Nucl - Phys.B v.646

Nuclear Physics B 646 (2002) 323
www.elsevier.com/locate/npe
Signals of anomaly mediated supersymmetry

breaking in an e collider
Debajyoti Choudhury a, , Dilip Kumar Ghosh b , Sourov Roy c
a Harish-Chandra Research Institute, Chhatnag Road, Jhusi, Allahabad 211 019, India
b National Center for Theoretical Sciences, National Tsing Hua University, Hsinchu, Taiwan, ROC
c Department of Physics, TechnionIsrael Institute of Technology, Haifa 32000, Israel
Received 10 September 2002; accepted 8 October 2002
Abstract
We study the signatures of minimal anomaly mediated supersymmetry breaking in an e collider.
We demonstrate that the associated production of a sneutrino with the lightest chargino leads to a
substantially large signal size. The background is negligibly small, though. Even more interestingly,
a measurement of the fundamental supersymmetry breaking parameters could be possible.
2002 Published by Elsevier Science B.V.
PACS: 12.60.Jv; 13.10.+q; 14.80.Ly
Keywords: Supersymmetry breaking; e collider
1. Introduction
The question of supersymmetry (SUSY) breaking is of burning relevance in high energy
physics today. The most general version of the minimal supersymmetric standard model
(MSSM), with the attendant large number of arbitrary SUSY breaking soft parameters,
makes itself intractable for experimental searches. However, if the mechanism of SUSY
breaking were known, then it would be possible to reduce these large number of parameters
into a much smaller set. With the corresponding ordering in the mass spectrum, the nature
of the lightest supersymmetric particle (LSP) would be determined, and the decay chains
established, thereby making the model much more predictive. Several such supersymmetry
* Corresponding author.
E-mail addresses: debchou@mri.ernet.in (D. Choudhury), dilip.ghosh@cern.ch (D.K. Ghosh),

roy@physics.technion.ac.il (S. Roy).
0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 9 0 2 - 1
D. Choudhury et al. / Nuclear Physics B 646 (2002) 323
breaking mechanisms have been discussed in the literature, along with their distinctive
phenomenological signatures. Anomaly mediated supersymmetry breaking (AMSB) [1,2]
is one such possibility and has attracted a lot of attention in recent times. Building on the
basic idea, wherein the supersymmetry breaking is conveyed to the observable sector by the
super-Weyl anomaly, a whole class of models have been constructed [320], and many of
the phenomenological implications discussed [5,16,2132]. For example, the characteristic
signatures of the minimal anomaly mediated supersymmetry breaking model (mAMSB)
have been studied in the context of hadronic colliders [2229], as well as for high energy
linear colliders, whether of the e+ e type [30,31], or photonphoton colliders [32]. In this
paper we study the unique signatures of the mAMSB model in an e collider.
In its original version, the AMSB scenario consisted of a higher-dimensional supergravity theory wherein the hidden sector and observable sector superfields are localized
on two distinct parallel three-branes separated by a distance rc (rc is the compactification radius) in the extra dimension [1]. Below the compactification scale (c rc1 ), only
four-dimensional supergravity fields are assumed to propagate in the bulk. Even if there
are additional bulk fields above this scale, their effects would, typically, be suppressed by
a factor em/c . As there is no tree level coupling between the hidden sector fields and
those in the observable sector, it had been assumed that the flavor changing neutral current
(FCNC) processes would be suppressed naturally, thereby solving a longstanding problem
of supersymmetric theories. Recently, though, it has been shown that the physical separation between the visible and hidden sectors in extra dimension is not sufficient to suppress
the FCNC processes except in some special cases [17]. However, it is possible to construct
models of AMSB, even in a completely four-dimensional framework, that can circumvent
the flavour problem [18,19]. For the purposes of the present study, we do not need to delve
into such details and would rather be concentrating on the minimal scenario of Ref. [1].
If one were to describe the AMSB scenario in terms of a four-dimensional effective
theory, applicable below c , then a rescaling transformation can be defined so as to
eliminate, from the classical Lagrangian, any tree-level interaction (except for the -term
in the superpotential) connecting the supergravity fields with the visible sector matter
fields. Such a scaling transformation, however, is anomalous and hence supersymmetry
breaking is communicated from the hidden sector to the visible sector through the
super-conformal anomaly [1]. The supersymmetry breaking soft mass parameters for the
gauginos and the scalars are generated at the same order in the corresponding gauge
coupling strength. The analytical expressions for the scalar and gaugino masses, in terms
of the supersymmetry breaking parameters, are renormalization group (RG) invariant, and,
thus, can be computed at the low-energy scale in terms of the appropriate beta functions
and anomalous dimensions. The minimal AMSB scenario suffers from a glaring problem
though. At low energies, it is beset with the existence of tachyonic sleptons. Several
solutions to this problem exist. In this work, we shall consider the minimal AMSB model
wherein a constant term m20 is added to all the scalar squared masses thereby making the
slepton mass-squareds sufficiently positive. While this may seem to be an ad hoc step,
models have been constructed that naturally lead to such an eventuality. A consequence is
that the RG invariance of the expression for scalar masses is lost and hence one needs to
consider the corresponding evolution down to the electroweak scale.
The minimal AMSB model has several unique features. The gravitino is very heavy, its
mass being in the range of tens of TeV. Left and right selectrons and smuons are nearly
mass-degenerate while the staus split into two distinct mass eigenstates. But perhaps the
most striking feature is that both the lightest supersymmetric particle (LSP) 10 and the
lighter chargino ( 1 ) are predominantly winos and hence nearly mass degenerate. Loop
corrections as well a small gauginohiggsino mixing at the tree level do split the two, but
the consequent mass difference is very small:
M < 1 GeV. The dominant decay mode of
the lighter chargino is 1 10 + and this long-lived chargino would typically result
in a heavily ionizing charged track and/or a characteristic soft pion in the detector [33].
Signals of supersymmetry in an e collider have been discussed in various contexts [3442]. In this paper, we consider the process e 1 to look for signals of
anomaly mediated supersymmetry breaking in such a collider. This choice has some advantages over the possibilities at an e+ e collider. The dominant production channel at
the latter machine, namely a 1+ 1 pair, is notoriously difficult to tag onto, and even if
detected, is hardly amenable to mass determination. And as the sleptons tend to be significantly heavier than 1+ , at least for a very large part of the parameter space, the 1
threshold tends to be quite a bit lower than those for + or . Moreover, the cross section for the first-mentioned process is, generically, much higher, and even more so close to
the production threshold.
Once produced, the sneutrino may decay into either an (e 1+ ) pair or a ( 10 ) pair.
Concentrating upon the former, we are left with a fast e (which serves as the trigger),
two heavily ionizing charged tracks coming from the long-lived 1 and/or two visible soft
pions with opposite charges1 and a large missing transverse momentum (/
p T ). This is a
very unique and distinct signature of anomaly mediated supersymmetry breaking and does
not readily arise in either of mSUGRA or GMSB scenarios.
What of other channels at an e collider itself? Continuing with the same production
process, had the selectron decayed into the alternate channel, namely a ( 10 ) pair, we
would have been left with + p
/ T final state, possibly associated with a single heavily
ionized track. The lack of a reliable trigger renders this channel of little use. The associated
production of a left2 selectron and the lightest neutralino (e eL 10 ), on the other hand,
would result in the same signal (e + p
/ T ) as in the case of mSUGRA. Though eminently
detectable, this is of hardly any use in establishing the nature of supersymmetry breaking,
and hence we desist from discussing it any further. It is, of course, true that, for some
choices of the SUSY parameters, the selectron may also decay into the heavier neutralinos
(eL e 20 ) thereby leading to final states with multifermions and p
/ T . However, the
signal cross section in this case will be small on account of the suppressions due to the
various branching ratios. And finally, one may also consider associated production of
heavier charginos (with sneutrino) and neutralinos (with selectron) and their subsequent
cascade decays. The signals for such processes would be rather complex, though. And the
1 While it might not very easy to measure the pion charges, they are nonetheless distinguishable, if only in a
statistical sense, from their rapidity distribution.
2 Note that e e 0 is suppressed in this model on account of 0 having a vanishingly small bino
R 1
1
component.
production cross sections typically smaller, both on account of reduced phase space as well
as smaller couplings. It may, thus, be said that the channel of our choice is the simplest as
well as the most promising one.
The paper is organized as follows. In Section 2, we give a brief description of the e
collider. Section 3 discusses the spectrum and the couplings within the mAMSB model
and the constraints on the parameter space from various experimental and theoretical
considerations. Numerical results of our computations and their discussions are presented
in Section 4. In Section 5, we discuss the possibility of determining the supersymmetry
breaking parameters. Finally, we conclude in Section 6.
2. e collider and the photon spectrum
While it is quite apparent that maximizing the signal cross sections implies using
perfectly polarized electron and photon beams, in reality, though, perfect polarizations
is almost impossible. Furthermore, even near monochromaticity for high energy photon
beams is extremely unlikely. In fact, the only known way to obtain very high energy photon
beams is to induce laser back-scattering off an energetic e beam [43]. The reflected
photon beam carries off only a fraction (y) of the e energy with
ymax =
z
,
1+z
4Eb EL
bL
,
cos2
m2e
2
(1)
where Eb(L) are the energies of the incident electron (positron) beam and the laser,
respectively, and bL is the incidence angle. In principle, one can increase the photon
energy by increasing the energy of the laser beam. However, a large EL (or, equivalently, a
large z) also enhances the probability of electronpositron pair creation through laser and
scattered-photon interactions,
and consequently results in beam degradation. An optimal
choice is z = 2(1 + 2), and this is the value that we adopt in our analysis.
The cross sections for a realistic electronphoton collider can then be obtained by
convoluting the fixed-energy cross sections (s , P , Pe ) with the appropriate photon
spectrum:

dn
(s) = dy ds (Pb , PL ) (s , P , Pe )(s ys),
(2)
dy
where the photon polarization is itself a function of Pb,L and the momentum fraction,
viz. P = P (y, Pb , PL ). For simplicity, we shall only consider circularly polarized lasers
scattering off polarized electron (positron) beams. The corresponding number-density n(y)
and average helicity for the scattered photons are then given by [43]
2 2
dn
= 2
C(y),
dy me zC

1
1
y
2
+ y(2r 1) PL (2r 1) 1 y +
P (y) =
Pb
,
C(y)
1y
1y
y
+ (1 y) 4r(1 r) 2Pb PL rz(2r 1)(2 y),
C(y)
1y
(3)
where r y/z/(1 y) and the total Compton cross section C provides the normalization.
A further experimental issue needs to be concerned at this stage. While the photon
spectrum of Eq. (3) has a long low-energy tail, in a realistic situation it might be that
they cannot participate in any interaction. There is a one-to-one relationship between the
energy of the back-scattered photons and their angle with respect to the direction of the
initial electron: harder photons are emitted at smaller angles whereas softer photons are
emitted at larger angles. For example, for small deflection angles, we have
(y)
me
Eb
z
z 1.
y
(4)
Since the photons are distributed according to an effective spectrum (Eq. (3)), this relation
effectively throws out the low energy photons, these being produced at too wide an angle
to contribute significantly to any interaction. The exact profile of this effective spectrum,
though, is not so simple and depends somewhat on the shape of the electron beam, and the
conversion distance, i.e., the distance between the interaction point and the point where
the laser photons are back-scattered. In the absence of a detailed (and machine specific)
study of this effect, we are, unfortunately, not in a position to include it in our simulations.
However, as the study of Ref. [38] had indicated, and as we have checked, neglecting this
effect does not change the total signal cross section to any significant extent. Elimination
of the low energy photons, however, can help in reducing the backgrounds, and thus our
approximation is a conservative one.
Before we end this section, we would like to point out that while perfect polarization for
the laser beam is relatively easy to obtain, the same is not true for electrons or positrons.
In our studies with polarized beams, we shall then use |PL | = 1, |Pb | = |Pe | = 0.8,
which, once again, reflects a conservative choice. Since we seek to produce the sneutrino,
it follows that the e should be preferentially left-polarized, or Pe = 0.8. Similarly,
choosing the laser and the e beam to be oppositely polarized (PL Pb < 0) improves the
monochromaticity of the outgoing photons [44]. For the sake of completeness, we shall use
both choices of polarizations consistent with PL Pb < 0. While the total cross sections
are obviously dependent on the polarization choice, the efficiencies of the kinematical cuts
are expected not to be.
3. Model parameters and constraints

The minimal AMSB model has a high degree of predictivity as it is described by
just three parameters (apart from the SM parameters, of course): the gravitino mass
m3/2 , the common scalar mass parameter m0 and tan , the ratio of the two Higgs
vacuum expectation values. In addition, there is a discrete variable, namely the sign of
the Higgs mass term (). As mentioned earlier, the soft supersymmetry breaking terms
in the effective Lagrangian are then determined solely in terms of the gauge (gi ) and
Yukawa (ya ) couplings. Denoting the generic beta-functions and anomalous dimension
by g (g, y) dg/dt, y (g, y) dy/dt and (g, y) d ln Z/dt (t being the logarithmic
scale variable), respectively, we have, for the gaugino () masses

M =
g
m3/2,
g
(5)
where the appropriate gauge coupling and -function are to be considered. Similarly, for
the trilinear soft breaking parameters, one has
Ay =
y
m3/2 .
y
(6)
The scalar masses, on the other hand, receive contributions from more than one source.
Apart from the individual contributions from each of the relevant gauge couplings, there is
also the universal contribution m20 . Symbolically, then,

1
2
2
g +
y m23/2 .
mf = m0
(7)
4
g
y
The detailed expressions for the gaugino masses at the one loop level and the squared
masses for the Higgs and the other scalars at the two loop level can be found in Refs. [23,
31,45].
In our analysis, we use two-loop renormalization group equations (RGE) [46] to evolve
the gauge and Yukawa couplings from the unification scale (MG 2 1016 ) down to
the electroweak scale. For the gauge and Yukawa couplings, the boundary conditions
are determined at the weak scale (with 3 (MZ ) 0.118), while for the scalar masses
these are given at the unification scale vide Eq. (7). The magnitude of the Higgsino
mass parameter is computed from the complete one-loop effective potential [47] and
imposing the requirement of radiative electroweak symmetry breaking. The optimal choice
for the renormalization scale is expressible in terms of the masses of the top-squarks, viz.
Q2 = mt1 mt2 . We also include the supersymmetric QCD corrections to the bottom-quark
mass [48] since this plays a significant role for large tan .
A particularly interesting feature of the mAMSB model is that the ratios of the gaugino
mass parameters, at low energies, turn out to be
|M1 | : |M2 | : |M3 | 2.8 : 1 : 7.1.
(8)
In Eq. (8), M1 , M2 and M3 refer to the U (1), SU(2) and SU(3) gaugino mass parameters,
respectively. An immediate consequence is that the lighter chargino 1 and the lightest
neutralino 10 are both almost exclusively a wino and, hence, nearly degenerate. A small
mass difference is generated though from the tree-level gauginohiggsino mixing as well
as from the one-loop corrections to the chargino and the neutralino mass matrices [23,49].
The mass splitting has an approximate form:

4 tan2
2
MW
M2 MW
W
2
M m + m 0 =
,
sin
2
1
+
O
1
1
(M1 M2 )2
M1
2
2
MW
MZ
M2
2
+
(9)
f
cos W f
,
2
2
M2
M22
sin W
with

2
x x2
1
x
1 (1 x) 4x x
2
f (x) +
ln(x) +
4x x tan
.
1+
4
8
2
2
3x x 2
(10)
For the range of m0 and m3/2 that we would be considering in our analysis,
M
500 MeV. And, for very large M2 , the mass difference reaches an asymptotic value of
165 MeV.
To determine the parameter space allowed to the theory, several experimental constraints
need to be considered, the most important ones being:
10 must be the LSP;
m > 86 GeV, when this chargino almost degenerate with the lightest neutralino
1
[50].3
The last constraint serves to rule out relatively low values of m3/2 irrespective of the value
of m0 . The width of this disallowed band depends, of course, on tan and sgn(). The
first constraint, on the other hand, restricts the parameter space through a linear relation:
m3/2 < am0 + b, where the numerical values of the constants a, b, depend, once again,
on tan and sgn(). Typically, the maximum possible value of m3/2 for a given m0 is
a decreasing function of tan . Note that the LEP2 constraints on the lighter stau mass,
namely m1 > 82 GeV [51], are subsumed by the ones listed above. A detailed discussion
of such issues can be found in Ref. [45].
Apart from direct bounds, one must also consider the constraints imposed on virtual
exchange contributions to low energy observables. The recent measurement of the muon
anomalous magnetic moment (g 2) is a case in point. The resultant constraints on the
mAMSB model parameters have been considered by several authors [24,45,52,53]. The
numerical results of these papers need to be modified though. For one, the light by light
hadronic contribution to (g 2) has since been reevaluated resulting in a reversal of
the sign of this particular contribution [54] and a consequent reduction of the discrepancy
with the SM result. Even more recently, the E821 experiment has published new data that
confirms their earlier result while increasing the precision significantly [55]. It must be
borne in mind, though, that the calculation of the SM contribution to (g 2) is beset with
many remaining theoretical uncertainties, and, hence, any such constraint must be treated
with due circumspection. Constraints are also derivable [24,52,53] from the measurement
of the rare decay rate (B Xs ), but, once again, they are not too restrictive and
many a loophole exists. Additional bounds may exist if one demands that the electroweak
vacuum corresponds to the global minimum of the scalar potential [56,57]. This restriction,
however, can be evaded as long as it can be ensured that the local minimum has a life time
longer than the present age of the Universe [58].
3 The lower limit mentioned in that paper is valid for heavier sfermion masses and is slightly above (88 GeV)
the value we consider here.
10
4. Signal and backgrounds

As explained in Section 1, we would be focussing on the production process e
1 . The sneutrino may subsequently decay into a (e + 1+ ) pair or through the
invisible (e + 10 ) channel4 with the branching fractions BR(e e + 10 ) 33% and
BR(e + 1+ ) 66% relatively well determined on account of the 1+ and the 10 being
predominantly winos. Decays into the heavier charginos (neutralinos) are kinematically
forbidden over most of the parameter space and, even if they are allowed, the branching
ratios in those channels are very small. And while three body decays of the form
e 1+ are allowed in principle, in practice they are too small to be of any relevance. For
reasons of detectability, we choose to neglect the invisible channel and concentrate on the
electronchargino one. The two charginos subsequently decay into a neutralinopion pair
each. The entire schematics of the signal is presented in Fig. 1. In subsequent discussions,
we shall denote 1 as the primary chargino and 1+ (arising from the decay of ) as the
secondary chargino.
It turns out that, for most of the mAMSB parameter space, the two charged pions are
well separated from each other. In particular, when m is small compared to both se
1
and m e , the boosts imparted to the charginos make the pions appear almost back-to-back.
The neutralinos, on the other hand, escape detection, thereby giving rise to an imbalance
in momentum.
The signal then is
/T ,
e e 1 e + + + p
(11)
with the energetic electron serving as the trigger for the event. The relatively small
decay width of the charginos is manifested in heavily ionizing charged tracks (one for
each chargino) terminating inside the detector after traversing a macroscopic distance and
ending in a soft pion (with pT > 200 MeV) in the Silicon Vertex Detector (SVD) located
very close to the beam pipe. The probability that the chargino decays before travelling a
distance is given by P () = 1 exp(/L), where L = c ( ) is the average decay
length of the chargino. Although a large fraction of the events do turn out to be associated
with a 1 decay that is so prompt as to make the charged track nearly invisible (the end
productsoft is still detectable), a substantial number of events do have a reasonably
large decay lengths for which the displaced vertex XD may be visible. Consequently, one
could find a thick ionizing charged track in the first few layers of the SVD with the track
terminating to give off a soft pion, which can be observed through its impact parameter.
In the worst case, when the chargino track is not seen, our signal can still be observed by
looking at the soft pion impact parameter b .
We select the signal events according to the following criteria:
The transverse momentum of the electron must be large enough: pTe > 20 GeV.
The transverse momentum of the pions must satisfy pT > 0.2 GeV.
The total energy of the pions should not be very large though: E < 2 GeV.
4 The charged slepton is always heavier than the sneutrino on account of electroweak D-term contribution.
11
Fig. 1. Schematic diagram for the signal process e e 1 e + + + p

/T . We define 1 as the
primary chargino and 1+ as the secondary one.
The electron and both the pions must be relatively central, i.e., their pseudorapidities
must fall in the range |e, | < 2.5.
The electron and
the pions must be well-separated from each other: i.e., the isolation
variable
R (
)2 + (
)2 (where and denote the separation in rapidity and
the azimuthal angle respectively) should satisfy
R > 0.4 for each combination.
The missing transverse energy p
/ T > 20 GeV.
Any heavily ionizing charged track would be an additional bonus. While the rationale
behind most of the kinematical requirements listed above is self-evident, the importance of
the upper bound on the pion energy would become clear shortly.
4.1. The SM background
Before we discuss the ensuing signal cross section, let us first examine the possible
Standard Model backgrounds. This would also help put into perspective the cuts listed
above. A heavily ionizing track may arise only from the motion of a relatively massive
charged object. The only particles in the SM spectrum that fit the bill are ones that decay
promptly (and hence leave no ionizing track). Thus, if macroscopic tracks are seen, the
signal is essentially free of backgrounds.
What if no tracks are seen at all? The signal then comprises of just a hard electron and
two soft pions and accompanied by a measure of missing energy. The backgrounds to this
are manifold:
e e + i i . While an accurate calculation of this is a difficult task, a fair
estimate may be made by the use of scalar-QED (as well as scalar QFD). To improve
accuracy, form factors may be used to parametrize the vertices involving the pions. It is
12
here that the upper bound on the pion energy, as discussed above, assumes importance.
Since the pionphoton vertices (as well as the pionZ or pionW vertices) arise from
derivative couplings, a process such as the one we are describing would be heavily
biased towards energetic pions. With the cuts that we have imposed, the total cross
section of this particular background is O(103 fb) and hence entirely negligible.
i i with the quarks hadronizing into soft pions.
e e q q
The largest contribution to the partonic process accrues from the production mode
e e ZZ with a Z each decaying into quarks and neutrinos. It is easy to see
that the quarks from such processes would be very energetic and would not lead to
just a pair of soft pions unaccompanied by any other hadronic activity. As for the
4 ) one and hence
(with Z i i ), it is a O(W
continuum process e e q qZ
highly suppressed.
It is often argued that a signal with just a pair of soft pions, an electron trigger and no
macroscopic ionization track would suffer from large, and almost incalculable, QCD
backgrounds. However, as explained above, the particular electroweak process that
can give rise to a event topology similar to our signal, is highly suppressed and even
non-perturbative QCD corrections cannot enhance it to significant levels.
e e + with the taus decaying appropriately.
While estimating this background, it is imperative that the -polarization information
be retained in the intermediate stages of the calculation, especially since the
distribution of the decay products depend crucially on the polarization. We include all
decay channels that involve two charged pions satisfying our selection criteria. Since
some of the events thus included may be vetoed by other criteria (such as the absence
of a 0 ), our choice is a conservative one. On imposition of our cuts, the surviving
background is 0.05 fb, with the exact numbers depending on the polarization choice.
e e W + W with W .
While the W -pair production cross section is sizeable, the s and hence the s tend
to be harder. This serves to suppress the background to below O(103 fb).
4.2. The signal profile
As we have seen in the previous section, we may safely conclude that, even in the
absence of macroscopic ionization tracks, our selection criteria serves to remove virtually
all of the SM background. What of the signal, then? Clearly, both the number of signal
events and the kinematical distributions thereof would depend crucially on the sneutrino
and chargino mass. We will illustrate the case for two widely separated points in the
parameter space:
(A) m0 = 210 GeV, m3/2 = 37 TeV, tan = 5 and > 0, which leads to (m + , m 0 , m )
1
1
= (119.64, 119.39, 139.11) GeV; and
(B) m0 = 450 GeV, m3/2 = 55 TeV, tan = 30 and > 0, leading to (m + , m 0 , m )
1
1
= (186.42, 186.23, 390.07) GeV.
Let us concentrate, for the moment, on the parameter space point (A). Had we
a monochromatic photon beam, the sneutrino and the chargino would have been
13
monochromatic too. And since the electron is a product of the two-body decay of the
sneutrino into the very same chargino, the electron energy would have been strictly bound
on both sides (with the bounds determined by m and m + ), and the distribution uniform
1
within. Although this simplistic situation is modified to a degree by the spread in the
photon spectrum, the main feature is still quite discernible (see Fig. 2(a)). The transverse
momentum of the electron (see Fig. 2(b)), too, shows unmistakable signs of its kinematic
origin, with the position of the peak being governed, once again, by m and m + . With its
1
rapidity distribution mirroring that of the sneutrino, the electron is still relatively central,
though boosted in the direction of the incoming electron beam (Fig. 2(c)). The same holds
for the + arising from the sneutrino decay. The , on the other hand, is a decay product
of the primary chargino ( 1 ) and is hence boosted somewhat in the direction of the
incoming photon beam (Fig. 2(c)). This immediately suggests that the would be well
separated from both the electron and the + , a conclusion supported by the distribution
displayed in Fig. 2(d). Once again, this conclusion holds almost over the entire SUSY
parameter range under consideration. The angular separation between the electron and the
+ , on the other hand, is a sensitive function of the difference m m + and, for larger
1
mass differences, peaks at larger values. Of crucial importance are the transverse momenta
of the pions, for a minimal value is necessary for them to be detected. As expected, the
pions are very soft (see Fig. 2(e)), with the extent of the pT (as well as energy) distributions
determined by m + m 0 . The very same quantity (along with the absolute magnitude of
1
1
the masses) also determine the decay length of the charginos. The slightly faster decay of
the primary chargino in this case (see Fig. 2(f)) is but a reflection of its having a lower
average momentum.
Having discussed the kinematics, it is now relatively easy to anticipate the effects of the
cuts that we have chosen to impose. As we shall shortly see, the efficiency has a non-trivial
dependence on the superparticle masses. While, in an actual experimental context, it might
be more prudent to employ flexible cuts, we shall, nevertheless, continue to work with
those that we have already described. As Table 1 shows, the requirement on the minimum
electron transverse momentum serves to reject quite a few signal events for a small m
and/or a small difference m m + . Were both these to be large (point B), the effect is
1
marginal at best. A similar statement may also be made for the angular separation between
the electron and the + . On the other hand, a small mass difference between the chargino
and neutralino renders the requirement on the pion transverse momentum a crucial one (a
larger fraction of the signal is lost for point B than for point A). And, as expected, the rest of
the cuts imposed have only marginal effect on the signal. Finally, a remark on the chargino
track lengths. As shown in Table 1, if we insist that both charginos leave a substantial track,
we would be discarding the majority of events. While this may still be fine for relatively
lighter superparticles, when the cross sections are large, such a requirement would severely
limit the reach of the experiment. In view of our demonstration that our signal is relatively
background-free even in the absence of macroscopic tracks, it is perhaps unnecessary to
impose such a requirement, at least in the discovery stage. Of course, if a signal were
to appear, one could always go back and look for the confirmatory tracks. This is the
standpoint we shall adopt in the remainder of the article.
14
Fig. 2. Normalized kinematic distributions of signal events at a machine operating with see = 1 TeV and
beam polarization choices: PL = 1.0, Pb = 0.8, Pe = 0.8. The AMSB parameters are m0 = 210 GeV,
m3/2 = 37 TeV, tan = 5, sgn() > 0 (point A). The different panels correspond to (a) electron energy;
(b) electron pT ; (c) rapidities; (d) cone separations; (e) pion pT ; and (f) chargino decay lengths.
15
Table 1
Illustrating the effects of various cuts on the signal cross sections for two specimen points in the parameter space
and for see = 1 TeV. In either case, > 0, and the cross sections are shown both for unpolarized beams as well
as for two particular choices for beam polarizations. Whenever nonzero, |PL | = 1 and |Pb | = |Pe | = 0.8
(m0 (GeV), m3/2 (TeV), tan )
(m + , m 0 , m ) (GeV)
1
A (210, 37, 5)
(119.64, 119.39, 139.11)
B (450, 55.0, 30)

(186.42, 186.23, 390.07)
(PL , Pb , Pe )
(PL , Pb , Pe )
Total (fb)
|e | < 2.5
pTe > 20 GeV
Both | | < 2.5
Both pT > 200 MeV
e > 0.4
[1 2 ] > 0.4
p
/T > 20 GeV
Decay length for one
chargino > 4.0 cm
Decay length for both
chargino > 4.0 cm
(, +, )
(+, , )
(0, 0, 0)
(, +, )
(+, , )
(0, 0, 0)
407.1
386.7
265.5
248.7
179.6
125.5
125.5
125.0
279.3
250.9
144.5
125.5
79.5
59.5
59.5
59.0
169.3
157.0
100.6
91.6
63.1
45.7
45.7
45.4
139.3
136.5
136.1
130.7
30.6
30.3
30.3
30.1
201.9
197.0
196.4
183.2
32.3
32.0
32.0
32.0
82.3
80.4
80.2
75.9
14.4
14.2
14.2
14.2
62.5
28.9
21.8
22.1
22.2
10.7
4.6
3.48
6.9
6.37
9.68
2.70
4.3. Signal strength and the parameter space

Until now, we have restricted ourselves to an examination of the signal profile for two
specific points in the parameter space. While the gross features can be inferred from the
discussion in the previous section, it is of interest to determine magnitude of the total signal
strength as a function of the parameters involved. We aim to do this now. Clearly, it is not
possible to depict the dependence on each of the parameters, and hence we shall restrict
ourselves to two discrete values of tan while allowing m0 and m3/2 to vary freely modulo
the experimental constraints. As for the beam polarization, we make a particular choice,
namely, PL = +1, Pb = Pe = 0.8.
In Fig. 3, we display our results for a machine operating at see = 500 GeV, in the
form of scatter plots for the cross section, after imposing the cuts, in the plane spanned
by m0 and m3/2 . The panels on the left (right) correspond to low (high) tan (5 and 30).
For each choice, we also depict the dependence on the sign of the -term. In each of the
individual graphs, the region marked by X corresponds to a chargino mass of less than
86 GeV, thereby falling foul of the ALEPH bound [50]. The region Y , on the other hand,
would correspond to the 1 being the LSP, a possibility not allowed phenomenologically
if R-parity is to be conserved. We, thus, need to consider ourselves only with the wedgeshaped region enclosed by the two straight lines. Note that the extent of the region Y is
much more sensitive to the value of tan than is the case for the region X. This is easy to
understand when one considers the fact that a large tan results in making the Yukawa
coupling stronger, which, in turn, drives down the mass of the lighter , thereby rendering
it the LSP.
It is obvious that the total cross section would fall monotonically with each of the two
mass parameters, and this is amply reflected in the figure. Note that the variation with m3/2
16
Fig. 3. Scatter plots for the signal cross section (fb) in the m0 m3/2 plane for a machine operating at
see = 500 GeV and PL = +1, Pb = Pe = 0.8. The values of tan and sgn() are as indicated. The regions
marked by X are ruled out by the experimental lower limit on the chargino mass, while those marked by Y are
ruled out by the requirement of 10 being the LSP. In each panel, the top three shaded regions correspond to cross
section ranges of 0.15, 550, 50150. The lowermost region corresponds to 150470 in (a), 150390 in (b),
150335 in (c) and 150350 in (d), respectively.
is much sharper for large m0 . This, again, is easy to understand as one is progressively
squeezing the available phase space for the production process. The variations with
tan and the sign of are of a subtler origin, namely the dependence of
M (the
mass-difference between the lighter chargino and the lightest neutralino) on these two
parameters. It can be easily ascertained that, for positive , the mass-difference
M
decreases with tan . This, in turn, implies that the pions resulting from the decay of the
chargino have a higher average transverse momentum for lower tan , thereby making them
easier to detect. For negative , the effect is opposite.
17
It is clear from Fig. 3 that, for the low tan case, an experiment such as this can easily
explore m3/2 as high as 60 (50) TeV for negative (positive) . For tan = 30, on the
other hand, the maximal reach in m3/2 is approximately 40 TeV irrespective of sgn().
Similarly, the reach in m0 has little dependence on either of tan and sgn(). Finally, it
must be borne in mind that Fig. 3 has been drawn keeping in mind a moderate luminosity
( 100 fb1 ). A significantly larger integrated luminosity would, of course, allow one to
explore beyond the topmost shaded band.
In Fig. 4, we show a similar plot for a machine operating at see = 1 TeV instead.
The features are very similar, with the obvious enhancement in the reach.
Fig. 4. As in Fig. 3, but for see = 1 TeV instead. In each panel, the top three shaded regions correspond to cross
section ranges of 0.15, 550, 50150. The lowermost region corresponds to 150215 in (a), 150205 in (b),
150185 in (c) and 150195 in (d), respectively.
18
5. Parameter determination
Once the existence of a new particle has been confirmed, it is almost contingent upon the
experimentalist to determine its mass, spin, etc., and, hopefully, glean further information
as to the possible values of the parameters within a particular theoretical framework. We
shall now investigate the possibility for such studies at an e collider.
Positing that the particles produced were the sneutrino and the chargino and that the
sneutrino subsequently decayed into a similar chargino and an electron, it is easy to see
that the energy of the decay electron is strictly confined [38] within the interval
m2 m2 +
1
2(Emax
+ kmax
E
e
m2 m2 +
1
2(Emax
kmax
(12)
is the maximum possible energy that the intermediate sneutrino may have
where Emax
carried, viz.

1
(1 + ymax ) ymax s + m2 m2 +
=
Emax
1
4ymax s

2
+ (1 ymax ) ymax s + m2 m2 + 4ymax sm2 ,
(13)
1
kmax
is the corresponding momentum.

and
e
and
As s and ymax are both known, it follows that an accurate measurement of Emax
e
Emin would allow us to determine both the masses. A very precise measurement of the
endpoints is unlikely, though. The reasons are manifold. Apart from the issue of energy
resolution, one has to take into account that the sharp edges, as displayed in Fig. 2(a),
do not remain as sharp once all the cuts are imposed. Rather, there is a more gradual
tapering off at the higher end, as displayed in Fig. 5. This, obviously, could result in an
e , and, from a random survey in the parameter space, we find that
underestimation of Emax
this effect is typically 20 GeV. To account for this, we consider a conservative error
e .
contribution of 25 GeV in the measurement of Emax
e
The determination of Emin , on the other hand, does not suffer from this problem, and,
hence, we only consider an error of 5 GeV to account for the finite resolution. There is a
caveat, though. Looking at Fig. 2(a) (or, equivalently, Eqs. (12) and (13)), one immediately
e
may very well be less than 20 GeV, and hence in contradiction
sees that the true Emin
with our requirement on the minimal transverse momentum for the electron. Clearly, for
e
20 GeV. Of course, for a given collider
such points in the parameter space, Emin
configuration, it is easy to determine the part of the parameter space that is beset with
this particular problem. In the rest of this section, we rather concentrate on a point that lies
outside this region, namely the point B discussed earlier.
In Fig. 6(a), we display the two bands obtained from the measurement of the two
electron energy endpoints corresponding to point B. The measurement errors are exactly as
described above, and the curious shape of the band is but a consequence of the non-linear
nature of Eq. (12). What is interesting is that the two measurements lead to very different
constraints in the parameter space, thereby facilitating a relatively good measurement of
the two masses. The error on the sneutrino mass, thus determined, is roughly 12 GeV,
19
Fig. 5. The normalized electron energy distribution for parameter point B (see text). The dashed (solid) lines
depict the cross section before (after) all the selection criteria are imposed.
while that on the chargino mass is roughly 6 GeV. Moreover, the errors are quite correlated
(see Fig. 6(b), which displays the region of interest on an expanded scale) and hence the
combined 1 ellipse would lie well within the overlap region. Note that all this information
has been gleaned from a single process and without an energy scanning.
Having determined the masses, it is now of interest to measure the remaining parameter,
namely, tan . Clearly, kinematical distributions are essentially independent of this quantity
and one should rather consider cross section measurement, or, in other words, a counting
experiment. For this purpose, we choose to work with a moderate choice of luminosity,
viz. L = 100 fb1 . The cross sections are converted to event numbers through the relation
n = >L,
where the overall detection efficiency > is but the product of the efficiency for electron
detection (assumed to be 95%) and those for pion detection (50% each). The corresponding
error in the cross section measurement (after imposing the cuts, naturally) is easily
determined on application of Poisson (or Gaussian) statistics. Armed with this, and
for a given value of tan , one could easily determine the part of the m m + space
1
that would be consistent with the measured cross section. In Fig. 6, we display these
constraints for two particular values of the ratio tan . One might wonder at the abrupt
end for the band corresponding to tan = 30. This, however, is but a consequence of
the aforementioned constraints on the mass of 1 . Note that, while some resolution is
possible, such experiments are not overly sensitive to this parameter. It is possible that
significant improvement would occur once other production processes are considered, but
that is beyond the scope of the present work.
20
Fig. 6. (a) The determination of the sneutrino and chargino masses from a measurement of the endpoints of the
electron energy spectrum (Eq. (12)). The width of the bands correspond to the error bars described in the text. For
points within the two horizontal bands, the resultant cross section would agree with the measured one to within
1 . (b) The overlap region has been shown on an expanded scale. The dark square denotes the reference point in
the parameter space (point B in the text).
6. Conclusions
To summarise, we discuss the feasibility of using an electronphoton collider to
investigate minimal models wherein supersymmetry breaking is mediated to the visible
sector through the super-Weyl anomaly. A very striking feature of such models, including
the minimal model we have studied here, is that the lightest chargino 1 and the lightest
neutralino 10 are nearly degenerate and predominantly winos. This leads to a long-lived
1 which then decays into 10 + (soft) resulting in a heavily ionizing charged track and
21
a soft whose impact parameter may be measurable. While signals for such scenarios
have been studied in the context of other colliders, this very feature often restricts the
detectability of the model.
We demonstrated though, that the associated production of the lighter chargino and the
sneutrino at an e collider could provide a very clean signature for such a scenario. The
signal event consists of an energetic electron (emanating from the decay of the sneutrino)
which serves as the trigger, two macroscopic charged tracks in the vertex detector and/or
two soft charged pions and, of course, a large missing transverse momentum. The possible
SM backgrounds are calculated and shown to be negligibly small once suitable selection
criteria are employed. Consequently, even with an integrated luminosity of only 100 fb1 ,
one could see as many as 1000 (background-free) signal events over a very large region of
the allowed parameter space.
Additional advantages of the mode we advocate are encapsulated in the kinematic
distribution of the signal events. The presence of distinct energy endpoints for the electron
allows one to determine the masses of both the chargino and the sneutrino to a fair degree
of accuracy. For a very large part of the parameter space, this is true even without an energy
scan. The latter technique, if employed, can only improve the measurements. In addition,
a measurement of the cross section can be used to constrain the possible range for tan .
Once the signal has been established and all possible information gleaned, the same
collider can easily be used to obtain confirmatory checks. The simplest example is the
associated production of the left-selectron with the lightest neutralino. A more non-trivial
example (and requiring a higher energy) is the associated production of the right-selectron
with the second-lightest neutralino. Using techniques similar to those discussed here, one
could also measure the masses of these two particles. Since AMSB models have very
definite predictions for the ratios of the masses of two lightest neutralinos as well as for the
splitting of slepton masses, such tests are likely to be crucial in establishing the mechanism
of supersymmetry breaking as well as the parameters of the theory. We hope to come back
to this issue in the near future.
Acknowledgements
D.K.G. would like to thank the Department of Physics, National Taiwan University,
Taipei, where the initial part of this work was done. S.R. thanks Aseshkrishna Datta
and Biswarup Mukhopadhyaya for very helpful discussions. D.C. would like to thank
the Department of Science and Technology, India, for financial assistance under the
Swarnajayanti Fellowship grant.
References
[1]
[2]
[3]
[4]
[5]
L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79.

G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, J. High Energy Phys. 12 (1998) 027.
J.A. Bagger, T. Moroi, E. Poppitz, J. High Energy Phys. 04 (2000) 009.
A. Pomarol, R. Rattazzi, J. High Energy Phys. 05 (1999) 013.
G.D. Kribs, Phys. Rev. D 62 (2000) 015008.
22
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
E. Katz, Y. Shadmi, Y. Shirman, J. High Energy Phys. 08 (1999) 015.

I. Jack, D.R.T. Jones, Phys. Lett. B 482 (2000) 167.
M. Carena, K. Huitu, T. Kobayashi, Nucl. Phys. B 592 (2000) 164.
Z. Chacko, M.A. Luty, I. Maksymyk, E. Ponton, J. High Energy Phys. 04 (2000) 001.
B.C. Allanach, A. Dedes, J. High Energy Phys. 06 (2000) 017.
N. Arkani-Hamed, D.E. Kaplan, H. Murayama, Y. Nomura, J. High Energy Phys. 02 (2001) 041.
Z. Chacko, M.A. Luty, E. Ponton, Y. Shadmi, Y. Shirman, Phys. Rev. D 64 (2001) 055009.
I. Jack, D.R.T. Jones, Phys. Lett. B 491 (2000) 151.
D.E. Kaplan, G.D. Kribs, J. High Energy Phys. 09 (2000) 048.
A.E. Nelson, N. Weiner, Phys. Rev. Lett. 88 (2002) 231802.
F. de Campos, M.A. Daz, O.J.P. boli, M.B. Magro, P.G. Mercadante, Nucl. Phys. B 623 (2002) 47.
A. Anisimov, M. Dine, M. Graesser, S. Thomas, Phys. Rev. D 65 (2002) 105011, hep-th/0201256.
M.A. Luty, R. Sundrum, Phys. Rev. D 65 (2002) 066004, hep-th/0111231.
R. Harnik, H. Murayama, A. Pierce, hep-ph/0204122.
I. Jack, D.R.T. Jones, R. Wild, Phys. Lett. B 535 (2000) 193.
R. Rattazzi, A. Strumia, J.D. Wells, Nucl. Phys. B 576 (2000) 3.
J.L. Feng, T. Moroi, L. Randall, M. Strassler, S. Su, Phys. Rev. Lett. 83 (1999) 1731.
T. Gherghetta, G.F. Giudice, J.D. Wells, Nucl. Phys. B 595 (1999) 27.
J.L. Feng, T. Moroi, Phys. Rev. D 61 (2000) 095004.
S. Su, Nucl. Phys. B 573 (2000) 87.
F. Paige, J. Wells, hep-ph/0001249.
H. Baer, J.K. Mizukoshi, X. Tata, Phys. Lett. B 488 (2000) 367.
A. Datta, P. Konar, B. Mukhopadhyaya, Phys. Rev. Lett. 88 (2002) 181802, hep-ph/0111012.
A.J. Barr, C.G. Lester, M.A. Parker, B.C. Allanach, P. Richardson, hep-ph/0208214.
D.K. Ghosh, P. Roy, S. Roy, J. High Energy Phys. 08 (2000) 031;
A. Datta, S. Maity, Phys. Lett. B 513 (2001) 130.
D.K. Ghosh, A. Kundu, P. Roy, S. Roy, Phys. Rev. D 64 (2001) 115001.
D. Choudhury, B. Mukhopadhyaya, S. Rakshit, A. Datta, hep-ph/0205103.
C.H. Chen, M. Drees, J.F. Gunion, Phys. Rev. Lett. 76 (1996) 2002;
C.H. Chen, M. Drees, J.F. Gunion, Phys. Rev. D 55 (1997) 330;
C.H. Chen, M. Drees, J.F. Gunion, Phys. Rev. Lett. 82 (1999) 3192, Erratum;
J.F. Gunion, S. Mrenna, Phys. Rev. D 64 (2001) 075002.
R.W. Robinett, Phys. Rev. D 31 (1985) 1657.
F. Cuypers, G.J. van Oldenberg, R. Rckl, Nucl. Phys. B 383 (1992) 45.
H.A. Knig, K.A. Peterson, Phys. Lett. B 294 (1992) 110.
T. Kon, A. Goto, Phys. Lett. B 295 (1992) 324.
D. Choudhury, F. Cuypers, Nucl. Phys. B 451 (1995) 16.
K. Kiers, J.N. Ng, G.H. Wu, Phys. Lett. B 381 (1996) 177.
V. Barger, T. Han, J. Kelly, Phys. Lett. B 419 (1998) 233.
A. Ghosal, A. Kundu, B. Mukhopadhyaya, Phys. Rev. D 57 (1998) 1972.
D.K. Ghosh, S. Raychaudhuri, Phys. Lett. B 422 (1998) 187.
I.F. Ginzburg, G.L. Kotkin, V.G. Serbo, V.I. Telnov, Nucl. Instrum. Methods 205 (1983) 47.
S. Berge, M. Klasen, Y. Umeda, Phys. Rev. D 63 (2001) 035003.
U. Chattopadhyay, D.K. Ghosh, S. Roy, Phys. Rev. D 62 (2000) 115001.
S.P. Martin, M.T. Vaughn, Phys. Rev. D 50 (1994) 2292.
R. Arnowitt, P. Nath, Phys. Rev. D 46 (1992) 3981;
V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 49 (1994) 4908.
L.J. Hall, R. Rattazzi, U. Sarid, Phys. Rev. D 50 (1994) 7048;
R. Hempfling, Phys. Rev. D 49 (1994) 6168;
M. Carena, M. Olechowski, S. Pokorski, C. Wagner, Nucl. Phys. B 426 (1994) 269;
D. Pierce, J. Bagger, K. Matchev, R. Zhang, Nucl. Phys. B 491 (1997) 3.
H.C. Cheng, B.A. Dobrescu, K.T. Matchev, Nucl. Phys. B 543 (1999) 47.
ALEPH Collaboration, A. Heister, et al., hep-ex/0203020.
DELPHI Collaboration, M. Elsing, et al., http://www.cern.ch/~offline/physics_links/lepc.html.
[52] J.L. Feng, K.T. Matchev, Phys. Rev. Lett. 86 (2001) 3480;
K. Choi, K. Hwang, S.K. Kang, K.Y. Lee, W.Y. Song, Phys. Rev. D 64 (2001) 055001.
[53] U. Chattopadhyay, P. Nath, Phys. Rev. Lett. 86 (2001) 5854;
H. Baer, C. Balazs, J. Ferrandis, X. Tata, Phys. Rev. D 64 (2001) 035004;
K. Enqvist, E. Gabrielli, K. Huitu, Phys. Lett. B 512 (2001) 107.
[54] M. Knecht, A. Nyffeler, hep-ph/0111058;
M. Knecht, A. Nyffeler, M. Perrottet, E. de Rafael, Phys. Rev. Lett. 88 (2002) 071802;
M. Hayakawa, T. Kinoshita, hep-ph/0112102;
I. Blokland, A. Czarnecki, K. Melnikov, Phys. Rev. Lett. 88 (2002) 071803;
J. Bijnens, E. Pallante, J. Prades, hep-ph/0112255;
M. Ramsey-Musolfand, M.B. Wise, hep-ph/0201297.
[55] Muon g 2 Collaboration, G.W. Bennett, et al., hep-ex/0208001.
[56] A. Datta, A. Kundu, A. Samanta, Phys. Rev. D 64 (2001) 095016.
[57] E. Gabrielli, K. Huitu, S. Roy, Phys. Rev. D 65 (2002) 075005.
[58] A. Riotto, E. Roulet, Phys. Lett. B 377 (1996) 60;
A. Kusenko, P. Langacker, G. Segre, Phys. Rev. D 54 (1996) 5824.
23

Supersymmetric Higgs bosons in a 5D

orbifold model
V. Di Clemente, S.F. King, D.A.J. Rayner
Department of Physics and Astronomy, University of Southampton, Southampton, SO17 1BJ, UK
Received 21 May 2002; accepted 19 September 2002
Abstract
We analyze the phenomenology of the Higgs sector in a 5D model compactified on an S1 /Z2
orbifold where the compactification scale MC is around the TeV scale. We show that the conventional
MSSM Higgs boson mass bounds in 4D can be violated when we allow the gauge sector, Higgs
and third family multiplets to live in the fifth extra dimension. Supersymmetry is broken at
an orbifold fixed point which is spatially separated from the Yukawa brane where two chiral
families are localized. When the brane-localized supersymmetry breaking term for the stop sector
is arbitrarily large, we find that the stop KK-mode mass spectrum is completely independent of
the Higgs fields. Hence, the Higgs masses only receive radiative contributions from the top KKmodes. The 1-loop effective potential is insensitive to the cutoff scale of the theory and yields
a negative Higgs squared-mass contribution that triggers electroweak symmetry breaking in the
range 1.5 tan 20, where bottom/sbottom loop effects can be ignored. The recent LEP Higgs
bound at mh0 > 114.1 GeV, in conjunction with naturalness arguments, allows us to bracket the
compactification scale 1.5 MC 4 TeV. Within this parameter space, we find that the lightest
Higgs boson mass has an upper bound mh0 160 GeV with the magnitude of the -parameter
restricted to the range 33 || 347 GeV.
PACS: 11.10.Kk; 11.25.Mj; 12.60.-i; 12.60.Jv
1. Introduction
Extra-dimensional supersymmetric models with a TeV compactification scale [1] have
recently offered an exciting new environment for investigating electroweak symmetry
E-mail address: vicente@hep.phys.soton.ac.uk (V. Di Clemente).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 3 9 - 8
V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442
25
breaking (EWSB) [26]. The main features of these types of models are the following:
(i) they provide a new mechanism to break supersymmetry (SUSY); (ii) the contribution of
quark/squark KaluzaKlein (KK) modes to the Higgs mass term is negative, thus triggering
EWSB at the TeV scale; (iii) the 1-loop radiative correction to the effective potential is free
of ultraviolet divergences, which implies that the Higgs physics is completely independent
of the high-energy physics above some cutoff scale. Remarkably, the requirement of N = 1
SUSY in 5D (which is equivalent to N = 2 SUSY in 4D after compactification) leads to a
finite 1-loop effective potential because even though SUSY is globally softly broken, it is
still preserved locally [4,7].1
Various models have been proposed to study EWSB in extra dimensions and their
phenomenological implications [26]. However, we will concentrate only on those models
which recover the MSSM below the compactification scale and are anomaly-free after the
orbifold compactification [9]. The authors of Ref. [4] have made a qualitative study of
the Higgs physics in their model, but there is little discussion of the phenomenological
implications. In contrast, the authors of Ref. [5] have made an extensive study of Higgs
phenomenology in their model and consider how radiative corrections modify the quadratic
and quartic couplings of the tree-level potential. However, their approach neglects the nonrenormalizable operators that arise at higher-order in the expansion of the full effective
potential at 1-loop and, as discussed in Appendices A and B, this brings into question the
reliability of their calculation.
In this paper we show that the conventional MSSM lightest Higgs boson mass bound
mh0 135 GeV [10] can be violated by allowing some of the fields to live in a fifth
extra dimension.2 Motivated by fine-tuning arguments, we are led to an upper bound on
the compactification scale MC 4 TeV. At this compactification scale, the lightest Higgs
boson mass can be pushed as high as mh0 160 GeV. We find that in order to have EWSB
through radiative corrections, tan can have a wide range 1.5 tan 20 rather than
the small range allowed by Ref. [5] 35 tan 40. We only limit tan 20 since we
neglect bottomsbottom loop effects. Note that, unlike the MSSM, tan 1.5 is not ruled
out by experiment in our extra-dimensional model.
The layout of the remainder of the paper is as follows. In Section 2 we review our 5D
orbifold model, and in Section 2.1 we discuss the top/stop sector KK mass spectra. In
Section 2.2 we discuss the Higgs SUSY breaking parameters. In Section 2.3 we discuss
how perturbativity and naturalness provide a constraint on the theory cutoff M . Then
in Section 3 we minimize the 1-loop effective potential and calculate the Higgs mass
eigenvalues. We find that experimental data and fine-tuning arguments allow us to constrain
the physical parameter space. Section 4 concludes the paper.
1 Recently, the finiteness of this kind of theory has been showed explicitly at 2-loops [8].
2 Higher upper bounds on the lightest boson mass have been recently calculated in the context of SUSY in
warped extra dimensions [11] and a (de)constructed model [12].
26
2. Our model
In this section we review some basic features of our model [6] in which the extra
dimension of a 5D theory is compactified on an S 1 /Z2 orbifold.
The orbifold compactification leads to a description of 5D bulk fields as infinite towers
of 4D KK-modes, and a classification of bulk fields into odd and even Z2 parity.3 The
kth KK-mode has a mass mk O(kMC ), where MC = 1/R is the compactification scale,
and the lowest KK resonance (k = 0) can be identified with the usual MSSM fields. It
is conventional in five-dimensional orbifold models to exploit the equivalence of N = 1
SUSY in 5D and N = 2 SUSY in 4D [13]. Therefore, an N = 1 multiplet in 5D can
be decomposed into a 4D N = 1 chiral multiplet and its CP-conjugate mirror which
together form a complete N = 2 hypermultiplet in 4D. Similarly, an N = 1 vector
multiplet in 5D can be decomposed into a vector and chiral multiplet in 4D.4 The setup
is given in Fig. 1 and shows the location of the superfields within the extra-dimensional
space.
We localized 4D 3-branes at the two fixed points y = 0, R arising from the orbifold
compactification. SUSY is broken on the source brane at y = R when a localized gauge
field singlet (
S) acquires a non-zero F -term vacuum expectation value (VEV) FS . The first
1,2 ) families are confined to the Yukawa brane5 at y = 0, while the third
two MSSM (F
Fig. 1. Our model showing the parallel 3-branes spatially separated along the extra dimension y. This extra
dimension is compactified on the orbifold S 1 /Z2 that leads to two fixed points at y = 0, R, where the two
1,2 ) live on the Yukawa brane at y = 0, while SUSY is
3-branes are located. The first two chiral families (F
3 ), gauge
broken by the F -term of a gauge singlet field (
S) on the source brane at y = R. The third family (F
MSSM ) and Higgs superfields (H
u,d ) live in the extra-dimensional bulk along with their N = 2 SUSY
sector (V
partners which are required for consistency.
3 Note that only even parity bulk fields have k = 0 KK-modes and have a non-vanishing wavefunction profile
at the fixed points.
4 The MSSM vector multiplets V
MSSM each contain a 4D gauge field (A ), two symplectic Majorana
gauginos (1,2 ) and a real scalar ( ) [13].
5 This is known as the Yukawa brane since the superpotential cannot be defined in the N = 2 supersymmetric
bulk, and we are forced to localize the N = 1 supersymmetric Yukawa couplings on the 4D brane at y = 0.
27
3 ), MSSM gauge sector (

u,d ) live in the extrafamily (F
VMSSM ) and Higgs superfields (H
6
dimensional bulk. These bulk fields acquire tree-level soft parameters due to their direct
coupling with the SUSY breaking brane through non-renormalizable operators defined at
y = R:
LR = (y R)

1
4
u H
u + cHd H
H

cF F F3 + cHu H
d
d d
M3 3 3

c
u H
d + h.c.
+ cB H
S
+
h.c.
,
S+
H
H
S
u
d
M2
(1)
where M is the cutoff of the theory, and is often identified with the string scale. The
coefficients cF3 (cHu,d , c , cB ) are the couplings of the third family (Higgs) superfields to
the SUSY breaking sector. There are also higher-dimensional Yukawa couplings localized
on the Yukawa brane at y = 0, but we will only consider the dominant top/stop sector
couplings:
LYuk
0 = (y)
ft
3/2

3L H
u U
c + h.c. ,
d 2 Q
3R
(2)
where ft = (RM )3/2 yt and ft (yt ) is the 5D (4D) Yukawa coupling. The zero modes
of the neutral components of the two complex (scalar) Higgs doublets can be expanded in
terms of real and imaginary parts
1
Hu0 = (hu + iu ),
2
1
Hd0 = (hd + id )
2
(3)
and electroweak symmetry is spontaneously broken when the real parts acquire non-zero
VEVs
hu ,
hd = 0 and
u =
d = 0.
This setup is like the gaugino mediated SUSY breaking (gMSB)
scenario [14], but

with the third family superfields moved from the Yukawa brane and placed in the extradimensional bulk. The spatial separation of MSSM fields on the Yukawa brane away
from the SUSY breaking sector alleviates the supersymmetric flavour-changing neutralcurrent (FCNC) problem since squark masses, arising from direct couplings between the
two sectors, are exponentially suppressed by the separation between branes. Instead, first
and second family squark masses are generated via flavour-blind loop-effects.7 Unlike
previous models, our setup is motivated by a type I string-inspired model [16] in which
the separation of the third family is generic, and often leads to realistic Yukawa textures.
6 The presence of the top/stop fields in the bulk is phenomenologically important for their dominant 1-loop
contribution to the Higgs potential.
7 Notice that the third family squarks acquire unsuppressed masses due to their direct coupling to the SUSY
breaking. However, this does not undermine the solution to the FCNC problem since there are much weaker
experimental constraints on third family contributions [15].
28
2.1. Top/stop KK mass spectra

We will summarize the KK mass spectra found in our earlier paper [6]. The nonrenormalizable operators in Eq. (1) contain a delta-function that induces mixing between
different KK-modes, where the mixing strengths (t, H ) are proportional to the SUSY
breaking VEVs:
F 2
2

FS,H
M
M
S,t
t = ct
(4)
,
H = cH
,
M4
MC
M4
MC
S field that couples to the
where FS,t(FS,H ) is the F -term VEV associated with the singlet
stop (Higgs) fields.8 We have made the simplifying assumption that cHu = cHd = cH . The
non-trivial mixing between different KK-modes requires that we diagonalize an infinite
mass matrix to obtain the KK mass eigenvalues.9 However, if the mixing is small, the
mass matrix is dominated by the diagonal components. In order to respect N = 1 SUSY
transformations on the brane, we use an off-shell formulation of N = 2 SUSY in the 5D
bulk [13] that mixes fields of different Z2 -parity and so the KK-summation runs over the
full tower ( < k < ).
The field-dependent top KK-mode mass eigenvalues are given by:10

yt hu
MC

(k = , . . . , ),
arctan
mt,k [hu ] = kMC +
(5)
2MC
where the eigenvalues only depend on the real part of Hu0 . The observable top mass is
identified with the k = 0 KK-mode,

yt hu
MC
mt [hu ] =
(6)
arctan
.
2MC
Notice that we can recover the usual MSSM relation mt [hu ] = yt hu / 2 in the limit that
MC . Eq. (6) is different due to the non-trivial mixing between KK-modes on the
Yukawa brane.
The field-dependent stop KK mass eigenvalues mt,k [hu ] are solutions of the following
transcendental equation

mt,k [hu ]
mt,k [hu ]
mt,k [hu ]
yt hu 2
cot
tan

MC
MC
MC
2 MC

2
yt hu
= t 1 +
(7)
2 MC
which can be solved by considering different limits of mixing t .
8 The extra index on the SUSY breaking VEVs allows for non-universal couplings in the hidden sector
(FS,H = FS,t ).
9 Details of the diagonalization procedure are given in our earlier paper [6].
10 Notice that the argument of the arctan function differs by a factor of 2 2 compared to the previously
published result [6].
29
Minimal mixing (t 1) is equivalent to a very small extra dimension (MC

1016 GeV) where we expect to recover the MSSM results. The stop KK mass eigenvalues
are:11

mt [hu ] 2 t
m2t,k [hu ] = k +
+ 2 + Zk2 [hu , t] MC2 1 + O t2
MC
(k = , . . . , ),
(8)
where we have used the field-dependent top mass mt [hu ] from Eq. (6), and the function
Zk2 [hu , t] is given by:

2
Zk2=0 [hu , t] = t/ , if
hu = 0,
0,
otherwise,
2
Zk=0
[hu , t] = 0.
(9)
In the limit that electroweak symmetry remains unbroken (i.e.,

hu = 0), we find that
only the even-parity bulk stop KK-modes acquire soft masses from the non-renormalizable
operators of Eq. (1), while the odd parity stops have the usual KK masses (modd
= kMC ).
t,k
Therefore, the KK-summation only runs over the positive tower (0 < k < ). Following
EWSB (
hu = 0), the Yukawa interaction of Eq. (2) induces mixing between different
parity stop KK-modes, and the KK-summation recovers the full infinite tower. Comparing
Eqs. (5) and (8), we can see that the form of the k = 0 mass eigenvalues resemble the
MSSM as expected12 where the stop mass squared is given by the top mass and SUSY
breaking mass added in quadrature. However, in this limit the compactification scale MC
can be at very high energies, and we recover the MSSM phenomenology at low energy.
Therefore, we consider maximal mixing in the stop sector (t 1) which is equivalent
to a large extra dimension. In this case, the solution of Eq. (7) is given by

1 2 2
(k = , . . . , ).
MC 1 + O t1
m2t,k = k +
(10)
2
Note that to leading order the stop mass eigenvalues are independent of the Higgs
background field (hu ) even when the electroweak symmetry is broken. This can be
attributed to the arbitrarily large mixing term on the source brane that makes the Yukawa
brane become transparent which washes out any field dependence [6]. The Yukawa
interaction induces mixing between different parity stop fields on the Yukawa brane such
that both odd- and even-parity stop KK-modes acquire the same mass from Eq. (10). We
find that the compactification scale should be MC O (TeV) to provide soft terms at the
electroweak scale.13
11 This case is particularly interesting for grand unified theories (GUT) in extra dimensions [17] since
2 TeV2 ) even when the
it is possible to generate soft parameters around the electroweak scale (t MC
16
compactification scale is as high as the GUT scale (MC 10 GeV).
12 For simplicity, we have neglected the trilinear mixing term A between the stop left and right fields.
t
13 This maximal mixing eigenvalue has a very weak dependence on the precise magnitude of the -mixing
t
and so the 1
corrections in Eq. (10) can be neglected.
30
We use dimensional regularization and zeta-function regularization techniques [18] to

evaluate the top/stop contributions to the 1-loop effective potential:
V1-loop = Vtop [hu ] + Vstop
p 2 + m2

1
d 4p
t,k
= Tr
ln
2
(2)4
p2 + m2t,k [hu ]
k=
=
9MC4 cos[2nmt [hu ]/MC ] 135 [5] 4

+
M ,
16 6
n5
256 6 C
(11)
n=1
where we have used Eqs. (5), (6) and (10), and the trace is over all degrees of freedom.
The top and stop contributions are found to be separately finite (and free of ultraviolet
divergences). However, we find the same result by using other regularization techniques
(i.e., KK-regularization [25]), where the UV divergences vanish due to supersymmetric
cancellations between the top and stop contributions. Note that only the top contribution
is field dependent, so the constant stop contribution can be absorbed into the cosmological
constant and dropped from the subsequent analysis.14
2.2. SUSY breaking Higgs parameters
We have seen in Section 2.1 that we are led to maximal mixing (t 1) in the
stop sector. We also know that for EWSB via top/stop radiative corrections, we require
a negative mass-squared to trigger spontaneous symmetry breaking. This is harder to
achieve when t H since this leads to tree-level and (negative) 1-loop contributions
of comparable magnitude. Therefore, we conclude that the KK-modes in the Higgs sector
must be minimally mixed (i.e., H 1).
Using Eq. (4), we have two options, either (i) the couplings in the higher-dimensional
operators are hierarchical (ct cH ), or else (ii) there is a non-universality in the hidden
sector where different SUSY breaking fields couple to the stop and Higgs sectors (FS,H =
FS,t). In this paper, we will assume that all hidden sector couplings are c O(1), and
instead have non-universal F -terms (FS,H FS,t).
For minimal mixing in the Higgs sector, the KK mass matrix is dominated by the
diagonal terms and we can decouple the non-zero KK excitations. We will impose the
EWSB conditions on the lightest (k = 0) KK-modes where the soft masses are taken
directly from Eq. (1):
m2Hu = m2Hd = m2soft =
=
2
cH FS,H
M3
MC ,
B =
2
cB FS,H
M3
MC ,
c FS,H
MC
M2
(12)
and we have assumed that cHu = cHd = cH for universal soft Higgs masses.
14 We show that the field-dependence in the stop sector vanishes using a diagrammatic approach in Appendix A.
31
2.3. Reliability and perturbativity

We have not yet imposed any constraint on the relationship between the compactification scale MC and the cutoff scale M . The requirement of perturbativity (where our
perturbative analysis is valid) allows us to find an upper bound on the ratio (M /MC ). It is
well known that in 5D (extra-dimensional) theories that the gauge and Yukawa couplings
exhibit power law running behaviour [19]. The beta functions of these couplings depend
on powers of the renormalization scale due to the inclusion of the KK-modes that makes
the physics highly sensitive to the renormalization scale. This implies that gauge coupling
unification and the emergence of the Landau pole in the Yukawa couplings are accelerated
with respect to the (logarithmically-running) 4D theory. The top Yukawa coupling is found
to become singular at energies close to the compactification scale MC . In our model, the
presence of the third family in the 5D bulk makes the Yukawa coupling beta function (yt )
depend quadratically on the ratio between the renormalization scale and the compactification scale MC

2
3
y t y t
+ .
(13)
MC
Note that the dependence on (/MC ) is stronger than the corresponding beta function for
the gauge coupling (yt ) which is only linearly-dependent

+ .
g g 3
(14)
MC
Suppose NP is the scale where the top Yukawa coupling becomes non-perturbative,
which we numerically found to be NP 5MC [5]. We can maintain a (reliable)
perturbative regime by imposing the following constraint on the cutoff scale of our
theory M
M NP
M 5MC .
(15)
Similarly it appears unnatural to have the cutoff of our theory M below the
compactification scale MC . Hence, we find that perturbativity and naturalness severely
constrain M to the range:
MC M 5MC .
(16)
We need to check that these constraints are consistent with maximal mixing in the stop
sector (t 1). However, we find that there is no inconsistency since our earlier work [6]
showed numerically that maximal mixing only requires t 12 since solutions of Eq. (7)
tend towards an asymptotic value with the stop mass eigenvalues given by Eq. (10). This
is achieved when FS,t (0.9/ct)M2 M2 for ct 1.
3. Higgs mass spectrum
In this section we calculate the mass eigenvalues in the Higgs sector, where the light and
heavy CP-even Higgs mass eigenstates (h0 , H 0 ) are linear combinations of the real fields
32
hu and hd . We can use the standard MSSM relations to find the masses of the charged (H )
and CP-odd (A0 ) Higgs fields. The tree-level potential in terms of the neutral components
of the Higgs doublets (Hu0 , Hd0 ) is:
2
2

m 2 0 2 2 2
Vtree = m21 Hd0 + m22 Hu0 B Hu0 Hd0 + h.c. + Z2 Hu0 Hd0 , (17)
2v
where we are free to define B as real and positive by absorbing any phase into Hu0 and
Hd0 ; and we have traded the U (1)Y and SU(2)L gauge couplings (g , g) for the physical Z 0
mass and the VEV v = 246 GeV. The soft parameters m21 = ||2 + m2Hd , m22 = ||2 + m2Hu
and B are given in Eq. (12).
Combining Eqs. (3), (11), (17), we find an expression for the total 1-loop effective
potential in terms of the real Higgs fields hu , hd . Notice that we have dropped the 1-loop
stop contribution since it is independent of the Higgs fields and can be absorbed into the
cosmological constant
m2Z 0 2

2
m2soft 2
2
hu + hd Bhu hd + 2 hu h2d
V=
2
8v

1 yt h
u
9MC4 cos 2n tan
MC 2
+
16 6
n5
(18)
n=1
and we have assumed that the Higgs doublets acquire universal soft masses m21 = m22 =
m2soft which we regard as input parameters along with B. Applying the EWSB conditions
at the usual minimum
hu = v sin ,
hd = v cos (v = 246 GeV)

V
V
=
=0
(19)
hu
hu ,
hd hd
hu ,
hd
allows us to eliminate B and m2Z 0 in terms of the other parameters. By imposing the
correct observable Z 0 mass (mZ 0 = 91.2 GeV), we can also eliminate msoft such that the
compactification scale MC and tan (or equivalently mA0 ) can be regarded as the two free
parameters. In Fig. 2, we plot msoft as a function of the compactification scale MC for two
different values of tan .
We construct the CP-even mass matrix from the second derivatives of the effective
potential at the minimum, which can be diagonalized to find the mass eigenvalues
(mh0 , mH 0 ):
2V
2V
M2even =
h2u
2V
hd hu
hu hd
2V
h2d
(20)
Notice that these CP-even eigenvalues include the 1-loop effects, but only m22 is 1-loop
improved since we are neglecting bottomsbottom loops for tan 20:

1 2 Vtop
2
2
m2,imp = m2 +
(21)
.
2 h2u
hu ,
hd
33
Fig. 2. The universal Higgs soft mass m2soft = ||2 + m2H against the compactification scale MC for tan = 1.5
u,d
and 20.
Fig. 3. The fine-tuning parameter as a function of tan and the compactification scale MC in GeV. is shown
to have a very weak dependence on tan in the range 2.5 tan 20 for which we can neglect bottomsbottom
effects. However, the fine-tuning becomes singular as tan 1.
However, B (and m21 ) are not 1-loop improved, so we can use the standard 4D tree-level
MSSM expressions to find the CP-odd (A0 ) and charged Higgs (H ) masses.
m2A0 =
2B
,
sin 2
m2H = m2A0 + m2W .
(22)
(23)
We can solve for m2Z 0 as a function of the free parameters (MC , tan ), and use the standard
definition of fine-tuning [20,21] (but neglecting the variation of tan with respect to
34
Fig. 4. The Higgs masses (mh0 , mH , mH 0 ) against mA0 for tan = 1.5 (upper panel) and tan = 20 (lower
panel). We have run 1 MC 4 TeV parametrically along each curve. The LEP candidate at 115 GeV (bold
dotted-line) [23] is shown for comparison with mh0 . For tan = 1.5 (upper panel), we also compare the MSSM
results taken from [22] (dotted-lines) against our model (bold-lines).
changes in MC ) to find an upper limit on the compactification scale MC

MC m2Z 0
.

= 2
m 0 MC
(24)
Motivated by the fine-tuning as shown in Fig. 3, we will investigate the parameter space
MC 4 TeV and 1.5 tan 20 where the fine-tuning O(102 ).
In Fig. 4, we plot the eigenvalues (mh0 , mH 0 ) of the CP-even mass matrix and the
charged Higgs mass (mH ) against the CP-odd mass (mA0 ) for two fixed values of
tan = 1.5 and 20. We run the value of the compactification scale parametrically along
each curve, 1 MC 4 TeV. We also include the MSSM predictions for tan = 1.5 taken
from [22] for comparison. Notice that unlike the MSSM predictions from [22], our model
is not excluded by the LEP signal for tan = 1.5. There are additional experimental lower
limits for the other Higgs masses
mA0 92 GeV
and mH 69 GeV
but these provide a much weaker constraint on our model.
(25)
35
Fig. 5. Mass contour plot for the lightest Higgs mass mh0 as a function of the compactification scale MC (GeV)
and tan in the absence of bottom sector effects (i.e., tan 20). The LEP candidate at 115 GeV [23] is easily
accommodated over the range of tan with a compactification scale MC 1.51.7 TeV. At large tan 20 and
MC 4 TeV, the lightest Higgs mass can be as large as mh0 164 GeV.
In Fig. 5, we plot the lightest Higgs mass (mh0 ) as a function of the compactification
scale MC and tan . The LEP data excludes the parameter space below the first contour at
mh0 = 115 GeV [23] which corresponds to a compactification scale MC 1.51.7 TeV
over the whole range of tan . Combining Figs. 3 and 5 we find an allowed window for the
compactification scale
1.5 TeV MC 4 TeV.
(26)
Notice that our model can easily accommodate the conventional 4D MSSM upper
bound on the lightest Higgs boson mass mh0 130 GeV, and can be pushed as high
as mh0 160 GeV with MC 4 TeV and 5 tan 20. Remember that including
additional matter (e.g., gauge singlets in the NMSSM) can also raise the upper bound on
the lightest Higgs mass, but our minimal extension of the MSSM achieves higher mass
bounds without adding extra matter content.
We can use Fig. 2 to find limits on the -parameter from the universal soft Higgs
mass msoft which is constrained when we impose EWSB at the electroweak minimum.
Using Eq. (12), we find the ratio between m2Hu = m2Hd = m2H and ||2 in terms of the
compactification and cutoff scales:
m2H
M
,
2
||
MC
where we have assumed that cH , c 1.
(27)
36
Table 1
Upper and lower limits on the size of the -parameter for two different values of tan = 1.5 and 20, and
compactification scales MC = 1.5 and 4 TeV
tan = 1.5
tan = 20
MC = 1.5 TeV
MC = 4 TeV
114 || 57
66 || 33
347 || 173
215 || 107
Recall from Section 2.2 that we assume non-universality in the SUSY breaking sector
(FS,H FS,t) to obtain maximal (minimal) mixing in the stop (Higgs) sectors for viable
radiative EWSB without hierarchical couplings ct cH . We can use the perturbativity and
naturalness constraints of Eq. (16) to find limits on the ratio between the soft Higgs mass
and the -parameter from Eq. (27)
m2H
5
||2
and therefore limits on msoft in terms of :
m2soft = m2H + ||2
( + 1)||2 m2soft (5 + 1)||2.
(28)
(29)
The constraints on the -parameter for different compactification scales and values of tan
are shown in Table 1. Therefore, the magnitude of the -parameter is constrained to the
range 33 || 347 GeV.
4. Conclusions and discussion

In conclusion we have considered the Higgs sector of an N = 1 supersymmetric
5D theory compactified on an orbifold S 1 /Z2 where the compactification scale MC
O (TeV). Orbifolding leads to fixed points at either end of the extra dimension (y) where
4D branes can be localized. Supersymmetry is broken by the F -term VEV of a gauge
singlet on the brane at y = R that is spatially separated from another Yukawa brane at
y = 0 where the first two MSSM families and Yukawa couplings are localized. Direct
coupling between the two sectors (and therefore soft squark masses) are suppressed by
the separation between the branes which alleviates the flavour-changing neutral-current
problem since the first and second family squark masses are only generated through
flavour-blind loops. The third family, gauge sector and Higgs fields live in the extradimensional bulk with their N = 2 supersymmetric partners (which are required for
consistency) and therefore receive unsuppressed soft masses due to their direct coupling to
the SUSY breaking sector.
We assume a non-universality in the SUSY breaking sector, where different gauge singlets couple separately to the stop (Higgs) fields and the associated F -term VEVs are hierarchical (FS,H FS,t ) which induces maximal (minimal) mixing between different KKmodes. The maximal mixing between stop modes requires that we use a matrix method
to diagonalize the infinite mass matrix. In contrast the Higgs KK-modes are minimally
mixed so that the mass matrix is dominated by the diagonal components and we can decouple the non-zero KK-modes from the analysis. We find that the soft Higgs parameters
37
are generated by non-renormalizable operators. The presence of the third family in the
bulk is particularly important for their dominant 1-loop contribution to the Higgs effective
potential. The full tower of top and stop KaluzaKlein modes contribute to the potential
and trigger radiative electroweak symmetry breaking. Following-dimensional regularization and zeta-function regularization techniques, we see that the 1-loop contributions to the
effective potential are separately finite and therefore insensitive to the high-energy cutoff
M . However, in the maximal mixing limit we find that the top contribution has a nontrivial dependence on the Higgs background fields, and the stop only contributes to the
cosmological constant.
We minimize the 1-loop effective potential and impose the conditions for electroweak
symmetry breaking to find the physical Higgs mass eigenvalues. Requiring the correct
physical Z 0 -mass allows us to eliminate parameters in terms of the compactification
scale MC and tan (or equivalently mA0 ). We use fine-tuning arguments to constrain our
parameter space MC 4 TeV, and we choose to study the region 1.5 tan 20 where
bottom sector effects can be neglected. We obtain physical Higgs mass eigenvalues for
different values of tan and find that the LEP signal [23] imposes a lower limit on the
compactification scale MC 1.5 TeV. We also find that, unlike the MSSM, tan = 1.5 is
not excluded by experiment and our model can accommodate the LEP signal over the full
parameter space. In fact the usual MSSM upper bound (mh0 130 GeV) and the extended
NMSSM bound (mh0 150 GeV) can be trivially exceeded and raised to mh0 164 GeV
for a compactification scale MC 4 TeV and 5 tan 20.
Note that radiative electroweak symmetry breaking is viable over a large range 1.5
tan 20 in comparison to an alternative model [5] that is severely constrained to the
smaller range 35 tan 40. In fact, we show in Appendix B that their diagrammatic
analysis is incomplete because it neglects higher-order non-renormalizable operators that
inevitably appear in the expansion of the 1-loop effective potential. The requirement
of perturbativity and naturalness in our model imposes a constraint on the relationship
between the compactification scale MC and the cutoff M scale (MC M 5MC ). We
use this constraint in combination with the universal soft Higgs mass to deduce limits on the
-parameter, and we find that the magnitude of is inside the range 33 || 347 GeV.
ScherkSchwarz (SS) boundary conditions have been used extensively in the literature
to break SUSY in extra-dimensional models [25,7]. Recently, the authors of Ref. [24]
demonstrated that the SS effects are always present, and a vanishing SS parameter at treelevel will be generated at 1-loop by bulk supergravity fields.15 In the case that SUSY
breaking is localized on a hidden sector brane, the SS breaking parameter SS is no longer
discrete, but may take a range of values that depend upon the relative strength of the SUSY
breaking and the matter content of fields living in the extra dimension.16 In our model, the
third family and Higgs hypermultiplets (NH = 19) live in the bulk with the MSSM vector
multiplets (NV = 12), and the SS breaking parameter tends towards SS = 0 (SS = 1/2)
15 We thank Antonio Riotto for clarifying this point.
16 Ref. [24] investigate two scenarios with SUSY breaking localized on hidden sector brane where the value of
SS depends on the relationship between NH and NV , where NH (NV ) are the number of hypermultiplets (vector
multiplets) in the bulk.
38
for weak (strong) SUSY breaking on the brane which is equivalent to minimal (maximal)
mixing in the stop sector.
We anticipate that non-trivial SS contributions will modify our expressions for the
stop KK-mode mass eigenvalues17 given in Eqs. (8), (10). However, we can see that the
additional SS contributions will not alter our analysis of the Higgs sector. For minimal
mixing the stop mass eigenvalues are given by Eq. (8), but we know that in this limit
the SS parameter vanishes. Also, for maximal mixing SS 1/2 and Eq. (10) becomes
m2t,k = (k + 1/2 + SS )2 MC2 + . However, we have already seen that the stop mass
eigenvalues are independent of the background Higgs field in this maximal mixing limit
and the stop sector contribution to the effective potential is constant. Therefore, we find that
ScherkSchwarz effects play no rle in our calculations and we are justified in ignoring
them.
Acknowledgements
S.K., V.D.C. and D.R. would like to thank PPARC for a Senior Fellowship, Research
Associateship and a Studentship. We would like to thank Antonio Riotto for useful
discussions and bringing Ref. [24] to our attention.
Appendix A. Finite Higgs mass from a diagrammatic approach

In Section 2.1 we have shown that in the limit the brane-localized SUSY breaking term
associated with the stop sector is arbitrarily large (maximal mixing), the 1-loop effective
potential only receives field-dependence from the top sector KK masses while the stop
sector contribution can be absorbed into the cosmological constant. In this appendix we
will show that the finite part of the 1-loop Higgs scalar two-point function at zero external
momenta, as shown in Fig. 6, only arises from the top KK-mode contributions. We illustrate
this with a toy model that we have already discussed in earlier work [6].
In Ref. [6] we made the simplifying assumption that the tree level potential only receives
D-term contributions by ignoring the Higgs soft parameters that are generated by non-
Fig. 6. One-loop diagrams contributing to m2h in Eq. (A.3). Notice that the second and third diagrams (b), (c)
appear in the MSSM while the requirement for N = 2 SUSY in the extra-dimensional bulk leads to an additional
diagram (a) involving CP-mirror partners.
17 Remember that the top spectrum remains unaffected by ScherkSchwarz boundary conditions.
39
renormalizable operators in Eq. (1). We made a further simplification by taking the limit
tan where the Hd doublet can be decoupled. Therefore, the tree-level potential is
only a function of the neutral component of the up-type Higgs Hu0 which we take to be real
hu from Eq. (3)
Vtree [hu ] =
g2 + g 2 4
hu ,
32
(A.1)
where g and g are the gauge couplings of SU(2)L and U (1)Y , respectively. We can see
that at tree-level the potential has a minimum at hu = 0. However, the 1-loop contributions
can trigger electroweak symmetry to be spontaneously broken. The 1-loop contribution is
given in Eq. (11). The full effective potential is then
V [hu ] = Vtree [hu ] + V1-loop[hu ].
(A.2)
We find that a negative quadratic mass term is given by:

d 2 V
9 (3) 2 2
2
=
yt MC 1.39 102 yt2 MC2 ,
mh =

2
4
dhu
hu =0
8
(A.3)
where (3) 1.202. Notice that Eq. (A.3) is identical to the result reported in Ref. [4],
where m2h was calculated using the diagrams shown in Fig. 6. The mass eigenvalues of
the even-parity stop KK-modes (tR,l and tL,l ) are given in Eq. (10); and the masses for the
mc and tmc ) and the tops (t
odd-parity (mirror) stops (tR,k
L,k , tR,k ) are kMC .
L,k
It is possible to see that the finite part of m2h effectively arises from diagram (b) in
Fig. 6 where only the top KK-modes propagate in the loop. Evaluating diagrams (a) + (c)
gives:

im2h (a) + (c)
= 3yt2

(k )2
k=0 l=0
+ 3yt2

k=0 l=0
m2tmc
d 4p
L,k
+ (L R)
(2)4 (p2 m2mc )(p2 m2 )
t
t
(k )2
R,l
L,k
d 4p
(2)4
1
(p2
m2t )
+ (L R),
(A.4)
R,l
where the factor of 3 is the colour multiplicity and k = (1/ 2)k0 . Numerous papers in
the literature [18,25] have demonstrated the validity of exchanging the sum and integral
in Eq. (A.4) despite the controversy surrounding this so-called KK-regularization
technique [26]. Performing a Wick rotation to Euclidean momentum space pE , and a
change of variables x = pE /MC gives
(k )2 x 2
d 4x
(2)4
(x 2 + (k + 1/2)2)(x 2 + l 2 )
k,l=0
2 d 4x
3iyt2 4
2 2
= i3yt MC
=
,
4
128 MC2
(2)4

im2h (a) + (c) = i3yt2MC2
(A.5)
40
where is the UV cutoff. The quartically-divergent contribution in Eq. (A.5) exactly

cancels with the infinite part of diagram (b). Therefore, m2h in Eq. (A.3) only arises from
diagram (b).18
We will conclude this appendix by calculating the compactification scale and Higgs
mass in this toy model. We use the MS running top mass mt,0 = 166 GeV to find the
compactification scale by imposing the following minimization conditions around the
minimum at v = 246 GeV:

d 2 V
dV
= 0,
= m2h .
(A.6)
dh
dh2
u
hu =v
u
hu =v
Numerically we find:
MC 830 GeV
(A.7)
which is approximately 2.5 times larger than the compactification scale calculated in
Ref. [3]. The second-derivative of the effective potential at the minimum yields a lightest
Higgs scalar mass19
mh 120 GeV.
(A.8)
This prediction for the lightest Higgs boson mass is possible because the effective
potential just depends on the compactification scale MC which has been fixed at a specific
value MC 830 GeV to obtain the correct minimum of the effective potential. However,
the more general two-Higgs doublet analysis in Section 3 depends on other soft parameters
including tan , the universal soft Higgs mass m2soft , B and the supersymmetric mass .
Appendix B. Truncation of the 1-loop effective potential
In this appendix we discuss the problems that arise when truncating the expression of
the full 1-loop effective potential in the context of extra dimensions. We will show that this
truncation leads to a significant discrepancy compared to using the full 1-loop expression.
This problem arises because all dimensionful parameters defined in the theory are of the
same order MC . In the effective field theory, it is impossible to distinguish between high
and low energy scales, and all non-renormalizable operators are found to be as important
as the renormalizable operators.
Expanding the 1-loop effective potential (11) around the origin hu = 0 we obtain:
1-loop[hu ] = 9 (5) MC4 + 1 m2h h2u
V
2
16 6

6
4
25 1
hu
2 yt hu
3yt 4
log
+O
,
+
h (3) +
16 2 u
24 2
MC
MC2
(B.1)
18 Notice that this is only true with maximal mixing in the stop sector. For example, when SUSY remains
unbroken (t = 0), we obtain a finite contribution from diagrams (a) + (c) which cancels exactly with the
contribution from diagram (b) such that m2h = 0. Also note that when SUSY is (only) broken by ScherkSchwarz
boundary conditions, the finite contribution to m2h arises from all three
diagrams in Fig. 6 [2,5].
19 Note that the published result in Ref. [6] differs by a factor of 2.
41
where we have used the expression for m2h from Eq. (A.3) and tildes denote truncated
results. Notice that all non-renormalizable operators O(h6u /MC2 ) vanish in the limit
MC .
However, for MC O (TeV), these higher-order operators will give an important
contribution to the effective potential. Therefore, if the expansion is truncated at O(h4u ) we
can no longer regard the results as reliable. For example, if we impose the minimization
conditions of Eq. (A.6) on the truncated potential from Eq. (B.1) we find:
C 1 TeV,
M
m
h 170 GeV.
(B.2)
Comparing with the exact results from Eqs. (A.7), (A.8), we see that the truncated
approximation from Eq. (B.1) leads to an error of (2040)%.
We can easily understand the origin of this discrepancy by noting that the only
dimensionful parameter that appears in the effective potential is the compactification scale
MC . Therefore, each operator in the expansion of Eq. (B.1) gives comparable contributions.
Notice that the truncation of the series is equivalent to integrating out the top KK-tower.
From the effective field theory perspective, this truncation must be controlled by an
expansion in = (m2 /M 2 ), where m is the low energy scale (for example, the electroweak
scale MW ) and M is the high energy scale associated with the masses of the heavy particles
MC .
In the case that 1, the truncation of the series is reliable. However, in this model m
is generated at 1-loop and is proportional to M = MC , where m2 = m2h from Eq. (A.3) but
without the loop factor.20 Therefore m2 M 2 1, and we conclude that truncation
of the effective potential expansion at finite order cannot be reliable. The same conclusion
was observed in Ref. [3].
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[2] A. Pomarol, M. Quiros, Phys. Lett. B 438 (1998) 255, hep-ph/9806263;
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410;
A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
[3] R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
[4] N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81, hepph/0102090.
[5] A. Delgado, M. Quiros, Nucl. Phys. B 607 (2001) 99, hep-ph/0103058.
[6] V. Di Clemente, S.F. King, D.A.J. Rayner, Nucl. Phys. B 617 (2001) 71, hep-ph/0107290.
[7] A. Masiero, C.A. Scrucca, M. Serone, L. Silvestrini, Phys. Rev. Lett. 87 (2001) 251601, hep-ph/0107201.
[8] A. Delgado, G.V. Gersdorff, M. Quiros, Nucl. Phys. B 613 (2001) 49, hep-ph/0107233.
[9] D.M. Ghilencea, S. Groot Nibbelink, H.P. Nilles, Nucl. Phys. B 619 (2001) 385, hep-th/0108184;
C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169, hep-th/0110073;
L. Pilo, A. Riotto, hep-th/0202144.
[10] For a recent calculation of the 2-loop correction to the CP-even Higgs mass in the MSSM, see, for example,
J.R. Espinosa, I. Navarro, Nucl. Phys. B 615 (2001) 82, hep-ph/0104047;
G. Degrassi, P. Slavich, F. Zwirner, Nucl. Phys. B 611 (2001) 403, hep-ph/0105096;
20 Note that all of the loop factors in the expansion from Eq. (B.1) factorize out.
42
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
A. Brignole, G. Degrassi, P. Slavich, F. Zwirner, hep-ph/0112177, and references therein.

