 565.
[51] C. Ahn, C. Kim, C. Rim, M. Stanishkov, Duality in N = 2 super-Liouville theory, hep-th/0210208.
[52] H. Saleur, Lectures on nonperturbative field theory and quantum impurity problems, cond-mat/9812110;
H. Saleur, Lectures on nonperturbative field theory and quantum impurity problems: part 2, cond-
mat/0007309.
[53] P. Fendley, H. Saleur, N.P. Warner, Nucl. Phys. B 430 (1994) 577.
[54] J.L. Cardy, G. Mussardo, Phys. Lett. B 225 (1989) 275.
[55] P. Dorey, A. Pocklington, R. Tateo, G.M.T. Watts, Nucl. Phys. B 525 (1998) 641.
[56] Z.S. Bassi, A. LeClair, Nucl. Phys. B 552 (1999) 643.
[57] P. Fendley, Adv. Theor. Math. Phys. 2 (1998) 987.
[58] V.A. Fateev, P. Wiegmann, Phys. Lett. A 81 (1981) 179.
[59] P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. B 52 (1995) 8934.
[60] J. Ashkin, E. Teller, Phys. Rev. 64 (1943) 178.
[61] L.P. Kadanoff, A.C. Brown, Ann. Phys. 121 (1979) 318.
[62] L.P. Kadanoff, Ann. Phys. 120 (1979) 39.
Free energy of the two-matrix model/dToda

tau-function
M. Bertola a,b
a Department of Mathematics and Statistics, Concordia University, 7141 Sherbrooke W., Montréal, PQ,
H4B 1R6 Canada
b Centre de Recherches Mathématiques, Université de Montréal, C.P. 6128, Succ. Centre Ville, Montréal, PQ,
H3C 3J7 Canada
Received 26 June 2003; received in revised form 22 July 2003; accepted 30 July 2003
Abstract
We provide an integral formula for the free energy of the two-matrix model with polynomial
potentials of arbitrary degree (or formal power series). This is known to coincide with the τ -function
of the dispersionless two-dimensional Toda hierarchy. The formula generalizes the case of conformal
maps of Jordan curves studied by Kostov, Krichever, Mineev-Weinstein, Wiegmann, Zabrodin and
separately Takhtajan. Finally we generalize the formula found in genus zero to the case of spectral
curves of arbitrary genus with certain fixed data.
PACS: 02.30.Jr; 05.90.+m
1. Introduction
Many instances of integrable systems are obtained by means of a suitable limit

(dispersionless or semi-classical) of a statistical theory. The departing point of our analysis
in this paper is the random 2-matrix model [9,21], which is attracting growing attention
due to its applications to solid state physics [15] (e.g., conduction in mesoscopic devices,
quantum chaos and, lately, crystal growth [22]), particle physics [29], 2d-quantum gravity
and string theory [3,11,13]. The model under inspection consists of two Hermitian matrices
E-mail address: bertola@mathstat.concordia.ca (M. Bertola).
doi:10.1016/j.nuclphysb.2003.07.029
436 M. Bertola / Nuclear Physics B 669 [FS] (2003) 435–461
M1 , M2 of size N × N with a probability distribution given by the formula

1 1
dµ(M1 , M2 ) = dM1 dM2 exp − Tr V1 (M1 ) + V2 (M2 ) − M1 M2 ,
ZN h̄
∞
uK ∞
vJ
V1 (x) = xK , V2 (y) = yJ , (1.1)
K J
K=1 J =1
where Vi are formal power series but soon will be restricted to polynomials for simplicity.
The partition function ZN is known to be a τ -function for the KP hierarchy in each
set of deformation parameters (coefficients of V1 or V2 ) and to provide solutions of the
two-Toda hierarchy [1,2,28]. This model has been previously investigated in the series of
paper [5–7] where a duality of spectral curves and differential systems for the relevant
biorthogonal polynomials has been unveiled and analyzed in the case of polynomial
potentials. In [8] the mixed correlation functions of the model (traces of powers of the
two non-commuting matrices) have been reduced to a formal Fredholm-like determinant
without any assumption on the nature of the potentials and using the recursion coefficients
for the biorthogonal polynomials. We briefly recall that the biorthogonal polynomials are
two sequences of monic polynomials ([5] and references therein)
πn (x) = x n + · · · , σn (y) = y n + · · · , n = 0, 1, . . . (1.2)
that are “orthogonal” (better say “dual”) w.r.t. to the coupled measure on the product space

dx dy πn (x)σm (y)e− h̄ (V1 (x)+V2(y)−xy) = hn δmn , hn = 0 ∀n ∈ N,
1
(1.3)
R R
where V1 (x) and V2 (y) are the functions (called potentials) appearing in the two-matrix
model measure (1.1). It is convenient to introduce the associated quasipolynomials defined
by the formulas
1
πn−1 (x)e− h̄ V1 (x) ,
1
ψn (x) := √ (1.4)
hn−1
1
σn−1 (y)e− h̄ V2 (y) .
1
φn (y) := √ (1.5)
hn−1
In terms of these two sequences of quasipolynomials the multiplications by x and y,
respectively, are represented by semi-infinite square matrices Q = [Qij ]i,j ∈N∗ and P =
[Pij ]i,j ∈N∗ according to the formulae

xψn (x) = Qn,m ψm (x), yφn (y) = Pm,n φm (y),
m m
Qn,m = 0 = Pm,n , if n > m + 1. (1.6)
The matrices P and Q have a rich structure and satisfy the “string equation”
[P , Q] = h̄1. (1.7)
We refer for further details to [4–7] where these models are studied especially in the case
of polynomial potentials. We also point out that the model can easily be generalized to
accommodate contours of integration other than the real axes [4,5].
M. Bertola / Nuclear Physics B 669 [FS] (2003) 435–461 437
The partition function is believed to have a large N expansion according to the formula
1 1
− 2
ln ZN = F = F (0) + 2 F (1) + · · · . (1.8)
N N
This expansion in powers of N −2 has been repeatedly advocated for the 2-matrix model on
the basis of physical arguments [13,14] and has been rigorously proven in the one-matrix
model [16]. In the two-matrix model this expansion is believed to generate 2-dimensional
statistical models of surfaces triangulated with ribbon-graphs [11,13,17], where the powers
of N −1 are the Euler characteristics of the surfaces being tessellated. From this point
of view the term F (0) corresponds to a genus 0 tessellation and the next to a genus one
tessellation.
The object of this paper is the leading term of the free energy, F (0) . It is the generating
function of the expectations of the powers of the two matrices in the model
K K
M1 = K∂uK F (0) + O N −2 , M2 = J ∂vJ F (0) + O N −2 . (1.9)
Integral formulas for these partial derivatives at the leading order are known and involve
integrals over a certain spectral curve (see below [18]), however a closed formula for the
function F (0) itself was so far missing; this paper fills the gap (Theorem 2.3). Remarkably,
an algorithm for the computation of the subleading terms is known and also a closed
expression of the genus 1 correction F (1) [14], and therefore this paper precedes logically
and complements [14]. The paper is organized as follows: in Section 2 we recall the
main formulas known in the literature and set up the notation, linking our result with the
relevant other approaches [19,20,23–27,30,31]. The core of the paper is Section 2.1 where
the formula for the leading term of the free energy (dToda tau function) is presented and
proved (Theorem 2.3). In Section 2.2 we apply our result to the generating function of the
canonical change of coordinates represented by the Lax operators of the dToda hierarchy.
Finally in Section 3 we extend the result obtained in Section 2.1 and find an integral
expression for the free energy of the two matrix model in the case the spectral curve is of
arbitrary genus (Theorem 3.1).
2. Planar limit (dToda hierarchy)
In this section we investigate the planar free energy (F (0) in the notation of the
introduction) and we will make soon the common “one-cut” assumption (to be lifted in
Section 3) which amounts to saying that the multiplication operators tend to meromorphic
functions over a spectral curve of genus zero. Indeed in this limit the two multiplication
operators for the wave-vectors defined by the biorthogonal quasipolynomials become
commuting functions in the shift operator here replaced by the variable λ ([7] and
references therein) and they corresponds to the Lax operators of the dToda hierarchy:
∞

Q(λ) = γ λ + αj λ−j , (2.1)
k=0
∞

γ
P (λ) = + βj λj . (2.2)
λ
j =0
The original non-commutativity of the operators now translates to the following Poisson-
bracket which is nothing but the dispersionless form of the string equation (1.7) [23–25]
∂P ∂Q ∂Q ∂P
{P , Q} := λ −λ = 1, (2.3)
∂λ ∂t ∂λ ∂t
where t = h̄N in the large limit. The parameter t is the (scaled) number of eigenvalues of
the matrix model and enters the relations

1 1
P dQ = t, Q dP = t. (2.4)
2iπ 2iπ
The contour chosen is a contour on the physical sheet of the Q (P , respectively) plane
around infinity (i.e., around λ = ∞, λ = 0, respectively).
The deformation equations describe the infinitesimal variations of the operators P , Q
under variations of the parameters of the potentials; they are known in the finite N regime
as well ([5,7] and references therein) whereas in the dispersionless limit are given by the
evolution equations [23–25,27]

(∂uK Q)λ = Q, QK +0 , (∂vJ Q)λ = Q, P J −0 , (2.5)

(∂uK P )λ = P , QK +0 , (∂vJ P )λ = P , P J −0 , (2.6)
where the subscript ±0 denotes the positive (negative) part of the Laurent polynomial plus
half the part constant in λ, viz., e.g.,
1
QK +0
(λ) = QK + (λ) + QK 0 . (2.7)
2
If the potentials Vi are polynomials of degrees di + 1, i = 1, 2 then both P and Q are
finite Laurent polynomials

d2
Q(λ) = γ λ + αj λ−j , (2.8)
k=0
γ 1 d
P (λ) = + βj λj . (2.9)
λ
j =0
In what follows we restrict to this case so as to avoid complication of convergence; however

one may replace the contour integrals that will follow with formal residues of formal
Laurent series and carry out the same computations.
Another reason why we prefer the truncated setting is that then the two functions
P , Q define a (singular) spectral curve of genus g = 0 which is given by the polynomial
locus (resultant-like) obtained from the determinant of the following Sylvester matrix ([6],
thanks to a remark by J. Hurtubise)
0 = E(P , Q)
 
γ β0 − P β1 ··· ··· βd 1 0 0 0
 0 γ β0 − P β1 ··· ··· βd 1 0 0 
 
 0 0 γ β0 − P β1 ··· ··· βd 1 0 
 
 0 0 0 γ β0 − P β1 ··· ··· βd 1 
1  
= det 
 αd2 ··· α1 α0 − Q γ 0 0 0 0  .
γ d1 +d2  0
 αd2 ··· α1 α0 − Q γ 0 0 0  
 0 ··· α1 α0 − Q 0 
 0 αd2 γ 0 
 0 0 0 αd2 ··· α1 α0 − Q γ 0 
0 0 0 0 αd2 ··· α1 α0 − Q γ
(2.10)
This spectral curve is precisely the (limit of the) spectral curve of the four finite-
dimensional folded differential systems for the quasipolynomials [5,7]. If we worked with
formal power series it is not known whether a spectral curve in this sense can be defined.
Taking this viewpoint λ ∈ CP 1 is the uniformizing parameter of the spectral curve
E(P , Q). The two potentials are related to the parameters γ , αj , βj by the relations
t
P = V1 (Q) − + O Q−2 , near ∞Q , (2.11)
Q
t
Q = V2 (P ) − + O P −2 , near ∞P , (2.12)
P
where the point ∞Q (∞P ) is the point on the spectral curve where Q (P , respectively)
has a simple pole; in the uniformization provided by the coordinate λ it corresponds to
λ = ∞ (λ = 0, respectively). By expanding both sides in powers of λ and matching the
coefficients one can realize that the coefficients of the two potentials are rational functions
of the parameters γ , αj , βj . More explicitly we have
1 +1
d d2 +1
uK K vJ J
V1 (q) = q , V2 (p) = p , (2.13)
K J
K=1 J =1

1 P 1 Q
uK = − dQ, vJ = − dP . (2.14)
2iπ QK 2iπ PJ
The leading term of the free energy of the model is then defined by the differential
equations

∂F 1 1
= UK := P QK dQ = res P QK dQ,
∂uK K2iπ K ∞Q

∂F 1 1
= VJ := QP J dP = res QP J dP . (2.15)
∂vJ J 2iπ J ∞P
These equations are precisely the same that define the τ -function of the dToda hierarchy.
In the relevant literature [23–25,27] the functions P , Q are the Lax operators denoted by
L,
L or L, L and the normalization is slightly different.
We should also remark the following link to the works [19,20,26,30,31] in that if we take
V1 = V = V 2 we then have γ ∈ R, αj = β̄j . The function Q(λ) is then the uniformizing
map of a Jordan curve Γ in the Q-plane (at least for suitable ranges of the parameters)
which is defined by either of the following relations
P (1/λ) = Q(λ̄), |λ| = 1. (2.16)

In the setting of [19,20,30] the function Q is denoted by z (and λ by w) so that then P is
nothing but the Schwartz function of the curve Γ , defined by
z̄ = S(z), z ∈ Γ. (2.17)
tk
The coefficients of the potential V (x) = k k x k are the so-called “exterior harmonic
moments” of the region D enclosed by the curve Γ

1
tK = z̄z−K dz, (2.18)
2iπ
Γ

1 1 Area(D)
t = t0 = dz ∧ dz̄ = z̄ dz = . (2.19)
2iπ 2iπ π
D Γ
By writing z̄ = S(z(w)) these integrals become residues in the w-plane. In our general
situation the representation of the coefficients uK , vJ (2.14) cannot be translated to a
surface integral, hence the need for a separate analysis. For conformal maps (i.e., Jordan
curves) the τ -function has been defined in [26] and given an appealing interpretation
as (exponential of the Legendre transform of) the electrostatic potential of a uniform 2-
dimensional distribution of charge in D [19]

1 1 1 2 2
ln(τΓ ) = −
ln − d z d z . (2.20)
π z z
D
It can be rewritten as a (formal) series in the exterior and interior moments as

1
2 ln(τΓ ) = − d2 z |z|2 + t0 w0 + (tK wK + t¯K w̄K ), (2.21)
4π
D K>0
where the interior moments are defined by (the normalization here differs slightly
from [30])

1
wK = zK d2 z, K > 0, (2.22)
πK
D

1
w0 = ln|z|2 d2 z. (2.23)
π
D
Unfortunately this formula is not exportable to our more general setting (in particular the
logarithmic moment above) and to the general setting of the dispersionless Toda hierarchy;
our principal objective is to fill this gap.
2.1. Free energy of the 2-matrix model in the planar limit
Let us focus on the function P as a (multivalued) function of Q (similar argument can

be reversed for Q as a function of P ); on the physical sheet P (Q) will have in general
some branch-cuts with square-root singularities at the endpoint of each cut. These cuts are
bounded in the physical sheet because P (Q) is analytic in a neighborhood of ∞Q . Note
that

1 q
V1 (q) = ln 1 − P dQ, (2.24)
2iπ Q

1 p
V2 (p) = ln 1 − Q dP , (2.25)
2iπ P
as one can immediately realize by expanding in powers of q, p. The integrals are well
defined provided that q (p, respectively) are kept inside the contour of integration. In the
following we will develop all the necessary arguments only for V1 (q) and related objects,
where the reader can obtain the relevant proofs for V2 (p) by interchanging the role of P
and Q.
Let us now introduce the exterior potentials1

1 Q
Φ1 (qout ) = − ln 1 − P dQ. (2.26)
2iπ qout
In this formula the contour of integration is such as to leave the point qout in the outside
region (whence the subscript). In general we will distinguish the choice of the contours by
a subscript qout or qin in what follows.
The coefficients of Φ1 (qout ) in inverse powers of qout are precisely (minus) the UK
defined in (2.15). Note that expanding Eq. (2.26) in inverse powers of qout the first d1 + 1
coefficients are the UK coefficients as defined in (2.15).
The first objective is to analytically continue Φ1 to the physical sheet so as to obtain a
different representation of it. For this purpose we compute

1 Q P dQ 1 1
Φ1 (qout ) = P dQ = + P dQ
2iπ qout(Q − qout) qout 2iπ 2iπ (Q − qout )

t 1 1 t
= + P dQ + P (q) = − V1 (q) + P (q), (2.27)
q 2iπ Q − qin q
where we have dropped the subscript outside of the integral as those terms are analytic
functions in the whole physical sheet. Integrating once we obtain

Φ1 (Q) = −V1 (Q) + t ln(Q) + P dQ, (2.28)
Xq
where Xq is a point defined implicitly by the requirement Φ1 = O(Q−1 ) near ∞Q (≡ (λ =

∞)). Note that, in spite of the ln(Q) term, this is an analytic function around ∞Q .
1 It has been pointed out to me by B. Eynard that they are sometimes referred to as effective potentials because
they can be thought of as the effective potential felt by one eigenvalue in the field of the others.
By similar reasoning we get

1 P
Φ2 (P ) = − ln 1 − Q dP = −V2 (P ) + t ln(P ) + Q dP . (2.29)
2iπ P
Xp
We need to introduce two more points (beside Xq and Xp ) on the spectral curve and a
lemma: those are the points Λq , Λp defined implicitly by the relations

P dQ = V1 Q(λ) >0 + t ln(λ) + O λ−1 , near ∞Q , (2.30)
Λq

Q dP = V2 P (λ) <0 + t ln(λ) + O(λ), near ∞P . (2.31)
Λp
By a simple inspection of the λ0 -coefficient in the LHS and RHS one finds that
Λq

P dQ = P dQ + P dQ = P dQ + V1 (Q) 0 − t ln(γ ), (2.32)
Xq Λq Xq Λq
Λp

Q dP = Q dP + Q dP = Q dP + V2 (P ) 0 − t ln(γ ), (2.33)
Xp Λp Xp Λp
where the subscript 0 denotes the constant part in λ (which can be written as a residue).
We now have the
Lemma 2.1. The following relation holds

Xq Xp
µ : = Q(Xp )P (Xp ) + P dQ = Q(Xq )P (Xq ) + Q dP
Xp Xq

= P Q − V1 (Q) − V2 (P ) 0 + 2t ln(γ ). (2.34)
Remark 2.2. The quantity µ will be proved in Corollary 2.8 to be the derivative of the
free energy w.r.t. t, i.e., the chemical potential. It therefore corresponds to the logarithmic
moment in the setting of [19,20,30,31].
Proof. The equivalence of the two integrals is immediate by an integration by parts.

Integration by parts is in fact the key to prove also the last part:

P dQ = P Q − P (Λq )Q(Λq ) − Q dP
Λq Λq
Λp
= P Q − P (Λq )Q(Λq ) − Q dP − Q dP . (2.35)
Λq Λp
Looking at the constant part in λ in both sides of the above equation we conclude that
Λp
(QP )0 = P (Λq )Q(Λq ) + Q dP . (2.36)
Λq
Therefore we have
Λp Λp Xp
(QP )0 = P (Λq )Q(Λq ) + Q dP = P (Λq )Q(Λq ) + Q dP + Q dP
Λq Xp Λq
Λp Xp
= P (Λq )Q(Λq ) + Q dP + Q(Xp )P (Xp ) − Q(Λq )P (Λq ) − P dQ
Xp Λq
Λp Xp Xq

= Q dP + Q(Xp )P (Xp ) − P dQ − P dQ
Xp Xq Λq
Xq

= V1 (Q) + V2 (P ) 0
− 2t ln(γ ) + Q(Xp )P (Xp ) + P dQ. (2.37)
Xp
This concludes the proof of the lemma. ✷
We are now in a position to formulate the first main result of this paper.
Theorem 2.3. The free energy is given by the formula (up to constant)
1
2F = res(P Φ1 dQ) + res (QΦ2 dP ) + res P 2 Q dQ
∞Q ∞P 2 ∞Q

V1 (Q) + V2 (P ) − P Q
+ t res dλ + 2t 2 ln(γ ) (2.38)
λ=∞ λ
1
= u K UK + vJ VJ + res P 2 Q dQ
2 ∞Q
K J

V1 (Q) + V2 (P ) − P Q
+ t res dλ + 2t 2 ln(γ ). (2.39)
λ=∞ λ
Before proceeding to the proof a corollary and some remarks are in order. First off from
the expressions in Theorem 2.3 we find the well-known scaling property of the free energy
[10,19,31]
Corollary 2.4. The free energy defined in Theorem 2.3 satisfies the scaling equation

4F = −t 2 + 2t∂t + (2 − K)uK ∂uK + (2 − J )vJ ∂vJ F . (2.40)
K J
(More general scaling equations will be introduced later in Corollary 3.7).
Proof. The proof in the context of the normal matrix model can be found in the references
quoted above and, in view of the formal equivalence of the normal matrix model with the
two-matrix model [19], the statement would follow also in our case.
In order to be self contained we rederive this property in the present context; one way
of proving formula (2.40) is from the expression (2.39) for F given in Theorem 2.3, by
computing the residues involved after rewriting symmetrically the term res∞Q P 2 Q dQ as
2 (res∞Q P Q dQ + res∞P Q P dP ) and computing it. A second, possibly instructive way
1 2 2
to derive it also directly from Eq. (2.39) is by using the scaling properties of the various
quantities involved. To this purpose we introduce the rescaling according to the formulæ

Q = δ Q, .
P = δP (2.41)
Under this change of the functions Q, P we have (using formulas (2.4), (2.15))
uK = δ 2−K ũK , vJ = δ 2−J ṽJ , t = δ 2 t˜. (2.42)
Moreover, computing explicitly the residue resλ=∞ P Q dλ we find that

γ 2 = −t + j αj βj , (2.43)
j
from which one immediately obtains that γ = δ γ̃ . The exterior potentials are also
conformally invariant, indeed

Φ1 = −V1 (Q) + t ln(Q) + P dQ
Xq

Xq
= −δ 2 V + δ 2 t˜ ln(Q)
1 (Q) + P dQ
+ δ 2 t˜ ln(δ) + P dQ

q
X Xq

Xq
1
= δ2Φ 2˜
+ δ t ln(δ) + P dQ.
(2.44)
Xq
The last two terms cancel each other out because both the LHS and RHS must be
O(Q−1 ) = O(Q−1 ). Repeating the argument for Φ2 we finally get
1 ,
Φ1 = δ 2 Φ 2 .
Φ2 = δ 2 Φ (2.45)
Plugging the relations (2.40)–(2.42), (2.45) into the expression (2.39) for the free energy
we obtain
+ δ 4 t˜ 2 ln δ = δ 4 F
F = δ4 F + t 2 ln(δ). (2.46)
Applying the operator δ∂δ |δ=1 to both sides we obtain

0 = 4F − (2 − K)uK ∂uK + (2 − J )vJ ∂vJ + 2t∂t F + t 2 , (2.47)
K J
whence the statement of the corollary. ✷
We also add a few remarks before moving on with the proof of Theorem 2.3.
Remark 2.5. Formulas (2.38), (2.39) are symmetric in the roles of P and Q: the only
non-immediately symmetric term is res∞Q P 2 Q dQ but an integration by parts restores
the symmetry.
Remark 2.6. The formula is derived for polynomial potentials but it could possibly be
extended to convergent or formal series.
Remark 2.7. The genus 1 correction to the above formula has been computed in [14] and
is given by
1 4
F (1) = − ln γ D , (2.48)
24
1
D=
γ d1 +d2 +2
 
−γ 0 β1 ··· · · · d 1 βd 1 0 0 0
 0 −γ 0 β1 ··· · · · d 1 βd 1 0 0 
 
 0 0 −γ 0 β1 ··· · · · d 1 βd 1 0 
 
 0 0 0 −γ 0 β1 ··· · · · d 1 βd 1 
 
× det 
 d2 αd2 · · · α1 0 −γ 0 0 0 0   . (2.49)
 0 d α · · · −γ 0 
 2 d2 α1 0 0 0 
 0 0 d2 αd2 · · · −γ 0 
 α1 0 0 
 0 0 0 d2 αd2 ··· α 0 −γ 0 
1
0 0 0 0 d2 αd2 ··· α1 0 −γ
Proof of Theorem 2.3. The equivalence of lines (2.38) and (2.39) follows from the
definition of the exterior potentials and Lemma 2.1.
Let us consider the derivative ∂K := ∂u∂K done at argument (P or Q) fixed, where the
value being kept fixed under differentiation is denoted by the corresponding subscript

QK
(∂K Φ1 )Q = − + (∂K P )Q dQ − P (Xq )∂K Q(Xq ) , (2.50)
K
Xq

(∂K Φ2 )P = (∂K Q)P dP − Q(Xp )∂K P (Xp ) . (2.51)
Xp
Let us set

1
4iπF0 := P Φ1 (Q) dQ + QΦ2 (P ) dP + P 2 Q dQ, (2.52)
2
∞Q ∞P ∞Q
and study the variation of this functional. First off we have the formulas
∞

t
P = V1 (Q) − + KUK Q−K−1 , (2.53)
Q
K=1
∞

Φ1 = − UK Q−K , (2.54)
K=1
∞
t
Q = V2 (P ) − + J VJ P −J −1 , (2.55)
P
J =1
∞

Φ2 = − VJ P −J , (2.56)
J =1
where the UK , VJ ’s have been defined in (2.15). We also need the variations of the
functions P , Q given here below

(∂K P )Q = QK−1 + O Q−2 , (∂K Q)P = O P −2 . (2.57)
With these formulæ we can now compute the variation of F0 (the subscript on the loop
integrals that follow specify the points around which we circulate):
=2iπUK =5 =0

4iπ∂K F0 = (∂K P )Q Φ1 dQ + P (∂K Φ1 )Q dQ + (∂K Q)P Φ2 dP
∞Q ∞Q ∞P

+ Q(∂K Φ2 )P dP + P (∂K P )Q Q dQ
∞P ∞Q

= 4iπUk + P (∂k P )Q dQ dQ + Q (∂k Q)P dP dP
∞Q Xq ∞P Xp

+ P (∂K P )Q Q dQ − 2iπt P (Xq )∂K Q(Xq ) + Q(Xp )∂K P (Xp )
∞Q

= 4iπUk + P (∂k P )Q dQ dQ + Q (∂k Q)P dP dP
∞Q Xq ∞P Xq

− P (∂k P )Q dQ dQ − Q (∂k P )Q dQ dP
∞Q Xq ∞Q Xq
Xq

− 2iπt P (Xq )∂K Q(Xq ) + Q(Xp )∂K P (Xp ) − (∂K Q)P dP ,
Xp
(2.58)
where

5 = 2iπUk + P (∂k P )Q dQ dQ − 2iπtP (Xq )∂K Q(Xq ) . (2.59)
∞Q Xq
We now use the “thermodynamic identity”
(∂K P )Q dQ = −(∂K Q)P dP , (2.60)

so that the last term on sixth line of Eq. (2.58) reads (notice the double change of sign
due to the definition of the circle around ∞Q , which is (homologically) minus the circle
around ∞P )

Q (∂k P )Q dQ dP = Q (∂k Q)P dP dP . (2.61)
∞Q Xq ∞P Xq
Plugging into the variation of F0 we get

4iπ∂K F0 = 4iπUK − 2iπt P (Xq )∂K Q(Xq ) + Q(Xp )∂K P (Xp )
Xq
− (∂K Q)P dP . (2.62)
Xp
Finally we claim that the term in the square brackets in (2.62) is

Xq

P (Xq )∂ Q(Xq ) + Q(Xp )∂ P (Xp ) − (∂Q)P dP
Xp
Xq
= ∂ Q(Xp )P (Xp ) + P dQ
Xp

= ∂ P Q − V1 (Q) − V2 (P ) 0 + 2t ln(γ ) , (2.63)
where ∂ denotes any vector field in the space of parameters uK , vJ (or even t). Indeed
Xq
∂ Q(Xp )P (Xp ) + P dQ
Xp
Xq

= ∂ Q(Xp )P (Xp ) + P (λ)Q (λ) dλ
Xp

= ∂ Q(Xp ) P (Xp ) + Q(Xp )∂ P (Xp )
Xq

+ (∂P )λ (λ)Q (λ) + P (λ)(∂Q )λ (λ) dλ
Xp
− P (Xp )Q (Xp )∂Xp + P (Xq )Q (Xq )∂Xq

= ∂ Q(Xp ) P (Xp ) + Q(Xp )∂ P (Xp )
Xq

+ (∂P )λ (λ)Q (λ) − P (λ)(∂Q)λ (λ) dλ + P (Xq )∂(Q)λ (Xq )
Xp
− P (Xp )∂(Q)λ (Xp ) − P (Xp )Q (Xp )∂Xp + P (Xq )Q (Xq )∂Xq . (2.64)
In order to proceed we note that if X is a point depending on the parameters, we have

∂ Q(X) = (∂Q)λ (X) + Q (X)∂X. (2.65)
Using this formula we get

(2.64) = Q(Xp )∂ P (Xp ) + ∂ Q(Xp ) P (Xp )
Xq

+ (∂P )λ (λ)Q (λ) − P (λ)(∂Q)λ (λ) dλ
Xp
Xq

= Q(Xp )∂ P (Xp ) + ∂ Q(Xp ) P (Xp ) + (∂P )Q dQ, (2.66)
Xp
which is the term in square brackets in (2.62). Using then Lemma 2.1 we have proven that