J.A. Casas, J.R. Espinosa, I. Navarro, Nucl. Phys. B 620 (2002) 195, hep-ph/0109127.
A. Falkowski, C. Grojean, S. Pokorski, hep-ph/0203033.
E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293;
Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003, hep-ph/9911323.
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hep-ph/9604387.
S.F. King, D.A.J. Rayner, Nucl. Phys. B 607 (2001) 77, hep-ph/0012076.
R. Barbieri, L.J. Hall, Y. Nomura, hep-ph/0106190;
R. Barbieri, L.J. Hall, Y. Nomura, Nucl. Phys. B 624 (2002) 63, hep-th/0107004.
V. Di Clemente, Y.A. Kubyshin, hep-th/0108117.
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55, hep-ph/9803466;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
S. Dimopoulos, G.F. Giudice, Phys. Lett. B 357 (1995) 573, hep-ph/9507282.
G.L. Kane, S.F. King, Phys. Lett. B 451 (1999) 113, hep-ph/9810374, and references therein.
D.E. Groom, et al., Particle Data Group Collaboration, Eur. Phys. J. C 15 (2000) 1.
LEP Higgs Working Group for Higgs Boson Searches Collaboration, hep-ex/0107029.
G.V. Gersdorff, M. Quiros, A. Riotto, hep-th/0204041.
A. Delgado, G. von Gersdorff, P. John, M. Quiros, Phys. Lett. B 517 (2001) 445, hep-ph/0104112;
R. Contino, L. Pilo, Phys. Lett. B 523 (2001) 347, hep-ph/0104130.
D.M. Ghilencea, H.P. Nilles, Phys. Lett. B 507 (2001) 327, hep-ph/0103151.

N = 1 and N = 2 super-YangMills theories from

wrapped branes
P. Di Vecchia a , A. Lerda b,c , P. Merlatti c
a NORDITA, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
b Dipartimento di Scienze e Tecnologie Avanzate, Universit del Piemonte Orientale, I-15100 Alessandria, Italy
c Dipartimento di Fisica Teorica, Universit di Torino,
and INFN, Sezione di Torino, Via P. Giuria 1, I-10125 Torino, Italy

Received 1 July 2002; accepted 7 October 2002
Abstract
We consider supergravity solutions of D5-branes wrapped on supersymmetric 2-cycles and use
them to discuss relevant features of four-dimensional N = 1 and N = 2 super-YangMills theories
with gauge group SU(N). In particular in the N = 1 case, using a gravitational dual of the gaugino
condensate, we obtain the complete NSVZ -function. We also find non-perturbative corrections
associated to fractional instantons with charge 2/N. These non-perturbative effects modify the
running of the coupling constant which remains finite even at small scales in a way that resembles to
the soft confinement scenario of QCD.
1. Introduction
Recently, it has become more and more evident that a lot of relevant information
about supersymmetric YangMills (SYM) theories can be obtained by studying their
dual supergravity backgrounds produced by stacks of D-branes. This gauge/gravity
correspondence has been thoroughly investigated in the maximally supersymmetric and
conformal case, where a precise duality can be established between the N = 4 SYM theory
in four dimensions with gauge group SU(N) and the type IIB supergravity in AdS5 S5
Work partially supported by the European Commission RTN programme HPRN-CT-2000-00131 and by
MIUR under contract 2001-025492.
E-mail addresses: divecchia@nbivms.nbi.dk (P. Di Vecchia), lerda@to.infn.it (A. Lerda),
merlatti@to.infn.it (P. Merlatti).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 9 0 6 - 9
44
P. Di Vecchia et al. / Nuclear Physics B 646 (2002) 4368
which is the near horizon geometry of N D3-branes in flat space [1]. Prompted by the
remarkable success of this AdS/CFT duality, a lot of activity has been devoted to extend the
gauge/gravity correspondence also to less supersymmetric and/or non-conformal theories.1
One way to reduce the amount of supersymmetry is to place a stack of D3-branes
at the apex of an orbifold [3] or of a conifold [4]. Depending on the details of the
background, the number of preserved supercharges can be reduced to eight or four, and
thus the SYM theories that correspond to these configurations will possess N = 2 or
N = 1 supersymmetry in four dimensions. Furthermore, both for the orbifold and the
conifold there is a very natural way to break conformal invariance, namely by means of
fractional D3-branes [5]. Therefore, in this way one can realize interesting non-conformal
SYM theories in four dimensions with N = 2 or N = 1 supersymmetry (for recent reviews
on this approach see, for example, Refs. [6,7]).
Another possibility to reduce the number of preserved supercharges is to consider
D-branes whose world-volume is partially wrapped on a supersymmetric cycle inside
a K3-manifold or a CalabiYau space. The unwrapped part of the brane world-volume
remains flat and supports a gauge theory. In order to preserve at least some supersymmetry,
the normal bundle to the wrapped D-branes has to be partially twisted [8], and as a
consequence of this twist, some world-volume fields become massive and decouple. This
procedure has been first used in Ref. [9] to study the pure N = 1 SYM theory in four
dimensions, and later it has been generalized to many other cases with different space
time dimensions and different amounts of supersymmetry [1014].2
In this paper we will use the gauge/gravity correspondence to study SYM theories
with N = 1 and N = 2 supersymmetry in four dimensions. In particular we reconsider
in some detail the supergravity solutions corresponding to N D5-branes wrapped on a
2-sphere that have been first found in Refs. [9] and [10,11] for the N = 1 and N = 2 case,
respectively. Even if these two solutions have been discussed already in several papers, here
we review again their derivation in order to setup the notation, to show their similarities
and differences, and also to be self-contained. In doing this, we also provide a derivation of
the MaldacenaNuez (MN) solution of Ref. [9] using the first order formalism.3 We then
consider the (bosonic) massless open-string modes that propagate on the flat part of the D5
world-volume and study their effective action at energies where the higher string modes, as
well as the KaluzaKlein excitations around the 2-cycles, decouple. This resulting theory
is simply a four-dimensional SYM theory with gauge group SU(N). In both the N = 2
and N = 1 case, it turns out that the coupling constant gYM and the vacuum angle YM of
this SYM theory can be expressed in terms of the ten-dimensional supergravity solution
representing N wrapped D5-branes in a very explicit and simple form.4 In particular, if
we denote by the radial coordinate on which the classical supergravity fields depend, we
1 See the introduction of Ref. [2] for a more detailed discussion.

2 We thank W. Lerche for pointing out that a different approach to wrapped branes has been considered in
Ref. [15].
3 A similar derivation in the first order formalism has been previously presented in Ref. [16].
4 This situation is very similar to the one encountered with fractional D3-branes in orbifold models (see, for
example, Refs. [2,1721]).
45
find that
1
2
gYM
= F (),
(1.1)
where the function F () is an explicitly computable function (see Eqs. (3.8) and (4.7)).
The other crucial ingredient for the gauge/gravity correspondence is the relation
between the radial parameter of the supergravity solution and the energy scales of the
gauge theory. For non-conformal theories in general it may be ambiguous and difficult to
establish such a relation [22]; however, for the two cases under consideration we manage
to find a supergravity realization of a protected operator of the gauge theory, so that a
definite radius/energy relation can be obtained.5 The protected operators we will consider
are the complex scalar field of the vector multiplet in the N = 2 theory, and the gaugino
condensate 2 in the N = 1 theory. In general, the energy/radius relation that we obtain
in this way takes the form
= G(),
(1.2)
where is the subtraction point at which the theory is defined, is the scale dynamically
generated by quantum corrections and G() is an explicit function which depends on the
model (see Eqs. (3.13) and (4.16)).
Once the functions F () and G() are determined, one can exploit them to find
explicit results for the gauge theory. For example, by eliminating between (1.1) and
(1.2), one can obtain the running coupling constant and hence the -function. In the
N = 2 case, these manipulations can be performed in an analytic way and lead to the
exact perturbative -function. In the N = 1 case, instead, these manipulations can only
be performed using asymptotic expansions but nevertheless lead to interesting results. In
particular, we will show that the complete NSVZ -function [23] of the N = 1 theory
can be obtained from the MN solution, once the appropriate energy/radius relation is
enforced. It is remarkable to see that the entire perturbative -function, and not only
the 1-loop approximation, is encoded in a classical supergravity solution! Actually,
the MN solution also shows the presence of non-perturbative effects in the form of
fractional instantons with charge 2/N which smooth out the running of the gauge coupling
constant that remains finite even at small energy. This smooth behavior as well as
the absence of the Landau pole indicate that the theory softly flows to the confining
phase, in analogy to the soft-confinement scenario of QCD [24]. Furthermore, the MN
solution also accounts for the chiral anomaly, the chiral symmetry breaking and the
correct action for the gauge instantons which we manage to realize explicitly as wrapped
D-strings.
The analysis of the IR regime of a gauge theory and the study of its non-perturbative
features by means of a supergravity solution are possible only if the latter is non-singular.
This is precisely the case of the MN solution which is regular even at small distances
due the presence of a field that, in Ref. [14], has been identified with the gravitational
5 Again, this situation is similar to the one of fractional D3-branes in orbifold models (see, in particular,
Ref. [21]).
46
dual of the gaugino condensate. In this respect, the MN solution is very different from
the N = 2 solution of Ref. [10,11] which exhibits a naked singularity of repulson type
and an enhanon locus [25] that prevent from obtaining information on the IR regime
of the dual N = 2 SYM theory and in particular on its instanton corrections.6 On
the contrary, the MN solution is similar to the KlebanovStrassler solution (KS) [26]
which describes a system of regular and fractional D3-branes on a deformed conifold.
The KS solution is free of singularities and can be successfully used to describe many
interesting features of a four-dimensional N = 1 gauge theory with chiral matter that,
through a series of duality cascades, eventually flows to the pure N = 1 SYM theory
in the deep IR to which our analysis can be extended [27]. A qualitative study of the
general implications of the MN and KS solutions for N = 1 gauge theories can be found
in Ref. [28].
The paper is organized as follows. In Section 2 we review the N = 2 supergravity
solution of Refs. [10,11] and the N = 1 solution Ref. [9] in a first order formalism. In
Section 3 we discuss the pure N = 2 SYM theory using the dual supergravity solution
of Section 2.1. In particular, we derive the running of the coupling constant and the
chiral anomaly, but differently from Ref. [10] we do not use the probe technique. In
Section 4 we study the pure N = 1 SYM theory and, by exploiting the supergravity
solution of Section 2.2, we discuss the chiral anomaly, the gaugino condensate, the running
coupling constant, the NSVZ -function and the instanton action. Finally, in Section 5 we
present our conclusions, while in Appendix A we give some technical details about the
parameterizations used to derive the supergravity solutions of Section 2.
2. The supergravity solutions

In this section we review the type IIB supergravity solutions that correspond to N = 1
and N = 2 SYM theories in four dimensions with gauge group SU(N) and no matter.
The starting point is a stack of N D5-branes with two longitudinal directions wrapped
on a supersymmetric 2-cycle. The difference between N = 1 and N = 2 arises from the
way in which the 2-cycle is embedded in the ambient space, a CalabiYau threefold for
N = 1 and a K3-manifold for N = 2. The four unwrapped longitudinal directions of the
D5-branes span a flat world-volume where the SYM theory lives. In order to preserve
the proper amount of supersymmetry the normal bundle to the five-branes has to be
topologically twisted [8]; this is achieved by embedding the U (1) spin-connection of the 2cycle into SO(4) which is the R-symmetry group of the D5-branes. As remarked in [9,29],
an efficient way to derive the geometry corresponding to this setup is to consider the SO(4)
gauged supergravity in seven dimensions, find domain-walls wrapped on a 2-sphere, and
then lift them up in ten dimensions.
The fields of the SO(4) gauged supergravity in d = 7 are the metric G , the gauge
ij
ji
fields A = A with i, j = 1, . . . , 4 being vector indices of SO(4), a symmetric matrix
Tij of scalars and a 2-form potential A(2)
[30]. The Lagrangian of this supergravity
6 The same problems occur in the fractional D3-brane solutions [2,1719].
theory is
L7 =
5
1
1
det G R(G) y y Tij1 D Tj k Tk
D Ti
16
4

1
1 y
ij k
1 F
e
ey/2 Tik1 T
F
H
H
V
,
j
8
12
47
(2.1)
where
Tij = ey/4 Tij ,
(2.2)

jk
ij + Aik

D Tij = T
Tkj + A Tik ,

ik kj
ij
ij
ik kj
F
= Aij
A + A A A A ,
(2.3)
ey = det T ,
and

2 y/2 ij
e
2Tij T (Tii )2 .
2
Finally, H are the components of the following 3-form

1
H (3) = dA(2) + !ij k F ij Ak Aij Akm Am .

8
3
V=
(2.4)
(2.5)
(2.6)
In these equations denotes the SO(4) gauge coupling constant which has dimensions of
a (length)1.
As mentioned above, we are interested in domain-walls of this supergravity theory that
are wrapped on a 2-sphere. Thus we look for metrics of the form
2

1
ds72 = e2f (r) dx1,3
(2.7)
+ dr 2 + 2 e2g(r) d22 ,
2 is the Minkowski metric on R , r is the transverse coordinate to the domainwhere dx1,3

1,3
wall, and d22 = d 2 + sin2 d 2 (with 0 and 0 2 ) is the metric of a
unit 2-sphere.7 To fully determine the domain-wall configuration we must specify also the
profile of the other supergravity fields. However, to do this we have to distinguish between
the N = 1 and N = 2 cases which require different types of ansatz. In the next subsection
we review the N = 2 case [10,11], while the N = 1 case [9] will be considered in the last
subsection.
2.1. The N = 2 solution
In order to obtain a solution corresponding to a gauge theory with eight supercharges in
four dimensions, one must first truncate the SO(4) gauge group of the d = 7 supergravity
to its diagonal U (1) subgroup, i.e., to the Abelian part of the diagonal SU(2)D SO(4).
This truncation can be achieved, for example, by setting
A12
A
7 Notice that the factor of 2 in (2.7) is necessary for dimensional reasons.
(2.8)
48

ij
with all other A s put to zero. The consistency of this Abelian reduction requires that the
ij takes the form [31]
unimodular matrix T

Tij = diag ex , ex , ex , ex .
(2.9)
With these positions it is easy to realize that the Lagrangian (2.1) becomes
5
1
L7 = det G R(G) y y x x e2xy/2F F
16
4

1
ey H H + 42 ey/2 ,
12
(2.10)
where F = [ A] and H = [ A(2)

] .
We now look for solutions of the field equations following from the Lagrangian
(2.10) in which the metric has the form (2.7) and the fields x and y are functions only
of the transverse coordinate r. To implement the topological twist that preserves eight
supercharges, we must identify the U (1) gauge field with the spin-connection on the
2-sphere and thus write
1
A = cos d ,
(2.11)
where the coupling constant has been introduced for dimensional reasons. Furthermore,
we can consistently set A(2) = 0.
The task is then reduced to find the four functions f (r), g(r), x(r) and y(r) that solve
the field equations. This has been done explicitly in Ref. [10]8 where it has been shown
that it is consistent to impose the relation y = 4f and that, after defining
3
k = f + g,
(2.12)
2
it is possible to derive the field equations for h(r), k(r) and x(r) from the auxiliary
Lagrangian

L = e2k 4k 2 2h 2 x 2 V ,
(2.13)
h = g f,
where dots denote derivative with respect to r and

2 4h2x
(2.14)
e
2
together with the additional condition that the Hamiltonian H associated to L vanishes.
This condition, which reads

1 2k 1 2 1 2
2
H= e
(2.15)
p p px + e2k V = 0,
4
4 k 2 h
V = 42 22 e2h +
is a signal of the fact that the second-order field equations can be derived from a system
of first-order (or BPS) equations, as one can show with a HamiltonJacobi approach [32].
8 See also Ref. [11].
49
F
F
In fact, by introducing the principal function F (k, h, x) such that pk = k
, ph = h
,
F
2k
px = x and assuming that F (k, h, x) = e W(h, x), it is easy to see that (2.15) becomes

1 W 2 1 W 2 1 2
+
W = V,
8 h
4 x
4
(2.16)
which is solved by
W = 4 cosh x e2hx .
(2.17)
The remaining Hamilton equations yield the following first-order system

1
k = W,
4
1 W
h =
,
4 h
x =
1 W
,
2 x
(2.18)
which, as shown in Ref. [10], can be conveniently solved in terms of the variable z e2h .
In fact, one finds
e2x = 1
1 + ce2z
,
2z
e2k+x = ze2z ,
(2.19)
where c is an arbitrary integration constant. Thus, choosing z as new radial coordinate, the
functions appearing in the domain-wall solution are
1
1
2
g(z) = f (z) + log(z),
y(z) = 4f (z)
f (z) = z x(z),
(2.20)
5
5
2
with x(z) given in (2.19).
We now up-lift this solution to ten dimensions in order to exhibit it as a D5-brane
configuration of type IIB supergravity. This up-lift can be explicitly performed using the
formulas of Ref. [31] which lead to a metric, a dilaton and a 2-form of magnetic type in
ten dimensions. In this respect, however, we would like to point out that in Ref. [31] only
the up-lift to NS5-brane configurations is considered. Indeed, the magnetic 2-form that
is produced in ten dimensions has to be identified with the NSNS antisymmetric tensor
B in view of the sign that is chosen for the exponential coupling with the dilaton in the
Einstein frame. However, it is straightforward to adjust the up-lift formulas of Ref. [31]
to the other sign and thus produce a magnetic 2-form which can be identified with a RR
potential C (2) in ten dimensions. In this way, the seven-dimensional solution can be directly
up-lifted to a D5-brane configuration of type IIB, which is what one needs to discuss the
correspondence with a dual SYM theory in four dimensions.
With this in mind, we parameterize the 3-sphere that leads from seven to ten dimensions
with the angles , 1 and 2 (with 0 2 and 0 1,2 2 ) as discussed in
Appendix A (see Eq. (A.2)), and use the (modified) up-lift formulas of Ref. [31] to find
a ten-dimensional (string frame) metric given by

1
z
2
2
+ 2 d 2 + sin2 d 2 + 2 e2x dz2
ds10 = e dx1,3

x

2 ex
1
e
cos2 d1 + cos d + sin2 d22 ,
+ 2 d 2 +
(2.21)
50
a dilaton given by

1 + ce2z
,
e2 = e2z 1 sin2
2z
and a magnetic RR 2-form given by
2

sin
1
(d
+
cos
d
)
,
C (2) = 2 2 d
1
ex
(2.22)
(2.23)
where
= ex cos2 + ex sin2 .
(2.24)
We remark that this D5-brane solution agrees with the one of Ref. [10] which is obtained by
performing a S-duality transformation on a NS5-brane configuration. However, the metric
(2.21) is written in a way in which the role of the different coordinates and factors is not
immediately clear. Thus, in analogy with what has been done in Ref. [13], we perform the
following change of variables
= z cos ezx
= sin ez ,
(2.25)
and rewrite Eq. (2.21) as follows

z 2
H 1/2
2
1/2
2
2
2
ds10 = H
dx1,3 + 2 d + sin d
+ 2 d 2 + 2 d22

H 1/2 2
+ 2 d + 2 (d1 + cos d )
2 ,
z
(2.26)
where H e2 . In this form the structure of the metric is much clearer. In fact, one
can distinguish the transverse space into a plane parameterized by and 2 (see the last
two terms in the first line of (2.26)), and a non-trivial twisted plane pertaining to the
K3-manifold (see the second line of (2.26)). Thus one can say that the two (dimensionless)
coordinates and defined in (2.25) represent two radial directions, respectively in the
flat transverse space and in the curved space transverse to the brane but non-trivially fibered
on the 2-cycle along which the brane is wrapped. In the next section we will show that the
flat radial coordinate is directly related to the energy scale of the dual four-dimensional
gauge theory, and that shifts in the angle 2 are directly related to chiral transformations.
We conclude this subsection by observing that the RR charge of the configuration
(2.21)(2.23) is given by

1
2 2
dC (2) = 2 ,
Q5 = 2
(2.27)
210
10 2
S3
where 10 = 8 7/2gs 2 is the gravitational coupling constant of the type IIB string theory
and S3 is the transverse 3-sphere at infinity. Since this charge must be an integer multiple
of the elementary D5 charge
5 = (2)5 gs1 3 ,
(2.28)
51
we deduce that the seven-dimensional coupling constant can be written in terms of string
theory parameters as follows
1
= Ngs ,
2
where N is the number of wrapped D5-branes.
(2.29)
2.2. The N = 1 solution

We now discuss the MN supergravity solution of Ref. [9] which corresponds to a gauge
theory with four supercharges in four dimensions. As before, we start from the SO(4)
gauged supergravity Lagrangian (2.1), but this time we gauge one of the two SU(2) factors
inside SO(4) (say, for example, SU(2)L ). This is achieved by imposing the anti-selfduality
constraint
1 ij k k
A
Aij
(2.30)
= !
2
which explicitly reads
12
3
A34
= A A ,
31
2
A24
= A A ,
23
1
A14
= A A .
(2.31)
The vectors Aa (with a = 1, 2, 3) are the gauge fields of SU(2)L whose field strengths are
F a = dAa + abc Ab Ac as one can see by inserting (2.30) and (2.31) into (2.4). The
consistency of this SU(2)L reduction requires that the unimodular matrix Tij be simply
[31]
Tij = ij .
(2.32)
With these positions the Lagrangian (2.1) becomes

5
1
a
F a
L7 = det G R(G) y y ey/2 F
16
2

1 y
2 y/2
e H H
+ 4 e
,
12
(2.33)
where H are the components of the 3-form

abc a
! A Ab Ac .
(2.34)
12
Like in the N = 2 case, we look again for domain-wall solutions where the metric is of
the form (2.7) and the scalar y is function only of the radial coordinate r. To implement the
topological twist that preserves four supercharges we should identify the spin-connection
on the 2-sphere with the gauge field of a U (1) SU(2)L . If we simply do this, the
corresponding supergravity solution turns out to be singular and unphysical. However, as
discovered in Ref. [9], the singularity can be smoothed out by considering a more general
ansatz in which all gauge fields of SU(2)L are switched-on, namely by taking [33]
H (3) = dA(2) + F a Aa
A1 =
a(r) d ,
2
A2 =
1
a(r) sin d ,
A3 =
1
cos d .
(2.35)
52
With this ansatz it is easy to realize that the only non-vanishing components of the field
strengths are
Fr1 =
1
a,
Fr2 =
1
sin a,
F3 =

1
sin 1 a 2 .
2
(2.36)
Moreover one finds that F a Aa = A1 A2 A3 = 0 so that the 3-form (2.34) simply

becomes H (3) = dA(2) . Therefore, it is consistent to set A(2) = 0 and look for a classical
configuration that depends only on the functions f (r), g(r), y(r) and a(r).
Inserting this ansatz into the supergravity field equations that follow from (2.33), we find
that it is possible to set y = 4f and derive the remaining equations from the auxiliary
Lagrangian

a 2
L = e2k 4k 2 2h 2 e2h V ,
(2.37)
2
where h and k are defined as in (2.12) and
2 (1 a 2 )2 4h
(2.38)
e
4
provided we require that the Hamiltonian H associated to L vanishes. This condition
explicitly reads

1 2k 1 2 1 2
2h 2
p p e pa + e2k V = 0.
H= e
(2.39)
2
8 k 4 h
V = 42 22 e2h +
We now proceed as in the N = 2 case with the HamiltonJacobi method and introduce the
F , p = F , p = F . Then, if we assume
principal function F (k, h, a) such that pk = k
h
a
h
a
that F (k, h, a) = e2k W(h, a), Eq. (2.39) becomes

1 W 2 1 W 2 2h 1 2
(2.40)
+
e W = V,
8 h
2 a
4
which is solved by (see also Ref. [16])

2

W = e2h 1 + 4e2h + 2 1 + 4e2h a 2 + a 4 .
(2.41)
Having the explicit form of W, the problem is reduced to a system of first-order (BPS)
equations given by
1 W
W
1
h =
,
a = e2h
k = W,
4
4 h
a
which can be explicitly solved. Indeed one finds
e2h = coth 2
sinh 2
,
2
2
a=
,
sinh 2
e2k = eh
1
,
sinh 2 4
2
(2.42)
(2.43)
(2.44)
(2.45)
53
where r. This solution has been first found in Ref. [33], although in a different
context. The functions originally appearing in the domain-wall solution are then

sinh 2
1
f () = log
(2.46)
,
g() = f () + h(),
y() = 4f ()
5
2eh
with h() given in (2.43). It is interesting to observe that a e2 for ; thus only
the A3 component of the SU(2)L gauge field effectively survives in the large region and
the gauge group reduces to U (1).
We now up-lift this solution to ten dimensions to exhibit it as a D5-brane configuration
of type IIB supergravity. To this aim we could choose the same parameterization of the
3-sphere that we used in the previous subsection when we up-lifted the N = 2 solution.
However, for the applications to the dual gauge theory that we will discuss in the following,
it is more convenient to choose a different parameterization and describe the 3-sphere with
the Euler angles , and (with 0 , 0 2 and 0 4 ) as described
in (A.4), and use the corresponding left-invariant 1-forms a given in (A.5). Then, using
the (modified) up-lift formulas of Ref. [31], we find a ten-dimensional (string frame) metric

3

a
2
e
e2h
2
2
Aa ,
ds10
= e dx1,3
+ 2 d 2 + sin2 d 2 + 2 d 2 +
a=1
(2.47)
a dilaton
sinh 2
,
2eh
and a magnetic RR 2-form
e2 =
(2.48)

1
sin d d sin d d cos cos d d
2
4

a
+ 2 d 1 sin d 2
2
with field strength
C (2) =

1 a
2 1
A1 2 A2 3 A3
F a.
2
(2.49)
F (3) =
(2.50)
a=1
By computing the RR charge Q5 of this configuration (see Eq. (2.27)) and imposing its
quantization in units of the elementary D5 charge 5 , we find again that
1
(2.51)
= Ngs ,
2
where N is the number of wrapped five-branes. The solution (2.47)(2.50) entirely agrees
with the one of Ref. [9].
We conclude this section by observing that there are many common features and
similarities between the ten-dimensional N = 2 and N = 1 solutions as we have presented
them here, despite the fact that we have used different angular variables in the two cases.
After all, this is not too surprising since both solutions originate from the same ansatz
54
(2.7) in seven dimensions; but nevertheless it is noteworthy to see that in the first-order
formalism the two solutions are very similar and that a flow between them could be
in principle established in terms of the superpotentials W defined in Eqs. (2.17) and
(2.41). Of course there are also many important differences between the two solutions. In
particular we would like to emphasize that in the N = 2 case one can distinguish a plane
in the transverse space where the wrapped D5-branes can move, while in the N = 1 case
no such plane exists. Moreover the N = 2 solution is singular at short distance while the
N = 1 solution is regular. These facts have important consequences for the dual SYM
theories as we shall discuss in the next sections.
3. The N = 2 super-YangMills theory
The supergravity solution presented in Subsection 2.1 describes the geometry produced
by N D5-branes wrapped on a supersymmetric 2-cycle in such a way that one quarter of
the 32 supercharges of type IIB are preserved. The four-dimensional part of the D5 worldvolume that is not involved in the wrapping process remains flat and is the Minkowski
spacetime R1,3 where a supersymmetric gauge theory with 8 supercharges is defined. To
specify this theory one must determine which massless open-string modes can propagate
in R1,3 . By applying for example the methods of Ref. [8], one can find that these modes fill
a N = 2 vector multiplet in the adjoint representation of SU(N), and thus the low-energy
four-dimensional gauge theory is a pure N = 2 SYM theory with gauge group SU(N).
We now show how the supergravity solution (2.22), (2.23) and (2.26) can be used to
extract information on the corresponding gauge theory. First of all, we observe that in the
transverse space to the D5-branes one can select a two-dimensional subspace in which
the branes can freely move. This is the locus = 0 and thus is the plane parameterized
by and 2 . To find this result we can use the probe technique [6,34] and study the
dynamics of a wrapped D5-brane in the background geometry (2.22)(2.26). The (string
frame) world-volume action for a probe D5-brane is

6
(n)
2 F

det(G + 2 F ) + 5
C e
, (3.1)
S = 5 d e
6-form
n
where 5 is defined in (2.28), F is a world-volume gauge field and all bulk fields are
understood to be the pull-backs onto the brane world-volume which is parameterized by
By expanding the action (3.1) in powers of and substituting
= {x 0 , x 1 , x 2 , x 3 , , }.
in it the solution (2.22), (2.23) and (2.26), we find that there is a static potential between
the probe and the source given by

2 H 2 1/2
z
6
1 .
5 d 2 sin 1 +
(3.2)
H
z2 tan2
In order to satisfy the zero-force condition and hence describe a BPS supersymmetric
configuration, this potential term should vanish. Thus, we should require that the probe
brane be placed at = 0. This means that in order to preserve supersymmetry the branes
cannot move arbitrarily in their transverse space but only in the locus = 0, i.e., in the
(, 2 )-plane.
55
From now on, we abandon the probe approach but still we set = 0. Then, the relevant
part of the transverse D5-brane metric (2.26) can be simply written as
2
= H 1/2 dZ dZ,
dstransv
(3.3)
where we have defined

1
Z = ei 2 ,
and the RR 2-form (2.23) simply reduces to
(3.4)
1
(3.5)
2 sin d d .
2
After these preliminaries we now study the dynamics of the SU(N) gauge fields that
propagate on the wrapped D5-branes. As discussed in Refs. [2,17,21] for the analogue
case of the fractional branes, the low-energy action for the bosonic degrees of freedom of
the non-Abelian gauge theory can be inferred from the Abelian world-volume action (3.1)
with the following procedure: take the limit 0 keeping fixed the combination
C (2) =
= (2 )1 Z
(3.6)
that has to play the role of the complex scalar of the N = 2 vector multiplet, and then
promote F and to SU(N) fields by giving them an adjoint index A and replacing all
derivatives with the covariant ones.9 If we normalize the SU(N) generators in such a way
that tr(T A T B ) = (1/2) AB for the fundamental representation, then the above procedure
leads to

1
1 A 1
YM
A
4
A
F FA + D D A +
SYM = 2
FA ,
d x
d 4 x F
2
4
2
32
gYM
(3.7)
where
1
2
gYM
5 (2)2 2
2
4 2
YM = 5 (2)
d
0
N
d e H det G =
log ,
4 2
(3.8)
d C = 2N2 .
(3.9)
(2)

In other words, the inverse square of the YM coupling constant is proportional to the
volume of the 2-sphere along which the D5-branes are wrapped computed in the tendimensional metric, whereas the YM -angle is proportional to the flux of the RR 2-form
across this 2-sphere. Notice that in deriving these results we have used Eqs. (2.28) and
(2.29), the explicit form of the metric (2.26), of the dilaton (2.22) and of the RR 2-form
9 In flat space or in orbifold backgrounds where the classical D-brane solutions can be explicitly obtained using
boundary states (see, e.g., Ref. [35]), this procedure can be made more rigorous by computing string scattering
amplitudes as done for example in Ref. [36].
56
(3.5), as well as the fact that Eq. (2.25) implies that z = log for = 0. In the following
two subsections we will use the results (3.8) and (3.9) to derive the -function and the
chiral anomaly of the N = 2 SYM theory.
3.1. Running coupling constant and -function
To obtain further information on the gauge theory we have to find the precise
relationship between the gravitational coordinates and the scales of the gauge theory which,
in some sense, is the essence of the gauge/gravity correspondence. In the present case, it is
not difficult to find this relationship. In fact, let us denote by the (arbitrary) mass-scale
at which the theory is defined and by the (fixed) dynamically generated scale which is
the analogue of QCD for quantum chromodynamics. Under a scale transformation with
parameter s the mass transforms according to
s
(3.10)
but the physical mass remains fixed. This is the signal of the lack of conformal invariance
in this theory. On the other hand, it is known that the complex scalar of the N = 2 vector
multiplet has a (protected) mass-dimension 1, that is
s.
(3.11)
If we now use Eqs. (3.6) and (3.4), we easily conclude that the scale transformation (3.11)
implies
s,
(3.12)
i.e., in the dual gravitational description the scale transformation of the gauge theory is
realized as a dilatation of the 2-plane in the transverse space of the wrapped D5-branes.
Combining the fact that is dimensionless with the transformations (3.10) and (3.12), we
are lead the following identification
= .
(3.13)
This is a sort of holographic relation, at least in an extended sense, because it establishes

a correspondence between a quantity of the gravity theory (the coordinate ) and a quantity
of the gauge theory (the scale ) even if it does not follow from a standard holographic
bulk/boundary relation. The fact that and are directly proportional to each other can
also be established by looking at the energy of a string that stretches up to a distance in
the transverse plane. The energy E of such a string is proportional to its world-volume per
unit time [7,22], namely

E
G00 G d ,
(3.14)
where we have used the metric (2.26). This energy is the natural scale to use in order to
regulate the theory and hence it is natural to take E , which in turn leads to (3.13).
Notice that this relation implies that the UV (IR) regime of the gauge theory corresponds
to the large (small) distance region in the dual gravitational description.
57
If we insert (3.13) into (3.8) we can obtain the expression of the running coupling
constant at the scale , namely
1
2 ()
gYM
log ,
2
(3.15)
from which we can derive the -function

N 3
(3.16)
g .
8 2 YM
This is the correct field-theory result, numerical factors included. We would like to stress
that this -function has been obtained without using the probe approach which instead
is appropriate to describe the theory in the Coulomb branch (for a recent review, see
Ref. [6]).10
(gYM ) =
3.2. Chiral anomaly

We now discuss the chiral anomaly from the dual gravitational point of view. It is well
known that the N = 2 SYM theory possesses at the classical level an Abelian U (1)R
symmetry which becomes anomalous at the quantum level. For definiteness, we fix the
R-charge of the complex scalar to be 2, so that the gauginos have the conventional
R-charge 1. This means that under a chiral transformation with parameter we have
e2i ,
(3.17)
and that the chiral anomaly corresponds to the following transformation of the -angle
YM YM 4N.
(3.18)
From Eqs. (3.6) and (3.4) it is natural to interpret the transformation (3.17) as due to a shift
in the angle 2 , namely
2 2 + 2,
(3.19)
which is also consistent with the anomaly transformation (3.18) in view of Eq. (3.9).
Thus, the chiral U (1)R transformations of the gauge theory are realized in the dual
gravitational description as phase rotations in the transverse plane where the branes can
move with a multiplicity factor (2 in our case) that depends on the charge assignments.
As one can easily see from the explicit expressions (2.22), (2.23) and (2.26), the
transformation (3.19) is an isometry of the metric but it is not a symmetry of the full
solution since the RR 2-form C (2) explicitly depends on the angle 2 and is not invariant
under (3.19). This fact is the gravitational counterpart of the occurrence of a chiral anomaly
in the dual gauge theory [21,37]. However, not all different values of the RR 2-form
C (2) are physically different: in fact what really matters is the value of the flux across the
10 The -function for the N = 2 SYM theory in the Coulomb branch has been first derived from the wrapped
D5-brane solution in Ref. [10] up to some numerical factors.
58
2-sphere
1
4 2 gs

C (2) =
N2
(3.20)
S2
which is allowed to change by integer values. Thus, the following shifts
2 2 + k
(3.21)
N
(with k integer) are true symmetries of the supergravity solution. On the gauge theory side,
these shifts correspond to the non-anomalous Z4N rotations with parameter = (/2N)k,
which change the -angle by integer multiples of 2 , as one can easily see from (3.18).
4. The N = 1 super-YangMills theory
The supergravity solution presented in Subsection 2.2 describes the geometry produced
by N D5-branes wrapped on a supersymmetric 2-cycle in such a way that one eighth of the
32 supercharges of type IIB are preserved. The four-dimensional unwrapped part of the D5
world-volume is the Minkowski spacetime where a supersymmetric gauge theory with 4
supercharges is defined. By using the methods of Ref. [8], one can see that the massless
open string modes that propagate in this Minkowski space fill a N = 1 vector multiplet
in the adjoint representation of SU(N), and thus the low-energy four-dimensional gauge
theory is a pure N = 1 SYM theory with gauge group SU(N).
We now show how to use the supergravity solution (2.47)(2.49) to extract information
on the corresponding gauge theory. To this aim, we consider again the world-volume action
(3.1) of a wrapped five-brane with a gauge field strength F and then extract from it the
quadratic terms in F . In contrast to the N = 2 case discussed in the previous section, this
time there is no direction in which the branes can move, since the entire transverse space
is curved as a consequence of the topological twist of the normal bundle that has been
performed. Thus, no scalars survive and the non-Abelian bosonic action that is obtained
with the procedure discussed in Section 3, is simply

1
YM
A
A
FA +
d 4 x F
d 4 x F
FA ,
SYM = 2
(4.1)
32 2
4gYM
where
1
2
gYM
5 (2)2 2
=
2
YM = 5 (2)4 2
d e3 det G,
(4.2)
d C .
(4.3)
(2)

Exactly as in the N = 2 case (see Eqs. (3.8) and (3.9)) the inverse square of the YM
coupling constant is proportional to the volume of the 2-sphere on which the D5-branes are
59
wrapped and the YM vacuum angle is proportional to the flux of the RR 2-form across
this 2-sphere. However, we would like to stress that our definition (4.2) of the YM coupling
constant is different from the one used in Ref. [9] and several other papers that followed
afterwards, because we compute the volume of the 2-sphere using the ten-dimensional
metric of the wrapped D5-branes, and not the metric (2.7) of the domain-wall solution of
the seven-dimensional gauged supergravity. Our definition is the natural one from a stringtheory point of view; on the other hand, no one doubts that the YM -angle is related
to the flux of the RR form in ten dimensions, and thus it appears more acceptable that
also the other parameter of the SYM action be related to quantities of the ten-dimensional
solution as in (4.2). The difference between our definition of the YM constant and the one
of Ref. [9] will have interesting consequences for the gauge theory, especially in the IR
regime.
If we insert the explicit form of the supergravity solution (2.47)(2.49) in Eq. (4.2), after
simple calculations we obtain

N
cot2 1/2
1
sin 1 +
=
Y
()
d
(4.4)
2
Y ()
32 2
gYM
0
with
Y () = 4e2h() + a()2 = 4 coth 2 1,
(4.5)
where in the last step we have used Eqs. (2.43) and (2.45). It is interesting to observe that
the right-hand side of Eq. (4.4) can be written in terms of the complete elliptic integral of
the second kind
/2
d 1 x 2 sin2 ;
E(x)
(4.6)
in fact, we find
1
2
gYM
NY ()
=
E
16 2

Y () 1
.
Y ()
(4.7)
Using the properties of the elliptic integral, it is easy to see that

N
(4.8)
, for ,
4 2
1
N
, for 0.
(4.9)

2
gYM 32
The large- behavior (4.8) is the same as in Ref. [9], but the small- behavior (4.9) is very
different, since we get a finite coupling at = 0, in contrast to the divergent one of Ref. [9].
Finally, inserting in Eq. (4.3) the classical profile of the RR 2-form (2.49), we find
1
2
gYM
N
(4.10)
.
2
In the following we will exploit these results to discuss some relevant features of the pure
N = 1 SYM theory from the gravitational point of view.
YM =
60
4.1. Chiral anomaly

As is well known, the N = 1 SYM theory possesses a classical Abelian U (1)R
symmetry which becomes anomalous at the quantum level. If we assign R-charge 1 to
the gaugino (x) so that under a chiral transformation with parameter
(x) ei (x),
(4.11)
then the presence of the anomaly implies that

YM YM 2N.
(4.12)
From Eq. (4.10) it is clear that the chiral transformations of the SYM theory must be
realized in the gravitational description as shifts in the angle . The only question is which
is the multiplicity factor. Now we argue that the correct transformation law of is
+ 4,
(4.13)
which indeed allows to obtain the anomaly rule (4.12) directly from (4.10). The factor
of 4 in (4.13) has a natural explanation: in fact it is /2 that is the appropriate angular
variable with period 2 which must transform with multiplicity 2, just like the angle 2 in
the N = 2 case (see Eq. (3.19)). This argument can be made more rigorous by observing
that there is a simple relation between the Euler angles that we used to parameterize the
3-sphere in the up-lifting process of the N = 1 solution and the angles that instead we
used in the up-lift of the N = 2 solution. In particular, as is discussed in Appendix A,
= 1 + 2 (see (A.8)), and since under a chiral transformation of parameter both 1
and 2 shift by 2, then Eq. (4.13) immediately follows.
The transformation (4.13) is not a symmetry of the supergravity solution (2.47)(2.49)
and is not even an isometry of the metric. This situation is thus very different from
the N = 2 case discussed in the previous section. However, there is a region where the
transformation (4.13) is an isometry, namely the large- region where the function a()
becomes exponentially small and can be neglected. In fact, if we remove a, then all explicit
dependence disappears from the metric (2.47) but still remains in the RR 2-form (2.49),
which therefore is not invariant under (4.13). However, as we discussed in Section 3.2, the
relevant quantity that should be considered is the flux of C (2) across the 2-sphere, i.e.,

1
N
(2)
(4.14)
C
=

2

4
4 gs
a=0
S2
which is allowed to change by integer values. Thus the transformations (4.13) with
= (/N)k and k integer are true symmetries of the supergravity background in the large region. These are precisely the non-anomalous Z2N transformations of the gauge theory
that correspond to shifts of the -angle by integer multiples of 2 . What we have described
here is therefore the gravitational counterpart of the well-known breaking of U (1)R down
to Z2N .
However, in the true supergravity solution a() is not vanishing and thus even the nonanomalous Z2N transformations are not symmetries of the background. In fact, both in
the metric and in the RR 2-form, a() appears in front of terms that explicitly involve
61
cos and sin , and these are clearly invariant only under shifts of 2 . Thus, the only nonanomalous symmetries of the supergravity background are given by (4.13) with = k and
k integer. This phenomenon is the gravitational counterpart of the spontaneous breaking of
the chiral symmetry from Z2N Z2 .
4.2. Gaugino condensate
In the previous subsection we have seen that the presence of a non-vanishing a() in the
supergravity solution is responsible for the spontaneous chiral symmetry breaking to Z2 ,
which, on the gauge theory side, is accompanied by the presence of a non-vanishing value
Tr 2
of the gaugino condensate 2 0| 16
2 |0. Thus, it appears very natural to conjecture
that the gravitational dual of this condensate is precisely the function a() that is present in
the supergravity solution (2.47)(2.49). As a matter of fact, this idea has been made more
precise in Ref. [14] where a direct relation between 2 and a() has been established by
adapting the techniques of the AdS/CFT correspondence to the wrapped branes.
The gaugino condensate 2 belongs to a class of gauge invariant operators which
do not acquire any anomalous dimensions. This happens because the gaugino condensate
is the lowest component of the so-called anomaly multiplet whose scale dimensions are
protected by virtue of the fact that its top component is the trace of the energy-momentum
tensor which is a conserved current. Thus, since the engineering dimension of 2 is 3,
we have
2
= c3 ,
(4.15)
where is the (exact) dynamical scale of the N = 1 SYM theory and c a computable
constant. Several explicit calculations lead to c = 1 [38].
In view of the findings of Ref. [14] and of our previous discussion, we now propose
to identify the function a() given in (2.45) with the gaugino condensate 2 measured
in units of the (arbitrary) mass scale that is introduced to regulate the theory. Thus, we
write
a() =
3
.
3
(4.16)
This crucial equation allows us to establish a precise relation between the supergravity
radial coordinate and the scales of the gauge theory, and is the strict analogue of the
holographic relation (3.13) of the N = 2 model. Notice that since a() 0 for ,
once again the large- region corresponds to the UV regime of the gauge theory, and vice
versa the small- region corresponds to the IR regime. In the following we will exploit the
relation (4.16) to compute the perturbative -function of the pure N = 1 SYM theory and
get some insights on non-perturbative corrections.
4.3. Running coupling constant and -function
The gauge coupling constant has been defined in (4.4) as a function of the radial
coordinate which, in turn, is related to the scales of the gauge theory as prescribed
by (4.16). Thus, by combining these two equations and eliminating the dependence,
62
we could in principle obtain an exact expression for the running coupling constant of
the N = 1 SYM theory. Unfortunately, it appears very difficult to manipulate Eqs. (4.4)
and (4.16) in an analytic way and exhibit the running coupling constant in a closed form.
However, despite this fact, many interesting results can still be derived.
First of all, if we use in (4.16) the explicit expression of a() given in (2.45), we easily
obtain
3
1
(4.17)
=
.
log(/) 2 1 (2)1 + 2e4 (1 e4 )1
The right-hand side has a nice interpretation in the large- region, i.e., in the UV. In fact,
for the fraction in square brackets receives two types of contributions: one from
terms that are negative powers of , and the other from terms that involve also powers of
e4 . If we recall that in the UV region is directly proportional to the inverse square
of the gauge coupling constant as shown in (4.8), it is very easy to realize that the first
type of terms corresponds to perturbative loop contributions while the second describes
non-perturbative instanton-like effects. Actually, we can be more precise and write explicit
formulas. Let us first concentrate on the power-like part and neglect all terms that vanish
exponentially for . Then, if we use the leading UV asymptotic behavior (4.8) and
2 ), on the one hand we have
thus trade for 4 2 /(NgYM
8 2
= 3 (gYM ),
log(/)
NgYM
(4.18)
where (gYM ) is the -function, and on the other hand from (4.17) we have
2 1
NgYM
3
= 1
.
log(/) 2
8 2
(4.19)
Combining these two equations, we obtain

(gYM ) =
2 1
NgYM
3NgYM
1
.
16 2
8 2
(4.20)
This is the complete perturbative NSVZ -function of the pure N = 1 SYM theory with
gauge group SU(N) in the PauliVillars regularization scheme [23]. It is remarkable to see
that a classical supergravity solution representing wrapped D5-branes is able to completely
reproduce it!
It is interesting to observe that the 1-loop approximation corresponds to keep in the
supergravity solution only the leading terms in the expansion. In fact, if in (4.17)
we approximate the square bracket to 1, we get 32 log(/) which is indeed the correct
relation between the radial coordinate and the scale at 1-loop [14]. If instead we keep
also the 1 term in (4.17), we obtain the complete perturbative result (4.19) and the NSVZ
-function. However, the full supergravity solution contains more than this, since there are
also terms that vanish exponentially for large and correspond to non-perturbative effects.
Keeping these terms in (4.17) amounts to replace the expression inside the square brackets
63
of (4.19) and (4.20) with

2
2 exp 162
2
NgYM
NgYM
1
+
.

2
8 2
1 exp 162
(4.21)
NgYM
From this formula we see that the non-perturbative corrections have the form of instantons
with fractional charge 2k/N where k is a positive integer. The fractional instantons have
recently been shown in Ref. [39] to play a crucial role in the pure N = 1 SYM theory
since they are the elementary field configurations that directly contribute to the gaugino
condensate. It would be very interesting to check with a field theory analysis whether they
also contribute to the -function as our supergravity analysis suggests.
We conclude by observing that the presence of these non-perturbative effects modifies
the running of the coupling constant which remains finite even at low energy. In fact, from
(4.16) we see that the IR limit corresponds to the limit 0 in which the coupling
constant smoothly tends to a finite value (see Eq. (4.9)). This smooth behaviour and the
absence of a Landau pole qualitatively resembles the soft confinement scenario of QCD.
4.4. Instantons
It well known that a system of N D3-branes and k D(1)-branes in flat space describes
the k-instanton sector of the maximally supersymmetric SU(N) YangMills theory in four
dimensions [40]. This idea has been successfully generalized to other less supersymmetric
configurations by considering, for example, systems of D3/D(1) fractional branes in
orbifold backgrounds. Thus, it appears very natural to think that the same happens also
in the context of wrapped branes. The idea is then to consider a system of N D5-branes
and k D1-branes wrapped on the same supersymmetric 2-cycle and see whether the k
wrapped D-strings branes account for the k instanton contributions. This analysis has been
performed in Ref. [28] where, however, negative conclusions have been reached. We now
show that this is not the case and that, also in the wrapped brane case, the instantons are
correctly represented by D-branes with four dimensions less than the branes which support
the gauge degrees of freedom.
The world-volume action of a Euclidean D1-brane in the string frame is given by

S = 1 d 2 e det G i1 C (2) ,
(4.22)
where
1 =
1
2gs
(4.23)
and the bulk fields are understood to be the pull-backs onto the brane world-volume which
is parameterized = { , }.
If the D1-brane is in the background geometry produced by N
wrapped D5-branes, we must use in the above action the metric, the dilaton and the RR
2-form of the solutions reviewed in Section 2. In particular in the N = 1 case, by using the
64
explicit form of the metric given in (2.47) we get

e
cot2
,
det G = 2 Y () sin 1 +
Y ()
4
(4.24)
where Y () is defined in (4.5). Inserting this result into the action (4.22), recalling from
(2.49) that the relevant component of the RR 2-form is
1
sin ,
42
and performing the integrals over the world-volume coordinates, we find

Y () 1
S = 1 2 Y ()E
+ i1 2 .
Y ()
C (2)
=

(4.25)
(4.26)
Finally, if we express 1 and in terms of the string parameters as in (4.23) and (2.51), and
use the definitions of the YM coupling constant and the -angle given in (4.7) and (4.10),
it is easy to see that
S=
8 2
iYM
2
gYM
(4.27)
which is the correct form of the instanton action! Clearly, to obtain the k-instanton action
we should consider k wrapped D1-branes in the background of N wrapped D5-branes,
which amounts simply to multiply the result (4.27) by k.
Therefore, we conclude that the gauge theory instantons of the N = 1 SYM theory
are indeed represented by D1-branes that wrap the same supersymmetric 2-cycle of the
D5-branes, as one should expect from general considerations.11 We also notice that the
above calculations can be performed in the N = 2 SYM theory using the supergravity
solution of Section 2.1 which leads to similar results.
5. Conclusions
We have shown that many interesting features of the pure N = 1 and N = 2 SYM
theories in four dimensions are quantitatively encoded in the supergravity solutions that
describe D5-branes wrapped on supersymmetric 2-cycles. These features comprise the
running of the gauge coupling constant, the -function, the chiral anomaly and instantons.
In the N = 1 case we have also discussed the gravitational dual of the gaugino condensate
and exploited it to obtain the complete NSVZ -function. Moreover we have found
the occurrence of non-perturbative corrections in the form of fractional instantons with
topological charge 2/N that smooth out the running of the coupling constant which never
11 The discrepancies with Ref. [28] originate from a different choice of normalization for the SU(N ) generators
as well as from a different definition of the YM coupling constant that for us is proportional to the volume of the
2-cycle computed with the ten-dimensional metric and not with the seven-dimensional domain-wall metric as in
Ref. [28].
65
diverges. The precise meaning of the fractional instanton corrections and their relevance
for the IR physics deserve, however, further study from a field theory point of view. Finally,
we observe that in the N = 2 case the supergravity analysis that we have presented allows
to obtain only the perturbative part of the -function and no instanton corrections. In
this respect it is worth pointing out that, differently from the N = 1 case, the N = 2
supergravity solution exhibits a naked singularity at small distances where it becomes
unphysical. It would very interesting to see whether it is possible to resolve this singularity
and eventually find the known non-perturbative instanton corrections of the N = 2 SYM
theory from a resolved supergravity solution. Another interesting development would be to
explore if the MN solution and the results we have found here can be used to quantitatively
investigate other features of the N = 1 SYM theory and if they can be extended also to
theories in which supersymmetry is completely broken.
Acknowledgements
We thank R. Auzzi, M. Bertolini, M. Bill, M. Frau, K. Konishi, E. Imeroni, L. Magnea,
R. Marotta and I. Pesando for useful discussions and exchange of ideas.
Appendix A. Parameterizations of the 3-sphere

There are several ways to describe a 3-sphere S 3 . Here we briefly describe the two
parameterizations we have used in Section 2 to up-lift the domain wall-solution from seven
to ten dimensions.
The defining equation for a unit 3-sphere is
21 + 22 + 23 + 24 = 1.
(A.1)
One explicit parameterization of the i s is the following

1 = cos cos 1 ,
2 = cos sin 1 ,
3 = sin cos 2 ,
4 = sin sin 2 ,
with 0
ds 2 =
4

(A.2)
and 0 1,2 2 . In this parameterization, the metric of the 3-sphere is

d2i = d 2 + cos2 d12 + sin2 d22 .
(A.3)
i=1
These are the coordinates that we used in Section 2.1 in order to up-lift to ten dimensions
the domain-wall solution in the N = 2 case.
A second convenient parameterization of the i s is in terms of the Euler angles, namely
+

cos
,
2
2

,
2 = sin cos
2
2
3 = cos
+

sin
,
2
2

1 = sin sin
,
2
2
4 = cos
(A.4)
66
with 0 , 0 2 and 0 4 . The corresponding left-invariant 1-forms

of S 3 are

1
1
1 = cos d + sin sin d ,
2 = sin d sin cos d ,
2
2

1
3 = d + cos d ,
(A.5)
2
which close the SU(2) differential algebra
d a = ! abc b c .
(A.6)
In this parameterization the metric of the unit 3-sphere is

ds 2 =
4

i=1
d2i =
3

2
a=1

2
1
d + cos d + sin2 d 2 + d 2 .
4
(A.7)
These are the coordinates that we used in Section 2.2 in order to up-lift to ten dimensions
the domain-wall solution in the N = 1 case.
By comparing (A.2) and (A.4), it is immediate to see that the two sets of angles are
related as follows
= 2,
= 1 2 ,
= 1 + 2 .
(A.8)
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, hep-th/9711200.
[2] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, N = 2 gauge theories on systems of fractional
D3/D7 branes, Nucl. Phys. B 621 (2002) 157, hep-th/0107057.
[3] S. Kachru, E. Silverstein, 4D conformal field theories and strings on orbifolds, Phys. Rev. Lett. 80 (1998)
4855, hep-th/9802183;
A. Lawrence, N. Nekrasov, C. Vafa, On conformal field theories in four dimensions, Nucl. Phys. B 533
(1998) 199, hep-th/9803015.
[4] I.R. Klebanov, E. Witten, Superconformal field theory on threebranes at a CalabiYau singularity, Nucl.
Phys. B 536 (1998) 199, hep-th/9807080.
[5] M.R. Douglas, G. Moore, D-branes, quivers and ALE instantons, hep-th/9603167;
I.R. Klebanov, N.A. Nekrasov, Gravity duals of fractional branes and logarithmic RG flow, Nucl. Phys.
B 574 (2000) 263, hep-th/9911096.
[6] M. Bertolini, P. Di Vecchia, R. Marotta, N = 2 four-dimensional gauge theories from fractional branes,
hep-th/0112195.
[7] C.P. Herzog, I.R. Klebanov, P. Ouyang, Remarks on the warped deformed conifold, hep-th/0108101;
C.P. Herzog, I.R. Klebanov, P. Ouyang, D-branes on the conifold and N = 1 gauge/gravity dualities, hepth/0205100.
[8] M. Bershadsky, C. Vafa, V. Sadov, D-branes and topological field theories, Nucl. Phys. B 463 (1996) 420,
hep-th/9511222.
[9] J. Maldacena, C. Nuez, Towards the large N limit of N = 1 super-YangMills, Phys. Rev. Lett. 86 (2001)
588, hep-th/0008001.
[10] J.P. Gauntlett, N. Kim, D. Martelli, D. Waldram, Wrapped fivebranes and the N = 2 super-YangMills
theory, Phys. Rev. D 64 (2001) 106008, hep-th/0106117.
67
[11] F. Bigazzi, A.L. Cotrone, A. Zaffaroni, N = 2 gauge theories from wrapped five-branes, Phys. Lett. B 519
(2001) 269, hep-th/0106160.
[12] C. Nunez, I.Y. Park, M. Schvellinger, T.A. Tran, Supergravity duals of gauge theories from F(4) gauged
supergravity in six dimensions, JHEP 0104 (2001) 025, hep-th/0103080;
J. Gomis, D-branes, holonomy and M-theory, Nucl. Phys. B 606 (2001) 3, hep-th/0103115;
J.D. Edelstein, C. Nuez, D6-branes and M-theory geometrical transitions from gauged supergravity,
JHEP 0104 (2001) 028, hep-th/0103167;
J. Gomis, T. Mateos, D6-branes wrapping Khler four-cycles, Phys. Lett. B 524 (2002) 170, hep-th/0108080;
J. Gomis, J.G. Russo, D = 2 + 1, N = 2 YangMills theory from wrapped branes, JHEP 0110 (2001) 028,
hep-th/0109177;
J.P. Gauntlett, N.W. Kim, D. Martelli, D. Waldram, Fivebranes wrapped on SLAG three-cycles and related
geometry, JHEP 0111 (2001) 018, hep-th/0110034;
K. Dasgupta, K. Oh, R. Tatar, Geometric transition, large N dualities and MQCD dynamics, Nucl. Phys.
B 610 (2001) 331, hep-th/0105066;
K. Dasgupta, K. Oh, R. Tatar, Open/closed string dualities and seiberg duality from geometric transitions in
M-theory, hep-th/0106040;
R. Hernandez, Branes wrapped on coassociative cycles, Phys. Lett. B 521 (2001) 371, hep-th/0106055;
K. Dasgupta, K. Oh, J. Park, R. Tatar, Geometric transition versus cascading solution, JHEP 0201 (2002)
031, hep-th/0110050;
O. Aharony, E. Schreiber, J. Sonnenschein, Stable non-supersymmetric supergravity solutions from
deformations of the MaldacenaNunez background, JHEP 0204 (2002) 011, hep-th/0201224;
N. Evans, M. Petrini, A. Zaffaroni, The gravity dual of softly broken N = 1 super-YangMills, JHEP 0206
(2002) 004, hep-th/0203203;
U. Gursoy, C. Nunez, M. Schvellinger, RG flows from Spin(7), CY 4-fold and HK manifolds to AdS,
Penrose limits and pp-waves, JHEP 0206 (2002) 015, hep-th/0203124.
[13] P. Di Vecchia, H. Enger, E. Imeroni, E. Lozano-Tellechea, Gauge theories from wrapped and fractional
branes, Nucl. Phys. B 631 (2002) 95, hep-th/0112126.
[14] R. Apreda, F. Bigazzi, A.L. Cotrone, M. Petrini, A. Zaffaroni, Some comments on N = 1 gauge theories
from wrapped branes, Phys. Lett. B 536 (2002) 161, hep-th/0112236.
[15] A. Klemm, W. Lerche, P. Mayr, C. Vafa, N.P. Warner, Self-dual strings and N = 2 supersymmetric field
theory, Nucl. Phys. B 477 (1996) 746, hep-th/9604034.
[16] G. Papadopoulos, A.A. Tseytlin, Complex geometry of conifolds and 5-brane wrapped on 2-sphere, Class.
Quantum Grav. 18 (2001) 1333, hep-th/0012034;
S.S. Gubser, A.A. Tseytlin, M.S. Volkov, Non-Abelian 4D black holes, wrapped 5-branes and their dual
descriptions, JHEP 0109 (2001) 017, hep-th/0108205.
[17] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, Fractional D-branes and their gauge
duals, JHEP 0102 (2001) 014, hep-th/0011077.
[18] J. Polchinski, N = 2 gauge-gravity duals, Int. J. Mod. Phys. A 16 (2001) 707, hep-th/0011193.
[19] M. Bill, L. Gallot, A. Liccardo, Classical geometry and gauge duals for fractional branes on ALE spaces,
Nucl. Phys. B 614 (2001) 254, hep-th/0105258.
[20] M. Bertolini, P. Di Vecchia, G. Ferretti, R. Marotta, Fractional branes and N = 1 gauge theories, Nucl. Phys.
B 630 (2002) 222, hep-th/0112187.
[21] M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, More anomalies from fractional branes, hepth/0202195, Phys. Lett. B, submitted for publication.
[22] A.W. Peet, J. Polchinski, UV/IR relations in AdS dynamics, Phys. Rev. D 59 (1999) 065011, hepth/9809022.
[23] V. Novikov, M. Shifman, A. Vainstein, V. Zakharov, Exact Gell-Mann-low function of supersymmetric
YangMills theories from instanton calculus, Nucl. Phys. B 229 (1983) 381;
V. Novikov, M. Shifman, A. Vainstein, V. Zakharov, Beta-function in supersymmetric gauge theories:
instantons versus traditional approach, Phys. Lett. B 166 (1986) 329;
M.A. Shifman, A.I. Vainshtein, Solution of the anomaly puzzle in susy gauge theories and the Wilson
operator expansion, Nucl. Phys. B 277 (1986) 456;
M.A. Shifman, A.I. Vainshtein, Sov. Phys. JETP 64 (1986) 428.
68
[24] M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, QCD and resonance physics: applications, Nucl. Phys. B 147
(1979) 448.
[25] C.V. Johnson, A.W. Peet, J. Polchinski, Gauge theory and the excision of repulson singularities, Phys. Rev.
D 61 (2000) 086001, hep-th/9911161.
[26] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and SBresolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[27] E. Imeroni, On the N = 1 -function from the conifold, hep-th/0205216.
[28] A. Loewy, J. Sonnenschein, On the holographic duals of N = 1 gauge theories, JHEP 0108 (2001) 007,
hep-th/0103163.
[29] J. Maldacena, C. Nuez, Supergravity description of field theories on curved manifolds and a no go theorem,
Int. J. Mod. Phys. A 16 (2001) 822, hep-th/0007018.
[30] A. Salam, E. Sezgin, SO(4) gauging of N = 2 supergravity in seven dimensions, Phys. Lett. B 126 (1983)
295.
[31] M. Cvetic, H. Lu, C.N. Pope, Consistent KaluzaKlein sphere reduction, Phys. Rev. D 62 (2000) 064028,
hep-th/0003286.
[32] V.L. Campos, G. Ferretti, H. Larsson, D. Martelli, B.E. Nilsson, A study of holographic renormalization
group flows in d = 6 and d = 3, JHEP 0006 (2000) 023, hep-th/0003151.
[33] A.H. Chamseddine, M.S. Volkov, Non-Abelian BPS monopoles in N = 4 gauged supergravity, Phys. Rev.
Lett. 79 (1997) 3343, hep-th/9707176;
A.H. Chamseddine, M.S. Volkov, Non-Abelian solitons in N = 4 gauged supergravity and leading order
string theory, Phys. Rev. D 57 (1998) 6242, hep-th/9711181.
[34] C.V. Johnson, D-brane primer, hep-th/0007170.
[35] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Classical p-branes from boundary states,
Nucl. Phys. B 507 (1997) 259, hep-th/9707068;
P. Di Vecchia, M. Frau, A. Lerda, A. Liccardo, (F, Dp)-bound states from the boundary state, Nucl. Phys.
B 565 (2000) 397, hep-th/9906214.
[36] P. Merlatti, G. Sabella, World volume action for fractional branes, Nucl. Phys. B 602 (2001) 453, hepth/0012193.
[37] I.R. Klebanov, P. Ouyang, E. Witten, A gravity dual of the chiral anomaly, Phys. Rev. D 65 (2002) 105007,
hep-th/0202056.
[38] For a recent review, for example, see: T.J. Hollowood, V.V. Khoze, W.J. Lee, M.P. Mattis, Breakdown of
cluster decomposition in instanton calculations of the gluino condensate, Nucl. Phys. B 570 (2000) 241,
hep-th/9904116.
[39] N.M. Davies, T.J. Hollowood, V.V. Khoze, M.P. Mattis, Gluino condensate and magnetic monopoles in
supersymmetric gluodynamics, Nucl. Phys. B 559 (1999) 123, hep-th/9905015.
[40] M.R. Douglas, Branes within branes, hep-th/9512077;
M.R. Douglas, Gauge fields and D-branes, J. Geom. Phys. 28 (1998) 255, hep-th/9604198.

On operator mixing in N = 4 SYM

Massimo Bianchi, Burkhard Eden, Giancarlo Rossi,
Yassen S. Stanev 1
Dipartimento di Fisica, Universit di Roma Tor Vergata, INFNSezione di Roma Tor Vergata,
Via della Ricerca Scientifica, 1 00133 Roma, Italy
Received 25 June 2002; received in revised form 1 August 2002; accepted 6 September 2002
Abstract
We resolve the mixing of the scalar operators of naive dimension 4 belonging to the representation
20 of the SU(4) R-symmetry in N = 4 SYM. We compute the order g 2 corrections to their
anomalous dimensions and show the absence of instantonic contributions thereof. Ratios of the
resulting expressions are irrational numbers, even in the large N limit where, however, we observe
the expected decoupling of double-trace operators from single-trace ones. We briefly comment on the
generalizations of our results required in order to make contact with the double scaling limit of the
theory conjectured to be holographically dual to type IIB superstring on a pp-wave.
2002 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Pg; 11.10.Gh; 11.30.Pb; 11.25.Hf
1. Introduction and summary

The holographic correspondence between superstring theory in anti-de Sitter (AdS)
spaces and superconformal theories (CFT) [13] has renewed the interest in N = 4
supersymmetric YangMills theory (SYM) and triggered the discovery of new unexpected
properties of its superconformal phase. An overview may be found in the review
articles [47], which also contain extensive lists of references.
Until very recently, most of the available results on the superstring side were confined
to the low energy (weak curvature L2 ) approximation where supergravity takes
over. This limit corresponds to the strong t Hooft coupling (g 2 N 1) regime of
E-mail address: massimo.bianchi@roma2.infn.it (M. Bianchi).
1 On leave of absence from Institute for Nuclear Research and Nuclear Energy, Bulgarian Academy of
Sciences, BG-1784, Sofia, Bulgaria.

0550-3213/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 1 7 - 9
70
M. Bianchi et al. / Nuclear Physics B 646 (2002) 69101
the dual gauge theory which is obviously inaccessible by perturbative means. Barring
few important exceptions, perturbative and non-perturbative tests were thus restricted
to protected quantities, i.e., observables which are actually independent of the coupling
constant. By now there is quite a long list of such protected quantities that includes
dimensions of operators belonging to short multiplets [8], certain OPE coefficients of chiral
primary operators (CPOs) [9], extremal [10] and next-to-extremal [11] correlators.
The first truly dynamical test of the correspondence emerged from the remarkable
agreement between SYM instanton effects and D-instanton corrections to higher derivative
terms in the type IIB superstring effective action [12]. Bonus symmetry [13] of up to
four-point functions of protected operators was another suggestive hint to the underlying
type IIB string description of N = 4 SYM. Another class of observables that should
clearly display stringy behaviour are MalcadenaWilson loops [14]. So far non-trivial
string predictions [15] have only received partial support from perturbation theory [16,
17] and seem to require a deeper understanding of D-instanton effects in order to
accommodate SYM instanton corrections [18]. Among the other achievements of N = 4
super-instanton calculus the two-line proof [7] of the partial non-renormalization of
the four-point function of operators in the supercurrent multiplet [19] stands out for its
simplicity. However, partial non-renormalization is essentially a consequence of SU(2, 2|4)
superconformal symmetry that severely constrains the dynamics, though it does not
completely trivialize it [19,20].
The main purpose of this paper is to resolve the mixing of the scalar primary operators
of naive dimension 0 = 4 in the (real) representation 20 of the SU(4) R-symmetry group
and compute their anomalous dimensions at order g 2 .
Relying on previous results on four-point functions of lowest CPOs Q with [21] or
without [2224] insertion of the lowest Konishi scalar K, we disentangle the mixing among
the scalar operators in the 20 representation. The vanishing of instanton contributions to
the relevant four-point functions [21] implies the absence of non-perturbative corrections
to the mixing coefficients and anomalous dimensions that we compute. Our analysis is
further simplified by the observation that operators which belong to the Konishi multiplet
and whose leading terms are generalized Yukawa couplings, decouple at the order at
which we work. The remaining operators show an intricate pattern of mixing at finite N
that simplifies significantly as N . In this analysis we exploit the vanishing of the
that appears in OPE of two Qs. This property
anomalous dimension of the operator D20
may be viewed as resulting from a generalized shortening condition of the linear type
that survives interaction as a consequence of certain differential constraints satisfied by
three point functions involving two protected operators [25,26].2 Our results for the oneloop anomalous dimensions definitely exclude the possibility, envisaged in [29] and not
disproved so far by any explicit computation, that ratios of anomalous dimensions be
rational even in the large N limit. This is not in conflict with any basic principle but
rather suggests that the theory behaves in a highly non-trivial fashion. Still, it is reassuring
to find that multi-trace operators that are dual to multi-particle bound states have in the
2 It should be kept in mind, however, that there are operators that satisfy the same shortening condition, i.e.,
saturate the same unitarity bound at tree level, but violate it after inclusion of radiative corrections [27,28].
71
limit N with g 2 N fixed anomalous dimensions that are given by the sums of the
anomalous dimensions of their constituents. This suggests that the dual bound states are
at threshold. Moreover, as observed in [21], the absence of non-perturbative instanton
corrections for the anomalous dimensions and OPE coefficients of the operators which we
study is in line with S-duality that maps operators dual to string excitations into operators
dual to dyonic string excitations.
The pattern of intricate mixings and irrational anomalous dimensions that we find, may
not necessarily prove to be an insurmountable obstacle towards the extrapolation of the
string spectrum and interactions from low energy (strong coupling) to large curvature
(weak coupling) at least in the Penrose limit of AdS5 S 5 which gives rise to a maximally
supersymmetric pp-wave [30,31]. String loop corrections, which seem to be calculable in
the pp-wave background [32], may play a crucial role in quantitatively establishing this
correspondence [33,34]. We will argue that our results can be generalized to yield further
insight into the properties of the set of operators dual to the low-lying string excitations.
The plan of the paper is as follows: after recalling some basic definitions and establishing our notation in Sections 2, 3 we briefly describe unitary irreducible representations
(UIRs) of the superconformal group SU(2, 2|4) and discuss the emergence of multiplet
shortening. In Section 4 we identify the scalar composite operators of naive dimension
0 = 4 belonging to the representation 20 . In Section 5 we perform to order g 2 the orthogonalization of the two-point functions and compute the anomalous dimensions of these
operators. In Sections 6 and 7 we follow a different route to the same results that requires
the computation of the four-point function of two Qs and two Ks at order g 4 , thus extending similar results previously obtained at order g 2 [21]. In Section 8 we briefly comment
on possible generalizations of our results to the double scaling limit of the theory which
is conjectured to be holographically dual to type IIB superstring on a pp-wave [30]. In
Appendix A we gather unwieldy formulae.
2. Notation and conventions

In this section, we summarize our notations and conventions and we recall some
relevant results of [21] concerning the structure and the renormalization properties of the
unprotected Konishi supermultiplet.
The field content of N = 4 SYM [35] comprises a vector, A , four Weyl spinors, A
(A = 1, 2, 3, 4), and six real scalars, i (i = 1, 2, . . . , 6), all in the adjoint representation
of the gauge group, that we take to be SU(N) for definiteness. In the N = 1 approach that
we shall follow the fundamental fields can be arranged into a vector superfield, V , and
three chiral superfields, I (I = 1, 2,3). The six real scalars, i , are
combined into three
complex fields, I = ( I + i I +3 )/ 2 and I = ( I i I +3 )/ 2 that are the lowest
components of the chiral and antichiral superfields, I and I , respectively. Three of the
Weyl fermions, I , are the spinors of the chiral multiplets. The fourth spinor, = 4 ,
together with the vector, A , form the vector multiplet. In this way only an SU(3) U (1)
subgroup of the full SU(4) R-symmetry is manifest.
The complete N = 4 SYM action in the N = 1 superfield formulation has a nonpolynomial form, as we do not work in the WessZumino gauge. A gauge fixing term
72
must anyway be added to the classical action. We shall use the FermiFeynman gauge, as it
makes corrections to the propagators of the fundamental superfields vanish at order g 2 [36
38]. Actually a stronger result has been proved in these papers, namely, the vanishing of the
anomalous dimensions of the fundamental fields up to O(g 4 ). With the FermiFeynman
gauge choice the terms relevant for the calculation of the Green functions we are interested
in are

S = d 4 x d 2 d 2

a
a
V a Va Ia aI 2igfabc I V b I c + 2g 2 fabe fecd I V b V c I d

ig 2 abc
(2)
f
I J K aI bJ cK (2) ( ) I J K aI
bJ
cK
( ) + , (1)
3!
where fabc are the structure constants of the gauge group. As neither the cubic and quartic
vector interactions nor the ghost terms will contribute to the calculations we will present
in this paper, we have omitted them in Eq. (1).
Since all superfields are massless, their propagators have an equally simple form in
momentum and in coordinate space and thus we choose to work in the latter which is more
suitable for the study of conformal field theories. In Euclidean coordinate space one finds

I J ab ( + 2 ) 1
e ii jj ji j 2 ,
Ia xi , i , i bJ xj , j , j =
4 2
xij
(2)

ab (2) (ij ) (2) (ij )
,
Va xi , i , i Vb xj , j , j = 2
8
xij2
(3)
where xij = xi xj , ij = i j , ij = i j .
The simplest protected (dimension two) CPOs

ij k k
(ij )
i j
Q20 = tr
,
6
(4)
belong to the representation 20 of SU(4) and are the lowest component of the N = 4
supercurrent multiplet.
(ij )
In terms of SU(3) U (1) the Q20 s decompose in

C I J (x) = tr I (x) J (x) ,
(5)
CIJ (x) = tr I (x)J (x)
and

I

VJI = tr e2gc(x)J (x)e2gc(x) I (x) J tr e2gc(x)L (x)e2gc(x) L (x) ,
(6)
3
where the exponentials (c(x) is the lowest component of the vector superfield) are included
to ensure gauge invariance and regularization of the operators is understood, e.g., by pointsplitting (see below). Note that no normal-ordering is needed because the VEVs of all the
above operators vanish, none of them being an SU(4) singlet.
73
The N = 4 Konishi multiplet is a long multiplet of SU(2, 2|4) [39]. Its lowest
component, K1 , is a scalar operator of (naive) conformal dimension 0 = 2, which is a
singlet of the SU(4) R-symmetry group. The (naive) definition of K1 is

1 i
:tr (x) i (x) : ,
2
6
K1 (x)|naive =
(7)
i=1
where the trace is over colour indices and the symbol : : stands for normal ordering. As
usual, normal ordering means subtracting the operator VEV or, in other words, requiring
K1 = 0. In terms of N = 1 superfields K1 can be written in the form
3

K1 (x)|formal =
:tr e2gc(x)I (x)e2gc(x) I (x) : .
(8)
I =1
Since the operator K1 has an anomalous dimension, K (g 2 ), it will suffer a non-trivial

renormalization. We assume (as is always the case in perturbation theory) that K (g 2 ) is
small and represented by the series expansion

K g 2 = g 2 1K + g 4 2K + .
(9)
From the results of Refs. [40] and [23,24], one gets for the first two coefficients of the
expansion
3N
(10)
,
4 2
3N 2
2K =
(11)
.
16 4
It is convenient to regularize operators by point splitting. In particular for K1 we write
1K =
K1 (x)|reg = a
K 2

3

. 2gc(x) I
.
2gc(x)
:tr e
I x +
x
e
:,
2
2
(12)
I =1
where . is an infinitesimal, but otherwise arbitrary, four-vector. Note that, due to our choice
of gauge-fixing, there is no need to point-split the vector field in the exponents, because
the c-field has vanishing propagator. Finally, the renormalized operator has the form
K1 (x)|ren = lim
.0
a K (g 2 )
1
(. 2 ) 2
K (g 2 )

3

. 2gc(x) I
.
:tr e2gc(x)I x +
x
e
:,
2
2
I =1
(13)
where a K (g 2 ) is a normalization factor that we choose of the form a K (g 2 ) = 1 +

g 2 a1K + g 4 a2K + . Unlike the operators corresponding to symmetry generators (like the
R-symmetry currents or the stress-energy tensor), the Konishi scalar K1 has no intrinsic
normalization, so we shall use this freedom in the normalization factor a K (g 2 ) of Eq. (13),
to make the two-point function of K1 depend on g 2 only through K . This is achieved at
the order we work by setting a1K = 3N/8 2 (other coefficients would require higher order
74
computations to be fixed). With this choice one gets

3(N 2 1)
1
.
K1 (x1 )K1 (x2 ) =
4(4 2 )2 (x12 2 )2+ K (g 2 )
(14)
3. Comments on the UIRs of SU(2, 2|4)

The unitary irreducible representations (UIRs) of SU(2, 2|4) have been classified
in [41].3 A general UIR is specified by a set of quantum numbers comprising the
dilation weight , the Lorentz spins (j1 , j2 ) and the Dynkin labels [k, l, m] of the
SU(4) R-symmetry. There are three unitary series, which are distinguished by different
relations between the dilation weight and the other quantum numbers.
It has been known for some time that generic UIRs can be obtained by tensoring the
so-called singleton representations [42].4 In [44] this has been elaborated in full detail
using the technique of harmonic superspace [45]. Within this approach, in addition to the
usual N = 4 super-Minkowski space one introduces 4 4 matrices, uA
r , parameterizing
the coset SU(4)/U (1)3 . We omit the details of the construction, but rather remark that
these matrices should be contracted on all free SU(4) indices in the constraints defining
the Grassmann analytic superfields [46]
W [1k] ,
1k3
(15)
which read
DA W [1k] = 0, 1 A k,
A
W [1k] = 0,
D
k + 1 A 4.
(16)
r
In Eq. (16) DA = uA
r D are projected N = 4 supercovariant derivatives. The constraints
express the fact that the W superfields depend on half of the spinor coordinates, i.e., they
are 1/2 BPS objects.
Besides the W s, the list of singletons additionally includes N = 4 chiral superfields,
whichunlike the formermay have either left or right handed spinor indices, but cannot
carry a non-trivial SU(4) representation. For the present purpose we only need to introduce
the scalar chiral superfield, , which satisfies
A
= 0,
D
D (A DB) = 0,
(17)
where the second (linear) constraint is a sort of field equation. In the tensoring
procedure it is assumed that the chiral superfield is on shell.
3 In (perturbative) N = 4 SYM-theory only UIRs of PSU(2, 2|4) are actually relevant. They are characterized
by the vanishing of the U (1)C central charge that extends PSU(2, 2|4) to SU(2, 2|4).
4 Alternatively the UIRs may be built by using the oscillator method (see [43] and references therein) or they
can be realized as analytic tensor fields on analytic superspace [26].
75
The statement that any UIR can be obtained as a product of singletons is formal in that
none of the multiplets W 1 , W [123] and can be expressed in terms of elementary fields.
On the contrary, W [12] is the fundamental N = 4 SYM multiplet and its square is the
supercurrent multiplet.
An important observation is that the component field content of the Konishi multiplet
in the non-interacting theory is correctly reproduced by the product of a field
satisfying (17) with its complex conjugate:
.
K1 |g 0 =
(18)
Using (17), one may verify that5

K1 0 = 0.
(A
D (A DB) K1 g 0 = 0,
D
D
B)
g
(19)
(A
on K1 |g 0 and performing some
By acting with the products of D (A D and D
D
B)
D-algebra, one realizes that both the singlet and the 15 components of the current
A

A

B
D , D
=
KB
(20)
K1 g 0 ,,=0
B)
are conserved. There are similar constraints on some of the higher components. Also, each
of the linear constraints (19) separately implies the absence of some component fields in
the supermultiplet.
In the interacting theory the Konishi multiplet has an anomalous dimension, so we may
formally write

(1+ ) .
K1 =
(21)
For = 0 it does not satisfy any differential constraint, consistently with it being a long
multiplet.
Next, we focus on operators of naive dimension 4 in the 20 . Operators in this
representation can be either single or double trace composites of the fields in the
fundamental N = 4 SYM multiplet. Independently of their actual expression, the
formalism of [44] allows to determine the component field content multiplets of this kind
by representing it in the form

(1+ ) W [12] 2 .
O20 =
(22)
Eq. (22) shows that it has the same field content as a product of the Konishi and
supercurrent multiplets.
If and only if = 0, the operator satisfies differential constraints which are in the
intersection of the conditions (16) and (19), i.e.,
D A DB O20 = 0,
A
B O20 = 0,
D
D
1 A, B 2,
3 A, B 4.
(23)
5 In the interacting theory the situation is more complicate, as the r.h.s. of in Eq. (19) do not vanish [28]. For
B)
CD ). These considerations are at the basis of the
instance, the first becomes D (A D K1 g tr([W AC , W BD ]W
derivation of the Konishi anomaly [47].
76
3 , D
4 mutually anticommute, so that we can never
Note that the derivatives D 1 , D 2 , D
derive a condition containing a spacetime derivative from them. Hence, even in the
non-interacting case, such multiplets do not contain conserved tensor currents. The
constraints (23) merely express the absence of some component fields.
If the conditions (23) happen to be enforced by some mechanism, e.g., because the
field occurs in the OPE of two supercurrent multiplets [25], then the operator will have
protected dimension. This is exactly what happens for the double-trace operator D20 . Its
explicit expression, obtained in [21], is given for convenience in Eq. (31). It may be argued
that the reason why this operator is protected is to be ascribed to the fact that it obeys
constraints not involving SYM covariant superderivatives [28].
4. Scalar operators in the 20 representation
For a sufficiently large number of colours, N 4, there are 6 distinct scalar primary
operators of naive scale dimension 0 = 4 in the real representation 20 of SU(4). Two
of them K20
belong to the Konishi multiplet. At leading order in g they correspond to
generalized Yukawa couplings

ij
(i
K20 = tAB Tr j A , B + ,
(24)
which are bilinear in the fermions. Hence they do not contribute at tree level and, in general,
2 )n , in the correlation functions that we shall analyze.
to the leading logarithms, g 2n (log x12
The other four operators are quartic in the fundamental scalars (at leading order in g) and
naively can be written in the form6
3

I
O911 (x)naive =
:a1 (x)b1 (x)
(x) + dI (x)I,c
(x) :X9abcd ,
c (x)I,d
a,b,c,d
(25)
I =1
where the index 9 = 1, 2, 3, 4 labels the four different operators characterized by the four
colour tensors X9abcd , given below (Eqs. (30) to (35)).
To properly define renormalized operators one has to make three modifications to the
naive formula (25). First, in order to ensure gauge invariance the fundamental scalars have
to be replaced by
I (x) egc(x)I (x)egc(x),
I (x) egc(x) I (x)egc(x),
(26)
respectively. Second, one has to regularize the operator. We choose to do this by point
splitting, separating the arguments of the four scalars in (25) by a small distance. There are
several, essentially equivalent, ways to do this, the most compact one is to position the four
scalars at the vertices of a tetrahedron so that all the separations have equal length which
6 We shall omit the 20 label from now on and choose to deal with one of the 20 components of the
multiplet that we take to be the one belonging to the representation 6+2 of SU(3) U (1) in the decomposition
20 80 6+2 62 , namely, O 11 .
77
we shall denote by .. As remarked after Eq. (12), there is no need to point split the different
gauge exponents in (26). Still one has to separate their arguments from the arguments of
the fundamental scalars. We decided to put them all at the center of the tetrahedron. Finally,
as in the case of K, operators that acquire anomalous dimension have to be renormalized.
Within the regularization prescription we chose this amounts to writing
O911 (x)|ren = lim
.0
a 9 (g 2 )
(. 2 )
1 O9 2
(g )
2

O911 (x)reg + ,
(27)
where dots in the r.h.s. denote possible subleading (in g) mixings which will not contribute
at the order we work. Here O9 (g 2 ) is the anomalous dimension of the operator O9 for
which as before we assume the power series expansion

O9 g 2 = g 2 1O9 + g 4 2O9 + ,
(28)
while a 9 (g 2 ) is a finite renormalization of the form

a 9 g 2 = 1 + g 2 a19 + g 4 a29 + ,
(29)
which depends on the regularization prescription.