2∂K F0 = 2UK − t∂K −V1 (Q) − V2 (P ) + QP 0 + 2t ln(γ )
= 2UK − t∂K µ. (2.67)
The second term is exactly the opposite of the variation of the last two terms in formula
(2.39), and this concludes the proof (the derivatives w.r.t. vJ are treated by interchanging
the roles of P and Q). ✷
The derivative w.r.t. the number operator t has to be treated in a separate way.
Corollary 2.8. The derivative of the free energy w.r.t. t is given by the formula
Xq
∂t F = Q(Xp )P (Xp ) + Q dP = µ. (2.68)
Xp
Proof. We start with the following observation

dλ
(∂t P )Q dQ = −(∂t Q)P dP = − . (2.69)
λ
In other words they are differential of the third kind with poles at ∞Q and ∞P and opposite
residues, to wit (using the thermodynamical identity (2.60))
1 1
− dP + O P −2 = (∂t Q)P dP = −(∂t P )Q dQ = dQ + O Q−2 (2.70)
P Q
and the differentials have no other singularities. We then proceed as in the proof of
Theorem 2.3

(∂t Φ1 )Q = ln(Q) + (∂t P )Q dQ − P (Xq )∂t Q(Xq ) , (2.71)
Xq

(∂t Φ2 )P = ln(P ) + (∂t Q)P dP − Q(Xp )∂t P (Xp )
Xp
Xq

= ln(P ) + (∂t Q)P dP + (∂t Q)P dP − Q(Xp )∂t P (Xp )
Xq Xp
Xq

= ln(P ) + (∂t Q)P dP − (∂t P )Q dQ − Q(Xp )∂t P (Xp ) , (2.72)
Xq Xp
1
(∂t P )Q = − + O Q−2 , (2.73)
Q
1
(∂t Q)P = − + O P −2 . (2.74)
P
We then find that
=− dλ/λ dλ/λ
=

4iπ∂t F0 = P ln(Q) + (∂t P )Q dQ dQ + Q ln(P ) + (∂t Q)P dP dP
∞Q Xq ∞P Xq
Xq = dλ/λ

− 2iπt∂t Q(Xp )P (Xp ) + Q dP + P Q (∂t P )Q dQ
Xp ∞Q
Xq
= −2iπt∂t Q(Xp )P (Xp ) + Q dP − 2iπ(P Q)0
Xp

+ P ln(Q) − ln(λ) dQ + Q ln(P ) + ln(λ) dP . (2.75)
∞Q ∞P
We now claim that

P ln(Q) − ln(λ) dQ = V1 (Q)0 − t ln(γ ), (2.76)
∞Q
and the symmetric formula for the other term. Indeed

Q
ln = ln γ + O Q−1 , near ∞Q , (2.77)
λ
and thus

P ln(Q) − ln(λ) dQ
∞Q

t −2
= V1 (Q) − + O Q ln(Q) − ln(λ) dQ
Q
∞Q

= V1 (Q) + O Q−2 ln(Q) − ln(λ) dQ + 2iπt ln(γ )
∞Q

dQ dλ
=− V1 (Q) + O Q−1 − + 2iπt ln(γ )
Q λ
∞Q

= −2iπ V1 (Q) 0 + 2iπt ln(γ ). (2.78)
Plugging this into (2.75) we obtain
Xq Xq
2∂t F0 = −t∂t Q(Xp )P (Xp ) + Q dP + Q(Xp )P (Xp ) + Q dP (2.79)
Xp Xp
so that
Xq
2∂t F = 2 Q(Xp )P (Xp ) + Q dP . (2.80)
Xp
2.2. Canonical transformations
In the Poisson structure (2.3) the coordinates ln(λ) and t are canonically conjugate, as
well as the functions P , Q. Therefore it makes sense to find the generating function of these
transformations. This was accomplished in the context of conformal maps in [30] but it is
probably a new result in this context, since now we can express explicitly the generating
function as integrals of the dToda operators P , Q. I recall that we are looking for a function
Ω(Q, t) with the property that
dQ,t Ω = P dQ + ln(λ) dt. (2.81)

The following proposition gives a representation of such function
Proposition 2.9. The generating function of the canonical change of coordinates

(ln(λ), t) → (P , Q) is given by
Xq
µ 1
Ω = P dQ − = P dQ − Q(Xp )P (Xp ) + P dQ
2 2
Xq Xq Xp

1
= P dQ + V1 (Q) + V2 (P ) − P Q 0 − t ln(γ ) (2.82)
2
Xq
is the generating function of the canonical transformation (ln(λ), t) → (P , Q), or,

dQ,t Ω = ∂Q Ω1 dQ + (∂t Ω)Q dt = P dQ + ln(λ) dt. (2.83)
Remark 2.10. The function that we get by interchanging the rôle of P and Q would
generate the transformation (− ln(λ), t) → (Q, P ) (or the anti canonical one).
Proof. The first part is obvious

∂Q Ω = P by definition. (2.84)
As for the second we compute

1 −1
(∂t Ω)Q = ∂t V1 (Q) + t ln(Q) − µ + O Q
2 Q
1 −1 1
= ln(Q) − ∂t µ + O Q = ln(λ) + ln(γ ) − ∂t µ + O λ−1 . (2.85)
2 2
It is well known in the theory of the dToda tau-function that
∂t2 F = ∂t µ = 2 ln(γ ), (2.86)
and hence the constant term in the above expression vanishes. On the other hand

dλ
(∂t Ω)Q = (∂t P )Q dQ − P (Xq )∂t Q(Xq ) = − P (Xq )∂t Q(Xq )
λ
Xq Xq

= ln(λ) − ln λ(Xq ) − P (Xq )∂t Q(Xq ) . (2.87)
By comparison we conclude not only that ∂t Ω = ln(λ) but also

ln λ(Xq ) = −P (Xq )∂t Q(Xq ) . (2.88)
The proof of the lemma is complete. ✷
3. Planar free energy for spectral curves of arbitrary genus
The situation in the case the spectral curve is of higher genus is only slightly different.
We will work with the following data: a (smooth) curve Σg of genus g is assigned with
two marked points ∞Q , ∞P . On the curve we are given two functions P and Q which
have the following pole structure:
(1) the function Q has a simple pole at ∞Q and a pole of degree d2 at ∞P ;

(2) the function P has a simple pole at ∞P and a pole of degree d1 at ∞Q .
From these data it would follow that P , Q satisfy a polynomial relation but we will not
need it for our computations. By their definition we have
1 +1
d
t t
P= uK QK−1 − + O Q−2 =: V1 (Q) − + O Q−2 , near ∞Q , (3.1)
Q Q
K=1
2 +1
d
t t
Q= vJ P K−1 − + O P −2 =: V2 (P ) − + O P −2 , near ∞P . (3.2)
P P
J =1
The fact that the coefficient of the power Q−1 or P −1 is the same follows immediately from
computing the sum of the residues of P dQ (or Q dP ). The formulas for the coefficients
uK , vJ , t are the same as in the genus zero case, viz.
uK = − res P Q−K dQ, vJ = − res QP −J dP ,
∞Q ∞P
t = res P dQ = res Q dP . (3.3)
∞Q ∞P
Note that the requirement that the curve possesses two meromorphic functions with this
pole structure imposes strong constraints on the moduli of the curve itself. In fact a
Riemann–Roch argument shows that the moduli space of these data is of dimension
d1 + d2 + 3 + g; so far our data show only (d1 + 1) + (d2 + 1) + 1 parameters and therefore
we add as parameters the following period integrals which, in the matrix-model literature
are referred to as filling fractions

1
8i := P dQ, i = 1, . . . , g. (3.4)
2iπ
ai
Here we have introduced a symplectic basis {ai , bi }i=1,...,g in the homology of the curve
Σg and the choice of the a-cycles over the b-cycles is purely conventional.
In this extended setting the free energy is defined by the equations
1 1
∂uK Fg = res P QK dQ, ∂vJ Fg = res QP J dP , (3.5)
K ∞Q J ∞P

∂8i Fg = P dQ =: Γi . (3.6)
bj
Once more we introduce the exterior potentials by the same requirements as in the genus 0
case:

1 Q
Φ1 = − ln 1 − P dQ
2iπ Q

= −V1 (Q) + t ln(Q) + P dQ = O Q−1 , near ∞Q , (3.7)
Xq

1 P dP
Φ2 = − ln 1 − Q
2iπ P

= −V2 (P ) + t ln(P ) + Q dP = O P −1 , near ∞P . (3.8)
Xp
In the loop integral formulas the contours wind around the marked points so as to leave
1. As in the
the point where the potentials are computed inside, i.e., for instance, Q/Q
genus 0 case the two points Xp , Xq are implicitly defined by the requirement O (local
parameter). Recall the definition of the “chemical” potential
Xq
µ = Q(Xp )P (Xp ) + P dQ. (3.9)
Xp
Theorem 3.1. The free energy over the curve Σg is given by
1 g
2Fg = res P Φ1 dQ + res QΦ2 dP + res P 2 Q dQ + tµ + 8i Γi . (3.10)
∞Q ∞P 2 ∞Q
i=1
Proof. It is essentially the same as in the genus zero case. We start from the same
expression F0 used in the proof there (2.52) and proceed to the variation w.r.t. uK .
Following the same steps we obtain the expression

4iπ∂K F0 = 4iπUk + P (∂k P )Q dQ dQ + Q (∂k Q)P dP dP
∞Q Xq ∞P Xq

+ QP d (∂k P )Q dQ
∞Q Xq
Xq

− 2iπt P (Xq )∂K Q(Xq ) + Q(Xp )∂K P (Xp ) − (∂K Q)P dP ,
Xp

=∂K µ
(3.11)
where the subscripts in the contour integrals specify the point around which we are
circulating. The integration by parts of the third term gives

4iπ∂K F0 = 4iπUk + P (∂k P )Q dQ dQ + Q (∂k Q)P dP dP
∞Q Xq ∞P Xq

− P (∂k P )Q dQ dQ − Q (∂k P )Q dQ dP − 2iπt∂K µ
∞Q Xq ∞Q Xq

= 4iπUk + Q (∂k Q)P dP dP + Q (∂k Q)P dP dP
∞P Xq ∞Q Xq
− 2iπt∂K µ, (3.12)
where we have used the thermodynamical identity (2.60). Due to the genus of the curve the
two contours are not homologically opposite and the sum of the two integrals gives finally
(using the Riemann bilinear identity which we recall in Appendix A)

g
4iπ∂K F0 = 4iπUk − 2iπ P dQ (∂K P )Q dQ − P dQ (∂K P )Q dQ
i=1 ai bi bi ai
− 2iπt∂K µ

g

= 4iπUk − 2iπ 8i ∂K Γi − Γi ∂K 8i − 2iπt∂K µ

i=1 =0

g
= 4iπUk − 2iπ 8i ∂K Γi − 2iπt∂K µ. (3.13)
i=1
This implies promptly that

2∂K Fg = 2∂K F0 + 8i ∂K Γi + t∂K µ = 2UK , (3.14)
i
as desired. The check for the variation w.r.t. the filling fractions is the same by noticing
that

(∂8i P )Q = O Q−2 , (∂8i Q)P = O P −2 , (3.15)
near the corresponding marked points. Moreover, by definition

(∂8j P )Q dQ = δij . (3.16)
aj
From this, following the same formal steps and using the Riemann bilinear identity one
easily finds

g
2∂8i F0 = Γi − 8j ∂8i Γj − t∂8i µ, (3.17)
j =1

g
2∂8i Fg = 2∂8i F0 + t∂8i µ + Γi + 8j ∂8i Γj = 2Γi . (3.18)
j =1
This concludes the proof. ✷
Remark 3.2. In the two-matrix model setting the moduli of the spectral curve are fixed
uniquely in terms of the potentials V1 , V2 by the requirement Γi = 0, i = 1, . . . , g (which
is a requirement of stationary point).
Remark 3.3. The case of the one matrix model is a subcase of this setting where one of the
two potentials is quadratic, say V2 ; in this case the spectral curve is hyperelliptic of genus
at most [(d1 + 2)/2].
As for the previous case some special care must be paid for the derivative w.r.t. t, but
the result is the same as in the genus zero case and it is contained in the next corollary.
Corollary 3.4. The chemical potential µ is indeed the derivative ∂t Fg . Moreover, we have
the formula

µ = res V1 (Q) − t ln(Q/λ) dS − res V2 (P ) − t ln(P λ) dS
∞Q ∞P

g
− res P Q dS + 8i dS, (3.19)
∞Q
i=1 bi
where dS is the normalized differential of the third kind with poles at ∞P ,Q and residues
±1 and the function λ is the following function (defined up to a multiplicative constant)
on the universal covering of the curve with a simple zero at ∞Q and a simple pole at ∞P

λ := exp dS . (3.20)
Proof. The proof is quite more involved than in genus zero and requires some preparation.
We introduce the normalized differential of the third kind with simple poles at ∞P ,Q
and residues ±1,

dS = dS∞Q ,∞P , dS = 0, i = 1, . . . , g,
ai
res dS = −1 = − res dS. (3.21)
∞Q ∞P
This provides a coordinate on the (covering of the) curve as follows

λ := e dS
. (3.22)
Quite clearly the parameter λ is defined up to a multiplicative constant and it is
multiplicatively multivalued on the curve Σg (around the b-cycles). This arbitrariness
and multivaluedness will not affect our computations because formula (3.21) remains
unchanged under the rescaling λ → δλ. Moreover, it has no branch-points at ∞P ,Q , where
instead it has a simple zero and a simple pole (and no other zeroes or singularities). By the
definition then
dS = d ln(λ). (3.23)
The rest mimics the genus zero case. We introduce the two points Λp,q such that
! "
res −V1 (Q) + t ln(λ) + P dQ dS = 0, (3.24)
∞Q
Λq
! "
res −V2 (P ) − t ln(λ) + Q dP dS = 0. (3.25)
∞P
Λp
From the definition of the points Xp,q and Λp,q and following the same steps that were
taken in genus 0 we have
Xq

P dQ = res V1 (Q) − t ln(Q/λ) dS, (3.26)
∞Q
Λq
Xp

Q dP = − res V2 (P ) − t ln(P /λ) dS. (3.27)
∞P
Λp
The next relation is different from the genus zero case:
Λp
P (Λq )Q(Λq ) + Q dP = − res P Q dS + 8i dS. (3.28)
∞Q
Λq i=0 bi
Indeed we have
! "
0= P dQ − t ln λ dS
∞Q Λq
Λp
= QP − Q(Λq )P (Λq ) − Q dP − t ln(λ) − Q dP dS
∞Q Λq Λp
Λp

= 2iπ res QP dS + 2iπ Q(Λq )P (Λq ) + Q dP
∞Q
Λq
! "

g
− Q dP − t ln λ dS + 8i dS, (3.29)
∞P Λp i=1 bi

=0

where we have used the bilinear Riemann identity as well as the fact that ai dS = 0 and
the fact that (by definition of the point Λp ) the term with the under-brace is residue-free at
∞P .2 We can now proceed as in Lemma 2.1 and obtain the formula