Let us now list the four (for N 4) colour tensors X9abcd . The operator, D,
corresponding to the tensor
1
1
abcd
= ac bd ab cd .
XD
(30)
2
6
is double trace and, in line with the discussion at the end of Section 3, turns out to be finite,
i.e., it does not need any infinite renormalization, and as such protected. In fact, anticipating
the results of the following sections, D does not even need any finite redefinition as it can
be expressed in terms of only the protected = 2 CPOs Q(ij ) of Eq. (4) as
D(ij ) =
Q(ik) Q(j k)
ij (k9) (k9)
Q Q .
6
(31)
k,9
We have to choose an appropriate basis for the remaining three operators, which we
denote by M, P and L. This choice is at this point purely conventional, since, as we
shall demonstrate, the correct operators, i.e., those that have well defined (anomalous)
dimensions, will turn out to be complicated linear combinations of the former. We denote
by M the double trace operator corresponding to the colour tensor
1
abcd
= ab cd .
XM
2
(32)
Note that M is the product of the lowest chiral primary operator Q(ij ) of Eq. (4) and the
lowest operator in the Konishi supermultiplet, K1 , given in Eq. (7), i.e.,
M(ij ) = Q(ij ) K1 .
We denote by P the single trace operator corresponding to
1
abcd
XP
=
face fbde .
2 e
(33)
(34)
78
Note that for SU(2) P is a linear combination of D and M. The fourth operator, L, is
constructed by saturating colour indices with the (traceless) quartic Casimir operator of
SU(N)

abcd
= Tr T a T b T c T d + permutations of b, c, d
XL

1 2N 2 3 ab cd
+ ac bd + ad bc ,
2
2 N(N + 1)
(35)
where T a are the generators of SU(N) in the fundamental representation. XL vanishes

both for SU(2) and SU(3).
From this general discussion it is clear that one has to treat separately the two low rank
cases of SU(2) and SU(3), since the number of independent operator varies with N .
5. Orthogonalization of the scalar operators

We shall consider only tree-level and order g 2 constraints, since higher loop quantities
will depend also on terms omitted in Eq. (27). The correctly renormalized operators
(i.e., those having well defined anomalous dimensions) must satisfy the following three
orthogonality properties.
(1) They have to be orthogonal to (i.e., they must have vanishing two-point functions
with) the two operators belonging to the Konishi multiplet, K20

. This is automatic both
2
at tree level and for the leading logarithms at order g , because K20
are bilinear in the
fermions at leading order.
(2) They have to be orthogonal to the operator D defined in (31) that turns out to be
protected at the order at which we work. An explicit calculation shows that both at tree
level and at order g 2 the operator L already enjoys this property, while for the other two
one needs to introduce the definitions
6
= M
M
(36)
D,
2
3N 2
= P + 2 N M N D.
P
(37)
3 N2 2
N2 2
and P
is orthogonal (in the sense explained above)
Indeed any linear combination of M
to D. Our choice is such that they are mutually orthogonal at tree level for any N and the
vanishes for SU(2). Thus for SU(2) the only relevant operator is M.
From the
operator P
presence of logarithmic terms at short distances in the expansion of the four-point function
has a non-vanishing one-loop

(x2 )K1 (x3 )K1 (x4 ) at order g 2 it follows that M
C 11 (x1 )C11
anomalous dimension. The precise value was computed in [21] with the result
Ng 2
(38)
, N = 2.
4 2
(3) Finally they have to be mutually orthogonal both at tree level and at one loop. In
other words we have to define new operators

1M = 5
+ 1P P
+ 1L L,
O1 = 1M M
+ 2P P
+ 2L L,
O2 = 2M M
+ 3P P
+ 3L L
O3 = 3M M
79
(39)
and require

Oi (x)Oj (y) = 0 if i = j
(40)
g2 .
both at tree level and at order

It follows from (27) that the order g 2 correction to the anomalous dimension of the
operator Oi is equal to the ratio of the coefficient of the ln(x y)2 term at order g 2 and
the tree level normalization of the two-point function of Oi , i.e.,
1Oi =
Oi (x)Oi (y)|g 2 ,ln(xy)2

Oi (x)Oi (y)|0
(41)
Hence it suffices to solve the orthogonality relations (40) to get the explicit expressions of
the Oi s anomalous dimensions.
Before proceeding to this rather long calculation let us make two comments. First, we
notice that the order g 2 contributions to these two-point functions come from two different
types of diagrams, corresponding to either chiral or vector internal lines. Both types of
diagrams can be expressed in terms of the massless box integral (see Appendix A). At any
order in g 2 each orthogonality relation (40) leads to one equation for the coefficients
in (39). These equations ensure the cancellation of all ln(.)-singularities. At this point, by
a suitable choice of the coefficients a19 in (29) one can also cancel the finite corrections at
the same order. Second, since the overall normalization of the operators Oi is arbitrary, 3
out of the 9 parameters in (39) can be chosen without loss of generality to be equal to 1.
Thus the orthogonality constraint leads to 6 quadratic equations for 6 variables. A careful
analysis shows that up to permutation symmetry there is only one acceptable (i.e., real)
solution.
The two-point function of the operator D is given by its tree-level value

10 2

N 1 3N 2 2 I0 (x12 ),
DD|0 = D(x1 )D(x2 ) =
9
where
1
.
I0 (x12 ) =
2 )4
2
4
(4 ) (x12
(42)
(43)
The absence of one-loop corrections either finite or infinite indicates that D defined in (31)
is indeed a protected operator. For the other operators at tree level we find

2
2
2

M
= 18(N 1)(N 2)(N + 1) I0 (x12 ),
M
0
2
(3N 2)
2
2
2

P
= 3N (N 1)(N 4) I0 (x12),
P
0
2
(N 2)
2

15(N 1)(N 2 4)(N 2 9)
LL 0 =
I0 (x12),
(N 2 + 1)
while the off-diagonal functions vanish by construction.
(44)
(45)
(46)
80
At order g 2 we obtain

M
2 = 18N N 2 1 N 2 + 1 I(x12 ),
M
g

2
2
2

M
2 = 18N (N 1)(N 4) I(x12 ),
P
g
2
(N 2)

LM g 2 = 0,
3
2
2
2

P
2 = 2N (N 1)(N 4)(5N 16) I(x12 ),
P
g
(N 2 2)2
2
2
2

2 = 5N(N 1)(N 4)(N 9) I(x12 ),
LP
g
(N 2 + 1)
2

25N(N 1)(N 2 4)(N 2 9)
I(x12 ),
LL g 2 =
(N 2 + 1)
(47)
(48)
(49)
(50)
(51)
(52)
where I(x12 ) is proportional to the short distance limit of the box integral. Using Eq. (A.6)
in Appendix A, one finds
I(x12 ) =
2

1
ln
x
+
1
.
12
2 )4
(4)5 (x12
(53)
Let us start by analyzing the generic case N 4. The cases N = 3 and N = 2 are much
simpler and we shall discuss them in Section 7. Since, as we said, all coefficients in (39)
turn out to be non-vanishing, without loss of generality we shall set i P = 1. With this
choice the orthogonality relations lead to the equation
i M =
i
,
(N 2 + 1)(N 2 2)
(54)
where i are the three roots of the cubic equation

8N 3 + N 6 + 2N 4 + 68N 2 40 2

+ 3N 9 16N 7 + 132N 5 80N 3 84N 10 + 64N 4 + 244N 8
224N 6 + 9N 12 = 0.
(55)
For the coefficients i L we get

i L =
24 2
(N 2 + 1)3
i M
5
(3N 2 2)(N 2 4)(N 2 9)
3
(N 2 2N + 2)(N 2 + 2N + 2)(N 4 19N 2 + 10)(N 2 + 1)
+ i M
5
N(N 2 2)(3N 2 2)(N 2 4)(N 2 9)
+
4 (N 2 + 1)2 N 2
.
5 (N 2 2)2 (N 2 9)
(56)
Inserting all these formulae into Eq. (41), we obtain for the order g 2 corrections to the
anomalous dimensions of the operators Oi the following expressions
81
Fig. 1. The ratios 1 i /1K , i = 1, 2, 3, as functions of N . 1K is the order g 2 coefficient of the Konishi multiplet
anomalous dimension, displayed in Eq. (10).
g2 N
4 2
8 2 N (N 6 2N 4 68N 2 + 40)i + 6N 9 64N 7 + 184N 5 96N 3
,
i
N 3 (N 2 4)(3N 2 2)(N 2 2)
(57)
where again the i s are the three roots of the cubic equation (55).
In Fig. 1 we plot the ratios 1Oi /1K , i = 1, 2, 3, as functions of N , where 1K is the
order g 2 coefficient of the Konishi multiplet anomalous dimension, displayed in Eq. (10).
1Oi =
5.1. The N limit

We would like to end this section by briefly discussing the peculiar properties of the
above orthogonalization procedure in the N limit. To this purpose we notice that
(1) multiple trace operators with different number of traces are mutually orthogonal in
this limit;
(2) in the space of multiple n-trace operators one can choose a basis of product
operators of the form

j ,
O
(58)
j =1n
j s are mutually orthogonal single trace operators of well defined anomalous

where the O
j ).
dimensions, (O
82
Note that both double trace operators defined above, i.e., D of Eq. (31) and M of
Eq. (33), have the form (58) (the SU(4) trace subtraction in D does not affect the following
argument). Since for large N the two-point function of a multiple trace product operator
is dominated by the most disconnected part (i.e., by the products of the two point functions
of the constituent operators), different product operators will be orthogonal for large N
j s. Moreover, it follows that in the N limit the
due to the orthogonality of the O
anomalous dimension of the product operator is the sum of the anomalous dimensions
of its constituent operators, namely,

j .
lim
(59)
Oj =
lim O
N
j =1n
j =1n
Let us now return to the case at hand. From Fig. 1 one can see that for large N the
anomalous dimension of one of the three operators tends to the anomalous dimension
of the Konishi supermultiplet, 1K . An explicit calculation confirms that this operator is
indeed M (see Eq. (33)). The anomalous dimensions of the other two operators, which in
line with the above discussion for large N will be dominantly single trace ones, tend to
lim =

1
5 5 1K .
6
The corresponding colour tensors Xabcd entering Eq. (25) are in the N limit

a b c d

abcd
X =
5 1 Tr T T T T
5 1 Tr T a T c T b T d .
(60)
(61)
Despite the appearance of the golden ratio7 the resulting expressions are not rational
numbers in units of the anomalous dimension of the Konishi multiplet. This kills any
residual hope that N = 4 SYM could behave as some sort of rational conformal field
theory at least in the large N limit. Barring S-duality miracles it is very difficult to imagine
that one could ever find the complete exact spectrum of dimensions of scaling operators
in this theory.
6. The four point function C 11 C11

K1 K1 at order g 4
Given the complexity of the results we got in the previous section, it is important to
rederive them from an alternative point of view, i.e., by performing an OPE analysis of
appropriate four-point functions.
Since the only operator of dimension 4 in the 20 that appears in the OPE of two CPOs Q
is the protected operator D20 (see Eqs. (25) and (30)) [29], little mileage can be gained by
studying only the correlation function of four Qs. The next simplest four-point functions
involve the lowest component of the Konishi multiplet. Correlation functions involving
two Konishi operators like QQKK meet all the necessary requirements. This kind of
7 We would like to thank I. Todorov for pointing this out to one of us (M.B.).
83
correlators have been calculated at lowest order in g 2 in [21], where it was also shown that
they receive no instanton corrections.
Since there is only one SU(4) tensor structure, without loss of generality we can use the
following representative for this correlator

(x2 )K1 (x3 )K1 (x4 ) ,

G(x1 , x2 , x3 , x4 ) = C 11 (x1 )C11
(62)
which with the help of the identity (no summation on J )

1
:tr J (x) J (x) : = VJJ (x) + K1 (x)
3
can be written also as
G(x1 , x2 , x3 , x4 )

(x2 ):tr 2 (x3 ) 2 (x3 ) : :tr 3 (x4 ) 3 (x4 ) :

= 9 C 11(x1 )C11

(x2 )K1 (x3 )V33 (x4 ) 3 C 11 (x1 )C11

(x2 )V22 (x3 )K1 (x4 )
3 C 11(x1 )C11

9 C 11(x1 )C11
(x2 )V22 (x3 )V33 (x4 ) + C 11 (x1 )C11
(x2 ) K1 (x3 )K1 (x4 ) .
(63)
(64)
For the purposes of our analysis only the first (connected) term in the r.h.s. of Eq. (64) has
2 )
really to be computed at order g 4 . In fact in the OPE limit x13 0, the leading ln2 (x13

contributions in the 20 channel that comes from the other correlators in Eq. (64) can be
obtained simply by using tree level and order g 2 data.
In order to simplify the calculation it is convenient to make use of conformal invariance
to map one of the insertion points to infinity, while at the same time appropriately rescaling
the operator inserted at that point. Introducing the abbreviated notation
() = lim x24 C11

(x2 ),
C11
x2
(65)
we find after a lengthy calculation (which is sketched in Appendix A)

11

C (x1 )C11
():tr 2 (x3 ) 2 (x3 ) : :tr 3 (x4 ) 3 (x4 ) : g 4
2

2
2
2 x 2
2
x13 x14
x34
x34
x14
N 2 (N 2 1) 1
13
=
B
,
+
2
4
2 x2
2
2
2
2
16(4 2)6 2x34
x34
x14
x13
x13
x14
34
2 2
2 2
x
x
x
x
1
1
+ 2 2 ln 13
ln 14
+ 2 2 ln 14
ln 13
2
2
2
2
2x13x34
x14
x34
2x14x34
x13
x34
2 2

x x
1 2
+ 2 2 ln x34
+ 1 ln 132 34 + 1
x13 x34
x14
2

2 2
2

x13 x14
x14 x34
1 2
4
2
ln x34
+ 2 2 ln x34 + 1 ln
+1
+1 + 4 B 2 , 2
2
x14 x34
x13
x34
x34 x34

2

1 2
+ 2 2 ln x13 + 1 ln x14 + 1 ,
(66)
x13 x14
where B is the box integral given in Eq. (A.3). In principle one can restore the
x2 -dependence, along the lines of [24], but we will refrain from doing so here since it
is not needed for our present limited purposes.
84
From Eq. (66) we can now determine the coefficient of the leading logarithmic terms
in G associated to the scalar operators of (naive) dimension 0 = 4. This requires to
identify also all the other operators with (naive) dimension smaller or equal to 4 and to
2 ) in (66), we find
subtract their contributions. In particular, for the coefficient of ln2 (x13
2
2
2
6
the factor N (N 1)/(32(4 ) ). Looking at the other correlators in Eq. (64), we
2
see that in the functions C 11 C11
K1 V33 and C 11 C11
V2 K1 only the protected operators
of dimension 0 = 2 and 0 = 4 can appear in the relevant intermediate channel, while
2 3
V2 V3 the component K84 of the Konishi multiplet alone contributes. Finally
in C 11 C11
2 ) term. The explicit
the disconnected parts in Eq. (64) give no contribution to the ln2 (x13
calculation of the residual coefficients after subtracting the known contribution of all these
operators confirms that for N = 2 at level 0 = 4 in the representation 20 of SU(4) there
having anomalous dimension given by
is only one unprotected scalar operator, namely M,
Eq. (38). This fact will allow us to compute in the next section the anomalous dimensions
of all the scalar operators appearing at the same level in the representation 20 of SU(4) for
N > 2.
7. A second derivation of the order g 2 anomalous dimension of the 20 operators

In this section we present an alternative computation of the anomalous dimensions of
the scalar operators with naive dimension 0 = 4 belonging to the representation 20 of
SU(4), that does not require their explicit construction in terms of fundamental fields.
The basic assumption is that, according to conformal invariance, logarithmic terms
exponentiate to a power law. This has been confirmed in all order g 2 - and g 4 -calculations
performed so far.
Another important ingredient is the observation that four-point functions of single trace
operators quadratic in the fundamental fields (like the protected operators C 11 , C11
as well
as the lowest component of the Konishi multiplet K ) have a particularly simple polynomial
behaviour in N . Indeed, expanding the interaction Lagrangian in powers of g one observes
that the coefficient of g 2n contains the product of exactly 2n SU(N) structure constants
f abc , hence the connected part of the perturbative amplitude can be represented as a
single trace of 2n SU(N) matrices in the adjoint representation. It is a rather long, but
straightforward, computation to evaluate all the traces we shall need (i.e., those that can
appear in perturbation theory up to order g 10 ). Below we list the types of N behaviour that
one encounters. The coefficients with which these power behaviours appear depend on the
particular trace one is considering and will be of no interest here. One gets
order g 2 : only N(N 2 1);

order g 4 : only N 2 (N 2 1);
order g 6 : only N 3 (N 2 1);
order g 8 : linear combination of N 4 (N 2 1) and N 2 (N 2 1);
order g 10 : linear combination of N 5 (N 2 1) and N 3 (N 2 1).
85
In the computation of the four-point function G there are both connected and
disconnected contributions (see Eq. (64)). The disconnected piece is effectively double
trace, so it is multiplied by one more factor of N 2 1 with respect to the connected term.
We consider the double OPE x13 0, x24 0, i.e., the exchange of operators
(x2 )K1 (x4 ), and concentrate on the leading

between the products C 11 (x1 )K1 (x3 ) and C11
2
ln(x34) behaviour only. To be more precise we shall analyze the coefficient of the terms
2 ) in the correlator (62). A convenient normalization is to extract a common
g 2n lnn (x34
factor equal to
0 K1 K1 0
(1K )n C 11 C11
,
2
n!
(N 1)
(67)
where

(N 2 1)

C 11 C11
=
,
0
2(4 2)2
3(N 2 1)
K1 K1 0 =
4(4 2 )2
(68)
(69)
are the tree level normalizations of the two-point functions of the operators C 11 and K1 ,
respectively, while 1K = 3N/(4 2 ) is the order g 2 correction to the anomalous dimension
of the Konishi multiplet (Eq. (10)).
2 )
With these choices the contribution from the disconnected diagrams to the g 2n lnn (x34
2
term is N 1 for any value of n. It is convenient to measure anomalous dimensions in
units of the anomalous dimension of the Konishi multiplet, by defining
i =
1Oi
(70)
1K
and to introduce the ratio

Fi =
N2 1
C 11 K1 Oi 0 C11
K1 Oi 0
C 11 C11
0 K1 K1 0
Oi Oi 0
(71)
to normalize the contributions of the different operators to the four-point function (64).
In Eq. (71) the notation X0 means that the corresponding expectation value has been
evaluated at tree level. We thus obtain the following conditions.
At tree level
F1 + F2 + F3 P0 =
3(N 2 2)(N 2 + 1)
.
(3N 2 2)
(72)
The complicated form of the r.h.s. is due to the subtraction of the contribution of the
protected operator D.
At order g 2 and g 4 the OPE analysis of (66) lead to the equations
g2 :
g4 :
F1 1 + F2 2 + F3 3 P1 = N 2 + 1,
13
F1 12 + F2 22 + F3 32 P2 = N 2 + .
3
(73)
86
At order g 6 and g 8 and g 10 one gets

g6 :
F1 13 + F2 23 + F3 33 P3 = N 2 + b3 ,
g8 :
F1 14 + F2 24 + F3 34 P4 = N 2 + b4 +
g 10 :
c4
,
N2
c5
F1 15 + F2 25 + F3 35 P5 = N 2 + b5 + 2 ,
N
(74)
where bi and ci are for the moment unknown coefficients.

It is convenient to eliminate the Fi from the above equations in favour of the PL ,
obtaining the system
PL+3 (1 + 2 + 3 )PL+2 + (1 2 + 1 3 + 2 3 )PL+1 (1 2 3 )PL
= 0,
(75)
for L = 0, 1, 2. Note that this equation actually holds in general (i.e., for any value of L).
Hence the knowledge of P0 P5 completely determines all PL .
In order to compute the unknown coefficients bi and ci recall that for N = 2 there is only
From Eq. (73) its relative anomalous dimension is determined
one possible operator, M.
to be M
=
5/3
for
N
= 2 in agreement with Eq. (38). Then from the system (74) for

N = 2 we find b3 = 89/9, so that P3 is completely determined, and we can express b4 and
b5 in terms of c4 and c5 as follows: b4 = 517/27 c4 /4, b5 = 2801/81 c5 /4.
The knowledge of P3 allows us to solvethe case N = 3, where only two operators are
present with anomalous dimensions (19 61)/18. Then we can also fix the coefficients
entering P4 and P5 to be c4 = 40/9 and c5 = 1720/81.
Substituting back these numbers into the system (75) gives for generic N 4 the
equations
8
1 + 2 + 3 = ,
3
1 2 3 =
1 2 + 1 3 + 2 3 =
5(3N 2 2)
,
27N 2
10(2N 2 1)
,
9N 2
(76)
whose solutions precisely yield the values of the anomalous dimension (57) previously
found in Section 5.
8. Relation with the Penrose double scaling limit

In this admittedly more speculative section we would like to discuss how one might
generalize our calculations in order to make contact with the double scaling limit that
corresponds to type IIB superstring around a pp-wave supported by a RR 5-form flux [30].
From the supergravity perspective this corresponds to performing a Penrose limit around
a null geodesic at the center of AdS5 S 5 . The resulting maximally supersymmetric
87
geometry8
2 2 + 2 2 2
+ Y dx
+ d Y ,
ds 2 = 4 dx + dx 2 X
+ d X
(77)
F+2356 = F+1478 = ,
(78)
e = gs ,
with all other fields set to zero, admits an exactly solvable worldsheet description in
the GreenSchwarz formalism in the light-cone gauge [31,48], see however [49]. In the
Penrose limit the SU(2, 2|4) super-isometry of AdS5 S 5 undergoes an InonWigner
contraction. In particular, SO(4, 2) SO(4)X U (1) and SO(6) SO(4)Y U (1)J but
at the same time a Heisenberg group, H (8), emerges so that the total number of generators
remains equal to 30 as for AdS5 S 5 . A similar rearrangement takes place for the 32
supersymmetry charges. In addition to the standard identifications
gs =
g2
,
4
L2
=

g2 N ,
at large N and large J , with J

=
(79)
N , the relevant coupling turns out to be [31]
g2 N
,
4J 2
(80)
where J is the U (1)J charge that appears in the above decomposition of SO(6).9 The
relevant null geodesic is identified by the light-cone coordinates x = 1 ( )/2L2 ,
where is the global time in AdS5 [6], conjugate to , and is an angular coordinate in
S 5 , conjugate to J . Thus in the Penrose limit the light cone momentum P + is essentially
proportional to J , i.e.,
p+ =
+J
J
.
2
2L
L2
(81)
Operators with = J and = J + 1 are known to be protected, as a consequence of

SU(2, 2|4) shortening conditions of BPS type that survive the relevant InonWigner
contraction. The simplest nearly protected operators, that are expected to correspond to
b |p+ , are of the form
the lowest type IIB superstring excitations Yna Yn
Aab
n =

qn9 Tr Z J 9 Y a Z 9 Y b ,
(82)
9=0
where qn = exp[2in/(J + 1)] and, in our previous notation, Z = 1 and Y a = a for

a = 2, 3, 5, 6. Notice that, at variance with what is done in Refs. [3134], we defined the
Fourier coefficients in (82) to be qn = exp[2in/(J + 1)], and not qn = exp[2in/J ], in
order to have a formula yielding J + 1, and not J , linearly independent operators which
are exactly orthogonal at tree level in the planar limit for different ns.
8 In our conventions, the indices of X
run over 1, 4, 7, 8 and those of Y over 2, 3, 5, 6.
9 This U (1) does not coincide with the U (1) in the N = 1 decomposition of N = 4 SYM used so far.
J
R
88
The knowledge of the free spectrum of the light cone Hamiltonian P gives a prediction
for the planar contributions to the anomalous dimensions of the operators Aab
n , i.e.,

4gs Nn2
p = ( J ) = 1 +
(83)
+ (non-planar).
2
J2
As noticed by [31], the effective string loop counting parameter is g2 = J 2 /N . In [32] the
first non-planar corrections in SYM-theory have been explicitly computed and matched
with string one-loop corrections.
The obvious difference between Aab
n and the operators
we have considered in the
previous sections is the presence of a large number (J N ) of Zs that account for
their large U (1)J charge. Aside from combinatorial factors that can be elegantly deduced
by resorting to a Gaussian matrix model [33], one can explicitly check that most of the Z
fields are spectator to lowest non-trivial orders in and g2 . A subtler difference pertains
to the SO(4) transformation properties. In the decomposition 20 = (1, 1)0 + (3, 3)0 +
(2, 2)+1 + (2, 2)1 + (1, 1)+2 + (1, 1)2 , the operators O911 , we have mostly concentrated
our attention on, transform in the (1, 1)+2 . On the other hand Aab
n for generic J belong
to (3, 3)J + (3, 1)J + (1, 3)J + (1, 1)J . Explicit perturbative analyses in Refs. [3134]
have so far dealt with the non-singlet components. It would be interesting to study how
the remnant of SU(4) R-symmetry after InonWigner contraction could help disposing
of any difference among the various SO(4) irreducible components.
In studying (non)renormalization properties, it is worth noticing that for generic J
the operators Aab
n can belong only to the SU(4) representations which appear in the
decomposition
[0, J, 0] [0, 1, 0] [0, 1, 0]
= [0, J + 2, 0] + [2, J 2, 2] + 2[1, J, 1] + [2, J 1, 0]
+ [0, J 1, 2] + 2[1, J 2, 1] + 3[0, J, 0] + [0, J 2, 0]
(84)
and observing that only the last representation in the r.h.s. does not saturate any unitarity
bound and necessarily belongs to a long multiplet. The first five, if super-primary, are
protected (1/2, 1/4, 1/4, 1/8 and 1/8 BPS, respectively). Only super-descendants in these
representations can acquire anomalous dimensions. The sixth and seventh representations
saturate the bound 2 + k + l + m that leads to a linear type shortening condition of the
type (23).
Although a detailed resolution of the mixing of these operators has so far proven to
be quite elusive and goes beyond the scope of the present investigation, we would like to
argue that our computations are the building blocks for the study of their renormalization
properties. The N = 1 formalism makes more transparent many laborious cancellations
found to take place in [3234]. For instance, the operator Aab
0 , being totally symmetric
in the SU(4) indices, is a CPO belonging to the representation [0, J + 2, 0], hence
it is protected (1/2 BPS) with = J + 2. Moreover the symmetric, A(ab)
n , and the
antisymmetric, A[ab]
n , parts of the operators defined in Eq. (82) are mutually orthogonal
at tree-level for all values of n, N and J .
89
Actually one can go one step further and include also multiple trace operators.
A preliminary analysis of the symmetric case for J = 4, including all possible mixings,
shows the existence, for any N , of eight protected and three unprotected operators. The
latter have exactly the anomalous dimensions reported in Eq. (57) and turn out to be
super-descendants of the operators, Oi , identified in Section 5. To be more explicit, for
any N , the three unprotected operators are linear combinations of only the following three
commutators (Z = 1 , Y = 2 )

W1 = Tr Y [Y, Z] ZZZ ,
(85)

W2 = Tr Y Z [Y, Z] ZZ ,
(86)

W3 = Tr(ZZ) Tr Y [Y, Z] Z .
(87)
For large N , the double trace operator W3 decouples from the single trace operators
W1 and W2 . The anomalous dimension of W3 tends to the anomalous dimension of the
Konishi supermultiplet, 1K . The anomalous dimensions of the other two operators, which
for large N are given by

5 1 W2
5 1W1 ,
(88)
(compare with Eq. (61)), tend to the values (60)

1
5 5 1K .
(89)
6
Even though multiple trace operators decouple from single trace ones at large N ,
mixing effects may compete with non-planar J 2 /N corrections that are dual to string
loop corrections to masses. Only after disentangling operator mixing, can the comparison
between gauge theory and string theory results be sensible. We will come back to this and
related issues in a forthcoming publication.
Despite the success of the proposal of Ref. [31], the way holography is realized
in the pp-wave background is still a matter of debate [50]. This prevents a naive
application of the procedure that for (asymptotically) AdS spaces has lead to holographic
renormalization [51]. However, conformal flatness of the background, that is made
manifest by the coordinate transformation [52]

u = tan x + ,
=
X

X
,
cos(x + )
1 2 2 +

X + Y tan x ,
2
Y
Y =
,
cos(x + )
v = x
(90)
guarantees the absence of higher derivative corrections to amplitudes with fewer than
four insertions, much in the same way as in AdS5 S 5 [53], and makes one hope that
a viable string description along the lines of [49] could be not far from reach. It would
then be interesting to study non-perturbative effects induced by instanton-like D-brane
solutions [54] in the double scaling limit. In particular, it should be possible to extend our
proof of the absence of instanton contributions to the anomalous dimensions of operators
dual to string excitations. We plan to come back to these and related issues in the near
future.
90
9. Conclusions
The centerpiece of this article is the resolution of the mixing of scalar operators of
naive dimension 4 in the 20 of SU(4) at order g 2 in perturbation theory. The problem
is investigated in two different ways. First, by a direct orthogonalization of all possible
single and double trace structures. Second, via an OPE analysis of a four point correlator
computed up to order g 4 . The agreement of the two approaches was for us an important
consistency check, because the expressions found for the mixing matrix and anomalous
dimensions are not rational in N . This certainly comes as an unexpected feature, since all
computations so far performed were compatible with the ratios of anomalous dimensions
being rational numbers, but it is not in conflict with any fundamental property of the field
theory.
Of the three unprotected operators not belonging to the Konishi multiplet that we have
identified, there is one whose anomalous dimension tends to 1K in the large N limit.
In this limit the operator under consideration can be identified with the product of the
lowest Konishi operator and the lowest component of the supercurrent multiplet. We find
it noteworthy that its anomalous dimension is the sum of the anomalous dimensions
of its two factors, thus perhaps hinting at a deeper physical significance of singleton
multiplication as a way of generating generic UIRs of SU(2, 2|4) and possibly of its Inon
Wigner contraction which is relevant for the double scaling Penrose limit dual the pp-wave
background.
Acknowledgements
We are grateful to Matthias Blau, Hugh Osborn, Augusto Sagnotti, Kostas Skenderis
and Marika Taylor for useful conversations. M.B. thanks Soo-Jong Rey and Dan Freedman
for clarifying e-mail exchanges. B.E. acknowledges many useful discussions with Emery
Sokatchev and Christian Schubert. This work was supported in part by INFN, by the EC
contract HPRN-CT-2000-00122, by the EC contract HPRN-CT-2000-00148, by the INTAS
contract 99-0-590 and by the MURST-COFIN contract 2001-025492.
Appendix A
A.1. The explicit calculation of the correlation function (66)
In this appendix we sketch the calculation of the connected part of the amplitude
A(x1 , x2 , x3 , x4 )

= C 11 (x1 )C11
(x2 ):tr 2 (x3 ) 2 (x3 ) : :tr 3 (x4 ) 3 (x4 ) :
(A.1)
at order g 4 .
In the manifestly N = 1 formalism, the N = 4 vector multiplet decomposes into
the three chiral superfields and an N = 1 YM superfield. In (2) and (3) we have
(a)
(b)
(c)
91
(d)
Fig. 2. Feynman diagrams contributing to Eq. (A.1).
given the corresponding propagators. Vertices can be deduced by expanding the (nonpolynomial) action to the required order. Perturbative calculations are expressible in terms
of Feynman graphs involving superfields, called superdiagrams. The diagrams below are
to be understood in this sense. One obtains the component field Feynman graphs by
expanding in the Grassmann variables s and s. In particular, component field graphs
for the lowest scalar components of chiral multiplets are obtained by simply putting to
zero the Grassmann variables of the external points. The technique is efficient in that the
number of diagrams to be drawn is generally smaller and in that the (partial) cancellation
of divergences typical of supersymmetric theories is made manifest.
Let us now discuss the set of superdiagrams that contribute to A(x1 , x2 , x3 , x4 ). There
are no connected diagrams with two vector lines, since A vanishes at tree level. A direct
calculation shows also that the diagrams with one chiral and one vector line sum up to zero.
Hence we shall consider only diagrams with all lines corresponding to chiral propagators.
We project on the lowest components of the supermultiplets, so graphs with a cubic chiral
vertex with all three lines attached to external points do not contribute. Moreover, each
internal chiral line gives rise to a delta function. The relevant order g 4 diagrams are shown
in Fig. 2.
The topologies (a), (b) and (c) also occur with points 3 and 4 flipped. All diagrams have
the same combinatorial weight and group factor. In order to evaluate the integral coming
, to
from diagram (a), we send the argument of one of the dimension two operators, say C11
infinity as in Eq. (65). Diagram (d) reduces to the product of two box integrals, so we shall
concentrate on the graphs (a) and the sum of (b) and (c).
A.2. Point splitting regularization
We shall first sketch the calculation of the most complicated diagram (a) regularized
by point-splitting as described in Section 2. The basic object is the massless scalar box
integral defined by

2
d 4 x5
= 2 2 B(r, s).
(A.2)
2
2
2
2
x15x25 x35 x45 x13 x24
A more explicit expression of B is given by

r +s1 p 2
1
B(r, s) =
ln(r) ln(s) ln
p
2

2
2
2Li2
2Li2
,
1+r s + p
1r +s+ p
(A.3)
92
where
p = 1 + r 2 + s 2 2r 2s 2rs
(A.4)
is a function of the two conformally invariant ratios

r=
2 x2
x12
34
2 x2
x13
24
s=
2 x2
x14
23
2 x2
x13
24
and the dilogarithm function is defined as Li2 (z) =

The small r limit of B(r, s) is
(A.5)

zn
n=1 n2 .
lim B(r, s) = ln(r) + 2 + O(r).
r0
(A.6)
The partial derivatives of B(r, s) have the following form

r +s 1
2
r s1
B(r, s)
ln(s) + ln(r),
p
pr
p
r +s 1
2
sr 1
B(r, s)
ln(r) + ln(s).
s B(r, s) =
p
ps
p
r B(r, s) =
If we introduce the notation

2
.4
2
(x4 5 ) = x4 x5 ,
2
2

.3
(x3 6 )2 = x3 x6 ,
2
the integral corresponding to diagram (a) has the form

d 4 x6 d 4 x5
Ia =
,
2 x 2 (x + )2 (x )2
(x3+ 6 )2 (x3 6 )2 x56
4 5
15 4 5
(A.7)
(A.8)
(A.9)
where thanks to conformal invariance we have sent x2 to infinity in order to simplify

the calculation. As stated in the discussion of Eq. (65), this limit is sufficient for our
present purposes. In order to evaluate (A.9) we shall add and subtract from Ia the auxiliary
factorized integral

d 4 x6
d 4 x5
2
I1 = x34
(A.10)
,
2
2
2
x46 (x3+ 6 )2 (x3 6 )2
x15(x4+ 5 )2 (x4 5 )2 x35
which in the limit of small .3 and .4 can be computed by the use of Eqs. (A.2)(A.6) to
give

2

2 2

.
. x
4
lim I1 = 2 2 ln 23 + 2 ln 24 13
(A.11)
+
2
.
2
.3,4 0
x14 x34
x34
x14 x34
Note that in this limit I1 has logarithmic divergences, which are cancelled by the factors
2 ) 12 K (g 2 ) in the definition of the renormalized Konishi operator, Eq. (13). We can
(.3,4
evaluate the integral over x5 in the difference Ia I1 and take the limit .4 0, using
93
again Eqs. (A.2)(A.6), obtaining

lim (Ia I1 ) =
.4 0
J=
2
J,
2
x14
2 2
x16 x34
d 4 x6
ln
.
2
2 x2
2
2
x46 (x3+ 6 ) (x3 6 )
x46
13
(A.12)
(A.13)
In order to calculate J let us first prove that it is finite in the limit .3 0. To this end
we compute the derivative of J with respect to x1 . Since under derivation the logarithm
disappears, we can express the derivative of J in terms of the standard box integral and
take the limit .3 0 with the finite result
2
2 2
2
x
x13 x14
2
lim x J = 2x13 2 2 ln 13
(A.14)
B
,
.
x3
2
2
2 x2
.3 0 1
x13x34
x14
x34
x34
34
Hence, if present, the divergent (for .3 0) part of J will be independent of x1 . But in
the limit x1 x4 , J is equal to zero, so it has a finite .3 0 limit. Scale invariance then
allows us to write it as
2
2
x13 x14
2
J= 2 f
(A.15)
,
.
2 x2
x34
x34
34
Formula (A.14) gives rise to two equations for the two partial derivatives of the function
f (a, b) defined above. These, together with the initial condition f (1, 0) = 0, determine
completely the function f , yielding

b
1
.
f (a, b) = (a b 1)B(a, b) + ln(a) ln
(A.16)
2
a
Inserting this solution in Eqs. (A.15), (A.13) and taking into account Eq. (A.11), we obtain
the complete expression of Ia (Eq. (A.9)).
The computation of the other three diagrams (b), (c) and (d) is simpler and makes use
of Eqs. (A.2)(A.6) only. The final result of the calculation is given in Eq. (66).
A.3. Dimensional regularization
The calculation is done in the Euclidean regime and with the spacetime dimension
changed to d = 4 + 2.. In order to preserve the validity of equation 2 G0 = (d) , one
has to appropriately modify the scalar propagator:
( d 1)
( d2 )
1
1 2
G0 =
d
2
2
2
4 x
(d 2)2 2 x

1 (1 + .) 1 (1+.)
=
.
4 2
.
x2
(A.17)
The regulator is introduced in the spirit of dimensional reduction [55], i.e., the x
dependence of the superpropagators is changed as indicated above, but their -dependence
94
is not. One can perform the graph calculation formally since the regulator preserves the
algebra by definition and is compatible with partial integration.
Renormalization introduces a renormalization factor Z(g 2 ) for the Konishi operator
in the operator sums :tr(2 (x) 2 (x)): = V22 + Z(g 2 )K1 /3 and :tr(3 (x) 3 (x)): = V33 +
Z(g 2 )K1 /3, with Z|g 0 = 1. The order g 2 part of the Z factor can be found by
reexpressing (14) in dimensional regularization. The order g 4 piece of Z does not play
any role in the current calculation because there are no connected tree diagrams between
, K1 and :tr(I (x) I (x)): for I = 2, 3. However, in order to obtain a finite

C 11 , C11
expression for the correlator (A.1) we have to take into account the subtractions at order
g 4 , coming from its expression at tree level and order g 2 .
A.3.1. Diagram (a)
The most complicated calculation concerns the contribution of this diagram. We need
to compute the x-space integral

d 4 x5
d 4 x6
,
Ja (1, 3, 4) =
(A.18)
2 x4
2 x4
x15
x56
45
36
where we have exploited conformal invariance to map the point x2 to infinity as in the
previous section. We shall first evaluate the second subintegral, e.g., by the standard
Feynman parameter trick. We find

d (4+2.)x6
1 (1+2.)
(A.19)
=
,
2 x 4 )(1+.)
2
. x35
(x56
36
where
=
(2+.)
.
(1 + 2.) (1 + .)
(A.20)
Note that the power of the propagator in (A.19) is not the standard 1 + .. We split
2 )(1+2.) = (x 2 )(1+.) (x 2 )(.) and expand the last factor, thus obtaining three
(x35
35
35
contributions to Ja

2 )
d 4 x5 ln(x35
d 4 x5
Ja (1, 3, 4) =
2 x 4 x 2 )(1+.)
2 x 4 x 2 )(1+.)
.
(x15
(x15
45 35
45 35

2 )
d 4 x5 ln2 (x35
.
+
+
2 x 4 x 2 )(1+.)
2
(x15
45 35
Ja1 + Ja2 + Ja3 + .
(A.21)
In all three pieces the remaining integral diverges and yields another simple pole in ..
Since we need to know Ja up to O(.), we have to compute the first of these integrals at
O(1/.), O(1) and O(.), the second at O(1/.), O(1), but only the pole part of the third is
needed and all higher terms can be ignored.
One can compute the integrals Jai by the method of Gegenbauer polynomials (see [56]
and references therein). This approach is particularly suited to three point integrals in which
only one propagator has an unusual exponent. The other two propagators are expanded
95
in orthogonal polynomials and the angular and radial integrations are carried out. This
produces two infinite series, which, however, can be written as an expansion in orthogonal
polynomials of a single propagator at the expense of introducing a parametric integral [57].
We first consider the contribution Ja1 . Let x = x14 , z = x34 , then for |x| < |z|:
Ja1 Ia1 =
.
.
1
2
=
2.
dt
d (4+2.)y
(y 4 (x y)2 (z y)2 )(1+.)
1
z2
(1+2.)
(1+3.)
x2
(1+2.)
t
(1+.)
1
.
((z xt)2 )(1+.)
(A.22)
The last term in the denominator can be factorized as

(1+.)
(1+.)
(t t )(t t+ )
= x2
,
(A.23)

where t = (x z (x z)2 x 2 z2 )/x 2 [57]. The first parameter integral in (A.22) is
regular, while the second may be performed after subtracting the singularity. The result
reads

(z xt)2
Ja1 =
(1+.)
2 ).
(x13
2
+ O(1).
2 x 2 )(1+2.)
. 2 (x14
34
(A.24)
The finite part of the integral contains the box integral B and some ln2 terms.
Remarkably, the knowledge of Ja1 is only necessary in as much as its infinities are
concerned. The renormalization of the Konishi operator produces a diagram yielding
Z|g 2 Ia1 , which exactly cancels the contribution Ja1 .10
Let us now focus on Ja2 . We shall write the integral as

Ja2 =
2 )
d (4+2.) x5 ln(x35
2 x 2 x 4 )(1+.)
(x15
35 45
=
1+.

d (4+2.) x5

.

2
2
4
(1+.)
(x15(x35 ) x45 )
=1
(A.25)
Define

Ka2 =
d (4+2.)x5
.
2
2 )(1+2.) (x 2 )(1+.)
(x15 )(1+.) (x45
35
(A.26)
We can isolate the divergence in Ja2 by writing

x14 4
d 4 x5
+ O(.),
1 Ka2 = (x14 4 1)
2
2 (x 2 )
1 + 2.
x15 x45
35
(A.27)
10 The finite part of the Z factor leads to a term bI . This has only a first order pole in . so that formula (A.24)
a1
suffices to account for this contribution.
96
since the integral is finite if is close to one. Swapping integration and differentiation in
the l.h.s. we derive (the integrals are finite as long as the regulator is not removed)

x14 4
1 Ka2
1 + 2.

d (4+2.) x5
d (4+2.) x5
2
= x14
.
(A.28)
2 x 4 )(1+.) (x 2 )(1+.)
2 ). (x 4 )(1+.) (x 2 )(1+.)
(x15
(x15
45
35
45
35
2 ). up to first order in .
Let us call the second term in the last line La2 . Expanding (x15
and recombining the pole parts of the resulting two integrals one finds

1
d (4+2.) x5
La2 = 2
(A.29)
+ O(.).
4
2 )(1+.)
.
(1+.)
(x14)
(x45 )
(x35
2 , we
We substitute this expression into (A.28) and equate with (A.27). After dividing by x14
obtain

d (4+2.) x5
2 (x 2 ) x 4 )(1+.)
(x15
35
45

1
x14 4 1
d (4+2.) x5
d 4 x5
= 2
+
+ O(.).
2
4
2
2
2 ) x 2
(x14 )(1+.)
((x35) x45)(1+.)
x14
x15 (x35
45
(A.30)
The first term in the r.h.s. contains a divergence, but as a two-point structure it is readily
calculable. We compute the second term in the r.h.s. again by means of Gegenbauer
polynomials. As above, we introduce a one parameter integral to rewrite the result of the
angular integration in closed form. One finds for = 1 +

d 4 x5
2 x 2 (x 2 )(1+)
x15
45 35

2
2
d 4 x5
= 1
+ O 2 ,
(A.31)
ln x13 + ln x34
2
2
2
2
x15x35 x45
which is sufficient to calculate the first order parametric derivative in (A.25). Finally, we
employ the identity
2

x
2
d 4 x5
x14 4 1
= 2 2 ln 14
.
(A.32)
2
2
2
2
2
x14
x15 x35 x45 x14x34
x13
Collecting terms we find
2 )
ln(x34
2
4 1
+
2
2
2 x2
. (x14 )(1+.) (x34)(1+2.)
2 x14
34
2

2 2 x2
2

2
x
x14 34
x14
x
34
2
B
+
ln
x
+
ln
x
ln
1 14
,
13
34
2
2
2 x2
2
x13
x13
x13
x
13
13
+ O(.).
(A.33)
Ja2 =
97
As for Ja3 , since we learned above how to extract a simple pole, without any further
calculation we get
Ja3 =
2 )2
4 ln(x34
+ O(.).
2
2
2 x14 x34
(A.34)
A.3.2. Diagrams (b) and (c)

It is convenient to compute directly the sum of these graphs, since compensations
between the two terms lead to a simpler result. The x-space integrals are of the form

d 4 x6
d 4 x5
Jb + Jc =
(A.35)
+ (3 4).
2 x2
2 x2 x4
x15
x56
35
36 46
The second subintegral is divergent. It is given by Ia1 = .Ja1 / which was calculated
already in Eqs. (A.22)(A.24). Its coordinate behaviour at orders O(1/.) and O(1) is
2 x 2 ), which does not lead to overlapping divergences in the second
essentially 1/(x34
45
integration. We can then expand Ia1 obtaining (one can neglect terms of order O(.) and
higher)
2 . 2 .

x
x45
d (4+2.)x5
1
Jb + Jc =
+ 35
2
2
2
2
2
2
. (x34)(1+2.)
(x15 x35x45 )(1+.) x35
x45

2
d (4+2.)x5
d 4 x5
2
2 ln(x34 )
+
2
.
=
(A.36)
2
2
2
2
2
2
2 x2
. (x34)(1+.)
(x15 x35 x45 )(1+.)
x34
x15 x35
45
Note that the first term contains trilogarithms but is again identically cancelled by Z|g 2
times an order g 2 diagram. The second integral is once again the box B. The end result of
the calculation is in exact agreement with (66).
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, hep-th/9711200.
[2] G.G. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111.
[5] D.Z. Freedman, P. Henry-Labordere, Field theory insight from the AdS/CFT correspondence, hepth/0011086.
[6] E. DHoker, D.Z. Freedman, Supersymmetric gauge theories and the AdS/CFT correspondence, hepth/0201253.
[7] M. Bianchi, (Non-)perturbative tests of the AdS/CFT correspondence, Nucl. Phys. B (Proc. Suppl.) 102
(2001) 56, hep-th/0103112.
[8] W. Skiba, Correlators of short multi-trace operators in N = 4 supersymmetric YangMills, Phys. Rev. D 60
(1999) 105038, hep-th/9907088;
S. Penati, A. Santambrogio, D. Zanon, Correlation functions of chiral primary operators in perturbative
N = 4 SYM, hep-th/0003026;
98
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]

S. Penati, A. Santambrogio, D. Zanon, More on correlators and contact terms in N = 4 SYM, Nucl. Phys.
B 593 (2001) 651, hep-th/0005223;
S. Penati, A. Santambrogio, Superspace approach to anomalous dimensions in N = 4 SYM, Nucl. Phys.
B 614 (2001) 367, hep-th/0107071.
D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Correlation functions in the CFT(d)/AdS(d + 1)
correspondence, Nucl. Phys. B 546 (1999) 96, hep-th/9804058;
S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in D = 4, N = 4
SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074;
E. DHoker, D.Z. Freedman, W. Skiba, Field theory tests for correlators in the AdS/CFT correspondence,
Phys. Rev. D 59 (1999) 045008, hep-th/9807098;
P.S. Howe, E. Sokatchev, P.C. West, 3-point functions in N = 4 YangMills, Phys. Lett. B 444 (1998) 341;
G.E. Arutyunov, S.A. Frolov, Three-point Green function of the stress-energy tensor in the AdS/CFT
correspondence, Phys. Rev. D 60 (1999) 026004, hep-th/9901121;
B. Eden, P.S. Howe, P.C. West, Nilpotent invariants in N = 4 SYM, Phys. Lett. B 463 (1999) 19, hepth/9905085.
H. Liu, A.A. Tseytlin, Dilaton-fixed scalar correlators and AdS5 S 5 SYM correspondence, JHEP 9910
(1999) 003, hep-th/9906151;
E. DHoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the AdS/CFT
correspondence, hep-th/9908160;
M. Bianchi, S. Kovacs, Non-renormalisation of extremal correlators in N = 4 SYM theory, Phys. Lett.
B 468 (1999) 102, hep-th/9910016;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Extremal correlators in four-dimensional SCFT,
Phys. Lett. B 472 (2000) 323, hep-th/9910150.
J. Erdmenger, M. Perez-Victoria, Non-renormalization of next-to-extremal correlators in N = 4 SYM and
the AdS/CFT correspondence, Phys. Rev. D 62 (2000) 045008, hep-th/9912250.
M. Bianchi, M.B. Green, S. Kovacs, G.C. Rossi, Instantons in supersymmetric YangMills and D-instantons
in IIB superstring theory, JHEP 9808 (1998) 013, hep-th/9807033;
N. Dorey, T.J. Hollowood, V.V. Khoze, M.P. Mattis, S. Vandoren, Multi-instanton calculus and the AdS/CFT
correspondence in N = 4 superconformal field theory, Nucl. Phys. B 552 (1999) 88, hep-th/9901128.
K. Intriligator, Bonus symmetries of N = 4 super-YangMills correlation functions via AdS duality, Nucl.
Phys. B 551 (1999) 575, hep-th/9811047;
K. Intriligator, W. Skiba, Bonus symmetry and the operator product expansion of N = 4 super-YangMills,
Nucl. Phys. B 559 (1999) 165, hep-th/9905020.
J. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002;
S. Rey, J. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter supergravity,
Eur. Phys. J. C 22 (2001) 379, hep-th/9803001.
N. Drukker, D.J. Gross, H. Ooguri, Wilson loops and minimal surfaces, Phys. Rev. D 60 (1999) 125006,
hep-th/9904191;
H. Ooguri, Wilson loops in large N theories, Class. Quantum Grav. 17 (2000) 1225, hep-th/9909040;
N. Drukker, D.J. Gross, An exact prediction of N = 4 SUSYM theory for string theory, J. Math. Phys. 42
(2001) 2896, hep-th/0010274.
J.K. Erickson, G.W. Semenoff, K. Zarembo, Wilson loops in N = 4 supersymmetric YangMills theory,
Nucl. Phys. B 582 (2000) 155, hep-th/0003055;
G.W. Semenoff, K. Zarembo, More exact predictions of SUSYM for string theory, Nucl. Phys. B 616 (2001)
34, hep-th/0106015.
J. Plefka, M. Staudacher, Two loops to two loops in N = 4 supersymmetric YangMills theory, JHEP 0109
(2001) 031, hep-th/0108182;
G. Arutyunov, J. Plefka, M. Staudacher, Limiting geometries of two circular MaldacenaWilson loop
operators, JHEP 0112 (2001) 014, hep-th/0111290.
M. Bianchi, M.B. Green, S. Kovacs, Instantons and BPS Wilson loops, hep-th/0107028;
M. Bianchi, M.B. Green, S. Kovacs, Instanton corrections to circular Wilson loops in N = 4 supersymmetric
YangMills, JHEP 0204 (2002) 040, hep-th/0202003.
B. Eden, A.C. Petkou, C. Schubert, E. Sokatchev, Partial non-renormalisation of the stress-tensor four-point
function in N = 4 SYM and AdS/CFT, Nucl. Phys. B 607 (2001) 191, hep-th/0009106;
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
99
E. DHoker, J. Erdmenger, D.Z. Freedman, M. Perez-Victoria, Near-extremal correlators and vanishing

supergravity couplings in AdS/CFT, Nucl. Phys. B 589 (2000) 3, hep-th/0003218.
F.A. Dolan, H. Osborn, Implications of N = 1 superconformal symmetry for chiral fields, Nucl. Phys. B 593
(2001) 599, hep-th/0006098;
F.A. Dolan, H. Osborn, Conformal four point functions and the operator product expansion, Nucl. Phys.
B 599 (2001) 459, hep-th/0011040;
F.A. Dolan, H. Osborn, Superconformal symmetry, correlation functions and the operator product expansion,
Nucl. Phys. B 629 (2002) 3, hep-th/0112251.
M. Bianchi, S. Kovacs, G.C. Rossi, Ya.S. Stanev, Properties of the Konishi multiplet in N = 4 SYM theory,
JHEP 0105 (2001) 042, hep-th/0104016.
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Four point functions in N = 4 supersymmetric
YangMills theory at two loops, Nucl. Phys. B 557 (1999) 355, hep-th/9811172;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Simplifications of four point functions in N = 4
supersymmetric YangMills theory at two loops, Phys. Lett. B 466 (1999) 20, hep-th/9906051;
B. Eden, P.S. Howe, A. Pickering, E. Sokatchev, P.C. West, Four point functions in N = 2 superconformal
field theories, Nucl. Phys. B 581 (2000) 523, hep-th/0001138;
B. Eden, C. Schubert, E. Sokatchev, Three-loop four-point correlator in N = 4 SYM, Phys. Lett. B 482
(2000) 309, hep-th/0003096.
M. Bianchi, S. Kovacs, G.C. Rossi, Ya.S. Stanev, On the logarithmic behaviour in N = 4 SYM theory,
JHEP 9908 (1999) 020, hep-th/9906188.
M. Bianchi, S. Kovacs, G.C. Rossi, Ya.S. Stanev, Anomalous dimensions in N = 4 SYM theory at order
g 4 , Nucl. Phys. B 584 (2000) 216, hep-th/0003203.
G. Arutyunov, B. Eden, A.C. Petkou, E. Sokatchev, Exceptional non-renormalization properties and OPE
analysis of chiral four-point functions in N = 4 SYM4 , Nucl. Phys. B 620 (2002) 380, hep-th/0103230;
G. Arutyunov, B. Eden, E. Sokatchev, On non-renormalization and OPE in superconformal field theories,
Nucl. Phys. B 619 (2001) 359, hep-th/0105254;
B. Eden, E. Sokatchev, On the OPE of 1/2 BPS short operators in N = 4 SCFT(4), Nucl. Phys. B 618
(2001) 259, hep-th/0106249;
S. Ferrara, E. Sokatchev, Universal properties of superconformal OPEs for 1/2 BPS operators in 3 D 6,
New J. Phys. 4 (2002) 2, hep-th/0110174;
P.J. Heslop, P.S. Howe, OPEs and three-point correlators of protected operators in N = 4 SYM, Nucl. Phys.
B 626 (2002) 265, hep-th/0107212.
P.J. Heslop, P.S. Howe, On harmonic superspaces and superconformal fields in four dimensions, Class.
Quantum Grav. 17 (2000) 3743, hep-th/0005135;
P.J. Heslop, P.S. Howe, Harmonic superspaces and superconformal fields, hep-th/0009217;
P.J. Heslop, Superfield representations of superconformal groups, Class. Quantum Grav. 19 (2002) 303,
hep-th/0108235.
S. Ferrara, A. Zaffaroni, N = 1, 2, 4D superconformal field theories and supergravity in AdS5 , Phys. Lett.
B 431 (1998) 49, hep-th/9803060;
S. Ferrara, A. Zaffaroni, Bulk gauge fields in AdS supergravity and supersingletons, hep-th/9807090;
S. Ferrara, A. Zaffaroni, Superconformal field theories, multiplet shortening and the AdS5 S 5 correspondence, hep-th/9908163.
P.J. Heslop, P.S. Howe, A note on composite operators in N = 4 SYM, Phys. Lett. B 516 (2001) 367,
hep-th/0106238.
G. Arutyunov, S. Frolov, A.C. Petkou, Perturbative and instanton corrections to the OPE of CPOs in N = 4
SYM 4 , Nucl. Phys. B 602 (2001) 238;
G. Arutyunov, S. Frolov, A.C. Petkou, Nucl. Phys. B 609 (2001) 540, hep-th/0010137, Erratum;
G. Arutyunov, S. Frolov, A.C. Petkou, Operator product expansion of the lowest weight CPOs in N = 4
SYM(4) at strong coupling, Nucl. Phys. B 586 (2000) 547, hep-th/0005182.
M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, A new maximally supersymmetric background of
IIB superstring theory, JHEP 0201 (2002) 047, hep-th/0110242;
M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry,
Class. Quantum Grav. 19 (2002) L87, hep-th/0201081;
100
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
M. Blau, J. Figueroa-OFarrill, G. Papadopoulos, Penrose limits, supergravity and brane dynamics, hepth/0202111.
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.
N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, pp-wave string
interactions from perturbative YangMills theory, hep-th/0205089.
C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, A new double-scaling limit of N = 4 super-Yang
Mills theory and pp-wave strings, hep-th/0205033.
D.J. Gross, A. Mikhailov, R. Roiban, Operators with large R charge in N = 4 YangMills theory, hepth/0205066.
L. Brink, J. Scherk, J.H. Schwarz, Supersymmetric YangMills theories, Nucl. Phys. B 121 (1977) 77;
F. Gliozzi, D.I. Olive, J. Sherk, Supersymmetry, supergravity and the dual spinor model, Nucl. Phys. B 122
(1977) 253.
M. Grisaru, M. Rocek, W. Siegel, Zero value of the three-loop function in N = 4 supersymmetric Yang
Mills theory, Phys. Rev. Lett. 45 (1980) 1063;
W.E. Caswell, D. Zanon, Zero three-loop beta-function in N = 4 supersymmetric YangMills theory, Nucl.
Phys. B 182 (1981) 125.
S. Penati, A. Santambrogio, D. Zanon, Two-point functions of chiral operators in N = 4 SYM at order g 4 ,
JHEP 9912 (1999) 006, hep-th/9910197.
S. Kovacs, A perturbative re-analysis of N = 4 supersymmetric YangMills theory, hep-th/9902047;
S. Kovacs, N = 4 supersymmetric YangMills theory and the AdS/SCFT correspondence, Ph.D. Thesis,
hep-th/9908171.
L. Andrianopoli, S. Ferrara, KK excitations on AdS5 S 5 as N = 4 primary superfields, Phys. Lett. B 430
(1998) 248, hep-th/9803171;
L. Andrianopoli, S. Ferrara, Non-chiral primary superfields in the AdSd+1 /CFT d correspondence, Lett.
Math. Phys. 46 (1998) 265, hep-th/9807150;
L. Andrianopoli, S. Ferrara, On short and long SU(2, 2|4) multiplets in the AdS/CFT correspondence, Lett.
Math. Phys. 48 (1999) 145, hep-th/9812067.
D. Anselmi, D.Z. Freedman, M.T. Grisaru, A.A. Johansen, Universality of the operator product expansions
of SCFT 4 , Phys. Lett. B 394 (1997) 329, hep-th/9608125;
D. Anselmi, D.Z. Freedman, M.T. Grisaru, A.A. Johansen, Non-perturbative formulas for central functions
of supersymmetric gauge theories, Nucl. Phys. B 526 (1998) 543, hep-th/9708042;
D. Anselmi, J. Erlich, D.Z. Freedman, A.A. Johansen, Positivity constraints on anomalies in supersymmetric
gauge theories, Phys. Rev. D 57 (1998) 7570, hep-th/9711035;
D. Anselmi, The N = 4 quantum conformal algebra, Nucl. Phys. B 541 (1999) 369, hep-th/9809192.
V.K. Dobrev, V.B. Petkova, All positive energy unitary irreducible representations of extended conformal
supersymmetry, Phys. Lett. B 162 (1985) 127.
M. Flato, C. Fronsdal, One massless particle equals two Dirac singletons: Elementary particles in a curved
space. 6, Lett. Math. Phys. 2 (1978) 421;
M. Flato, C. Fronsdal, On dis and racs, Phys. Lett. B 97 (1980) 236;
M. Flato, C. Fronsdal, Quantum field theory of singletons: The rac, J. Math. Phys. 22 (1981) 1100;
M. Flato, C. Fronsdal, Quarks or singletons?, Phys. Lett. B 172 (1986) 412.
M. Gunaydin, N. Marcus, The spectrum of the S 5 compactification of the chiral N = 2, D = 10 supergravity
and the unitary supermultiplets of U (2, 2/4), Class. Quantum Grav. 2 (1985) L11.
L. Andrianopoli, S. Ferrara, E. Sokatchev, B. Zupnik, Shortening of primary operators in N -extended
SCFT 4 and harmonic superspace analyticity, Adv. Theor. Math. Phys. 3 (1999) 1149, hep-th/9912007;
S. Ferrara, E. Sokatchev, Short representations of SU(2, 2|N ) and harmonic superspace analyticity, Lett.
Math. Phys. 52 (2000) 247, hep-th/9912168;
S. Ferrara, E. Sokatchev, Superconformal interpretation of BPS states in AdS geometries, Int. J. Theor.
Phys. 40 (2001) 935, hep-th/0005151.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge Univ. Press,
Cambridge, 2001, p. 306.
M.F. Sohnius, Bianchi identities for supersymmetric gauge theories, Nucl. Phys. B 136 (1978) 461;
M.F. Sohnius, Supersymmetry and central charges, Nucl. Phys. B 138 (1978) 109;
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
101
W. Siegel, On-shell O(N ) supergravity in superspace, Nucl. Phys. B 177 (1981) 325;
P.S. Howe, K.S. Stelle, P.K. Townsend, Nucl. Phys. B 191 (1981) 445;
P.S. Howe, K.S. Stelle, P.K. Townsend, Nucl. Phys. B 192 (1981) 332.
K. Konishi, Anomalous supersymmetry transformation of some composite operators in SQCD, Phys. Lett.
B 135 (1984) 195;
K. Konishi, K. Shizuya, Functional integral approach to chiral anomalies in supersymmetric gauge theories,
Nuovo Cimento A 90 (1985) 111.
R.R. Metsaev, Type IIB GreenSchwarz superstring in plane wave RamondRamond background, Nucl.
Phys. B 625 (2002) 70, hep-th/0112044;
R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in plane wave RamondRamond
background, hep-th/0202109.
N. Berkovits, Conformal field theory for the superstring in a RamondRamond plane wave background,
JHEP 0204 (2002) 037, hep-th/0203248;
N. Berkovits, Covariant quantization of the superstring, Int. J. Mod. Phys. A 16 (2001) 801, hep-th/0008145.
S.R. Das, C. Gomez, S.J. Rey, Penrose limit, spontaneous symmetry breaking and holography in pp-wave
background, hep-th/0203164;
E. Kiritsis, B. Pioline, Strings in homogeneous gravitational waves and null holography, hep-th/0204004;
R.G. Leigh, K. Okuyama, M. Rozali, pp-waves and holography, hep-th/0204026.
See, e.g., M. Bianchi, D.Z. Friedman, K. Skenderis, How to go with an RG flow, JHEP 0108 (2001) 041,
hep-th/0105276;
M. Bianchi, D.Z. Friedman, K. Skenderis, Holographic renormalization, hep-th/0112119, and references
therein.
D. Berenstein, H. Nastase, On lightcone string field theory from super-YangMills and holography, hepth/0205048.
T. Banks, M.B. Green, Non-perturbative effects in AdS5 S 5 string theory and d = 4 SUSY YangMills,
JHEP 9805 (1998) 002, hep-th/9804170.
M. Billo, I. Pesando, Boundary states for GS superstrings in an Hpp-wave background, hep-th/0203028;
K. Skenderis, M. Taylor, Branes in AdS and pp-wave spacetimes, hep-th/0204054;
O. Bergman, M.R. Gaberdiel, M.B. Green, D-brane interactions in type IIB plane-wave background, hepth/0205183.
W. Siegel, Supersymmetric dimensional regularization via dimensional reduction, Phys. Lett. B 84 (1979)
193.
K.G. Chetyrkin, A.L. Kataev, F.V. Tkachov, New approach to evaluation of multiloop Feynman integrals:
The Gegenbauer x-space technique, Nucl. Phys. B 174 (1980) 345.
D.Z. Freedman, K. Johnson, J.I. Latorre, Differential regularization and renormalisation: a new method of
calculation in quantum field theory, Nucl. Phys. B 371 (1992) 353.

The -measure and the generalised dipoles

in the Lund model
Bo Andersson, Sandipan Mohanty, Fredrik Sderberg
Department of Theoretical Physics, Lund University, Slvegatan 14A, S-223 62 Lund, Sweden
Received 5 July 2002; received in revised form 18 September 2002; accepted 30 September 2002
Abstract
We demonstrate that the multiple gluon emission phase space in the dipole cascade model has a
strong linear correlation with the number of gluons emitted. The number of gluons per unit available
phase space at a certain resolution scale is found to be remarkably independent of the cms energy and
global event properties like thrust, and even changes in the ordering variable or resolution scale. We
show that the distribution of sizes of gluongluon dipoles in a parton cascade has stability properties
which are sufficient to account for those of the phase space variable. We observe that certain more
abstract entities, defined in the context of hadronisation and related to the gluon emission phase
space, share those properties of colour dipoles and name them generalised dipoles. We also present
an analytical model to qualitatively describe our findings.
PACS: 13.65.+i
Keywords: Lund model; QCD; Dipole cascade model; -measure; -measure; Generalised dipole
1. Introduction
The phase space for multiple bremsstrahlung emission of photons in QED is entirely
given by the properties of the original current. The reason is that the photon-quanta are
uncharged and therefore the currents are not changed because of emission of photons
(besides the recoil effects from hard emissions).
On the other hand the QCD field quanta, the gluons, are charged. Already the emission
of a first QCD gluon in e+ e -annihilation means that the original qq dipole is changed.
It so happens that the change is (to a very good approximation) from one dipole to two
E-mail addresses: sandipan@thep.lu.se (S. Mohanty), fredrik@thep.lu.se (F. Sderberg).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 9 7 - 0
B. Andersson et al. / Nuclear Physics B 646 (2002) 102126
103
independent dipoles, one between the quark (q) and the first gluon and the other between
the gluon and the anti-quark (q). The initial dipole is at rest in the total cms whereas the
two new dipoles move away from each other. The combined phase space for emitting
from either one or the other of these dipoles is found to be larger. In rapidity space the
increase can be described as an extra region of a size corresponding to the logarithm of the
squared transverse momentum of the first emitted gluon.
In the dipole cascade model [1], as implemented in the event generator A RIADNE [2],
a second emission leads to three independent dipoles and so on. The ordering variable
is an invariantly defined transverse momentum of the emitted gluons. The dipole masses
decrease quickly with successive emissions.
H ERWIG [3], and P YTHIA [4], subdivide the emerging dipole angular regions into
independent cones for each emitting parton. One way to state the coherence conditions
is that they do not permit double-counting in the emission process. H ERWIG and P YTHIA
implement the coherence conditions by means of angular ordering to avoid such overlaps.
The ordering variable for H ERWIG is just the angle (or the rapidity variable) occurring
in the coherence conditions. P YTHIA uses the virtuality along the emission lines as an
ordering variable and only afterwards introduces an angular ordering condition.
In this paper we will mainly describe the emerging features in terms of the notions of
the dipole cascade model (A RIADNE) and only perform a cursory comparison to similar
results from P YTHIA. Some of the ideas presented here were also discussed in [20] in
their preliminary form. Our aim is to investigate a set of distributions stemming from the
perturbative parton cascades. We will show that the cascades result in a local structure,
corresponding to sets of independent entities, which we are going to call Generalised
Dipoles (GDs). The GDs are linked together along the colour lines of the QCD force field.
They have a common distribution in a generalised rapidity range , and in the (local)
transverse momentum. This structure is independent of the total cms energy, the global
event variables like thrust and the number of hard emissions. Further the GDs show a
surprisingly small scale dependence, i.e., almost the same distributions occur inside a wide
range of the ordering variable.
We will concentrate upon e+ e -annihilation events, where the gluon cascade is
produced through the bremsstrahlung from an originally produced qq pair (although we
expect that the same structures will emerge also in other dynamical situations). Such a
multi-gluon state is conveniently described by means of a four-vector valued function, the
directrix A [5]. The directrix is obtained by laying out the energymomentum vectors,
kj , of the emitted partons in colour-order, starting with the q and ending with the q (we will
neglect the production of new qq pairs through the gluon splitting process). As we will
use massless partons, the directrix is a curve with a tangent that is light-like everywhere.
As mentioned earlier the phase space available for gluon emission at a certain value
of the ordering variable k depends on the partons present and their colour order. It
could, therefore, be regarded as a functional of the directrix A . In Section 2, we will
discuss a measure of this phase space, to be called , for multiple gluon emission in
perturbative QCD. An infrared stable generalisation of this phase space, involving a
resolution parameter m0 leads to the concept of the -measure [6].
The -measure was introduced many years ago [6], as a generalisation of the rapidity
range, to describe the phase space for hadronisation. At the same time [6], a curve similar
104
in appearance to a set of connected hyperbolae along the directrix was introduced, the X curve. Its length is related to in the same way as the ordinary rapidity variable measures
the length of a hyperbola spanned between two light-like vectors.
The tangent vector, to be called qT , at a point on the X -curve that reaches out to a
point on the directrix quickly reaches a constant length, qT2 = m20 , just like the constant
distance in Minkowski metric between the lightcones and an ordinary hyperbola. The
parameter m0 is a resolution parameter for the properties of the directrix. In the context of
hadronisation, it is fixed by the properties of the spectrum of hadrons produced. Thought
of as a generalisation of , the -measure has a resolution parameter which can be directly
related to the ordering variable in the perturbative cascade. When the tangent vector qT
follows the directrix it will span a surface between the directrix and the X -curve. The
-measure is also proportional to the area of this surface.
In the dipole model, the available phase space for gluon emission, , in the Leading Log
Approximation (LLA) [7] and in the Modified Leading Log Approximation (MLLA) [7]
schemes is a sum of a series of terms, ()j , one for each pair of consecutive vectors along
the directrix [6]. These terms are contributions to the phase space from individual dipoles
in the dipole cascade model. We will show that the -measure can be similarly subdivided
into a number of parts ()j , one for each vector along the directrix, and that they are
strongly related to the contribution from the dipoles, ()j . The process of subdividing
corresponds to a partitioning of the surface mentioned above into plaquettes1 that are
enclosed by one initial qT and one final qT vector together with a hyperbolic segment
of the X -curve and the parton energymomentum vector in between, cf. Fig. 1 in Section
2. It is the regularity in the () distribution that will lead to the definition of the GDs.
For the surface mentioned above this regularity implies a simple general structure, which
shows only a slowly varying dependence on the ordering variable.
The directrix also plays a major role in the description of the state of the massless
relativistic string, which is used as a model for the QCD force fields in string fragmentation
[5]. The surface spanned by the string during one period is a minimal surface. This has
two consequences. On the one hand the surface is completely determined by its boundary
curve. On the other hand the surface is stable against small deformations (the model is
infrared stable). In the motion of a massless relativistic string, a wave moves across the
string surface bouncing at the endpoints defined by the orbits of the q and the q . This
means that the internal excitations (which in the Lund model are identified as the gluons)
will reach and affect the endpoints in turn, i.e., in the colour order of their emission. The
corresponding orbit of the q endpoint is the directrix.
It is interesting to note that the X -curve (or rather a close relative, called the P-curve
in [8]) and the -measure play a major role in the string fragmentation process for multigluon string states. In the process that we devised in [8], the final state hadron energy
momenta, laid out in rank order, constitute a curve, the X-curve. The X-curve follows the
directrix at a typical stochastically fluctuating distance and the area in between the two
curves could be thought of as the area in the Lund model area law [8]. It turns out that the
1 In this paper, we will use the word plaquette to denote this area even though one part of the boundary curve
is a part of a hyperbola. The plaquettes defined in this paper are not related to the plaquettes used in lattice QCD.
105
average X-curve is just the P-curve with a length given by . The generalised dipoles to be
defined in this note also play an interesting role in the fragmentation process. Just as they
emerge as independent entities from the partonic cascades, they also fragment essentially
independently of each other into the final state hadrons, in a fragmentation scheme along
the directrix.
The -measure has been used in many different contexts in investigations over the years.
It was quickly found, that the X -curve shows properties of a fractal character [9] and the
multi-fractal dimensions were shown to be identical to the so-called anomalous dimensions
of QCD [10].
In Section 2, we recapitulate some features of the gluon emission phase space -measure
and the -measure with the aim of pointing out similarities and differences between the two
quantities.2 We also introduce the quantities and present a geometrical interpretation
for them. In Section 3, we describe our findings and define the generalised dipoles. In
Section 4, we present a model to understand the distributions we have obtained. At the
same time we will consider some earlier results and the theoretical analysis. Finally, in
Section 5 we make some concluding remarks.
2. The -measure and the perturbative cascades

2.1. The phase space in a dipole cascade and the -measure
The probability for the emission of gluons in the dipole model can be described in terms
of an inclusive density of gluons (cf. [1]):

dk 2
,
dn = eff dy 2
(1)
p
k

with eff the effective running coupling and ( p ) the polarisation sum, i.e., the coupling
between the spin of the emitters and the gluon. The coherence conditions in this case
are identical to energymomentum conservation, i.e., the cms energy of the gluon cannot
exceed half the total cms energy
s
.
k cosh(y)
(2)
2
A convenient approximation, corresponding to the leading log approximation, is |y| <
1
2
2
2 ln(s/k ) so that the total rapidity range is y ln(s/k ). As we mentioned above the
emission of a gluon will change the current. But it is an immense simplification that this
change is simply a change from one dipole to two independent dipoles (to a very good
approximation) [7,11]. Thus the density for the emission of two gluons is factorisable, so
that we obtain

dn qg1 g2 q dn qg1 q dn(qg2 g1 ) + dn g1 g2 q
(3)
2 In the past, like in the references listed in this paper, the gluon emission phase space has often been called
the -measure. But since both this phase space and the phase space for hadronisation are relevant for this work,
we make the distinction here and give them different names.
106
(with a negative correction that formally is of the order of 1/Nc2 for a finite number of
colours, but reduced further in practice due to kinematics). Eq. (3) is valid if the transverse
momenta are ordered so that k1 > k2 or else the two gluons are exchanged. While the
original (qq) dipole is at rest in the cms the two new dipoles move apart. Each of them
will decay in its own restframe with a phase space given by Eq. (2) and this means that the
combined rapidity region for emitting the second gluon with transverse momentum k is

s12
s
s01 s12
s01
.
()01 + ()12 = ln 2 + ln 2 = ln 2 + ln
(4)
2
k
k
k
sk
The expression s01 s12 /s is the conventionally defined invariant squared transverse
2 . Similarly, for an emission at k 2 from a system of n
momentum of the first gluon, k1
gluons, we can write the phase space as:

n
n
n

si,i+1
si,i+1
=
(5)
()i,i+1 =
ln
= ln
.
2
2
k
k
i=0
i=0
i=0
At this point, it is necessary to differ between two definitions of the transverse momentum.
The one we have used up to now is the ordering variable k , i.e., the value used for the
emission of a gluon and defined by the partitioning of a dipole of mass M into a pair of
2 = m2 m2 /M 2 . There is another local variable, to
dipoles with masses (m1 , m2 ) so that k
1 2
be called kt , which is defined along the directrix so that we have
ktj2 =
sj 1,j sj,j +1
(2kj 1 kj )(2kj kj +1 )
,
(2kj 1 kj ) + (2kj kj +1 ) + (2kj 1 kj +1 )
sj 1,j,j +1
(6)
in terms of the energymomentum vectors of the (colour) adjacent gluons.