Xq
µ = Q(Xp )P (Xp ) + P dQ
Xp

= res V1 (Q) − t ln(Q/λ) dS − res V2 (P ) − t ln(P λ) dS
∞Q ∞P

g
− res P Q dS + 8i dS. (3.30)
∞Q
i=1 bi
The formula (3.30) is the equivalent in higher genus of the formula in Lemma 2.1.
Coming back to the proof of the corollary we note as in Corollary 2.8 that
(∂t P )Q dQ = −(∂t Q)P dP = dS (3.31)
is the normalized Abelian differential of the third kind with poles at the marked points for
we have

0 = ∂t 8i = (∂t P )Q dQ. (3.32)
ai
Computing the variation of F0 we now obtain
Xq
2∂t F0 = −t∂t Q(Xp )P (Xp ) + Q dP − res P Q dS
∞Q
Xp

Q
+ res ln P dQ + res ln(P λ)Q dP . (3.33)
∞Q λ ∞P
We now have

Q
res ln P dQ
∞Q λ

t Q
= res V1 (Q) − + O Q−2 ln dQ
∞Q Q λ

Q Q dQ
= res V1 (Q) + O Q−2 ln dQ − t res ln
∞Q λ ∞Q λ Q

= res V1 (Q) dS − t ln(Q/λ) dS , (3.34)
∞Q
where, in the second residue, we can replace dQ/Q by dS because ln(Q/λ) = O(1).
A similar argument goes for the term involving P so that we finally have

2∂t F0 = −t∂t µ − res P Q dS + res V1 (Q) − t ln(Q/λ) dS
∞Q ∞Q

− res V2 (P ) − t ln(P λ) dS. (3.35)
∞P
2 Note that Q dP − t dS is a meromorphic Abelian differential without residues at the poles ∞

P ,Q .
Inserting this into the full expression of Fg we obtain

g
2∂t Fg = 2∂t F0 + ∂t (tµ) + 8i ∂t Γi
i=1

= µ − res P Q dS + res V1 (Q) − t ln(Q/λ) dS
∞Q ∞Q

g

− res V2 (P ) − t ln(P λ) dS + 8i dS
∞P
i=1 bi
= 2µ, (3.36)
where we have used the expression (3.30). This concludes the proof of the corollary. ✷
Remark 3.5. The formula for µ seems not symmetric in the roles of P and Q only
superficially. In fact, if we exchanged the roles, the 3rd kind differential should also change
sign.
We now investigate the scaling property of this Free energy. First off we have the simple
and P = σ P the
Lemma 3.6. Under the change of scale for the functions Q = δ Q
chemical potential rescales with an anomaly as follows
µ = δσ µ̃ + δσ t˜ ln(δσ ). (3.37)
Proof. The proof is almost immediate from the expression (3.19) considering the fact that
under that rescaling we have
uK = σ δ 1−K ũK , vJ = δσ 1−J ṽJ , t = δσ t˜, 8i = δσ 8̃i . (3.38)
On the other hand the differential dS (and hence the function λ) are invariant as follows
from its expression
dS = (∂t P )Q dQ = (∂t˜P)Q
dQ. (3.39)
The proof follows then immediately from (3.19). ✷
With the above lemma we can immediately find the scaling properties of the genus g
free energy
Corollary 3.7 (Scaling properties). The free energy of the genus g data above satisfies the
scaling constraints

g
1
2Fg = (1 − K)uK ∂uK + vJ ∂vJ + 8i ∂8i + t∂t Fg − t 2 , (3.40)
2
K J i=1

g
1
2Fg = uK ∂uK + (1 − J )vJ ∂vJ + 8i ∂8i + t∂t Fg − t 2 . (3.41)
2
K J i=1
Proof. The proof is immediate from the definitions of the various objects and using the
anomalous scaling of the chemical potential µ in Lemma 3.6. The constraint (3.40) is
obtained by keeping σ = 1 and differentiating w.r.t. δ at δ = 1, and (3.41) is obtained
similarly by interchanging the roles of δ and σ in the above procedure. ✷
The two scaling constraints (3.40), (3.41) form the “two halves” of the scaling

g
4Fg = (2 − K)uK ∂uK + (2 − J )vJ ∂vJ + 2 8i ∂8i + 2t∂t Fg − t 2
K J i=1
(3.42)
(obtained by adding (3.40) and (3.41)) which is the translation of the well-known property
(2.40) for higher genus free energies. Of course, the same properties (3.40), (3.41) hold
also for the free energy in Section 2.1 (in which case the part involving the filling fraction
would be missing).
We conclude with the remark that—quite clearly—the (multivalued) function λ is
playing essentially the same role of the uniformizing parameter in genus zero. The
quantities
ln(γ ) := − res ln(Q/λ) dS, ln(γ̃ ) := res ln(P λ) dS, (3.43)

∞Q ∞P
are the translation in this setting of the homonymous quantity in the genus zero case,
except that they need not be equal. However, since now λ = exp( dS) is defined up to
a multiplicative constant depending on the base-point of the integral, there would be a
choice for the base-point which makes γ = γ̃ .
4. Conclusion
The formulas we have presented fill a gap in both the theory of the dispersionless Toda
hierarchy and the two-matrix model, where the tau-function (free energy in the planar limit)
is known only through its partial derivatives but no closed formula for the tau-function itself
is known.
The derivation and the technique of the proof emphasizes the importance of the spectral
curve of the model, at least in the case of polynomial potentials or, in the dToda language,
finite Laurent polynomials for the Lax operators.
On a slightly different perspective, we have computed the free energy of the matrix
model in the case where the spectral curve is of genus g > 0; the computation is less
explicit than in genus zero but the formula is closed.
It would be interesting to explore further the remnant of the Poisson structure (2.3) in
this context. In fact it is almost immediate to verify that still
{P , Q} = λ∂λ P ∂t Q − λ∂λ Q∂t P = 1. (4.1)

The investigation of the Poisson structure will be the topic of subsequent publications.
Acknowledgements
I would like to thank Bertrand Eynard for discussion and pointing out many relevant
references, and Dmitri Korotkin for help in the computation of the higher genus case. Also
I would like to thank the anonymous referee who helped me realize a mistake in a first
version, which prompted me towards formulation of Corollary 3.7.
Many thanks go also to my daughter Olenka for anticipated inspiration.
Appendix A. The Riemann bilinear identity
It is a classical identity but it may be useful to recall it here. Let be given a curve Σg
of genus g and a symplectic basis in the homology {ai , bi }i=1,...,g . Let η and ω be two
meromorphic Abelian differentials and ω be without residues and define

Ω = ω. (A.1)
P
g be a simply connected fundamental domain on the universal covering of the curve
Let Σ
then the function Ω is single-valued inside Σg with possibly poles. Let ∂ Σg be the
boundary of the domain constituted by the cycles of the chosen basis. Then the bilinear
Riemann identity claims
g
2iπ ηΩ = η ω− ω η . (A.2)
residues i=1 ai bi ai bi
The proof is very easy and can be found in [12].
References
[1] M. Adler, P. Van Moerbeke, String-orthogonal polynomials, string equations and 2-Toda symmetries,
Comm. Pure Appl. Math. J. 50 (1997) 241–290.
[2] M. Adler, P. Van Moerbeke, The spectrum of coupled random matrices, Ann. Math. 149 (1999) 921–976.
[3] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[4] M. Bertola, Bilinear semi-classical moment functionals and their integral representation, J. Appl. Theor. 121
(2003) 71–99, math.CA/0205160.
[5] M. Bertola, B. Eynard, J. Harnad, Differential systems for biorthogonal polynomials appearing in 2-matrix
models and the associated Riemann–Hilbert problem, nlin.SI/0208002, Commun. Math. Phys, in press.
[6] M. Bertola, B. Eynard, J. Harnad, Duality of spectral curves arising in two-matrix models, Theor. Math.
Phys. 134 (1) (2003) 25–36.
[7] M. Bertola, B. Eynard, J. Harnad, Duality, biorthogonal polynomials and multi-matrix models, Commun.
Math. Phys. 229 (2002) 73–120.
[8] M. Bertola, B. Eynard, Mixed correlation functions of the two-matrix model, J. Phys. A 36 (2003) 7733–
7750, hep-th/0303161.
[9] P.M. Bleher, A.R. Its (Eds.), Random Matrix Models and their Applications, in: MSRI Research
Publications, Vol. 40, Cambridge Univ. Press, Cambridge, 2001.
[10] A. Boyarsky, O. Ruchayskiy, Integrability in SFT and new representation of KP tau-function, J. High Energy
Phys. 3 (2003) 027.
[11] F. David, Planar diagrams, two-dimensional lattice gravity and surface models, Nucl. Phys. B 257 (1985)
45.
[12] B.A. Dubrovin, A.T. Fomenko, S.P. Novikov, Modern Geometry-Methods and Applications. Part III.
Introduction to Homology Theory, in: Graduate Texts in Mathematics, Vol. 124, Springer-Verlag, New York,
1990.
[13] P. Di Francesco, P. Ginsparg, J. Zinn-Justin, 2D gravity and random matrices, Phys. Rep. 254 (1995) 1.
[14] B. Eynard, Large N expansion of the 2-matrix model, J. High Energy Phys. 1 (2003) 051.
[15] T. Guhr, A. Mueller-Groeling, H.A. Weidenmuller, Random matrix theories in quantum physics: common
concepts, Phys. Rep. 299 (1998) 189.
[16] N.M. Ercolany, K.T.-R. McLaughlin, presentation at the Montréal 2002 AMS meeting.
[17] V.A. Kazakov, Ising model on a dynamical planar random lattice: exact solution, Phys Lett. A 119 (1986)
140–144.
[18] V.A. Kazakov, A. Marshakov, Complex curve of the two matrix model and its tau-function, hep-th/0211236.
[19] I.K. Kostov, I. Krichever, P. Wiegmann, A. Zabrodin, The τ -function for analytic curves, random matrix
models and their applications, in: Math. Sci. Res. Inst. Publ., Vol. 40, Cambridge Univ. Press, Cambridge,
2001, pp. 285–299.
[20] A. Marshakov, P. Wiegmann, A. Zabrodin, Integrable structure of the Dirichlet boundary problem in two
dimensions, Commun. Math. Phys. 227 (1) (2002) 131–153.
[21] M.L. Mehta, Random Matrices, 2nd Edition, Academic Press, New York, 1991.
[22] M. Praehofer, H. Spohn, Universal distributions for growth processes in 1 + 1 dimensions and random
matrices, Phys. Rev. Lett. 84 (2000) 4882, cond-mat/9912264.
[23] K. Takasaki, T. Takebe, SDiff(2) Toda equation-hierarchy, tau function, and symmetries, Lett. Math.
Phys. 23 (3) (1991) 205–214.
[24] K. Takasaki, T. Takebe, SDiff(2) KP Hierarchy, Infinite Analysis, Part A, B, in: Advanced Series in
Mathematical Physics, Vol. 16, World Sci. Publishing, River Edge, NJ, 1992, pp. 889–922.
[25] K. Takasaki, T. Takebe, Quasi-classical limit of Toda hierarchy and W-infinity symmetries, Lett. Math.
Phys. 28 (3) (1993) 165–176.
[26] L.A. Takhtajan, Free bosons and tau-functions for compact Riemann surfaces and closed smooth Jordan
curves. Current correlation functions, Lett. Math. Phys. 56 (2001) 181–228.
[27] L.-P. Teo, Analytic functions and integrable hierarchies-characterization of Tau functions, hep-th/0305005.
[28] K. Ueno, K. Takasaki, Toda lattice hierarchy, Adv. Stud. Pure Math. 4 (1984) 1–95.
[29] J.J.M. Verbaarshot, Random matrix model approach to chiral symmetry, Nucl. Phys. B (Proc. Suppl.) 53
(1997) 88.
[30] P. Wiegmann, A. Zabrodin, Conformal maps and integrable hierarchies, Commun. Math. Phys. 213 (3)
(2000) 523–538.
[31] A. Zabrodin, The dispersionless limit of the Hirota equations in some problems of complex analysis, Teor.
Mat. Fiz. 129 (2) (2001) 239–257.
The Ginzburg–Landau theory and the surface energy

of a colour superconductor
Ioannis Giannakis, Hai-Cang Ren
Physics Department, The Rockefeller University, 1230 York Avenue, New York, NY 10021-6399, USA
Abstract
We apply the Ginzburg–Landau theory to the colour superconducting phase of a lump of dense
quark matter. We calculate the surface energy of a domain wall separating the normal phase from the
super phase with the bulk equilibrium maintained by a critical external magnetic field. Because of the
symmetry of the problem, we are able to simplify the Ginzburg–Landau equations and express them
in terms of two components of the di-quark condensate and one component of the gauge potential.
The equations also contain two dimensionless parameters: the Ginzburg–Landau parameter κ and ρ.
The main result of this paper is a set of inequalities obeyed by the critical value of the Ginzburg–
Landau parameter—the value of κ for which the surface energy changes sign—and its derivative
with respect to ρ. In addition we prove a number of inequalities of the functional dependence of the
surface energy on the parameters of the problem and obtain a numerical solution of the Ginzburg–
Landau equations. Finally a criterion for the types of colour superconductivity (type I or type II) is
established in the weak coupling approximation.
1. Introduction
The Ginzburg–Landau theory provides a powerful tool for exploring systems with
inhomogeneous order parameters near the critical temperature. For instance, its application
to the surface energy of a normal-superconducting interface led to the discovery of the two
types of superconductors. When an external magnetic field whose magnitude is less than a
critical value Hc , is applied to type I superconductors, the field is expelled from the interior
of the sample (the Meissner effect). As the magnitude of the magnetic field increases above
E-mail addresses: giannak@summit.rockefeller.edu (I. Giannakis), ren@summit.rockefeller.edu