We note that the two values of k and kt coincide for the last emitted gluon. But in
general the transverse momentum kt of a gluon may be much smaller than the k value at
which it was emitted, because of the recoils from subsequent emissions. In fact, in this
way kt can even become smaller than the resolution scale at which the state is being
sampled. But the emission k and the concepts of the first emitted gluon and the
second emitted gluon etc., are necessities of the strategies used to simulate this quantum
mechanical process, and not properties of the final state of partons. When one asks what the
probability is for the production of a state with a certain number of partons, where colour
connected partons with transverse momenta smaller than a certain value would not be
considered as resolved emissions, the only relevant transverse momentum is the kt , which
is calculated using vectors in the final state. Therefore, it is undesirable that this variable is
pushed to values smaller than the resolution parameter. We have used two different methods
to avoid such situations. The first is to veto such emissions along the cascade3 and the
second to change the ensuing directrix vectors so that three neighbouring light-like vectors
are combined into two. All the results of this investigation are independent of the method
that is used. Furthermore, such a procedure is essential for the inclusion of the second order
matrix element corrections into A RIADNE [12].
3 This feature will be made available in the next version of A RIADNE.
107
2.2. The -measure, its properties and its connection with the -measure
When a string without gluonic excitations fragments under the usual Lund model
assumptions, hadrons are formed by breaking of the string field at vertices which, on
the average, lie on a typical hyperbola parametrised by a squared proper time from the
origin, usually called 0 or m20 . The rapidity range available along this hyperbola is given
by ln(s/m20 ).
In the Lund model interpretation for a string with a single gluon excitation, there will be
two parts of the string; one string piece spanned between the q and the g and one between
the g and the q . Therefore there will be two hyperbolic angular ranges (y)01 and (y)12
for hadron formation. The sum of these ranges is

s12
s
s01 s12
s01
(y)01 + (y)12 = ln
(7)
+ ln
= ln
+ ln
.
2m20
2m20
m20
4sm20
We draw attention to the factor of 4 in Eq. (7), and its absence in Eq. (4). In the Lund
model, when the partons move apart stretching a string-like field between them, they lose
energymomentum to the string field. A gluon is attached to two string pieces while a quark
or an anti-quark is attached to one. Therefore a gluon will lose energymomentum twice as
quickly as a quark or an anti-quark. Flat string regions will therefore be formed, bounded by
the whole of quark or anti-quark momenta, but half of gluon momenta. Since hadrons form
from this field, the phase space is calculated by adding up the lengths of typical hyperbolae
in each of these flat string regions. For instance, the invariant mass of the region involving
the quark and the gluon, for the system in Eq. (7), is (kquark + 12 kgluon)2 = s201 , and its
contribution to the phase space for hadronisation is s012 .
2m0
Such a factor was not necessary in Eq. (4). A gluon is shared between two dipoles, but
both of those dipoles are modified by the emission of a single gluon from either of these
dipoles. This is because the gluon in common between them would receive a recoil and
that would change the invariant masses of both the dipoles. Formation of a single hadron
in string fragmentation does not give recoils to the partons making up the string. Hadron
formation occurs from the string, which has a spatial extent. Gluon emission, on the other
hand, is a perturbative phenomenon and the new emission is thought to come directly from
one or the other of the two partons making a dipole, or from the dipole as a whole regarded
as a unit. Therefore, in the implementation of the dipole cascade model in A RIADNE, the
full gluon momentum is used for both the dipoles involving the gluon. There is no problem
in conserving energymomentum because both the dipoles cannot emit simultaneously.
As soon as there is an emission from one dipole, the whole system is updated into a new
chain of dipoles.
The expression in Eq. (7) does not make sense for soft or collinear gluons, as these
would give negative contributions to the effective rapidity range. An infrared stable
generalisation applicable for arbitrarily many gluons was presented in [6] and its relation to
the hadronisation process was discussed in [13]. For a state with a single gluon we replace
the expression in Eq. (7) with:

s01 s12
s
+
.
ln
(8)
m20
4m40
108
This is a nicely interpolating expression between the results for a string with and without a
gluon. It can be generalised to an arbitrary number of gluons by introducing the following
quantities, defined as integrals along the directrix A( ) (with a suitable parameter, here
taken to be the energy along the directrix):
2j2

2

Etot
3
dA dA
dA dA
d1 d2
d3 d4
d2j 1
tj =
d1 d2
d3 d4
0
2j1
d2j
0

dA
dA
.
d2j 1 d2j
(9)
This corresponds to a 2j -fold partitioning of the directrix into non-overlapping pieces and
then to multiplying the adjacent vector differentials two by two. For a massless relativistic
string, a point on the string (parametrised by the energy ( ) between the point and the
quark end) at time t will be at x(, t):
[A(t + ) + A(t )]
,
(10)
2
i.e., it is given by one left-moving and one right-moving vector defined by points on the
directrix. As the differentials of the directrix are light-like the quantities dA( ) dA( )
occurring in the integrals in Eq. (9) are surface elements on the string. The integrals in
this way correspond to all the possibilities of obtaining j -fold partitionings into nonoverlapping areas on the string surface.
We may then define a functional T exp() as
x(, t) =
T =1+
tj
(m20 )j
j =1
(11)
i.e., as the generating function of the partitioning functionals tj in Eq. (9) with m0 an
energy scale to make it dimensionless. We note that the integrals tj will vanish if j > n + 1
where n is the number of gluons emitted. The term involving one power of m0 is always
equal to s 2 and the last (non-vanishing) term will have the generic form
2m0
1 s01 s12 s23

sn,n+1
,
2 2m20 4m20 4m20
2m20
(12)
with (j, j + 1) indices of colour-consecutive partons (index 0 for the q, indices 1, . . . , n for
gluons, and n + 1 for the q ). Besides the addition of 1 and a factor of 1/2 (both introduced
to obtain simple factorisation properties, cf. below) the T functional in Eq. (11) coincides
with the argument of the logarithm in Eq. (8).
It is useful to define a generalisation, T ( ), by exchanging the upper limit in Eq. (9)
from Etot to [6]. This functional fulfills

1
T ( ) = 1 +
0 0
dA(1 ) dA(2 )
T (2 ),
m20
(13)
109
and together with the four-vector valued functional qT :4

1
qT ( ) =
T ( )

dA T ,
(14)
we obtain the differential equations

dT =
qT dA
T,
m20
dqT = dA
qT dA
qT ,
m20

q2
dqT2 = 2 1 T2 qT dA. (15)
m0
They can be integrated to

T ( ) = exp
0

qT ( ) dA( )
,
m20

qT2 ( ) = m20 1 T ( )2 .
(16)
Finally, we define a four-vector valued function X = A qT 5 so that

dX =
(qT dA)
qT ,
m20
(17)
i.e., the vector qT is the tangent to the X -curve reaching out to the directrix. From Eq. (16)
we conclude that the functional T is the exponential of an area (note that the vector dA
is light-like, and qT dA is an area element between the directrix and the X -curve). This
area is proportional to the functional as mentioned in the introduction. We also note that
the vector qT quickly reaches the length m0 . Integrating the differential equations for a
directrix built up from a set of light-like vectors {kj } we obtain the iterative equations [6]:

Tj
Tj +1 = 1 + qTj kj +1 /m20 Tj
,
j +1
1 + j +1
kj +1 .
qTj+1 = j +1 qTj +
(18)
2
The starting values are T0 = 1 and qT0 = k0 .
The factorisability of the functional T means that its logarithm, , can be written as a
sum containing one term for each kj :
n+1
n+1

j =
ln 1 + qTj1 kj /m20 .
j =1
(19)
j =1
The j defined in this way is the size of a subarea just as the total represents the
area spanned between the X and the directrix curves. In Fig. 1 we exhibit such a region
bordered by the initial qTj1 and the final qTj together with a hyperbolic segment from
the X -curve and the gluon energymomentum vector kj from the directrix.
4 Called q in [6].
5 Called X in [6].
110
Fig. 1. A plaquette bordered by qTj , qTj+1 , kj and a hyperbolic segment from the X -curve, as described in the
text.
If we define 0 1 and the vectors kj 12 kj (1 + j ), then it is easy to see that the

vector qTj is a weighted sum of the kl vectors (l j ). The weight for any vector km is
j 1
exponentially suppressed with l=m l+1 .
To see that j as defined in Eq. (19) actually has the meaning of a rapidity we go
to the rest frame of the vector qTj1 and assume for simplicity that it has the length m0 .
Further assume that the light-like vector kj is directed along the 3-axis with the length
so that j = ln(1 + /m0 ). Then it is easy to show that the next qT -vector qTj will
have the lightcone coordinates along the 3-axis equal to (m0 exp(j ), m0 exp(j )),
i.e., the qT vector has been accelerated from rest to the rapidity j by the step kj along
the directrix. The corresponding segment of the X -curve is a part of a hyperbola.
For sufficiently well ordered emissions and a resolution scale much smaller than the
smallest transverse momentum present, the last term (Eq. (12)) will dominate in Eq. (11)
and therefore (m0 = 12 k ) will be approximately equal to (k ) defined in Eq. (5).
From the definition in Eq. (19) it is easy to show (cf. the definition of the vectors qTj in
Eq. (18)) that the argument in the logarithm of the j is equal to (1 + sj 1,j /4m20 ) plus
terms down by at least a factor of . From the definition of the local kt variables, Eq. (6),
we may also conclude that
ktj2 <
sj 1,j sj,j +1
,
(sj 1,j + sj,j +1 )
(20)
and consequently we obtain

j > ln
sj 1,j
4m20

= ln
sj 1,j
2
k

j 1,j ,
ln
ktj2
2
k

< min(j 1,j , j,j +1 ).
(21)
This implies that there is a minimum positive value of , if a procedure to exclude or

filter out emissions creating gluons with kt < k has been used.
111
3. The properties of the distributions

We will now exhibit a very noticeable regularity of the dipole cascades, cf. Fig. 2. In
this figure, we show a plot of the average
number of gluons as a function of k , n(k ),
versus (k ) for e+ e -events with s = 200 GeV and for a range of values of k from 4
down to 1 GeV. We have switched off the gluon splitting process so that we end up with
one colour connected set of partons.
There are several interesting properties of this plot:
There is a linear correlation between the average gluon multiplicity n(k ) and the
phase space variable (k ) inside a very narrow band. The fact that the different lines
obtained for different values of k are so close to each other means that the slope of
this linear correlation is rather insensitive to changes in the ordering variable in this
range. It should be noted that the (k )-distributions are widely varying for these k values as we show in Fig. 3. We have used a running coupling (with QCD = 0.22 GeV,
i.e., the default value in A RIADNE). The region with (k ) > 20 is mainly populated
by the results from small k -values, k < 2 GeV.
We find that the linear correlation in Fig. 2 is independent of the cms energy (we have
used different cms energies ranging from 40 GeV and up to even 2000 GeV, though for
small energies there are too few gluons emitted to allow detailed studies), cf. Fig. 4.
Not only do we find a linear correlation at each value of cms energy (when the number
of gluons is more than 2), the slope of the correlation varies rather weakly with respect
to energy. This property is also not a function of the total
thrust, or the presence of
particularly hard emissions (e.g., gluon of k > 7 GeV for s = 200 GeV). There is
a minimum value of 10, corresponding to no gluon emission (for k = 1 GeV and
Fig. 2. A plot of n(k ) versus (k ) for an e+ e -event with
s = 200 GeV from A RIADNE.
112
Fig. 3. The (k )-distribution for different k -values at
s = 200 GeV.
Fig. 4. A plot of n(k ) versus (k ) (k = 1.0 GeV) for different cms energies from A RIADNE.
s = 200 GeV), and there is a (small) transition region for a few more units, until the
number of gluons exceeds two. But after that dn/d is a constant.
As a consistency check we have used two procedures for sampling the events. It is
possible to run the parton cascade in A RIADNE down to a certain value of the ordering
variable k , i.e., emit all gluons with larger k s, sample the emerging directrix and
investigate its properties, e.g., calculate the number of gluons emitted to that point,
n(k ), the total length of the X -curve, (k ), the individual (m0 )s and the local
kt s. After that we may continue the cascade for the same event down to smaller
113
Fig. 5. A plot of n(k ) versus (k ) (left) for an e+ e -event with s = 200 GeV from P YTHIA. A plot
n(k ) versus (k ) (right) at s = 200 GeV, k = 1.0 GeV for coherent and incoherent showers in
P YTHIA.
k -values and do the same exercise. Another possibility is to run the cascade down
to a certain value k kcut and collect a large sample of events to investigate. After
this we can pick a new value of kcut and repeat the process. The result in Fig. 2 does
not depend upon which of the two methods for sampling the events we use.
We observe the same effect in the parton shower in P YTHIA as in A RIADNE. The
2 = z(1 z)Q2
ordering variable in P YTHIA is the parton virtuality Q2 , and the variable k
(where z is the lightcone fraction taken by the emitted parton) is used as the argument for
the running coupling. This is a good approximation for the transverse momentum in the
case of massless parton kinematics. However, since transverse momentum itself is not the
ordering variable in this case, there is no real lower limit to it. Therefore, in order to make
the comparisons we filtered the events obtained from P YTHIA and combined partons such
that all gluons had kt above a certain value proportional to Q.
It is then possible to use P YTHIA to obtain an inclusive sample of ng (CQ) and (CQ).
We just note that it is possible to tune the constant C such that, as shown in Fig. 5, there is
the same dn/d as in A RIADNE. We find that C 0.5.
It is possible to switch off the coherence conditions in the parton shower in P YTHIA,
since the strong angular ordering is implemented only after the generation using the
ordering variable Q2 . We show the effect of this feature in Fig. 5. For the same C we
obtain many more partons per unit phase space for an incoherent shower, although they
still show a linear correlation to the -measure.
The linear relation between n and can be understood as follows. If one considers a
variable such as defined as in Eq. (5) where the parts j are distributed with the averages
j and the variances j2 , then itself will mostly have a Gaussian distribution according

to the central limit theorem. The average is = j and the variance 2 = j2 . The
numbers and 2 are in general well-defined for large n unless the values of the j
are correlated. We find that the j s indeed show very little correlation, even between
adjacent dipoles, cf. Fig. 6. We also find that the average j shows only a weak dependence
upon the total number of gluons in the event, n, for a wide range of n.
114
Fig. 6. Scatter plot of two adjacent values.
We will therefore go over to the investigation of the distribution of the j defined in

Eq. (5). We notice that the first and the last term in Eq. (5) are different from the other
terms in one important aspect. They are contributions from dipoles involving a quark
or an anti-quark (with different colour charges than gluons), whereas all the other terms
come from purely gluonic dipoles. It turns out that the dipoles at the ends have slightly
different distributions than all the rest. However, if the gluon splitting process is switched
off, there are only two such dipoles in an event. This means that we can write for the
average = 2q + (n 1)gg . In Fig. 4 we see this effect clearly, because it is only
for n > 2 that we find a constant slope, i.e., when we have purely gluonic dipoles present.
The properties we noted above are about variation of the number of gluons or dipoles with
respect to (k ) and the stability of the slope of the correlations, and should correspond
to the properties of purely gluonic dipoles. Therefore, from now on we will not include
the very first and the very last -values in the investigations. We have seen that purely
gluonic dipoles, irrespective of their position along the directrix, have the same distribution
dP
P () from A RIADNE and observe
in size. In Fig. 7, we show the distribution d()
that:
The distribution P has an average and a variance 2 , consistent with the results
in Fig. 2. We have found that it is independent of the cms energy, of the global
event variables like thrust and the total -value for the event, and also of hard gluon
emissions.
The distribution P is rather insensitive to the value of k . In Fig. 8, we show the
2 /2 ) and we find that it changes about
average value of as a function of ln(k
5%, when the ordering variable goes from 0.8 to 3 GeV.
115
dP P () distribution from A RIADNE.

Fig. 7. The d()
2 /2 ). Note that the variation in the average is

Fig. 8. The average value of as a function of ln(k
much smaller than the width of the distribution of as seen in Fig. 7.
There are no noticeable correlations between adjacent s as we have noted before.

In Fig. 6, we show the scatter plot between two adjacent , taken from stochastically
chosen pairs in many
different events. In Fig. 9, we show the values of the average a(n)
(defined as a(n) nj=1 j ), and the standard deviations around it, for chains of
n connected . For this figure the statistics was collected from events with a fixed
number of gluons.
116
Fig. 9. The
average (and the standard deviations around it) for chains of n connected called a(n)
(a(n) nj=1 j ). The statistics is collected from events with a fixed number of gluons.
In order to further investigate the independence of the values in an event we

have examined events containing a fixed number (N ) of gluons. We may then ask
about the multiplicity distribution, P(N ), of events with N values of .
We find a binomial distribution with the mean Np and variance Np(1 p) where

p = 0 dx P (x) as expected from uncorrelated dipoles. We have also investigated
the average number of -values, which are consecutive and satisfy . We
find a geometric distribution, weighted with the average number of not fulfilling
the condition (which is, once again, expected when we have uncorrelated dipoles). We
also find that the distribution Pn obtained from the sum of the lengths of n consecutive
is an n-fold convolution of the distribution P .
The corresponding results in P YTHIA are very similar to the results in A RIADNE (with
the appropriate choice of the constant C as discussed above). To the left in Fig. 10, we
show the distributions from P YTHIA and note that P YTHIA is somewhat more concentrated
towards small values and also has a somewhat wider tail so that the average values are
the same as in A RIADNE. The incoherent option gives a much more narrow distribution
centred around smaller values, as expected.
It is also possible to investigate the distributions in terms of the -measure defined in
Eq. (19). In Fig. 10 to the right, we show two distributions (one from A RIADNE and one
from P YTHIA), to be called P () using as before 2m0 = k = CQ where m0 is the
argument of the -measure. Once again we find the same properties as for the distributions
P ().
We conclude that, inside the whole region relevant to the hadronisation procedure in
string fragmentation, the partonic states are dominated by this structure of independent
entities. It is tempting to consider them as a kind of collective variables for the QCD force
117
Fig. 10. The plot to the left shows the P ()-distribution from P YTHIA compared to the distribution obtained
from A RIADNE. The plot to the right shows the corresponding plot for the P () distribution.
fields. The variables represent the local size of the dipoles between adjacent parton
energymomentum vectors, whereas the values correspond to a similar although more
non-local property depending upon a set of adjacent vectors.
These properties of the distributions of s are properties of gluonic dipoles, and result
from perturbative QCD. In the next section we will make models in the LLA and the MLLA
schemes based on simple gain loss considerations on the dipole cascade model, and show
that it is possible to understand many qualitative features of these distributions.
The s on the other hand are complicated objects, and there is no obvious way to
calculate their distributions except by relying on the similarity between the -measure and
. But the fact that there are as many s as s, and the s have parallels for each
of the properties of the s discussed here, lead us to introduce the notion of Generalised
Dipoles (GD) to be the source of the contributions to the j s. Whereas a dipole is
spanned between two colour connected partons, a GD corresponds to one flat block or
plaquette in the surface traced between the directrix and the X -curve by the tangent to
the X -curve. Regularities in the properties of dipoles which are mirrored in the GDs would
be passed on to the hadronised state in a hadronisation scheme based on the -measure,
such as the one introduced in [8].
4. Models to describe the dipole distributions

4.1. A simple cascade model
Multiplicity distributions were presented in the parton cascade formalism in [14] and
in the dipole cascade formalism in [15] to leading logarithmic order. In this paper we will
follow the approach presented in [16].
In order to provide a theoretical framework for our findings in the previous section,
we will consider an analytical model for a dipole cascade. We will assume that there is a
2 /2 )
distribution, f (, ) d, that describes the dipoles with size at the scale ln(k
2
2
(where k is the ordering variable and ln(Md /k ), in terms of the dipole masses
Md ). We will consider the change in f when we make a small change in the scale . We
118

M2
note that + = ln 2d is independent of . If dipoles never decayed, the distribution

f (, ) would not change if + is kept constant as varies. Therefore any change in
the distribution along the lines + = constant must purely come from decays.
There are two contributions to this. A dipole of size may decay into two smaller
dipoles. A dipole of size bigger than may decay such that one of the daughter dipoles is
of size . Gathering these terms together we get a partial integro-differential equation, cf.
[16,17], based upon Eq. (1) with eff = 0 / for gluonic dipoles (0 = 112N6 f /Nc ):
Df

f
0
f
=
f (, ) 2 d f , .

(22)
The factor of 2 in the contribution from the larger dipoles stems from the fact that we are
only considering central or purely gluonic dipoles here. When a gluonic dipole decays,
there are 2 equivalent ways to obtain a gluonic dipole of size .
This integro-differential equation can be made into two coupled differential equations
in terms of the first two moments of f , N1 and N2
Nj (, ) = d1 (1 )j 1 f (1 , ),
0
DN1 = (N2 2N1 ),
0
DN2 = N1 + 2 N1 .
(23)
For the normalisation we note that N1 (0, ) is the total number of dipoles available at the
The
scale , to be called n in this section. N2 (0, ) is their combined length, to be called .
equations for these quantities can be obtained directly from Eq. (23) in the LLA-scheme:
0
d n
= ,
d
d
= n.
(24)
These equations are solvable in terms of combinations of the modified Bessel functions
(cf. the next subsection) as soon as we have defined a starting value, max L0 , for the
cascade.
In order to solve the coupled equations in Eq. (23), we introduce the combination
g(, ) = N2 N1 , which fulfills the equation:
Dg =
0
(g),
(25)
with the boundary value
g(0, x) = (x).
It is related to N1 through N1 (, ) =
g(, ).
(26)
Including the boundary condition above, we obtain:

+

0
dy ( + y) .
N1 (, ) =
( + ) exp
119
(27)
This is a derivative of two contributing factors. The first is evaluated at +

ln(Md2 /2 ), i.e., at the largest value where the dipole could have been produced. It is
multiplied by the probability that the dipole has not decayed until . In field theoretical
language this second contribution, the Sudakov form factor, corresponds to the sum of all
the virtual corrections during the lifetime of the dipole. This Sudakov form factor can be
easily calculated

+
0
exp(0 )
dy ( + y) =
.
S(, ) = exp
(28)
y
(1 + )0 (+)
It fulfills
S(0, ) = 1,

S
S.
(, ) = 0 ln 1 +
(29)
The derivative vanishes at = 0.

The distribution f is obtained by a partial differentiation of N1 , cf. Eq. (23)
N1
,
(30)
so that using Eqs. (24), (27) and (29) we obtain:

0 , + )0 ln(1 + /)2 + 2n(L
f (, ) = (L
0 , + )0 ln(1 + /) S(, ).
f (, ) =
(31)
The result can be understood in a simple and useful way. The logarithmic factors in the
parenthesis are equal to
+
0 ln(1 + /) =
0 dx
,
x
(32)
i.e., the probability that there will be a dipole breakup somewhere between + , the
largest virtuality where a dipole with size at can be produced, and , where it is found.
There are two such factors multiplying all interior points in the dipoles at + , counted
+ ), and there is one factor multiplying twice the number of dipoles at +
by (
(because if there is already a dipole one can keep either its left side or its right side and
produce the other side of the dipole with size by a breakup in between). Finally, the
Sudakov form factor describes the probability that there is no decay affecting the existence
of the dipole at .
In the limit when L0 ( + ), and + is significantly positive, the two functions
n and factorise and we may write in the LLA

0 , x) x/0 n(L
(L
0 , x).
n(L
0 , x) V (L0 )(x)1/4 exp 2 0 x ,
(33)
120
Fig. 11. The f () function, obtained from Eq. (31), normalised to unity.
In Fig. 11, we show the function f (), obtained from f in Eq. (31) by normalising
it to unity, for a set of values of the parameter . There is a weak dependence upon the
ordering variable and for large -values it has an essentially exponential falloff with a
slope very similar to the ones we find in the distributions P () and P () in Section 3.
For small values of there are quantitative differences. The reason is that the LLA
is not sufficient to describe the behaviourof small dipoles. In that case it is necessary to
include both the contributions from the ( p ) term in the dipole cross section in Eq. (1),
and also to take into account the recoil contributions which are particularly noticeable for
small dipoles. We will consider this question in the next subsection.
4.2. The MLLA and earlier investigations
In the so-called modified leading log approximation, subleading effects from the
polarisation
sum are included [7,18]. In general, the assumptions have been that the
cms energy s (or rather its logarithm L ln(s/2 )) is very large and that the main
contributions from the cascades will come from well-ordered emissions, i.e., that kj
kj +1 (at least for adjacent emissions).
Using this approximation, it is possible to write out analytical equations for the expected
changes in the average (to be called ())

and the average n (similarly n())
for a given
2 /2 ) [7,17,18]:
ordering variable ln(k
d
= n,
d

d n
= eff n
.
d
(34)
121
Eq. (34) are simplified versions of the modified leading log approximation (MLLA)
formulae. A more elaborate treatment should include the difference between the quark and
anti-quark endpoint dipoles with a different effective coupling and a different correction
, cf. [18], but these are small corrections if the starting value L = ln(s/2 ) is large.
The interpretation is that in a step d, changes by n contributions from the scale
change according to Eq. (5), just like the second line in Eq. (24). Further, the change in n
in that step is given by the coupling times the available phase space. This phase space is
given
by to first approximation; butthere is a correction from each dipole from the term
in
Eq.
(1). The polarisation sum p 2 near the centre of a dipole but decreases in the
p
neighbourhood of the emitters. It has been shown [18] that the suppression of emissions
because of this can be approximated by (and this is essentially the MLLA) a constant
decrease in the dipole size = 11/6 for gluonic dipoles (this number only depends upon
the triple gauge boson vertex for 3 colours). In Eq. (34) it is also assumed that the gluon
splitting process g qq is neglected and therefore eff 0 / with 0 = 6/11 for all
dipoles and not only the purely gluonic ones.
The decay of one dipole also affects the two adjacent ones due to recoils. A method to
estimate this effect was suggested in [19]. Therefore an extra loss-term is added to the
first line in Eq. (34):
d n
d
= n Cr
,
(35)
d
d
with a constant recoil correction estimated to be Cr 2.
These equations can be solved using the following combinations of the modified Bessel
functions:

I1 (x) = 2 x /2 I 2 0 x ,

I2 (x) = 20 x ( 1)/2I 1 2 0 x ,
(36)
and similarly for K1 and K2 in terms of the exponentially falling Bessel function K. Note
that they are normalised so that
I1 (x)K2 (x) + I2 (x)K1 (x) = x 1 ,
(37)
and that there are simple differential relations between them:

dI1
= I2 (x),
dx
( 1)
dI2 0
= I1 (x) +
I2 (x).
(38)
dx
x
x
Similar relations hold for exponentially falling modified Bessel function pair (K1 , K2 ).
If we start the cascade at a large value of L ln(s/2 ) we obtain, with the
boundary values = 0 and n = 1 at L, the results:

a = L1 Cr K2 (L) K1 (L) ,

b = L1 Cr I2 (L) + I1 (L) ,
n = bK2 () aI2 (),
+ Cr n = aI1 () + bK1 (),
(39)
122
using the notations = 1 + 0 ( + Cr ).

It is difficult to obtain an analytically solvable partial integro-differential equation
corresponding toEq. (22). The method of moments runs into difficulties because the
variation in the p term in the cross section (Eq. (1)) affects the integral contribution,
i.e., the gain from the decays of larger dipoles, in such a way that the equations for the first
few moments are no longer independent of higher moments. Therefore, we will directly
go over to the solution in Eq. (31) instead, and modify the ingoing terms according to the
MLLA, keeping in mind the interpretation given there.
We assume that each dipole of size can decay only in the interior excluding a region
/2 on each extreme. Then the effective phase space at ( + ) in Eq. (31) is changed from
( + ) into [( + ) n( + )]. A dipole of size is produced by a breakup first
at a point = 1 and then at 2 < 1 . The region where the no emission probability is to
be calculated in the Sudakov form factor is changed in the MLLA, first by a loss of /2 at
1 and then by a loss of a further /2 at 2 :

1

2
0 dz( + z)
0 dz( + /2 z)
exp
S1 ,2 = exp
z
z
exp
0 dz( + z)
z

=
1 2
2
0 /2
S(, ),
(40)
where S(, ) is the Sudakov form factor without the term as given in Eq. (28). Since this
is a function of 1 and 2 now, we obtain, for the contribution to f (, ) from situations
where both gluons involved in a dipole are emitted below + , the result:

f1 2 n ( + )
0 d1
1
0 d2
S1 ,2
2

4((1 + /)0 /2 1)2
= n ( + )
S(, ).
(41)
2
The fact that the first break may be either on the left or on the right side contributes a
factor of 2 in the above equation.
In the same way, for the second contribution related to the number of dipoles at + ,
i.e., the situations where only one gluon involved in a dipole is produced below the scale
+ , we obtain,
4((1 + ( /2)/)0 /2 1)
S( /2, ).
(42)
In this case there can never be a breakup until the dipole size has at least reached /2,
because one of the end points is already fixed at the virtuality + .
We note for consistency that the result f f1 + f2 has the limit f in Eq. (31)
when vanishes. In Fig. 12, we show the result of this MLLA version of the dipole size
distribution. We have used a fixed value = 11/6 but we have allowed the parameters 0
and Cr vary. The reason for allowing 0 to vary is that it is feasible to shut off the gluon
+ )
f2 = ( /2)n(
123
Fig. 12. The plot to the left shows the distribution f (, ), obtained from the MLLA approach at k = 1 GeV,
compared with the P () distribution from A RIADNE and P YTHIA (cf. Fig. 10). The plot to the right shows
the average value of as a function of , from the same distribution f (, ). In the plot to the right, we also
compare with the average value of (cf. Eq. (21)) as a function of obtained from A RIADNE and P YTHIA, and
we also include the standard deviation in the result from P YTHIA and A RIADNE. This result is obtained using
0 = 0.7 and Cr = 2 in the MLLA distribution as explained in the text.
splitting process in A RIADNE, but the running coupling will always receive contributions
from the different number of active flavours at different virtualities. From the figure to
the left, we conclude that it is in between the distributions from A RIADNE and P YTHIA for
the central dipoles when we use 0 = 0.70 (i.e., for three to four active flavours), and
Cr = 2 (according to [19]). This result is rather insensitive to Cr and it can therefore not
be used as a test for the claims in [19]. We have also checked the dependence upon the
ordering variable. In Fig. 12 to the right, we show that f also exhibits a slow change of
its average value as a function of the (relevant) ordering variable like the A RIADNE and
P YTHIA P distributions.
5. Concluding remarks
The number of gluons emitted above a certain value of the ordering variable shows an
interesting linear correlation (with Gaussian fluctuations) with the phase space available
at that scale. We have called the phase space variable the -measure in this paper to
distinguish it from the -measure, which is an infrared stable generalisation of the measure, useful in the context of hadronisation.
Both these quantities are functions of not only the resolution scale (kcut for a dipole
cascade, 0 in string fragmentation), but also of the precise geometry of the state under
consideration. Therefore they change with every new gluon emitted in the system.
The resolution parameter relevant for hadronisation is a parameter fixed by the break-up
properties of the string and the mass spectrum of the hadrons produced. The -measure is
constructed in such a way that addition of gluons at transverse momenta much smaller than
the scale 0 does not change its value. In that way it is infrared stable.
However, since the resolution parameter for the -measure is the transverse momentum
of the gluon to be emitted, the transverse momenta of all gluons already emitted must
be higher in an ordered dipole cascade. Therefore, the only kind of infrared stability
124
relevant for this measure would be the stability, when the ordering variable, which is also
the resolution parameter, is small. Technically this is not the case. However, since in this
approach we will never compute phase space available at a value of the ordering variable
below the QCD , there is an effective infrared stability for the phase space, if all local
transverse momenta ktj are required to be above k . It is easily shown that the mass of
any dipole with a gluon at one end is greater than the transverse momentum kt of that
gluon, which in turn, will be greater than QCD . Besides, we must remember that it is
kinematically not possible for dipoles of mass smaller than 2QCD to emit gluons above
QCD .
The two measures discussed here are quite strongly related. If one keeps only the last
(often the dominant) term in the argument of the logarithm in the -measure, one gets an
expression very similar to the -measure.
The linear correlation between the -measure and the number of gluons is a direct
consequence of the following facts that we have demonstrated. The -measure is the sum
of a series of terms j , one for each dipole. A given j is the contribution from one
particular dipole to the -measure, and is in some sense the size of that dipole. There
are two kinds of dipoles. The first kind involves either the quark or the anti-quark energy
momentum. The second kind involves only the gluon energymomenta. These two kinds
of dipoles have different distributions in sizes.
It turns out that the size of the dipoles do not show any significant correlations.
Therefore the -measure of a state consisting of a certain number of dipoles, which is
just the sum of the corresponding j s, is distributed like the sum of several independent
random variables, i.e., like a Gaussian with a mean that is the sum of the means of the
individual independent distributions. This shows up as a linear correlation between the
number of partons and the -measure.
More interesting than the linear relation is the stability of the slope of these lines, which
represents the number of gluons per unit phase space, with respect to changes in global
event parameters like cms energy or thrust and even with changes in the ordering variable.
It seems that parton cascades tend to form structures at one (relative) size when sizes are
measured with the -measure with a resolution parameter proportional to the ordering
variable. As we go down in the ordering variable, the running coupling should increase
the number of emissions per unit phase space. However, the phase space variable itself
scales appropriately to effectively absorb any increase of emissions per unit phase space.
We have also made another observation in this paper. Just like the -measure, the measure can be thought of as a sum of terms j , one for each vector along the directrix
in colour order. Out of the two vectors involved in a particular j , only one is a vector
along the directrix. The other vector, in a sense a cumulative variable, is a weighted average
of the vectors ki s along the directrix with i < j .
It is interesting that all the properties of the -measure we have discussed here are also
reflected in the -measure. There is a similar linear correlation to the number of gluons,
similar lack of correlation among adjacent terms and similar stability with respect to
change of global event variables and the ordering variable. Partitioning the -measure into
a series of j s corresponds to partitioning the surface spanned between the directrix and
the X -curve into a series of plaquettes or flat regions along the directrix. Each plaquette
is spanned between a parton energy momentum vector kj , a vector qTj1 (weighted average
125
of ki s along the directrix with i < j ), another vector qTj and a hyperbolic section of the
X -curve. The size of the plaquette j is the length of the section of the X -curve, and is
determined by kj and qTj1 . Since these plaquettes share so many properties with dipoles,
we will call them generalised dipoles.
We note in passing that the possibility of partitioning the contributions to the measure as contributions from connected flat regions has an interesting consequence for
the fragmentation of a string according to the Lund model area law. When formulated as a
process along the directrix [8,21], string fragmentation is closely related to the -measure
and the X -curve. Particle production in this scheme can be thought of as partitioning the
plaquettes mentioned here into smaller plaquettes, one for each hadron. Since the hadron
energymomenta will mostly come from these flat regions in the string, particles stemming
from one GD will have energymomenta aligned in a plane in Minkowski space, up to
transverse momentum fluctuations. We have found such chains of particles and examined
their properties. These chains, which we will call coherence chains are important for the
study of BoseEinstein correlations in the Lund model for multi-gluon systems, which we
will elaborate on in a forthcoming publication.
We have also presented here an analytical model for the observations we have made
about dipoles, in the LLA and MLLA schemes. There are limitations to this approach,
since complications such as the polarisation sum make it very difficult to obtain differential
equations which can be solved analytically. Nevertheless, our analysis based on simple
gainloss considerations broadly reproduces the qualitative features of the observations.
Acknowledgement
We would like to thank G. Gustafson, T. Sjstrand and L. Lnnblad for helpful
discussions.
This work was completed by S. Mohanty and F. Sderberg, about two months after the
untimely death of our collaborator Bo Andersson.
References
[1] G. Gustafson, U. Pettersson, Nucl. Phys. B 306 (1988) 746.
[2] L. Lnnblad, Comput. Phys. Commun. 71 (1992) 15.
[3] G. Corcella, I.G. Knowles, G. Marchesini, S. Moretti, K. Odagiri, P. Richardson, M.H. Seymour, B.R.
Webber, J. High Energy Phys. 0101 (2001) 010.
[4] T. Sjstrand, P. Edn, C. Friberg, L. Lnnblad, G. Miu, S. Mrenna, E. Norrbin, Comput. Phys. Commun. 135
(2001) 238.
[5] X. Artru, Phys. Rep. 97 (1983) 147.
[6] B. Andersson, G. Gustafson, B. Sderberg, Z. Phys. C 20 (1983) 317.
[7] Yu.L. Dokshitzer, V.A. Khoze, A.H. Mueller, S.I. Troyan, Basics of Perturbative QCD, Edition Frontires,
1991.
[8] B. Andersson, S. Mohanty, F. Sderberg, Eur. Phys. J. C 21 (2001) 631.
[9] B. Andersson, G. Gustafson, P. Dahlqvist, Nucl. Phys. B 328 (1989) 76.
[10] G. Gustafson, A. Nilsson, Z. Phys. C 52 (1991) 553.
[11] Ya.I. Azimov, et al., Phys. Lett. B 1658 (1985) 147.
126
[12] L. Lnnblad, hep-ph/0112284.

[13] B. Andersson, P. Dahlqvist, G. Gustafson, Z. Phys. C 44 (1989) 455;
B. Andersson, P. Dahlqvist, G. Gustafson, Z. Phys. C 44 (1989) 461.
[14] Yu.L. Dokshitzer, V.S. Fadin, V.A. Khoze, Z. Phys. C 15 (1982) 325;
A. Bassetto, M. Ciafaloni, G. Marchesini, A.H. Mueller, Nucl. Phys. B 207 (1982) 189.
[15] B. Andersson, P. Dahlqvist, G. Gustafson, Phys. Lett. B 214 (1988) 604.
[16] B. Andersson, Nucl. Phys. B 360 (1991) 109.
[17] B. Andersson, The Lund Model, Cambridge Univ. Press, Cambridge, 1998.
[18] G. Gustafson, Nucl. Phys. B 392 (1993) 251;
G. Gustafson, M. Olsson, Nucl. Phys. B 406 (1993) 293.
[19] P. Edn, G. Gustafson, J. High Energy Phys. 09 (1998) 015.
[20] B. Andersson, Closing talk for the XXXI International Symposium on Multiparticle Dynamics, Datong,
China, September 2001.
[21] B. Andersson, Acta Phys. Pol. B 32 (2001) 3993.

Non-commutative gauge theory of twisted D-branes

Anton Yu. Alekseev a,b , Stefan Fredenhagen c , Thomas Quella c ,
Volker Schomerus c
a Universit de Genve, Section de mathmatiques, 2-4, rue du Livre, CH-1211 Genve 24, Switzerland
b ITP Uppsala University, S-75108 Uppsala, Sweden
c MPI fr Gravitationsphysik, Albert-Einstein-Institut, Am Mhlenberg 1, D-14476 Golm, Germany
Received 17 June 2002; received in revised form 19 September 2002; accepted 25 September 2002
Abstract
In this work we propose new non-commutative gauge theories that describe the dynamics of
branes localized along twisted conjugacy classes on group manifolds. Our proposal is based on a
careful analysis of the exact microscopic solution and it generalizes the matrix models (fuzzy gauge
theories) that are used to study, e.g., the bound state formation of point-like branes in a curved
background. We also construct a large number of classical solutions and interpret them in terms of
condensation processes on branes localized along twisted conjugacy classes.
PACS: 11.25.-w; 11.25.Hf; 11.15.-q
Keywords: D-branes; Conformal field theory models in string theory; Non-commutative geometry
1. Introduction
Branes on group manifolds have received a lot of attention throughout the last years.
Even though the group manifold of SU(2)
= S 3 is the only one that can appear directly as
part of a string background (e.g., in the background of NS5-branes), other group manifolds
provide simple toy models for studying the behavior of branes in curved backgrounds. In
fact, the symmetry of group manifolds allows to control string perturbation theory beyond
the supergravity regime and hence it renders investigations of brane spectra and brane
dynamics at finite curvature possible. Moreover, group manifolds are the starting point for
E-mail addresses: anton.alekseev@math.unige.ch (A.Yu. Alekseev), stefan@aei-potsdam.mpg.de
(S. Fredenhagen), quella@aei-potsdam.mpg.de (T. Quella), vschomer@aei-potsdam.mpg.de (V. Schomerus).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 7 3 - 8
128
A.Yu. Alekseev et al. / Nuclear Physics B 646 (2002) 127157
coset and orbifold constructions and thereby allow to obtain many less symmetric string
backgrounds.
Open strings on group manifolds are modeled with the help of the WessZumino
Witten (WZW) theory on the upper half = {z 0} of the complex plane. Boundary
conditions giving rise to a consistent open string spectrum were first proposed by Cardy
[16] but their geometric interpretation in terms of brane geometry was only uncovered
much later in [9]. There it was shown that Cardys boundary theories describe branes that
wrap certain integer conjugacy classes on the group manifold G (see also [36,68]).
The BornInfeld action has been used in [10,53] to explain the stability of such branes
in a weakly curved background (see also [14] for more general cases). Once several branes
are placed on a group manifold, they tend to form new bound states. These dynamics
are described by non-commutative gauge theories on the quantized (fuzzy) conjugacy
classes. The latter were derived from the exact boundary conformal field theory in [5].
Even though these actions are applicable only at large level k, many of the condensation
processes they encode are known to possess a deformation to finite k (see [4,27]).
Branes localized along conjugacy classes are not the only ones that admit an exact
solution. In [13] Birke, Fuchs and Schweigert constructed the open string spectrum
for new sets of branes that were shown later to be localized along so-called twisted
conjugacy classes [25]. These extra series of branes come associated with non-trivial outer
automorphisms G of the group G.
In the present work we aim at understanding the dynamical properties of such twisted Dbranes on group manifolds. After a brief introduction to the geometry and conformal field
theory of maximally symmetric branes on group manifolds, we will use the microscopic
information on open string spectra to construct the space of functions on all these branes
in the limit where the level k is sent to infinity. We show that the space of functions can be
equipped with an associative and non-commutative product. For non-trivial twists G = id,
the geometry of these branes is not just a simple matrix geometry as in the case of branes
on conjugacy classes but is based on a certain algebra of matrix valued functions on the
group. The associated gauge theories are then easily obtained by copying the computations
from [5]. In the final section we analyze classical solutions. Most importantly, we shall
establish that for a given twist G , all the branes appear as bound states from stacks of
one distinguished elementary twisted brane. This is analogous to the situation of branes
wrapping ordinary conjugacy classes which can all be built from condensates of pointlike branes [6] (see also [38,40]) by some analogue of Myers dielectric effect [52]. Let
us remark that the new condensation processes we find are consistent with the charge
conservation laws proposed in [27].
2. D-branes on group manifolds

This section is devoted to the description of maximally symmetric branes on group
manifolds. Following [9,25], we will begin with a brief review of their classical geometry.
Then, in the second subsection, we shall present some basic results on the boundary
conformal field theory of such branes.
129
2.1. The geometry of branes on group manifolds

Strings on the group manifold of a simple and simply connected group G are described
by the WZW-model. Its action is evaluated on fields g : G taking values in G and it
involves one (integer) coupling constant k, which is known as the level. For our purposes
it is most convenient to think of k as controlling the size (in string units) of the background.
Large values of k correspond to a large volume of the group manifold. When dealing with
open strings at tree level, the 2-dimensional world sheet is taken to be the upper half
plane = {z C | z 0}.
Along the boundary of this world sheet we need to impose some boundary condition.
Here we shall analyze boundary conditions that preserve the full bulk symmetry of the
model, i.e., the affine algebra g k . These boundary conditions are formulated in terms of the
chiral currents
J (z) = kg 1 (z, z )g(z, z ),
J (z) = k g(z,
z )g 1 (z, z ).
(1)
Note that J and J take values in the finite-dimensional Lie algebra g of the group G. Along
the real line we glue the holomorphic and the anti-holomorphic currents according to
J (z) = J (z)
for all z = z ,
(2)
where is an appropriate automorphism of the current algebra g k (see, e.g., [60]). The
choice of is restricted by the requirement of conformal invariance which means that
T (z) = T (z) all along the boundary. Here T , T are the non-vanishing components of the
stress energy tensor. They can be obtained, as usual, through the Sugawara construction.
The allowed automorphisms of the affine Lie algebra g k are easily classified. They
are all of the form
= Adg
for some g G.
(3)
Here, Adg denotes the adjoint action of the group element g on the current algebra g k .
It is induced in the obvious way from the adjoint action of G on the finite-dimensional
Lie algebra g. The automorphism does not come from conjugation with some
element g. More precisely, it is an outer automorphism of the current algebra. Such outer
automorphisms = come with symmetries of the Dynkin diagram of the finitedimensional Lie algebra g. One may show that the choice of and g G in Eq. (3)
exhausts all possibilities for the gluing automorphism (see, e.g., [43]).
So far, our discussion of the admissible types of gluing automorphisms has been fairly
abstract. It is possible, however, to associate some concrete geometry with each choice
of . This was initiated in [9] for = id and extended to non-trivial symmetries = id in
[25] (see also [36,68]).
Let us assume first that the element g in Eq. (3) coincides with the group unit g = e.
This means that = = is determined by alone. The diagram symmetry
induces an (outer) automorphism of the finite-dimensional Lie algebra g through the
unique correspondence between vertices of the Dynkin diagram and simple roots. After
exponentiation, furnishes an automorphism G of the group G. One can show that
the gluing conditions (2) force the string ends to stay on one of the following -twisted
130
conjugacy classes

Cu := huG h1 h G .
The subsets Cu G are parametrized by equivalence classes of group elements u where
the equivalence relation between two elements u, v G is given by u v iff v Cu .
Note that this parameter space U of equivalence classes is not a manifold, i.e., it contains
singular points.
To describe the topology of Cu and the parameter space U (at least locally), we need
some more notation. By construction, the action of on g can be restricted to an action
on the Cartan subalgebra T . We shall denote the subspace of elements which are invariant
under the action of by T T . Elements in T generate a torus T G. One may
show that the generic -twisted conjugacy class Cu looks like the quotient G/T . Hence,
the dimension of the generic submanifolds Cu is dim G dim T and the parameter space
has dimension dim T almost everywhere. In other words, there are dim T directions
transverse to a generic twisted conjugacy class. This implies that the branes associated with
the trivial diagram automorphism = id have the largest number of transverse directions.
It is given by the rank of the Lie algebra.
As we shall see below, not all these submanifolds Cu can be wrapped by branes on
group manifolds. There exists some integrality requirement that can be understood in
various ways, e.g., as quantization condition within a semiclassical analysis [9] of the
branes stability [10,14,53]. This implies that there is only a finite set of allowed branes
(if k is finite). The number of branes depends on the volume of the group measured in
string units.
Let us become somewhat more explicit for G = SU(N). The simplest case is certainly
N = 2 because there exists no non-trivial diagram automorphism . The conjugacy classes
Cuid are 2-spheres S 2 S 3
= SU(2) for generic points u and they consist of a single
point when u = e in the center of SU(2). More generally, the formulas dim SU(N) =
(N 1)(N + 1) and rank SU(N) = (N 1) show that the generic submanifolds Cuid have
dimension dim Cuid = (N 1)N . In addition, there are singular cases, N of which are
associated with elements u in the center ZN SU(N). The corresponding submanifolds
Cuid are 0-dimensional. Note that all the submanifolds Cuid are even-dimensional. Similarly,
the generic manifolds Cu for the non-trivial diagram symmetry have dimension
dim Cu = (N 1)(N + 1/2) for odd N and dim Cu = N 2 N/2 1 whenever N is even.
For some exceptional values of u, the dimension can be lower. For a complete discussion
we refer the reader to [69].
So far we restricted ourselves to = being a diagram automorphism. As we stated
before, the general case is obtained by admitting an additional inner automorphism of
the form Adg . Geometrically, the latter corresponds to rigid translations on the group
induced from the left action of g on the group manifold (see, e.g., [61,62]). The freedom
of translating branes on G does not lead to any new charges or to essentially new physics
and we shall not consider it any further, i.e., we shall assume g = e in what follows.
131
2.2. The conformal field theory description

The branes we considered in the previous subsection may be described through an
exactly solvable conformal field theory. In particular, there exists a detailed knowledge
about their open string spectra based on the work of Cardy [16] and of Birke, Fuchs and
Schweigert [13].
We shall use , , . . . Bk to label different conformal boundary conditions of the
boundary conformal field theories associated with the gluing conditions Eq. (2) on the
currents. The set Bk depends on the choice of the diagram automorphism and on the
level k. For the trivial diagram automorphism = id, Bkid = Pk+ coincides with the set of
primaries of the affine KacMoody algebra g k . The latter is well known to form a subset in
the space P + of dominant integral weights which label equivalence classes of irreducible
representations for the finite dimensional Lie algebra g. To keep notations simple, we will
not distinguish in notation between elements of P + and Pk+ and denote them both by
capital letters A, B, C, . . . .
The automorphism generates a map k : Pk+ Pk+ . In fact, given an irreducible
representation of g, we can define another representation by composition . The class
of is independent of the choice of [ ] and so we obtain a map : P + P + .
The latter descends to the subset Pk+ P + . A weight A Pk+ is said to be (-)symmetric
if it is invariant under the action of , i.e., if (A) = A. According to the results of [13,
16], the labels for branes associated with the diagram automorphism take values in
a certain subset Bk B = P (P + ) of dominant fractional symmetric weights. Here
P = N1 (1 + + N1 ) is the projection of weights onto their (-)symmetric part
with N denoting the order of . Finally let us briefly mention that the number of boundary
conditions Bk is equal to the number of symmetric weights in Pk+ . On the other
hand, Bk = P (Pk+ ) in contrast to what one might expect naively. We will not need
these details here and refer the interested reader to [13] (see also [58]). Our considerations
below will mostly take place in the limiting regime of large level k where we can identify
= B = P (P + ).
B
Before we continue to outline the conformal field theory results let us briefly summarize
some conventions we will be using in the limit k . In this case, the fractional
symmetric weights which label boundary conditions can be described explicitly by

B = =

i i i = (i) , li i N0 .
(4)
The numbers li denote the length of the orbit of the fundamental weights i under the
automorphism . A distinguished element of this set of fractional symmetric weights is
= (1/ l1 , 1/ l2 , . . .) B . It can be considered as a twisted counterpart of the Weyl
vector = (1, 1, . . .) P + . Under the assumption of infinite level k, the representations
of g and g k are both labeled by the same set P + . With the identifications of this section in
mind, we will always assume that we fix the weights before we let the level run to infinity.
Our main goal now is to explain the open string spectra that come with the maximally
symmetric branes on group manifolds. For a pair of boundary labels , Bk associated
132
with the same diagram automorphism , the partition function is of the form

Z
(q) =
nA A (q).
(5)
APk+
Here, A (q) denote the characters of the current algebra g k and , Bk . Their
appearance in the expansion (5) reflects the fact that all the (twisted) conjugacy classes
admit an obvious action of the Lie group G by (twisted) conjugation. Consistency requires
the numbers (nA ) to be non-negative integers.
There exists a very simple argument due to Behrend et al. [11,12] which shows that the
matrices (nA ) give rise to a representation of the fusion algebra of g k . This means that
they obey the relations

(6)
nA nB =
NAB C nC ,
CPk+
Bk
where NAB C are the fusion rules of the current algebra g k . The argument of [11,12] starts
from a general ansatz for the boundary state assigned to Bk . Using world sheet duality,
one can express the matrices nA in terms of the coefficients of the boundary states and the
modular matrix S for the current algebra g k . The general form of this expression is then
sufficient to check the relations (6) (see [11,12] for details).
The expression for the matrices nA that is given in [13,33] resembles the familiar
Verlinde formula,
nA
S S

S
A
Pk+
S 0
for , Bk and A Pk+ .
(7)
()=
are indexed by two -symmetric labels

It contains a unitary matrix S whose entries S
+
Pk , Bk , i.e., they obey () = and () = . When = id, the matrix S

coincides with the usual S-matrix so that Verlindes formula [70] implies

nA = NA for all , , A Pk+ = Bkid.
This reproduces Cardys results on the boundary partition functions [16]. For non-trivial
automorphism , the matrix S describes modular transformations of twined characters.
An explicit formula for S can be found in [13]

2i
S
(8)
w( + ), + .
) (w) exp
k + g
wW
Here, W W is the -invariant part in the Weyl group of g. As W can be considered as

the Weyl group of another Lie algebra [32] it comes with a natural sign function ) .
There exists a generalized state-field correspondence that associates to every highest
weight state A Pk+ in the Hilbert space (5) of the (, ) boundary conformal field theory
()
a boundary primary field A living between boundaries , Bk . The general structure
of the boundary operator product expansion (OPE) is given by

()
( )
( ) C ( )
A (x) B (y)
(x y)hC hA hB CAB
C (y) for x < y,
133
(9)
where the numbers hA denote the conformal weights of the fields which, in the case at
hand, are given by
hA =
CA
.
2(k + g )
(10)
Here, CA is the quadratic Casimir of the representation A Pk+ (see Eq. (23) below)
and g denotes the dual Coxeter number of g. For a consistent conformal field theory,
( )C
the structure constants CAB
have to satisfy so-called sewing constraints [45] (see also
[56,63,64]). One of these constraints expresses the associativity of the OPE,
( )H ()D + ()0
C
( )E ( )E + ( )0
,
CAB
(11)
CCD
CEE + =
CBC
CAH
CD + D FH E B
AD
H
where the symbol F denotes the fusing matrix. For the g k -WZW model, this fusing-matrix
is closely related to the 6j -symbol of the corresponding quantum group at (k + g )th root
of unity. In the limit k , it thus reduces to the 6j -symbol of g. Solutions of the sewing
constraints for the standard case G = id can be found in [5,12,25,63]. The boundary OPE
for non-trivially twisted branes, on the other hand, is not yet known. Our results below
implicitly contain these solutions in the limit k .
The spectrum of ordinary conjugacy classes can be explained in detail. For simplicity,
we shall restrict to G = SU(2). In this case, generic conjugacy classes are 2-spheres and
j/2
the space of functions thereon is spanned by spherical harmonics Ym , |m| j/2 and
1
j = 0, 2, 4, . . . . The space of spherical harmonics is precisely reproduced by ground
states in the boundary theory when we send (and hence k) to infinity. For finite ,
the angular momentum j is cut off at a finite value j = min(2, 2k 2) 2. This
means that the branes world-volume is fuzzy, since resolving small distances would
require large angular momenta. The relation between branes on SU(2) and the familiar
non-commutative fuzzy 2-spheres [41,46] was fully analyzed in [5] and it provides a very
important example of an open string non-commutative geometry that goes beyond the
familiar case of branes in flat space [17,21,65]. The analysis of [5] goes much beyond
the study of partition functions as it employs detailed information on the operator product
expansions of open string vertex operators based on [63]. Using the results in [2325] it
is easy to generalize all these remarks on ordinary conjugacy classes to other groups (see
also [41] for more details and explicit formulas on fuzzy conjugacy classes).
Twisted conjugacy classes are more difficult to understand. This is related to the fact that
they are never small. More precisely, it is not possible to fit a generic twisted conjugacy
class into an arbitrarily small neighborhood of the group identity unless the twist is
is not cut off before it
trivial. This implies that the spectrum of angular momenta in Z
1 To be consistent with our treatment of other groups below, we use a convention in which the representations
are labeled by Dynkin labels rather than spins.
134
reaches the obvious large momentum cut-off that is set by the volume of the group, i.e., by
the level k. For large Bk (and large k) the ground states in the boundary theory span the
space of functions on the generic twisted conjugacy classes Cu [25]. The non-commutative
geometry associated with twisted conjugacy classes with finite , however, was unknown
and it is the main subject of the next section.
3. Non-commutative geometry and gauge theory

The existing information on the open string spectra of twisted D-branes on group
manifolds can be turned into a proposal for the algebra of functions on these branes. The
algebras turn out to be non-commutative due to the presence of a non-vanishing B-field.
In the second subsection we shall spell out gauge theories that encode the dynamics of
twisted D-branes.
3.1. The world-volume algebra
According to the procedure suggested in [5,65] (see also [29,30] for similar proposals
in case of closed strings), the world-volume geometry of branes can be read off from
the correlators of boundary operators in the decoupling regime k .2 Note that the
conformal dimensions (10) of the boundary fields vanish in this limit so that the operator
product expansion (9) becomes independent of the world sheet coordinates. In particular,
all conformal families in the boundary theory contribute to the massless sector and thus to
the gauge theory which governs the low energy dynamics of the D-branes we are studying.
The program we have just sketched has been carried out successfully for the untwisted
branes on compact group manifolds. In case of A1 , this leads to the well-known fuzzy
spheres [5,46]. We will now describe the generalization to arbitrary -twisted D-branes on
compact simply-connected simple group manifolds G. Note that all simple Lie groups of
ADE type (except from A1 ) admit non-trivial outer automorphisms.
In Section 2.2 we explained that all possible -twisted D-branes are labeled by the
set B of fractional symmetric weights of g. In this section we propose an alternative in
which the same boundary conditions are labeled by representations PG+ of the invariant
subgroup G = {g G | (g) = g}. Let us stress that we use representations of the group
G and not of its Lie algebra. The two sets of representations agree only if G is simplyconnected. This is the case for all Lie algebras but the A2n series where G = SO(2n + 1)
and PG+ = {A PG+ | An even}. An overview over the relevant groups and Lie algebras
can be found in Table 1. It is far from obvious that such a labeling of twisted boundary
conditions through representations of the invariant subgroup exists. But as we explain in
Appendix A one can indeed construct a structure preserving map : PG+ B between
the set of irreducible representations of G and the boundary labels for -twisted branes.
2 When taking the limit k , we keep the open string data stable. In particular, the low energy spectrum of
open string modes does not change. A limit where the closed string data is kept stable instead has been considered
in [25].
135
Table 1
Simple Lie algebras, groups and data related to outer automorphisms
g
Order
xe
A2
A2n1
A2n
D4
Dn
E6
2
2
2
3
2
2
A1
Cn
Bn
G2
Bn1
F4
4
1
2
1
1
1
SU(3)
SU(2n)
SU(2n + 1)

Spin(8) = SO(8)

Spin(2n) = SO(2n)
6
E
SO(3)
Sp(2n, C) SU(2n)
SO(2n + 1)
2
G

Spin(2n 1) = SO(2n
1)
4
F
For this reason, we will henceforth use both kinds of labels a = 1 (), b = 1 (), . . .
equivalently when referring to boundary conditions.3
Once the existence of is established, we can formulate our proposal for the worldvolume algebra. To this end, let Va , Vb be two representation spaces for irreducible
representations a, b PG+
= B of G . As we will argue below, the relevant algebraic
structure governing strings stretching between two D-branes of type a and b, respectively,
is given by

A(a,b)
(12)
= InvG F (a,b) , where F (a,b) := F (G) Hom(Va , Vb ).
Here F (G) denotes the algebra of functions on the group G and Hom(Va , Vb ) is the vector
space of linear transformations from Va to Vb . The auxiliary space F (a,b) can be regarded
as a vector space of matrix valued functions on the group G. It carries an action of the
product group G G defined by
A(g,h) (g ) = Rb (h)A(g 1 g h)Ra (h)1 ,
(13)
where Ra (h) GL(Va ), Rb (h) GL(Vb ) are the representation matrices of h. In our
construction of A(a,b) , we restrict to matrix valued functions InvG (F (a,b) ) which are
invariant under the action of {id} G G G . Let us note that this leaves us with
an action of G on the space A(a,b) of G invariants. The latter will become important later
on.
We can realize the G-module A(a,b) explicitly in terms of G -equivariant functions on
the group G,

A(a,b)
(14)
= A F (a,b) A(gh) = Rb (h)1 A(g)Ra (h) for h G .
When the two involved representations are trivial, i.e., a = b = 0, elements of A(a,b) are
simply invariant under right translations with respect to G G.
There exists more structure on the spaces A(a,b) if a = b. In fact, A(a) = A(a,a) inherits
an associative product from the pointwise multiplication of elements in F (a,a). This turns
the subspace A(a) of G -invariants into an associative matrix algebra.
The constructions we have outlined so far may easily be generalized to arbitrary
superpositions of D-branes. To this end we replace the irreducible representations Va , Vb
3 The map specifies the way in which we treat the boundary labels while sending k to infinity. We refer the
reader to Appendix A for details.
136
Qa Va .
in (12) by reducible ones. Let V
Q be such a reducible representation, i.e., VQ =
It represents a superposition of Qa D-branes in which Qa branes of type a PG+
= B
are placed on top of each other. Strings ending on such a brane configuration Q give
rise to an algebra A(Q) = A(Q,Q) analogous to (12). For a stack of N identical branes
of type a PG+
= B , the constructions specialize and produce the typical ChanPaton
factors,

AN(a)
(15)
= InvG F (a,a) Mat(N).
Obviously, the left translation of the group G turns this into a G-module with trivial action
of G on Mat(N).
This concludes the formulation of our proposal for the algebra of functions on twisted
D-branes. There exist two different kinds of evidence which we can use to motivate and
support our claim. Let us begin with a simple semiclassical argument. It is not difficult to
see that a twisted conjugacy class close to the twisted conjugacy class of the group unit
can be represented in the form C = G G C . Here, C is a conjugacy class of G close
to the group unit and G acts on G by multiplications on the right. In other words, C can
be considered as a bundle over G/G with fiber C . In the k limit, we keep C small
by rescaling its radius. As discussed in [5], C then turns into a co-adjoint orbit. At the
same time, the volume of G/G grows with k so that the corresponding Poisson bi-vector
scales down. Hence, the G/G part of C becomes classical. After quantization, we obtain
a bundle with non-commutative fibers. While the co-adjoint orbits turn into Hom(Va , Va ),
the base G/G stays classical in agreement with our claim for a = b.
More substantial support comes from the exact CFT results described in the last section.
In particular, we shall confront the formula (14) with the CFT-spectrum of boundary
fields (5). Before we carry out the details, let us note that the sewing constraints (11) are
automatically satisfied by our construction if we manage to show that the spectra match.
In fact, associativity is manifest in our proposal and it is the only content of the sewing
constraints when we send the level k to infinity.
Hence, it remains to discuss the spectrum of open strings. The CFT description
provides an expression Eq. (5) for the spectrum of strings stretching between D-branes of
type , Bk in terms of characters of g k which explicitly shows the G-module structure
of the space of ground states that emerges in the limit k . We claim that in this limit,
the G-module of ground states is isomorphic to the G-module A(a,b) where a = 1 ()
and b = 1 () are the pre-images of the boundary labels and under the map
: PG+ Bk (see above). We will prove this by decomposing A(a,b) into irreducibles.
To do so, let us note that there is a canonical isomorphism Hom(V , W )
= V W.
Furthermore, we may apply the PeterWeyl theorem to decompose the algebra F (G) with
respect to the regular action of G G into

F (G)
UA UA ,
=
A
where A runs over all irreducible representations of G and the two factors of G G act on
the two vector spaces UA , UA , respectively. To make contact with our definition of A(a,b) ,
we have to restrict the right regular G action to the subgroup G , which leaves us with the
137
(G G )-module

F (G)
bA c UA Vc .
=
A,c
The numbers bA c N0 are the so-called branching coefficients which count the multiplicity
of the G -module Vc in UA . Combining these remarks we arrive at

F (a,b)
bA c UA Vc Va Vb .
=
A,c
It remains now to find the invariants under the G -action. Note that G acts on the last
three tensor factors. The number of invariants in the triple tensor product of irreducible
representations is simply given by the fusion rules Ncb a of G . Hence, as a G-module, we
have shown that

A(a,b)
bA c Ncb a UA .
=
A,c
This decomposition is now to be compared with the formula (5) for the CFT partition
functions. A careful analysis shows that both decompositions agree in the limit k
provided one uses the appropriate identification map , i.e.,
c
nA =
(16)
bA Nca b ,
c
where the indices are related by = (a) and = (b). The proof is quite technical and
not relevant for the further developments in this paper. It can be found in Appendix A. Let
us only mention at this place that our proof takes some benefit of results found in [57,58].
As a simple cross-check we consider the case of trivial automorphism = id where we
can make contact to well-known results (see [5]). First we observe that the construction
above simplifies considerably since G
= G. This implies that all the lower case labels
can be replaced by capital letters. In particular, the boundary conditions are now labeled by
representations of G itself which is a well established fact. The corresponding G-module
structure is now given by

NA+ B C UC .
A(A,B)
=
C
This is in complete agreement with the known CFT results in Cardys case [16].
We close this section with an interesting side-remark. Since we were interested in the
analysis of twisted branes, our presentation has focused on the subgroup G in G. The
right-hand side of our central formula (16), however, gives rise to an infinite-dimensional
analogue of a NIM-rep4 for the fusion algebra5 of the Lie group G and any choice of a
subgroup H > G. This has been discussed extensively in [57] where also the connection
of the k limit of NIM-reps for twisted boundary conditions in WZW models to
4 Non-negative Integer valued Matrix Representation.
5 The latter is obtained from the fusion algebra of the WZW model in the limit k .
138
these NIM-reps has been established. Recently, new explicit expressions for NIM-reps of
twisted D-branes have been proposed in [35,55] for finite level k. They bear some similarity
with our formula (16) but generically do not involve branching coefficients of the invariant
subgroup G . Where they do, i.e., for the An -series in [55] as well as the Dn -series and
the A2n -series in [35], they reduce to relation (16) in the limit k . It seems to us that
so far a systematic understanding of the proper choice for the relevant subgroup is only
available at infinite level.
3.2. The non-commutative gauge theory
Equipped with the exact solution of the boundary WZW model in the limit k , we
are finally prepared to calculate the low-energy effective action for massless open string
modes. Compared to the case of D-branes in flat space with background B-field which
leads to a YangMills theory on a non-commutative space [66], there are two important
changes in the computation. First, the non-commutative MoyalWeyl product gets replaced
by the product of G -equivariant matrix valued functions (15) described in the previous
subsection. Moreover, there appears a new term f J in the operator product expansion
of the currents. This term leads to an extra contribution of the form f A A A in
the scattering amplitude of three massless open string modes. Consequently, the resulting
effective action is not only given by YangMills theory on a non-commutative space but
also involves a ChernSimons like term.
For N branes of type a PG+
= B on top of each other, the fields A (g) are elements
of AN(a) , i.e., they are functions on G with values in End(Va ) Mat(N) and equivariance
property as formulated in Eq. (14). We denote the dimension of Va by da . The results of
the computation [6] may easily be transferred to the new situation and can be summarized
in the following action

i

2
1
CS ,
SN(a) = SYM + SCS = 2
(17)
tr F F
tr f
k da N 4
2
where we defined the curvature form F by the expression
F (A) = iL A iL A + i[A , A ] + f A
(18)
and a non-commutative analogue of the ChernSimons form by

1
i
CS (A) = L A A + A [A , A ] f A A .
3
2
Gauge invariance of (17) under the infinitesimal gauge transformations
A = iL + i[A , ]
for AN(a)
(19)
(20)
follows by straightforward computation. The operator L is the usual Lie derivative as

defined in (25) below. Note that the mass term in the ChernSimons form (19) guarantees
the gauge invariance of SCS . On the other hand, the effective action (17) is the unique
combination of SYM and SCS in which mass terms cancel.
In contrast to earlier work, the trace is now normalized by tr(id) = da N . Moreover, we
use conventions in which only pure Lie algebraic quantities appear. The remaining part of
139
this section is devoted to presenting these conventions which follow [31]. Indices are raised
and lowered using the Killing form
(x, y) = tr(adx ady ),
1
1
T , T = f f .
I
I
(21)
Here, IB denotes the Dynkin index of a representation B P + and T are the generators
of g satisfying

T , T = if T .
(22)
For the adjoint representation of g entering (21), the Dynkin index may also be expressed
by the quadratic Casimir and by the dual Coxeter number according to I = C = 2g . The
quadratic Casimir and the index of an arbitrary representation B P + with dimension dB
are given by
CB = (B, B + 2) and
IB =
dB
CB .
dim g
(23)
It will be of particular importance for us to evaluate the quadratic Casimir in a given

representation B P + according to

tr RB T RB T = IB .
(24)
Finally, let us note that the G-action on the algebra AN(a) allows us to define the derivatives

1 d it T
L A(g) =
(25)
Ae
g
i dt
t =0
for A AN(a) . These Lie derivatives appear in the construction of the action above and
they satisfy the same Lie algebra relations as in Eq. (22).
For a calculation in the next section we will also need derivatives on functions K : G
Mat(N) which are defined via the group action on the argument from the right,

1 d it T
LR K(g) =
(26)
K ge
.
i dt
t =0
The connection between the different derivatives (25), (26) is given by the adjoint
representation of G,

L K(g) = Ad g 1 LR K(g).
(27)
4. Condensates of branes on group manifolds
The study of classical solutions of the non-commutative gauge theories constructed
in the previous sections provides insights into the dynamics of twisted branes on group
manifolds. In the first subsection we shall exhibit a rather large class of solutions. The
symmetry preserving solutions are then interpreted geometrically in the second subsection.
In particular, we shall see that all twisted D-branes appear as bound states from stacks of a
distinguished elementary twisted brane.
140
4.1. Classical solutions

A simple variation of the action (17) with respect to the gauge fields allows us to derive
the following equations of motion,

L F + A , F = 0,
(28)
which express that the curvature has to be covariantly constant. Note that the space of
gauge fields, and hence the equations of motions, depends on the brane configuration we
are looking at.
Solutions to Eq. (28) describe condensation processes on a brane configuration Q which
drive the whole system into another configuration Q . To identify the latter, we have two
different types of information at our disposal. On the one hand, we can compare the
tension of D-branes in the final configuration Q with the value of the action SQ (A) at
the classical solution A(g). On the other hand, we can look at fluctuations around the
chosen solution and compare their dynamics with the low energy effective theory SQ of
the brane configuration Q . In formulas, this means that
!
SQ (A + A) = SQ (A) + SQ (A)
with SQ (A) = ln
gQ
.
gQ
(29)
The second requirement expresses the comparison of tensions in terms of the g-factors of
the involved conformal field theories (see, e.g., [6] for more details). All equalities must
hold to the order in (1/k) that we used when constructing the effective actions. We say that
the brane configuration Q decays into Q if Q has lower mass, i.e., whenever gQ < gQ .
In terms of the world-sheet description, each classical solution of the effective action is
linked to a conformal boundary perturbation in the CFT of the brane configuration Q.
Adding the corresponding boundary terms to the original theory causes the boundary
condition to change so that we end up with the boundary conformal field theory of another
brane configuration Q . Recall, however, that all these statements only apply to a limiting
regime in which the level k is sent to infinity.
We are especially interested in processes that connect maximally symmetric configurations of D-branes, i.e., those configurations for which the spectrum decomposes into
representations of g k . For such configurations we know the low energy effective actions so
that we are able to compare them with the dynamics of fluctuations. We shall argue that
solutions possessing a G-symmetric fluctuation spectrum are associated with gauge fields
A(g) whose curvature is proportional to the identity, or equivalently
[F (A), ] = 0.
(30)
The G-module structure of the fluctuation spectrum around the solution can be characterized by the simple rule

Q
LQ
:= L + [A , ].
(31)
Q
Our notation suggests that we want to think of L as derivative associated with some new
brane configuration Q . Note, however, that by construction it acts on the space A(Q) of
gauge fields on Q. To check our claims, we must show that the derivatives (31) satisfy
141
the Lie algebra commutation relations and that the action for the fluctuation fields involves
these new derivatives rather than LQ . With the ansatz (31) it is easy to show that
Q Q

L , L = if LQ
i F (A), .
Furthermore, the expansion of the action around the solution A(g) reads

i
SQ (A + A) = SQ (A) + SQ (A) + tr F (A), A A .
2
In a slight abuse of notation we have denoted the action for the fluctuation fields A
A(Q) by SQ . For the moment, this only means that all derivatives LQ are replaced by

LQ . As one reads off from the previous equation, gauge fields satisfying (30) do indeed
lead to a G-symmetric fluctuation spectrum. A more careful investigation shows that the
converse is also true, which is to say, any solution which produces a symmetric fluctuation
spectrum is necessarily of the form we have described. The crucial observation in proving
this statement is that the relation (31) between the derivatives is dictated by comparing
third-order terms in (29).
Before we dive into the discussion of these symmetric solutions, let us briefly mention
some solutions of (28) which break the G-symmetry. One particular type of non-symmetric
solution can be obtained for each choice of a subalgebra h in g. We label the subalgebra
generators by i, j, . . . , the directions orthogonal to h by r, s, . . . . Now let R be a dR dimensional representation of h. Then the equations of motion for a stack of dR identical
branes possess a constant solution of the form Ai = id R(Ti ) and Ar = 0, where id is the
identity matrix in the spacetime degrees of freedom. The curvature of these solutions
is given by Fij = 0, Fir = 0 and Frs = frs i R(Ti ). Obviously, this solution does not
satisfy (30) and hence it breaks the G-symmetry. One can identify solutions of this type
with some of the symmetry breaking boundary theories that were obtained in [59]. The
associated boundary states preserve the affine Lie algebra of H along with the coset chiral
algebra for G/H .
There is another more trivial class of constant solutions which break the G-symmetry.
In fact, any set of commuting constant matrices A is a solution to (28). The corresponding
curvature is given by F = f A , so that in general [F , ] = 0. The solutions have
vanishing action and allow for an easy geometric interpretation [6,61]: they describe
translations of the branes in the group manifold. It is not surprising that these translations
break the symmetry, unless the whole stack is translated at once. In the latter case, the
fields A and thus F are multiples of the identity as it should be the case for a symmetric
solution. Let us observe that uniform translations on the group manifold are a rather trivial
operation, and we are only interested in analyzing solutions up to translations. One can
show that it suffices to study solutions with traceless fields A(g). From these it is possible
to generate any other solution by a uniform translation.
After this digression we come back to the study of symmetric solutions. It follows from
the last remark in the previous paragraph that there is no loss in restricting ourselves
to symmetric solutions with vanishing curvature F = 0. Let us now start with a stack
of N branes corresponding to the trivial representation 0 PG+ of G . A whole class of
solutions of (28) with F = 0 can be constructed out of a function K : G GL(N) satisfying
K(gh) = K(g)R(h)
for all h G ,
142

A (g) := (L K)(g)K(g)1 = K(g) L K1 (g)
(32)
G .
with R being an N -dimensional representation of

The first property guarantees the
necessary invariance property A(gh) = A(g) of the physical fields A(g). Moreover, the
curvature of these gauge fields vanishes because
F = iL A iL A + i[A , A ] + f A

= iL L KK1 + iL L K K1

i L KL K1 L KL K1 f L KK1
= i[L L ]KK1 f L KK1 = 0.
We should mention at this point that F = 0 is precisely the integrability condition of
the linear system L K = A K so that any solution to F = 0 is of the form (32).
The covariance law K(gh) = K(g)R(h) for h G is then necessary to ensure the
invariance property A(gh) = A(g) for h G . Consistency implies that R(h) satisfies
the representation property.
Our formula (32) for symmetric solutions leaves us with the problem to find a GL(N, C)
valued function satisfying the covariance condition K(gh) = K(g)R(h) for h G . As we
shall argue now, such a function does always exist.
To begin with, let us rephrase the existence of K in the language of vector bundles.
For a given representation R, we consider the vector bundle E G/G . Here E is the
associated bundle G G VRC which consists of pairs (g, v) along with the identification
(g, v) (gh, v R(h)). Suppose now that we are given the function K : G GL(N).
By construction, the N row-vectors ki (g) are linearly independent and satisfy ki (gh) =
ki (g)R(h). In other words, they provide a linearly independent set of global sections of E,
i.e., a global trivialization. The converse is also true. Hence, the existence of K is equivalent
to the triviality of E.
It is well known that the bundle G G VR is trivial if the representation R of G can be
obtained by restricting a representation of G. This condition is too strong for our purposes.
Note, however, that we construct our bundle E with the complexified space VRC and this
makes a huge difference. In fact, E turns out to be trivial for all representations R.
The proof proceeds in two steps. First one shows that all the bundles E are stably
equivalent to a trivial bundle. Since stable equivalence is preserved under tensoring
representations of G , one can concentrate on the fundamental representations and hence
reduce the problem to a finite number of checks which have to be performed case-by-case.
Passing from stable equivalence to isomorphism is possible if the rank of the bundle is
sufficiently large compared to the dimension of the base manifold. Again, this has to be
decided by going through all the cases separately. The details of this proof can be found in
Appendix B.
After these remarks on the existence of K, let us briefly address the issue of uniqueness.
On the face of it, formula (32) seems to provide a very large set of solutions with many
continuous parameters. For a given representation R, however, all these solutions are gauge
equivalent. To see this, let K, A and K , A be two such solutions with a covariance law
involving the same R. Then we find
A = U A U 1 (L U )U 1
(33)
143
with U = K K1 , i.e., A and A are related by a gauge transformation U (note that this
requires invariance of our theory under global gauge transformations which is granted as
long as we do not consider the automorphism of order 3). In conclusion, our discussion in
this subsection provides for each R a unique solution with vanishing curvature.
The case of K R being a representation of G can be considered as a special example
of our general construction as it may be restricted to a (reducible) representation of G .
The corresponding field A(g) turns out to be constant and it is given by

A = R T ,
where T are generators of the Lie algebra g. It is easy to see that all constant solutions
to F = 0 have this form. This class of solutions and the corresponding brane processes
have already been discussed in [27]. In the case of untwisted branes, these solutions form a
complete set of solutions for a stack of branes of type 0 PG+ . For twisted branes, however,
we just presented new non-constant solutions which give rise to new brane processes.
4.2. Interpretation of the solutions
In the last section we presented a large number of new stationary points for the action
of N = da branes of type 0 PG+
= B . Here, da is the dimension of an irreducible
representation Ra of G . We shall present evidence that such solutions correspond to
processes of the type
Stack of da branes of type 0 PG+
single brane of type a PG+ .
(34)
Any twisted D-brane can be obtained as a condensate of a stack of elementary branes of

type 0 PG+ . This implies in particular that any two configurations whose corresponding
representations of G have the same dimension can be related by a process. A similar
observation has been made for untwisted branes [6,27] and the new processes might
provide further insights to the contribution of twisted D-branes to the twisted K-groups
which seem to describe the D-brane charges on group manifolds (see [15,27,47]).
According to Eq. (29), there are essentially two checks that we must perform in order
to test the conjecture (34). We will compare the value of the action at the solution to the
CFT-prediction below and start with an analysis of the fluctuations around the solution.
The general form of the fluctuation spectrum follows from the discussion in Section 4.1
and is summarized in Eq. (29). It is implicitly contained in the G-action entering the
derivatives (31) which we used to compute the action SQ (A) of the fluctuation fields.
But as we have stressed earlier, the fields A are elements of the algebra Ada (0) . Let us be
da (0) for the G-module (Ada (0) , LQ ). Our aim then is
precise and introduce a new symbol A
da (0) with the action functional S(a) on A(a) which
to identify the functional SQ (A) on A
governs the dynamics of a brane of type (a). This comes down to providing a G-module
da (0) and A(a) . In particular, the derivatives have to match,
isomorphism between A
da (0)

L A = L(a)
(A).
Such an isomorphism can be obtained with the help of the function K from which we
constructed our solution,
(A)(g) = K1 (g)A(g)K(g) A(a) .
144

(a)
One can check that the action of the derivatives L on A(a) precisely coincides with the
da (0) ,
action of the shifted derivatives on A
da (0)

= Lda (0) + [A , ]
L
= L(a)
,
where the shift is given by our solution A = L K K(g)1 of the equations of motion.
Thus we have proved that the theory of fluctuations around our solution coincides with
S(a) as we anticipated in Eq. (34).
Our second check involves a comparison between the value of the action at the solution
and the g-factors of a CFT description. For technical reasons, we restrict ourselves to
automorphisms of order 2, thereby excluding only the case of triality for D4 . The
existence of implies strong constraints on the form of the structure constants. To be
specific, by diagonalization of , we may choose a basis in which only the constants
fij k , frsi and cyclic permutations thereof do not vanish. Here, i, j, k, . . . denote indices
for elements in the invariant subalgebra g and r, s, t, . . . label directions orthogonal to g .
We are now able to compute the action using no more than the properties of the solution
we have specified. For a solution with F = 0, the action reduces to

tr A A A
Sda (0) (A) = f
with = i 2 /6k 2 da . We now express the solution (32) in terms of the right derivatives
(26),