(H.-C. Ren).
doi:10.1016/j.nuclphysb.2003.07.022
I. Giannakis, H.-C. Ren / Nuclear Physics B 669 [FS] (2003) 462–478 463
Hc , superconductivity disappears completely and the normal phase is restored. In the case
of type II superconductors, the sample behaves in a similar manner as long as the external
magnetic field remains below a lower critical magnitude, Hc1 . A qualitatively different
behaviour emerges when the applied magnetic field lies in the range Hc1 < Hc < Hc2 ,
where it becomes energetically favourable for the magnetic field to penetrate the sample in
the form of quantized flux lines called vortices. Beyond the upper critical magnitude of the
field, the vortices are too dense to maintain the condensate, and the normal phase is once
again restored.
The criterion that determines whether a superconductor is of type I or type II is the
Ginzburg–Landau parameter. It is defined as the ratio of the penetration depth of the
magnetic field δ over the coherence length ξ , the distance over which changes in the
order parameter occur, i.e., κ = δ/ξ . These characteristic
√
lengths imply that fluctuations
in the magnitude of the condensate decay as e − 2 x/ξ while the magnetic field inside
the superconductor falls off as e−x/δ . In their original paper, Ginzburg and Landau [1]
analyzed the energy of the interface between a normal and a superconducting phase, kept
in equilibrium in the bulk by an external magnetic field at the critical value. They found
analytically that the surface energy vanishes at
1
κ = κc = √ ∼ = 0.707. (1.1)
2
The physical meaning of this critical value was clarified√ further by Abrikosov [2]. √ It
represents the demarcation line between type I (κ < 1/ 2 ) and type II (κ > 1/ 2 )
superconductors.
In the present paper we shall carry out a similar analysis of the Ginzburg–Landau theory
for colour superconductors. Colour superconductivity is essentially the quark analog of
BCS superconductivity [3–6]. Because of attractive quark–quark interactions in QCD,
the Fermi sphere of quarks becomes unstable against the formation of Cooper pairs and
the system becomes superconducting at sufficiently low temperatures. In this paper we
shall consider a domain wall separating the normal and colour superconducting phases
in equilibrium under the influence of an external magnetic field and calculate the surface
free energy per unit area. The Ginzburg–Landau equations for this system are considerably
more complicated than the analogous equations for an ordinary superconductor [7–11],
because of the non-Abelian nature of the problem. By using symmetry arguments, we were
able to simplify the problem of the domain wall and express the GL equations in terms
of one gauge potential and two components of the di-quark condensate. When expressed
in terms of dimensionless quantities, the GL equations contains two dimensionless
parameters: the usual Ginzburg–Landau parameter κ and another parameter ρ. Weak
coupling calculations fix the value of ρ to be equal to −1/2. Although we were unable
to derive an analytical result analogous to (1.1) we proved the following inequalities for
the critical GL parameter and its derivative with respect to ρ:
1 dκc
κc (ρ) √ , 0. (1.2)
2 dρ
We also derived a number of inequalities of the functional dependence of the surface energy
on the parameters of the problem and obtained a numerical solution for κc (ρ). For weak
464 I. Giannakis, H.-C. Ren / Nuclear Physics B 669 [FS] (2003) 462–478
coupling we found that
κc 0.589, (1.3)
in contrast to (1.1).
In Section 2 of this paper, we shall review the Ginzburg–Landau theory of colour
superconductivity. The analytical and numerical solution of the domain wall problem will
be presented in Section 3. In Section 4 of the paper we shall summarize our results and in
Appendix A we shall discuss the validity of the Ginzburg–Landau theory in the presence
of the fluctuations of the gauge field.
2. The Ginzburg–Landau theory of colour superconductivity
The symmetry group of QCD in the chiral limit is
SU(3)c × SU(3)fR × SU(3)fL × U (1)B , (2.1)

where the subscript c denotes colour, the subscript fR (fL ) the right(left)-hand flavour and
B the baryon number. The electromagnetic gauge group U (1)em is not a separate symmetry
group, but a subgroup of SU(3)fR × SU(3)fL . The dominant pairing channel consists
of two quarks of the same helicity and the corresponding order parameters in a colour
superconducting quark matter will be denoted by ΨR and ΨL . These condensates transform
into each other under space inversions. Neglecting the parity violating processess, the
Ginzburg–Landau free energy functional which is consistent with the symmetry group (2.1)
reads

1 1 2
Γ = d 3 r Fijl Fijl + (∇ × A)
4 2
1 † L )† (DΨ
L)

+ Tr (DΨ R ) (DΨ R ) + (DΨ
2
1 1 2 2
+ a Tr ΨR† ΨR + ΨL† ΨL + b Tr ΨR† ΨR + ΨL† ΨL
2 4
1
† 2 † 2 1 † †
+ b Tr ΨR ΨR + Tr ΨL ΨL + c Tr ΨR ΨR Tr ΨL ΨL ,
4 2
(2.2)
where the gauge covariant derivative of the di-quark condensate Ψ reads

R(L))c1 c2 = ∇(Ψ
(DΨ R(L))c1 c2 − ig Ac1 c (ΨR(L) )c c2 − ig Ac2 c (ΨR(L))c1 c
f1 f2 f1 f2 f1 f2 f1 f2
R(L))c1 c2 ,
− ie(qf1 + qf2 )A(Ψ (2.3)
f1 f2
where A = Al T l denotes the classical vector potential of the SU(3) colour gauge field,
A the electromagnetic field, and eqf is the electric charge of the f th flavour quark. The
SU(3) generator T l is in its fundamental representation.
At weak coupling, the parameters are calculable from either the perturbative one-gluon
exchange interaction of QCD or from the Nambu–Jona-Lasinio effective action. Both
approaches lead to the same expression [7,8]:
48π 2 2
a= k Tc (T − Tc ),
7ζ (3) B

576π 4 kB Tc 2
b= ,
7ζ (3) µ
b
= c = 0, (2.4)
where µ is the chemical potential and Tc the transition temperature.
Since in this paper we address the domain wall problem—the interface between
a normal and a superconducting phase in an external magnetic field—the boundary
conditions select the even parity sector of (2.2)
ΨR = ΨL ≡ Ψ. (2.5)
The one-gluon exchange process that dominates the di-quark interaction at ultrahigh
chemical potential is attractive for quarks within the colour antisymmetric channel. Thus
the Cooper pairs realise the colour antisymmetric representation to the leading order of
the QCD running coupling constant at T = 0 [12,13], or to the leading order of (1 − TTc )
near Tc [8]. Assuming such a pairing pattern persists even at moderately high chemical
potential, the di-quark condensate Ψ can be expressed in terms of a 3 × 3 complex
matrix, Φ
Ψfc11fc22 = ) c1 c2 c )f1 f2 f Φfc . (2.6)

The gauge covariant derivative of Φ can be written as
c − ig Ācc
Φ c
− ieQf AΦ
c = ∇Φ
(DΦ) c, (2.7)
f f f f
where
Ā = Al T l , Φ = Φ0 + Φ l T l (2.8)

with T l = −T l∗ being the generator of the 3̄ representation, Qf = qf1 + qf2 and ff1 f2
represent a cyclic permutation of 1, 2, 3. Arranging the flavour index in the conventional
order of u, d, s we find the diagonal electric charge matrix

2 1 1 2
Q = −diag , − , − = − √ T 8 . (2.9)
3 3 3 3
Here we have adapted an expression of T l that differs from the standard one by a cyclic
permutation of rows (columns), in which T 8 = √ 1
diag(2, −1, −1). Eq. (2.7) takes the
2 3
matrix form
= ∇Φ
DΦ + i √2 eAΦ
− ig ĀΦ T 8 , (2.10)
3
and the Ginzburg–Landau free energy (2.2) becomes

1 1 2
Γ = d r Fijl Fijl + ∇
3
× A + 4 tr(DΦ) † (DΦ)
+ 4a tr Φ † Φ
4 2

† 2 2
+ b1 tr Φ Φ + b2 tr Φ Φ ,†
(2.11)
where the parameters b1 and b2 are related to the parameters of Eq. (2.2) in the following
manner
b1 = b, b2 = b + 8b
+ 8c. (2.12)
First, let us review the case with a homogeneous condensate, i.e., A = A = ∇Φ
= 0,
for which the Ginzburg–Landau free energy (2.11) becomes

2 2
Γ = d 3 r 4a tr Φ † Φ + b1 tr Φ † Φ + b2 tr Φ † Φ . (2.13)
We distinguish the following three regions of the parameter space b1 − b2 , following the
treatment of [7].
(1) b1 > 0 and b1 + 3b2 > 0: the minimum free energy corresponds to the colour-
flavour locked condensate [4],
Φ = φ0 U, (2.14)
where

2a
φ0 = − (2.15)
b1 + 3b2
and U is a unitary matrix.1 Consequently, we find that

12a 2
Γmin = −Ω ; (2.16)
b1 + 3b2
(2) b1 < 0 but b1 + b2 > 0: in this case the colour-flavour locked condensate (2.14)
becomes a saddle point of the free energy, which is nevertheless bounded from below. The
minimum corresponds to an isoscalar condensate, given by

Φ = diag φ
0 eiα , 0, 0 , (2.17)

where φ
0 = − b12a+b2 . We find then that
4a 2
Γmin = −Ω ; (2.18)
b1 + 3b2
1 It is very important to maintain this general form of the CFL condensate, since its non-trivial winding onto
the gauge group, SU(3)c × U (1)em gives rise to vortex filaments.
(3) For b1 and b2 outside the region specified by (1) and (2), the free energy is no longer
bounded from below. Higher powers of the order parameter have to be included and the
superconducting transition becomes first order. Since case (2) is mathematically identical
to a metallic superconductor, we shall not address it in this paper.
In order to identify the various characteristic lengths of the system, we consider fluc-
tuations about the homogeneous condensate (2.14). We parametrize the order parameter
by

1 X + iY l Xl + iYl
Φ = φ0 + √ +T (2.19)
6 2 2
with X’s and Y ’s real, and form linear combinations of the ordinary electromagnetic gauge
potential with the eighth component of the colour gauge potential citejur,
θ,
V = A8 cos θ + Asin
8 θ,
V = −A sin θ + Acos (2.20)
where the “Weinberg angle” is given by
2e
tan θ = − √ . (2.21)
3g
and A. We find
Next we expand (2.11) to quadratic order in X’s Y ’s A’s

1 2
Γ = Γmin + d 3 r (∇ × V) + tr (∇ ×W )2 + m2W W
2
2
1
2 + 1 (∇X)
2 + m2Z Z 2 + m2H X2

+ (∇ × Z)
2 2
1 l l
2 l l
1

+ ∇X ∇X + mH X X + (∇Y ) , 2
(2.22)
2 2
7
= l=1 T̄ W
where W l and
l
l = Al − 1 ∇Y
W l for l = 1, 2, . . . , 7,
mW
= V − 1 ∇Y
Z 8. (2.23)
mZ
The masses of the excitations provide us with the relevant length scales which are the
coherence lengths, ξ and ξ
defined by
2 2
m2H = (b1 + 3b2 )φ02 = , m
H2 = b1 φ02 = , (2.24)
ξ2 ξ
2
that indicate the distances over which the di-quark condensate varies and the magnetic
penetration depths, δ and δ
by
1 1
m2Z = 4g 2 φ02 sec2 θ = , m2W = m2Z cos2 θ = . (2.25)
δ2 δ
2
The excitations that violate (2.5) correspond to the Goldstone bosons associated with chiral
symmetry breaking, and the η particle that becomes massive through the anomaly.
In the rest of this section, we shall determine the critical magnetic field of a
homogeneous colour superconductor. For simplicity, the diamagnetic response of the
quarks in the normal phase will be neglected. In the presence of an external magnetic
field, the thermodynamic function to be minimized is the Gibbs free energy Γ, which is a
Legendre transformation of the Helmholtz free energy Γ , i.e.,

Γ = Γ − H · d 3 r ∇ × A. (2.26)
Following the decomposition (2.20), we write

Γ = Γ1 + Γ2 (2.27)
with

1 l l
7
1
Γ1 = d 3 r Fij Fij + (∇ × V )2 + 4 tr(DΦ)
† (DΦ)
+ 4a tr Φ † Φ
4 2
l=1

† 2 2
+ b1 tr Φ Φ + b2 tr Φ Φ − H · ∇×V sin θ ,
†

1 2
Γ2 = d 3 r (∇ × V) − H · ∇×V cos θ . (2.28)
2
The minimization of Γ2 gives rise to
× V = H cos θ,
∇ (2.29)
and no distinction will be made between the super-phase and the normal phase. In the
remaining of the section we shall concentrate solely on Γ1 .
A homogeneous super-phase is characterized by a perfect Meissner effect, ∇ × V = 0

and the CFL condensate Φ = φ0 , which gives rise to the minimum of Γ , Γs = Γmin . given
× V = H sin θ and the
by (2.16). Γ1 for a homogeneous normal phase is minimized by ∇

minimum reads Γn = −Ω 2 H sin θ . The critical magnetic field H = Hc is determined
1 2 2
from the equilibrium condition

12a 2
Γn = Γs = Γmin = −Ω , (2.30)
b1 + 3b2
which leads to

6a 2
Hc = 2 | csc θ |. (2.31)
b1 + 3b2
Substituting the parameters (2.4) which were calculated using weak coupling approxima-
tion into (2.30) yields

3 2 k B Tc T
Hc = 4 µ 1− |csc θ |. (2.32)
14ζ (3) µ Tc
Extrapolating this expression to a realistic value for the quark chemical potential, for
example, µ = 400 MeV, and using the one-loop running coupling constant
12π 2
g 2 = 11 µ (2.33)
2 Nc − Nf ln Λ
with Nc = Nf = 3 and Λ = 200 MeV, we find
θ∼
= −5.6◦ (2.34)
and

k B Tc Tc
Hc ∼
= 1.47 × 1020 1− Gauss, (2.35)
µ T
which is much higher than the typical magnetic field inside a neutron star. This estimation
of course depends on the validity of the extrapolation of the weak coupling formulas to
realistic values of the chemical potential.
3. The domain wall problem
Let us consider an interface between the normal phase and the colour superconducting
phase, for example, the yz plane. The equilibrium in the bulk is maintained by a uniform
external magnetic field of critical magnitude, H = Hc ζ̂ . All quantities then in both phases
will depend solely on x. It is evident from symmetry that the gauge field lies in one plane,
let this be the xy plane, so that V = V (x)ŷ and ∇ × V = dVdx ζ̂ . The boundary conditions
on the Ginzburg–Landau equations in the problem that we are considering (corresponding
to the normal and colour superconducting phases as x → −∞ and x → ∞) are
dV
φ → 0, χ → 0, Hc sin θ at x → −∞,
→
dx
√ dV
φ → φ0 , χ → 2 φ0 , → 0 at x → ∞. (3.1)
dx
The surface energy σ is defined as
Γ1 − Γs
σ= , (3.2)
Area of yz plane
where the Gibbs free energy of a homogeneous superphase, Γs is given by (2.30), and it is
equal to that of the normal phase in the presence of H = Hc .
The minimization of Γ1 generates a set of coupled non-linear differential equations—the
Ginzburg–Landau equations-subject to the boundary conditions (3.1). Now we divide the
while
field variables in Eq. (2.8) into two groups, the first group contains Φ0 , Φ 8 , A 8 , A,
the second one includes Φl , Al with l = 1, 2, . . . , 7. A closer inspection of the structure of
reveals the absence of terms in the free energy (2.2) that are linear in the variables of
DΦ
the second group. Therefore a solution of the equations of motion exists in which only
the fields of the first group acquire non-zero values. Since this ansatz implements the
maximum symmetry allowed by the boundary conditions, we expect that it includes the
solution that minimizes the free energy of the domain wall.
By setting Φl = A l = 0 for l = 1, 2, . . . , 7 and transforming the remaining variables