1

A = Ad g 1 LR
K K .
Since we have to contract the A s with f , the adjoint action of G can be dropped to
obtain

Sda (0) (A) = f tr A A A
with A = K1 (LR K). Our previous remarks on the form of the structures constants
suggest to split this formula for the action into two terms

Sda (0) (A) = f ij k tr Ai Aj Ak + 3f rsi tr Ar As Ai .
(35)
The first term can be computed easily since Ai (g) = Ra (Ti ). As for the second term, one
proceeds by expressing the gauge field components Ar (g) through K(g). Integration by
parts and use of the anti-symmetry of the structure constants results in

ij k
tr Ra (Ti )Ra (Tj )Ra (Tk )
Sda (0) (A) = f

3 rsi
R
f
+
tr K1 LR
r , Ls K Ra (Ti ).
2
Because of the constraints on the structure constants, the commutator of the derivatives lies
in the direction of g so that we obtain a factor ifrs j Ra (Tj ) within the trace. After a bit of
algebra using Eqs. (21), (23) and (24), the two terms in the expression for the action can
be combined into the following result

2 Ca
C
2
Sda (0) (A) =
3C
.
12k 2 xe
xe
145
(36)
The numbers Ca , C , C are the quadratic Casimirs of the representation a of g and of

the adjoint representations of g, g , respectively (, are the highest roots). The constant
xe denotes the embedding index of the embedding g > g. It appears due to different
normalization of the quadratic forms g and g |g . It turns out that the value of the
action (36) is always negative.
As we recalled before, this result for the value of the action must be compared with
the difference between two logarithms of the g-factors [3] in the CFT-description. For the
branes (a), the g-factor is given by [12, Eq. (4.2)]
S0
(a)
g(a) =
,
S00
(37)
where we explicitly used the identification : PG+ B to emphasize that the second
argument of the twisted S matrix is a fractional symmetric weight. This leads to the
following expression for the logarithm of ratio of g-factors
S0
g(a)
2 dim g Ca
(a)
ln
= ln
=
C .
da g(0)
da S0 (0)
12k 2 dim g xe
(38)
To derive this result one shows first that the quotient in the argument of the logarithm is an
ordinary character of g ,
S0
2i 1
(a)
=
P()
.
S0
k + g xe
(0)
Using (a generalized version of) the asymptotic expansion in [20, Eq. (13.175)] one then
recovers (38) to lowest order in 1/k.
Although it is not obvious, formula (38) agrees with the previous expression (36) for the
value of the effective action at the solution which was obtained under the assumption that
is an order 2 automorphism. This can be checked with the help of Table 1 and well known
Lie algebra data (see for example [31, p. 44]). To summarize we have shown (except for
the case G2 > D4 ) that our solutions describe processes of the type (34).
5. Conclusions and outlook

In this work we have extended the previous analysis of brane dynamics [57] to brane
configurations of all maximally symmetric branes on group manifolds including the socalled twisted D-branes. In particular, we exploited the CFT data to construct the algebra of
gauge fields on such branes and provided a formula for the effective action. Moreover, we
found a large number of solutions and their geometric interpretation. The new condensation
processes turn out to be consistent with the charge conservation laws formulated in [27]
146
(see also [47]). In particular they suggest that the charge of an arbitrary -twisted D-brane
of type a PG+
= B is given by the dimension of the group representation a.
All this analysis is performed in the limiting regime where the level k is sent to infinity.
It would be interesting to investigate how the described condensation processes deform
when we go to finite values of the level k. For constant gauge fields, such deformations
have been studied in [4,27] based on the absorption of the boundary spin-principle (see
[1,2]). These investigations lead to very strong constraints on the structure of the charges
that are carried by untwisted D-branes. For non-trivial , however, constant condensates
provide only a small number of processes which are difficult to evaluate. The bound state
formation (34) we found in the last section of this paper suggests some obvious extensions
that might place the analysis of conservation laws for twisted and untwisted D-branes on
an equal footing.
Geometrically, the situations with finite k are associated with a non-vanishing NSNS
3-form H-field. As has been argued in [5,19,39] this might lead to new phenomena in the
world-volume geometry. For branes on group manifolds they seem to be related to quantum
groups or appropriate modifications thereof (see [5,8,54]).
Let us also mention that the results we have obtained here can be of direct relevance
for the study of branes in coset models that has recently attracted some attention [22,
26,28,37,48,67]. In fact, it was shown in [26,28] that many results on the dynamics of
branes on group manifolds descend to cosets through some reduction procedure. The
main ideas of this reduction are not specific to trivial gluing conditions and generalize
immediately to twisted branes in coset theories. Finally, following the ideas of [48,59], the
coset construction is an essential tool to obtain symmetry breaking boundary conditions,
e.g., on group manifolds. It is therefore tempting to conclude that a combination of all these
results can provide a rather exhaustive picture of brane dynamics on group manifolds, even
when we go beyond maximally symmetric branes.
Acknowledgement
We would like to thank V. Braun, G. Moore, I. Runkel and Ch. Schweigert for their
comments and useful discussions. The work of T.Q. and S.F. was supported by the
Studienstiftung des deutschen Volkes. Part of the research was done during a stay of T.Q. at
the University of Geneva which was supported by the Swiss National Science Foundation.
Appendix A. Correspondence between boundary labels and representations

The aim of this section is to provide the correspondence between exactly solvable
boundary conditions and representations of the invariant subgroup G . This is needed
to relate the results of the more geometric picture to those of CFT. We will proceed in the
first subsection by comparing the explicit expressions for the NIM-reps (7) to recent results
for branching coefficients of arbitrary semi-simple Lie algebras [57]. We are thus able to
distinguish a certain subalgebra h and identify boundary labels with representations of
h such that we get a purely Lie algebra theoretic expression for the NIM-reps in the
147
limit k . In almost all cases h is given by the invariant subalgebra g and we also
have PG+ = Pg+ . The only case where this procedure does not work is g = A2n where we
are lead to the orbit Lie algebra h = Cn and not to the invariant subalgebra g = Bn . It
was argued in [57], however, that in this specific case a second identification is possible
which leads to the invariant subalgebra. We reserve a second subsection to review these
results and to discuss the case of A2n in detail. Note that results closely related to those
presented in this section have also been found independently in [35,55] for finite k using
different methods. In all cases for which such a finite k extension exists, i.e., for the An series in [55] as well as the Dn -series and the A2n -series in [35], our results may also be
recovered from the existing literature by taking k to infinity. Let us emphasize, however,
that in the cases of A2n1 , D4 (triality) and E6 our treatment seems to indicate stronger
statements, i.e., larger subgroups, for finite k than those proposed in [35]. The geometrical
interpretation of these results which was described in the main text and the identification of
the invariant subgroup as the relevant structure seem to be new. The results of this appendix
have already been announced in [57] where they were used to derive new representations
for branching coefficients.
A.1. The generic correspondence
In this first subsection we propose an identification of boundary labels with representations of a distinguished subalgebra h of g such that formula (16) is satisfied for the
embedding h > g after taking the limit k . Neither is obvious that this will work
a priori nor is clear which subalgebra one should take. Starting from certain assumptions
we will first derive a set of consistency relations. Afterwards we will show that for each g
there is indeed a unique solution h to these consistency equations and that in most cases
it is given by the invariant subalgebra g .
Let us begin our discussion with a few general remarks about embeddings of Lie
algebras. Any embedding map : h > g induces a projection (or rather restriction)
(g)
(h )
in weight space P : Lw Lw . A representation is completely defined by its character,
i.e., by the weight system of the given highest weight. Under the projection P, this weight
system is mapped to a sum of weight systems of highest weights of
the subalgebra. This
process may be summarized in the decomposition formula UA = c bA c Vc of modules
(both sides considered as modules of h ) where we introduced the so-called branching
coefficients bA c N0 .
It is now important to specify which structure our identification of boundary labels
and representations is supposed to preserve. Let B ! be the integer linear span of the
set of boundary conditions B , i.e., the full lattice generated by elements of B . Both
(h )
the lattice B ! and the weight lattice Lw permit an action of Weyl groups. In the
first case this group is given by the symmetric part W = {w Wg | w = w } of
the Weyl group of g and in the other case it is naturally given by Wh . Furthermore, in
(g)
(h )
(g)
both cases we have a projection P : Lw Lw and P : Lw B !, respectively. The
latter is given by the projection P = N1 (1 + + + N1 ) onto the symmetric part
of the weights, N being the order of . As we will see, we have to find an isomorphism
(h )
: Lw B ! between the fractional lattice generated by the boundary conditions and
148
the weight lattice of the subalgebra which preserves all of these structures. In particular it
should be accompanied with an isomorphism : Wh W of the corresponding Weyl
groups. To summarize, we have to find a subalgebra h and a functor-like map such that
the diagrams (A.1) commute. It turns out that the answer for both h and is unique.
P
(g)
Lw
(h )
Lw
(h )
Lw
B !
(h )
Lw
wWh
(h )
Lw
B !
(w)W
(A.1)
B !
In the following we restrict ourselves to some non-trivial outer automorphism = id since

our statements become trivial otherwise.
Remember that we only consider the limit k . For = (a), = (b) we want
to proof the equality
c
nA =
bA Ncb a ,
c
where the last expression contains some branching coefficients for h > g and the tensor
product coefficients of the subalgebra. This relation explicitly assumes the existence of the
bijection : a Ph+ B conjectured above. Using a result of [58] and a useful
identity for branching coefficients (which is related to [44], see however [57] for a recent
proof using affine KacMoody algebra techniques) we may write

nA =
) (w),w(P B++ ) ,

Bwts(A),wW
bA Ncb a =
c
)(w)a,w(P B+b+) .
(A.2)
Bwts(A),wWh
The abbreviation B wts(A) means that B runs over all weights in the weight system
of A. Both expressions are obviously equal to each other if the bijection is structure
preserving, i.e.,
() = ,
P = P ,
(wa) = (w) (a).
The last condition already constrains the possible subalgebras to a large extent. Indeed, the
Weyl group W can be described as the Weyl group of the so-called orbit Lie algebra of g
with respect to the automorphism (see [32]). In some special cases this orbit Lie algebra
coincides with the invariant subalgebra g while it does not for the whole An and Dn
series. A survey of these relations can be found in Table 2 which in part has been taken
from [13]. Note however that the Weyl groups of Bn and Cn are isomorphic to each other
(see, e.g., [31, p. 74]) which can most easily be seen by treating them as abstract Coxeter
groups. Thus by imposing the last constraint we only have to decide which of possibly
two subalgebrasthe orbit Lie algebra or the invariant subalgebraand which specific
149
Table 2
More data related to outer automorphisms of simple Lie algebras
g
Order
Orbit Lie algebra g
Relevant subalgebra h
A2
A2n1
A2n
D4
Dn
E6
2
2
2
3
2
2
A1 (xe = 4)
Cn
Bn
G2
Bn1
F4
A1
Bn
Cn
G2
Cn1
F4
A1 (xe = 1)
Cn
Cn > A2n1
G2 > B3
Bn1
F4
kind of embedding we should take. This choice is uniquely determined by the other two
conditions. In the cases of A2n1 and Dn the orbit Lie algebra not even is a subalgebra so
that this possibility is ruled out immediately.
We will show in the most simple example of the Lie algebra A2 how this procedure
works and then state only results for all the other cases. Let us consider g = A2 with outer
automorphism (a1 , a2 ) = (a2 , a1 ). The relevant subalgebra is given by h = A1 and the
projection to fractional symmetric weightswhich describe the boundary conditions of the
theoryreads P (a1 , a2 ) = 12 (a1 + a2 , a1 + a2 ). There are two inequivalent embeddings
A1 > A2 given by projections Pxe (a1 , a2 ) = xe (a1 + a2 ) with embedding index xe = 1

and xe = 4, respectively [20, p. 534]. Imposing the first condition we see that
Pxe (a1 , a2 ) =
xe (a1 + a2 ).
(A.3)
This only equals P (a1 , a2 ) for

1
(a) = (a, a).
2 xe
(A.4)
The condition () = = 12 (1, 1) forces us to use the projection with xe = 1. One

can also check explicitly that the Weyl groups correspond to each other. This is the first
example where the relevant subalgebra is not given by the invariant subalgebra (which
has embedding index xe = 4) but by the orbit Lie algebra. The same statement holds for
the whole A2n series as we will see. As mentioned already the results of this section are
summarized in Table 2.
One can treat the whole ADE series using a case by case study. Let us emphasize that we
use the labeling conventions for weights which can be found in [20, p. 540]. The projections
have been found using [50, pp. 5761] and the programs LiE [18] and SimpLie [51]. Note
that LiE uses a different labeling convention for the weights. For a useful table of branching
rules see also [49].
(1) The case of A2n1 is straightforward. The relevant subalgebra is given by the
invariant subalgebra Cn > A2n1 . This is a maximal embedding and the identification
150
reads
1
1
P =
2
..
..
..
.
1
1 1
1
.
.
=
.. ..
= P.
2
2
2
1 1
1
1
..
.
..
..
.
.
.
(2) The case A2n is exceptional. Here the relevant subalgebra is given by the orbit Lie
algebra which can be described by the sequence of maximal embeddings Cn > A2n1 >
A2n (for n = 1 we have A1 > A2 ). The identification reads
1
1
P =
2
..
..
.
..
.
.
11
11
..
1
=
2
.
1
..

. 1
1
1
..
.
.
= P.
. .
1
1
1
.
.
There is a second identification related to the subalgebra Bn > A2n which will be
discussed in the next subsection and which is the relevant one for the main part of the
paper.
(3) The order 3 diagram automorphism of D4 leads to the sequence of maximal
embeddings G2 > B3 > D4 and to the identification
1 0 1 1
0 1
1 0 3 0 0 1 3 00 1 0 0
= P.
=
P =
1011
3 1011
3 01
1011
01
(4) For the order 2 automorphism of the D-series one obtains the maximal embedding
Bn1 > Dn and
2
P =
..
.
2
11
11
1
=
2
..
.
2
1
1
...
= P.
11
(5) Also the last case E6 behaves regular and yields the maximal embedding F4 > E6
with
1 0 0 0 1 0
0 0 0 1
010100
0 0 1 0 0 0 0 0 0 1
1
0 0 2 0 0 0 1 0 2 0 0 0 0 1 0 0 0
P =
= P.
=
2 0 1 0 1 0 0 2 0 0 1 0 0 1 0 1 0 0
100010
000002
0001
2000
100010
151
These considerations show that in all cases but A2n we may identify the boundary labels
with representations of the invariant subgroup G . Let us emphasize that it is possible to
identify the set of Lie algebra representations Pg+ with the set of group representations
PG+ in these cases as the corresponding groups G are all simply-connected. A detailed
discussion of the identification one should use in the exceptional case of A2n is postponed
to the next subsection.
A.2. The special case A2n
In the last section it was shown that under certain natural and well-motivated
assumptions the relevant subalgebra h in the case of A2n which describes the boundary
labels is uniquely given by the orbit Lie algebra Cn and not by the invariant subalgebra
Bn . This contradicts, however, our geometrical intuition as we are expecting the invariant
subalgebra (or even better the invariant subgroup) to be the relevant structure. In [57],
however, it was argued that one is lead to the invariant subalgebra by taking a different
inductive limit, i.e., an identification which involves the level k explicitly. Indeed, in
writing (A.2) we already took a very special limit implicitly. That taking the limit k
may have non-trivial effects can already be seen from simple current symmetries in fusion
rules. In this limit all simple current symmetries get lost and it becomes important on which
branch of the simple current orbits one sits while taking the limit. There is also another
point on which we have not been careful enough in the last subsection. The geometric
picture suggests that we should work with representations of the invariant subgroup, not
necessarily with those of the invariant subalgebra. These two sets may differ as can easily
be seen from the familiar example of SO(3) which only allows SU(2) representations of
integer spin. Note that the automorphism in the case of SU(3) is just given by charge
conjugation and that SO(3) exactly is the invariant subgroup. Similar remarks hold for the
whole series SO(2N + 1) SU(2N + 1), i.e., the whole A2n series. All this should be
reflected in the new identification in a certain way.
Let us review the new identification of [57] and see whether it fits our requirements.
The construction only works for even values of the level k. This fact may be reminiscent
of the D-series modular invariants of SU(2) describing a SO(3) WZW model. Therefore
we will assume k to be even in what follows. This restriction will not be relevant in the
limit k . The labels for the twisted boundary conditions in the WZW model based
on A2n are given by half-integer symmetric weights of A2n . To be more
specific, the
Dynkin labels have to satisfy the relations 2i N0 , i = 2n+1i and ni=0 i k/4.
These labels may be interpreted as labels of the invariant subalgebra Bn of A2n . The map
from weights of Bn to the boundary labels is given by [57]

n1

1
ai an , . . . , 2an1 .
(a1 , . . . , an ) =
(A.5)
2an1 , . . . , 2a1 , k 2
4
i=1
Note that this map involves k explicitly and is only well-defined for weights whose
last Dynkin label an is even. This last condition has two interpretations. From the
group theoretical point of view it restricts to representations of the Lie algebra Bn
which may be integrated to single-valued representations of the group SO(2n + 1).
152
From the Lie algebra point of view it corresponds to the branching selection rule of the
embedding Bn > A2n . The relevant projection for this embedding reads P(i1, . . . , i2n ) =
(i1 + i2n , i2 + i2n1 , . . . , 2(in + in+1 )).
Actually, there is a geometric reason why we should use a k-dependent identification
map in the case of A2n . In the limit k , the twisted conjugacy classes in the vicinity
of the group unit have boundary labels close to (0, . . . , 0, k/4, k/4, 0, . . ., 0). This can be
inferred from an analysis of brane geometry seen by closed strings [25].
Let us summarize these results. Using the new identification map (A.5) we can
identify the labels of twisted branes in SU(2n + 1) sitting close to the group unit with
representations of the invariant subgroup G > G. In particular this identification knows
about the fact that certain representations of g may not be lifted to representations of G
and drops them from the set of boundary labels.
Appendix B. Proof of bundle triviality

In this appendix we will prove that all complexified vector bundles G G VRC over
the base manifold G/G associated to representations VR of G are trivial. Here, G is
any simple simply-connected compact Lie group and G the subgroup invariant under a
diagram automorphism. All possible cases are summarized in Table 1.
Before we start with the actual proof, let us note that representations VR which arise
by restricting representations of G to G always lead to trivial bundles. We will use this
extensively to proof the triviality of all other bundles.
The proof can be devided into five parts. We will first present these five main statements
and then enter the detailed discussion of the single steps.
Statement 1. Consider the K-ring KC (G/G ) of complex vector bundles over the base
manifold G/G . The map K : VR K(G G VRC ) which sends a representation VR of
G to the K-class of its associated complexified vector bundle, is a ring homomorphism
from the representation ring Rep(G ) to KC (G/G ).
Statement 2. The representation ring of G is a polynomial ring on the fundamental
representations, Rep(G ) = Z[V1 , . . . , Vr ], r = rank G .
Therefore any K-class of a bundle G G VRC can be expressed as a polynomial in the
K-classes of G G VCi .
Statement 3. All complexified vector bundles associated to fundamental representations of
G have trivial K-class, K(G G VCi ) = 0, i.e., all these bundles are stably equivalent
to trivial bundles.
From the previous remark it follows then that all bundles G G VRC are stably equivalent
to trivial bundles.
Statement 4. Two stably equivalent complex vector bundles of rank d over a base manifold
of dimension n satisfying 2d n are isomorphic.
153
Statement 5. All representations VR that are not a restriction of a representation of G

obey the inequality
2 dim VR dim G/G .
(B.1)
We had seen that all bundles are stably equivalent to trivial bundles, from the last two
statements we can thus conclude that all bundles are trivial. This ends the main line of
argumentation. Note that it was important that we considered complexified vector bundles,
otherwise there would appear a much stronger inequality in Statement 4 which in many
cases could not be fulfilled any more.
Let us now take a closer look at the single statements. The first statement follows from
the fact that the bundles associated to the tensor product of two representations VR VR
is the tensor product of the associated bundles,
G G (VR VR ) $ (G G VR ) G/G (G G VR ).
Statement 2 is a structure theorem which can be found, e.g., in [34, Theorem 23.24].
Statement 3 is much more technical. We have to check its validity case by case. We
know that restrictions of representations of G give rise to trivial bundles and thus to trivial
K-classes. Studying the appearance of fundamental representations in the decomposition
of G-representations, we deduce in an inductive way that all fundamental representations
of G correspond to bundles of trivial K-classes. Let us discuss the way it works in an
example.
B3 > D4 : the fundamental representations of B3 have the dimensions 7, 8 and 21.
The first fundamental representation of D4 is 8-dimensional and decomposes as 8
7 + 1. The corresponding bundle is trivial, as well as the bundle associated to the trivial
representation 1, hence the bundle associated to the 7-dimensional representation is stably
equivalent to a trivial bundle. The next fundamental representation of D4 decomposes
as 28 21 + 7 so that we find by an analogous argument as above that the bundle
associated to the fundamental 21 of B3 is stably equivalent to a trivial bundle. The
remaining 8-dimensional fundamental representation of B3 is the restriction of one of the
8-dimensional representations of D4 and thus gives rise to a trivial bundle.
In a similar way we will prove Statement 3 for all cases at the end of this section. But
before we do so, we want to discuss the last two statements. After we have shown that the
fundamental and thus all representations give rise to bundles which are stably equivalent
to trivial bundles, we want to show that they are actually really trivial. Statement 4 is a
theorem that can be found, e.g., in [42, Theorem 9.1.5] which tells us that s-equivalent
and isomorphic have the same meaning if the rank of the bundles is high enough. The
last Statement 5 ensures that all our bundles indeed comply with this requirement, hence
they are trivial. To prove this statement we determine for all cases the lowest-dimensional
representation not arising as a restriction of a G-representation. We show then that these
satisfy the inequality (B.1).
As an example take again B3 > D4 . The lowest-dimensional non-trivial representation
has dimension 7 which is not a restriction of a representation of D4 . The base manifold has
dimension dim D4 dim B3 = 28 21 = 7. As 2 7 = 14 7, the inequality holds.
We will now show the validity of this statement for all cases together with Statement 3
in a case-by-case study.
154
Bn > A2n [SO(2n + 1) SU(2n + 1)]:

Decomposition of fundamental representations of A2n (Dynkin label notation):
(0 . . 1 . . . . . . . . . . . 0)
(0 . . . 1 . . 0)
i <n
(0 . . . . . 1 . . . . . . . . 0)
(0 . . . . . . .2)
n
All fundamental weights of SO(2n + 1) are restrictions of fundamental representations of
SU(2n + 1). Statement 3.
The lowest-dimensional representation not obtained from a restriction is the representation
(20 . . . 0) of Bn with dimension 2n2 + 3n. The base manifold has dimension dim SU(2n +
1) dim SO(2n + 1) = 2n2 + 3n, so inequality (B.1) holds. Statement 5.
Cn > A2n1 [Sp(2n) SU(2n)]:
Decomposition of fundamental weights of A2n1 :
(1 . . . . . . . . . . . . . . . . . 0)
(1 . . . . . . . . . . 0)
(01 . . . . . . . . . . . . . . . . 0)
(01 . . . . . . . . . 0) (0 . . . . . . . . . . 0)
(001 . . . . . . . . . . . . . . . 0)
(001 . . . . . . . . 0) (1 . . . . . . . . . . 0)
(0 . . . . . . 1 . . . . . . . . . 0)
(0 . . . . . . 1 . . . 0) (0 . . . . . 1 . . . . 0)
i 2
i

(i n)
(1 . . . . . . . . . . 0) i odd
(0 . . . . . . . . . . 0) i even
Proceeding inductively, we see that all bundles associated to fundamental representations

of Cn have trivial K-class. Statement 3.
The lowest-dimensional representation that cannot be obtained from a restriction is
(010 . . .0) and has dimension 2n2 n 1. The base manifold has dimension dim SU(2n)
dim Sp(2n) = 2n2 n 1. Statement 5.
Bn1 > Dn [Spin(2n 1) Spin(2n)]:
Decomposition of fundamental representations of Dn :
(10 . . . . . . . . . 0) (10 . . . . . . . . . 0) (0 . . . . . . . . . . . 0)
(01 . . . . . . . . . 0) (01 . . . . . . . . . 0) (10 . . . . . . . . . 0)
(0 . . . . . . 1 . . . 0) (0 . . . . . . 1 . . . 0) (0 . . . . . 1 . . . . 0)
i (i < n 1)
(0 . . . . . . . . . 10)
(0 . . . . . . . . . 01)
i
(0 . . . . . . . . . . . 1)
i 1
155
Statement 3.
The lowest-dimensional non-trivial representation of Bn1 is (10 . . .0) and has dimension
2n 1. The dimension of the base manifold is dim Spin(2n) dim Spin(2n 1) = 2n 1.
Statement 5.
F4 > E6 :
Decomposition of representations of E6 :
(000010) (0001) (0000)
(000001) (1000) (0001)
(000100) (0010) (1000) (0001)
So we see that the three fundamental representations (1000), (0010), (0001) of F4 give
rise to bundles of vanishing K-class. From the tensor product
(0001) (1000) = (0001) (0010) (1001)
we can deduce that the same is valid for (1001).
Now we look at the decomposition
(001000) (0100) (1001) (0010) (0010) (1000)
and we find a vanishing K-class also for the fourth fundamental representation (0100).
Statement 3.
The lowest-dimensional representation of F4 is (0001) and has dimension 26. The
dimension of the base manifold is dim E6 dim F4 = 26. Statement 5.
G2 > D4 :
Decompositions of representations of D4 :
(1000) (01) (00)
(0100) (10) (01) (01)
Statement 3.
The lowest-dimensional non-trivial representation of G2 is (01) and has dimension 7. The
base manifold has dimension dim D4 dim G2 = 28 14 = 14. Statement 5.
References
[1] I. Affleck, A.W.W. Ludwig, Critical theory of overscreened Kondo fixed points, Nucl. Phys. B 360 (1991)
641696.
[2] I. Affleck, A.W.W. Ludwig, The Kondo effect, conformal field theory and fusion rules, Nucl. Phys. B 352
(1991) 849862.
[3] I. Affleck, A.W.W. Ludwig, Universal noninteger ground state degeneracy in critical quantum systems,
Phys. Rev. Lett. 67 (1991) 161164.
[4] A.Yu. Alekseev, V. Schomerus, RR Charges of D2-branes in the WZW Model, 2000.
[5] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Non-commutative world-volume geometries: branes on su(2)
and fuzzy spheres, JHEP 09 (1999) 023.
[6] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Brane dynamics in background fluxes and non-commutative
geometry, JHEP 05 (2000) 010.
156
[7] A.Yu. Alekseev, A. Recknagel, V. Schomerus, Open strings and non-commutative geometry of branes on
group manifolds, Mod. Phys. Lett. A 16 (2001) 325336.
[8] A.Yu. Alekseev, V. Schomerus, Representation theory of ChernSimons observables, Duke Math. J. 85
(1996) 447510.
[9] A.Yu. Alekseev, V. Schomerus, D-branes in the WZW model, Phys. Rev. D 60 (1999) 061901.
[10] C. Bachas, M.R. Douglas, C. Schweigert, Flux stabilization of D-branes, JHEP 05 (2000) 048.
[11] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, On the classification of bulk and boundary conformal
field theories, Phys. Lett. B 444 (1998) 163166.
[12] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Boundary conditions in rational conformal field
theories, Nucl. Phys. B 570 (2000) 525589.
[13] L. Birke, J. Fuchs, C. Schweigert, Symmetry breaking boundary conditions and WZW orbifolds, Adv. Theor.
Math. Phys. 3 (1999) 671726.
[14] P. Bordalo, S. Ribault, C. Schweigert, Flux stabilization in compact groups, JHEP 10 (2001) 036.
[15] P. Bouwknegt, V. Mathai, D-branes, B-fields and twisted K-theory, JHEP 03 (2000) 007.
[16] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[17] C.-S. Chu, P.-M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151168.
[18] A.M. Cohen, M. van Leeuwen, B. Lisser, LiE, 1996, this computer program can be obtained from
http://young.sp2mi.univ-poitiers.fr/~marc/LiE/.
[19] L. Cornalba, R. Schiappa, Nonassociative star product deformations for D-brane worldvolumes in curved
backgrounds, Commun. Math. Phys. 225 (2002) 3366.
[20] P. Di Francesco, P. Mathieu, D. Senechal, Conformal Field Theory, in: Graduate Texts in Contemporary
Physics, Springer, New York, 1999.
[21] M.R. Douglas, C.M. Hull, D-branes and the noncommutative torus, JHEP 02 (1998) 008.
[22] S. Elitzur, G. Sarkissian, D-branes on a gauged WZW model, Nucl. Phys. B 625 (2002) 166178.
[23] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, Correlation Functions and Boundary Conditions in RCFT
and Three-Dimensional Topology, 1999.
[24] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, Conformal boundary conditions and three-dimensional
topological field theory, Phys. Rev. Lett. 84 (2000) 16591662.
[25] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, The geometry of WZW branes, J. Geom. Phys. 34 (2000)
162190.
[26] S. Fredenhagen, V. Schomerus, Brane Dynamics in CFT Backgrounds, 2001.
[27] S. Fredenhagen, V. Schomerus, Branes on group manifolds, gluon condensates, and twisted K-theory,
JHEP 04 (2001) 007.
[28] S. Fredenhagen, V. Schomerus, D-branes in coset models, JHEP 02 (2002) 005.
[29] J. Frhlich, K. Gawedzki, Conformal Field Theory and Geometry of Strings, 1993.
[30] J. Frhlich, O. Grandjean, A. Recknagel, Supersymmetric Quantum Theory, Noncommutative Geometry,
and Gravitation, 1995.
[31] J. Fuchs, Affine Lie Algebras and Quantum Groups, in: Cambridge Monographs on Mathematical Physics,
Cambridge Univ. Press, Cambridge, 1995.
[32] J. Fuchs, B. Schellekens, C. Schweigert, From Dynkin diagram symmetries to fixed point structures,
Commun. Math. Phys. 180 (1996) 3998.
[33] J. Fuchs, C. Schweigert, Symmetry breaking boundaries I: General theory, Nucl. Phys. 419 (1999) 419483.
[34] W. Fulton, J. Harris, Representation Theory: A First Course, in: Graduate Texts in Mathematics, Vol. 129,
Springer, 1991.
[35] M.R. Gaberdiel, T. Gannon, Boundary states for WZW models, Nucl. Phys. 639 (2002) 471501.
[36] K. Gawedzki, Conformal Field Theory: a Case Study, 1999.
[37] K. Gawedzki, Boundary WZW, G/H, G/G and CS Theories, 2001.
[38] K. Hashimoto, K. Krasnov, D-brane solutions in non-commutative gauge theory on fuzzy sphere, Phys.
Rev. 64 (2001) 046007.
[39] K. Hayasaka, R. Nakayama, An associative and noncommutative product for the low energy effective theory
of a D-brane in curved backgrounds and bi-local fields, Nucl. Phys. 624 (2002) 307326.
[40] Y. Hikida, M. Nozaki, Y. Sugawara, Formation of spherical D2-brane from multiple D0-branes, Nucl. Phys.
B 617 (2001) 117150.
[41] J. Hoppe, Diffeomorphism groups, quantization and su(infinity), Int. J. Mod. Phys. A 4 (1989) 5235.
157
[42] D. Husemoller, Fibre Bundles, in: Graduate Texts in Mathematics, Vol. 20, 3rd Edition, Springer, 1994.
[43] V.G. Kac, Infinite Dimensional Lie Algebras, Cambridge Univ. Press, Cambridge, 1990.
[44] A. Klimyk, Multiplicities of weights of representations and multiplicities of representations of semisimple
Lie algebras, Soviet Math. Dokl. 8 (1967) 15311534, published originally in Dokl. Akad. Nauk
SSSR 177 (5) (1967) 10011004, in Russian.
[45] D.C. Lewellen, Sewing constraints for conformal field theories on surfaces with boundaries, Nucl. Phys.
B 372 (1992) 654682.
[46] J. Madore, The fuzzy sphere, Class. Quantum Grav. 9 (1992) 6988.
[47] J.M. Maldacena, G.W. Moore, N. Seiberg, D-brane instantons and K-theory charges, JHEP 11 (2001) 062.
[48] J.M. Maldacena, G.W. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models,
JHEP 07 (2001) 046.
[49] W.G. McKay, J. Patera, Tables of Dimensions, Indices, and Branching Rules for Representations of Simple
Lie Algebras, Dekker, New York, 1981.
[50] W.G. McKay, J. Patera, D.W. Rand, in: Tables of Representations of Simple Lie Algebras, Vol. I, Centre de
Recherches Mathematiques, Montreal, 1990.
[51] R.V. Moody, J. Patera, D.W. Rand. SimpLie, 2000, this computer program can be obtained from
http://omega.CRM.UMontreal.CA/~rand/simpLie.html.
[52] R.C. Myers, Dielectric-branes, JHEP 12 (1999) 022.
[53] J. Pawelczyk, SU(2) WZW D-branes and their noncommutative geometry from DBI action, JHEP 08 (2000)
006.
[54] J. Pawelczyk, H. Steinacker, A Quantum Algebraic Description of D-branes on Group Manifolds, 2002.
[55] V.B. Petkova, J.B. Zuber, Boundary Conditions in Charge Conjugate sl(n) WZW Theories, 2002.
[56] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Completeness conditions for boundary operators in 2D conformal field
theory, Phys. Lett. B 381 (1996) 97104.
[57] T. Quella, Branching rules of semi-simple Lie algebras using affine extensions, J. Phys. A 35 (2002) 3743
3754.
[58] T. Quella, I. Runkel, C. Schweigert, An Algorithm for Twisted Fusion Rules, 2002.
[59] T. Quella, V. Schomerus, Symmetry breaking boundary states and defect lines, JHEP 06 (2002) 028.
[60] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185225.
[61] A. Recknagel, V. Schomerus, Boundary deformation theory and moduli spaces of D-branes, Nucl. Phys.
B 545 (1999) 233282.
[62] A. Recknagel, V. Schomerus, Moduli spaces of D-branes in CFT-backgrounds, Nucl. Phys. B 48 (2000)
195198.
[63] I. Runkel, Boundary structure constants for the A-series Virasoro minimal models, Nucl. Phys. B 549 (1999)
563578.
[64] I. Runkel, Structure constants for the D-series Virasoro minimal models, Nucl. Phys. B 579 (2000) 561589.
[65] V. Schomerus, D-branes and deformation quantization, JHEP 06 (1999) 030.
[66] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 09 (1999) 032.
[67] S. Stanciu, D-branes in KazamaSuzuki models, Nucl. Phys. B 526 (1998) 295310.
[68] S. Stanciu, D-branes in group manifolds, JHEP 01 (2000) 025.
[69] S. Stanciu, An Illustrated Guide to D-branes in su(3), 2001.
[70] E. Verlinde, Fusion rules and modular transformations in 2D conformal field theory, Nucl. Phys. B 300
(1988) 360.

Fermion-induced quantum action of vortex systems

Kurt Langfeld a , Laurent Moyaerts a , Holger Gies b
a Institut fr Theoretische Physik, Universitt Tbingen, D-72076 Tbingen, Germany
b CERN, Theory Division, CH-1211 Geneva 23, Switzerland
Abstract
The quantum action generated by fermions which are minimally coupled to abelian vortex
background fields is studied in D = 2 + 1 and D = 3 + 1 Euclidean dimensions. We present a
detailed analysis of single- and binary-vortex configurations using the recently developed method
of worldline numerics. The dependence of the fermion-induced quantum action on the fermion
mass and the magnetic fluxes carried by the vortices is studied, and the binary-vortex interaction
is computed. Additionally, we discuss the chiral condensate generated from a dilute gas of vortices
in the intermediate fermion mass range for the case D = 3 + 1. As a byproduct, our findings provide
insight into the validity limits of the derivative expansion, which is the standard analytical approach
to inhomogeneous backgrounds.
PACS: 12.20.-m; 11.15.Ha
Keywords: Worldline; Fermion-induced effective action; Vortex; Monte Carlo simulation
1. Introduction
The dynamics of surfaces plays an important role in many fields of physics ranging
from solid state physics and chiral quantum field theories [13] to string theory [4]. The socalled vortices, (d 2)-dimensional surfaces of the d-dimensional space, are of particular
interest in view of their phenomenological importance: in solid state physics, freely moving
vortices give rise to a nonvanishing conductance of high Tc -superconductors, thereby
limiting their technical applicability [5]. In the context of YangMills theory, it has recently
been observed in lattice gauge simulations that, in the continuum limit [6], the YangMills
E-mail address: kurt.langfeld@uni-tuebingen.de (K. Langfeld).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 3 5 - 0
K. Langfeld et al. / Nuclear Physics B 646 (2002) 158180
159
vacuum is populated by center vortices the core of which can be detected in a certain gauge
by a projection technique [7,8]. A random gas of these vortices can grasp the essence of
quark confinement at zero temperature, and the deconfinement phase transition at finite
temperatures can be understood as a vortex de-percolation transition [9,10]. Although
rigorous estimates signal that the core size of the full (unprojected) field configurations
is spread over the whole spacetime [11], an effective vortex model supplemented by a
particular core size describes certain low-energy observables remarkably well [12,13].
In many applications, the surfaces appear as classical background fields (solitons) of
an underlying field theory. The surface properties as well as the surfaces interactions are
determined by thermal and quantum fluctuations, respectively [14,15]. In the context of the
layered superconductors, the spectrum of the quasi-particles encodes phenomenologically
relevant information of the vortex background [16,17]. In the context of YangMills theory,
a deep knowledge of the interplay of quarks with vortex-like solitons would help to
describe hadron properties in the vortex picture.
The determination of the vortex free energy generically requires the calculation of
functional determinants where the vortex profile enters as a background field. Whereas
the eigenvalue problem of the single-vortex case with a -function profile can be solved
analytically [18], a numerical treatment seems inevitable for realistic profiles and multivortex configurations. One possible technique aims at the numerical integration of a
KleinGordon or Dirac type of equation associated with the differential operator under
consideration. For example, in [14,15], this approach has been further developed and the
numerical burden has been reduced to a quantum mechanical computation of boundstate energies and scattering phase shifts; in particular, the usually delicate issue of
renormalization can advantageously be inherited from standard perturbation theory in this
approach. The disadvantage of this approach is that numerical complications increase with
the degree of complexity of the background field; up to now, only highly symmetric
backgrounds have been treated (see [19] and [20] for the single-vortex case). In this
work, we put forward a recently proposed numerical technique [21,22] which is based
on the string-inspired worldline method [23]. In this formalism, the effective action
(functional determinant) is represented in terms of first-quantized particle path-integrals,
which have turned out to be highly convenient for analytical computations involving
constant background fields (see [24] for a review). In our numerical realization of this
formalism, Monte Carlo techniques can be exploited to estimate expectation values of
background-field-dependent operators with respect to scalable worldline ensembles, socalled loop clouds. This worldline numerical scheme is formulated in coordinate space,
which facilitates a transparent renormalization, since the divergencies are associated with
local operators in coordinate space and the required counterterms can easily be read off.
The most important advantage of this formalism is marked by the fact that the numerical
algorithm can be realized without any reference to the specific choice of the background
field. In particular, a high degree of symmetry of the background is not required at all;
on the other hand, if a symmetry is present, it can, of course, be exploited for reducing
numerical efforts. Most recently, this technique was adapted for worldline loops on a
cubic lattice, most convenient in the case of lattice gauge simulations [25]. Therein, it was
pointed out that a random walk easily generates loop clouds with the appropriate measure
required for the worldline approach.
160
In this paper, we will explore the free energy of vortex-type solitons, being supported
by a U (1) gauge field which is minimally coupled to a fermion. We will discuss the case
of 2 + 1 and 3 + 1 spacetime dimensions and restrict ourselves to vortex configurations
which are static with respect to one or two directions of spacetime, respectively. The free
energy will be studied for the single and the binary-vortex configuration. In the latter case,
we will obtain the fermion-induced vortex-vortex interaction.
The paper is organized as follows. In Section 2, the worldline approach to the fermion
determinant is introduced and the renormalization procedure is discussed in detail for the
case D = 3 + 1. The vortex configurations under investigation are introduced in Section 3.
In Section 4, our numerical result for the single-vortex configuration is presented and
compared with the estimate obtained from existing analytical studies. The binary-vortex
interaction is presented in Section 5. Fermion condensation due to a dilute gas of vortices
is addressed in Section 6. Conclusions are left to the final section.
2. Fermion determinants in the worldline approach

2.1. Setup
In this paper, we shall investigate the case of one flavor of a 4-component Dirac fermion
in D = 2 + 1 and D = 3 + 1 spacetime dimensions. The fermion field is minimally
coupled to a U (1) gauge field A (x), which is considered as a background field. The
fermion-induced effective action ferm , which is a free-energy functional, determines the
probabilistic weight exp{ferm }. It is given by
ferm = ln det(m iD),
/
D
/ := / + i/
A,
(1)
where m is the fermion mass, A

/ = A (x) , and are the anti-hermitean Euclidean
-matrices, = . As a consequence, D
/ is hermitean.
Since the Dirac operator exhibits no spectral asymmetry in the 4-component formulation, the effective action is real, and we obtain in Schwinger proper-time regularization
1
ferm =
2
2
dT m2 T
e
Tr exp D
/ T ,
T
(2)
1/2
where the scale acts as a UV regulator. The idea of the worldline approach consists of
rewriting Eq. (2) in terms of 1-dimensional path integrals (for a comprehensive review,
see [24]). The worldline representation of Eq. (2) is given by
ferm =
1
1
2 (4)D/2

d D x0
dT
em
(D/2)+1
2T

4 Wspin [A] 1 x ,
(3)
1/2
where a gauge-field-independent constant has been dropped, and we have defined the
spinorial counter gauge factor of the Wilson loop including the Pauli term (spin-field
161
coupling),
T

T

1
1
d F
.
Wspin [A] = exp i A (x)x d tr PT exp
4
2
0
(4)
is the field strength tensor, and := i[ , ]/2 are hermitean Dirac

Here
algebra elements. The average x in (3) is performed over an ensemble of closed
worldlines, or loop clouds. A single loop is characterized by its worldline x ( ), [0, T ]
in D dimensions. PT denotes path ordering with respect to the propertime T . The loops
are centered upon a common average position,
F (x( ))
x0
1
:=
T
T
d x ( )
(center of mass).
The loop ensemble is generated according to the Gaussian weight:
1
exp
4
T

2

2
d x1 + + xD .
(5)
In practice, one greatly reduces the numerical work by generating a loop gas of unit
propertime T = 1 only, and by producing loop ensembles for a given propertime T = 1
by rescaling. This numerical method is discussed in some detail in [21,22].
Finally, we point out that the symmetry of the background field can easily be exploited
to reduce the numerical burden. In our case, the background field is static with respect
to D d spacetime dimensions, i.e., the gauge field depends only on the coordinates
x1 , . . . , xd , d < D. Since the weight of the loop clouds is Gaussian, the loop average in
D dimensions can be evaluated by using d < D dimensional loop clouds only, i.e.,

2

1 T
2
1
F [A] x =
(6)
Dx1...d F [A]e 4 0 d x1 ++xd ,
N
where
N=
Dx1...d e 4
T
0
d x 12 ++x d2
In particular, it is sufficient for the examples discussed below to generate a single (large)
2-dimensional loop cloud to address the properties of static vortices in D = 2 + 1 and
D = 3 + 1, respectively.
2.2. Renormalization
In the case of D = 2 + 1 dimensions, the propertime integration in Eq. (3) is finite at
the lower bound. This implies that one might safely remove the regulator, i.e., , in
this case. We therefore confine ourselves in this subsection to the case D = 3 + 1, where
the fermion-induced effective action discussed above is actually the bare action and given
162
by
B
ferm
1 4
=
2 (4)2

4
d x0

dT m2 T
e
Wspin [A] 1 x .
3
T
(7)
1/2
For small values of the propertime T , the expectation value Wspin [A] 1 x can be
calculated analytically for the case of a loop cloud with x0 being the center of mass. One
finds

1
Wspin [A] x = 1 + F (x0 )F (x0 )T 2 + ,
(8)
6
where the ellipsis denotes higher-order invariants that can be formed out of the field
strength tensor, its dual, and derivatives thereof. Rewriting (7) as
B
ferm
1
=
8 2
+

4
d x0
1/2

dT m2 T
1 2
2
F
W
e
[A]
(x
)T
spin
0
6
T3
x
dT m2 T
e
T
1 1
6 8 2
(9)

d 4 x0 F 2 (x0 ),
(10)
1/2
B is finite if the regulator is removed, . Using
we observe that the part (9) of ferm
2
2
dT m2 T
m
m
e
= ln
+
O
E
T
2
2
(11)
1/2
(E is the Euler constant), we neglect irrelevant terms which are suppressed by powers of
1/2 in part (9) and (10) and obtain
B
ferm
1
=
8 2

4
d x0
0

dT m2 T
1 2
2
F
W
e
[A]
(x
)T
spin
0
6
T3
x
2

1
m
ln
+
d 4 x0 F 2 (x0 ).
E
48 2
2
(12)
(13)
The renormalized effective action eff is obtained by adding the bare classical action

1
B
eff = ferm + 2
(14)
d 4 x0 F 2 (x0 ),
4gB
where gB2 is the bare coupling constant, and we enforce the renormalization condition
2

1
1
1
+
.
ln
E = 2
(15)
2
2
2
gB () 12
gR ()
Here gR () denotes the renormalized coupling at a given renormalization point , and we
rediscover the QED function (gR2 ) = gR2 () = gR4 /(6 2 ). Inserting (12) and (15)
163
into (14), the renormalized effective action is given by

eff = F 2 + ferm ,

dT m2 T
1
1 2
4
2
ferm =
e
,
d x0
Wspin [A] 1 F (x0 )T
8 2
T3
6
x
(16)
F 2
2

1
1
m
1
=
ln
d 4 x0 F 2 (x0 ).
2
2
4 gR () 12
2
(17)
In these equations, F 2 denotes the renormalized Maxwell action; here we can impose
fermion-mass-shell renormalization by choosing = m, so that the log term in (17)
drops out. In addition to the (trivial) classical term (17), worldline numerics provides
us with an explicit answer for the renormalized fermion-induced quantum contribution
ferm = d D x Lferm given by (16).
The renormalization scheme employed here can easily be related to standard renormalization prescriptions for Feynman amplitudes. The latter are formulated, for instance,
by considering the renormalized photon propagator D (p) = P D(p2 ) in the Landau
gauge, where P is the transverse projector, and specifying a number R,

R := 2 D 1 (p2 )
(18)
,
p
p 2 =0
which determines the wave function renormalization constant in a particular renormalization scheme. Since the effective action eff given above represents nothing but the generating functional for one-particle irreducible Greens functions, it is related to the renormalized photon propagator by

D 1
(x, y) =
2 eff [A]
.
A (x)A (y)
(19)
We observe that the part (16) does not contribute to the right-hand side of Eq. (18) in the
limit p2 0; this leads us to the desired result

2

1
1
m
1 2
= 2
ln
,
R = 2 D (p )
(20)
2
p
2
gR () 12
p 2 =0
which determines the value of the coupling at any renormalization scale for a given value
of R.
3. Vortex interfaces
The core of a vortex corresponds to a (D 2)-dimensional surface of D-dimensional
Euclidean spacetime. The vortex field which we will discuss below is represented by a
U (1) gauge potential A (x). In an idealized case, the gauge potential is singular at the
vortex surface and can locally be represented by a pure gauge. Considering a closed line C,
164
the vortex can be characterized by the holonomy

exp i A (x) dx = exp{iL},
(21)
where is the magnetic flux carried by the surface, and L is the linking number of C with
the (D 2)-dimensional vortex surface. In the case = , the gauge potential describes a
so-called center vortex in analogy to SU(2) YangMills theory.
In many physical applications (for instance, in the context of center vortices in Yang
Mills theories) one observes that the vortex surface possesses a finite thickness d with
respect to the directions perpendicular to the flux, implying that the gauge potential
singularity in the surface is smeared. We point out that the physics of the extended object
can be quite different from the physics of the idealized vortex. In the latter case, one
removes the (D 2)-dimensional vortex surface from the base manifold. Quantum fields
are acting in the reduced coordinate space and obey certain constraints at the singular
subspace. For purposes of illustration consider the case = 2 . One expects that the free
energy of the vortex is degenerated with the vacuum because the quantum fields of both
cases are related by a gauge transformation (see [26] for an illustration at the 1-loop level).
By contrast, in the case of the smeared vortex surface, the quantum fields experience the
magnetic flux of = 2 . Therefore, one does not expect a degeneracy of the smeared
vortex configuration with the vacuum.
In this first investigation, we will concentrate on plane vortex surfaces. For definiteness,
we consider the gauge potential of a vortex with flux = and core size d of the form
A (x) =
(x2 , x1 , 0, . . . , 0)T ,
2 d 2 + x12 + x22
= 1, . . . , D.
(22)
Since the worldline approach is manifestly gauge invariant, any other choice of a
gauge-equivalent configuration would give the same fermionic interface energy. The
corresponding field strength is given by
F12 (x) =
d 2
.
[d 2 + x12 + x22 ]2
(23)
In the case of D = 2 + 1 dimensions, the fermion-induced effective action is

proportional to the extent Lt of the spacetime in time direction,
(3)
ferm = ELt = Lt 2
(3)
d Lferm (),
(24)
where is the radial coordinate of the xy-plane. E can be interpreted as the energy of the
static vortex soliton. L(3)
ferm () is a radial energy density, i.e., the amount of energy which is
stored in a cylindric shell [, + d]. In the case of D = 3 + 1 dimensions and at a given
time slice, the vortex core is given by a straight line of length Lz . The effective action is
165
proportional to the extent Lt Lz ,

(4)
ferm

= Lt Lz = Lt Lz 2
(4)
d Lferm (),
(25)
where is the string tension of the vortex line in the 3D hypercube.
4. Quantum energy of the single-vortex configuration

4.1. The derivative expansion
The standard analytical approach to effective actions and quantum energies for
nonhomogeneous backgrounds is the derivative expansion. Here, the desired answer is
expanded in terms of a small parameter constructed from derivatives of the background
field. In the present case, there are two options for the expansion parameter, which
can symbolically be written as 2 /m2 1 or 2 /B(x) 1, i.e., the derivatives of the
background field should be smaller than the fermion mass or the local field strength. It
is remarkable that closed-form expressions at the next-to-leading order (NLO) level in
D = 2 + 1 and D = 3 + 1 have been found in recent years which are applicable in both
cases; only one of the conditions mentioned above has to be satisfied. The leading-order
derivative expansion, which agrees with the EulerHeisenberg effective Lagrangian for
constant background fields, is given by [27,28]

Ld0
ferm (x) = Cd0

dT m2 T
B(x)T coth B(x)T 1 + {c.t.} .
e
T
(26)
The next-to-leading (NLO) correction can be written as [29,30]1

Ld1
ferm (x) = Cd1
(i B(x))2
B (x)
3
m2

d B(x)
d
coth()
.
e
d3
(27)
The sum of both Lagrangians provides us with the NLO derivative expansion of the
fermion-induced effective Lagrangian. The prefactors Cd0,1 and exponents , and
depend on the number D of spacetime dimensions and are summarized in Table 1. Note
that the counter term {c.t.} = 13 (B(x)T )2 must be added to the integrand in Eq. (26) for
D = 3 + 1, as discussed in Section 2.2 (cf. Eq. (9)); for D = 2 + 1, there is no counterterm,
{c.t.} = 0.
The expression (27) is valid for any value of the ratio m2 /B, i.e., either m2 or B has
to be large compared with the inverse length scale squared set by the variation of the
background field. Near the vortex core, this length scale is naturally given by the core
1 A representation of the integral in terms of the Hurwitz zeta function as well as asymptotic expansions have
been found; see, for example, [30].
166
Table 1
Prefactors and exponents of the derivative expansion
D =2+1
D=3+1
Cd1
1
4 3/2
1
4(4 )3/2
1
8 2
1
(8 )2
1.5
2.5
0.5
1
3
1
Cd0
size d. For a flux = O(1) considered in the following, we find from Eq. (23) that
B(x)d 2 = O(1) near the core. Simultaneously assuming small masses, md 1, neither
of the possible derivative expansion parameters is small. The NLO Lagrangian reaches
its validity limit, and one cannot expect reliable results from the derivative expansion.
Below, it will turn out that the quality of the NLO approximation strongly depends on
the number of spacetime dimensions. By contrast, for large masses md 1 or for large
radial distances from the vortex core d, we expect the NLO Lagrangian to give a
reasonable answer, since one of the possible expansion parameters is small. Below, we
will use the regime of large masses (or large ) to gain insight into the range of parameters
to which worldline numerics is reliably applicable (see [21] and [25] for a first comparison
of worldline numerics to the constant-field case).
4.2. Worldline numerics: D = 2 + 1
Closely following the procedure outlined in [21], we generate a loop ensemble with
ne loops, where each loop is represented by a set of N spacetime points with coordinates
x1 , . . . , xN . The loop ensembles are generated with appropriate measure (see discussion
in Subsection 2.1). We have used 10 000 and 20 000 dummy sweeps and found that for a
proper thermalization, 10 000 dummy sweeps are sufficient for the accuracy achieved in the
results shown below. Furthermore, we have used N 100, 150, 200 points specifying each
loop. Whereas N = 100 seems sufficient for our purposes in the (2 + 1)-dimensional case,
we take N = 200 to generate high-precision data for the case D = 3 + 1. Generically, we
average over ne = 1000 loop ensembles in the (2 + 1)-dimensional case, and exceptionally
use ne = 16 000 for certain applications in D = 3 + 1. Any dimensionful quantity quoted
in the following is calculated in the simulation in units of vortex thickness d.
In order to avoid a violation of gauge invariance in the numerical computation, we have
to deal with a subtlety concerning the discretization of the integrals along the worldlines:
a particular loop in our ensemble is considered to be a polygon with straight lines Ci
connecting the points xi and xi+1 . In order to evaluate the holonomy in Eq. (4), we consider
the infinitesimal part

exp i A (x)x d
Ci
(28)
167
Fig. 1. Effective Lagrangian as a function of the radial distance to the vortex core for the case m = 1, = 1,
D = 2 + 1. The NLO derivative expansion (solid line) is compared with the numerical computation (squares with
error bars).
for each Ci separately and evaluate the integral analytically using the vortex profile under
consideration. This procedure guarantees that our numerical result is invariant under gauge
transformations of the background gauge field A (x) for any number N of points defining
the polygons. Moreover, the flux enclosed by the polygon is exactly taken into account
as desired (and not only within discretization errors). Of course, this procedure can be
generalized to arbitrary vortex profiles for which the infinitesimal integrations along
the Ci s can be performed numerically; thereby, the properties mentioned above can be
preserved to any numerically desired precision. However, we should stress that the use of
a smooth gauge (e.g., covariant gauges) for the background field is recommended in this
case; this facilitates a fast convergence of the numerical integration.
In order to test our numerical approach, we calculate the effective Lagrangian Lferm ()
for large values of , d, where the derivative expansion is expected to give reliable
results. The result of the numerical worldline approach is compared with the derivative
expansion in Fig. 1 for D = 2 + 1. The agreement between the two curves is satisfactory;
in particular, the numerical approach is able to compute Lferm over a range of many orders
of magnitude. In the region close to the core of the vortex, i.e., d, the gradients of
the background field are as large as the field itself, so that the reliability of the derivative
expansion now depends on the value of the mass. Our numerical findings for this regime
are shown in Fig. 2 for m2 = 1 (part (a)) and m2 = 5 (part (b)). For larger masses, we
observe a good qualitative and a reasonable quantitative agreement. But even for the small
mass value m2 = 1, there is at least qualitative agreement between the numerical result and
the derivative expansion, indicating that the applicability of the derivative expansion can
be pushed to its formal validity limit in D = 2 + 1. This observation is also supported by
the fact that the NLO term (27) is only a small correction to the zeroth-order result (26).
168
(a)
(b)
Fig. 2. Effective Lagrangian in the small region for the cases m = 1 (a) and m2 = 5 (b), = 1, D = 2 + 1.
Moreover, we expect that the (up to now unknown) NNLO correction, which is sensitive to
the curvature of the field strength, improves the result near the vortex core. In this sense, it
is reassuring to observe that the numerical result agrees with the NLO derivative expansion
precisely at the turning point of the curve at 0.5, because the NNLO correction must
vanish here. We should finally stress that for even smaller masses m < 1, the discrepancy
between our numerical result and the derivative expansion increases, so that the derivative
expansion should be abandoned here.
169
(a)
(b)
Fig. 3. Quantum energy as a function of the mass m2 for = 1 (a) and as a function of the flux carried by the
vortex for m = 1 (b) in comparison with Frys upper bound given in Eq. (30).
Let us now examine the fermion-induced quantum energy E of the vortex soliton as
defined in Eq. (24),
(3)
ferm
= 2
E=
Lt

d Lferm (),
(29)
which we obtain by numerically integrating the effective Lagrangian. Our result for
this energy is shown in Fig. 3 as function of the fermion mass m in units of the
170
vortex thickness d. Since fermion fluctuations are suppressed with increasing mass, the
quantum energy decreases with increasing m, and vanishes in the large mass limit. For
phenomenological purposes, it is important to notice that the quantum energy is positive.
This implies that potential effective models for vortex dynamics have to account for the fact
that vortex nucleation is suppressed by the fermion-induced effective action in D = 2 + 1
dimensions.
Furthermore, let us consider the variation of the quantum energy with respect to the
flux , which is carried by the vortex. Our numerical result is shown in Fig. 3(b). We find
that the energy is monotonically increasing with the flux . For = 2 , the vortex
configuration approaches asymptotically ( ) a pure gauge. As in the instanton case,
the energy is nonvanishing due to the finite extension of the vortex core. It is interesting to
compare our result for E() with a general result for fermionic determinants described by
M. Fry in [31]; therein a lower bound has been derived for unidirectional magnetic fields in
D = 2 + 1, which translates into an upper bound Eb for the quantum energy of our vortex
configuration given by
E() Eb (),

1 1

Eb () =
2 3 2 1 + + 1 + + 3 Arsinh( )
d 6
(30)
for the case md = 1 and Dirac 4-component spinors. For other values of md, this formula
receives a total factor of (md)3 and the flux has to be replaced by /(md)2 . As shown
in Fig. 3, our numerical result lies well within this bound. More remarkable is the fact that
the functional dependence of our result agrees with the bound within the error bars, if the
bound is scaled by a factor of roughly 0.65.
As a further check, we have compared all our above-mentioned results with those
of [19], where the single-vortex case with profile functions different from ours was
considered within the phase-shift approach. We find good agreement within the error bars
except for a global factor of 2 by which the result of [19] for E() is larger.
Let us finally point to a problematic feature of our present approach. Note that our
statistical error bars increase in the regime of small fermion masses in Fig. 3(a), which has
the following origin: the Pauli term in Eq. (4),

T
1
,
d F
tr PT exp
2
0
can favor large loops (T d 2 ) for the present unidirectional magnetic field. On the other
hand, the holonomy factor,

T
exp i A (x)x d ,
0
changes its sign rapidly in this case when the loop under consideration is slightly deformed.
This implies that a finite result in the small-mass regime arises after subtle cancellations,
and is therefore hardly accessible to the present numerical formulation. For finite values
171
(a)
(b)
Fig. 4. Effective Lagrangian for m2 = 1 (a) and m2 = 3 (b), = 1, D = 3 + 1.
of the mass, potentially large Pauli term contributions are suppressed by the factor
exp(m2 T ), which solves the numerical problem for large loops (large T ).2 However,
this cancellation problem can become serious for the approach to the chiral limit; here,
a solution to the problem has to be implemented in the algorithm on the analytical level
rather than by brute-force numerical means.
2 In the constant-field case, this cancellation problem already occurs for m2 B as observed in [22]. In the
present vortex case, we can afford much smaller masses. In general, the smaller the coherence length of the
field, the less serious the cancellation problem. The usually considered homogeneous and unidirectional field
configurations are therefore rather pathological in this respect.
172
4.3. Worldline numerics: D = 3 + 1

Similarly to the previous studies of the D = 2 + 1 case, we investigate the effective
Lagrangian Lferm as a function of and compare the result with the one obtained from
the derivative expansion (26), (27) (see Fig. 4). Contrary to D = 2 + 1, we observe that
unless d, the leading order (LO) of the derivative expansion falls far too short of
reproducing the numerical result for m2 = O(1). For instance, for m2 = 1, we find an
order-of-magnitude difference at = 0 (Fig. 4(a)). Moreover, the NLO contribution of the
derivative expansion also exceeds the leading order result by almost an order of magnitude
for d. This signals the break-down of the derivative expansion for the case of the field,
its gradient and the mass being all of the same order. Contrary to the case of D = 2 + 1,
the applicability of the NLO derivative expansion cannot be pushed to its formal validity
limits md 1. Even for larger values of the mass, m2 = 3 (right panel), there is only a
minor improvement of the quality of the NLO derivative expansion. Of course, the NNLO
contribution could, in principle, improve the results of the derivative expansion, but this
would only emphasize the fact that there is no clear hierarchy from term to term in the
derivative expansion.
We believe that the striking difference to the D = 2 + 1 dimensional case is indeed
remarkable and points to a deeper reason in terms of a renormalization effect. To illustrate
this, we note that the quantity Wspin 1 occurring in the propertime integrand is positive
for the vortex background, whereas the counterterm 13 B 2 (x)T 2 is negative. Since the
effective Lagrangian is largely negative as seen in Fig. 4, it is mainly driven by the
counterterm. Now the leading-order derivative expansion obviously overestimates the
value of Wspin 1 near the vortex core, since it is a local expansion. The true value
as seen in the numerical computation is much smaller because it is a nonlocal average over
the extended loop cloud that also feels the much weaker field at a radial distance from the
core. The final value of the total effective Lagrangian at a point x therefore results from a
nontrivial interplay between nonlocal (and nonlinear) vacuum polarization ( Wspin 1 )
and a local definition of the coupling giving rise to a local counterterm. In regions where
the background field varies rapidly, such as the near vortex core in our case, this interplay
can lead to an order-of-magnitude enhancement of the effective Lagrangian as compared
with the constant-field approximation (leading-order derivative expansion). In our opinion,
this phenomenon clearly deserves further investigation.
Returning to our numerical study of the one-vortex background, we calculate the string
tension as defined in Eq. (25) as a function of the fermion mass m and plot it in Fig. 5
for = 1. The negative values of show that the fermion-induced effective action ferm
favors the nucleation of vortices. Since the modulus of this effective action increases if
the vortex thickness d is decreased, the fermionic part of the vortex action supports the
existence of thin vortices. These results are in contrast to those of the case D = 2 + 1,
where the effective action turned out to be positive (see Fig. 3). This sign difference is again
related to renormalization effects mediated by the counterterm. However, we expect that,
in a complete quantum field theoretical treatment, the properties of the vortices are mainly
governed by the pure gauge field part of the effective action, whereas the contribution from
the fermion determinant provides for corrections.
173
(a)
(b)
Fig. 5. Effective action as a function of the dimensionless quantity m2 d 2 , = 1, D = 3 + 1 (a), and as function
of the flux, m2 d 2 = 1 (b).
5. Binary-vortex interactions
Since we are investigating the case of Abelian gauge configurations, the binary-vortex
configuration is given by the superposition of two single vortex gauge fields A (x) (22),

l
l
(2)
A x +
,
A (x) = A x
(31)
2
2
where l denotes the vortex distance. Below, we will study the case of the so-called center
vortices the flux of which is given by = 1. The relative sign between the gauge fields on
the right-hand side of Eq. (31) corresponds to the relative orientation of the fluxes: the plus
sign corresponds to an equal orientation of the flux in each vortex, while the minus sign
signals an opposite orientation.
174
(a) Anti-aligned flux, E = 0.9 2n Emax , n = 0, . . . , 9
(b) Aligned flux, E = 0.9 2n Emax , n = 0, . . . , 9
(3)
Fig. 6. Effective Lagrangian Lferm (x, y) as function of the xy-plane for the two vortex configuration: parallel (a)
and anti-parallel (b) orientation of the flux, m2 d 2 = 0.5, = 1, D = 2 + 1.
Fig. 7. The interaction of two parallel vortex lines in D = 2 + 1, = 1, m2 d 2 = 1.
Fig. 6 shows the lines of equal effective Lagrangian L(3)

eff (x, y) in the xy-plane which is
perpendicular to the vortex fluxes. The vortices are located at the x axis at a distance l = 3d.
It is straightforward (but computer time consuming) to integrate the effective action
over the xy-plane in order to derive the quantum energy E (24) of the binary-vortex
configuration. The result is shown in Fig. 7 for the case D = 2 + 1. For large distances
l d, the quantum energy approaches twice the value of a single vortex. For l = 0, the
vortices fall on top of each other. If the fluxes of the vortices are oppositely oriented, the
175
vortices annihilate each other, and the quantum energy of the configuration vanishes. If
the vortex fluxes are equally oriented, the configuration is equivalent to the single vortex
configuration with flux = 2. Since the quantum energy is roughly proportional to 2 (see
Fig. 3), the vortex configuration with flux = 2 possesses a higher energy than twice
the energy of a single vortex, carrying flux = 1. Hence, vortices with an equal flux
orientation repel each other in D = 2 + 1, while vortices with oppositely oriented flux
attract each other. The same line of argument applies to the case D = 3 + 1. Since the
fermionic contribution to the effective action is negative in this case, the fermion-induced
force is attractive (repulsive) for equally (oppositely) oriented vortices, contrary to the
D = 2 + 1 case.
6. Fermion condensation in a vortex gas

The fermion condensate can be related to the effective action eff (17) by differentiation
with respect to the fermion mass m,

eff
.
d D x0 =
(32)
m
Let us concentrate on D = 3 + 1 dimensions in the following. Here the condensate reads
in worldline representation:

m
m2
d x0
=
4 2
4

4
d x0

dT m2 T
Wspin [A] 1 x ,
e
2
T
(33)
where we have first performed the mass differentiation at an arbitrary renormalization point
and then implicitly chosen on-shell renormalization.
In the present section, we intend to estimate the quark condensate which is generated by
a dilute gas of vortices of flux = 1 in D = 3 + 1. For this purpose, we first calculate the
contribution of the single-vortex configuration with the help of Eq. (33). Due to translation
invariance of our single-vortex background, the string tension (25) is provided in units
of the vortex core size only. This allows us to define the dimensionless function c0 by

m2 2 2
(1) = Lt Lz
c0 m d ,
m d 4 x0
(34)
4 2
where c0 can directly be obtained from worldline numerics,

d T (md)2 T

Wspin [Ad] 1 x .
c0 (m d ) = 2 d
e
2

d d
T
2 2
(35)
= T /d 2 , and
Hereall dimensionful quantities are scaled in units of the core size d, e.g., T
2
2
= x + y measures the radial distance from the vortex core. In the following, we are
interested in a dilute gas of vortices which are static and aligned in the z direction, but
intersect the xy plane at random locations with random fluxes = 1. In this plane the
vortex gas can be characterized by a planar vortex area density V , which is the average
176
number of vortices per xy unit area. The total number of vortices within the 4-dimensional
spacetime is given by NV = V Lx Ly . We expect the dilute-gas approximation to give
reasonable results, if the average distance between two neighboring vortices is at least
2d, where the fermion-induced vortex interactions become small (see, e.g., Fig. 7 for the
D = 2 + 1 dimensional analogue). In other words, there should be less than one vortex
per core-size area, V (d 2 ) 1. In this dilute-gas approximation, the fermion condensate
averaged over spacetime volume V is therefore given by

1
1
(1) =
NV d 4 x0
(36)
mV c0 m2 d 2 .
V
4 2
In the limit of large fermion masses (m 1/d), Eq. (35) can be studied analytically with
the aid of the heat-kernel expansion (8), and we find a contribution to the condensate which
is subleading in 1/m:

1 1 1
d 4 x0 F 2 (x0 ) + O 1/m2 .

(37)
2
24 m V
Here, we recover the familiar result [32] that in the large-m limit the fermion condensate is
proportional to the field strength squared of the background field (i.e., gluon condensate
in a QCD-like language).
We investigate numerically the interesting regime of small masses m, where the heatkernel expansion breaks down. The representation (35) is highly convenient for this
purpose. However, as already pointed out, the loop average in Eq. (34) is plagued from
a severe cancellation problem which in the present case of the loop parameters employed
here limits the region of validity to T /d 2 100. In order to ensure this limit, we confine
ourselves to the mass range m2 Tmax 10, i.e., m2 > 0.1/d 2 . The investigation of the
small mass regime (m2 < 0.1/d 2 ) clearly needs further study. Work in this direction is
in progress [33].
Our numerical result for c0 is presented in Fig. 8. For small masses, i.e., 0.1 < m2 d 2 <
0.5, we find that c0 (m2 d 2 ) 1/(md), which indicates that the condensate approaches a
plateau in this regime according to Eq. (36),

2 2
1
1

m2 d 2 =O(0.1)
(38)
mc0 m d
0.058 2 ,
2
4
4 d
m2 d 2 =O (0.1)
as depicted in Fig. 8(b). However, we do not believe that this result extends to the chiral
limit m 0. Guided by the case of a constant background field in D = 3 + 1 [35], we
expect that the chiral condensate vanishes according to

m ln m for m 0. As
seen from Fig. 8, our numerical result for c0 does not discriminate between the ln m and
the desired 1/m behavior in the small-mass regime.
Let us stress the two main findings of this subsection: first, the vortex-gas-induced
condensate is characterized by a comparably low value of c0 for all mass values depicted
above and an onset of the plateau value for comparably large masses m2 d 2 0.1; we
point out that it is not possible to fit the form of the condensate depicted above within
a derivative expansion even qualitatively by arbitrarily varying B. Second, in the context
of a center vortex model for low-energy QCD, the current quark masses are in fact finite,
though small. The present result indicates that contributions to the chiral condensate have to
177
(a)
(b)
Fig. 8. c0 (Eq. (34)) (a) and the fermion condensate (b) as function of the fermion mass m, = 1, D = 3 + 1.
be expected from an interaction of the massive quarks with the gauge vacuum (modeled by
a random vortex background). As a first estimate, we may insert parameter values known
from lattice calculations. In the case of an SU(2) gauge theory, the density of the center
vortices is roughly given by V = 3.6/fm2 . The planar vortex correlation function was
also studied in SU(2) lattice gauge theory [34]. One finds an exponential decrease of the
correlation function. This allows for a definition of a vortex thickness of d 0.30.4 fm.
Since V d 2 = O(0.1) in this case, the dilute-gas approximation should be applicable.
From Eq. (38), we find for the vortex-induced condensate qq
[50 MeV]3 , for masses
2
2
m d = 0.1, i.e., m 100 MeV. As discussed above, the present algorithm cannot treat the
case of light mass values, m 510 MeV.
The vortices considered here obviously possess a trivial topology, i.e., they do not
contribute to the topological charge. The origin of a chiral condensate here is entirely
relegated to the field strength carried by the vortex cores. In a fully non-Abelian QCD-like
context, topologically nontrivial vortex configurations (due to the presence of low-lying
178
fermionic modes of the Dirac operator) as well as multi-gluon exchange interactions can
be expected to provide for a drastic enhancement of the condensate.
7. Conclusions
The fermion-induced quantum action of Abelian vortex configurations has been studied
in the case of D = 2 + 1 and D = 3 + 1 dimensions using worldline numerics [21,22,
25]. The quantum action of a single-vortex configuration is characterized by the fermion
mass m, the vortex thickness d and the flux carried by the vortex. Our numerical approach
has been successfully tested in the parameter regime where the derivative expansion is
expected to provide reliable results: in the large-mass regime md 1 or for strong-field
suppression of the inhomogeneities 2 /B 1.
Compared with our numerical results in D = 2 + 1, the derivative expansion provides
for a reasonable approximation to the quantum energy of a single vortex configuration even
for smaller masses, md 1. The next-to-leading order only adds a small correction to the
leading-order result. By contrast, in D = 3 + 1, the derivative expansion is insufficient
even for comparably large masses, md = O(5). Only for very large masses, md 1, can
the derivative expansion be trusted. We have argued that this disparity between D = 2 + 1
and D = 3 + 1 arises from renormalization effects in the latter case, which can occur near
regions where the background varies rapidly. In particular, these renormalization effects
can lead to an effective enhancement of the quantum action. Further investigations are
planned to settle this issue.
From a physical point of view, we find that the quantum action is positive in D = 2 + 1,
implying that the presence of vortices is suppressed, and large vortex core sizes d are
preferred. In D = 3 + 1, the properly renormalized quantum action turns out to be negative:
the nucleation of thin (d 0) vortices in D = 3 + 1 is supported by the fermion induced
quantum action.
Subsequently, the binary-vortex interaction was investigated. We found that the
fermion-induced interaction favors vortices with an opposite orientation of the fluxes in
D = 2 + 1, while in D = 3 + 1, a unique orientation of the fluxes is preferred.
It should be stressed that these statements refer to and are derived from the fermioninduced action. In a pure QED context, the classical action has to be taken into account. The
latter will dominate the fermion-induced action by far at weak coupling, reflecting the usual
hierarchy between classical and quantum-induced nonlinear electrodynamics. Formally,
the classical and quantum action can become comparable in magnitude for exponentially
strong fields in the vortex core (exponentially small d for fixed flux); however, this is
nothing but a manifestation of the Landau pole of QED and thus should be rated as
unphysical.
Finally, the fermion condensate which is generated by a dilute gas of vortices was
studied as function of the fermion mass md > 0.1 in D = 3 + 1. We found that the
condensate decreases like 1/m for large values of the mass and that it reaches a plateau
for md 0.5. Similarly to the case of a constant magnetic field, we expect that the
condensate vanishes like m ln m in the chiral limit m 0. However, our results point to an
interesting phenomenon: although chiral symmetry breaking might be tied to topological
179
properties of the background fields not covered by the present considerations, the field
strength carried by the vortex cores enhances the chiral condensate in the intermediate
fermion mass regime. In a center vortex model of low-energy QCD, this effect may lead
to a nonnegligible contribution to the condensate. We regret that our worldline numerical
algorithm in its present form cannot address the small-mass regime md < 0.1 due to severe
cancellations. The study of the chiral limit in general and chiral symmetry breaking by
vortex background fields in particular requires improved algorithms and is left to future
work [33].
Acknowledgements
The authors are grateful to W. Dittrich, H. Reinhardt and H. Weigel for helpful
discussions and detailed comments on the manuscript. H.G. acknowledges the support of
the Deutsche Forschungsgemeinschaft under contract Gi 328/1-1. L.M. was supported by
the Deutsche Forschungsgemeinschaft under contract GRK683.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
D.B. Kaplan, Phys. Lett. B 288 (1992) 342, hep-lat/9206013.

Y. Kikukawa, Nucl. Phys. B 584 (2000) 511, hep-lat/9912056.
P. Chen, et al., Phys. Rev. D 59 (1999) 054508, hep-lat/9807029.
V.A. Rubakov, Phys. Usp. 44 (2001) 871, Usp. Fiz. Nauk 171 (2001) 913, hep-ph/0104152.
G.W. Crabtree, D.R. Nelson, Phys. Today 50 (1997) 38.
K. Langfeld, H. Reinhardt, O. Tennert, Phys. Lett. B 419 (1998) 317, hep-lat/9710068.
L. Del Debbio, M. Faber, J. Greensite, S. Olejnik, Nucl. Phys. Proc. Suppl. 53 (1997) 141, hep-lat/9607053.
L. Del Debbio, M. Faber, J. Giedt, J. Greensite, S. Olejnik, Phys. Rev. D 58 (1998) 094501, hep-lat/9801027.
K. Langfeld, O. Tennert, M. Engelhardt, H. Reinhardt, Phys. Lett. B 452 (1999) 301, hep-lat/9805002.
M. Engelhardt, K. Langfeld, H. Reinhardt, O. Tennert, Phys. Rev. D 61 (2000) 054504, hep-lat/9904004.
T.G. Kovacs, E.T. Tomboulis, Nucl. Phys. Proc. Suppl. 106 (2002) 670, hep-lat/0110123;
T.G. Kovacs, E.T. Tomboulis, hep-lat/0108017.
M. Engelhardt, H. Reinhardt, Nucl. Phys. B 585 (2000) 591, hep-lat/9912003.
M. Engelhardt, Nucl. Phys. B 585 (2000) 614, hep-lat/0004013.
N. Graham, R.L. Jaffe, H. Weigel, Int. J. Mod. Phys. A 17 (2002) 846, hep-th/0201148.
N. Graham, R.L. Jaffe, M. Quandt, H. Weigel, Phys. Rev. Lett. 87 (2001) 131601, hep-th/0103010.
F. Gygi, M. Schlter, Phys. Rev. B 43 (1991) 7609.
N. Schopohl, K. Maki, Phys. Rev. B 52 (1995) 490;
N. Schopohl, cond-mat/9804064.
P. Gornicki, Ann. Phys. 202 (1990) 271;
Y.A. Sitenko, A.Y. Babansky, Mod. Phys. Lett. A 13 (1998) 379, hep-th/9710183.
P. Pasipoularides, Phys. Rev. D 64 (2001) 105011, hep-th/0012031.
M. Bordag, K. Kirsten, Phys. Rev. D 60 (1999) 105019, hep-th/9812060.
H. Gies, K. Langfeld, Nucl. Phys. B 613 (2001) 353, hep-ph/0102185.
H. Gies, K. Langfeld, Int. J. Mod. Phys. A 17 (2002) 966, hep-ph/0112198.
Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 1669;
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451;
M.J. Strassler, Nucl. Phys. B 385 (1992) 145, hep-ph/9205205;
A.M. Polyakov, Gauge Fields and Strings, in: Contemporary Concepts in Physics, Vol. 3, Harwood, Chur,
Switzerland, 1987.
180
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
C. Schubert, Phys. Rep. 355 (2001) 73, hep-th/0101036.

M.G. Schmidt, I.O. Stamatescu, hep-lat/0201002.
D. Diakonov, Mod. Phys. Lett. A 14 (1999) 1725, hep-th/9905084.
W. Heisenberg, H. Euler, Z. Phys. 98 (1936) 714.
J. Schwinger, Phys. Rev. 82 (1951) 664.
D. Cangemi, E. DHoker, G.V. Dunne, Phys. Rev. D 51 (1995) 2513, hep-th/9409113.
V.P. Gusynin, I.A. Shovkovy, J. Math. Phys. 40 (1999) 5406, hep-th/9804143.
M.P. Fry, Phys. Rev. D 54 (1996) 6444, hep-th/9606037.
M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147 (1979) 385.
L. Moyaerts, H. Reinhardt, H. Gies, K. Langfeld, in preparation.
M. Engelhardt, K. Langfeld, H. Reinhardt, O. Tennert, Phys. Lett. B 431 (1998) 141, hep-lat/9801030.
W. Dittrich, M. Sieber, J. Phys. A 21 (1988) L711;
W. Dittrich, H. Gies, in: Springer Tracts in Modern Physics, Vol. 166, Springer, Berlin, 2000, p. 1.