1 √ 1
φ = Φ0 + √ Φ 8 , χ = 2 Φ0 − √ Φ 8 , (3.3)
3 2 3
we arrive at the following expression for the Ginzburg–Landau free energy functional (2.2)

2 2

Γ1 = d 3 r (∇
1
× V )2 + 4 ∇ − i √g sec θ V φ + 4 ∇ +i √g
sec θ χ
V
2 3 2 3

1 2
+ 4a |φ|2 + |χ|2 + b1 |φ|4 + |χ|4 + b2 |φ|2 + |χ|2
2

× V )sin θ .
− Hc ζ̂ · (∇ (3.4)
The condition for the colour-flavour locking (that the 3 × 3 matrix Φ is proportional to a
unitary matrix) amounts to |χ|2 = 2|φ|2 . Consequently, we shall use instead of the variable
x and the functions V , φ and χ the dimensionless quantities
√
√ −3a
s = δx, φ = φ0 u, χ = 2 φ0 v, V =− Acos θ . (3.5)
g
The boundary conditions in terms of the new dimensionless quantities become
u → 0, v → 0, A
→ 1 at s → −∞,

u → 1, v → 1, A → 0 at s → ∞, (3.6)
where prime indicates differentiation with respect to s. The surface energy per unit area in
terms of the dimensionless quantities becomes
∞
6a 2 1
1 1 1
σ= ds (A − 1)2 + 2 u
2 + 2v
2 + A2 2u2 + v 2 − u2 + 2v 2
b 2 3κ 6 3
−∞

1 4 1 2
+ 2u + 2u2 v 2 + 5v 4 + ρ u2 − v 2 , (3.7)
18 18
where
δ
κ= (3.8)
ξ
is the Ginzburg–Landau parameter, b = 14 (b1 + 3b2 ), and
b1 − 3b2
ρ= (3.9)
b1 + 3b2
is another dimensionless parameter. The weak coupling calculation fixes ρ to be equal
to − 12 .2 The Ginzburg–Landau equations which determine the profile of the colour

2 Colour neutrality provides ρ = − 1 + 12π 2 kB Tc 2 ln2 µ .
2 7ζ (3) µ kB Tc
condensate u, v and the magnetic field in the colour superconductor are determined by
minimizing the surface energy with respect to functions u, v and A
1
−A
+ A 2u2 + v 2 = 0,
3
1
2 1 1
− 2 u + A − 1 u + 2u2 + v 2 u + ρ u2 − v 2 u = 0,
κ 3 3
1
1 2 1 2 1
− 2 v + A − 4 v + u + 5v 2 v − ρ u2 − v 2 v = 0. (3.10)
κ 4 6 6
In the presence of a non-zero A, as it is required by the boundary conditions, this set of
equations does not admit a solution in which u = v everywhere. Stated differently, in the
presence of an inhomogeneity and a non-zero gauge potential the unlocked condensate-
the octet-has to acquire a non-zero value somewhere. This situation is analogous to the
Ginzburg–Landau theory of a cuprate superconductor [15], where the s-wave condensate
becomes non-zero in the vicinity of a vortex filament while the d-wave condensate
dominates in the bulk. It is easily verified that Eqs. (3.10) have the first integral
1
2 1 1 1
A + 2 u
2 + 2v
2 − A2 2u2 + v 2 + u2 + 2v 2
2 3κ 6 3
1 4 1 2
2 2 1
− 2u + 2u v + 5v − ρ u − v
2 2 4
= (3.11)
18 18 2
which implies, according to the boundary conditions ((3.6)), that
A(∞) = 0. (3.12)
Let σmin be the minimum surface energy. It is obtained from the substitution of the field
variables u, v, A that satisfy the GL equations (3.10) into (3.7), and is a function of the
parameters κ and ρ. We proceed now to prove the following lemma.
Lemma 1.

∂σmin
0, (3.13)
∂κ ρ

∂σmin
0. (3.14)
∂ρ κ
Proof.
According to (3.7),
∞
∂σmin 4a 2
=− 3 ds u
2 + 2v
2
∂κ ρ bκ
−∞
∞

δσmin ∂A(s) δσmin ∂u(s)
+ ds +
δA(s) ∂κ ρ δu(s) ∂κ ρ
−∞

δσmin ∂v(s)
+ . (3.15)
δv(s) ∂κ ρ
The second term vanishes because of the equation of motion (3.10) and the boundary
conditions (3.6), (3.12). Consequently, we find that
∞
∂σmin 4a 2
=− 3 ds u
2 + 2v
2 0. (3.16)
∂κ ρ bκ
−∞
Similarly
∞
∂σmin a2 2
= ds u2 − v 2 0. (3.17)
∂ρ κ 3b
−∞
The lemma is then proved. ✷
Furthermore lets denote the critical value of the Ginzburg–Landau parameter for which
σmin vanishes by κc (ρ). We shall prove then the following theorem
Theorem 2.
1
κc (ρ) √ (3.18)
2
and
dκc
0. (3.19)
dρ
Proof. Consider a special field configuration u = v, which satisfies the boundary

conditions (3.6) but not the Ginzburg–Landau equations (3.10) with A = 0. Therefore
∞

6a 2 1
1
2 1 2 2 1 4
σmin ds (A − 1) + 2 u + A u − u + u ≡ σ̄ .
2 2
(3.20)
b 2 κ 2 2
−∞
The minimization of σ̄ with respect to u and A yields the set of differential equations,
−A
+ V u2 = 0,
1 1
− 2 u
+ A2 − 2 u + u3 = 0, (3.21)
κ 2
subject to the boundary conditions (3.6). We recognize that (3.20) and (3.21) correspond
exactly to the domain wall problem for an ordinary superconductor analyzed √ by Ginzburg
and Landau in their original work. It follows then that σ̄ = 0 at κ = 1/ 2. On the other
hand, this field configuration
√ does not satisfy the equations of motion (3.7) and therefore
σmin 0 at κ = 1/ 2. Following (3.13) of Lemma 1 we establish the first equation of the
theorem.
The second equation of the theorem follows from the identity
∂σmin
∂κ ∂ρ κ
= − ∂σ (3.22)
∂ρ σmin min
∂κ ρ
and Lemma 1. The theorem is then proved. ✷
The solution to the Ginzburg–Landau equations (3.10) and the critical value of the
Ginzburg–Landau parameter κ for various values of ρ ∈ (−1, ∞) were also determined
numerically. The continuous range s ∈ (−∞, ∞) was replaced by a finite lattice with
sn = −L + n), (3.23)
where n = 0, 1, 2, . . . , N + 1 and (N + 1)) = 2L. A good approximation amounts to ) and
L that satisfy ) min(1, 1/κ) and L max(1, 1/κ). The field variables un , vn , and An
are assigned to each site and their derivatives can be approximated by u
n = (un+1 − un )/),
vn
= (vn+1 − vn )/), and A
n = (An+1 − An )/). The surface energy is then written
N

6a 2 1
1
σ= ) (An − 1)2 + 2 u
n2 + 2vn
2
b 2 3κ
n=0

N+1
1 2 2 1
+ )n An 2un + vn2 − u2n + 2vn2
6 3
n=0

1 4 1 2
+ 2un + 2u2n vn2 + 5vn4 + ρ u2n − vn2 (3.24)
18 18
with )0 = )N+1 = 12 ) and )n = ) for n = 1, . . . , N . The multivariable function (3.24) is

then minimized by iteration subject to the conditions u0 = v0 = 0, uN+1 = vN+1 = 1,
A
0 = 1 and AN+1 = 0.
The critical GL parameter, κc versus ρ is plotted in Fig. 1. The critical value of
GL parameter in the weak coupling approximation (ρ = − 12 ), is κc = 0.589. The two
components of the di-quark condensate, the colour-flavour locked and the colour-flavour
octet, and the corresponding magnetic field, that correspond to the solution with ρ = −1/2
and κ = κc are plotted in Fig. 2. The numerical results are consistent with the inequalities
(3.18) and (3.19) that we derived analytically.
Substituting the weak coupling expression of the parameters, the coherence length reads
2
7ζ (3) 1 Tc
ξ =
2
, (3.25)
48π 2 kB Tc Tc − T
and the penetration depth
3π Tc
δ2 = cos2 θ (3.26)
2αs µ2 Tc − T
g2
with αs = 4π .
The Ginzburg–Landau parameter is given then by

δ2 72π 3 kB Tc 16.3 kB Tc
κ= 2 = cos θ ∼
=√ . (3.27)
ξ 7ζ (3)αs µ αs µ
If we consider a realistic value for the chemical potential, for example, µ = 400 MeV,
it follows from (1.3) that the colour superconductor will be of type I if kB Tc < 14 MeV
Fig. 1. The critical Ginzburg–Landau parameter κc vs. the parameter ρ.
Fig. 2. The solutions to the Ginzburg–Landau equations (3.10) for ρ = − 12 and κ = 0.589. The solid line
represents the colour-flavour locked component, 13 (u + 2v), the dashed line represents the colour-flavour octet
component, √2 (v − u) and the dotted line the colour-magnetic field, A
.
3
and of type II otherwise. Furthermore if we extrapolate the weak coupling formula for Tc
[16–20], which is valid at asymptotic densities, to µ = 400 MeV we find kB Tc = 3.5 MeV
and the colour superconductor is of type I. Of course if non-perturbative effects raise Tc
significantly, for example, by one order of magnitude, the colour superconductor could be
of type II.
4. Concluding remarks
In this paper we have applied the Ginzburg–Landau theory to the problem of a domain
wall which separates the normal phase from the colour superconducting phase of dense
quark matter. The main purpose of this work was to establish a criterion that determines
whether the colour superconductor is type I or type II with respect to an external magnetic
field. We initially derived the Ginzburg–Landau equations of motion for the gauge fields
and the order parameter by minimizing the surface energy of the interface between the two
phases. The condition that the surface energy is a minimum is equivalent to the requirement
that the normal and the colour superconducting phases are in stable equilibrium with
each other. This equilibrium is maintained by an external magnetic field. Using symmetry
arguments we were able to simplify the original set of equations and rewrite them in
terms of only two components of the order parameter and one gauge potential. The
equations also contain two dimensionless parameters: the Ginzburg–Landau parameter
κ and ρ. Therefore the problem is more complicated than the problem of the metallic
superconductor which was addressed by Ginzburg and Landau. Although we were unable
to derive an analytical result analogous to the Ginzburg–Landau criterion (1.1), we derived
several rigorous inequalities about the critical parameter κc , the value for which the surface
energy vanishes, and its derivative with respect to ρ. Those analytical results are supported
by a numerical solution of the Ginzburg–Landau equations. By extrapolating the weak
coupling approximation formulas for the parameters to realistic values of the chemical
potential, for example, µ = 400 MeV, we find that the colour superconductor is of type I
(type II) if the transition temperature is below(above) 14 MeV.
The value of our work is mainly theoretical. It addresses the broken U (1) gauge field
onto which the electromagnetic field has a small projection. Although we have been
referring to the typical chemical potential of a neutron star, the implications of our work
might be quite limited [14]. First of all, there has not been any evidence yet that colour
superconductivity is realised in the core of a neutron star. Even if this is the case, the
critical magnetic field (2.35) is too high for the distinction between type I and type II to
be observed unless the pairing force is strong enough to reduce the lower critical field by
several orders of magnitude. In addition the small mixing angle (2.34) and the dependence
of the magnetic response on the thickness of the crust of the CSC core make it difficult to
observe the partial magnetic Meissner effect and the appearance of vortex filaments.
Because of the strength of the QCD coupling and the ultra-relativistic Fermi sea, the
fluctuations of the gauge field could be more significant than those in a non-relativistic
superconductor [21]. These fluctuations contribute to the Ginzburg–Landau free energy
functional an energy density
AΓ ∼ kB Tc δ −3 . (4.1)

As the condensate energy (2.16) is proportional to (1 − TTc )2 while δ ∼ 1 − TTc , the
contribution from the fluctuations will eventually dominate as T approaches Tc and modify
the nature of the phase transition.
The fluctuation energy is estimated in Appendix A within the framework of the
Ginzburg–Landau theory, where we demonstrate that the inclusion of this energy modifies
the phase transition from second order to first order. The parameter
T
t= −1 (4.2)
Tc
which describes the deviation of the first order transition temperature T from the second
order one Tc , is used as a measure of the importance of the fluctuations. We find that
for realistic values of the quark chemical potential, µ = 400 MeV and kB Tc 13 MeV,
t 0.1. In that case, the Ginzburg–Landau free energy functional (2.2) represents a
reasonable approximation.
For t ∼ 1 the approximation employed to derive the Ginzburg–Landau free energy (2.2)
with (2.4) from QCD or from the Nambu–Jona–Lasinio effective action breaks down.
Higher powers of the order parameter have to be included. It would be very interesting
to develop a systematic weak coupling approximation in this case and we hope to be able
to report progress in the future.
Acknowledgements
We would like to thank S. Catto, G. Dunne, H. Hansson and D. Rischke for useful
discussions. This work is supported in part by US Department of Energy, contract number
DE-FG02-91ER40651-TASKB.
Appendix A
In this appendix we shall estimate the effect of the fluctuations of the gauge field.
A systematic evaluation of the contribution from the fluctuations to the free energy requires
a resummation of a set of ring diagrams of QCD at finite temperature. Here we shall
follow the treatment of Bailin and Love [21] and estimate the fluctuation terms within the
framework of the Ginzburg–Landau theory. Due to the Debye screening the contributions
of the longitudinal components of the gauge potentials are suppressed, and only the
transverse components contribute in the static limit. We expect that our simple-minded
estimation captures the fluctuation terms to the leading order of the condensate.
Assuming that the small fluctuations are governed by the quadratic form of the
Ginzburg–Landau free energy functional, the shift of the free energy due to the Meissner
effect is given by

[dW ][dZ]e−S
AΓ = −kB T ln , (A.1)
[dW ][dZ]e−S0
and Z
where W represent the transverse degrees of freedom and

β
S= d 3 r (∇ ×W )2 + m2W W
2 + (∇ × Z)
2 + m2Z Z
2 ,
2

β
S0 = d 3 r (∇ ×W )2 + (∇ 2 .
× Z) (A.2)
2
Completing the Gaussian integral, we find that

d 3 k m2W m2Z
AΓ = kB T 7 ln 1 + + ln 1 +
(2π)3 k2 k2
kB TC
−3 4g 3
=− 7δ + δ −3 = − 7 + sec2 θ kB Tc |φ|3 . (A.3)
6π 3π
In the previous calculation, dimensional regularization was employed to eliminate
ultraviolet divergences. By including this term to the homogeneous Ginzburg–Landau free
energy with Φ = φ, we find
4g 3
Γ = 12a|φ|2 + 12b|φ|4 − 7 + sec2 θ kB Tc |φ|3 . (A.4)
3π
The presence of the cubic term will induce a first-order phase transition at T > Tc . The
temperature is determined by the condition that the free energy curve Γ vs. |φ| becomes
tangent to the φ-axis at some φ = 0, i.e.,
16παs3 2
a= 7 + sec2 θ (kB TC )2 . (A.5)
81b
Substituting the weak coupling expression for a and b we obtain that

T 49ζ 2(3)(7 + sec2 θ)2 α 3 s µ 2
t≡ −1= (A.6)
TC 139968π 5 k B Tc
with µ = 400 MeV, and kB TC > 13 MeV, t < 0.1. As a result the fluctuations are
not significant and the Ginzburg–Landau energy functional (2.2) represents a reasonable
approximation.
References
[1] V.L. Ginzburg, L.D. Landau, Zh. Eksp. Teor. Fiz. 20 (1950) 1064.
[2] A.A. Abrikosov, Sov. Phys. JETP 5 (1957) 1174.
[3] B. Barrois, Nucl. Phys. B 129 (1977) 390;
S. Frautschi, in: N. Cabibbo (Ed.), Proceedings of the Workshop on Hadronic Matter at Extreme Energy
Density, Erice, Italy, 1978;
D. Bailin, A. Love, Phys. Rep. 107 (1984) 325, and references therein for early works.
[4] M. Alford, K. Rajagopal, F. Wilczek, Nucl. Phys. B 537 (1999) 443.
[5] R. Rapp, T. Schafter, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 (1998) 53.
[6] K. Rajagopal, F. Wilczek, in: M. Shifman (Ed.), At the Frontier of Particle Physics/Handbook of QCD, B.L.
Ioffe Festschrift, World Scientic, Singapore, 2001;
T. Schafer, Quark matter, hep-ph/0304281.
[7] K. Iida, G. Baym, Phys. Rev. D 63 (2001) 074018.

[8] I. Giannakis, H.-C. Ren, Phys. Rev. D 65 (2002) 054017.
[9] K. Iida, G. Baym, Phys. Rev. D 65 (2002) 014022;
K. Iida, G. Baym, Phys. Rev. D 66 (2002) 014015.
[10] E. Nakano, T. Suzuki, H. Yabu, J. Phys. G 29 (2003) 491.
[11] D. Blaschke, D. Sedrakian, Ginzburg–Landau equations for superconducting quark matter in neutron stars,
nucl-th/0006038.
[12] T. Schafer, Nucl. Phys. B 575 (2000) 269.
[13] I. Shovkovy, L.C.R. Wijewardhana, Phys. Lett. B 470 (1998) 189.
[14] M. Alford, J. Berges, K. Rajagopal, Nucl. Phys. B 571 (2000) 269.
[15] Y. Ren, J.H. Xu, C.S. Ting, Phys. Rev. Lett. 74 (1995) 3680.
[16] D.T. Son, Phys. Rev. D 59 (1999) 094019.
[17] T. Schafer, F. Wilczek, Phys. Rev. D 60 (1999) 114033.
[18] R.D. Pisarski, D.H. Rischke, Phys. Rev. D 61 (2000) 051501;
R.D. Pisarski, D.H. Rischke, Phys. Rev. D 61 (2000) 074017.
[19] W. Brown, J.T. Liu, H.-C. Ren, Phys. Rev. D 61 (2000) 114012;
W. Brown, J.T. Liu, H.-C. Ren, Phys. Rev. D 62 (2000) 054016.
[20] Q. Wang, D.H. Rischke, Phys. Rev. D 65 (2002) 054005.
[21] D. Bailin, A. Love, Phys. Lett. B 109 (1982) 501.
CUMULATIVE AUTHOR INDEX B661–B669
Abel, S.A. B663 (2003) 197 Bellucci, S. B663 (2003) 605

Acquaviva, V. B667 (2003) 119 Bellucci, S. B665 (2003) 402
Adams, D.H. B662 (2003) 220 Bena, I. B664 (2003) 45
Aglietti, U. B668 (2003) 3 Benakli, K. B662 (2003) 40
Akbar, M.M. B663 (2003) 215 Benson, D. B665 (2003) 367
Akemann, G. B664 (2003) 457 Berezin, V. B661 (2003) 409
Alexandrov, S.Yu. B667 (2003) 90 Berg, M. B669 (2003) 3
Alishahiha, M. B661 (2003) 174 Bernabéu, J. B669 (2003) 255
ALPHA Collaboration B663 (2003) 3 Bertola, M. B669 (2003) 435
ALPHA Collaboration B669 (2003) 173 Bhattacharyya, T. B668 (2003) 415
Álvarez, E. B663 (2003) 365 Bigi, I.I. B665 (2003) 367
Álvarez-Gaumé, L. B668 (2003) 293 Bijnens, J. B669 (2003) 341
Antoniadis, I. B662 (2003) 40 Bilal, A. B663 (2003) 343
Armoni, A. B664 (2003) 233 Blažek, T. B662 (2003) 359
Armoni, A. B667 (2003) 170 Blumenhagen, R. B663 (2003) 319
Arnaudon, D. B668 (2003) 469 Boer, D. B667 (2003) 201
Arutyunov, G. B663 (2003) 163 Bonciani, R. B661 (2003) 289
Arutyunov, G. B665 (2003) 273 Bonciani, R. B668 (2003) 3
Astefanesei, D. B665 (2003) 594 Bonelli, G. B669 (2003) 159
Boonekamp, M. B669 (2003) 277
Avan, J. B668 (2003) 469
Bourbonnais, C. B663 (2003) 568
Axenides, M. B662 (2003) 170
Bouttier, J. B663 (2003) 535
Brandenburg, A. B667 (2003) 394
Bais, F.A. B666 (2003) 243
Brax, P. B667 (2003) 149
Balog, J. B668 (2003) 506
Brignole, A. B666 (2003) 105
Banerjee, R. B668 (2003) 179
Brower, R.C. B661 (2003) 344
Barbieri, R. B663 (2003) 141
Brower, R.C. B662 (2003) 393
Barbieri, R. B668 (2003) 273 Bruckmann, F. B666 (2003) 197
Bardakci, K. B661 (2003) 235 Buchbinder, I.L. B665 (2003) 402
Bartolo, N. B667 (2003) 119 Buchmüller, W. B665 (2003) 445
Baseilhac, P. B669 (2003) 417
Basu-Mallick, B. B668 (2003) 415 Cacciari, M. B664 (2003) 299
Batchelor, M.T. B669 (2003) 385 Campanario, F. B663 (2003) 280
Beccaria, M. B663 (2003) 394 Cao, J. B663 (2003) 487
Becchi, C. B664 (2003) 371 Casas, J.A. B666 (2003) 105
Becker, K. B666 (2003) 144 Chakraborty, B. B668 (2003) 179
Becker, M. B666 (2003) 144 Chandrasekharan, S. B662 (2003) 220
Beenakker, W. B667 (2003) 359 Chapovsky, A.P. B667 (2003) 359
Beisert, N. B664 (2003) 131 Chernodub, M.N. B669 (2003) 233
Belitsky, A.V. B667 (2003) 3 Chetyrkin, K.G. B666 (2003) 289
0550-3213/2003 Published by Elsevier B.V.

doi:10.1016/S0550-3213(03)00749-1
480 Nuclear Physics B 669 (2003) 479–483
Chitov, G.Y. B663 (2003) 568 Gardi, E. B664 (2003) 299

Choi, K.-S. B662 (2003) 476 Geyer, B. B662 (2003) 531
Conde, J. B663 (2003) 365 Ghodsi, A. B661 (2003) 174
Crampé, N. B668 (2003) 469 Giannakis, I. B669 (2003) 462
Cvetič, M. B662 (2003) 89 Giedt, J. B668 (2003) 138
Giudice, G.F. B663 (2003) 377
Daleo, A. B662 (2003) 334 Giusto, S. B664 (2003) 371
Darriulat, P. B661 (2003) 3 Goldstein, K. B669 (2003) 325
Dasgupta, K. B666 (2003) 144 Golec-Biernat, K. B668 (2003) 345
de Azcárraga, J.A. B662 (2003) 185 Gomis, J. B665 (2003) 49
de Boer, J. B665 (2003) 545 González, J. B663 (2003) 605
de Haro, S. B664 (2003) 45 Gorsky, A.S. B667 (2003) 3
D’Elia, M. B661 (2003) 139 Gracey, J.A. B662 (2003) 247
Demasure, Y. B661 (2003) 153 Gracey, J.A. B667 (2003) 242
Denner, A. B662 (2003) 299 Graham, N. B665 (2003) 623
Derkachov, S.É. B661 (2003) 533 Groot Nibbelink, S. B663 (2003) 60
Deser, S. B662 (2003) 379 Groot Nibbelink, S. B665 (2003) 236
Di Bari, P. B665 (2003) 445 Grozin, A.G. B663 (2003) 280
Di Francesco, P. B663 (2003) 535 Grozin, A.G. B666 (2003) 289
Dinh, P.N. B661 (2003) 3 Guan, X.-W. B669 (2003) 385
Dobashi, S. B665 (2003) 94 Guitter, E. B663 (2003) 535
Doikou, A. B668 (2003) 447 Gustavsson, A. B667 (2003) 111
Doikou, A. B668 (2003) 469
Dolan, F.A. B665 (2003) 273
Haack, M. B669 (2003) 3
Dorey, P. B661 (2003) 425
Haba, N. B669 (2003) 381
Dorey, P. B661 (2003) 464
Hagiwara, K. B668 (2003) 364
Dorogovtsev, S.N. B666 (2003) 396
Hall, L.J. B663 (2003) 141
Dotsenko, V.S. B664 (2003) 477
Harada, M. B669 (2003) 381
Dubovsky, S.L. B664 (2003) 407
Harmark, T. B662 (2003) 3
Dung, N.T. B661 (2003) 3
Heitger, J. B669 (2003) 173
Herdeiro, C.A.R. B665 (2003) 189
Emmanuel-Costa, D. B661 (2003) 62
Hernández, L. B663 (2003) 365
Engquist, J. B664 (2003) 439
Hieu, B.D. B661 (2003) 3
Espinosa, J.R. B666 (2003) 105
Hikida, Y. B669 (2003) 57
Etesi, G. B662 (2003) 511
Evlampiev, K. B662 (2003) 120 Hiller, J.R. B661 (2003) 99
Hofmann, R. B668 (2003) 151
Faisst, M. B665 (2003) 649 Holland, K. B668 (2003) 207
Falkowski, A. B667 (2003) 149 Hollik, W. B666 (2003) 305
Farhi, E. B665 (2003) 623 Honecker, G. B666 (2003) 175
Feng, B. B661 (2003) 113 Hosotani, Y. B669 (2003) 381
Feverati, G. B663 (2003) 409 Hou, B.Y. B663 (2003) 467
Floratos, E. B662 (2003) 170 Huber, P. B665 (2003) 487
Foerster, A. B669 (2003) 385 Huber, S.J. B666 (2003) 269
Forger, M. B667 (2003) 435 Hwang, K. B662 (2003) 476
Frappat, L. B668 (2003) 469
Freidel, L. B662 (2003) 279 Ichinose, I. B663 (2003) 520
Freitas, A. B666 (2003) 305 Imai, T. B665 (2003) 520
Frolov, S. B668 (2003) 77 Imbimbo, C. B664 (2003) 371
Fursaev, D.V. B664 (2003) 403 Intriligator, K. B667 (2003) 183
Fyodorov, Y.V. B664 (2003) 457 Isaev, A.P. B662 (2003) 461
Ito, M. B668 (2003) 322
Ganjali, M.A. B661 (2003) 174 Ivanov, N.Ya. B666 (2003) 88
García Canal, C.A. B662 (2003) 334 Izquierdo, J.M. B662 (2003) 185
Nuclear Physics B 669 (2003) 479–483 481
Jack, I. B662 (2003) 63 Lipatov, L.N. B661 (2003) 19

Jacobsen, J.L. B664 (2003) 477 Livine, E.R. B663 (2003) 231
Jaffe, R.L. B665 (2003) 623 Louapre, D. B662 (2003) 279
Janik, R.A. B661 (2003) 153 Lowe, D.A. B667 (2003) 55
Janssen, B. B669 (2003) 363 Lowe, D.A. B669 (2003) 325
Jones, D.R.T. B662 (2003) 63 Lozano, Y. B669 (2003) 363
Jung, E. B669 (2003) 306 Lü, H. B662 (2003) 89
Lü, H. B668 (2003) 237
Kamimura, K. B662 (2003) 491 Ludwig, A.W.W. B661 (2003) 577
Kaminsky, K. B663 (2003) 33 Lüst, D. B663 (2003) 319
Kanaki, A. B667 (2003) 359 Lynker, M. B667 (2003) 484
Kawai, H. B664 (2003) 185
Kawamura, Y. B669 (2003) 381 Maillard, T. B662 (2003) 40
Kazakov, V.A. B667 (2003) 90 Majumdar, P. B664 (2003) 213
Kehagias, A. B662 (2003) 170 Manashov, A.N. B661 (2003) 533
Khater, W. B661 (2003) 209 Mannel, T. B663 (2003) 280
Khemani, V. B665 (2003) 623 Mannel, Th. B665 (2003) 367
Kim, J.E. B662 (2003) 476 Manvelyan, R. B667 (2003) 413
Kim, S.-H. B669 (2003) 306 Marandella, G. B663 (2003) 141
Kimura, Y. B664 (2003) 512 Marandella, G. B668 (2003) 273
King, S.F. B662 (2003) 359 Marchi, M. B665 (2003) 425
Kitazawa, Y. B665 (2003) 520 Martin, L.C. B668 (2003) 335
Klebanov, I.R. B664 (2003) 3 Martucci, L. B666 (2003) 230
Kleinert, H. B666 (2003) 361 Masina, I. B661 (2003) 365
Knechtli, F. B663 (2003) 3 Mastrolia, P. B661 (2003) 289
Koizumi, K. B669 (2003) 417 Mastrolia, P. B664 (2003) 341
Korchemsky, G.P. B661 (2003) 533 Matarrese, S. B667 (2003) 119
Korchemsky, G.P. B667 (2003) 3 Mathur, S.D. B661 (2003) 344
Körs, B. B669 (2003) 3 Mawatari, K. B668 (2003) 364
Kostov, I.K. B667 (2003) 90 Meggiolaro, E. B665 (2003) 425
Kotikov, A.V. B661 (2003) 19 Melles, M. B662 (2003) 299
Kraus, E. B661 (2003) 83 Melnikov, K. B662 (2003) 409
Kristjansen, C. B664 (2003) 131 Mendes, J.F.F. B666 (2003) 396
Krykhtin, V.A. B665 (2003) 402 Metzger, S. B663 (2003) 343
Kühn, J.H. B665 (2003) 649 Meyer, H.B. B668 (2003) 111
Kumar, K. B668 (2003) 179 Minkowski, P. B668 (2003) 207
Kuroki, T. B664 (2003) 185 Moore, J.E. B661 (2003) 514
Kurth, M. B669 (2003) 173 Morita, T. B664 (2003) 185
Kurylov, A. B667 (2003) 321 Moriyama, S. B665 (2003) 49
Kutasov, D. B666 (2003) 56 Morozov, A. B666 (2003) 311
Mulders, P.J. B667 (2003) 201
Lalak, Z. B667 (2003) 149 Mülsch, D. B662 (2003) 531
Laliena, V. B668 (2003) 403
Larosa, M. B667 (2003) 261 Nadolsky, P.M. B666 (2003) 3
Laugier, A. B662 (2003) 40 Nadolsky, P.M. B666 (2003) 31
Lee, H.K. B665 (2003) 153 Naón, C.M. B663 (2003) 591
Lee, K. B665 (2003) 179 Nason, P. B667 (2003) 394
Lehners, J.-L. B661 (2003) 273 Nastase, H. B667 (2003) 55
Leonhardt, T. B667 (2003) 413 Navarro, I. B666 (2003) 105
Li, Y.Q. B666 (2003) 337 Niarchos, V. B666 (2003) 56
Lin, H.-Q. B663 (2003) 487 Nicolai, H. B668 (2003) 167
Lin, H.Q. B666 (2003) 337 Niemi, A.J. B666 (2003) 311
Lindner, M. B665 (2003) 487 Nilles, H.P. B665 (2003) 236
Lindström, U. B662 (2003) 147 Nógrádi, D. B666 (2003) 197
482 Nuclear Physics B 669 (2003) 479–483
Nogueira, F.S. B666 (2003) 361 Rey, S.-J. B669 (2003) 57