Non-singlet structure functions at three loops:

Fermionic contributions
S. Moch a,b , J.A.M. Vermaseren c , A. Vogt c
a Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
b Deutsches Elektronensynchrotron DESY, Platanenallee 6, D-15738 Zeuthen, Germany
c NIKHEF Theory Group, Kruislaan 409, 1098 SJ Amsterdam, The Netherlands
Received 17 September 2002; accepted 25 September 2002
Abstract
We compute the fermionic (nf ) contributions to the flavour non-singlet structure functions in
unpolarized electromagnetic deep-inelastic scattering at third order of massless perturbative QCD.
Complete results are presented for the corresponding nf -parts of the three-loop anomalous dimension
and the three-loop coefficient functions for the structure functions F2 and FL . Our results agree with
all partial and approximate results available in the literature. The present calculation also facilitates
a complete determination of the threshold-resummation parameters B2 and D2DIS of which only
the sum was known so far, thus completing the information required for the next-to-next-to-leading
logarithmic resummation. We find that D2DIS vanishes in the MS scheme.
PACS: 12.38.Bx; 12.38.Cy; 13.60.Hb
1. Introduction
Structure functions in deep-inelastic scattering (DIS) form the backbone of our
knowledge of the protons parton densities, which are indispensable for analyses of
hard scattering processes at proton(anti-)proton colliders like the T EVATRON and the
future LHC. Structure functions are also among the quantities best suited for precisely
measuring the strong coupling constant s . Over the past twenty years DIS experiments
have proceeded to a high (few-percent) accuracy and a wide kinematic coverage [1]. More
results, especially at high scales Q2 , can be expected from the forthcoming high-luminosity
E-mail address: avogt@nikhef.nl (A. Vogt).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 7 0 - 2
182
S. Moch et al. / Nuclear Physics B 646 (2002) 181200
phase of the electronproton collider HERA at DESY. On the theoretical side, at least the
next-to-next-to-leading order (NNLO) corrections of perturbative QCD need to be taken
into account in order to make full use of these measurements and to make quantitatively
reliable predictions for hard processes at hadron colliders.
The calculation of NNLO processes in perturbative QCD is far from easy. For deepinelastic structure functions, in particular, the current situation is that, while the coefficient
functions are known to two loops [26], only six/seven integer Mellin moments of
the corresponding three-loop anomalous dimensions have been computed for lepton
hadron [79] and leptonphoton DIS [10], together with the same moments of the threeloop coefficient functions. The hadronic results have been employed, directly [1114] and
indirectly [15,16] via x-space approximations constructed from them [1719], to improve
the data analysis and some hadron-collider predictions. However, the number of available
moments is rather limited, and hence these results cannot provide sufficient information at
small values of the Bjorken variable x.
For the complete information one needs to obtain either all even or odd (depending
on the quantity under consideration) Mellin moments, or do the complete calculation in
Bjorken-x space. We have adopted the first approach. Following the formalism of Refs. [7,
8,20,21] to obtain the lower fixed moments, we have used recursive methods to extend the
calculation to all values of the Mellin moment N . This is by no means trivial, since before
the start of the calculation the mathematics of the answer was still poorly understood [22].
Hence it was first necessary to develop the understanding of harmonic sums [2326] and
harmonic polylogarithms [2729]. In addition the Mellin transforms and the inverse Mellin
transforms from Bjorken-x-space to Mellin space and back had to be solved [29]. These
conceptual problems have been overcome and the method has been shown to work in a
complete re-calculation of the two-loop coefficient functions [6].
The concept of working in Mellin space is not new. This method was already used in the
early QCD papers [3032]. But even in the case of the two-loop anomalous dimensions it
was still possible to do the resulting sums in a rather direct manner [23,33]. This changed
with the two-loop evaluation of L /T in which Kazakov and Kotikov [34,35] managed to
obtain some of the integrals via recursion relations or first order difference equations.
In this article, we present the fermionic (nf ) corrections to the flavour non-singlet
structure functions in electromagnetic DIS at the three-loop level. This includes the threeloop anomalous dimensions which are needed for the completion of the NNLO calculation,
as well as the three-loop coefficient functions which are, at least at large x, the most
important contribution to the next-to-next-to-next-to-leading order (N3 LO) correction [36].
Of course, the nf -part is not the complete calculation. Yet we decided to present it already
now, because the complete calculation (including the singlet part) will still take quite some
(computer) time. Thus, for the time being, NNLO calculations of the DrellYan process
[37,38] (which are relevant for luminosity monitoring at T EVATRON and LHC [3941])
and of Higgs production [38,42] have to rely on parton distributions evolved with the
approximate splitting functions of Ref. [19]. Our present calculation provides a first check
of the reliability of these approximations. More importantly, it turns out that already this
calculation is sufficient to provide the last relevant missing information for the extension
of the soft-gluon (threshold) resummation for DIS [4345] to the next-to-next-to-leading
183
logarithmic accuracy [46]. In fact, our result for the resummation parameter D2DIS is most
intriguing and calls for further studies.
This article is organized as follows. In Section 2 we outline those parts of the calculation,
which differ from previous two-loop [6] and fixed-N three-loop [79] analyses. The
present calculation does not yet involve the full complexity of the method, as the most
difficult diagram topologies do not occur. Therefore we postpone a full account to a later
publication. In Section 3 we present our explicit even-N Mellin-space results, except for
the rather lengthy expressions for the three-loop coefficient functions which are deferred to
Appendix A. The corresponding three-loop quantities in x-space can be found in Section 4.
Here, instead of writing down the cumbersome exact expressions for the coefficient
functions, we follow the procedure applied in Refs. [17,18] to the two-loop coefficient
functions, and provide approximate parametrizations which are compact and sufficiently
accurate for all numerical applications. In Section 5 we then discuss the implications of our
calculation on the soft-gluon resummation, before we summarize our results in Section 6.
2. Method
Because we are considering only the non-singlet structure functions in this paper, the
method for the calculation of their moments closely follows Ref. [7]. Hence there is not
much need to explain the physics of the method here again. Thus we will discuss only
the differences introduced by the fact that we now compute all moments simultaneously
as a function of the moment number N . Since N is not a fixed constant, we cannot resort
to the techniques of Ref. [7], where the Mincer program [47,48] was used as the tool to
solve the integrals. Instead, we will have to introduce new techniques. However, we can
give N a positive integer value at any point of the derivations and calculations, after which
the Mincer program can be invoked to verify that the results are correct. From a practical
point of view this is the most powerful feature of the Mellin-space approach, as it greatly
simplifies the checking of all programs.
Similar to the fixed-N computations of Refs. [8,9], the diagrams are generated
automatically with a special version of the diagram generator QGRAF [49]. For all
the symbolic manipulations of the formulae we use the latest version of the program
FORM [50]. The calculation is performed in dimensional regularization [5154] with
D = 4 2. Hence the unrenormalized Mellin-space results will be functions of , N,
and the values 3,...,5 of the Riemann -function. The renormalization is carried out in the
MS-scheme [55,56] as described in Ref. [7].
We distinguish three categories of diagrams: complete diagrams, composite building
blocks and basic building blocks. A complete diagram is a Feynman diagram with all
its structure like traces and dotproducts in the numerator. Such a diagram may lead to
a large number of more fundamental integrals that cannot be reduced by considerations
of momentum conservation only. For the understanding of composite and basic building
blocks, one has to realize that in the framework of the operator-product expansion we
eventually have to take N derivatives with respect to the parton momentum P after
which P is put equal to zero. This projects out the N th Mellin moment [57] and it
effectively amputates the legs of the parton, leaving us with propagator-type diagrams.
184
Fig. 1. The topologies BE (left) and O4 (right) of propagator-type diagrams, with the line numbering as employed
in Figs. 2 and 3 below. The external lines carry the momentum Q.
Fig. 2. The diagrams of topology BE which contribute to the fermionic part of the non-singlet structure functions
F2 and FL . The subtopologies are BE1Q (left) and BE13 (right).
Therefore, we define the topology of a diagram as the propagator topology when the
P -momentum legs have been amputated. The three-loop propagator topologies of the BE
(Benz) type and of the O4 type are shown in Fig. 1. For the notation we refer to Refs. [47,
48]. The external lines in the propagator topology are referred to as Q.
When the numbers of the position of the incoming and outgoing momenta have been
attached we are referring to subtopologies. For instance, BE13 is a subtopology of type BE
in which the momentum P comes in line 1 and leaves in line 3, assuming the numbering
of the BE topology as in Fig. 1. We define basic building blocks (BBB) as integrals in
which both the incoming and the outgoing P -momentum are attached to the same line
as, e.g., in BE22 . In composite building blocks (CBB), on the other hand, the incoming
and the outgoing P -momentum are attached to different lines, as in the case of BE13
mentioned above. Of course we could have introduced names for all these three-loop
four-point functions, but since eventually the P -momentum legs get amputated the above
notation seems the clearest scheme. In this way, it is only a small step to an easy pictorial
representation of the integrals as used in Ref. [6].
For the calculation of the nf -parts of the non-singlet structure functions F2 and FL , one
does not need to consider all three-loop topologies. In fact, the only genuine three-loop
subtopologies one has to solve are of the BE-type and of the O4-type, with two complete
diagrams of either type entering the calculation. These diagrams are displayed in Figs. 2
and 3.
The main problem we are faced with as compared to the corresponding two-loop
calculation [6] is that the necessary reduction equations become much more complicated.
In addition the set of equations that was available at that moment was not maximal, and the
185
Fig. 3. The diagrams of topology O4, which contribute to the fermionic part of the non-singlet structure functions
F2 and FL . These diagrams are examples of the subtopologies O41Q and O418 .
structure of the more complicated topologies needs the maximal set of reduction equations.
This involves some equations in which the vanishing of P P is an issue that should be
considered with care, a point that will be explained in full detail in a later publication in
which we have to deal with all topologies.
For this calculation we first studied the basic building blocks. Here it is easy to expand
the single propagator that contains the momentum P to sufficient powers in P for the N th
moment. This number of powers can be less than N as there may be some powers of P in
the numerator already. It should be noted that if we have a power of P Q in the numerator
we are effectively computing the moment N 1 of the integral. At this point we write
down all equations based on integration by parts, all scaling relations and all form-factor
equations that can be constructed. Next follows a potentially difficult process in which we
have to combine these relations to construct equations that can systematically bring the
powers of the denominators in the integral down, either reducing them to zero or leaving
them at a fixed unique value. When a line is eliminated a simpler topology is reached and
we can refer to the reduction equations for that topology. This will eventually lead to a
topology that is simple enough so that we can calculate the integral.
A problem arises when a reduction equation cannot take the power n of a denominator
beyond one, because it is of the type nI (n + 1) I (n). Also when the power of a
propagator is not an integer, one can only try to reduce this power to a fixed value for which
the line does not vanish. In these cases we leave the power either at one or at this fixed value
(usually 1 + ) and we continue with the next propagator. Eventually we will have integrals
in which only the line with the power that involves N will not have a standard value. The N
will occur both in the power of the denominator p p and in the numerator 2P p when p
is the momentum associated to this line. By this time our remaining equations which have
also been treated to bring the powers of their propagators to standard values may have
become rather lengthy. But we still need them to bring the difference of the powers in the
numerator and the denominator to a fixed value, which can be either 0, 1 or 2, something
we have to leave open for reasons explained below. Now there are several possibilities:
if there is only one integral left of the type to be solved, the equation directly determines
the solution. This is, however, rarely the case. Usually there are several terms left, each
with a different power of 2P Q, leaving us with an equation of the type
a0 (N)I (N) + a1 (N)I (N 1) + + am (N)I (N m) = G(N)
(1)
in which the function G refers to a potentially horrendous combination of integrals of

simpler topologies. Eq. (1) defines a recursion relation or difference equation of order m.
186
For the present calculation we did not have to go beyond order 2. It should be noted that
recently difference equations have been encountered by Laporta in Refs. [58,59].
These difference equations can be solved by making the ansatz that the solution will
be a combination of harmonic sums. If the proper combination has been selected, the
coefficients of the harmonic sums can be obtained by substituting the trial solution into
the equation and solving the resulting, potentially large set of linear equations. There can
be several thousand equations in such a system. Of course one has to have a solution for
the function G(N) in Eq. (1) first, and m 1 boundary values are required. Because these
boundary values are basically fixed integer-N moments, they can be obtained using the
Mincer program. The need for knowing the function G puts a rigid hierarchy in the order
in which we have to treat the topologies.
Solving the difference equations is rather slow work. Hence we compute their solutions
only once and tabulate the results. We usually do this for several values of the difference of
the powers of the numerator and the denominator as mentioned above Eq. (1). The reason
behind this is that the equations we use for either raising or lowering this difference may
contain a so-called spurious pole in when we try to bring this value to one. The concept of
spurious poles is rather important. The rule of the triangle [60,61], for instance, can involve
a factor proportional to 1/ when an integral is being reduced. Close inspection reveals
that when powers of the loop momentum are present in the numerator, it is possible that
more than one of such poles is generated before a denominator is removed. The resolved
triangle [62] shows, however, that it is possible to sum all contributions of such a reduction
and that eventually there can be no more than one pole per eliminated line. This means that
the extra poles should cancel between the many generated terms. But when we work only a
to given cutoff in powers of (both for reasons of economy and because we cannot evaluate
some integrals easily beyond certain powers in ) these temporary poles could spoil the
final result in the same way as such things can happen in numerical calculations at fixed
precision. We call them spurious poles, because in principle they can be avoided. One of the
greatest difficulties in deriving reduction equations is to indeed avoid such spurious poles.
For some integrals, no spurious-pole free formula could be found for the last reduction
when bringing the difference of the power of the denominator and the numerator either
from 2 to 1 or from 0 to 1. This problem has been circumvented by solving the resulting
difference equations for the three cases 0, 1 and 2 separately.
Since the evaluation of all the basic building block integrals that can occur requires
very much computer time, we decided to tabulate all these integrals for the complicated
topologies. This saves much computer time, because each integral is typically used many
times. Also the more complicated topologies use many integrals of a less complicated type,
hence their evaluation becomes much faster once the latter integrals have been tabulated.
The next step is the evaluation of the composite building blocks. Here again we first
construct all possible equations. It turns out to be most economic to leave the propagators
with the momentum P unexpanded. The fact that eventually an expansion to N powers
of P will take place then requires some special calculational rules. If, for instance, an
equation is multiplied by P Q, we have to replace N by N 1 at the same time. Such
rules can be major sources of errors. Hence it is very fortunate that at any moment we
can decide that N has a fixed value like four or five and then evaluate the integrals in the
equation with the Mincer program to see whether it is still correct.
187
The equations are used to set up a reduction scheme that is similar to the one for the
BBBs. If a line is eliminated we obtain either a simpler topology or a BBB. Otherwise we
reduce the power of a denominator to a standard value and continue with the next line.
In numerous cases a reduction can only be done by means of a difference equation (1). If
this is a first order difference equation it can be solved directly, introducing one sum. Such
sums are of a benign type and can be evaluated afterwards. If the difference equation is
of a higher order we have to consider all more fundamental integrals first before we can
solve the equation. The further reduction scheme becomes then rather complicated, but not
impossible. On the average each subtopology can require several weeks of work before it
has been completely solved by these methods.
Analogous to the BBB case discussed above, we have tabulated the more complicated
CBB integrals entering the calculation. It is not only a matter of computer time that renders
this necessary. Also the size of the intermediate expressions becomes a most relevant
factor: if one is not careful, even a hard disk of 100 GBytes can become restrictive.
In practice, already expressions significantly larger than 10 GBytes took too long for
evaluation without further optimization. A careful hierarchy of tabulation managed to avoid
these problems.
Having programs for all basic and composite building blocks renders the remainder
of the calculation rather straightforward. The major difference to the fixed-moment
calculations is now that we obtain much longer results due to the presence of the
parameter N in the answer. We have checked the correctness of each individual diagram
for several values of N , by comparing with the results of a Mincer calculation. In addition
we have compared the complete renormalized results with the results in the literature for
the available values of N .
3. Results in Mellin space

Here we present the N -space coefficient functions and the anomalous dimensions up
to the third order in the renormalized coupling s . All results are given in the MSscheme with the renormalization scales identified with the physical hard scale Q. Thus the
perturbative expansion of the non-singlet coefficient functions and anomalous dimensions
can be written as

s n (n)
ci,ns (N),
Ci,ns (s , N) =
(2)
4
n=0

s n+1 (n)
ns (s , N) =
(3)
ns (N)
4
n=0
with i = 2, L in Eq. (2). There is no need to consider different choices of the scales, as they
do not require functions beyond those introduced in Eqs. (2) and (3). Recall that for F2 the
(n + 1)-loop anomalous dimensions and the n-loop coefficient functions together form the
Nn LO approximation of (renormalization-group improved) perturbative QCD.
There are, actually, three different non-singlet combinations of coefficient functions and
splitting functions. These combinations all coincide at order s , but they all differ beyond
188
the second-order. Only the so-called +-combinations (involving sums over quarks and
antiquarks) are probed in electromagnetic DIS, hence only these quantities are addressed
in the present article. Consequently our results below apply directly only to all even-integer
values of N from which, however, the results for arbitrary N can be uniquely inferred by
analytic continuation.
Our N -space results are expressed in terms of harmonic sums Sm (N). In the following
all harmonic sums are understood to have the argument N , i.e., we employ the shorthand notation Sm Sm (N). In addition we use operators N and Ni which shift the
argument N of a given function by 1 or a larger integer i,
N f (N) = f (N 1),
Ni f (N) = f (N i).
(4)
We normalize the trivial leading-order (LO) coefficient function and recover, of course,
the well-known result for the LO anomalous dimension [30,31]
(0)
c2,ns (N) = 1,

(0)
(N) = CF 2(N + N+ )S1 3 .
ns
(5)
(6)
In our notation, the next-to-leading order (NLO) non-singlet coefficient function for F2
[56] and the corresponding anomalous dimension [23,33] read

(1)
c2,ns
(N) = CF 7N+ S1 + 2S1 9 + (N + N+ )[2S1,1 3S1 2S2 ] ,

(1)
28
ns
(N) = 4CA CF 2N+ S3 17
24 2S3 3 S1

11
+ (N + N+ ) 151
S
+
2S
S
1
1,2
2
18
6

1
1
+ 43 S1 (N + N+ ) 11
S
S
+ 4CF nf 12
9 1
3 2

+ 4CF2 4S3 + 2S1 + 2S2 38 + N [S2 + 2S3 ]

(N + N+ )[S1 + 4S1,2 + 2S1,2 + 2S2,1 + S3 ] .
(7)
(8)
At next-to-next-to-leading order (NNLO, N2 LO), we have re-calculated the coefficient

(2)
function c2,ns
of Refs. [2,6], and we have computed all fermionic contributions to the
(2)
, the nf1 terms being a new result of this article. The fermionic NNLO
splitting function ns
correctionsthe complete expression for the N -space coefficient function can be found in
Ref. [6]are given by

89
(2)
7
c2,ns (N) = 4CF nf (1 N+ ) 122
27 S1 + 6 S1,1 (N 1) 108 S1 S2

1
2
1
(N + N+ ) 56 S3 + 13
18 S1,1 + 3 S1,1,1 3 S2,1 3 S1,2

247
19
16 S1,1 + 457
S
+
N
S
,
1
+
2
144
108
6
(9)

(2)
(N) = 16CA CF nf
ns

32 3 54 +
+ 43 S1,2 23 S4 + 2S1,1 25
9 S2

2
2
2
23
+ 257
27 S1 3 S3,1 N+ S2,1 3 S3,1 3 S4 + (1 N+ ) 18 S3 S2

11
317
16
2
(N + N+ ) S1,1 + 1237
216 S1 + 18 S3 108 S2 + 9 S1,2 3 S1,2,1

13 S1,3 12 S1,3 12 S2,1 13 S2,2 + S1 3 + 12 S3,1

2
17
2
11
1
13
S
+
S
+
(N
+
N
)
S
S
+
S
+ 16CF nf2 144
1
2
+
1
2
3
27
9
9
54
18

2
23
3
4
59
4
20
20
+ 16CF nf 16 2 3 + 3 S3,1 36 S2 + 3 S4 9 S3 + 9 S1

4
1
83 S1,2 83 S1,1 43 S1,2 + N+ 25
9 S3 3 S3,1 3 S4

4
4
+ (1 N+ ) 67
36 S2 3 S2,1 + 3 S3

2
32
4
4
+ (N + N+ ) S1 3 325
144 S1 3 S1,3 + 9 S1,2 3 S1,2,1 + 3 S1,1

4
11
2
10
1
2
8
+ 16
9 S1,2 3 S1,3 + 18 S2 3 S2,2 + 9 S2,1 + 2 S4 3 S2,2 9 S3 .
10
9 S3
189
10
9 S3
(10)
The corresponding formulae for the longitudinal coefficient function CL,ns are deferred
to the appendix, together with the rather lengthy N -space results for the fermionic parts of
(3)
the third-order coefficient functions ci,ns
, i = 2, L, which partly also represent new results
(3)
of this article (the nf2 term for F2 has already been presented in Ref. [63]). Notice that cL,ns
(3)
can be considered a NNLO quantity, since CL vanishes at order s0 . On the other hand c2,ns
represents, at least at large N , the dominant part of the N3 LO corrections to F2 [36].
As briefly mentioned at the end of Section 2, we have subjected our results to a number
of checks. First of all, we have calculated some lower even moments in an arbitrary
covariant gauge with the Mincer program [47,48], keeping the gauge parameter in the
gluon propagator. All dependence on does cancel in the final results. Secondly the
(2)
nf2 -contribution to ns is known from the work of Gracey [64] and we agree with his
(3)
result. Furthermore the coefficients of lnk N , k = 3, . . . , 5, of c2,ns (N) agree with the
prediction of the soft-gluon resummation [65]. Finally, we have checked the result of
each individual diagram for several integer values of N by comparing with the results
of a Mincer calculation. Thus, as the strongest check, our results reproduce the fixed even
moments N = 2, . . . , 14 computed in Refs. [79].
4. Third-order results in x-space

The x-space coefficient functions and the splitting functions are obtained from the
results of the previous section by an inverse Mellin transformation, which maps the
harmonic sums of moment space [2326] to harmonic polylogarithms in x-space [2729].
This transformation can be performed by a completely algebraic procedure [6,29] based
190
(2)
(2)
Fig. 4. The nf1 and nf2 parts P+,1 (x) and P+,2 (x) of the three-loop non-singlet splitting function (12), multiplied
by (1 x) for display purposes. Also shown in the left part (dashed curve) is the uncertainty band derived in
Ref. [19] from the lowest six even-integer moments [79].
on the fact that the harmonic sums also occur as coefficients of the Taylor expansion of
harmonic polylogarithms.
Here we confine ourselves to the third-order results; for the two-loop non-singlet
splitting functions and coefficient functions the reader is referred to Refs. [2,6,67]. For
brevity the exact results are written down only for the splitting function, conventionally
related to the anomalous dimension (3) by
1
(n)
(N) =
ns
(n)
dx x N1 Pns
(x).
(11)
0
(n)
The fermionic part of Pns involves only simpler harmonic polylogarithms which can
be expressed in terms of the usual (poly-)logarithms. The x-space analogue of Eq. (10),
graphically displayed in Fig. 4, can thus be written as
(2)
(x) = 16CA CF nf
Pns

pqq (x) 59 2
ln(x) + 13 ln(x)2 14 ln2 (x) ln(1 x)

7
1
12
ln2 (x) 18
ln3 (x) 12 ln(x)Li2 (x) + 13 Li3 (x)

+ pqq (x) 12 3 59 2 23 ln(1 + x)2 + 16 ln(x)2
209
216
32 3
167
108
2
3
10
5
1 2
1
9 ln(x) ln(1 + x) + 18 ln (x) 6 ln (x) ln(1 + x) + 18 ln (x)

1
1
2
10
9 Li2 (x) 3 Li3 (x) 3 Li3 (x) + 3 H1,0,1 (x)

+ (1 + x) 16 2 + 12 ln(x) 12 Li2 (x) 23 Li2 (x) 23 ln(x) ln(1 + x)

2
1
17
1
+ 24
ln2 (x) + (1 x) 13 2 257
54 + ln(1 x) 9 ln(x) 24 ln (x)
191

1 2
25
+ (1 x) 54 167
54 2 + 20 2 + 18 3

5

1
1
1
+ 16CF nf2 pqq (x) 54
ln(x) 54
+ 36
ln2 (x) + (1 x) 13
54 + 9 ln(x)

17
5
(1 x) 144
27
2 + 19 3

5
1
10
+ 16CF2 nf pqq (x) 53 3 55
48 + 24 ln(x) + 3 ln(x)2 + 9 ln(x) ln(1 x)

1
ln3 (x)
+ 14 ln2 (x) + 23 ln2 (x) ln(1 x) + 23 ln(x)Li2 (x) 23 Li3 (x) 18

4
1
5 2
+ pqq (x) 10
9 2 3 + 3 ln(1 + x)2 3 ln(x)2 9 ln (x)
+
20
1 3
1 2
20
9 ln(x) ln(1 + x) 9 ln (x) + 3 ln (x) ln(1 + x) + 9 Li2 (x)

7
ln2 (x) 67
+ 23 Li3 (x) + 23 Li3 (x) 43 H1,0,1 (x) + (1 + x) 36
72 ln(x)

1
+ 43 ln(x) ln(1 + x) + 12
ln3 (x) + 23 Li2 (x) + 43 Li2 (x)
1

4
2
1 2
+ (1 x) 9 ln(x) 10
9 3 ln(1 x) + 3 ln(x) ln(1 x) 3 ln (x)
(1 x)
23
16
5
12 2
29 2
30 2
17
6 3

,
(12)
where we have introduced

pqq (x) = 2(1 x)1 1 x
(13)
and all divergences for x 1 are understood in the sense of +-distributions. In Eq. (12) we
have left one particular harmonic polylogarithm, H1,0,1 (x), unsubstituted. This function
is given by
x
H1,0,1 (x)
0
dz
Li2 (z)
1+z

1
= Li2 (x) ln(1 + x) + S1,2 x 2 S1,2 (x) S1,2 (x),
2
(14)
where the representation by the Nielsen function S1,2 has been derived in Ref. [66].
H1,0,1 (x) can also be expressed in terms of trilogarithms, albeit with more complicated
arguments [6].
The x-space coefficient functions involve harmonic polylogarithms of weight four,
which in general cannot be expressed in terms of standard polylogarithms and Nielsen
functions anymore. Instead of writing down the cumbersome exact expressions, we prefer
to present sufficiently accurate, compact parametrizations in terms of the +-distributions
and end-point logarithms
k

ln (1 x)
Dk =
(15)
,
L1 = ln(1 x),
L0 = ln x.
1x
+
It is convenient to apply this procedure (which has been employed in Ref. [17] for the twoloop coefficient functions) also to the nf1 part of the splitting function (12). Inserting the
192
numerical value of the QCD colour factors, this function can be approximated by

(2)
(x)
Pns
= nf 183.187D0 173.927(1 x) 5120
81 L1 197.0 + 381.1x
+ 72.94x 2 + 44.79x 3 1.497xL30 56.66L0L1 152.6L0

51
2
2
64 3
2608
81 L0 27 L0 + nf D0 16 + 33 52 (1 x)

3 2 64
+ x(1 x)1 L0 32 L0 + 5 + 1 + (1 x) 6 + 11
2 L0 + 4 L0 81 .
(16)
Corresponding parametrizations for the three-loop coefficient functions read

(3)
6592
c2,ns (x)
= nf 640
81 D4 81 D3 + 220.573D2 + 294.906D1 729.359D0
+ 2572.597(1 x)
640 4
81 L1
+ 167.2L31 315.3L21 + 4742L1 + 762.1
+ 7020x + 989.4x 2 + L0 L1 (326.6 + 65.93L0 + 1923L1)

3
728 4
+ 260.1L0 + 186.5L20 + 12224
243 L0 + 243 L0

464
+ nf2 64
81 D3 81 D2 + 7.67505D1 + 1.00830D0 103.2655(1 x)
+ 15.46L21 51.71L1 + 59.00x + 70.66x 2 + L0 L1 (80.05

2
368 3
10.49L0 + 41.67L1) 8.050L0 1984
(17)
243 L0 243 L0 ,

(3)
3
2
2
3
1792
cL,ns
(x)
= nf 1024
81 L1 112.4L1 + 340.3L1 + 409 210x 762.6x 81 xL0
64 3
81 L1
2
+ L0 L1 (969.2 + 304.8L0 288.2L1) + 200.8L0 + 64
3 L0

+ 0.046(1 x) + nf2 3xL21 + (6 25x)L1 19 + 317
6 122 x

6xL0 L1 + 6xLi2 (x) + 9xL20 (6 50x)L0 64
(18)
81 .
(2)
(3)
and cL,ns
, the +-distribution contributions (up to a numerical
The nf2 parts of Pns
truncation of the coefficients involving i ), and the rational coefficients of the (sub-)leading
regular end-point terms are exact in Eqs. (16)(18). The remaining coefficients have been
determined by fits to the exact results, for which we have used the Fortran package of
Ref. [68]. The above parametrizations deviate from the exact expressions by one part in
thousand or less, an accuracy which should be amply sufficient for foreseeable numerical
applications.
5. Implications for the threshold resummation

The large-N /large-x behaviour of the three-loop splitting functions and coefficient functions is of special interest in connection with the soft-gluon (threshold) exponentiation [43
45] at next-to-next-leading logarithmic (NNL) accuracy. Here the coefficient function for
F2,ns can, up to terms which vanish for N , be written as

C2,ns (s , N) = 1 + as g01 + as2 g02 +

exp Lg1 (as L) + g2 (as L) + as g3 (as L) +
(19)
193
with as = s /(4) and L = ln N . The functions gl depend on (universal) coefficients Ail

DIS
and Bil1 and process-dependent parameters Dil1
as described in Ref. [46], where
also the explicit expressions for the functions g1,2,3 can be found. Hence the NNL function
g3 involves the new coefficients A3 , B2 and D2DIS . These coefficients can be fixed by
expanding Eq. (19) in powers of s and comparing to the result of the full fixed-order
calculations.
In the MS scheme adopted in this article, the parameter A3 is simply the coefficient of
(2)
(2)
ln N in ns
(N) or, equivalently, of 1/(1 x)+ in Pns
(x). Its fermionic part is thus known
from Eq. (12),

110

16
2
2
160
112
A3 |nf = CA CF nf 836
27 + 9 2 3 3 + CF nf 3 + 323 + CF nf 27 .
(20)
(2)
The numerical value can be read off from Eq. (16). Like for the whole of Pns (x), as shown
in Fig. 4, this result is consistent with, but supersedes the estimate derived in Ref. [19] from
the first six even-integer moments. Parallel to our work A3 |nf has also been calculated in
Ref. [69].
The combination B2 + D2DIS has been determined in Ref. [46] by comparing the
(2)
expansion of Eq. (19) to the ln N term of the two-loop coefficient function c2,ns of
(3)
Ref. [2]. As the ln2 N (or D1 ) contribution to c2,ns involves a different linear combination,
0 (B2 + 2D2DIS ), of the very same coefficients, B2 and D2DIS can be disentangled using
the three-loop coefficient function. The analytic results for the two new +-distribution
coefficients read

83

(3)
2
512
112
c2,ns D n = CA CF nf 15062
81 + 9 2 + 163 + CF nf 9 + 1682 + 3 3
1 f

32
+ CF nf2 940
(21)
81 9 2 ,
160906 9920

(3)
776
208 2
c2,ns D n = CA CF nf 729 81 2 9 3 + 15 2
0 f

2
4226
+ CF2 nf 2003
108 27 2 603 + 162

232
32
+ CF nf2 8714
(22)
729 + 27 2 27 3
(the coefficients of D2,...,5 can be found in Ref. [65]). In fact, due to the prefactor 0 ,
the complete results for B2 and D2DIS can already be inferred from fermionic result (21),
yielding

44
B2 = CF2 32 243 + 122 + CF CA 3155
54 + 403 + 3 2

8
+ CF nf 247
(23)
27 3 2 ,
D2DIS = 0.
(24)
D1DIS
D2DIS in
The vanishing of
and
contrast to the DrellYan process, where D2 is
different from zero [46]calls for a deeper explanation, possibly offering an all-order
generalization.
Finally we note that, once the non-fermionic contributions to the 3-loop nonsinglet splitting functions and coefficient functions are completed, the NNL threshold
194
resummation facilitates a prediction of the first six towers of logarithms, i.e., the
coefficients of sn ln2ni N , i = 0, . . . , 5, of C2,ns at all orders n > 3. We will return to
this issue in a later publication.
6. Summary
We have computed the fermionic (nf -enhanced) third-order contributions to the

structure functions F2 and FL in electromagnetic deep-inelastic scattering. The calculation
has been carried out for all even-integer Mellin moments N , by solving the three-loop
integrals by means of recursion relations (difference equations) in N . This progress with
respect to previous computations restricted to some fixed moments N is especially due to
an improved understanding of the mathematics of harmonic sums and difference equations,
and the implementation of corresponding tools in the symbolic manipulation program
FORM which we employed to handle the huge intermediate expressions. We are confident
that our approach will enable us to compute all three-loop corrections in DIS.
We have thus been able to derive the complete expressions for the corresponding nf parts of the NNLO anomalous dimensions and splitting functions and the N3 LO coefficient
functions for F2 and FL . The results have been presented in both Mellin-N and Bjorken-x
space, in the latter case we have also provided easy-to-use accurate parametrizations. Our
results agree with all partial and approximate results available in the literature for these
quantities, in particular we reproduce the even-integer moments N = 2, . . . ,14, computed
before.
The present results for the three-loop splitting function represent a step towards
completing the ingredients required for NNLO calculations of hard-scattering processes
involving initial-state hadrons in perturbative QCD. The three-loop coefficient functions
for the most important structure function F2 form the dominant part of the N3 LO
corrections at large x, thus facilitating extractions of s with a distinctly reduced theoretical
uncertainty. Already the nf -part computed in this article leads to a complete determination
of the threshold-resummation parameters B2 and D2DIS including the non-fermionic
contributionsof which only the sum was known so far, thus practically completing the
information required for the next-to-next-to-leading logarithmic resummation.
Acknowledgements
The work of S.M. at the Institut fr Theoretische Teilchenphysik at Karlsruhe University

was supported by the German Research Society (DFG) under contract No. FOR 264/2-1.
The work of J.V. and A.V. is part of the research program of the Dutch Foundation for
Fundamental Research of Matter (FOM).
195
Appendix A
All results for the non-singlet anomalous dimensions and coefficient functions presented
in this article can be obtained as a FORM file from the preprint server http://arXiv.org by
downloading the source file. Furthermore they are available from the authors upon request.
The fermionic parts, i.e., all terms proportional to CA CF nf , CF2 nf and CF nf2 of the
three-loop coefficient function for the electromagnetic structure function F2 are given by
(3)
c2,ns (N)

= 16CA CF nf (N 2) 53 5
119
300 3
142883
7776
1051
72 3
+ 34 4 +
83
54 S4
2S1,4 49 S4,1 +
14
3 S2 3
199819
8100 S1
56
9 S1,2,1
191
16
20
4
29
81 S3 3 S3,2 + 27 S3,1 9 S3,1,1 18 S2
16
13
8
23
3 S2,3 + 3 S2,2 + 3 S2,2,1 + 9 S5
101
181
83
4132
8
18 S1,2 + 108 S3,1 54 S4 + 135 S1,2 + 3 S1,2,2
10
58
23
3 S1,1 3 4S1,1,3 + 9 S1,1,1 18 S1,2,1
+ 10S1,4 + 83 S1,1,1,2 + 13 S1,1,1,2 4S2,2,1 13 S1,1,2,1 83 S1,2,2
5
1
36719
+ S1,2,2 + 35
9 S1,3 + 3 S1,3,1 12 S2,2 16200 S2
218
23
263
8
+ 32
3 S2 3 + 10S2,3 9 S2,2 + 18 S1,1,2 60 S2,1 3 S2,1,2
21463
1080 S1,1
28
208
112
25S1 3 23 S2,3 4537
1620 S3 + 3 S3,2 9 S1,3 9 N S1,1,2
4

2
2
+ N+ 9 S3,1,1 53 S2,1,1 + 49 S4,1 23
9 S5 + (N3 N2 ) 5 S1 3 + 5 S1,3
119
2
2
2
1
1
450 S1,2 15 S1,2,1 3 S1,1 3 15 S1,1,2 + 3 S1,1,3 3 S1,3,1

4
2
1
59
1
+ 15
S2,2 + (N2 N ) 25 S3 179
450 S1 + 3 S1 3 + 5 S1,1 450 S2 + 5 S2,1

2
13 S1,3 + 15
S1,2 + 13 S3,1 + (N 1) 4S1,4 + 2S1,2,2 + 23 S1,2,2

21
4
7
14
+ (1 N+ ) 11057
324 S1 2 S1 3 + 9 S1,3 3 S1,2 3 S1,2,2
176
217
83 S1,2,1 + 3559
216 S1,1 8S1,1 3 8S1,1,3 + 9 S1,1,2 + 36 S1,1,1
16
2
7
2
2
55
3 S1,1,1,2 + 3 S1,1,1,2 36 S1,1,2 3 S1,1,2,1 + 3 S1,1,3 9 S1,2
7
4
38
8
231037
16
3 S1,2,2 + 36 S1,2,1 3 S1,2,2 + 9 S1,3 + 3 S1,3,1 4S1,4 5400 S2
20
793
16
1
+ 3S2 3 6S2,3 + 118
9 S2,2 + 3 S2,2,1 90 S2,1 3 S2,1,2 + 6 S2,1,2
22
47
1
49 S2,2 16 S2,2,1 2S2,3 + 49717
1620 S3 3 S3,2 54 S3,1 6 S3,2
166
5
219
6
27 S4 + 6 S4,1 + (N+ N+2 ) 10S1 3 50 S1 + 5 S1,2 + 4S1,2,1
21
+ 24
5 S1,1 + 4S1,1 3 4S1,1,2 2S1,1,3 + 3S1,2 3S1,3 + 2S1,3,1 50 S2

3
4S2 3 + 4S2,2 24
5 S2,1 + 2S2,3 5 S3 S3,1 2S4,1

+ (N+2 N+3 ) 3S2,2 3S1,1,2
159
6
50 S1,2 5 S1,2,1
6S1,1 3 65 S1,1,2 + 3S1,1,3 + 3S1,2,1 + 6S2 3 3S1,3,1 + 65 S2,2 + 3S4,1
1711

21
18
5608067
3S2,3 159
50 S3 5 S3,1 + 5 S4 + (N + N+ ) 108 S1 3 291600 S1
72
5 S1 3
18
5 S1,3
196
12 S1 4
+
+
79
392
2
6613
9 S1,4 + 4S1,1,2 + 27 S1,3 + 9 S1,3,1 405 S1,2
104
4
8
25511
52
184
27 S1,2,1 + 9 S1,2,1,1 + 3 S1,1 3 1620 S1,1 + 9 S1,1,3 27 S1,1,1
32
11
17
1
4
13
9 S1,1,1,2 9 S1,1,1,1 18 S1,1,1,2 + 36 S1,1,2 + 9 S1,1,2,1 + 18 S1,1,3
661
34
29
1
8
108 S1,2 + 9 S1,2,2 + 12 S1,2,1 + 2 S1,2,1,1 4S1,3 + S4,1 3 S1,3,1
17
44537
23
79
377
9 S1,4 + 2700 S2 3 S2 3 9 S2,3 + 27 S2,2 +
34
59
5
53
5
9 S2,1,2 + 18 S2,1,1 + 9 S2,1,2 18 S2,2 9 S2,2,1 +
67
9 S3,2
83
18 S3,1
56 S3,1,1 + 12 S3,2 +
1
29
1
43
18
3 757
648 S1 108 S1,1 18 S1,2 324 S2
13
19
265
7
7
16 S2,1 + 19
54 S3 + (1 N+ ) 18 S2,1 18 S3 + 108 S2 + 18 S1,2 18 S1,1,1

1421
5585
1
161
13
133
108 S1,1 648 S1 + (N + N+ ) 5832 S1 + 27 S1 3 + 324 S1,1 + 54 S1,1,1
+ 16CF nf2
1
18 S1,1,1
253
54 S4
20
5731
9 S2,2,1 + 540 S2,1
13
4511
18 S2,3 405 S3
9517
7776
+ 19 S1,1,1,1 19 S1,1,2
29
19 S1,2,1 + 19 S1,3 301
324 S2 54 S2,1

1
23
29 S2,1,1 + 29 S2,2 + 62
81 S3 + 3 S3,1 54 S4

341
139
3
46
8
+ 16CF2 nf 119
150 3 (N 2) 576 + 6 3 4 4 9 S5 + 9 S4,1
13
54 S1,2
382
83
32
40
8
29
28
81 S3 + 27 S4 + 3 S3,2 27 S3,1 + 9 S3,1,1 + 9 S2 + 3 S2 3
66367
32
26
301
16
3 S2,2,1 + 129600 S1 + 3 S2,3 3 S2,2 + 6 S1 3 4S1,4
8264
8
112
83
20
+ 416
9 S1,3 135 S1,2 + 3 S1,2,2 9 S1,2,1 + 144 S1,1 + 3 S1,1 3
557
16
2
31
+ 8S1,1,3 224
9 S1,1,2 72 S1,1,1 3 S1,1,1,2 3 S1,1,1,2 18 S1,1,2
16
7
2
235
10
+ 2171
216 S1,2 + 3 S1,2,2 + 3 S1,2,1 + 3 S1,2,2 36 S1,3 + 4S1,4 3 S1,3,1
140237
436
7133
16200 S2 22S2 3 20S2,3 + 9 S2,2 + 8S2,2,1 + 1080 S2,1

16
50
7
7627
3 S2,1,2 + 9 S2,1,1 4S2,2 + 3 S2,3 3240 S3 + N+ S2,1,1,1 S2,1,2
2
56
329
29
7
401
3 S1,1,2,1 S2,2,1 3 S3,2 27 S3,1 9 S3,1,1 + 3 S3,2 + 54 S4

53
1
119
4
4
4
18 S4,1 + 18 S5 + (N3 N2 ) 225 S1,2 5 S1 3 + 15 S1,2,1 5 S1,3

4
8
179
4
4
59
15 S1,1,2 15 S2,2 + (N2 N ) 225 S1 15 S1,2 + 15 S1,1 + 225 S2
4
15 S2,1
+
+
+

45 S3 +
15439
219
12
1440 (N 1)S1,1 + (N+ N+2 ) 25 S1 5 S1,2
154
+ 4S1 3 8S1,2,1 13
5 S1,1 + 8S1,1,2 + 2S1,1,2 5S1,2 2S1,2,1 25 S2

13
+ 36
5 S3 + 10S3,1 + 5 S2,1 2S2,2 8S2,2 + (N+2 N+3 ) 5S1,1,2
37
159
159
12
12
5 S2,2 5S2,2 + 5 S3,1 + 25 S1,2 + 25 S3 5S1,2,1 + 5 S1,1,2
114
36
36
10337
5 S1 3 5 S1,3 5 S4 + (1 N+ ) 4S4 576 S1
16
16
40
8S1,4 89 S1,3 + 14
3 S1,2 + 3 S1,2,2 + 3 S1,2,1 + 3 S1,1 3
128
179
32
14
+ 16S1,1,3 76157
4320 S1,1 9 S1,1,2 + 72 S1,1,1 3 S1,1,1,2 + 3 S1,1,1,1
295
32
17
43 S1,1,1,2 + 565897
16200 S2 + 2S1,1,2,1 72 S1,2 + 3 S1,2,2 3 S1,2,1
12
5 S1,2,1
197
+ 43 S1,2,2
115
29
2
20
36 S1,3 + 8S1,4 6 S1,1,2 3 S2 3 3 S1,3,1 + 12S2,3
236
40
4663
32
37
22817
32
9 S2,2 3 S2,2,1 1080 S2,1 + 3 S2,1,2 3 S2,1,1 810 S3 + 3 S2,2
86

5
65
85
91
3 S2,3 4S3,2 + 18 S3,1 + 6 S1 3 + (N + N+ ) 9 S1,4 36 S5
4S3,1,1 89 S1,2,1,1 +
4S1,2,2 +
+
+
+
13226
104
152
784
405 S1,2 9 S1,1,3 + 9 S1,1,2 27 S1,3
41929
244
1
208
4
129600 S1 9 S1 3 + 2 S1 4 + 27 S1,2,1 9 S1,3,1
13
7
64
35
10
44
3 S1,1 3 8 S1,1,1 + 9 S1,1,1,2 6 S1,1,1,1 3 S1,1,1,1,1 + 9 S1,1,1,2
25
91
577
40
107
3 S1,1,2 + 3S1,1,2,1 18 S1,1,3 + 648 S1,2 + 3 S2 3 + 18 S1,2,1
34
79151
68
239
49
275
9 S1,2,1,1 64800 S2 9 S1,2,2 54 S2,2 9 S1,2,2 108 S1,3
73
31
158
754
40
133
18 S2,1,1,1 6 S2,1,2 + 9 S2,3 27 S2,2 9 S2,2,1 + 36 S4,1
403
68
79
67
17
5
5
810 S2,1 9 S2,1,2 + 18 S2,1,1 18 S2,2,1 + 18 S1,4 + 72 S4 18 S1,3,1
25
9 S2,3
7871
3240 S3
134
9 S3,2
241
72 S3,1
38
9 S3,2
(A.1)
For the sake of completeness, we include the result for the complete first and second-order
(1)
(2)
longitudinal coefficient functions cL,ns
and cL,ns
known from Refs. [2,6,56,70]
(1)
cL,ns(N) = 4CF (1 N+ )S1 ,

(2)
(N)
cL,ns
= 4CA CF
12
5 3 (N
2) +
12
5 (N+2
16
5 (N2 N )[S1
24
5 (N+ N+2 )[S1
(A.2)
12
5 (N+
N+2 )[S1 S2 ]
+ 85 S2 + 85 (N3 N2 )S1,2

+ 8(N 1)S1,2 + 85 (N2 N )[S1 + S2 ] + (1 N+ ) 12S1 3 4S1,2

287
176
23
3 S1,1 8S1,1,2 4S1,3 18 S1 + 4S1,3 + 15 S2 4S3

4
+ (N + N+ ) 49
S
S
1
2
15
5

2
2
+ 4CF nf (1 N+ ) 3 S1,1 43 S2 + 25
S
(N
1)S
1
1
9
3

22
4
16
+ 4CF2 24
5 3 (N 2) + 5 S1 + 5 S2 5 (N3 N2 )S1,2
N+3 )[S1,2 + S3 ]
+ (1 N+ )
33
2
98
15 S1
+ S2 ] + (N 1)[2S1,1 16S1,2 ]
S2 ]
24
5 (N+2
N+3 )[S1,2 + S3 ]
S1 + 8S1,3 24S1 3 8S1,3 + 8S1,2
+ 7S1,1 + 16S1,1,2 4S1,1,1 + 4S1,2

2
(N + N+ ) 11
5 S1 + 5 S2 .
54
5 S2
+ 6S2,1 + 4S3

(A.3)
The contributions to the 3-loop longitudinal coefficient function corresponding to Eq. (A.1)
198
read
(3)
cL,ns(N)

= 16CA CF nf (N 2) 20
3 5
149
75 3
+ (N3 N2 )
16
15 S2,2
8
8
43 S1,3,1 + 43 S1,1,3 83 S1,1 3 15
S1,1,2 15
S1,2,1 + 85 S1,3

8
4
418
8
8
298
225 S1,2 + 5 S1 3 + (N2 N ) 3 S1,1 3 225 S1 + 5 S3 + 15 S1,2
+ 43 S1,2,1 43 S1,1,2 + 45 S1,1
+ 43 S3,1 + 4S1 3

+ 45 S2,1 23 S1,1,3 43 S1,3 + 23 S1,3,1 + (N 1) 16
3 S2,2
178
225 S2
13033
1350 S1
+ 8S1 3
518
2
8
254
2
45 S1,2 5 S2,1 3 S1,2,1 45 S1,1 15 S3

8
8
8
7
+ 8S1,3 + 76
25 S2 3 S1,1,2 + (N+ N+2 ) 3 S1,2,1 3 S2 3 25 S2
8
4
83 S1,1,2 2S1,3 + 43 S1,3,1 43 S1,1,3 + 2S1,2 + 16
5 S1,1 + 3 S1,1 3 + 5 S1,2
20
73
2
2
4
4
8
3 S1 3 25 S1 5 S3 3 S3,1 3 S4,1 + 3 S2,3 + 3 S2,2
4
+ (N+2 N+3 ) 5 S2,2 + 2S2,2
53
12
48
12
45 S1,2,1 2S2,3 14
5 S3,1 25 S3 + 5 S4 + 2S4,1 5 S1 3 + 5 S1,3
+ 2S1,1,3 4S1,1 3 45 S1,1,2 53
25 S1,2 + 2S1,2,1
16
5 S2,1

2S1,3,1 + 4S2 3 2S1,1,2

40
+ (1 N+ ) 125599
4050 S1 3 S1 3 + 4S1,4
+
68
413
8
9 S1,3 + 45 S1,2 3 S1,2,2
88
40
16
9 S1,1,2 + 9 S1,1,1 + 3 S1,1,1,2
4469
270 S1,1 8S1,1 3 8S1,1,3 +
2
2
2
2
40
16
10
3 S1,1,1,2 + 3 S1,1,2 3 S1,1,2,1 + 3 S1,1,3 9 S1,2 3 S1,2,2 3 S4
2
2
2
8
20953
4
3 S1,2,1 3 S1,2,2 + 3 S3,1 + 3 S1,3,1 4S1,4 675 S2 + 8S2 3 + 3 S2,3
2
8
238
16
824
3 S2,2 + 3 S2,2,1 45 S2,1 3 S2,1,2 2S2,3 + 45 S3
38
9 S1,3

+ 23 S4,1

2
2
2
317
+ 16CF nf2 (N 1) 19
27 S1 + 9 S1,1 9 S2 + (1 N+ ) 9 S1,2 162 S1

2
50
4
2
25
S
S
+
S
+
S
S
1,1
1,1,1
2
2,1
3
27
9
27
9
3

16
16
+ 16CF2 nf 298
75 3 (N 2) + (N3 N2 ) 15 S1,1,2 5 S1,3

596
32
16
+ 16
15 S1,2,1 + 225 S1,2 15 S2,2 5 S1 3

16
356
836
8
16
+ (N2 N ) 16
15 S2,1 5 S3 + 225 S2 + 225 S1 3 S1 3 + 15 S1,1

2141
83 S1,2,1 + 83 S1,1,2 16
15 S1,2 + (N 1) 1350 S1 16S1 3 16S1,3
+
1036
16
73
16
5
45 S1,2 + 3 S1,2,1 + 45 S1,1 + 3 S1,1,2 3 S1,1,1

1
32
14
+ 53 S1,2 + 232
225 S2 + 10 S3 3 S2,2 + 5 S2,1

10
308
20
4
+ (N+ N+2 ) 146
25 S1 3 S1,2 75 S2 + 3 S3,1 3 S2,2
+ 83 S1 3
199
16
4
16
4
8
16
3 S2,2 3 S1,2,1 + 3 S1,1,2 + 3 S1,1,2 5 S1,2 3 S1,2,1

26
24
26
76
106
8
15 S1,1 + 5 S3 + 15 S2,1 + (N+2 N+3 ) 5 S1 3 + 25 S1,2 + 5 S1,2,1
24
10
8
10
8
5 S1,3 + 3 S1,1,2 + 5 S1,1,2 3 S1,2,1 5 S2,2

74
74
24
106
10
173653
15 S3,1 5 S4 + 25 S3 3 S2,2 + (1 N+ ) 3 S1 3 10800 S1
8S1,4 +
136
826
16
1609
40
9 S1,3 45 S1,2 + 3 S1,2,2 135 S1,1 + 3 S1,1 3
7
32
8
+ 16S1,1,3 176
9 S1,1,2 + 18 S1,1,1 3 S1,1,1,2 + 3 S1,1,1,1
17
43 S1,1,1,2 + 8S1,4 4S1,1,2 + 43 S1,1,2,1 23
18 S1,2 + 6 S4
4
49
20
30737
40
+ 32
3 S1,2,2 2S1,2,1 + 3 S1,2,2 9 S1,3 3 S1,3,1 + 1350 S2 3 S2 3
17
32
83 S2,3 + 43 S2,2 16
3 S2,2,1 10 S2,1 + 3 S2,1,2
13
3 S2,1,1
13
3 S2,2
+ 83 S2,3
913
45 S3
13
3 S3,1
(A.4)
Recall that in all our formulae the expansion parameter is normalized as in Eq. (2). The
operators N and Ni have been defined in Eq. (4).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
K. Hagiwara, et al., Particle Data Group, Phys. Rev. D 66 (2002) 010001.

W.L. van Neerven, E.B. Zijlstra, Phys. Lett. B 272 (1991) 127.
E.B. Zijlstra, W.L. van Neerven, Phys. Lett. B 273 (1991) 476.
E.B. Zijlstra, W.L. van Neerven, Phys. Lett. B 297 (1992) 377.
E.B. Zijlstra, W.L. van Neerven, Nucl. Phys. B 383 (1992) 525.
S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 573 (2000) 853.
S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 427 (1994) 40.
S.A. Larin, P. Nogueira, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 492 (1997) 338.
A. Retey, J.A.M. Vermaseren, Nucl. Phys. B 604 (2001) 281.
S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 621 (2002) 413.
A.L. Kataev, A.V. Kotikov, G. Parente, A.V. Sidorov, Phys. Lett. B 417 (1998) 374.
A.L. Kataev, G. Parente, A.V. Sidorov, Nucl. Phys. B 573 (2000) 405.
J. Santiago, F.J. Yndurain, Nucl. Phys. B 563 (1999) 45.
J. Santiago, F.J. Yndurain, Nucl. Phys. B 611 (2001) 447.
S.I. Alekhin, Phys. Lett. B 519 (2001) 57.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 531 (2002) 216.
W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263.
W.L. van Neerven, A. Vogt, Phys. Lett. B 490 (2000) 111.
S.A. Larin, F.V. Tkachev, J.A.M. Vermaseren, Phys. Rev. Lett. 66 (1991) 862.
S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 259 (1991) 345.
J.A.M. Vermaseren, S. Moch, Nucl. Phys. (Proc. Suppl.) 89 (2000) 131.
A. Gonzalez-Arroyo, C. Lopez, F.J. Yndurain, Nucl. Phys. B 153 (1979) 161.
A. Gonzalez-Arroyo, C. Lopez, Nucl. Phys. B 166 (1980) 429.
J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037.
J. Blmlein, S. Kurth, Phys. Rev. D 60 (1999) 014018.
A.B. Goncharov, Math. Res. Lett. 5 (1998) 497, available at http://www.math.uiuc.edu/K-theory/0297.
J.M. Borwein, D.M. Bradley, D.J. Broadhurst, P. Lisonek, math.CA/9910045.
E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725.
200
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
D.J. Gross, F. Wilczek, Phys. Rev. D 8 (1973) 3633.

H. Georgi, H.D. Politzer, Phys. Rev. D 9 (1974) 416.
D.J. Gross, F. Wilczek, Phys. Rev. D 9 (1974) 980.
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 129 (1977) 66.
D.I. Kazakov, A.V. Kotikov, Nucl. Phys. B 307 (1988) 721.
D.I. Kazakov, A.V. Kotikov, Nucl. Phys. B 345 (1990) 299, Erratum.
R. Hamberg, W.L. van Neerven, T. Matsuura, Nucl. Phys. B 359 (1991) 343.
R.V. Harlander, W.B. Kilgore, Phys. Rev. Lett. 88 (2002) 201801.
M. Dittmar, F. Pauss, D. Zurcher, Phys. Rev. D 56 (1997) 7284.
V.A. Khoze, A.D. Martin, R. Orava, M.G. Ryskin, Eur. Phys. J. C 19 (2001) 313.
W.T. Giele, S.A. Keller, hep-ph/0104053.
C. Anastasiou, K. Melnikov, hep-ph/0207004.
G. Sterman, Nucl. Phys. B 281 (1987) 310.
S. Catani, L. Trentadue, Nucl. Phys. B 327 (1989) 323.
S. Catani, L. Trentadue, Nucl. Phys. B 353 (1991) 183.
A. Vogt, Phys. Lett. B 497 (2001) 228.
S.G. Gorishnii, S.A. Larin, L.R. Surguladze, F.V. Tkachev, Comput. Phys. Commun. 55 (1989) 381.
S.A. Larin, F.V. Tkachev, J.A.M. Vermaseren, NIKHEF-H-91-18.
P. Nogueira, J. Comput. Phys. 105 (1993) 279.
J.A.M. Vermaseren, math-ph/0010025.
G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
C.G. Bollini, J.J. Giambiagi, Nuovo Cimento 12B (1972) 20.
J.F. Ashmore, Lett. Nuovo Cimento 4 (1972) 289.
G.M. Cicuta, E. Montaldi, Lett. Nuovo Cimento 4 (1972) 329.
G. t Hooft, Nucl. Phys. B 61 (1973) 455.
W.A. Bardeen, A.J. Buras, D.W. Duke, T. Muta, Phys. Rev. D 18 (1978) 3998.
S.G. Gorishnii, S.A. Larin, F.V. Tkachev, Phys. Lett. B 124 (1983) 217.
S. Laporta, Phys. Lett. B 504 (2001) 188.
S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087.
F.V. Tkachov, Phys. Lett. B 100 (1981) 65.
K.G. Chetyrkin, F.V. Tkachev, Nucl. Phys. B 192 (1981) 159.
F.V. Tkachov, Theor. Math. Phys. 56 (1983) 866.
S. Moch, J.A.M. Vermaseren, hep-ph/0208050.
J.A. Gracey, Phys. Lett. B 322 (1994) 141.
A. Vogt, Phys. Lett. B 471 (1999) 97.
J. Blmlein, Comput. Phys. Commun. 133 (2000) 76.
G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 141 (2001) 296.
C.F. Berger, hep-ph/0209107.
J. Sanchez Guillen, et al., Nucl. Phys. B 353 (1991) 337.

More chiral operators on the lattice

Werner Kerler
Institut fr Physik, Humboldt-Universitt, D-10115 Berlin, Germany
Received 10 June 2002; received in revised form 30 September 2002; accepted 7 October 2002
Abstract
Instead of the GinspargWilson (GW) relation we only require generalized chiral symmetry and
show that this results in a larger class of Dirac operators describing massless fermions, which in
addition to GW fermions and to the ones proposed by Fujikawa includes many more general ones.
The index turns out to depend solely on a basic unitary operator. We use spectral representations to
analyze the new class and to obtain detailed properties. We also show that our weaker conditions still
lead properly to Weyl fermions and to chiral gauge theories.
1. Introduction
In a large part of the works on chiral fermions the Dirac operator D is required to satisfy
the GinspargWilson (GW) relation [1]
{5 , D} = 1 D5 D,
(1.1)
with a real constant , and to be 5 -Hermitian,

D = 5 D5 .
(1.2)
Conditions (1.1) and (1.2) imply that the operator 1 1 D is unitary and 5 -Hermitian.
Therefore, instead of imposing (1.1) and (1.2) as is done in such works, it is completely
equivalent to require D to have the form
D = (1 V ),
with V = V 1 = 5 V 5 .
E-mail address: kerler@physik.hu-berlin.de (W. Kerler).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 9 0 7 - 0
(1.3)
202
W. Kerler / Nuclear Physics B 646 (2002) 201219
In this notation the condition for the generalized chiral symmetry of the classical action
of Lscher [2] in the form used in Ref. [3] reads
5 D + D5 V = 0,
with V = V 1 = 5 V 5 .
(1.4)
In the present paper we observe that (1.4) provides a weaker condition than the GW
relation (1.1). Accordingly, instead of (1.1) and (1.2) as are usually imposed, we here only
require (1.4) and (1.2) and investigate the resulting larger class of Dirac operators.
We note that using (1.2) one gets from (1.4)
D + D V = 0,
D + DV = 0,
(1.5)
by which it follows that V and D commute and that D is normal,

[V , D] = 0,
DD = D D.
(1.6)
To account for this we require D to be a function of V ,

D = F (V ),
(1.7)
i.e., to depend on V and possibly on constants, however, not on any other operator.
We show that in addition to GW fermions with (1.1) and to the ones proposed by
Fujikawa [4] and their extension [5] the new class includes many more general ones. We
nevertheless find that our weaker conditions still lead properly to Weyl fermions and to
chiral gauge theories. Thus it becomes obvious that these conditions constitute the more
general principle.
In addition to the general conditions (1.4) and (1.2), we require that F allows for a
nonvanishing index of the Dirac operator. Our respective analysis is based on spectral
representations. The index turns out to depend solely on the basic unitary operator V .
We introduce spectral functions which on the one hand side allow for a construction,
which gives insight into possible forms of F , and on the other side describe the location
of the spectrum. We also give a realization of the unitary operator V related to our general
construction of the Dirac operator D.
In Section 2 we obtain fundamental properties of the functions D = F (V ). In Section 3
we discuss the special cases of GW fermions, of the ones proposed by Fujikawa
(bringing D of them also into the form F (V )) and of the extension of the Fujikawa type.
In Section 4 we develop realizations of subclasses. We first study the one where F (V ) can
be represented by a power series. We then give a more general construction for operators
of the class, which is based on spectral functions. In a next step we further extend this
construction applying appropriate functions. We then address the question which sets of V
are related to the subclasses and give a realization of V holding for the general construction
of D. Further, by working out a particular example, the considerable freedom in this
construction is illustrated. In Section 5 we show that our conditions still lead to Weyl
operators and to chiral gauge theories. Section 6 contains our conclusions.
203
2. Properties from spectral representation

On the finite lattice V has the spectral representation V = k vk Pk with eigenvalues
projections Pk = Pk . Therefore the functions F (V ) can
satisfying |vk | = 1 and orthogonal

be represented by F (V ) = k f (vk )Pk .
Using 5 -Hermiticity we get the more detailed form

(+)
()
(+)
()
(I)
(II)
eik Pk + eik Pk ,
V = P1 + P1 P2 P2 +
(2.1)
k (0<k <)
in which the projections satisfy

5 Pj() = Pj() 5 = Pj() ,
5 Pk(I) = Pk(II) 5 .
The spectral representation of D = F (V ) then becomes

D = f (1) P1(+) + P1() + f (1) P2(+) + P2()

(I)
(II)

+
f eik Pk + f eik Pk ,
(2.2)
(2.3)
k (0<k <)
in which D is characterized by the function f (ei ). Clearly this function enters only at the
values ei = eik . However, since we want to define D in general (in particular, for any
gauge field configuration), we have to specify f (v) for all v = ei .
Inserting the general form (2.3) into (1.2) and (1.5) we obtain the conditions
f (v) = f (v ),
f (v) + f (v) v = 0,
(2.4)
(2.5)
respectively, on the functions f (v). These conditions imply

f (1) = 0,
f (1) real.
The operator form of (2.4) is

F (V ) = F V .
(2.6)
(2.7)
In addition to giving D by (2.3), the functions f (v) obviously describe the location of
its spectrum. For continuous f (ei ) the eigenvalues of D reside on a closed curve in the
complex plane which is symmetric to the real axis and meets this axis at zero and at the
value f (1) (which is required to be nonzero below).
Denoting the dimensions of the right-handed and of the left-handed eigenspace for
eigenvalue 1 of V by N+ () and N (), respectively, we have from (2.2)

()
()
= N (1),
= N (1),
Tr 5 P1
Tr 5 P2

(I)
(II)
Tr 5 Pk = Tr 5 Pk
(2.8)
= 0.
With this because of f (1) = 0, using the resolvent (D 1)1 , we obtain the index of D,

for f (1) = 0,
N+ (1) N (1),
lim Tr 5
=
(1)
N
(1)
+
N
(1)
N
(1),
for f (1) = 0,
N
0
D 1
+
(2.9)
204
and also find

lim Tr 5
0

D
N+ (1) N (1), for f (1) = 0,
=
0,
for f (1) = 0.
D 1
(2.10)
Adding up (2.9) and (2.10) the sum on the l.h.s. gets Tr(5 1) = 0 so that in any case
N+ (1) N (1) + N+ (1) N (1) = 0.
(2.11)
Because of (2.9) and (2.11), to admit a nonvanishing index we have to impose the condition
f (1) = 0.
(2.12)
After having (2.12), according to (2.11) to allow for a nonvanishing index one has
also to require that in addition to 1 the eigenvalue 1 of V occurs. The sum rule (2.11)
corresponds to the one found in Ref. [6] for the special case of Dirac operators which
satisfy the GW relation (1.1).
Using (2.1) with (2.8) we find
Tr(5 V ) = N+ (1) N (1) N+ (1) + N (1),
(2.13)
so that with (2.11) we generally get for the index of the Dirac operators D
1
Tr(5 V ).
(2.14)
2
Thus it turns out that solely the operator V enters for the whole class. This generalizes the
results obtained in the GW case [2,7] and in the overlap formalism [8] before.
On the infinite lattice the unitary space, in which the operators act, gets of infinite dimension and, including limit elements, a Hilbert
space. Then
to the spectral representa
tions (2.1) and (2.3) the continuous parts ei dE and f (ei ) dE , respectively,
are to be added in which the projector function E is purely continuous. However, the relations for the index do not change since the continuous spectrum does not contribute to
them and because the trace Tr for the expressions of interest can still be defined [9].
We note that condition (2.11) reflects a fundamental difference to the case of the Atiyah
Singer (AS) Dirac operator [10], for the nonvanishing eigenvalues AS j of which one
always has
N+ (1) N (1) =
(AS j ) = 0,
+ (AS j ) N
N
for AS j = 0,
(2.15)
are the dimensions of its right-handed and its left-handed eigenspaces,

+ and N
where N
respectively. For zero eigenvalue of the AS Dirac operator one gets
(0) = dim ker D (+) dim ker D (+) ,
+ (0) N
N
AS
AS
(2.16)
(+)
is the related Weyl operator (the AS Dirac operator being a composition of
where DAS
(+)
(+)
()
DAS and DAS = DAS ). Thus to obtain a nonvanishing index there it is inevitable to
allow the space structure itself to vary and to get chirally asymmetric, which implies its
dependence on the particular gauge field configuration. In contrast to this on the lattice
there is no such dependence, which is made possible by having (2.11).
205
3. GW-type special cases

3.1. GinspargWilson relation
The Dirac operators satisfying the GW relation (1.1) and (1.2) are the simplest special
case of the class defined by (1.4) and (1.2). For these operators F has the form (1.3). As
in the more general cases, given the function F (V ), suitable V remain to be determined.
Thus looking for solutions of the GW relation (1.1), from the present point of view means
to look for the appropriate set of operators V .
The overlap Dirac operator of Neuberger [11] is well-known to be of form (1.3) and to
provide correct results.
The underlying principle of the construction of V there is by the
normalization X( X X )1 of an operator X, where in that case X is the WilsonDirac
operator. Another type of construction occurs in an example of Chiu [12], in which V can
be identified as the Cayley transform V = (Y i1)(Y + i1)1 of a suitable Hermitian
operator Y . This, however, because bounded Y do not allow to reach the eigenvalue 1
of V , can provide proper topological properties only in the limit [9]. Thus for the study of
theoretical issues, where an explicit construction is desirable, up to now one has to rely on
the normalization method.
We note that operators D satisfying the general GW relation [1]
{5 , D} = 2D5 RD,
with [R, D] = 0,
(3.1)
(where R = R and [ , R] = 0) do not belong to the class. In fact, with 5 -Hermiticity

of D and [5 , R] = 0 from this relation one gets [D, D ] = 2D [R, D]D . Thus for
[R, D] = 0 the Dirac operator D is not normal, which is in contrast to what is required
in (1.6).
In this context it is to be mentioned that for the general GW relation (3.1) the analysis
of the index gets rather subtle. One obtains the relation

Tr 5 (Pj + RQj ) + Tr(5 RD) = 0,
(3.2)
in which the projections Pj need not to be orthogonal and where eigennilpotents Qj can
occur if the dimensions of the respective algebraic and geometric eigenspaces differ. In a
lengthy proof [13] it has been shown that for the eigenvalue k = 0 of D these dimensions
are equal so that Qk = 0 and Pk = Pk . Thus for zero eigenvalue the unwanted term with
Qk disappears and Pk gets orthogonal as needed.
3.2. Proposal of Fujikawa
The Dirac operators proposed by Fujikawa [4] satisfy
{5 , 5 D} = 2a02k+1(5 D)2k+2 ,
k = 0, 1, 2, . . . .
(3.3)
They are normal [14], D D = DD . One may also consider them as satisfying the
general GW relation with R = a02k+1 (DD )k , where, however, in contrast to (3.1) one
has [R, D] = 0. Using 5 -Hermiticity one gets from (3.3)
k+1

.
D + D = 2a02k+1 D D
(3.4)
206
Comparing this with (1.5), D + D V = 0, it follows that

k
V = 1 2a02k+1D D D = 1 2a02k+1D(5 D)2k ,
(3.5)
which obviously is 5 -Hermitian and using (3.4) can be checked to be indeed unitary. With
this it follows that (1.4) is fulfilled so that the operators proposed in Ref. [4] are a special
case of our general class.
For D one obtains from (3.5)
D = a01
1/(2k+1)
1/(2k+1)

1
5 5 (1 V )
,
2
which is the form of Ref. [4]. The operator V here is given by the normalization

(2k+1)
(2k+1) (2k+1) 1
V = DW
DW
DW
(3.6)
(3.7)
of the generalized WilsonDirac operator

(2k+1)
DW

2k+1

2k+1
1
r

+

+ m2k+1 ,
2
2
(3.8)
where ( )n n = n4 n Un n4 ,n+ and 2r < m < 0. For k = 0 obviously (3.6) becomes
the overlap Dirac operator of Neuberger [11].
The form (3.6), involving in addition operators 5 , is not yet the one which according
to our considerations should occur. To get the latter we note that because of D = DV
we can also write (3.5) as
V = 1 2a02k+1(V )k D 2k+1 .
With this we obtain

1/(2k+1)
1 1
k
(1 V )(V )
D = a0
,
2
(3.9)
(3.10)
where of the (2k + 1)th roots the one which satisfies (2.7) is to be chosen. Thus with (3.10)
we have indeed the form D = F (V ) required by the general concept, which appears more
convenient in practice, too.
The functions f of the spectral representation (2.3) for D, in the present case are
obtained from (3.10) as

i
1/(2k+1)
1 i()/2
f e = a0 e
,
sin
2
(3.11)
where of the (2k + 1)th roots the real one is to be taken. They give the curves describing
the location of the eigenvalues of D, which for k > 0 arise as deformations of the circle for
k = 0. All of them meet the real axis at zero and at the value f (1) = a01 .
207
3.3. Extension of Fujikawa type

Recently an extension of the proposal of Fujikawa has been reported [5], the Dirac
operators of which satisfy

{5 , 5 D} = 2a1 D D a12 D D ,
(3.12)
where the operator function is subject to (X) = (X) for X = X. From the identity
[5 , D D] = [{5 , 5 D}, 5 D] with (3.12) since D D = (5 D)2 one gets

5 , D D = 0.
(3.13)
With this one has D D = 5 D D5 and finds

D , D = 0.
(3.14)
Using 5 -Hermiticity of D to write (3.12) in the form D + D = 2a1 D D (a12 D D)

and comparing with the basic relation (1.5), D + D V = 0, it follows that

V = 1 2a1D a12 D D .
(3.15)
Because of (3.13) and (a12 D D) = (a12 D D) this is seen to be 5 -Hermitian and
using (3.14) and (3.12) it can be checked to be unitary. It thus turns out that the operators
satisfying (3.12) are also a special case of our general class.
With (3.15) and (1.5) one gets the equation

2a1 D a12 V 1 D 2 + V = 1,
(3.16)
for D and V , the solution of which gives D = F (V ).
4. Realizations of subclasses
4.1. Expansion in powers of V

We first realize functions D = F (V ) of the class in terms of expansions D = c V .
According to (2.7) the coefficients c must be real and (1.5) requires them to satisfy
c1 = c . We thus obtain

c V V +1 ,
D=
(4.1)
0
for which f (1) = 0 is seen to hold. Condition (2.12) now gets the form

f (1) = 2
(1) c = 0.
(4.2)
0
Obviously (1.3) is a special case of (4.1) and arises by putting c = 0 . Other choices of
the coefficients are seen not to allow GW-type relations.
208
For the functions f in the spectral representation of the Dirac operators (4.1) we obtain
the form

c sin(2 + 1) .
f ei = 2ei()/2
(4.3)
2
0
The functions
+ 1)/2 in (4.3) describes rosettes in the complex plane.
The eigenvalues of D thus reside on a closed curve which is given by a linear combination
of rosette functions and which meets the real axis at zero and at the value (4.2).
In case of an infinite number of terms of theexpansion (4.1), convergence properties
can be studied considering the Fourier series 0 c sin(2 + 1) 2 in (4.3). Uniform
convergence then is guaranteed by the condition

(4.4)
|c | < .
ei()/2 sin(2
0
Having F (V ) it remains to specify V . Below

the overlap V is appropriate
we show that
(1)
(1) (1) 1
(1)
DW DW
with DW
given by (3.8) is
here, for which the explicit form V = DW
available. In principle this could also be done for other V suitable in the GW case (1.3),
since with respect to the tests considered they should be equivalent to the overlap V .
The continuum limit of the propagator gives us further conditions on the coefficients c .
To study the propagator we note that in the free case and infinite volume with the Fourier
() 4 on gets for the powers of V with = ap
representation V = V
() V
+1 () a 2 + 1 i
V
|m|
p ,
for a 0,
(4.5)
while at the corners of the Brillouin zone one has

V
+1 = 2(1) .
V
(4.6)
According to (4.1) and (4.5) for the usual normalization of the continuum propagator

(2 + 1)c |m|a 1 , for a 0,
(4.7)
0
is needed, while with (4.1) and (4.6) suppression of doublers requires

(1) c , for a 0.
2
(4.8)
0
It is seen that (4.2) is a prerequisite for (4.8). In the special case c = 0 (4.7) is satisfied
putting = |m|a 1 , by which (4.8) gets 2|m|a 1 and is fulfilled, too. In the general case a
simple way to satisfy (4.7) and (4.8) as well as (4.2) and (4.4) is to put
c = |m|a 1 c ,
and to require

(2 + 1)c = 1,
0
(4.9)

0
(1) c = 0,

0
|c | < .
(4.10)
209
Obviously also weaker conditions than this can be formulated.

With respect to topological charge and chiral anomaly nothing remains to be shown. The
reason for this is that only V enters the relation for the index (2.14) and that the overlap V
is safely known to give the correct limit of the topological charge (see Ref. [15] for a proof
and a discussion of literature). Thus the correct results are also guaranteed here.
4.2. Construction with spectral functions
We next give a more general construction of Dirac operators of the class using
the functions f of the spectral representation (2.3) as a tool. We start noting that
condition (2.5), f (ei ) + f (ei ) ei = 0, can be written as
i/2 i

ie
(4.11)
f e
= iei/2 f ei .
This shows that f is of form

f ei = ei()/2 g(),
g() real.
(4.12)
Then from condition (2.4), f (ei ) = f (ei ), the requirement

g() = g(),
(4.13)
follows. Further, with (2)-periodicity in of f (ei ), the function g() has to satisfy
g( + 2) = g().
(4.14)
We now see that forms of D = F (V ) can be obtained by determining functions g() which
are real, odd, and satisfy (4.14).
With the indicated requirements the basic building blocks for the construction of g
are the functions sin(/2) and cos(/2) with integer and . The conditions on g
are fulfilled by sin(/2) with odd and sin(/2) cos(/2) with even and odd ,
where the latter form, however, can be expressed by terms of former type. Further, after
multiplication with a function of the cos(/2) with even integers one has still the
required properties. Thus, with the notation
t = cos , , integer,
s = sin(2 + 1) ,
(4.15)
2
we arrive at the form

g=
(4.16)
s w (t1 , t2 , . . .),
where the w are real functions. Because of the identity

t ,
s = s0 1 + 2
(4.17)
=1
(4.16) can be simplified to

g = s0 w(t1 , t2 , . . .),
(4.18)
210
with a real function w. Further, since the t are given by polynomials of t1 ,

2
22
t2 = d2 t1 + d22 t1
2+1
t2+1 = d2+1 t1
+ + d0 ,
21
+ d21 t1
+ + d1 t1 ,
(4.19)
(4.16) can be cast into the still simpler form

g = s0 w(t1 ).
(4.20)
Condition (2.12), f (1) = g() = 0, in case of (4.20) is seen to require

w(1) = 0.
(4.21)
Further, w(t1 ) should, of course, be finite in the region of interest.

The form (4.20) essentially extends the one found in Section 4.1 to the case where no
series expansion in V is available. To see this one has to note that there according to (4.3)
we have

c s = s0 b0 /2 +
b t = s0 w(t1 ),
g=2
(4.22)
0
1
where the second form

follows inserting (4.17) into the first one and introducing
coefficients b = 2 c . Then using (4.19) from the second one the last one is
obtained, which is (4.20) specialized to the case where w(t1 ) is given by a power series.
The operator D for the construction (4.12) with (4.20),

1
f ei = ei()/2 s0 w(t1 ) = 1 ei w(cos ),
2
inserting (4.23) into (2.3) becomes

1
1
D = (1 V )W
V +V ,
2
2
(4.23)
(4.24)
where the operator function W has to satisfy

W (X) = W X ,
W (P ) = 0,
for P = P 2 = P > 0.
(4.25)
+ V )) which
To check the continuum limit it remains to consider the
as compared to the GW case (1.3) replaces the constant 2. This can again be done with
the overlap V . For 12 (V + V ) in the free case and infinite volume we get with = ap
operator W ( 12 (V

1
() 1, for a 0,
V () + V
2
while at the corners of the Brillouin zone we have
1
V + V = 1.
2
Thus the requirements (4.7) and (4.8) generalize to
(1) 2|m|a 1,
W
(1) ,
W
for a 0.
(4.26)
(4.27)
(4.28)
211
These conditions can be satisfied by putting

= 2|m|a 1W
,
W
(1) = 1,
W
(1) = 0,
W
(4.29)
though again also weaker conditions can be formulated.

With respect to the meaning of these conditions we note that the eigenvalues of V in the
,
free case are those of V

2 + s2,
ei = i s 2
(4.30)
where
s2 =
sin 2 ,
=m+r

(1 cos ),
2r < m < 0.
(4.31)
The real eigenvalues in (4.30) obviously occur for = 0, and one gets +1 if all = 0
and 1 at each corner of the Brillouin zone. Noting that

1
() = / 2 + s 2 ,
() + V
cos = V
(4.32)
2
(1) and W
(1) are seen to be related to the behavior of V at its
the conditions on W
eigenvalues +1 and 1, respectively.
4.3. Extension of spectral construction
To extend the construction of the preceding section we note that given a function g with
the required properties, then h(g) is again a function with such properties provided that h
is odd and real,
h(x) = h(x),
h(x) = h(x),
for real x.
(4.33)
With this the form (4.20) generalizes to

g = h s0 w(t1 ) .
(4.34)
Now condition (4.21) extends to

h w(1) = 0,
(4.35)
which is satisfied anticipating the requirement of strict monotony of h implied by (4.52).

Looking for functions h an important observation is that because of the identity

k
k

2k + 1
2k+1
s0
(4.36)
= s0
t , k = 0, 1, 2, . . . ,
(1) 1 + 2
=0
=1
(4.34) reduces to the form (4.20) if h(x) is an integer power, is a polynomial, or allows
an expansion in powers of x. This greatly restricts nontrivial choices of h(x), i.e., ones by
which (4.34) gives something beyond (4.20).
An example of nontrivial choices in this sense, which can be expressed in an elementary
way, is
h(x) = x 1/(2k+1),
k = 1, 2, . . . ,
(4.37)
212
where the real one of the (2k + 1)th roots is to be taken. One should note that with this one
has a different nontrivial choice for each k.
The nontrivial choices of h actually are equivalence classes, i.e., equivalent ones are
(2k+1) r(x) with
not to be counted as different. For example, to (4.37) the form h(x)
=x
r(x) = r(x) is equivalent, because with r(x) = r(|x|) and |s0 | = 12 (1 t1 ) it gives the
same form in (4.34) as (4.37). Also forming in addition odd powers because of (4.36) gives
nothing new.
The number of nontrivial choices of h(x) which can be explicitly expressed in terms
of elementary functions appears to be limited. We give a further explicit example in
Section 4.7 and also point out strategies for more general constructions there.
The operator D for the construction (4.12) with (4.34),

f ei = ei()/2 h s0 w(t0 ) ,
(4.38)
inserting (4.38) into (2.3) becomes

1
1 1/2
V
V +V
V 1/2 W
D = iV 1/2 H
,
2i
2
(4.39)
where V 1/2 is defined with +eik /2 in its spectral representation, W has to satisfy (4.25)
and for the function H one needs
H (X) = H (X),
H (X) = H (X),
for X = X.
(4.40)
Eq. (4.39) demonstrates the drastic increase of possible forms of Dirac operators which
occurs as compared to the GW case.
4.4. Example h(x) = x 1/(2k+1)
To check the continuum limit for a particular choice of a nontrivial h, we use (4.37)
as an example, where h is conveniently represented in terms elementary functions. With it
(4.12) becomes
1/(2k+1) 1

1/(2k+1)

= 1 ei
w(cos )1/(2k+1). (4.41)
f ei = ei()/2 s0 w(t1 )
2
The operator D inserting (4.41) into (2.3) and putting W 1/(2k+1) = Wk becomes
1/(2k+1)

1
1
(1 V )(V )k
V +V ,
Wk
D=
2
2
(4.42)
where of the (2k + 1)th roots the one which satisfies (2.7) is to be chosen and where W is
subject to (4.25).
Obviously (4.42) generalizes our form (3.10) of Fujikawas proposal replacing the
constant a01 there by a function Wk ( 12 (V + V )). Using V given by (3.7) with (3.8),
in the free case and infinite volume one gets with = ap

2k+1
2k+1
() 1 a
V
p
, for a 0,
(4.43)
i
|m|
213
= 1. The requirements (4.28) thus

while at the corners of the Brillouin zone one has V
generalize to
k (1) ,
W
k (1) 21/(2k+1)(1)k |m|a 1 ,

W
for a 0.
(4.44)
k (1) = 0,
W
(4.45)
This can be satisfied by putting

k = 21/(2k+1)(1)k |m|a 1 W
k ,
W
k (1) = 1,
W
or by some weaker condition. In the special case where Wk is replaced by the constant a01
according to (4.45) one gets1
a01 = 21/(2k+1)(1)k |m|a 1.
(4.46)
The eigenvalues of V in the free case here are

2k+1 2 2k+1
ei = k i s 2
k + s 2
,
with
s2 =
sin 2 ,
k = m
(4.47)
2k+1

2k+1
+ r
(1 cos )
,
2r < m < 0.
(4.48)
In (4.47) the eigenvalues also get +1 if all = 0 and 1 at each corner of the Brillouin
zone. Noting that

1
() = k / 2 + s 2 2k+1 ,
() + V
cos = V
(4.49)
k
2
(1) and W
(1) are again seen to be related to the behavior of V at its
the conditions on W
eigenvalues +1 and 1, respectively.
4.5. Comparison of conditions on H and W
The example of the Dirac operator (4.42) in Section 4.4 contains the operator related
to the GW relation (1.1), Fujikawas operator (3.6), and our constructions (4.1) and (4.24)
(where (4.1) has been seen to be the expansion version of (4.24)) as special cases. Thus it
allows us to compare the features of the continuum limit for V from (3.7) with (3.8) (the
case with more general V being considered in Section 4.6).
The behavior of V is essentially determined by the choice of h(x), which for h(x) =
x 1/(2k+1) is given by (4.43),
2k+1
2k+1

() 1 a
i
p
, for a 0.
V
(4.50)
|m|
from (3.6) as in literature the factor (1)k also arises by properly calculating
(2k+1) = (1)k ( s )(2k+1) before taking the (2k + 1)th root.
s
)
1 Starting
5 (
214
This makes the form for k > 0 as well as for k = 0 explicit and illustrates the impact of the
function H (which is more generally seen in (4.56)).
With respect to the condition on W , which leads to the usual normalization of the
continuum propagator, one has from (4.44),
k (1) 21/(2k+1)(1)k |m|a 1 ,
W
for a 0,
(4.51)
which as well as the suppression of doublers is guaranteed by (4.45). In Fujikawas

case Wk = a01 is a constant, (4.51) becomes a01 = 21/(2k+1)(1)k |m|a 1 and (4.45) is
trivially satisfied. In case of the GW relation (1.1), where W = a01 = 2, this reduces to
= |m|a 1 . Thus the straightforward connection to the special cases becomes obvious and
it is seen that the essential requirements are the same ones in all cases.
4.6. Related sets of V
Clearly the determination of the functions F (V ) and the selection of the operators V are
distinct issues. Nevertheless they are not completely independent. For the choice h(x) = x
the set of V is appropriate which is known to be suitable already in the GW case. However,
for h(x) = x 1/(2k+1) for different k different V are needed. This suggests that the nontrivial
choices of h relate to distinct sets of V .
Indeed, what occurs in the above examples extends to other functions h for which a
strictly monotonous inverse function exists,

h (x) = x,
(y) > (x), for y > x.
(4.52)
In terms of operators one then has a corresponding Hermitian function E with H (E(X)) =
X. This allows us to introduce the general normalization-type definition of V ,

()
() () 1
DW DW
,
V = DW
(4.53)
with

()
DW = iE

1
r
+ E
+ E(m1),
2i
2
(4.54)
where the function E of the operators (which are Hermitian) is well-defined. This definition
of V is to be inserted into the general expression for D,

1
1 1/2
V
V +V
,
V 1/2 W
D = iV 1/2 a 1 H
(4.55)
2i
2
which is (4.39) up to a factor a 1 . In the free case for this V at the corners of the Brillouin
= 1 and at zero
zone one gets V

i

E a
p , for a 0.
V 1
(4.56)
|(m)|
(1) = 0, because of the monotony of E(X) doublers are suppressed for

Requiring W
(1) = 2|(m)| the correct limit of
2r < m < 0 as usual. Since H (E(X)) = X, putting W
the propagator is obtained in this quite general way.
215
We note that the strict use of Hermitian functions of Hermitian operators in (4.53)
with (4.54) is what makes it applicable in a general way. In contrast to this the formulation
of (3.7) with (3.8) appears restricted to the choice h(x) = x 1/(2k+1). Treating the latter
choice with the more satisfactory form (4.53) with (4.54), instead of (4.43) one gets

2k+1

a 2k+1

p
, for a 0,
V () 1 i
(4.57)
|m|
with different position of the factor i. On the other hand, instead of the first relation
in (4.44) one then has
(1) = 2|m|2k+1,
W
k (1) = 21/(2k+1)|m|,
W
(4.58)
in which the factors (1)k and a 1 of (4.44) do not occur. The latter factor is now
incorporated in the definition (4.55) of D while the former one is accounted for by the
different position of i in (4.57). Thus for the choice h(x) = x 1/(2k+1), for which both
formulations of V are applicable, one gets only minor differences in the representations.
4.7. Chirality of D 1
If the inverse D 1 exists one may multiply (1.1) from both sides with it, which gives
1
D , 5 = 1 .
(4.59)
This relation indicates that the propagator is chiral up to a local contact term, which is
desirable to get the Ward identities of chiral symmetry in the continuum limit. In case of
Fujikawas operator this relation becomes

1/(2k+1)
1
,
D , 5 = 2a02k+1(D5 )2k = a0 2(1 V )2k (V )k
(4.60)
so that the locality of interest follows from that of D. For the example of the Dirac operator
(4.42), which contains these operators as well as our constructions (4.1) and (4.24) as
special cases, we obtain

1
1
1
2k
k 1/(2k+1)
V +V
Wk
.
D , 5 = 2(1 V ) (V )
(4.61)
2
Since this differs from (4.60) only in that a0 is replaced by the function Wk ( 12 (V + V ))1 ,
the additional requirement is that the properties of W should be such that the product on
the r.h.s. of (4.61) remains local.
The above strategy to relate the locality of {D 1 , 5 } to that of D relies on the
introduction of an inverse function to D = F (V ),
V = 1 I (D),
with which from the basic relation (1.4) one generally obtains
1
D , 5 = 5 I (D)D 1 .
(4.62)
(4.63)
216
In the special case of (4.60) according to (3.5) one has I (D) = 2a02k+1(DD )k D so that
I (D)D 1 = 2a02k+1(DD )k . Analogously in the general case it is seen that I (D) must be
such that locality of D implies that of I (D)D 1 .
To study this in more detail we consider the subclass of Dirac operators for which
I (D)D 1 = (1 V )D 1 is a polynomial of DD ,
I (D)D
N

C DD ,
(4.64)
=0
with real coefficients C . In terms of the spectral functions this means that

1
1 ei f ei
=
C f ei f ei
,
(4.65)
=0
which inserting (4.12), f (ei ) = ei()/2 g(), becomes

2s0 () =
N
C g()2+1 ,
=0
s0 = sin .
2
(4.66)
To obtain D by inserting f into (2.3), we have to determine g, which according to (4.66)

requires to solve the algebraic equation
N
C g 2+1 2s0 = 0.
(4.67)
=0
The solution of (4.67) is trivial if only one of the coefficients C is different from zero,
which gives Fujikawas proposal.
In order to derive a more general explicit solution we consider the case where the two
coefficients C0 and C1 are nonzero. We then have the cubic equation
g 3 + 3pg + 2q = 0,
p=
C0
s0
, q = .
3C1
C1
(4.68)
Requiring C0 > 0 and C1 > 0 it follows that p3 + q 2 > 0, which implies that one gets the
real solution

1/3
1/3
g = q + q 2 + p3
(4.69)
+ q q 2 + p3
.
Thus in more detail we arrive at

s0
g=h
C1

2
1/3
2
1/3
C0 3
s0
C0 3
s0
s0
s0
+
+
+
+
, (4.70)
=
C1
C1
3C1
C1
C1
3C1
where the new nontrivial function h has the required properties of being odd and strictly
monotonous. The associated inverse function according to (4.68) is (g) = 12 (g 3 + CC0 g).
1
217
In the related operator expression (4.55) for D now H is of form (4.70) with s0
replaced by 2i1 (V 1/2 V 1/2 ) and W is constant, W = 1/C1 . The condition for the usual
normalization of the propagator in the continuum limit thus becomes 1/C1 = 2|(m)| or
C1 |m|3 + C0 |m| = 1.
It is now seen how one could proceed to still more general cases. Firstly, one
could solve (4.67) for more coefficients. Secondly, one could consider more complicated
expressions on the r.h.s. of (4.64).
It should, furthermore, be noted that instead of the indicated strategy in many cases it
may be more convenient to analyze directly

1/2

1
1
1 1/2
1
1/2
1/2
V
V +V
H
W
V
V
,
D , 5 = ia V
2i
2
(4.71)
which follows from the general form (4.55) of D. Generally the desired locality obviously
puts some additional conditions on the choices of the functions H and W , however, leaving
still much freedom.
5. Weyl operators and chiral gauge theories

The chiral projection operators implicit in the overlap formalism of Narayanan and
Neuberger [8] and used in the formulation of Lscher [3] in our notation read
1
1
P = P = (1 5 V )1.
P = P = (1 5 )1,
(5.1)
2
2
Obviously only 5 and V are involved in them and we can start with them only requiring
generalized chiral symmetry. From condition (1.4) we get the identity D = 12 (D
5 D5 V ) and inserting 5 = P+ P and 5 V = P+ P into it we obtain
D = P+ D P + P D P+ .
(5.2)
With this we have the relations

P D P = D P = P D,
(5.3)
which generalize the expressions for the Weyl operators in terms of the Dirac operator
familiar in continuum theory.
With respect to possible forms of (5.3) one should be aware of the fact that the relations
P 5 = 5 1,
5 V P = P ,
(5.4)
allow to absorb parts of D. In the special case of the Dirac operator (1.3), with (5.4) one
gets P+ (1 V )P = 2P+ P , which relates the different forms of the chiral determinant
in Ref. [3] and in Ref. [8], commented on in Refs. [16,17]. Considering the general class
of operators D here, we have to observe that (5.3) is the generally valid form and that
modifications by (5.4) depend on the particular choice of D. Furthermore, we also note
that for any operator C satisfying 5 C C5 V = 0 one gets P D P = P (D + C)P ,
which provides another possibility to modify (5.3).
218
For the numbers of the degrees of freedom Tr P+ and Tr P of the Weyl fermions in
P+ D P one gets from (5.1)
1
Tr(5 V ),
(5.5)
2
which obviously depends only on V for the whole class and agrees with the result (2.14)
for the index of the Dirac operator D.
The degrees of freedom are exhibited in more detail representing the projections by

=
P+ =
(5.6)
uj uj ,
P
u k u k ,
ui uj = ij ,
u k u l = kl ,
Tr P+ Tr P =
with basis vectors relating the degrees of freedom to the representation in full space.
Associating Grassmann variables j and k to the degrees of freedom, the fermion field
variables then get

=
(5.7)
j uj ,
=
u k k .
j
With this for Tr P = Tr P+ correlation functions are given by

(d l dl ) exp D
n(1) n(r1 ) n(2) n(r2 ) n(f ) n(rf )
l

=s1 s2 ...sf P D 1 P+ n(s
s1 ,...,sf
1 )n(r1 )

P D 1 P+ n(s
f )n(rf )
(5.8)
det M
(5.9)
where the matrix M occurring in the chiral determinant is

Mj k = uj D u k .
(5.10)
We thus have arrived at the usual basic relations of chiral gauge theories using only
generalized chiral symmetry and not the GW relation.
6. Conclusions
We have shown that generalized chiral symmetry provides the general principle for the
description of massless fermions on the lattice and for the formulation of chiral gauge
theories there, while the GW relation and its generalization by Fujikawa occur only as
special cases in this context. It has turned out that this principle gives a much larger class of
operators than one has in the GW and Fujikawa cases. We have given a general construction
of these operators and an associated realization of the basic unitary operator V involved.
The index of the Dirac operator has been found to depend only on V .
Acknowledgement
I wish to thank Michael Mller-Preussker and his group for their kind hospitality.
219
References
[1] P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
[2] M. Lscher, Phys. Lett. B 428 (1998) 342.
[3] M. Lscher, Nucl. Phys. B 549 (1999) 295;
M. Lscher, Nucl. Phys. B 568 (2000) 162.
[4] K. Fujikawa, Nucl. Phys. B 589 (2000) 487.
[5] K. Fujikawa, M. Ishibashi, H. Suzuki, Phys. Lett. B 538 (2002) 197.
[6] T.-W. Chiu, Phys. Rev. D 58 (1998) 074511.
[7] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125.
[8] R. Narayanan, H. Neuberger, Phys. Rev. Lett. 71 (1993) 3251;
R. Narayanan, H. Neuberger, Nucl. Phys. B 412 (1994) 574;
R. Narayanan, H. Neuberger, Nucl. Phys. B 443 (1995) 305.
[9] W. Kerler, JHEP 0210 (2002) 019.
[10] M.F. Atiyah, I.M. Singer, Ann. Math. 87 (1968) 546, Section 5.
[11] H. Neuberger, Phys. Lett. B 417 (1998) 141;
H. Neuberger, Phys. Lett. B 427 (1998) 353.
[12] T.-W. Chiu, Phys. Lett. B 521 (2001) 429.
[13] W. Kerler, Phys. Lett. B 510 (2001) 325.
[14] T.-W. Chiu, Nucl. Phys. B 588 (2000) 400.
[15] W. Kerler, Int. J. Mod. Phys. A 16 (2001) 3117.
[16] H. Neuberger, in: V. Mitrjushkin, G. Schierholz (Eds.), Lattice Fermions and Structure of the Vacuum,
Kluwer Academic Publishers, 2000, p. 113.
[17] M. Golterman, Nucl. Phys. (Proc. Suppl.) 94 (2001) 189.

Higgs boson production at hadron colliders in

NNLO QCD
Charalampos Anastasiou, Kirill Melnikov
Stanford Linear Accelerator Center, Stanford University, Stanford, CA 94309, USA
Abstract
We compute the total cross-section for direct Higgs boson production in hadron collisions at
NNLO in perturbative QCD. A new technique which allows us to perform an algorithmic evaluation
of inclusive phase-space integrals is introduced, based on the Cutkosky rules, integration by parts and
the differential equation method for computing master integrals. Finally, we discuss the numerical
impact of the O(s2 ) QCD corrections to the Higgs boson production cross-section at the LHC and
the Tevatron.
1. Introduction
The Higgs boson is currently the only missing particle in the minimal Standard Model
(SM) of electroweak interactions. Its discovery will be one of the final steps toward the
experimental verification of the SM, and will provide useful input for detailed studies of
the mass generation mechanism and for physics beyond the SM.
Direct searches at LEP restrict the Higgs boson mass to be greater than 114.1 GeV [1],
while a global fit to precision electroweak measurements [2] favors a value around 90 GeV.
In addition, the requirement that the SM remains perturbative up to relatively high energy
scales sets an upper bound at approximately 1 TeV [3]. Although the above evidence is not
completely conclusive, it indicates a relatively light Higgs boson which could be observed
at either the Tevatron or the LHC. At both of these facilities, gluon fusion through top-quark
loops is expected to be the dominant Higgs production mechanism. All other channels,
such as vector boson fusion qq H qq and associated Higgs production q q H W , are
E-mail addresses: babis@slac.stanford.edu (C. Anastasiou), melnikov@slac.stanford.edu (K. Melnikov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 3 7 - 4
C. Anastasiou, K. Melnikov / Nuclear Physics B 646 (2002) 220256
221
suppressed by about an order of magnitude (see Ref. [4] for a review). We, therefore, focus
upon the process gg H in this paper.
The theoretical estimates of the cross-section for the Higgs boson production via gluon
fusion, based on computations through to the next-to-leading order (NLO) in perturbative
QCD, turn out to be insufficient. The leading-order (LO) cross-section is proportional to
s2 (2 ), and for this reason exhibits a strong dependence on the choice of the scale .
Including the O(s ) corrections [5,6] decreases the scale dependence, but the cross-section
increases by a very large amount, approximately 70%. It is therefore important to evaluate
the next order in the perturbative expansion, since this is the only way to enhance the
credibility of the theoretical predictions.
To compute the cross-section to next-to-next-to-leading order (NNLO), we must
combine: the matrix elements for the O(s2 ) virtual corrections to gg H ; the matrix
elements for the O(s ) virtual corrections to gg H g, qg H q, and q q H g; and
finally the tree-level matrix elements for the processes gg H gg, gg H q q,
qg
H qg, q q H gg, and q q H q q.
For the inclusive cross-section we must integrate over
the loop-momenta in the virtual amplitudes and the phase-space of the real particles in the
final state. Both real and virtual corrections are divergent in four dimensions. We regularize
the amplitudes using conventional dimensional regularization (d = 4 2), and remove the
ultraviolet divergences by renormalizing in the MS scheme. The remaining divergences
arise from initial state collinear radiation and are absorbed into the parton distribution
functions, yielding a finite cross-section.
The calculation can be simplified substantially by considering the limit where the Higgs
boson is much lighter than twice the mass of the top-quark. In this limit, the top-quark
loops are replaced by point-like vertices. The corresponding effective Lagrangian is known
to provide a satisfactory description of the cross-section for a light Higgs boson at NLO
[5,6].
In the heavy top-quark limit, the NNLO contributions to the direct Higgs production
cross-section are topologically similar to the O(s2 ) corrections to the DrellYan process
which have been calculated in the past [8]. The phase-space and loop integrals required for
the calculation of the Higgs boson production cross-section could in principle be obtained
in a similar fashion. However, such an approach is impractical for the Higgs production
cross-section due to the larger number of Feynman diagrams with a considerably more
complicated tensor structure. For a problem of this complexity a highly automated
algorithm which treats virtual and real corrections in a unified manner is desirable.
It is well known how to construct algorithms which in principle can perform multiloop integrations. First, one can employ the method of integration by parts (IBP) [9] in
order to reduce the number of integrals involved in such computations. Algorithms which
find the solutions of IBP identities in a process and topology independent manner are
available [11,12]. After the application of IBP, a small number of remaining integrals
which are not reducible further (master integrals), must be evaluated explicitly. Powerful
techniques such as the differential equation method [10,11] and the MellinBarnes integral
representation [13] can then be employed to derive an expansion of the master integrals
in . The above methods provide general purpose tools for the evaluation of virtual
corrections. However, similar methods do not exist for computing phase-space integrals;
they are usually calculated manually and on a case by case basis. In this paper we present
222
an algorithmic procedure for evaluating phase-space integrals, based on the Cutkosky

rules [14], integration by parts [9] and the differential equation method [11].
Partial results for the NNLO corrections to the Higgs boson production cross-section
are available in the literature. The NNLO virtual corrections were computed in [15]
by Harlander. The soft part of the cross-section at NNLO was derived in [16,17] by
extracting contributions
to the partonic cross-sections that are singular when the partonic
center of mass energy s equals the mass of the Higgs boson mH .
Recently, Harlander and Kilgore [18] obtained an excellent approximation to the
complete NNLO result by expanding the phase-space integrals around the kinematic point
s = m2H . In this paper we present the full analytic result for the NNLO corrections to the
Higgs boson production cross-section. In our derivation we do not need to resort to an
expansion around a special kinematic point and our expressions are therefore valid for an
arbitrary ratio m2H /s .
The paper is organized as follows. In Section 2 we briefly review the effective
Lagrangian for describing gluon interactions with the Higgs boson. We also introduce
our notations and present all basic formulae and definitions for the total cross-section.
In Section 3 we describe our method for solving multi-particle phase-space integrals
in an algorithmic fashion and illustrate its application with a few typical examples. We
present the analytic expressions for the renormalized partonic cross-sections ij H + X
in Section 4. In Section 5 we discuss the impact of the O(s2 ) corrections on the Higgs
boson production cross-section at the Tevatron and at the LHC. We present our conclusions
in Section 6. Some useful formulae, including the complete list of master integrals, are
collected in Appendix A.
2. Effective Lagrangian
The Higgs boson interaction with gluons is a loop induced process and is therefore
sensitive to all colored particles which get their masses through the Higgs mechanism. In
this paper we restrict ourselves to the Standard Model where the top-quark contribution
dominates.
Although the Born cross-section is known as a function of the top mass mt and the
Higgs boson mass mH , it is much harder to obtain the exact analytic dependence of the
cross-section on the mass of the top-quark in higher orders of perturbation theory. However,
since it is most probable that the Higgs boson is light, it is sufficient to work in the infinite
top-quark mass limit.
For Higgs boson masses in the range 100200 GeV, we can describe the Higgs gluon
interaction by introducing the effective Lagrangian [57]
Leff =
1
C1 Ga Ga H,
4v
(1)
where Ga is the gluon strength tensor, H is the Higgs field and v 246 GeV is the Higgs
boson vacuum expectation value. The Wilson coefficient C1 , defined in the MS scheme,
223
is [19]
C1 =

2

11 s
s
2777 19
67 1
1
1+
+
+ Lt + nf + Lt
+ O s3 ,
3
4
288
16
96 3
(2)
where s () is the MS strong coupling constant, nf = 5 is the number of active flavors

and Lt = log(2 /m2t ).
It is expected that the effective Lagrangian of Eq. (1) is a valid approximation to
the Higgs gluon interaction for small values (mH < 2mt ) of the Higgs boson mass. It
can be checked that at leading order and for mH 150 GeV, the effective Lagrangian
approximation is accurate within 5%, whereas for mH 200 GeV, the accuracy drops to
10%. The precision of the approximation improves for the Higgs boson production crosssection computed at NLO accuracy [6]. The effective Lagrangian description therefore
seems accurate in the entire range of phenomenologically interesting Higgs boson masses,
and we adopt it for the calculation of the NNLO corrections.
The effective Lagrangian approach separates the short-distance ( m1
t ) and the longdistance ( m1
)
scales,
simplifying
the
calculation
of
the
Higgs
boson
production crossH
section. For example, the original one-loop triangle diagram of the gluongluon Higgs
interaction vertex at LO is now replaced by the simple tree-level vertex derived from
Eq. (1). The effective Lagrangian approach also yields the correct H ggg and H gggg
interaction vertices, as a consequence of gauge invariance. In addition, in the limit of
vanishing fermion masses, there is no direct interaction between massless quarks and the
Higgs boson.
The partonic cross-sections for the production of the Higgs boson, up to NNLO in
perturbation theory, receive the following contributions: (a) virtual corrections to gg H ,
up to O(s2 ); (b) virtual corrections to single real emission processes gg H g, qg H q,
qg
H q,
q q H g, up to O(s ); and (c) double real emission processes gg H gg,
gg H q q,
qg H qg, qg
H qg,
q q H gg, q q H q q to LO. The effective
Lagrangian of Eq. (1) and the corresponding matrix elements should be renormalized in
the MS scheme by a global renormalization factor [19]:
2 2

0
s
s 0
1
+
+ O s3 ,
Z(s ) = 1
(3)

where 0 and 1 are the two first coefficients of the QCD -function:
11 1
51 19
(4)
1 =
nf ,
nf .
4
6
8
24
This additional renormalization, together with the standard renormalization of the strong
coupling constant removes all ultraviolet divergences in the cross-section. However, since
we work in the approximation of massless colored partons in the initial state, the total crosssection is not finite even after the ultraviolet renormalization has been performed. The
remaining singularities are associated with the collinear radiation off the colliding partons.
It is well-known that these singularities factorize and can be removed by renormalizing the
parton distribution functions in a manner consistent with the DGLAP evolution equation.
0 =
224
The general factorization formula for the cross-section of the Higgs boson production
from the collision of two hadrons h1 (p1 ) and h2 (p2 ), is
h1 +h2 H +X =

dx1 dx2 fi(h1 ) (x1 )fj(h2 ) (x2 )ij H m2H , x1 x2 s ,
(5)
ij 0
(h)
where fi (x) is the standard distribution function for a parton i in the hadron h, ij is the
partonic cross-section for i + j H + X and s (p1 + p2 )2 is the square of the total
center of mass energy of the hadronhadron collision. Using dimensional analysis we can
write the partonic cross-sections in terms of the single dimensionless variable z = m2H /s ,

ij 1/v 2 g m2H /s
(6)
with s sx1 x2 . Introducing x = m2H /s, we then rewrite Eq. (5) in the form
(h )

fi 1 fj(h2 ) ij (z)/z (x),
h1 +h2 H +X = x
(7)
ij
where the standard convolution is defined as
1
[f1 f2 ](x) =
dx1 dx2 f1 (x1 )f2 (x2 )(x x1 x2 ).
(8)
The collinear singularities are factored out from the partonic cross-sections with the
following procedure. Denoting the unrenormalized (in the sense of collinear singularities)
partonic cross-sections by ij , the MS renormalized partonic cross-sections ij are
implicitly given by:

ij (x)/x =
(9)
kl (z)/z ki lj ,
k,l
where the kernels ij are

2
(0)

s
s Pij (x)
1 (0)
(0)
(0)
+
ij = ij (1 x)
Pik Pkj (x) + 0 Pij (x)
2
2

3
1 (1)
Pij (x) + O s .
(10)
2
The standard space-like splitting functions Pij [20,21] are listed in Appendix A. We can
easily solve Eq. (9) for ij (x)/x = ij (x) order by order in s . It is convenient to introduce
a matrix notation and rewrite Eq. (9) as
= T ,
(11)
where is the matrix of partonic cross-sections in flavor space and is the matrix with
components ij (x) as in Eq. (10). We then write
2

s
s
2 + O s3 ,
= (1 x)U 1 +
(12)
225
where the matrix U has the components Uij = ig jg and 1,2 can be read off from
Eq. (10). Inverting Eq. (11) to obtain
(1)
[ ]1 ,
= T
and expanding in s
= (0) +
2
s (1)
s
+
(2) ,
(13)
we find
(0) = (0) ,
(1) = (1) + 1T (0) + (0) 1 ,
(2) = (2) 2T (0) (0) 2 1T (0) 1 + 1T (1) + (1) 1 .

(14)
Having derived the finite partonic cross-sections ij , we must convolute them with the
MS parton distribution functions fi to obtain the total hadronic cross-section:
(h )

(h )
fi 1 fj 2 ij (z)/z (x).
h1 +h2 H +X = x
(15)
ij
We present our results for the partonic cross-sections ij (z) in Section 4. In Section 5 we
use Eq. (15) to calculate the Higgs boson production cross-section at the Tevatron and at
the LHC.
3. Method
In this section we describe the method employed to compute the partonic cross-sections
to NNLO. At this order we must calculate three distinct contributions:
double-virtual: the interference of the Born and the two-loop amplitude as well as the
self-interference of the one-loop amplitude for gg H ,
+ 148 terms;
real-virtual: the interference of the one-loop and the Born amplitudes for gg H g,
gq H q, and g q H q,
+ 635 terms;
double-real: the self-interference of the Born amplitudes for gg H gg, gg H q q,
gq H gq, g q H g q,
qq H qq, and q q H q q,
+ 594 terms.
226
The above interference terms are produced in a form convenient for further evaluation
using the QGRAF package [22] for generating Feynman graphs.
In the following subsections we briefly describe the available techniques for evaluating
virtual corrections and explain our method for integrating over the phase-space of the final
state particles.
3.1. Virtual corrections
There currently exists a general method which permits the systematic evaluation of
multi-loop virtual corrections. In order to calculate a multi-loop amplitude we must
first reduce the number of Feynman integrals. The hypergeometric structure of Feynman
integrals guarantees that simple algebraic relations between various scalar integrals exist,
making such a reduction possible. One method of producing these relations is the
integration by parts (IBP) technique [9]. In cases where the system of equations is
not complete, it can be supplemented with additional identities that exploit the Lorentz
invariance (LI) of scalar integrals [11].
In general, IBP and Lorentz Invariance (LI) identities relate integrals of differing
complexity. For example, it is possible that a single IBP equation relates an integral
with an irreducible scalar product to integrals with no irreducible scalar products or to
integrals with fewer propagators. A typical situation, however, involves multiple IBP
and LI identities relating several equally complicated integrals to a set of simpler ones.
In such cases, every integral must be written exclusively in terms of simpler ones and,
eventually, expressed in terms of a few master integrals which cannot be reduced further.
Unfortunately, finding recursive solutions of the IBP and LI identities is tedious, and
may be impossible in complicated cases. Also, a separate treatment of each different
topology in a Feynman amplitude is required. Consequently, the whole procedure becomes
increasingly cumbersome with the introduction of more kinematic variables and loops.
We may alternatively consider a sufficiently large system of explicit IBP and LI
equations which contains all the integrals that contribute to the multi-loop amplitude of
interest. It should then be possible to solve the system of equations in terms of the master
integrals [11,12] using standard linear algebra elimination algorithms. In this approach, the
number of loops, the topological details, and the number of kinematic variables, affect only
the size of the system of equations and the number of terms in each of the equations; they
have no bearing on the construction of the elimination algorithm. This in principle allows
us to express any multi-loop amplitude in terms of master integrals.
One possible elimination algorithm has been proposed by Laporta [12]. This algorithm
exploits the fact that Feynman integrals can be ordered by their complexity; for example,
they can be arranged according to the number of irreducible scalar products and the
total number and powers of propagators. This observation distinguishes the IBP and LI
systems of equations from algebraic systems with no intrinsic ordering, and it becomes
possible to solve them iteratively, starting with the simpler equations and progressing to
more complicated. We use a variant of this algorithm, implemented in FORM [23] and
MAPLE [24].
After the reduction we must compute the analytic expansion in of the master integrals.
The coefficients of the expansion are typically expressed in terms of polylogarithms whose
227
rank and complexity depends on the number of loops and kinematic variables of the
integral in question. The MellinBarnes representation [13] and the differential equation
method [11] can be used to evaluate master integrals explicitly.
3.2. Reduction of phase-space integrals
In this subsection we extend the application of the above techniques to calculate phasespace integrals for inclusive cross-sections. To the best of our knowledge the method we
present is new, however, a somewhat related discussion has been given earlier in [25].
To illustrate our method, we consider the following double-real contribution at NNLO:

d
d
2
2
2
2

d q1 d q2 (q1 )(q2 )(qH mH )[ ] .

[(qH p1 )2 ]2 [(q2 p2 )2 ]2

(16)
Using the Cutkosky rules [14], we can replace the delta-functions in the above integral by
differences of two propagators:

2i p2 m2
p2
1
1
2
.
2
m + i0 p m2 i0
(17)
The r.h.s. of Eq. (16) is now equal to a forward scattering diagram:

=
(18)
where a cut propagator should be replaced by the r.h.s. of Eq. (17).

We have exchanged the square of a Born amplitude for a two-loop diagram, in contrast
to the usual application of the Cutkosky rules. We do this in order to utilize IBP and LI
relations between multi-loop integrals. The phase-space integrals can then be evaluated in
the same algorithmic fashion as the multi-loop integrals.
We begin our calculation by summing over the colors and spins of the external particles
in the cut two-loop integral on the right-hand side of Eq. (18). The original diagram is
then expressed in terms of a large number of scalar two-loop integrals to which the same
cutting rules apply. Crucially, we can use the IBP method to reduce the cut scalar integrals.
This is a consequence of the fact that the delta-function in Eq. (17) is represented in a very
simple manner by the difference of two propagators with opposite prescriptions for their
imaginary parts. We derive the IBP equations by integrating over total derivatives which
act on the propagators of the cut scalar integrals. The prescription for the imaginary part of
the two propagators in the r.h.s. of Eq. (17) is irrelevant for the differentiation. Therefore
the IBP relations for the two descendants of these two terms have the same form as the
IBP relations for the original integral without the cut. It is then allowed to commute the
application of IBP reduction algorithms with the application of the Cutkosky rules.
After the IBP reduction, the original phase-space integral is expressed in terms of a
small number of master integrals cut through the same three propagators as the initial
228
diagram:1
+ A1
+ A2
+ .
(19)
During the reduction, integrals with one or more of the cut propagators eliminated are
produced. From Eq. (17) we observe that such terms do not contribute to the original phasespace integrals. Therefore, we can immediately discard them simplifying the reduction
process.
A similar procedure can be applied to the virtual-real contributions. In this case, since
we perform the phase-space integration over two final state particles, the resulting master
integrals should be cut through two of the propagators:
= B1
+ B2
+ .
(20)
In order to have a unified algorithm for all three types of interferences, we treat the doublevirtual corrections as integrals with a single cut through the propagator of the Higgs boson.
= C1
+ C2
+ .
(21)
Finally we must evaluate the master integrals as a series expansion in = (4 d)/2.

Since each of the cut master integrals represents a well-defined phase-space integral, we
could compute them using brute force techniques similar to the ones described in Ref. [8].
However, we can instead utilize the IBP reduction algorithm in order to produce a set
of coupled first order differential equations [11] that the master integrals satisfy. It is
simpler to solve the differential equations than to reinstate the delta-functions for the cut
propagators and perform the integrations over the phase-space.
3.3. Evaluation of phase-space master integrals
To explain how the system of differential equations for the master integrals is obtained,
let us consider a two-loop scalar integral with a single Higgs boson propagator

dd k dd l
1
I s, m2H =
(22)
.
d
d
2
2
(2) (2) [k mH ] A11 Ann

1 Bold lines represent a massive Higgs propagator. Normal lines denote massless scalar propagators.
By differentiating with respect to m2H we obtain:
I(s, m2H )
dd k dd l
1
=
.
2
d
d
2
(2) (2) [k 2 mH ]+1 A11 Ann
mH
229
(23)
After applying the IBP algorithm, we can rewrite the r.h.s. of the last equation in terms of
the master integrals {Xi }, yielding

I
=
cj Xj .
(24)
2
mH
j
By identifying I = Xi , we derive a closed system of differential equations for the master
integrals,

Xi
=
cij Xj .
(25)
2
mH
j
These differential equations can be solved up to a constant in terms of logarithms and
generalized Nielsen polylogarithms, order by order in . This constant is obtained by
evaluating the master integral at a specific kinematic point. This is typically simpler and
can often be avoided using general arguments, as shown in an example below.
As an explicit example we discuss the calculation of the master integrals for the realvirtual corrections. The IBP reduction produces six master integrals which depend on two
variables: the mass of the Higgs boson, mH , and the square of the sum of the incoming
momenta, s = (p1 + p2 )2 . Note that the integrals depend on a single Mandelstam variable,
since they correspond to forward scattering diagrams with the same incoming and outgoing
momenta. It is convenient to express the master integrals in terms of the dimensionless ratio
z = m2H /s . We can further simplify our results by setting s = 1. The full dependence on s
can be restored by simple dimensional analysis.
Three of these master integrals are combinations of the one-loop massless bubble
integral and the two-body phase-space integral and can be easily evaluated, yielding

= PL Re ei (1 z)12 ,
= PL
1 2
(1 2)2
(1 z)13 ,
1 3 (1 ) (1 3)

= PL Re ei z (1 z)12 ,
(26)
(27)
(28)
where
3
P=
2
( 32 )232
L=
() (1 )2
.
(2 2)
(29)
230
The remaining three master integrals,
(30)
have a more complicated dependence on z and we compute them using the method of
differential equations. As an example, we discuss the differential equation for the first
integral in Eq. (30):
+
z
z

=
(1 2)2
z(1 z)
(1 2)(1 3)
z(1 z)
(31)
This differential equation is of the form

(z) f (z) = (z),

z
(32)
and has the general solution

z
x
z
dx (x)
dx (x )
C + dx (x)e
.
f (z) = e
(33)
Using Eq. (33) and the expressions for the two boundary master integrals in Eqs. (27), (28),
we derive
=z
2 (1 2)
Re e
dx x 1 (1 x)2
(1 2)2
+
(1 3) (1 )
z
dx x
12
(1 x)
+C .
(34)
We compute the value of the integral at a specific kinematic point in order to determine the
constant C. A convenient choice is the threshold for Higgs production, z = 1, where the
integral vanishes:
(z = 1) = 0.
(35)
231
The value of the integral at z = 1 can be inferred without an explicit calculation by

observing that, in the z 1 limit the two particle phase-space scales like (1 z), and
the one-loop triangle diagram on the r.h.s. of the cut scales like (1 z)0 . Requiring that
Eq. (34) vanishes at z = 1, one finds:

i
(1 2)2
B(, 1 2) +
C = Re e
(36)
B(2, 1 3) .
(1 3) (1 )
We now evaluate the integrals in Eq. (34). First, by changing the integration variables, we
isolate the singularity at x = 0:

dx x 1a (1 x)b = za
1
+
a

b 1
(1
zy)
dy y a
.
y
(37)
The integrand on the r.h.s. can then be expanded in , yielding generalized Nielsen
polylogarithms,
Snp (z) =
(1)n+p1
(n 1)!p!
1
dy
logn1 (y) logp (1 zy)

,
y
(38)
which reduce to usual polylogarithms for p = 1:

Sn1,1 (z) Lin (z).
(39)
We finally obtain
z
dx x
0
1+a
(1 x)

za
n p n+p
1
=
a b
Snp (z) .
a
n,p
(40)
Substituting this result in Eq. (34) and truncating the series at the order where polylogarithms of rank n + p > 3 start to appear, we arrive at the result:

z2 1
log2 (z)
Li2 (z) +
2
= P (1 + )
1 2
2
+ 5S12 (z) 4 Li3 (z) + 4 log(z) Li2 (z)

log3 (z)
+ 22 log(z) 3 + O() .
+
2
(41)
We repeat the same procedure for the differential equations for the two remaining integrals
in Eq. (30), where the master integral we have just calculated enters as a boundary term.
It is important to extract the singular behavior of the master integrals around z = 1 before
expanding in . This is essential since terms of the form (1 z)1+a in the cross-section
232
are expanded in in terms of plus distributions,

1
1
(a)2 log(1 z)
+
(1 z)1+a = (1 z) + a
a
1x +
2!
1z
+
2

3
4
(a) log (1 z)
+
+O ,
3!
1z
+
(42)
facilitating the cancellation between real and virtual soft and collinear singularities prior to
integration over z.
Finally, we apply the same technique to compute the double real master integrals.
Explicit formulae for the required master integrals are given in Appendix B.
4. Partonic cross-sections
In this section we present analytic expressions for the partonic cross-sections i + j
H + X of Eq. (15). We write

2

3
s (1)
s
(0)
(2)
ij = 0 ij +
(43)
ij + O s ,
ij +
where
0 =
2
s
,
576v 2
and s is the MS strong coupling constant evaluated at the scale r = mH . For simplicity,
the factorization scale is also set equal to the mass of the Higgs boson f = mH .
At leading order we find
ij(0) = (1 x)ig jg .
(44)
At next-to-leading order there are contributions from the gluongluon, quarkgluon and
quarkantiquark channels:

ln(1 x)
11
(1)
+ 62 (1 x) + 12
12x x + x 2 + 2 ln(1 x)
gg =
2
1x +
11
6(x 2 + 1 x)2
ln(x) (1 x)3 ,
1x
2

x
2
1
(1)
qg
= 1 + (1 x)2 ln
1 + 2x x 2 ,
2
3
3
(1 x)
(45)
q(1)
q =
32
(1 x)3 .
27
(46)
The main result of this paper is the next-to-next-to-leading order corrections, which we
separate according to their dependence on the number of quark flavors:
ij(2) = (2)A
+ nf (2)F
ij
ij .
(47)
233
For the gluongluon channel we find

11399 133
165
9 2 19
+
L
(2)A
=
+
2
3
t (1 x)
gg
144
2
4
20 2
8

1
ln(1 x)
101
351
+ (133 902)
+ 332 +
3
+
(1 x) +
3
2
(1 x) +

ln(1 x)2
ln(1 x)3
33
+ 72
(1 x) +
(1 x) +
+
+
+
+
9(38x 2 20x 3 + 18x 39x 4 + 14 + 7x 5 )

Li3 (x)
1 x2

18(x 2 + x + 1)2
S12 x 2
1+x
9(4x 4 + 8x 3 + 21x 2 + 14x + 7)
S12 (x)
1+x
9 (5x 5 51x 4 57x 3 + 53x 2 + 59x 11)
S12 (x)
2
1 x2
9 (8x 4 + 8x 3 3x 2 2x 1)
Li3 (x)
2
1+x
9 (16 + 13x 5 40x 3 67x 4 + 64x 2 + 36x)
Li2 (x) ln(x)
2
1 x2
9 (2x 4 15x 2 10x 5)
Li2 (x) ln(x)
2
1+x
9 (59 + 177x 2 116x 3 + 59x 4 118x)
ln(x) ln2 (1 x)
4
1x
27(3x 2 + 2x + 1)
Li2 (x) ln(1 + x)
1+x
9(6 11x 3 + 18x 2 12x + 6x 4 ) 2
ln (x) ln(1 x)
1x
9 (3 8x 3 + 3x 4 6x + 9x 2)
Li2 (x) ln(1 x)
2
1x
3 (7x 7x 3 + 4 + 18x 2 17x 4 + 9x 5 ) 3
ln (x)
2
1 x2
9 (8x 4 + 16x 3 + 33x 2 + 22x + 11)
2 ln(1 + x)
2
1+x
36(x 2 + x + 1)2
Li2 (x) ln(1 + x)
1+x
9 (4x 4 + 8x 3 + 27x 2 + 18x + 9)
ln(1 + x) ln2 (x)
4
1+x
234

63
33
+ 21 + x 2 18x + x 3 ln(1 + x) ln(x)
2
2
27 (3x 2 + 2x + 1) 2
ln (1 + x) ln(x)
2
1+x
3 (280x 3 + 143x 4 + 394x 289 + 21x 2)
Li2 (x)
4
(1 x)

33
63
+ 21 + x 2 18x + x 3 Li2 (x)
2
2

2559 3 1079 2 2687
2027
ln(1 x)
+
x +
x
x+
4
2
4
4
+
3 (374x 4 389x + 154 + 699x 2 827x 3) 2

ln (x)
8
1x

+ 330x 3 348x 2 + 381x 297 ln2 (1 x)
3 (1180x 3 + 641 1238x + 1227x 2 + 605x 4)

ln(x) ln(1 x)
4
1x

72 2 x + x 2 x ln3 (1 x)
+
+
+
+
and
(2)F
gg =
1 (4318x 4 6955x 3 + 6447x 2 5611x + 2333)

ln(x)
8
1x
3 (495x 4 886x 3 + 564x 2 200x + 16)
2
4
1x
9(6x + 18x 2 + 2 + 10x 5 6x 3 19x 4)
2 ln(x)
1 x2
9 (48x 3 + 23x 4 46x + 3 + 69x 2)
2 ln(1 x)
2
1x
9 (36 15x 4 52x + 19x 2 + 13x 3 + 33x 5)
3
2
1 x2
18157
7539 3 24107 2 22879
x
x +
x
,
16
48
48
48
(48)

1189 5
5
2
10 ln(1 x)
+ 3 2 + Lt (1 x)
144
6
3
3
3
1x +

1
14
ln(1 x)2
22
+
+2
9
1x +
1x
+

1 65 2
31
31 2 1 17
x + + x S12 (x) + x + x Li3 (x)
+
6
6 12
12
6
6

1
1
47 2 25
1 2 1
x + x
Li2 (x) ln(x) + x + x
2 ln(1 x)
+
12
6
6
12
6
6
235

1
1
1
4x(1 + x)2 ln(x) + x + + x 2 Li2 (x) ln(1 x)
6
6 12

1
1 2
1
1x
x + x ln(1 x) ln(x) ln
+
12 12
24
x

1
5
17
7
3
+ x(1 + x) ln3 x + x 2 x
9
6
3
3

16 2
34
2
8
ln (1 x) 2
+ x3 + x2 x +
9
3
3
9
2 (21x 2 + 7x + 25x 4 + 17 61x 3)
ln(x) ln(1 x)
9
1x

785 3 83 2 49
461
+
x x + x
ln(1 x)
54
36
18
54
1 (351x 3 + 117x 2 + 68 + 132x 4 + 52x) 2

ln (x)
72
1x
1 (227x 3 + 68 + 4x 4 302x + 21x 2 )
Li2 (1 x)
+
36
1x
1 (333x 2 + 2384x 4 598x 3041x 3 + 1282)
ln(x)
+
216
1x
8887 3 1267 2 497
12923
(49)
x +
x
x+
.
648
432
216
1296
For the quarkgluon channel we obtain

338 119 2
170
(2)A
x
+
+
x
=
Li3 (x) + 4x + 4 + 2x 2 Li3 (x)
qg
3
9
3

269 2 74
614
S12 (x)
x
x
+ 16 + 8x 2 + 16x S12 (x) +
9
9
9

367 367 2 367
+ 2x 2 4 4x S12 x 2 +
+
x
x ln3 (1 x)
27
54
27

214
281
446
x+
+
x 2 ln(x)
+ 2 + x 2 2x ln(1 x)
9
9
9

8 + 4x 2 + 8x ln(1 + x) Li2 (x)

1+x
Li2 (x)
+ 8 + 8x + 4x 2 ln
x

230
115 2 230
x +
x
+
ln(x) ln2 (1 x)
9
9
9

107 107 2 107
+
x
x ln2 (x) ln(1 x)
+
9
18
9

145 2 71
x x 2 ln(x)3 + 3x 2 6 6x ln(1 + x) ln(x)2
+
54
27
+
236

+ 4x + 4 + 2x 2 ln(1 + x)2 ln(x)

74
166
4
11
+ x3 x x2
Li2 (x)
27
9
9
27

74
2605 146
79
x + x 3 x 2 Li2 (x)
+
54
9
27
6

37 2
1139
3
x + x + 8x 72 ln2 (1 x)
+
18
12

5935
121 2 326 3 826
x
x
x+
ln(x) ln(1 x)
+
18
27
9
54

113 3 244
13
31
+
x +
x x2
ln2 (x)
27
9
3
2

11
4
74
166
+ x3 x x2
ln(1 + x) ln(x)
27
9
9
27

118
59
118
ln(1 x)
x
+ 2 x 2 +
9
9
9

128 2 52
140
+ 2
x+
x +
ln(x) + 2 12 + 12x + 6x 2 ln(1 + x)
9
9
9

392 3 49 2 23671
x x +
106x ln(1 x)
+
81
3
162

1985 2 800
12209 616 3
+
x +
x
+
x ln(x)
108
9
162
81

16
221
292 3 82
2
x x + x2 +
+
27
3
3
27

92
210115 1537 3
+
x
+ 18x + 10x 2 +
3
9
1944
486
2393 2
16465
x+
x ,
+
162
648
and

1
19
209
1 2 1
38
29
x x+
ln2 (1 x) + x + x 2 +
ln(x)
x
18
9
9
27
27
27
81

265
4 4
2
13
16
+
+
+ x x 2 ln(x) x 2 + x
ln(1 x)
162
9 9
9
9
9

2
1 2 2
179 2
x +
x x+
+
(51)
ln2 (x).
162
9
9
9

(2)F
qg =
(50)
For the scattering of two identical quarks we obtain:

104 2 400
368
32
(2)A
x
+
x
Li3 (x) (x + 2)2 S12 (x)
=
+
qq
27
27
27
9

4
4
2 + x 2 2x ln2 (x) ln(1 x) (x + 2)2 ln3 (x)
27
27

32
16
2
19 + 5x + 17x Li2 (x) ln(x) (x + 3)(1 x) ln2 (1 x)
27
9

16
4
+ (x + 3)(1 x) ln(x) ln(1 x) +
26x 18 + 9x 2 ln2 (x)
3
27

4
8
2
6 + x + 4x Li2 (x) + (5x + 17)(1 x) ln(1 x)
9
3

248
46 2
8
118
2
(x + 2) 2
+
x + x ln(x)
+
9
9
27
9

8 2 16 16
8 2
16 32
+ x
+ x 3 +
x x 2
27
27 27
3
9
3
4
(27x + 160)(1 x),
27
237
(52)
and
(2)F
qq = 0.
(53)
For the scattering of distinct quarks we find

(2)A
qq =

8
32
(x + 2)2 Li3 (x) S12 (x) (x + 2)2 ln(x) Li2 (x)
9
3

8
4
2 3
(x + 2) ln (x) 4x 6 + x 2 Li2 (x)
27
9
32
16
(x + 3)(1 x) ln2 (1 x) + (x + 3)(1 x) ln(x) ln(1 x)
9
3
2
8 2
8
+ x + 4x 3 ln (x) + 2 (x + 2)2 ln(x)
9
9

2
4
+ (5x + 17)(1 x) ln(1 x) + 29x 2 + 44x 59 ln(x)
3
9

8 2
2
16 32
x x 2 (11x + 105)(1 x),
+
3
9
3
9
(54)
and
(2)F
qq = 0.
Finally, for the quarkantiquark channel the NNLO contribution is

8 2
16 16
16 2 32 32
(2)A
x S12 (x)
q q = x x Li3 (x) + x
9
9
9
27
27 27
32
32
4
+ (x + 2)2 Li3 (x) (x + 2)2 S12 (x) (x + 2)2 ln3 (x)
9
9
27

8
4
+ 2 + 2x + x 2 ln(1 + x) ln2 (x) +
2 + 2x + x 2 ln2 (1 + x)
9
27
(55)
238

8
8
(x + 2)2 Li2 (x) + 2 + 2x + x 2 Li2 (x) ln(x)
3
9

16
2 + 2x + x 2 Li2 (x) ln(1 + x)
27

32
+ (1 x) 13x 2 35x 14 ln2 (1 x)
81

16
(1 x) 37x 2 101x 44 ln(x) ln(1 x)
81

8
44x 3 + 39x 81x 2 + 27 ln2 (x)
81

16
+ x x + 6x 2 + 2 ln(1 + x) ln(x)
27

8
42x 87x 2 + 12 + 10x 3 Li2 (x)
+
81

4
16
+ x x + 6x 2 + 2 Li2 (x) (1 x) 384x 2 967x 75 ln(1 x)
27
81

16 2 32 32
x 3
+ x
27
27 27

512 3 10
8
4222 2 2896
2
ln(x)
(x + 2) 2 +
x
x
x
+
9
81
81
27
3

8
752 3 544 2 80 400
2
2 + 2x + x 2 ln(1 + x) +
x
x +
+
x 2
27
81
27
81
27

4
+ (1 x) 783x 2 1925x + 373 ,
(56)
81
and

32
64
64
16 80
(1 x)3 ln(1 x) + x 2 + x 3
+ x ln(x)
81
27
81
27 27

8
(57)
(1 x) 41x 2 88x + 23 .
243
The above results are valid if the renormalization and factorization scales are equal to
the mass of the Higgs boson, r = f = mH . It is easy to restore the complete functional
dependence of the partonic cross-sections on these scales using the fact that the total
hadronic cross-section is independent of them.
We first find the dependence of the partonic cross-sections on a scale which is equal
to both the renormalization and factorization scales r = f = . To do so, we restore the
full dependence in Eq. (15):
(h )

(h )
fi 1 () fj 2 () ij (z, )/z (x).
h1 +h2 H +X = x
(58)
(2)F
q q =
ij
Since the physical cross-section h1 +h2 H +X does not depend on ,

2
d
h +h H +X = 0,
d2 1 2
(59)
239
(h)
and the derivatives of the structure functions fi with respect to can be determined from
the DGLAP evolution equation:

d
s ()
Pij fj (z, ) (x),
fi (x, ) =
d2
we derive the relation

(h ) s ()

d
0=
fi 1
Pik kj (z, )/z + 2 2 ij (z, )/z
d
ij k

s ()
Pkj fj(h2 ) .
+ ik (z, )/z
(60)
(61)
This equation should hold for arbitrary and x; therefore, the expression in the square
brackets should be identically zero for any choice of i and j , yielding the following
evolution equation for the cross-section:
2

d
s ()
Pik kj (z, )/z + ik (z, )/z Pkj . (62)
ij (z, )/z =
d2
We can solve Eq. (62) order by order in s () using the partonic cross-sections at = mH
as the boundary condition. Having obtained the explicit dependence of the partonic crosssections on ,

2

s () (1)
s
ij () +
() ij(2) () + O s3 ,
ij () = 0 ij(0) () +
(63)
we can find their dependence on two independent renormalization and factorization scales
by expressing the strong coupling constant s ( = f ) through s (r ).
We have checked that our expressions for the partonic cross-sections, derived by
explicitly evaluating the Feynman amplitudes with their full scale dependence, are in
agreement with Eq. (62). Our results are also in complete agreement with Ref. [18]
where the first sixteen terms of an expansion of the partonic cross-sections in 1 x were
computed. In the limit, x 0, only the leading logarithmic corrections are known [26].
We can easily reproduce this result by expanding our formulae for partonic cross-sections
around x = 0.
5. Numerical results
We can now discuss the numerical impact of the NNLO corrections on the Higgs
boson production cross-section at the LHC and the Tevatron. To calculate the crosssection we must convolute the hard scattering partonic cross-sections of Section 4 with
the appropriate parton distribution functions. For a self-consistent calculation at NNLO,
we need the parton distribution functions at a given factorization scale at the same order.
At present, the NNLO evolution of the distribution functions cannot be performed since the
required three-loop splitting functions are not known. Nevertheless, a significant number of
moments of the splitting functions is available [27], and this information can be combined
240
Fig. 1. The Higgs boson production cross-section at the LHC at leading (dotted), next-to-leading (dashed-dotted)
and next-to-next-to-leading (solid) order. The two curves for each case correspond to r = f = mH /2 (upper)
and r = f = 2mH (lower).
with the known behavior at small x [28], to obtain a useful approximation for the NNLO
splitting functions [29]. In Ref. [30] this approach was used to determine the NNLO MRST
parton distribution functions; we use these for the numerical evaluation of the Higgs boson
production cross-section.
To demonstrate the convergence properties of the perturbative series for the hadronic
cross-section, we present the LO, NLO and NNLO results for both the LHC and the
Tevatron. In order to improve on the heavy top-quark approximation, we normalize the
cross-section to the exact Born cross-section with the full mt dependence. We use the
mode = 1 parton distribution functions (see Ref. [30] for the notation). For the NNLO
set, this mode provides the average of two extreme cases, the so-called fast and slow
evolutions. For the evaluation of the strong coupling constant we use LO, NLO and NNLO
running accordingly, with the Z-pole values used in the parton distribution functions as
initial conditions (see [30] for details). The total cross-section for the LHC is shown in
Fig. 1. We note that the NNLO cross-section does not vary significantly if we choose a
different mode for the MRST parton distribution functions; the observed changes are less
than 1%.
From Fig. 1 we observe that the scale variation of the Higgs production cross-section
at NNLO in the range r = f = mH /2 2mH is approximately 15%; this is a factor
of two smaller than the NLO scale dependence and a factor of four smaller than the
LO variation. Despite the scale stabilization, the corrections are rather large; the NLO
corrections increase the LO cross-section by about 70%, and the NNLO corrections
increase it further by approximately 30%. The K factor, defined as the ratio of the NNLO
cross section and the LO cross-section at r = f = mH , is approximately two. In Fig. 2
we plot the values of the Higgs production cross-section at the Tevatron. The NNLO K
factor is approximately three, and the residual scale dependence is approximately 23%.
These numerical results for the NNLO Higgs boson production cross-section are in very
good agreement with the corresponding results in [18].
241
Fig. 2. The Higgs boson production cross-section at the Tevatron at leading (dotted), next-to-leading
(dashed-dotted) and next-to-next-to-leading (solid) order. The two curves for each case correspond to
r = f = mH /2 (upper) and r = f = 2mH (lower).
A few remarks concerning the magnitude of the corrections are appropriate. Despite
the fact that the mass of the Higgs boson is much smaller than the total center of mass
energy, the production cross-section is dominated by partonic processes with s m2H .
This is because the gluongluon luminosity is a rapidly decreasing function of the partonic
center of mass energy. The agreement between our numerical results based on the complete
expressions of Section 4 with the approximate results of [18], where an expansion in
1 m2H /s was employed, demonstrates this indirectly.
The dominance of the threshold region renders resummation methods applicable
[31,32]. However, threshold dominance should also affect the cross-section estimates based
on fixed order calculations where there is freedom in the choice of the factorization
scale. Since the production process is dominated by the region x 1, the appropriate
factorization scale should be parametrically smaller than the mass of the Higgs boson;
choosing a factorization scale near the Higgs boson mass may not capture the essential
physics of the process.
We illustrate this point by considering the NLO correction to the Higgs production
cross-section. Concentrating on the gluongluon subprocess, and keeping the most singular
terms in the x 1 limit, we can write

2
s
1
11
(1)
+ 62 (1 x) 6
ln
(x) =
+
.
gg
2
1x
m2H (1 x)2 +
(64)
It is obvious from the above expression that if the dominant contribution to the integrated
cross-section comes from the region x 1, then choosing = mH leaves large logarithmic
corrections of the form log(1 x) in the hard scattering cross-section. To avoid this
problem, we should choose mH (1 x), which is parametrically smaller than the
mass of the Higgs boson. While it is not possible to use an x-dependent factorization
scale without resorting to a full resummation program, in the fixed order calculation we
can attempt to do this on average. This choice decreases the NNLO corrections and the
242
Fig. 3. The Higgs boson production cross-section at the LHC at leading (dotted), next-to-leading (dashed-dotted)
and next-to-next-to-leading (solid) order as the function of factorization and renormalization scale . The mass
of the Higgs boson is 115 GeV for the left and 275 GeV for the right plot.
Higgs boson production cross-section increases as compared to conventional choice of the

scales, r = f = mH .
We demonstrate this behavior with two examples in Fig. 3, where we plot the production
cross-sections for mH = 115 GeV and mH = 275 GeV. We equate the renormalization and
factorization scales and vary the factorization scale from = 15 GeV up to the mass of
the Higgs boson. These plots illustrate that for smaller values of , the NLO cross-section
increases more rapidly than the NNLO cross-section, and the difference between the NLO
and the NNLO results becomes smaller. Therefore, the convergence of the perturbative
series is improved for smaller values of the factorization scale.
If we adopt this argument and restrict our analysis to small we find a Higgs production
cross-section of 55 5 pb for mH = 115 GeV, a somewhat larger value than obtained
with the conventional scale choice = mH . It is interesting that recent studies [32] of
the threshold resumed cross-section for Higgs boson production, matched to the NNLO
calculation, detect a similar increase as compared to fixed order calculations with = mH .
6. Conclusions
In this paper, we have studied the Higgs boson production cross-section in hadronhadron collisions. The main contribution to the hadronic cross-section originates from
gluongluon fusion, which we have computed at NNLO (O(s4 )) in perturbative QCD.
The other partonic production channels, qg H + X, q q H + X, qq H + X and
qq H + X, were also studied to order s4 .
We have presented explicit analytic expressions for the partonic cross-sections valid
within the heavy top-quark approximation. Finally, we have calculated the cross-section
for direct Higgs boson production at the Tevatron and the LHC by performing a numerical
convolution of the partonic cross-sections with the MRST 2002 NNLO set [30] of
parton distribution functions. The residual scale dependence of the NNLO cross-section is
approximately 15% for the LHC and 23% for the Tevatron. The NNLO K-factors are
243
fairly large for both the LHC and the Tevatron. Nevertheless, the cross-section increases
less from NLO to NNLO than from LO to NLO, indicating a slow convergence of the
perturbative expansion.
While this calculation was in progress, Harlander and Kilgore [18] obtained an
approximation of the partonic cross-sections by expanding around the Higgs boson
production threshold. Our results for the partonic cross-sections are consistent with their
expansion. We also find a very good agreement of our numerical results for the NNLO
Higgs production cross section with the results reported in [18].
In Section 5, we have argued that it is more appropriate to choose smaller values of the
factorization scale than the conventional choice = mH . Then, the NNLO corrections
decrease, indicating a better convergence of the perturbative series. Moreover, with this
choice, the fixed order results are in better agreement with recent estimates of the crosssection based on threshold resummation [32].
We have suggested a method for the algorithmic evaluation of inclusive phase-space
integrals. This method combines the Cutkosky rules with integration-by-parts reduction
algorithms to achieve a systematic reduction of phase-space integrals to a few master
integrals. We have also shown how to compute these master integrals using differential
equations produced with the IBP reduction algorithms.
The techniques discussed in this paper can be used to compute higher order corrections
to other inclusive processes of direct phenomenological interest. We are also confident
that this approach can be generalized to enable the calculation of differential distributions.
In fact, a connection between phase-space integrals with a modified measure, such as the
integrals that appear in the evaluation of invariant mass, energy and angular distributions,
and loop integrals with unconventional propagators exists. This connection can be used to
automate the calculation of differential distributions following the lines of Section 3. The
above ideas will be the subject of more detailed studies in future work.
Acknowledgements
We would like to thank Robert Harlander and Bill Kilgore for comparisons with their
results of Ref. [18]. We are grateful to Frank Petriello for careful reading of the manuscript
and invaluable suggestions. This research was supported by the DOE under grant number
DE-AC03-76SF00515.
Appendix A. Splitting functions

In this appendix, some formulae used in our calculation are given.
We first summarize the formulae for the space-like splitting functions.

1
11 1
(1 x)
(0)
Pgg
nf (1 x) + 3 x(1 x) +
1 +3
=
,
4
6
x
1x +
(0)
=
Pgq
2 (1 + (1 x)2 )
,
3
x
244

1 2
x + (1 x)2 ,
4

1
2
3
(0)
(1 x) 1 x + 2
Pqq =
,
3 2
1x +

67 9
1
27
9(x 2 + x + 1)2
(1)
2
Li2 (x)
Pgg
= 6 + 3 (1 x) +
+
4
4
2
1x +
x(1 + x)
(0)
Pqg
=
25 109x
9(3 + 2x 2 + 4x + 2x 3)2 9(x 2 x 1)2 2
+
ln x
2(1 + x)
2(1 x 2 )
8
8

2
2
2
2
9(x x + 1)
9(x + x + 1)
ln(1 + x)
ln(1 x)
+
x(1 + x)
x(1 x)

75 33
2
+ x 33x ln(x)
4
4

5
1+x 2
3 13x
2
1
ln (x)
+
+ nf (1 x)
ln(x)
3
6 1x +
3
2
6

(9x 9x 2 + 109x 3 61)
+
,
36x
+
(1)
=
Pgq
9x + 23x 2 + 9 + 44x 3 4nf (4x 2 + 5 5x) 2(2 + 2x + x 2 ) Li2 (x)
+
9x
27x
x
5(2 2x + x 2 ) log2 (1 x)
9x

2(2 + 2x + x 2 ) log(1 + x) 100 31x 8x 2
+ log(x)
x
9
9
3

2
2
nf
31x + 42 42x
2(2 2x + x )
log(x) +
+
+ log(1 x)
x
9
9x
+ 42 +
(14 + 11x) log2 (x)

.
9
(A.1)
Appendix B. Master integrals

2 =
Below we list the master integrals required in this calculation. We denote s = p12
2
2
(p1 + p2 ) , z = mH /s and for simplicity we set s = 1.
B.1. Double-virtual
The master integrals for the double-virtual corrections can be expressed in terms of
Gamma functions with the exception of the cross-triangle which has been calculated
245
in [33]. For completeness, we list their -expansion below:
2 m2 )
(p12
H
2
2
2
2
2
2
12 ) (l + p1 ) (k l p1 ) (k l)
i i k (k + p12 ) (l + p

2i 1
72
3 19 4
2
= (1 z) (1 + ) Re e
2 27 + O() ,
4

60
dd k dd l
d
2
d
2
2 m2 )
(p12
H
d
d
2
2
)2 (k l)2
i 2 i 2 k (k + p12 ) (l + p12

1 1

5
= (1 z) (1 + )2 Re e2i
+ + 19 + (65 83 )
2
2

4

2
+ 211 403
2 + O 3 ,
15
dd k dd l
2 m2 )
(p12
H
d
d
2
2
2
i 2 i 2 k (l + p12 ) (k l)

1

13
2 115
1
+
= (1 z) (1 + )2 Re e2i

2
4
8
2
16

5971 1152 653 1122 3
53 865 132 2
+
+
+
+
+

+
2
32
4
64
8
4
10

+ O 4 ,
(B.1)
(B.2)
dd k dd l
(B.3)
2
1
=
d
2
2
i 2 k (k + p12 )
1

4
= (1 z) (1 + )2 Re e2i
+ + 12 22 + (32 82 43 )
2

3
422 2
+O ,
+ 80 163 242
(B.4)
5

2
p12
m2H
dd k
246
2

dd k
1

=
d

2
2
k
(k
+
p
)
12
2
i

4
1
= (1 z) (1 + )2 2 + + 12 22 + (32 82 43)

4 2
+ 80 163 242 2 2 + O 3 .
5

2
p12
m2H
(B.5)
B.2. Real-virtual
For the real-virtual contributions we find six master integrals:
(k 2 m2H )((k + p12 )2 )

l 2 (l + p12 )2
i

= PL Re ei s 2 (1 z)12 ,
dd k
d
2
dd k
i
d
2
dd l
dd l
= PLs 2
(k 2 m2H )((k + p12 )2 )

(k l)2 (l + p1 )2
(1 2)2
1 2
(1 z)13 ,
1 3 (1 ) (1 3)
(k 2 m2H )((k + p12 )2 )

(k l)2 l 2
i

= PLs 2 Re ei z (1 z)12 ,
dd k
d
2
(B.6)
(B.7)
dd l
(B.8)
247
(k 2 m2H )((k + p12 )2 )

(k l)2 (l + p1 )2 l 2
i

z2 1
log2 (z)
Li2 (z) +
2 + 5S12 (z)
= P (1 + )
1 2
2
=
dd k
dd l
d
2
4 Li3 (z) + 4 log(z) Li2 (z) +

log3 (z)
+ 22 log(z) 3 + O() ,
2
(k 2 m2H )((k + p12 )2 )

d
(k + p1 )2 (l + p12 )2 (l + p2 )2 l 2 (k l)2
i 2

(1 + )
22
1
63 114 + O 2
= P
(1 z)13 3
3(1 2)

log(z) Li2 (z) log2 (z) 2
+
+ (1 z)12
2
3

S12 (z) 2 Li3 (z) 82 log(z) + 3
+ O() ,
+
3
=
dd k
(B.9)
dd l
(B.10)
(k 2 m2H )((k + p12 )2 )

d
(l + p12 )2 (l + p1 )2 (k l p1 )2 (k l)2
i 2

1
(1 + )
62
2
(1 z)14
16
= 2P
18

+
O(
)
3
4
1 2

3

log(z) Li2 (z) 2
log3 (z)
+ S12 (z) 2 Li3 (z)
+ 3 + O() ,
+
2

3
(B.11)
where the common factors P and L are:
=
dd k
dd l
2
P=

232 32
(B.12)
248
and
L=
() (1 )2
.
(2 2)
(B.13)
B.3. Double-real
We find the following master integrals for the double-real contributions:

dd k dd l (k l)2 m2H (l + p1 + p2 )2 k 2

P 2 (1 2)2
1 2
(1 z)34
(1 4)
2(3 4)(1 4)

1
4
(1 z) z2 5z 2 + z log(z)
+ (1 z)
6

z(z + 2)
5

(1 z) 4z2 17z 5 +
log(z)
36
2

z log(z)2
+
+ 4z Li2 (1 z) + 2 8zS12 (1 z) 16z Li3 (1 z)
2
z log(z)3 56z3 z(2 + z)
691 2
+
+
log(z)2
z
6
27
4
72

56z 2
5z + 8
z log(z)
+ z + 4z + 1 Li2 (1 z) + 6 Li2 (1 z)
+
9
4

281
+
(B.14)
,
+ O 3
216
+

dd k dd l (k l)2 m2H (l + p1 + p2 )2 k 2 l 2

P 2 (1 2)2
(1 z)4 (1 z)3 + (3 z)z2 log(z)
(1 4)

2

1
2 1
(z 3)z2 log(z)2
+ (1 z) 5z 6z + 5 +
2
2

1
5z2 7z + 4 z log(z) + 2(z + 1) z2 4z + 1 Li2 (1 z)
2
249

1
1
+ (1 z) 27z2 26z + 27 + 3 (z 3)z2 log3 (z)
4
6

1
6(z 3)z2 Li2 (1 z) log(z) + z 5z2 7z + 4 log2 (z)
4

1
27z2 25z + 28 z log(z) + (1 + z) 5z2 8z + 5 Li2 (1 z)
4

+ 8(1 + z) z2 4z + 1 Li3 (1 z) + 10z3 + 30z2 6z + 2 S12 (1 z)

1
+ (1 z) 153z2 142z + 153 + O 4 ,
(B.15)
8
((k l)2 m2H )((l + p1 + p2 )2 )(k 2 )

l 2 (k + p1 + p2 )2

log(z)
= P 2 z (1 + z) 2 Li2 (z) 2 +
log(z) 4 log(1 + z)
2

log(z) 2
log (z) 6 log(z) log(1 + z)
+ 8 log(1 + z) Li2 (z) +
6

6 log(1 + z)2 6 log(z) + 24 log(1 + z) + 4 Li3 (z)

+ 4 3 log(z) Li2 (z) + 10S12 (z) 4 z2 + 8S12 (z) + 3 Li3 (z)

+ 2 + 8 log(1 + z) 4 log(z) 2 73 + O 2 ,
(B.16)
dd k dd l
((k l)2 m2H )((l + p1 + p2 )2 )(k 2 )

l 2 (l + p2 )2 (k + p1 )2 (k + p1 + p2 )2

log(z) 1
= P 2 z12 2 + 22 2 z2 4 log(z) log(1 + z) 1

log(z) 7 log(z) log(1 + z) 2 log(1 + z)2 16 log(1 + z) + 4 + 20 Li3 (z)

+ 16 10 log(z) 16 log(1 + z) Li2 (z) + 16 14 log(z)

+ 4 log(1 + z) Li2 (z) + 20S12(z) 8S12 z2 + 14 Li3 (z)

+ 2 6 log(z) + 18 log(1 + z) 8 43 + O() ,
(B.17)
dd k dd l
250
((k l)2 )((l + p1 + p2 )2 )(k 2 m2H )

l 2 (l + p1 )2 (k + p1 )2 (k + p1 + p2 )2

2 log(z) 1
= P 2 (1 + z)12
+ 102 + 8 log(z) log(z)2

2
4
+ 4 Li2 (z) 8 Li2 (z) + log3 (z) + 4 log2 (z) + 343 12 Li3 (z)
3

24S12 (z) + 8 Li2 (z) 8 4 Li2 (z) + 102 log(z) + 8S12 z2

48S12 (z) 402 16 Li2 (z) + 32 Li2 (z) + O() ,
(B.18)
dd k dd l
((k l)2 )((l + p1 + p2 )2 )(k 2 m2H )

l 2 (l + p2 )2 (k + p1 )2 (k + p1 + p2 )2

4
4 42
1
2
14
+ 16(3 2 ) + 4[42 163
= P (1 z)
3+ 2+

2 log(z) 8 log(z) 4 Li2 (z) + 42
+ 94 ] + O 2 + P 2 (1 z)12
+

2

+ 6 Li3 (z) 10S12(z) + 43 + 16 2 log(1 z) 4 log(z) Li2 (z)

+ 2 6 log(z) + 2 log(1 z) 16 + O() ,
(B.19)
dd k dd l
((k l)2 )((l + p1 + p2 )2 )(k 2 m2H )

(l + p2 )2 (l + p1 )2 (k + p1 )2 (k + p2 )2

2
42 10
20
1
4
3 3z
+ (1 z) + 163
= P 2 (1 z)14 3 + 2 +

9

8
224
(1 z) + (5 + z) + 364 643 + O 2
162 +
27
3

2 log(z) 8 log(z) 4 Li2 (z) + 42 23 (1 z)
+
+ P 2 (1 z)12

2
dd k dd l
251
20
log3 (z)
(1 z) + 6 Li3 (z) + 16 Li2 (z) 10S12(z) +
9
3

2
2
log (z) log(1 z) 162 + log(z) 62 log(1 z) 8 4 Li2 (z)

4(1 z)
+ log(1 z)
(B.20)
+ 22 2 Li2 (z) + O() ,
3
+ 43
((k l)2 m2H )((l + p1 + p2 )2 )(k 2 )

(l + p2 )2 (l + p1 )2 (k + p1 )2 (k + p2 )2

4
(1 2)2 P 2
16 z(z 8) 89
(1 z)14 3 + 2 +
=
(1 4)
6

2
2
1
+ (z 1)(1 + 2z) + (z 1)(13z 16) + O
9
27

2
2
(1 2) P
1
12 2 log(z)
+
(1 z)
+ 2 log(z)2
(1 4)

2

1
4
+ log(z) 4 log(1 z) + 8 + (1 z)(z 7) log3 (z)
6
3

1
log2 (z) 8 + 2 log(1 z) + 36 Li3 (z) 3 + z 2z2 + 1
9

log(z) 202 + 16 Li2 (z) + 16 log(1 z) 4 log(1 z)2 8

1
+ (z 1)(z 7) log(1 z) + O() ,
3
dd k dd l
(B.21)
((k l)2 m2H )((l + p1 + p2 )2 )(k 2 )

(k + p1 )2 (k + p2 )2

log2 (z)
2
12
+ 6 Li3 (1 z)
= P z (1 z)
Li2 (1 z) +
2

1
7S12 (1 z) 2 Li2 (1 z) 1 + log(1 z) + 2 log(z) + log3 (z)
6

log2 (z) 1 + log(1 z) + O 2 ,
dd k dd l
(B.22)
252
(k 2 m2H )((l p1 )2 )((k + l + p2 )2 )

(k + p1 )2 (k + p2 )2 (k + l)2 l 2

2
2
(1 2) P
11 + z(z 4)
1
(1 z)14 z12 3 +
=
(1 4)

2 2

82z 21z2 85 211
587
263 53 2
811 121 2
+
z+
+ z +
z
z
6
9
18
6
27
54
18

2
log(z) + 12 (z 1)(z 3) 1 1
+
(z 1)(21z 61)
+O
+
6
2

+ 4 log(z) 4 Li2 (1 z) 4S12 (1 z) 10 Li3 (1 z)
dd k dd l

1
+ 16 10 log(z) Li2 (1 z) 4 log(z) + (1 z)(159z 263) + O() ,
18
(B.23)
(k 2 m2H )((l p2 )2 )((k + l + p1 )2 )

(k + p1 + p2 )2 (k + l)2 l 2

3
142
14
2 2
+ 583 + 1194 + O()
3+
= (1 z)
(1 2) P

log3 (z)
log(z)

+z
2 + 20 Li3 (z) 3 62 log(z) 10 log(z) Li2 (z)

6

+ O() ,
(B.24)
dd k dd l
(k 2 m2H )((k + l + p1 + p2 )2 )(l 2 )

(l + p1 )2 (k + p2 )2 (k + l + p2 )2 (l + p2 )2

log(z) 42 Li2 (z) log2 (z)
= P 2 (1 2)2 z2 2 +

dd k dd l
253
2
+ 12 Li3 (z) 16S12(z) + 43 + 22 log(z) 10 log(z) Li2 (z) log3 (z)
3

+ O() ,
(B.25)
(k 2 )((l p2 )2 )((k + l + p1 )2 m2H )

(k + p1 + p2 )2 (k + l)2 l 2

log(z) 62 + 4 Li2 (z) 4 Li2 (z)
2
14
2
= P (1 + z)
z (1 2) 2 +

2

3 log3 (z)
+ 2 log(z) z + 62 log(z) +
8 Li3 (z) 32S12(z) + 8S12 z2
2

32S12 (z) 8 Li3 (z) + 303 + O() ,
(B.26)
dd k dd l
dd k dd l
= P 2 z2

log(z)2
log3 (z)
43 42 log(z) + 4 Li3 (z) +
+ O() ,
2
2
dd k dd l
2 2
=P z
(k 2 )((l + p1 + p2 )2 )((l k)2 m2H )

(l + p1 + p2 )2 (k + p1 + p2 )2 (l + p1 )2
(B.27)
(k 2 )((l + p1 + p2 )2 m2H )((l k)2 )

(l + p1 + p2 )2 (k + p1 + p2 )2 (l + p1 )2

log3 (z)
+ log(z) Li2 (z) 2 Li3 (z) + O() ,
23 + 2 log(z) +
6
(B.28)
dd k dd l
(k 2 )((l + p1 + p2 )2 )((l k)2 m2H )

(l + p1 )2 (k + p1 )2
254

log(z) 42 4 Li2 (z)
+ 12 Li3 (z) 16S12(z)
= P 2 z (1 2)2 2 +

log3 (z)
+ 43 + O() ,
6 log(z) Li2 (z) 22 log(z)
6
(B.29)
(l 2 )((k + p1 + p2 )2 )((l k)2 m2H )(l + p1 )2

(l + p1 + p2 )2 (k + p1 )2

3

log2 (z)
log (z)
2
(1 z) 1 +
+ Li2 (1 z) log(z) 1 + Li2 (1 z)
=P
6
2

2S12 (1 z) + O() ,
(B.30)
dd k dd l
((k + l + p1 + p2 )2 m2H )(k 2 )(l 2 )

(l + p1 )2 (l + p1 + p2 )2 (k + l + p2 )2 (k + p2 )2

3622
1
(1 2)2 P 2 12
z
+
8
=
(1 2)2
+
32
2
3
(1 4)
5
2 3

2
(1 + 2z) log(z) 1 3z log2 (z) (1 5z)2
4
+ O (1 z)
+
+
+
(1 + z) 2

1+z
1+z

4(1 + 2z)
4z Li2 (z) 2(1 z) Li2 (z)
+
+ log(z)
4 log(1 z)
+
1+z
1+z
1+z

1
2(1 z) log(1 + z)
+
+
(26z 6)S12 (z) + (16z 16)S12(z)
1+z
1+z
2
+ 4(1 z)S12 z (6z + 14) Li3 (z) 4(z + 3) Li3 (z) + (9 21z)3

+ 10(1 z) log(z) 12(1 z) log(1 + z)

2(1 + z) log(1 z) 2

+ (12 4z) log(z) + 8(z 1) log(1 + z) Li2 (z) + 3z log3 (z)

+ 4(1 + 3z) log(z) + 8(1 z) log(1 + z) + 2(1 + z) log(1 z) Li2 (z)

+ 6(1 z) log(1 + z) 3(1 + z) log(1 z) log2 (z)

+ 4(z 1) log2 (1 + z) + 9(1 + z) log2 (1 z)

+ 16(1 + z) log(1 z) log(z) .
(B.31)
dd k dd l
255
References
[1] See, LEP Higgs Working Group for Higgs boson searches, in: Proceedings International Europhysics
Conference on High Energy Physics, Budapest, Hungary, July 2001, hep-ex/0107029, and references therein.
[2] G. Degrassi, hep-ph/0102137;
J. Erler, hep-ph/0102143;
D. Abbaneo, et al., ALEPH, DELPHI, L3, OPAL Collaborations, LEP Electroweak Working Group, and
SLD Heavy Flavor and Electroweak Groups, hep-ex/0112021.
[3] For example, B.A. Kniehl, Phys. Rep. 240 (1994) 211, and references therein.
[4] M. Spira, Fortschr. Phys. 46 (1998) 203.
[5] S. Dawson, Nucl. Phys. B 359 (1991) 283.
[6] A. Djouadi, M. Spira, P.M. Zerwas, Phys. Lett. B 264 (1991) 440;
D. Graudenz, M. Spira, P.M. Zerwas, Phys. Rev. Lett. 70 (1993) 1372;
M. Spira, A. Djouadi, D. Graudenz, P.M. Zerwas, Nucl. Phys. B 453 (1995) 17.
[7] J. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 106 (1976) 292;
M.B. Voloshin, Yad. Fiz. 44 (1986) 738;
M. Shifman, Usp. Fiz. Nauk 157 (1989) 561.
[8] T. Matsuura, S.C. van der Mack, W.L. van Neerven, Nucl. Phys. B 319 (1989) 570.
[9] F.V. Tkachov, Phys. Lett. B 100 (1981) 65;
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
[10] A.V. Kotikov, Phys. Lett. B 254 (1991) 158;
A.V. Kotikov, Phys. Lett. B 259 (1991) 314.
[11] T. Gehrmann, E. Remiddi, Nucl. Phys. B 580 (2000) 485.
[12] S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087.
[13] V.A. Smirnov, Phys. Lett. B 460 (1999) 397;
J.B. Tausk, Phys. Lett. B 469 (1999) 225.
[14] R.E. Cutkosky, J. Math. Phys. 1 (1960) 429.
[15] R.V. Harlander, Phys. Lett. B 492 (2000) 74.
[16] R.V. Harlander, W.B. Kilgore, Phys. Rev. D 64 (2001) 013015.
[17] S. Catani, D. de Florian, M. Grazzini, JHEP 0105 (2001) 025.
[18] R.V. Harlander, W.B. Kilgore, Phys. Rev. Lett. 88 (2002) 201801.
[19] K.G. Chetyrkin, B.A. Kniehl, M. Steinhauser, Phys. Rev. Lett. 79 (1997) 353;
K.G. Chetyrkin, B.A. Kniehl, M. Steinhauser, Nucl. Phys. B 510 (1998) 61.
[20] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[21] G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27;
W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 129 (1977) 66;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 139 (1978) 545, Erratum;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 152 (1979) 493.
[22] P. Nogueira, J. Comput. Phys. 105 (1993) 279.
[23] J.A.M. Vermaseren, math-ph/0010025.
[24] Maple 7.0, Waterloo Maple Inc., see http://www.maplesoft.com.
[25] P.A. Baikov, V.A. Smirnov, Phys. Lett. B 477 (2000) 367.
[26] F. Hautmann, Phys. Lett. B 535 (2002) 159.
[27] S.A. Larin, P. Nogueira, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 492 (1997) 338.
[28] S. Catani, F. Hautmann, Nucl. Phys. B 427 (1994) 475;
V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127;
M. Ciafaloni, G. Camici, Phys. Lett. B 430 (1998) 349;
J. Blumlein, A. Vogt, Phys. Lett. B 370 (1996) 149;
J.A. Gracey, Phys. Lett. B 322 (1994) 141;
J.F. Bennett, J.A. Gracey, Nucl. Phys. B 517 (1998) 241.
[29] W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263;
W.L. van Neerven, A. Vogt, Nucl. Phys. B 588 (2000) 345;
W.L. van Neerven, A. Vogt, Phys. Lett. B 490 (2000) 111.
256
[30] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 531 (2002) 216.
[31] M. Kramer, E. Laenen, M. Spira, Nucl. Phys. B 511 (1998) 523.
[32] S. Catani, D. de Florian, M. Grazzini, P. Nason, in: The QCD/SM Working Group Summary Report, hepph/0204316.
[33] R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542;
G. Kramer, B. Lampe, J. Math. Phys. 28 (1987) 945.