Nomura, Y. B663 (2003) 141 Riccioni, F. B663 (2003) 60
Nyawelo, T.S. B663 (2003) 60 Riotto, A. B667 (2003) 119
Roček, M. B662 (2003) 147
Okawa, Y. B663 (2003) 33 Rodríguez-Gómez, D. B669 (2003) 363
Okui, T. B663 (2003) 141 Roiban, R. B664 (2003) 45
Oleari, C. B667 (2003) 394 Roiban, R. B665 (2003) 211
Olechowski, M. B665 (2003) 236 Royon, C. B669 (2003) 277
Oliver, S.J. B663 (2003) 141 Rühl, W. B667 (2003) 413
Onorato, P. B663 (2003) 605 Rupp, C. B661 (2003) 83
Ooguri, H. B663 (2003) 33 Russo, R. B669 (2003) 207
Oriti, D. B663 (2003) 231
Osborn, H. B665 (2003) 273 Sakaguchi, M. B662 (2003) 491
Oshimo, N. B668 (2003) 258 Saleur, H. B663 (2003) 443
Osland, P. B661 (2003) 209 Salvay, M.J. B663 (2003) 591
Owen, A.W. B663 (2003) 197 Samtleben, H. B668 (2003) 167
Samukhin, A.N. B666 (2003) 396
Paccetti Correia, F. B668 (2003) 151 Santachiara, R. B664 (2003) 477
Palomares-Ruiz, S. B669 (2003) 255 Sasaki, R. B663 (2003) 467
Papadopoulos, C.G. B667 (2003) 359 Sassot, R. B662 (2003) 334
Papucci, M. B663 (2003) 141 Savoy, C.A. B661 (2003) 365
Papucci, M. B668 (2003) 273 Schimmrigk, R. B667 (2003) 484
Park, D.K. B669 (2003) 306 Schmidt, M.G. B668 (2003) 151
Park, J. B665 (2003) 49 Schubert, C. B668 (2003) 335
Parvizi, S. B661 (2003) 174 Schwetz, T. B665 (2003) 487
Pearce, P.A. B663 (2003) 409 Sciuto, S. B669 (2003) 207
Pepe, M. B668 (2003) 207 Scrucca, C.A. B669 (2003) 128
Periwal, V. B667 (2003) 484 Segal, A.Y. B664 (2003) 59
Peschanski, R. B669 (2003) 277 Seidensticker, T. B665 (2003) 649
Petcov, S.T. B669 (2003) 255 Seki, S. B661 (2003) 257
Phuong, P.T. B661 (2003) 3 Serbo, V.G. B662 (2003) 409
Picón, M. B662 (2003) 185 Serone, M. B669 (2003) 128
Pijlman, F. B667 (2003) 201 Sezgin, E. B664 (2003) 439
Pinsky, S.S. B661 (2003) 99 Sezgin, E. B668 (2003) 237
Pittau, R. B667 (2003) 359 Sfetsos, K. B669 (2003) 103
Plümacher, M. B665 (2003) 445 Shafi, Q. B665 (2003) 469
Pocklington, A. B661 (2003) 425 Sharpe, E. B664 (2003) 21
Pocklington, A. B661 (2003) 464 Shi, K.-J. B663 (2003) 487
Pons, J.M. B665 (2003) 129 Shifman, M. B664 (2003) 233
Pope, C.N. B662 (2003) 89 Shifman, M. B667 (2003) 170
Pope, C.N. B668 (2003) 237 Shik, H.Y. B666 (2003) 337
Pozzorini, S. B662 (2003) 299 Shimada, H. B665 (2003) 94
Pradisi, G. B667 (2003) 261 Shin, H. B669 (2003) 78
Prokushkin, S. B666 (2003) 144 Sibiryakov, S.M. B664 (2003) 407
Sibold, K. B661 (2003) 83
Radu, E. B665 (2003) 594 Siegel, W. B665 (2003) 179
Ragoucy, E. B668 (2003) 469 Silva, P.J. B666 (2003) 230
Ramgoolam, S. B667 (2003) 55 Silvestrini, L. B669 (2003) 128
Ramsey-Musolf, M.J. B667 (2003) 321 Sin, S.-J. B667 (2003) 310
Ravindran, V. B665 (2003) 325 Singh, H. B661 (2003) 394
Remiddi, E. B661 (2003) 289 Skenderis, K. B665 (2003) 3
Remiddi, E. B664 (2003) 341 Smith, J. B665 (2003) 325
Ren, H.-C. B669 (2003) 462 Sokatchev, E. B663 (2003) 163
Renard, F.M. B663 (2003) 394 Sokatchev, E. B665 (2003) 273
Nuclear Physics B 669 (2003) 479–483 483
Solodukhin, S.N. B665 (2003) 545 Varela, O. B662 (2003) 185

Sommer, R. B669 (2003) 173 Vázquez-Mozo, M.A. B668 (2003) 293
Staśto, A.M. B668 (2003) 345 Veneziano, G. B667 (2003) 170
Staudacher, M. B664 (2003) 131 Veretin, O. B665 (2003) 649
Stefański Jr., B. B666 (2003) 71 Verzegnassi, C. B663 (2003) 394
Stelle, K.S. B661 (2003) 273 Villanueva Sandoval, V.M. B668 (2003) 335
Stelle, K.S. B662 (2003) 89
Striet, J. B666 (2003) 243 Walcher, J. B665 (2003) 211
Strumia, A. B663 (2003) 377 Waldron, A. B662 (2003) 379
Su, S. B667 (2003) 321 Walter, M.G.A. B665 (2003) 236
Sudbø, A. B666 (2003) 361 Walter, W. B666 (2003) 305
Sugawara, Y. B661 (2003) 191 Wang, W. B667 (2003) 349
Sugiyama, K. B669 (2003) 78 Wang, Y. B663 (2003) 487
Sundell, P. B664 (2003) 439 Wecht, B. B667 (2003) 183
Wehefritz-Kaufmann, B. B663 (2003) 443
Takayama, Y. B665 (2003) 520 Weigel, H. B665 (2003) 623
Takayanagi, T. B662 (2003) 3 Weiglein, G. B666 (2003) 305
Takeda, K. B663 (2003) 520 Weisz, P. B668 (2003) 506
Talavera, P. B665 (2003) 129 Wells, J.D. B663 (2003) 123
Talavera, P. B669 (2003) 341 Wiese, K.J. B661 (2003) 577
Tamai, K. B668 (2003) 385 Wiese, U.-J. B668 (2003) 207
Tan, C.-I. B661 (2003) 344 Wiesenfeldt, S. B661 (2003) 62
Tan, C.-I. B662 (2003) 393
Winter, W. B665 (2003) 487
Tanaka, T. B662 (2003) 413
Winterhalder, A. B667 (2003) 435
Tatar, R. B665 (2003) 211
Witten, E. B664 (2003) 3
Tateo, R. B661 (2003) 425
Wolff, U. B663 (2003) 3
Tateo, R. B661 (2003) 464
Wu, J.-B. B663 (2003) 79
Tavartkiladze, Z. B665 (2003) 469
Wu, J.-B. B663 (2003) 95
Tavartkiladze, Z. B668 (2003) 151
Wu, X. B665 (2003) 153
Taylor, M. B665 (2003) 3
Taylor, T.R. B663 (2003) 319
Yang, W.-L. B663 (2003) 467
Teper, M.J. B668 (2003) 111
Yılmaz, N.T. B664 (2003) 357
Thao, N.T. B661 (2003) 3
Yokoya, H. B668 (2003) 364
Thieu, D.Q. B661 (2003) 3
Thorn, C.B. B661 (2003) 235 Yoneya, T. B665 (2003) 94
Thuan, V.V. B661 (2003) 3 Yoshida, K. B669 (2003) 78
Tobe, K. B663 (2003) 123 Yuan, C.-P. B666 (2003) 3
Tomino, D. B665 (2003) 520 Yuan, C.-P. B666 (2003) 31
Tonel, A.P. B669 (2003) 385 Yue, C. B667 (2003) 349
Trittmann, U. B661 (2003) 99 Yung, A. B662 (2003) 120
Tseytlin, A.A. B664 (2003) 247
Tseytlin, A.A. B668 (2003) 77 Zakharov, V.I. B669 (2003) 233
Tsutsui, I. B662 (2003) 447 Zeuthen–Rome (ZeRo) Col-
laboration B664 (2003) 276
Uchino, T. B662 (2003) 447 Zheng, Z.-J. B663 (2003) 79
Uraltsev, N. B665 (2003) 367 Zheng, Z.-J. B663 (2003) 95
Zhou, H.-Q. B669 (2003) 385
van Baal, P. B666 (2003) 197 Zhu, C.-J. B663 (2003) 79
van Holten, J.W. B663 (2003) 60 Zhu, C.-J. B663 (2003) 95
van Neerven, W.L. B665 (2003) 325 Znojil, M. B662 (2003) 554
van Nieuwenhuizen, P. B662 (2003) 147 Zong, H. B667 (2003) 349

Nucl - Phys.B v.669 PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Nucl - Phys.B v.669 PDF

Caricato da

Copyright:

Formati disponibili

Nuclear Physics B 669 (2003) 3–56

An orientifold with fluxes and branes via T-duality

PACS: 11.25.-w; 11.25.Mj; 04.65.+e

E-mail addresses: berg@roma2.infn.it (M. Berg), haack@roma2.infn.it (M. Haack), kors@lns.mit.edu

0550-3213/$ – see front matter  2003 Published by Elsevier B.V.

[2,3], moduli stabilization through flux-induced potentials [4–48], Randall–Sundrum-like

2. Construction of the action via T-duality

with ω3 the gauge CS 3-form

5 The issue of the self-duality constraint on F will be dealt with later.

2.1. T-duality rules

and Fp is now defined more generally as

6 A similar democratic version of type II supergravity was discussed in [77].

2.2. T-duality without vectors

which we split into

SI [A = 0] = SNSNS + SRR . (24)

|Fp |2 = |Fp |2Gij ,gµν (27)

Fp = Fp |dy i →dy i −Aiµ dx µ . (28)

(29) in (23) maps the RR kinetic term according to

2.3. Coupling to vector fields

Ãaµ = Aaµ − Aai Aiµ , (53)

Ãaµ → Ãaµ + ∂µ / a + f abc Ãbµ / c , Aai → Aai + f abc Abi / c , (59)

Together with (21) we can then write

with the abbreviation

Fj µν = 2∂[µ Cν]j , (69)

(ω3 )µνρ = 6Aa[µ ∂ν Aaρ] + 2f abc Aaµ Abν Acρ ,

S = SEH + Skin + Spot + SCS . (81)

2.4. Additional modifications from the non-Abelian DBI

2.5. The action in the Einstein frame

2.6. Comparison to gauged supergravity

18 We are using gauge group indices a, b, . . . instead of i, j, . . . and A, B, . . . instead of I, J, . . . to avoid

F ABC = Lα fαABC = Lα fαΛΣΓ EΛ

C ABC = L2 f abc q aA q bB q cC = L2 f abc a aΛ a bΣ a cΓ EΛ

5F3 − e−Φ H3 = 0. (117)

21 We absorb factors of p! into our definition of the norm (A.1).

Rewriting this via (103) one finds a relation

f1ΛΣΓ − C0 f2ΛΣΓ + e−Φ (5f2 )ΛΣΓ = 0, (118)

3.1. The role of the superpotential

proposed by independent arguments in [11,12], with Ω3 the holomorphic 3-form. In that

of complex structure and Kähler moduli as on a Calabi–Yau. On a Calabi–Yau manifold,

The (2, 0) part can be written24

4. Vacua with fluxes and warped metric

24 Note the misprint in formula (12) of [35].

4.1. Generalized ansatz and modified action

Sop = SD3 + SO3 . (136)

4.2. Equations of motion and constraints

and explicitly we have

4.3. Large volume scaling limit: separation of mass scales

5. Generalization of the open string sector

as effective world-volume action. Considering D5- or D7-branes should pose no further

5.1. Supersymmetry and κ-symmetry

In this section we discuss the conditions for preserving world-volume supersymmetry

Γ = e−a/2(Γ0...p ⊗ iσ2 )ea/2 (155)

5.2. Calibrated branes

which leads to the condition

It is a great pleasure to acknowledge stimulating discussions and helpful explanations

A.1. Conventions and notation

Tangent-frame indices are generally denoted I, J, . . . = 0, . . . , 9, which decompose

A.2. Einstein equations with warp factors

A.3. Energy–momentum tensors

Note Tij3-flux g ij = TI5J-flux g I J = 0. If one specializes to b = 1, α = −∆−2 and ĝµν = ηµν

One then also has TIYM

A.4. T-duality of RR forms

where ψ i† and ψi are raising and lowering operators

comoving observer located at O. The