The 11-dimensional metric for AdS/CFT RG flows

with common SU(3) invariance
Changhyun Ahn a , Taichi Itoh b
a Department of Physics, Kyungpook National University, Taegu 702-701, South Korea
b Department of Physics, Hanyang University, Seoul 133-791, South Korea
Received 21 August 2002; received in revised form 17 September 2002; accepted 25 September 2002
Abstract
The compact 7-manifold arising in the compactification of 11-dimensional supergravity is
described by the metric encoded in the vacuum expectation values (vevs) in d = 4, N = 8 gauged
supergravity. Especially, the space of SU(3)-singlet vevs contains various critical points and RG
flows (domain walls) developing along AdS4 radial coordinate. Based on the nonlinear metric ansatz
of de WitNicolaiWarner, we show the geometric construction of the compact 7-manifold metric
and find the local frames (siebenbeins) by decoding the SU(3)-singlet vevs into squashing and
stretching parameters of the 7-manifold. Then the 11-dimensional metric for the whole SU(3)invariant sector is obtained as a warped product of an asymptotically AdS4 space with a squashed
and stretched 7-sphere. We also discuss the difference in the 7-manifold between two sectors, namely
(SU(3) U (1))-invariant sector and G2 -invariant sector. In spite of the difference in base 6-sphere,
both sectors share the 4-sphere of CP2 associated with the common SU(3)-invariance of various
7-manifolds.
PACS: 11.25-w; 11.25.Hf; 11.25.Mj; 11.30.Pb
1. Introduction
One of the crucial points of de WitNicolai theory [1] is the presence of warp
factor (x, y). When one reduces 11-dimensional supergravity theory to four-dimensional
N = 8 gauged supergravity, the four-dimensional spacetime is warped by this factor
which depends on both four-dimensional coordinate x and 7-dimensional internal
E-mail addresses: ahn@knu.ac.kr (C. Ahn), taichi@hepth.hanyang.ac.kr (T. Itoh).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 7 1 - 4
258
C. Ahn, T. Itoh / Nuclear Physics B 646 (2002) 257280
coordinate y m . This warp factor provides an understanding of the different relative scales
of the 11-dimensional solutions corresponding to the critical points in N = 8 gauged
supergravity. One writes down 11-dimensional metric as warped product ansatz
2
ds11
= ds42 + ds72 =
1
g (x) dx dx + Gmn (x, y) dy m dy n ,
(x, y)
(1.1)
where , = 1, 2, . . . , 4 and m, n = 1, 2, . . . , 7. The nonlinear metric ansatz in [2]

provides the explicit formula (5.1) for the 7-dimensional inverse metric Gmn (x, y),
which is encoded by the warp factor (x, y), 28 Killing vectors K mI J (y) on the round
7-sphere (expressed through the Killing spinors) and 28-beins uij I J (x), vij I J (x) in fourdimensional gauged N = 8 supergravity: the dependence of x appears in the first and last
one.
An important aspect of the holographic duals [35] is the notion that the radial
coordinate of AdS4 can be viewed as a measure of energy. A supergravity kink description
interpolating between r and r can be interpreted as an explicit construction
of the renormalization group (RG) flow between the ultraviolet (UV) fixed point and the
infrared (IR) fixed point of the three-dimensional boundary field theory. To construct the
superkink corresponding to the supergravity description of the nonconformal RG flow,
the most general four-dimensional bulk metric which has a three-dimensional Poincare
invariance takes the form

g (x) dx dx = e2A(r) dx dx + dr 2 ,
(1.2)
where = (, +, +), r = x 4 is the coordinate transverse to the domain wall and the
scale factor A(r) behaves linearly in r at UV and IR regions.
The 70 real scalars of N = 8 supergravity parametrize [6] the coset space E7(7)/SU(8).
The scalar potential is a function of these 70 scalars and this number is too large to be
managed practically. From the possible embeddings of SU(3), the 70 scalars contain 6
singlets of SU(3) (three from 35 scalars 35v plus three from 35 pseudo-scalars 35c ). But
the scalar potential depends on only four real fields due to the SO(8)-invariance of the
potential and a larger invariance of the SU(3)-invariant sector. The explicit construction
of 28-beins uij I J (x) and vij I J (x) in terms of these fields has been found in [7]. It is
known [8] that there exist five nontrivial critical points for the scalar potential of gauged
N = 8 supergravity: SO(7)+ , SO(7) , G2 , SU(4) and SU(3) U (1). Among them G2 invariant 7-ellipsoid and (SU(3) U (1))-invariant stretched 7-ellipsoid are stable and
supersymmetric.
By writing de WitNicolai potential in terms of a superpotential encoded in the T -tensor
of a theory, the holographic RG flow equations from N = 8, SO(8)-invariant UV fixed
point to N = 2, (SU(3)U (1))-invariant IR fixed point were constructed in [9]. Moreover,
the holographic RG flow equations from N = 8, SO(8)-invariant UV fixed point to
N = 1, G2 -invariant IR fixed point were obtained in [7,10] (see also [11]). Contrary to
(SU(3) U (1))-invariant case, there exists an algebraic relation in a complex eigenvalue
of A1 tensor for the G2 -invariant case. This observation was crucial for the correct
minimization of the energy-functional.
The M-theory lift of a supersymmetric RG flow is achieved as follows. First we
impose the nontrivial r-dependence of vacuum expectation values (vevs) subject to the
259
four-dimensional RG flow equations. Then the geometric parameters in the 7-manifold

metric at certain critical point are controlled by the RG flow equations so that they
can be smoothly extrapolated from the critical point. Secondly we make an appropriate
ansatz for the 11-dimensional three-form gauge field. If the ansatz is correct, the 11dimensional EinsteinMaxwell (bosonic) equations can be finally solved by using the RG
flow equations to complete the M-theory lift. Based on this prescription, an exact solution
to the 11-dimensional bosonic equations corresponding to the M-theory lift of the N = 2,
(SU(3) U (1))-invariant RG flow was found in [12]. Its Khler structure was extensively
studied in [13]. Similarly, the M-theory lift of the N = 1, G2 -invariant RG flow was done
in [10].
In AdS/CFT context [35], the above two membrane flows are holographic dual of
flows of the maximally supersymmetric N = 8 scalar-fermion theory in three dimensions.
It is still unclear how they are related both in the bulk supergravity and in the boundary
field theory. In order to answer this question, we would like to achieve the M-theory lift
of whole SU(3)-invariant sector, including the above five critical points and RG flows in
d = 4, N = 8 gauged supergravity. As the first step toward this goal, we need to know the
complete metric (1.1) together with (1.2). To solve the 11-dimensional bosonic equations
by utilizing the RG flow equations in [7], we have to further make an appropriate ansatz
for the 11-dimensional three-form gauge field. It will be a natural extension of the Freund
Rubin parametrization [14] but will be more complicated due to its nontrivial AdS4 radial
coordinate dependence. Even though the nonlinear metric ansatz in [2] provides the 11dimensional four-form gauge field strengths also,1 they are encoded in SU(3)-singlet vevs
in a complicated way. In this paper, we concentrate on the 11-dimensional metric for the
SU(3)-invariant sector and postpone the field strength ansatz as well as solving the 11dimensional bosonic equations to future work.
We begin our analysis in Section 2 by summarizing relevant aspects of the 11dimensional metric for N = 2, (SU(3) U (1))-invariant membrane flow found in [12].
In Section 3, we review the 11-dimensional metric for N = 1, G2 -invariant membrane
flow found in [10]. In Section 4, we summarize the four-dimensional RG flow equations
[7], namely the first-order BPS equations for the SU(3)-invariant sector in d = 4, N = 8
gauged supergravity. In Section 5, that is the main content of this paper, we show
the geometric construction of the compact 7-manifold metric and find the local frames
(siebenbeins) by decoding the SU(3)-singlet vevs into squashing and stretching parameters
of the 7-manifold. We mainly use Hopf fibration on CP3 as global 7-dimensional
coordinates. In Section 6, we summarize our results and will discuss about future direction.
In Appendix A, we list all the components of the 7-dimensional inverse metric generated
from the metric formula (5.1) by using R8 embedding. In Appendix B, some preliminaries
of Hopf fibration are presented. In Appendix C, we show the other set of global 7dimensional coordinates useful to describe the G2 -invariant critical point.
1 They are given by Eq. (7.6) in [15].
260
2. The 11-dimensional metric for N = 2, (SU(3) U (1))-invariant membrane flow

The 28-beins (u, v) or two vevs (, ) 2 are given by functions of the AdS4 radial
coordinate r = x 4 . The metric formula (5.1) generates the 7-dimensional metric from the
two input data of AdS4 vevs (, ). The (SU(3) U (1))-invariant RG flow subject to
the first order differential equations on two vevs [9] is a trajectory in (, )-plane and is
parametrized by the AdS4 radial coordinate.
Let us introduce the standard metric corresponding to an ellipsoidally squashed 7-sphere
with a stretched Hopf fiber. The stretching factor is characterized by the vev and
= 0 corresponds to the round 7-sphere. Then the isometry SO(8) of 7-sphere breaks
into U (4). The metric is also ellipsoidally squashed and it reduces the broken isometry to
SU(3) U (1) U (1). Let us introduce a diagonal 8 8 matrix

QAB = diag (r)2 , . . . , (r)2 , (r)6 , (r)6 .
Recall that A1 tensor of this theory has two distinct eigenvalues with degeneracies 6, 2.
This behavior reflects also in the deformation matrix here.
on the deformed R8
The metric
8
A
A
2
(Cartesian coordinates on R are denoted by X with A=1 (X ) = 1) can be written as
(see also [13])

2
sinh2 (r) A
B
X JAB dXB ,
dX
+
ds 2 (r), (r) = dXA Q(r)1
AB
2
(r, )
(2.1)
where the quadratic form (r, )2 XA QAB XB is now given by using the parametrization
of [12]
(r, )2 = (r)2 cos2 + (r)6 sin2 ,
where is one of the 7-dimensional internal coordinates and (r, )2 becomes 1 when
= 1. For = 1, = 0, it provides the trivial vacuum of SO(8) maximal supersymmetric
critical point. The antisymmetric Khler form JAB has nonzero elements J12 = J34 =
J56 = J78 = 1.
Applying the Killing vector together with the 28-beins (u, v) to the metric formula (5.1),
we obtain a raw inverse metric (x, y)1 Gmn (x, y) including the warp factor (x, y)
not yet determined. Substitution of this raw inverse metric into the definition of warp factor

(x, y)1 det Gmn (x, y) g np (y) ,

(2.2)
where g np (y) is a metric of the round 7-sphere, will provide a self-consistent equation for
(x, y). For the (SU(3) U (1))-invariant RG flow, solving this equation gives rise to the
warp factor

4/3
.
(r, ) = (r, ) cosh (r)
(2.3)
2 The variables and in [9] with = 0, = /2 are related to = 4 2 ln , =

and d we will define later, = a 1/4 = b1/4 and = cosh1 c = cosh1 d.
2 . In terms of a, b, c
261
Then we substitute this warp factor into the raw inverse metric and obtain the 7dimensional metric:

ds72 = Gmn (x, y) dy m dy n = (r, ) L2 ds 2 (r), (r)
(2.4)
together with (2.1) and(2.3) where L is a radius of round 7-sphere. The metric (2.1) now
is warped by a factor (r, ). The nonlinear metric ansatz combines the 7-dimensional
metric with the four-dimensional metric with warp factor to yield the 11-dimensional
warped metric characterized by (1.1), (1.2), (2.4), (2.3) and (2.1).
3. The 11-dimensional metric for N = 1, G2 -invariant membrane flow
The two vevs (, ) 3 are given by functions of the AdS4 radial coordinate r. The metric
formula (5.1) generates the 7-dimensional metric from the two input data of AdS4 vevs
(, ). The G2 -invariant RG flow subject to the first order differential equations on these
vevs [7,10] is a trajectory in (, )-plane and is parametrized by the AdS4 radial coordinate.
We will use (a, b) defined by

(r)
(r)
a(r) cosh
+ cos (r) sinh
,
2
2

(r)
(r)
b(r) cosh
(3.1)
cos (r) sinh
.
2
2
Let us introduce the standard metric of a 7-dimensional ellipsoid. Using the diagonal
8 8 matrix QAB given by4

QAB = diag b(r)2 , . . . , b(r)2, a(r)2 ,
(3.2)
the metric of a 7-dimensional ellipsoid can be written as

2
B
a(r), b(r) = dXA Q(r)1
dsEL(7)
AB dX .
(3.3)
This can be rewritten in terms of the 7-dimensional coordinates y m (see [10] for the explicit
relations between R8 coordinates XA and y m ) such that

2
dsEL(7)
(3.4)
a(r), b(r) = b(r)2 a(r)2 (r, )2 d 2 + sin2 d62 ,
where = y 7 must be identified with the fifth coordinate in 11 dimensions and the
quadratic form (r, )2 is given by
(r, )2 = a(r)2 cos2 + b(r)2 sin2
(3.5)
which becomes 1 for a = b = 1 corresponding to SO(8) trivial vacuum.

3 One can obtain these vevs from SU(3)-invariant sector by restricting four arbitrary fields , , , to =
and = . This is equivalent to c = a and d = b where a, b, c and d are defined as the one in Section 5.
4 Note that the diagonal matrix here is different from the one in [10] and is defined as the old one multiplied
by b(r)2 .
262
Note that the geometric parameters (a, b) for the 7-ellipsoid can be identified with the
two vevs (a, b) defined in (3.1). This was the reason why we prefer the vevs (a, b) rather
than (, ). Applying the Killing vector together with the 28-beins to the metric formula
(5.1) like we did before, we obtain a raw inverse metric (x, y)1 Gmn (x, y) including
the warp factor (x, y). For the G2 -invariant RG flow, solving the condition yields the
warp factor
(r, ) = a(r)1 (r, )4/3 .
(3.6)
Then we substitute this warp factor into the raw inverse metric and obtain the 7dimensional metric as follows:

2
ds72 = Gmn (x, y) dy m dy n = (r, )a(r) b(r)2 L2 dsEL(7)
(3.7)
a(r), b(r) .
The 7-dimensional metric (3.3) is warped by a factor (r, ) a(r). The nonlinear metric
ansatz with the warp factor yields the warped 11-dimensional metric described by (1.1),
(1.2), (3.7), (3.3) and (3.6).
4. Holographic RG flow for SU(3)-invariant sector

It is known [8] that SU(3)-singlet space with a breaking of the SO(8) gauge group into
a group which contains SU(3) may be written as four real parameters (r), (r), (r)
and (r). The vacuum expectation value of 56-bein for the SU(3)-singlet space, that is an
invariant subspace under a particular SU(3) subgroup of SO(8), can be parametrized by
+
ij kl = (r) cos (r)Yij1 kl

+ (r) sin (r)Yij1 kl
+
+ (r) cos (r)Yij2 kl

+ (r) sin (r)Yij2 kl
,
where the scalar and pseudo-scalar singlets of SU(3) are given by

1256
3456

5678
3478
1278
,
= ij1234
Yij1 kl
kl ij kl + ij kl ij kl + ij kl ij kl

1467

2468
2457
1368
2367
1458
2358
.
Yij2 kl
= ij1357
kl ij kl + ij kl ij kl + ij kl ij kl + ij kl ij kl
Here + = 1 and = i and + gives the scalars and gives the pseudo-scalars of
N = 8 supergravity. The four scalars (r), (r), (r) and (r) in the SU(3)-singlet vevs
parametrize an SU(3)-invariant subspace of the complete scalar manifold E7(7)/SU(8).
The 56-bein V(x) preserving the SU(3)-singlet space is a 56 56 matrix whose elements
are some functions of four fields (r), (r), (r) and (r) obtained by exponentiating the
above vacuum expectation value ij kl . Then the 28-beins, u and v in terms of these fields,
can be obtained as the 28 28 matrices given in Appendix A of [7].
It turned out [7] that A1 tensor has three distinct complex eigenvalues, z1 , z2 and z3
with degeneracies 6, 1, and 1 respectively and has the following form
A1 I J = diag(z1 , z1 , z1 , z1 , z1 , z1 , z2 , z3 ),
263
where the eigenvalues are some functions of (r), (r), (r) and (r). In particular,

z3 (, , , ) = 6ei(+2) p2 qr 2 t 2 + 6e2i(+) pq 2 r 2 t 2 + p3 r 4 + e4i t 4

+ e3i q 3 r 4 + e4i t 4 ,
and we denote hyperbolic functions of (r) and (r) by the following quantities for
simplicity

(r)
(r)
q sinh ,
p cosh ,
2 2
2 2

(r)
(r)
r cosh ,
t sinh .
2 2
2 2
We refer to [7] for explicit expressions of z1 and z2 .
The superpotential is one of the eigenvalues of A1 tensor and the supergravity potential5
can be written in terms of superpotential as follows
W (, , , ) = |z3 |,
16
2
( W )2 + 2 2 ( W )2 + 4( W )2
V (, , , ) = g 2
3
3p q

1
+ 2 2 ( W )2 6W 2 .
2r t
(4.1)
The flow equations [7] we are interested in are
8 2
g W (, , , ),
r (r) =
3
r (r) = 2 2g W (, , , ),
2
r (r) = 2 2 g W (, , , ),
3p q
2
r (r) = 2 2 g W (, , , ),
4r t
r A(r) = 2 gW (, , , ).
(4.2)
There exist two supersymmetric critical points of both a scalar potential and a superpotential: N = 1 supersymmetric critical point with G2 -symmetry and N = 2 supersymmetric
5 The scalar potential can be written, by combining all the components of A , A tensors, as
1 2
V (, , , ) =

1 2 4 2
x + 3 c3 + 4x 2 v 3 s 3 3v x 2 1 s 3 + 12xv 2 cs 2
g s
2

6(x 1)cs 2 + 6(x + 1)c2 sv

+ 2s 2 2 c3 + v 3 s 3 + 3(x + 1)vs 3 + 6xv 2 cs 2 3(x 1)cs 2 6c 12c ,

where we introduce the following quantities: c cosh , s sinh , c cosh
, s sinh
,
v cos , and x cos 2.
264
one with (SU(3)U (1))-symmetry. Also there are three nonsupersymmetric critical points
with SO(7)+ -, SO(7) - and SU(4) -symmetries.
5. The 11-dimensional metric for AdS4 RG flows with common SU(3) invariance
In this section, we will give an ansatz for the generic SU(3)-invariant metric in terms
of squashing deformation and the Khler form J by using the R8 vector X = R x. Then
we invert the metric and change the R8 basis from X to x to compare the ansatz with the
inverse metric generated by the de WitNicolaiWarner formula [2,15].
5.1. The compact 7-manifold metric encoded in data of d = 4, N = 8 gauged
supergravity
The consistency under the KaluzaKlein compactification of 11-dimensional supergravity, or M-theory, requires that the 11-dimensional metric for RG flows with common SU(3)
invariance is not simply a metric of product space but Eq. (1.1) for the warped product of
an asymptotically AdS4 space, or a domain wall, with a compact 7-dimensional manifold.
Moreover, the 7-dimensional space becomes a warped, squashed and stretched S7 and its
metric is uniquely determined through the nonlinear metric ansatz developed in [2,15]. The
warped 7-dimensional inverse metric is given by

1
Gmn (x, y) = (x, y) K mI J K nKL + (m n)
2

uij I J (x) + vij I J (x) u ij KL (x) + v ij KL (x) ,
(5.1)
where K mI J denotes the Killing vector on the round S7 with 7-dimensional coordinate
indices m, n = 5, . . . , 11 as well as SO(8) vector indices I, J = 1, . . . , 8. The uij I J and
vij I J are 28-beins in 4-dimensional gauged supergravity and are parametrized by the
AdS4 vacuum expectation values (vevs), , , and , associated with the spontaneous
compactification of 11-dimensional supergravity.
The 28-beins (u, v) or four vevs (, , , ) are given by functions of the AdS4 radial
coordinate r = x 4 . The metric formula (5.1) generates the 7-dimensional metric from the
four input data of AdS4 vevs (, , , ). The RG flow subject to Eq. (4.2) is a trajectory
in the SU(3)-singlet space spanned by (, , , ) and is parametrized by the AdS4 radial
coordinate. Hereafter, instead of (, , , ), we will use (a, b, c, d) defined by

(r)
(r)
a(r) cosh
+ cos (r) sinh
,
2
2

(r)
(r)
cos (r) sinh
,
b(r) cosh
2
2

(r)
(r)
c(r) cosh
+ cos (r) sinh
,
2
2

(r)
(r)
cos (r) sinh
.
d(r) cosh
2
2
265
The inverse metric for the 7-manifold generated by the de WitNicolaiWarner

(dWNW) formula (5.1) is encoded in the data of 28-beins, namely a, b, c, d given above.
Therefore, to get the metric available for practical purposes, we have to decode the 28beins into the deformation parameters of the compact 7-manifold. This can be done by
comparing the encoded inverse metric with the inverse metric given by some appropriate
ansatz. Since we do not know what are the proper coordinates describing the compact
7-manifold yet, we will consider the R8 embedding by using the Cartesian coordinates on
R8 , namely x A , A = 1, . . . , 8. Now the Killing vectors defined on the round S7 with radius
L are given by

K A I J = L I J BC x B A x C x C A x B = L x B I J BA x C I J AC ,

where the R8 coordinates x A s are constrained on the unit round S7 , 8A=1 (x A )2 = 1,
and I J are the SO(8) generators given in [10,16]. Substituting this into the formula (5.1)
generates the inverse metric divided by the warp factor, namely 1 GAB , described by the
R8 coordinates x A s. We list all of its components in Appendix A. Our goal is to reproduce
the same inverse metric via purely geometric construction and finally to determine the
metric and the warp factor separately.
The complication coming from using the R8 coordinates is that the Khler form J
defined by J 2 = I is not standard form in the coordinates x A s. By taking the SU(4) invariant limit a = b = 1, d = c [17] in the raw inverse metric in Appendix A, we obtain

1 GAB = c2 AB x A x B + 1 c2 JAC x C JBD x D
from which one can read the Khler form J as an 8 8 matrix given by Eq. (A.1) in
Appendix A. Therefore to get the standard Khler form J given by
J12 = J34 = J56 = J78 = 1,
(5.2)
we must transform the

vector x to another one X = R x by using the 8 8 orthogonal
matrix R given by Eq. (A.2) in Appendix A.
R8
5.2. Geometric construction of the compact 7-manifold metric

Geometric implication of N = 2, (SU(3) U (1))-invariant metric is as follows. First,
turning off the ellipsoidal deformation, the metric of compact 7-manifold is given by
Hopf fibration on CP3 with a stretched Hopf fiber. FubiniStudy metric on CP3 has
SU(4) invariance as can be seen from CP3 SU(4)/(SU(3) U (1)). Then turning on
the ellipsoidal deformation breaks the SU(4) invariance of CP3 down to its SU(3) U (1)
subgroup preserving the SU(3) invariance of FubiniStudy metric on CP2 CP3 as well
as the U (1) symmetry along the stretched Hopf fiber. Therefore to get the SU(3)-invariant
metric, one can further break the U (1) symmetry as long as the FubiniStudy metric on
CP2 is preserved. Since the S5 given by Hopf fibration on CP2 is embedded in R6 spanned
by X1 , . . . , X6 , the first six diagonal components of the deformation matrix Q must be the
same. This is the same as in (SU(3) U (1))-invariant case. Recall that the S1 of U (1)
Hopf fiber on CP3 is embedded in R2 spanned by X7 , X8 . Now one can break this U (1)
symmetry by choosing different scaling factors in the last two diagonal components of Q.
266
One can also decompose the Hopf fiber (X, J dX) XA JAB dXB into SU(3)-invariant
pieces.
Thus we are led to the ansatz for the SU(3)-invariant unwarped metric:

2

ds02 = dX, Q1 dX + 2 U, J dU + 1 (V1 , J dV2 ) + 2 (V2 , J dV1 ) , (5.3)
where the deformation matrix Q is given by

Q = diag(, . . . , , 1 , 2 ),
norm on the 7-sphere. The R8 coordinates
so that 2 (X, QX) can be the SU(3)-invariant
8
7
A
2
X are restricted on the round S , A=1 (X ) = 1, and subject to the SU(3)-invariant
decomposition X = U + V1 + V2 with

V1 = 0, . . . , 0, X7 , 0 ,
V2 = 0, . . . , 0, 0, X8 .
U = X1 , . . . , X6 , 0, 0 ,
By introducing two more parameters 1 , 2 , the Hopf fiber (X, J dX) is decomposed into
the SU(3)-invariant pieces (U, J dU ), (V1 , J dV2 ), (V2 , J dV1 ). Now the U (1) symmetry
of (X, J dX) is fully broken unless 1 = 2 = 1.
From the ansatz (5.3), one can read the matrix g as follows.
g = Q1
XXT + 2 F,
2
(5.4)
where we have defined F as

F = (J U + 1 J V1 + 2 J V2 ) (J U + 1 J V1 + 2 J V2 )T .
(5.5)
We notice that the first two terms in g combine into the projection operator for the direction
transverse to QX. This must be the case for the whole of g since the metric g describes
the embedding of the 7-sphere into R8 by restricting the R8 into the subspace transverse to
QX. Thus we require that F must project out the deformed vector QX, namely F QX = 0
which restricts 1 , 2 to be 2 1 = 1 2 and makes (F Q)2 become proportional to F Q.
We further require that
(F Q)2 = 2 F Q,
(5.6)
which can be achieved by choosing

1
1
=
.
1 =
2
2
The deformation parameters , 1 , 2 , must be determined by comparing the inverse
of g with the inverse metric generated by dWNW formula. However, the 8 8 matrix g is a
projection operator describing the R8 embedding of the compact 7-manifold and is rank 7.
It does not have its inverse in ordinary sense. Therefore we have to define the inverse of g,
say g 1 , as an 8 8 matrix satisfying
g 1 gg 1 = g 1
gg 1 g = g.
(5.7)
By looking at the SU(3)-invariant norm

= (X, QX), we realize that QX is a vector
dual to X. This implies that g 1 must be a projection operator transverse to X as g is so
2
for QX. Now we suppose that g 1 is given by

1
1
1
T
QF Q,
g = Q 2 (QX)(QX)
+ 1 2
267
(5.8)
which in fact projects out X by using F QX = 0. One can see by using Eq. (5.6) that g,
g 1 in Eqs. (5.4), (5.8) indeed satisfy the inversion relation (5.7).
Thus the inverse metric g 1 in Eq. (5.8) can be compared with the warped inverse metric
1
G obtained by dWNW formula. They are related via
1 G1 = 2 NR T g 1 R,
(5.9)
where 2 in the right-hand side is necessary to get the left-hand side just as a polynomial
of the R8 coordinates x A s as in Appendix A. The normalization factor N , as well as the
deformation parameters , 1 , 2 and , will be determined as a function of a, b, c, d.
We also have to replace the R8 vector X with R x in the right-hand side. Substituting the
results in Appendix A into the left-hand side, one can get
1 =
ac
,
bd
2 =
ad
,
bc
N=
bcd
,
2
cd
1,
ab
where can be fixed by writing it like as = a m1 bm2 cm3 d m4 and using the two limit
values: (SU(3) U (1))-invariant case and G2 -invariant case. For the former, we have
b = 1/a = 4 and d = c = cosh . In this case, was given as 2 . Therefore we should
have m3 = m4 and m1 m2 = 1/2. For the latter, one has c = a and d = b and was
given as b2 . Therefore m1 + m3 = 0 and m2 + m4 = 2. The exponents of m1 , m2 , m3 and
m4 are uniquely determined to yield
a 7/4 b1/4c1/4
a 3/4b5/4 d 3/4
,
1 =
,
3/4
c
d 1/4
a 7/4 b1/4d 7/4
c5/2
2 =
(5.10)
,
N
=
L2 .
c7/4
a 3/2b3/2d 1/2
Here a comment on G2 -invariant limit must be in order. In the limit, we have c = a
and d = b so that Eq. (5.10) provides = 2 = b2 and 1 = a 2 . Hence we get Q =
diag(b2 , . . . , b2 , a 2 , b2 ) which is different from Q in Section 3 (see Eq. (3.2)) by the
interchange between the seventh and the eighth components. To fix this mismatch, we
only have to interchange X7 with X8 by an orthogonal transformation. However, if we
do so the Khler form J is also transformed and is no longer the standard form given in
Eq. (5.2). Nevertheless, this does not cause any problem in G2 -invariant case as well as in
SO(7) -invariant cases since the 7-dimensional metric has no contribution from the Khler
form J .
=
5.3. The local frames for the compact 7-manifold

Now that we have decoded the encoded output of dWNW formula given in Appendix A
by determining all the geometric parameters in the SU(3)-invariant ansatz (5.3). They
are given by Eq. (5.10). However, the dWNW formula generates the involved inverse
metric 1 G1 without giving the warp factor separately. Therefore to get the full
268
7-dimensional metric, we have to separate out the warp factor from the obtained results.
Recall that the inverse metric g 1 is related to 1 G1 via Eq. (5.9). Inverting Eq. (5.9)
provides the warped 7-dimensional metric G involving the warp factor. Since g 1 is the
inverse of g in Eq. (5.4), G is given in the R8 basis X by

1
1
L2 a 3/2b3/2d 1/2
1
T
G=
(5.11)
Q 2 XX + 2 F .
g 2
2 N

c5/2
To determine the warp factor , the easiest way is to find out the 7-dimensional local
frames (or siebenbeins) E i (i = 1, . . . , 7) defined as
ds72 (dX, G dX) = L2
7

Ei Ei .
i=1
By using the wedge product of E i s, the defining equation (2.2) of the warp factor can
be written as
(7 )1
7

Ei ,
(5.12)
i=1
where 7 is the volume element of the unit round S7 . For convenience, let us introduce the
unwarped local frames e i (i = 1, . . . , 7) defined by
ds02 (dX, g dX) =
7

ei ei ,
(5.13)
i=1
factorizing the dependence of E i s. Eq. (5.12) then turns out to be the self-consistent
equation for :
3/4 3/4 1/4 7

7

1
a b d
1
=
(5.14)
(
)
ei ,
7
1/2
c5/4
i=1
e i s
where the wedge product of

is calculable without knowing the warp factor. Hence in
order to determine the warp factor , we only have to find out the unwarped frames e i s
and to calculate the wedge product of them.
In terms of deformation parameters and the Khler form J , the line element ds02 in
Eq. (5.13) is written explicitly as (see Eq. (5.3))
ds02 =
1
1
1
(dU )2 +
(dV1 )2 +
(dV2 )2
1
2
2

1
2
+ 2 (U, J dU ) +
(V1 , J dV2 ) +
(V2 , J dV1 ) ,
2
1
(5.15)
with = cd/ab 1. We will use the same 7-dimensional coordinatization of U , V1 , V2 as

in [12], that is
U = u cos ,
V1 = (0, . . . , 0, cos sin , 0),
V2 = (0, . . . , 0, 0, sin sin ),
269
where u = (u1 , . . . , u6 , 0, 0) subject to the constraint (u, u) = 1 describing the unit round
S5 . The deformed norm 2 (X, QX) then becomes

2 = cos2 + 1 cos2 + 2 sin2 sin2
(5.16)
which becomes 1 in the SO(8)-invariant limit = 1 = 2 = 1 to ensure the correct
normalization of the unwarped metric g. The vector u spans the S5 given by Hopf fibration
on CP2 base. This can be understood by rewriting (du)2 as
2
+ (u, J du)2 ,
(du)2 = (du)2 (u, J du)2 + (u, J du)2 dsFS(2)
2
where dsFS(2)
(du)2 (u, J du)2 denotes the FubiniStudy metric on CP2
= S4 and
(u, J du) is the Hopf fiber on it. As mentioned before, the SU(3)-invariant deformation
must preserve at least the FubiniStudy metric on CP2 . The U (1) symmetry associated
with the Hopf fiber (u, J du) is a maximal circle of S5 CP3 and is always preserved.
The U (1) symmetry preserved in the (SU(3) U (1))-invariant sector but broken in other
cases is another U (1) symmetry related to (X, J dX), namely the Hopf fiber on CP3 .
In SU(3) U (1) limit, one of the local frames must be fixed to the direction of the
Hopf fiber (X, J dX) [12]. Therefore it seems plausible that one of the local frames for the
SU(3)-invariant 7-manifold, say e7 , is given by

1
2
7
1
+ 1 (U, J dU ) +
(V1 , J dV2 ) +
(V2 , J dV1 )
e =
(5.17)
2
1
which in fact turns to be 1 cosh (X, J dX) in the SU(3) U (1) limit 1 = 2 ,
= sinh2 . Then one can rewrite the metric (5.15) such that
ds02
3

1
1
2
2
2
= cos dsFS(2) + 2 cos
Mij i j + e7 e7 ,
i,j =1
where
= d,
= d,
= (u, J du) and Mij s are components of the mixing matrix
M given by
f3
f0 f1
f1
f2
M = f0 f1 f0 f2
(5.18)
1
f1 f2 f2 f0
with polynomials
f0 =
,
1 2
(1 2 ) cos sin cos sin
f1 =
,
1 2
1
(1 cos2 + 2 sin2 ) sin2

,
1 2

f3 = 2 sin2 +
cos2 1 sin2 + 2 cos2
1 2

1
+ tan2 sin2 1 cos2 + 2 sin2 .
f2 =
(5.19)
270
If the seventh frame e7 we chose is indeed correct, the mixing matrix M must be rank 2
so that its eigenvalues can be + , , 0 including zero. Equivalently, the bilinear form must
be of the form
3

Mij i j = + + + + ,
i,j =1
where + , are eigenvectors for + , respectively. This is in fact the case as one can
see by diagonalizing M. Two nonzero eigenvalues + , are solutions to the quadratic
equation

f () 2 f3 + f2 f0 + f01 f0 + f01 f12 f0 f2 f3 = 0
which says that both eigenvalues + and are always positive and the corresponding
eigenvectors + , are determined as

1
=
(f3 )1 f1 f0 2 + f1 3 .
( )(f3 )
Thus we arrive at the unwarped frames e i s given by
e1 = 1/2 cos d,
e2 = 1/2 cos 12 sin 1 ,
e3 = 1/2 cos 12 sin 2 ,
e4 = 1/2 cos 12 sin cos 3 ,

+
5
1
e = cos
f1 (u, J du) f1 f0 d (f3 ) d ,
(+ )(f3 )

e6 = 1 cos
f1 (u, J du) f1 f0 d (f3 + ) d ,
( + )(f3 + )

7
1
e =
(5.20)
+ 1 cos2 (u, J du) + f2 d + f1 d ,
where the first four frames are those on S4
= CP2 and e5 , e6 are oriented to + ,
6
respectively.
Now we can calculate the warp factor by using the local frames (5.20). Some
straightforward calculations show

7

+1
c7/2
ei =
=
7 ,
(5.21)
7
6 1 2
a 9/2 b9/2d 5/2
i=1
where 7 is identified with

7 =
1
sin(2) cos4 sin3 cos d 1 2 3 d d (u, J du).
16
6 Note that the warped frames (E 1 , E 2 , E 3 , E 4 , E 5 , E 6 , E 7 ) here are denoted by (e6 , e7 , e8 , e9 , e5 , e10 , e11 )
in Eq. (4.23) in [12].
271
This identification is correct since the prefactor in Eq. (5.21) becomes 1 in the maximally
symmetric SO(8)-invariant limit a = b = c = d = 1. By using Eq. (5.21) in the selfconsistent equation (5.14), the warp factor is determined as
1/6
ab
c1 4/3 .
=
(5.22)
cd
By using this, in Eq. (5.11) can be solved for to yield
5/4 5/4 3/4
1
a b d
G=
L2 g,
g=
2N
c3/4
which determines the warping of compact 7-dimensional space. Finally the warped line
element ds72 (dX, G dX) is determined as
ds72
5/4 5/4 3/4

7
a b d
2
L
Gmn dy dy =
ei ei ,
c3/4
m
(5.23)
i=1
where substitution of the local frames ei given in Eq. (5.20) produces the warped metric
Gmn described by the 7-dimensional global coordinates y m .
5.4. The compact 7-manifold metric for various critical points in SU(3)-invariant sector
There must be some comments on the local frames (5.20) here. First, e5 , e6 given above
are not well defined in SU(3) U (1) limit. We have f1 = 0 in this limit so that the matrix
M shows the mixing between 1 and 2 only. Diagonalizing the matrix M provides two
independent frames

1 1
sin(2) 4 (u, J du) 4 d
2
as shown in [12]. However, since either of f3 as well as f1 becomes 0 in this limit,
some of the coefficients in e5 , e6 become indefinite if we naively take the limit. The careful
analysis shows that

5 6
2 d, 1 for f3 > f2 f0 + f01 ,
e , e

1 , + 2 d for f3 < f2 f0 + f01 .
2 d
and 1
Because of this switching, e5 and e6 do not have definite physical meaning in SU(3)U (1)
limit although they are well defined in generic SU(3)-invariant cases.
Secondly, the set of local frames (5.20) is not unique and any orthogonal transformation
on three frames e5 , e6 and e7 is possible to produce the other frames e5 , e 6 and e7 satisfying
e5 e5 + e6 e6 + e 7 e7 e5 e5 + e6 e6 + e7 e7 .
In G2 limit, we have = 2 = b2 , 1 = a 2 and = 0 in Eq. (5.15) so that ds02 has no
contribution of the last term. To preserve the FubiniStudy metric on the common CP2 as
well as the ellipsoidal deformation along V1 direction, one can choose e5 and e7 so as to be
fixed to (u, J du) and the seventh component of dV1 , respectively. The remaining frame
272
e6 is determined by completing squares in ds02 . Then one can immediately see that e5 , e6 ,
e7 are identified respectively with
1
cos (u, J du),
b
sin d + sin cos cos d

,
b 1 sin2 cos2
d(cos sin )

.
ab 1 sin2 cos2
(5.24)
However, even if we take G2 limit in the local frames (5.20), e5 , e6 , e7 there cannot
reproduce the above three frames without recombining into new frames via some
orthogonal transformation. In fact, when we derived Eq. (5.20) we picked up the broken
Hopf fiber (5.17) as one of the local frames, expecting the restoration of the U (1) symmetry
along the Hopf fiber in SU(3) U (1) limit. However in G2 limit there is no restoration of
the U (1) symmetry and we have no reason to choose the broken Hopf fiber as one of the
frames. We will discuss more on this shortly.
Let us look at the consistency of our results in both SU(3) U (1) and G2 -invariant
sectors by reconstructing the 7-manifold metric from the local frames (5.20).
(SU(3) U (1))-invariant sector: in SU(3) U (1) limit, we have b = 1/a and
d = c = cosh so that Eqs. (5.22) and (5.23) reproduce the correct warping of the
7-dimensional metric in Section 2 [12]:
= ( cosh )4/3 ,
ds72 = L2 ds02 with 2 = 2 cos2 + 6 sin2 .
In this limit, we have = 2 , 1 = 2 = 6 and e5 , e6 become

2 d, and 1 cos sin 4 (u, J du) 4 d
as mentioned before. The first four frames are those for CP2 and e7 = 1 cosh (X, J dX)
with = 0 shows the stretching of Hopf fiber on CP3 . Combining the first six frames
provides
4
6

2
2
ei ei = 2 d2 + cos2 6 dsFS(2)
+ 2 sin2 6 (u, J du) 2 d ,
i=1
which is nothing but the ellipsoidal deformation of the FubiniStudy metric on CP3 (see
Eq. (B.6) in Appendix B). Therefore the compact 7-manifold is given by the ellipsoidal
deformation of stretched S7 as shown in [12].
There is an N = 2 supersymmetric critical point specified by = 31/8 , cosh(2) = 2.
This is the (SU(3) U (1))-invariant critical point found in [18]. The N = 2 RG flow in
Section 2 carries this critical point to the SO(8)-invariant critical point specified by = 1,
= 0 [9,12]. The SU(4) -invariant critical point in [17] is obtained by further taking the
limit = 1 in (SU(3) U (1))-invariant sector. Since we have = 1, the warp factor has
no dependence on and is just a scaling factor in the sense of 7 dimensions. The compact
7-manifold is a stretched S7 and its CP3 base is described by the FubiniStudy metric
given in Eq. (B.6). The SU(4) critical point is nonsupersymmetric and there is no RG
flow carrying it to the SO(8) critical point [7].
273
G2 -invariant sector: to look at this sector, it is better to use the other global
coordinates given in Appendix C. It is obtained by doing the replacement
cos = sin 6 sin 7 ,
sin cos = cos 7 ,
and
sin sin = cos 6 sin 7 ,
with + = 5 in the previous coordinates of Hopf fibration on CP3 . The deformed norm
(5.16) is now rewritten as

2 = 1 cos2 7 + sin2 6 + 2 cos2 6 sin2 7 .
In G2 limit, we have c = a, d = b so that Eqs. (5.22) and (5.23) reproduce the correct
warping of the 7-dimensional metric in Section 3 [2,10]:
= a 1 4/3 ,
ds72 = a b2 L2 ds02 , with 2 = a 2 cos2 7 + b2 sin2 7 .
Note that = 2 = b 2 , 1 = a 2 and = 0 in this limit and there is no stretching of S7 . As
mentioned before, it is better to transform the last three of the generic frames (5.20) into
e5 , e6 , e 7 in Eq. (5.24). They are now simply given by
e6 = b1 sin 7 d6 ,
e7 = (ab)1 d7 ,
e5 = b1 sin 7 sin 6 (u, J du),
7

and are subject to the identity i=5 e i e i 7i=5 ei ei . Thus the unwarped metric for
the compact 7-manifold is obtained as
7

ei ei =
i=1
1
a 2 b2
2 d72 +

2

1
sin2 7 d62 + sin2 6 dsFS(2)
+ (u, J du)2 ,
2
b
which describes the ellipsoidally deformed S7 [2]. Note that the inside of the square bracket
is the metric on S6
= G2 /SU(3) preserving the FubiniStudy metric onCP2 .
There is an N = 1 supersymmetric critical point specified by a = 6 3/5 and b =

2 3/5. This is the G2 -invariant critical point found in [2]. The N = 1 RG flow in
Section 3 carries this critical point to the SO(8)-invariant critical point specified by a =
b = 1 [10]. The SO(7)+ -invariant critical point corresponds to a = 1/b = 51/4 showing
the ellipsoidal deformation of the 7-manifold. The difference from the G2 critical point
is that there is no field strength of 11-dimensional gauge field in SO(7)+ [19]. Although
this critical point is nonsupersymmetric, there exists an RG flow connecting it to the
SO(8)
-invariant critical point corresponds to a = b = 5/2 so
critical point
[7].
The
SO(7)
that = 5/2 showing no deformation of S7 . The compact 7-manifold is in fact the

parallelized round S7 characterized by the nontrivial field strength of 11-dimensional gauge
field [20]. The SO(7) critical point is nonsupersymmetric and has no RG flow carrying it
to the SO(8) critical point.
6. Discussions
In this section, we will discuss our obtained results (5.20), (5.22) and (5.23) in the point
of view of 11-dimensional supergravity. First of all, by looking at the deformed norm
in Eq. (5.16) we notice that coordinate dependence of the warp factor is not so simple
274
for G2 -invariant sector. We derived the local frames (5.20) such that the restoration of
U (1) symmetry associated with the Hopf fiber on CP3 is obvious in SU(3) U (1) limit.
However, such an U (1) symmetry does not exist in G2 limit and the local frames are not
appropriate to look at the ellipsoidal deformation of the G2 -invariant 7-manifold.
Therefore, the global coordinates appropriate to describe the compact 7-manifold
crucially depends on the base 6-sphere which is CP3 for the (SU(3) U (1))-invariant
sector, whereas G2 /SU(3) for the G2 -invariant sector. The base manifold CP3
= S6 is
nothing but the homogeneous space SU(4)/[SU(3) U (1)] characterized by the Khler
form J [17]. Both SU(4) and (SU(3) U (1))-invariant 7-manifolds share the same U (1)
symmetry along the Hopf fiber (X, J dX) so that Hopf fibration on CP3 is useful in those
cases. On the other hand, the 7-manifolds with at least G2 invariance, namely SO(7)
and G2 -invariant 7-manifolds, share the 6-sphere given by G2 /SU(3) described as an S6
embedded in R7 spanned by imaginary octonions [21]. Since the ellipsoidal deformation
is transverse to this S6 , it is well described by using the 7-dimensional coordinates in
Appendix C. Therefore, the complication in the warp factor and in the local frames is due
to the difference in S6 base between the two sectors.
However, in spite of the difference in the S6 base, both sectors share the same S4
= CP2 .
4
This S is obvious in (SU(3) U (1))-invariant sector, while it is implicit in the 6-sphere
of G2 /SU(3) in G2 -invariant sector. It was pointed out in [12,13] that replacing the
S4 with S2 S2 provides another 11-dimensional solution corresponding to the d = 4,
N = 2, (SU(3) U (1))-invariant RG flow. It may be interesting to look at whether such
a replacement yields another 11-dimensional solution corresponding to the d = 4, N = 1,
G2 -invariant RG flow.
As summarized in Section 5, the SU(3)-invariant sector contains various critical
points. It is still unclear why some of critical points have holographic RG flows to the
SO(8) critical point but is not so for others. To answer this question, we have to solve
the 11-dimensional EinsteinMaxwell equations to complete the 11-dimensional lift of
whole SU(3)-invariant sector including RG flows. The 11-dimensional metric is given by
Eq. (1.1) where the compact 7-manifold metric Gmn and the warp factor are completely
determined by Eqs. (5.23), (5.22) in the local frames (5.20). The geometric parameters a,
b, c, d depend on the AdS4 radial coordinate r and are subject to the RG flow equations
(4.2) in 4-dimensional gauged supergravity [7]. The local frame (5.20) found in this paper
will be useful to achieve this work. For example, as performed in [12], one can easily make
an ansatz for the 3-form gauge field by using the local frames. We postpone this subject
for future study.
Acknowledgements
This work by C.A. was supported by Korea Research Foundation Grant (KRF-2002015-CS0006). This research by T.I. was supported by grant 1999-2-112-003-5 from the
Basic Research Program of the Korea Science & Engineering Foundation. C.A. is grateful
to B. de Wit, H. Nicolai and N.P. Warner for discussions and thanks Max-Planck-Institut
fr Gravitationsphysik, Albert-Einstein-Institut where part of this work was undertaken.
275
Appendix A. The 7-dimensional inverse metric encoded in SU(3)-singlet vevs of

d = 4, N = 8 gauged supergravity
According to the formula (5.1) of 7-dimensional inverse metric, one gets all the elements
of 1 GAB as follows. For simplicity, we did not write them completely but some of them
can be read off from the known expressions. For example, 1 G13 can be written as
1 G12 by replacing x2 with x3 and x7 with x6 . We list them below. For simplicity,
L2 1 GAB is denoted by (AB).

1 2
1
ac + bcd x22 + ac2 + bcd x32
2
2

1
+ a 3 + 2a 2b + ab2 + ac2 2acd + ad 2 x42
4

1
1
+ ac2 x52 + ac2 + bcd x62 + ac2 + bcd x72
2
2

1 3
+ a ab2 ac2 + ad 2 x4 x8
2

1
+ a 3 2a 2b + ab2 + ac2 + 2acd + ad 2 x82 ,
4

1
1
(12) = ac2 bcd x1 x2 + ac2 + bcd x2 x5
2
2

1 2
2
+ a b + ab acd bcd x4 x7
2

1
+ a 2 b ab2 acd + bcd x7 x8 ,
2

1
1
(13) = ac2 bcd x1 x3 + ac2 + bcd x3 x5
2
2

1 2
2
+ a b ab + acd + bcd x4 x6
2

1
+ a 2 b + ab2 + acd bcd x6 x8 ,
2
= (12) with (x2 x3 , x7 x6 ),

1
1
(14) = a a 2 + 2ab + b2 + c2 2cd + d 2 x1 x4 + a a 2 b2 + c2 d 2 x4 x5
4
4

1 2
1
+ a a b2 c2 + d 2 x1 x8 a a 2 2ab + b2 c2 + 2cd d 2 x5 x8 ,
4
4

1
(15) = ac2 bcd x22 + ac2 bcd x32
2

1
+ a 3 ab2 + ac2 ad 2 x42 ac2 x1 x5
4

1
1
1
+ ac2 bcd x62 + ac2 bcd x72 + a 3 + ab2 ac2 ad 2 x4 x8
2
2
2

1 3
+ a ab2 + ac2 ad 2 x82 ,
4
(11) =
276
(16) = (12) with (x7 x3 , x2 x6 ),

(17) = (12) with (x2 x7 , x7 x2 ),

1
1
(18) = a a 2 + b2 + c2 d 2 x1 x4 + a a 2 2ab b2 + c2 + 2cd + d 2 x4 x5
4
4

1
+ a a 2 + 2ab b2 c2 2cd d 2 x1 x8
4

1
+ a a 2 + b2 c2 + d 2 x5 x8 ,
4

1
1 2
(22) = ac + bcd x12 + bcdx32 + bcd + ad 2 x42
2
2

1
+ ac2 + bcd x52 + ac2 bcd x1 x5
2

1
+ bcdx62 + ab2x72 + bcd + ad 2 x4 x8 + bcd + ad 2 x82 ,
2

(23) = bcdx2x3 + ab 2 + bcd x6 x7 ,

1
1
(24) = bcd ad 2 x2 x4 + a 2b ab2 + acd + bcd x1 x7
2
2

1
1
2
+ bcd ad x2 x8 + a 2 b + ab2 + acd bcd x5 x7 ,
2
2
(25) = (12) with (x5 x1 , x4 x8 ),
(26) = (23) with (x3 x6 , x6 x3 ),
(27) = ab2x2 x7 ,
(28) = (24) with (x4 x8 , x1 x5 ),
(33) = (22) with (x2 x3 , x7 x6 ),
(34) = (24) with (x2 x3 , x7 x6 ),
(35) = (12) with (x5 x1 , x4 x8 , x2 x3 , x7 x6 ),
(36) = ab2x3 x6 ,
(37) = (23) with (x2 x7 , x7 x2 ),
(38) = (24) with (x4 x8 , x1 x5 , x2 x3 , x7 x6 ),

1
1
(44) = a 3 + 2a 2b + ab2 + ac2 2acd + ad 2 x12 + bcd + ad 2 x22
4
2

1
1
+ (bcd + ad 2 )x32 + a 3 ab2 + ac2 ad 2 x1 x5
2
2

1 3
2
2
+ a 2a b + ab + ac2 + 2acd + ad 2 x52
4

1
1
+ bcd + ad 2 x62 + bcd + ad 2 x72 + ad 2x82 ,
2
2
(45) = (18) with (x4 x8 , x1 x5 ),
(46) = (24) with (x2 x6 , x7 x3 ),
277
(47) = (24) with (x2 x7 , x7 x2 ),

(48) =

1 3
1
1
a ab2 ac2 + ad 2 x12 + bcd + ad 2 x22 + bcd + ad 2 x32
4
2
2
+

1 3
1
a + ab2 ac2 ad 2 x1 x5 + a 3 ab2 ac2 + ad 2 x52
2
4

1
1
bcd + ad 2 x62 + bcd + ad 2 x72 ad 2 x4 x8 ,
2
2
(55) = (11) with (x5 x1 , x4 x8 ),

(56) = (12) with (x4 x8 , x1 x5 , x2 x6 , x7 x3 ),
(57) = (12) with (x2 x7 , x5 x1 , x4 x8 , x7 x2 ),
(58) = (14) with (x1 x5 , x4 x8 ),
(66) = (22) with (x6 x2 , x7 x3 ),

(67) = ab2 + acd x2 x3 bcdx6x7 ,
(68) = (24) with (x1 x5 , x4 x8 , x2 x6 , x7 x3 ),
(77) = (22) with (x7 x2 ),
(78) = (24) with (x4 x8 , x1 x5 , x2 x7 , x7 x2 ),
(88) = (44) with (x5 x1 , x4 x8 ).
In the SU(4) -invariant limit (a = b = 1, d = c) the above generated results can be
arranged into the 8 8 matrix

1 G1 = c2 I xx T + 1 c2 (Jx)(Jx)T
by introducing the Khler form J given by
0
0
0
0
0
0
1 0
J=
0
0
0
0
0 1
0
0
0
0
0
0
0
1
0
0
1
0
0
0
0
0
0
0
0
0 0
0
0 1
0 1 0
0
0 0
0
0 0
0
0 0
0
0 0
1 0 0
0
0
.
1
0
0
(A.1)
278
This Khler form is transformed into the standard one J given by Eq. (5.2) in the text via
J= R T J R with the orthogonal matrix
0
0 0
0
0
0 1
0
0
1 0
0
0
0 0
0
0
0 0
0
2 0 0
2
1
0
0 0
0 0 1
0
2
2 .
R=
(A.2)
0
0 1
0
0
0 0
0
0 0
0
0
1 0
0
0
1
0 0
0
1 0 0
0
2
2
0
0 0 1
0
0 0 1
2
Appendix B. The global coordinates for S7 as Hopf fibration on CP3

In this appendix, we summarize basic properties of Hopf fibration and FubiniStudy
metric on CPn (n = 2, 3) used in the text.
R8 embedding of Hopf fibration on CP3 :

X = u cos + v sin with u = u1 , . . . , u6 , 0, 0 , v = 0, . . . , 0, v 7 , v 8 .
i

u1 + iu2 = sin cos 12 1 e 2 (2 +3 ) ei(+) ,
i

u3 + iu4 = sin sin 12 1 e 2 (2 3 ) ei(+) ,
u5 + iu6 = cos ei(+) ,
v 7 + iv 8 = ei .
(B.1)
Imaginary quaternion basis:

1 = cos 3 d1 + sin 1 sin 3 d2 ,
2 = sin 3 d1 sin 1 cos 3 d2 ,
3 = d3 + cos 1 d2 ,
(B.2)
satisfying the MaurerCartan equation:

di =
1
Jij k j k .
2
Hopf fiber on CP2 :

(u, J du) = d( + ) +
1 2
sin 3 .
2
(B.3)
279
FubiniStudy metric on CP2 :

2
(du)2 (u, J du)2 = d 2 +
dsFS(2)

1 2 2
sin 1 + 22 + cos2 32 .
4
(B.4)
Hopf fiber on CP3 :

1 2
(X, J dX) = sin d + cos (u, J du) = d + cos d + sin 3 .
2
2
(B.5)
FubiniStudy metric on CP :
3
2
dsFS(3)
(dX)2 (X, J dX)2

= d + cos
2
2
dsFS(2)
2

1 2
.
+ sin d + sin 3
2
2
(B.6)
To get understanding of Hopf fibration, let us demonstrate to reconstruct the compact 7manifold metric for the SU(4) -invariant critical point. At this critical point, we have a =
b = 1, d = c and = 1 = 1 = 2 . Due to the common singularity in the (SU(3) U (1))invariant sector, we get either (e5 , e6 ) = (d, ) or (e5 , e6 ) = (, +d) where we used
the fact that = 1. Then e5 e5 + e6 e6 will lead to

2
1
d2 + sin2 cos2 d + sin2 3 .
2
Combining other ei , i = 1, 2, 3, 4 with e5 , e6 , the metric can be written as

2

6

1
2
ei ei = d2 + cos2 dsFS(2)
+ sin2 d + sin2 3
2
i=1
which is nothing but the standard FubiniStudy metric on CP3 [17]. Moreover, e7 in this
limit will be e7 = c(cos2 (u, J du) + sin2 d). Using the explicit form of (u, J du) one
gets

2
1
e7 e7 = c2 d + cos2 d + sin2 3
2
that is equal to the U (1) bundle and is the coordinate on the U (1) fibers. According to
the result of [17], the compact 7-manifold metric in the above will be an Einstein which
2
corresponds to the trivial SO(8)-invariant metric on the round
7-sphere when c = 1.3 On
the other hand, at theSU(4) -invariant critical point (c = 2 ), U (1) fibers over CP are
stretched by a factor 2 and the compact 7-manifold metric is not Einstein.
Appendix C. The global coordinates for S7 with G2 /SU(3) base
In this appendix, we describe the 7-dimensional coordinatization appropriate for the
base 6-sphere of G2 /SU(3).
280
R8 embedding of S7 with G2 /SU(3) base:

X = u sin 6 sin 7 + v

with u = u1 , . . . , u6 , 0, 0 , v = (0, . . . , 0, cos 7 , cos 6 sin 7 ),
where u is the same as above except for the replacement + 5 .
Relation to the Hopf fibration on CP3 :
cos 7 = sin cos ,
cos 6 sin 7 = sin sin ,
sin 6 sin 7 = cos , and 5 = + ,

d(cos sin )
sin d + sin cos cos d

d7 =
, sin 7 d6 =
,
1 sin2 cos2
1 sin2 cos2
(C.1)
sin d = cos 6 sin 7 d6 + sin 6 cos 7 d7 ,

sin2 d = cos 6 d7 sin 6 cos 7 sin 7 d6 .
The metric on S6
= G2 /SU(3):
2

+ (u, J du)2 .
d62 d62 + sin2 6 dsFS(2)
References
[1] B. de Wit, H. Nicolai, Nucl. Phys. B 208 (1982) 323;
B. de Wit, H. Nicolai, Phys. Lett. B 108 (1982) 285.
[2] B. de Wit, H. Nicolai, N.P. Warner, Nucl. Phys. B 255 (1985) 29.
[3] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[4] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[5] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[6] E. Cremmer, B. Julia, Phys. Lett. B 80 (1978) 48.
[7] C. Ahn, K. Woo, Nucl. Phys. B 599 (2001) 83, hep-th/0011121.
[8] N.P. Warner, Phys. Lett. B 128 (1983) 169.
[9] C. Ahn, J. Paeng, Nucl. Phys. B 595 (2001) 119, hep-th/0008065.
[10] C. Ahn, T. Itoh, Nucl. Phys. B 627 (2002) 45, hep-th/0112010.
[11] C. Ahn, S.-J. Rey, Nucl. Phys. B 572 (2000) 188, hep-th/9911199.
[12] R. Corrado, K. Pilch, N.P. Warner, Nucl. Phys. B 629 (2002) 74, hep-th/0107220.
[13] C.V. Johnson, K.J. Lovis, D.C. Page, JHEP 0110 (2001) 014, hep-th/0107261.
[14] P.G.O. Freund, M.A. Rubin, Phys. Lett. B 97 (1980) 233.
[15] B. de Wit, H. Nicolai, Nucl. Phys. B 281 (1987) 211.
[16] C. Ahn, K. Woo, Nucl. Phys. B 634 (2002) 141, hep-th/0109010.
[17] C.N. Pope, N.P. Warner, Phys. Lett. B 150 (1985) 352.
[18] H. Nicolai, N.P. Warner, Nucl. Phys. B 259 (1985) 412.
[19] B. de Wit, H. Nicolai, Phys. Lett. B 148 (1985) 60.
[20] F. Englert, Phys. Lett. B 119 (1982) 339.
[21] M. Gnaydin, N.P. Warner, Nucl. Phys. B 248 (1984) 685.
(C.2)
(C.3)

A local effective action for photongravity

interactions
G.M. Shore
Department of Physics, University of Wales, Swansea, Singleton Park, Swansea, SA2 8PP, UK
Received 28 August 2002; accepted 17 September 2002
Abstract
Quantum phenomena such as vacuum polarization in curved spacetime induce interactions
between photons and gravity with quite striking consequences, including the violation of the strong
equivalence principle and the apparent prediction of superluminal photon propagation. These
quantum interactions can be encoded in an effective action. In this paper, we extend previous results
on the effective action for QED in curved spacetime due to Barvinsky, Vilkovisky and others, and
present a new, local effective action valid to all orders in a derivative expansion, as required for a full
analysis of the quantum theory of high-frequency photon propagation in gravitational fields.
1. Introduction
In classical electrodynamics in curved spacetime, the interaction of the electromagnetic
and gravitational fields is encoded in the Maxwell action

1
(0) =
(1.1)
dx g F F ,
4
where the field strength F = D A D A is defined with covariant derivatives. The
interaction involves only the connection and is independent of the spacetime curvature.
This action therefore embodies the strong equivalence principle (SEP), which states that
the laws of physics should be the same in the local inertial frames at each point in spacetime
and reduce to their special relativistic form at the origin of each of these frames. This is
assured since the connections (though not of course the curvatures) vanish locally in an
appropriate frame in a Riemannian manifold.
E-mail address: g.m.shore@swansea.ac.uk (G.M. Shore).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 3 3 - 7
282
G.M. Shore / Nuclear Physics B 646 (2002) 281300
The picture is rather different in quantum theory. In quantum electrodynamics in curved

spacetime, vacuum polarization loops involving the electron induce interactions between
photons and the background gravitational field that depend explicitly on the curvature, with
the necessary scale being set by the Compton wavelength of the electron, c = m1 . These
interactions therefore effectively violate the SEP at the quantum level.
The simplest effective action which encodes some of the essential physics of vacuum
polarization was constructed explicitly by Drummond and Hathrell [1], who used standard
SchwingerDe Witt proper time/heat kernel methods (see, e.g., [26]) as well as an
independent diagrammatic technique to find

1

DH = (0) + 2 dx g aRF F + bR F F
m

+ cR F F + dD F D F .
(1.2)
Here, a, b, c, d are perturbative coefficients of O(), viz.
13
1
1
1
,
b=
,
c=
,
d =
.
(1.3)
144
360
360
30
As well as being a one-loop result, there are two other significant limitations on DH . First,
it is at most of third order in the generalized curvatures (including both the Riemann
curvature and the electromagnetic field strengths). Second, it is just the lowest-order term
in a derivative expansion involving higher orders in O(D/m) acting on the curvatures.
If we restrict ourselves to the interesting case of terms of O(F 2 ), which determine the
photon propagator and thus the quantum effects on the propagation of light in gravitational
fields, the first restriction is equivalent to working to first order in O(R/m2 ), or O(2c /L2 )
where L is a typical curvature scale, i.e., weak background gravitational fields. In terms
of photon propagation, the second condition implies a restriction to low frequencies, i.e.,
lowest order in O(c /) where is the photon wavelength.
By far the most dramatic consequence of Eq. (1.2) is its prediction of superluminal
photon velocities. As explained in Ref. [1] (see Refs. [712] for a selection of subsequent
work), the new SEP-violating interactions induce curvature-dependent shifts in the light
cones, which become
a=
1
8c
(2b + 4c)R k k + 2 C k k a a = 0,
(1.4)
m2
m
where k is the photon momentum and a is its polarization. Note that we have written
Eq. (1.4) in a convenient form involving the Weyl tensor C . For suitably chosen
trajectories in certain curved spacetimes, this condition remarkably permits k 2 to be spacelike.
Although photon propagation is the most striking application of the effective action, it is
clearly important much more generally in encoding the often unintuitive effects of gravity
on quantum fields. In this paper, we present an extension of the DrummondHathrell action
in which we relax the restriction to lowest order in the derivative expansion, while retaining
the weak gravitational field approximation. The application of this improved effective
action to the photon propagation problem and the nature of dispersion in gravitational
fields is considered separately in Ref. [12]; here we discuss the technical issues arising in
the construction of the effective action itself.
k2
283
At one-loop order, the QED effective action is given (in Euclidean space) by
= (0) + ln det S(x, x ),
(1.5)
where (0) is the free Maxwell action and S(x, x ) is the Green function of the Dirac
operator in the background gravitational field, i.e.,
1
(iD
/ m)S(x, x ) = (x, x ).
g
(1.6)
In fact it is more convenient to work with the differential operator corresponding to the
scalar Green function G(x, x ) defined by
S(x, x ) = (iD
/ + m)G(x, x ),
so that

1
1
2
2
D + ie F R + m G(x, x ) = (x, x ).
4
g
(1.7)
(1.8)
Then we evaluate from the heat kernel, or proper time, representation

1
= (0)
2
ds m2 s
e
Tr G(x, x ; s),
s
(1.9)
where
G(x, x ; s),
(1.10)
s
with G(x, x ; 0) = G(x, x ). Here, D is the differential operator in Eq. (1.8) at m = 0.
Our principal result is summarized in the following expression for the one-loop QED
effective action, to first order in O(R/m2 ):

1
1
= dx g ZF F + 2 D F G0 D F + G1 RF F
4
m

+ G2 R F F + G3 R F F
1
+ 4 G4 RD F D F + G5 R D F D F
m
+ G6 R D F D F + G7 R D D F F

(1.11)
+ G8 R D D F F + G9 R D F D F
.
DG(x, x ; s) =
The Gn (n 1) are form factor functions of three operators:

2
2
2
D(2)
D(3)
D(1)

Gn Gn 2 , 2 , 2 ,
m
m
m
(1.12)
2 ) acts on the first following term (the curvature), etc., G

0 is
where the first entry (D(1)
similarly defined as a single variable function. The explicit expressions for these form
factors are presented later in the paper.
284
Although the entire effective action is a non-local object, for theories such as QED with
a massive electron, it should permit a local expansion in inverse powers of m. A crucial
feature of the above form of the effective action is that it is indeed manifestly local, in
2 . This
the sense that the form factors Gn have an expansion in positive powers of the D(i)
depends on making the particular choice of basis operators above.
A very general effective action calculation, which encompasses the more specialized
results we need, has been given some time ago by Barvinsky, Gusev, Zhytnikov and
Vilkovisky (BGZV) in Ref. [13], building on methods developed earlier by two of the
authors [1416]. Closely related results have also been obtained by Avramidi [17]. The
present paper is essentially a translation and specialization of the BGZV result to the
O(RF F ) terms in the QED effective action. However, the BGZV result is presented in
an apparently non-local form and the translation into a manifestly local effective action
through an appropriate choice of basis operators involves a number of subtle manoeuvres.
Together with the complexity of the manipulations on the form factors, we feel this justifies
our independent presentation, in which we explain the delicate technical points in the
translation and quote explicit results for the form factors in the local effective action (1.11).
In the Section 2, we introduce the BGZV effective action and show how it is recast
in local form and reduced to the O(RF F ) action we are looking for. Then in Section 3,
we reconsider the O(F 2 ) terms and rewrite them in a convenient form consistent with the
DrummondHathrell action. Section 4 contains a summary of our final result, with explicit
algebraic expressions for the form factors collected in Appendix A. A numerical analysis
of the form factors is given in Section 5. The paper is presented so that the final results for
the local effective action and the form factors are entirely self-contained in Section 4 and
Appendix A.
2. The BGZV effective action

Barvinsky et al. evaluate the heat kernel G(x, x ; s) (denoted by K in Refs. [13,15])
corresponding to the generic second-order elliptic differential operator
1
H = 1 + P R 1,
6
g D D ,
(2.1)
acting on small fluctuations A of an arbitrary set of fields A (x). The hat notation
denotes matrices acting on the vector space of the A , i.e., P = P A B , 1 = A B , etc.
and the matrix trace is denoted tr. The Euclidean metric g describes the background
spacetime, with gravitational Ricci and Riemann curvatures R and R , respectively.
A further curvature is defined by the commutators of the covariant derivatives acting on
the fields A :
(D D D D ) A = RA B B ,
(2.2)
= RA B . The potential P is an arbitrary matrix. BGZV introduce

and we denote R
the collective notation R for the full set of generalized curvatures:

, P ,
R = R , R
(2.3)
285
and calculate the heat kernel G up to O(R3 ).

The first step in extracting the QED effective action from the BGZV results is to identify
the curvatures R by comparing the differential operator D of Eqs. (1.8), (1.10) with the
standard form H . In this case, the field A is the electron spinor field (x), so we identify
A, B as Dirac spinor indices , and identify
1
P = ie F R 1.
12
The (gauge and gravitational) covariant derivative acting on spinors is
1
D = + ab ab + ieA ,
2
where a b is the spin connection

a
e ,
a b = eb ea
(2.4)
(2.5)
(2.6)
with ea the vierbein and ab = 14 [ a , b ]. Evaluating the commutator, we find

1
[D , D ] = ieF 1 + ab Rab ,
2
and so, comparing with Eq. (2.2), we identify
(2.7)
= ieF 1 + 1 R .
R
(2.8)
2
Having established this identification between the BGZV generalized curvatures and the
QED field strength and gravitational curvature tensors, we can now summarize their
result
for Tr G(x, x ; s). Here, Tr denotes the full functional trace, i.e., Tr G(x, x ; s) =

dx g tr G(x, x ; s)|x =x . The result is [13]:

Tr G =
1 1
(4)2 s 2

5

fi (s(2))R1 R2 (i)
dx g tr 1 + s P + s 2
i=1
+ s3
11

Fi (s(1), s(2) , s(3))R1 R2 R3 (i)
i=1
+ s4
25

Fi (s(1), s(2) , s(3))R1 R2 R3 (i)
i=12
+ s5
28

Fi (s(1), s(2) , s(3))R1 R2 R3 (i)
i=26

+ s 6 F29 (s(1) , s(2), s(3))R1 R2 R3 (29) + O R4 .
(2.9)
286
Here, R1 R2 (i) denote independent basis operators formed from terms of secondorder in the generalized curvatures (2.3). Similarly for the independent third-order terms
R1 R2 R3 (i). Notice that these appear with different powers of s, determined by their
dimension. The form factors fi and Fi are functions of the operators (1) , etc., with the
suffix indicating which curvature term is acted upon. Thus, for example, since R1 R2 (5) =
2 , the corresponding contribution to Tr G is
1 R
R

1

2 .
(2.10)
dx g R
1 f5 (s)R
2
(4)
We will return to the detailed expressions for the form factors later. First, we consider the
basis of 5 O(R2 ) and 29 O(R3 ) operators and, with the identifications in Eqs. (2.4) and
(2.8), pick out those from the full list which produce the terms of O(F 2 ) and O(RF F )
relevant for our QED effective action. The remaining terms involving purely gravitational
curvatures of O(R 2 ) and O(R 3 ) are neglected. In BGZV notation, the relevant basis
operators are as follows:

1
tr R1 R2 (4) = tr P1 P2 = e2 F F tr 1 + O R 2 ,
2

2 = e2 F F tr 1 + O R 2 ,
1 R
tr R1 R2 (5) = tr R
1
tr R1 R2 R3 (1) = tr P1 P2 P3 = e2 RF F + F RF
24

+ F F R tr 1 + O R 3

P3 = 1 e2 F F R + 1 e2 R F F
1 R
tr R1 R2 R3 (3) = tr R
2
12
4

1 2
+ e F R F tr 1 + O R 3 ,
4

1
tr R1 R2 R3 (6) = tr P1 P2 R3 = e2 F F R tr 1 + O R 3 ,
2

2 R
= e2 RF F tr 1 + O R 3 ,
tr R1 R2 R3 (7) = tr R1 R
3

R
= e2 R F F tr 1 + O R 3 ,
tr R1 R2 R3 (8) = tr R1 R
3
2
D R
P3
tr R1 R2 R3 (14) = tr D R
2
1

1 2
1
e D F D F R + e2 D R D F F
=
12
4

1 2
tr 1 + O R 3 ,
+ e D F D R F
4

2 P3 = 1 e2 R D D F F tr 1 + O R 3 ,
tr R1 R2 R3 (17) = tr R1 D D P
2

D R
= e2 R D F D F tr 1 + O R 3 ,
tr R1 R2 R3 (18) = tr R1 D R
2
3

D R
3 = e2 R D F D F tr 1 + O R 3 ,
tr R1 R2 R3 (19) = tr R1 D R
2

D R
3 = e2 RD F D F tr 1 + O R 3 ,
tr R1 R2 R3 (20) = tr R1 D R
2
287

2 R
= e2 R D D F F tr 1 + O R 3 .
tr R1 R2 R3 (21) = tr R1 D D R
3
(2.11)
Notice that for clarity we have omitted the subscripts 1, 2, 3 on the terms in the right-hand
column, but must remember that the order is still significant since they are to be acted on
by the form factors. 1 is the unit Dirac matrix, so in four dimensions tr 1 = 4.
Collecting these terms and substituting into Eq. (2.9) then gives the complete O(F 2 )
and O(RF F ) terms in the BGZV effective action. However, as it stands this is expressed
in non-local form, whereas we require a manifestly local action. As explained in Ref. [15],
the key is to reintroduce the Riemann tensor into the basis of operators. (Notice that none
of the above set of terms in the form tr R1 R2 R3 involves the uncontracted Riemann tensor
and P
according
R , which only enters in the right-hand column when we expand R
to Eqs. (2.4), (2.8). This is a feature of the covariant perturbation theory construction
developed by Barvinsky and Vilkovisky.) We therefore introduce the further operator

R
= e2 R F F tr 1 + O R 3 .
tr R1 R
(2.12)
2
3
The Riemann tensor admits the following non-local re-expression in terms of the Ricci
tensor, up to O(R 2 ):

1
R =
D D R D D R D D R + D D R + O R 2 . (2.13)

This is proved [15] by differentiating the Bianchi identity
R[; ] = 0
(2.14)
to obtain an equation for R , which is solved iteratively in powers of curvature using

the Green function 1/ to obtain (2.13).
We can then rewrite the four BGZV basis operators which turn out to have non-local
form factors as:

R
+ s 2 F18 R D R
D R

dx g tr sF8 R R

D R
+ s 2 F21 R D D R
R

+ s 2 F19 R D R

R
+ s 2 F
D R

18 R D R
= dx g tr s F
8 R R
D R
+ s 2 F
R

19 R D R
21 R D D R
+ s2F

R
,
0 R R
+ sF
with the following relation between the form factors:

8 (x1 , x2 , x3 ) + 2 1 + x2 + x3 F
0 (x1 , x2 , x3 ),
F8 (x1 , x2 , x3 ) = F
x1
0 (x1 , x2 , x3 ),
18 (x1 , x2 , x3 ) 4 F
F18 (x1 , x2 , x3 ) = F
x1
19 (x1 , x2 , x3 ) + 2 F
0 (x1 , x2 , x3 ),
F19 (x1 , x2 , x3 ) = F
x1
21 (x1 , x2 , x3 ) 8 F
0 (x1 , x2 , x3 ),
F21 (x1 , x2 , x3 ) = F
x1
(2.15)
(2.16)
288
where we have used the notation xi = (i) . (Notice that BGZV [13] use i = s(i) in
discussing the form factors.)
0 (x1 , x2 , x3 ) is defined so as to remove the singular terms in
The new form factor F
8,18,19,21 form factors regular as x1 , x2 , x3 0. Of course this
F8,18,19,21 leaving the F
0 , but any ambiguity only amounts to a reshuffling of contributions
under-determines F
amongst the various terms in the final expression for the effective action.
In the local basis, we therefore find that all the O(F 2 ) and O(RF F ) terms in the heat
kernel are contained in the following BGZV terms (again suppressing the arguments of the
form factors, i.e., fi fi (s(2)) and Fi Fi (s(1), s(2) , s(3))):

1 R
2
dx g tr f4 P1 P2 + f5 R
1 R
P
2 R

1 P
2 P
3 + sF3 R
3 + sF6 P1 P2 R3 + sF7 R1 R
+ sF1 P
2
3
D R
P3 + s 2 F17 R1 D D P2 P3
+ s 2 F14 D R
2
1
D R
3 + s F
R

8 R1 R
+ s 2 F20 R1 D R
3
2
2
D R
+ s 2 F
D R
3
18 R1 D R
19 R1 D R
+ s2F
2
3
2

2 R
+ s F
R
.
21 R1 D D R
0 R R
+ s2F
3
(2.17)
We can now re-express this explicitly in terms of the O(F 2 ) and O(RF F ) basis operators
of Eq. (1.11) using the set of relations (2.11). We also need to use the following identity,
which is proved by integrating by parts and using the Bianchi identity for F :
dx g D R D F F

1
= dx g R D F D F + R F F + O R 2 .
(2.18)
2
This gives:
e2
tr 1
(4)2

dx g F h0 F

+ s h1 RF F + h2 R F F + h3 R F F

+ s 2 h4 RD F D F + h5 R D F D F
Tr G =
+ h6 R D F D F + h7 R D D F F

+ h8 R D D F F + h9 R D F D F ,
(2.19)
where h0 h0 (s) and hi hi (s(1), s(2), s(3) ), i 1. The form factors are
related to the BGZV definitions as follows (where we assume appropriate symmetrizations
on x1 , x2 , x3 ):
1
h0 (x) = f4 (x) + f5 (x),
2
289
1
1
h1 (x1 , x2 , x3 ) = F1 (x1 , x2 , x3 ) F3 (x2 , x3 , x1 )
8
12
1
F6 (x2 , x3 , x1 ) + F7 (x1 , x2 , x3 ),
2
8 (x1 , x2 , x3 ),
h2 (x1 , x2 , x3 ) = F
1
1
0 (x1 , x2 , x3 ),
h3 (x1 , x2 , x3 ) = F3 (x1 , x2 , x3 ) F14 (x1 , x2 , x3 )(x2 + x3 ) + F
2
8
1
h4 (x1 , x2 , x3 ) = F14 (x2 , x3 , x1 ) + F20 (x1 , x2 , x3 ),
12
18 (x1 , x2 , x3 ),
h5 (x1 , x2 , x3 ) = F
19 (x1 , x2 , x3 ),
h6 (x1 , x2 , x3 ) = F
1
h7 (x1 , x2 , x3 ) = F17 (x1 , x2 , x3 ),
2
21 (x1 , x2 , x3 ),
h8 (x1 , x2 , x3 ) = F
1
h9 (x1 , x2 , x3 ) = F14 (x1 , x2 , x3 ).
(2.20)
2
3. Second-order terms
While Eq. (2.19) is a perfectly satisfactory, local expression for Tr G, it is not the most
convenient form for many applications, e.g., photon propagation. Nor is it the form which
permits the simplest comparison with standard SchwingerDe Witt results. The key step
in transforming
to this standard
form is to rewrite the second-order contributions of the

type dx g F F as dx g D F D F , plus higher-order terms of O(RF F )

which of course modify the corresponding form factors. This second choice for the O(F 2 )
basis operator also matches the term in the DrummondHathrell action (1.2) and turns out
to be the more useful representation.
This transformation can be made already at the level of the BGZV curvatures using the
identity [13,15]:

R
D R
+ 4R
R

tr dx g R R = tr dx g 2D R

R
R R
R
. (3.1)
+ 2R R
In terms of the electromagnetic field strength and gravitational curvature, this is
equivalent to
dx g F F

= dx g 2D F D F + 2R F F

R F F + O R 2 .
(3.2)
The proof is straightforward, relying on the Bianchi identity for F and the cyclic
symmetry of the Riemann tensor.
290
We now need to extend this identity to include the full form factors. The first step is to
separate off the constant term in h0 () and define
dx g F h0 ()F

= h0 (0) dx g F F + dx g F h()F .
(3.3)
The first term on the r.h.s., where h(0) = 1/6, eventually gives rise to a divergent
contribution to the effective action which is removed by the conventional electromagnetic
field renormalization. The remaining term simplifies in two stages. First, write
dx g F h()F

= dx g 2D F h()D F + 2R F h()F

R F h()F + 2D F D , h() F .
(3.4)
The second and third terms are already in the required form of O(RF F ) basis operators.
Neglecting higher-order terms in the curvature simplifies the commutator, and with
repeated use of the Bianchi identity for the Riemann tensor we eventually find
dx g F h()F

= dx g 2D F h()D F + 2R h() + h () F F
R h()F F 2R h ()D F D F
+ 4R h ()D D F F + 2R h ()D F D F

+ O R2 F F .
(3.5)
The transformation between the alternative choices for the O(F 2 ) basis operator therefore
affects the O(RF F ) form factors h2 , h3 , h5 , h8 and h9 , in the notation of (2.19).
With this transformation, we can rewrite Tr G into the preferred form:
e2
tr 1
(4)2

1
dx g F F + sD F g0 D F
6

+ s g1 RF F + g2 R F F + g3 R F F

+ s 2 g4 RD F D F + g5 R D F D F
Tr G =
+ g6 R D F D F + g7 R D D F F
+ g8 R D D F F + g9 R D F D F

,
(3.6)
291
with g0 g0 (s) and gi gi (s(1), s(2) , s(3)), i 1 as before. The new form
factors are related to the initial ones by:

2
1
1
g0 (x) = 2h(x) =
f4 (x) + f5 (x) +
,
x
2
6
g1,4,6,7 = h1,4,6,7,
g2 = h2 + h(x2 ) + x2 h (x2 ) + h(x3 ) + x3 h (x3 ),

1
g3 = h3 h(x2 ) + h(x3 ) ,
2

g5 = h5 h (x2 ) + h (x3 ) ,
g8 = h8 4h (x2 ),
g9 = h9 + 2h (x2 ),
(3.7)
where gi gi (x1 , x2 , x3 ), i 1 and we have symmetrized where appropriate.
4. The local effective action

Collecting these results and evaluating the effective action from
1
= 0
2
ds m2 s
e
Tr G(x, x ; s),
s
(4.1)
we find the renormalized expression:

1
1
= dx g ZF F + 2 D F G0 D F , +G1 RF F
4
m

+ G2 R F F + G3 R F F
1
+ 4 G4 RD F D F + G5 R D F D F
m
+ G6 R D F D F + G7 R D D F F

+ G8 R D D F F + G9 R D F D F ,
(4.2)
2
2
2

where the form factors Gn = Gn ((1)/m , (2)/m , (3)/m ) are defined as proper
time integrals of the gn , viz.
1
Gn (x1 , x2 , x3 ) =
2
ds s p
e s gn (sx1 , sx2 , sx3 ),
s
(4.3)
where p = 1 for n = 0, . . . , 3 and p = 2 for n = 4, . . . , 9. Note that we have rescaled s

here so that it is dimensionless, while now xi = (i) /m2 . The divergent contribution to
the F F term has been evaluated using dimensional regularization and cancelled by
292
the standard electromagnetic field renormalization. The residual finite O() contribution
2
gives Z = 1 + 16 ln m
for a suitable choice of RG scale .
2
The form factors are given in terms of the BGZV definitions by the following set of
relations:

1
2
1
g0 (x) = 2h(x) =
f4 (x) + f5 (x) +
,
x
2
6
1
1
g1 (x1 , x2 , x3 ) = F1 (x1 , x2 , x3 ) F3 (x2 , x3 , x1 )
8
12
1
F6 (x2 , x3 , x1 ) + F7 (x1 , x2 , x3 ),
2
8 (x1 , x2 , x3 ) + h(x2 ) + x2 h (x2 ) + h(x3 ) + x3 h (x3 ),
g2 (x1 , x2 , x3 ) = F
1
1
g3 (x1 , x2 , x3 ) = F3 (x1 , x2 , x3 ) F14 (x1 , x2 , x3 )(x2 + x3 )
2
8

1
0 (x1 , x2 , x3 ) h(x2 ) + h(x3 ) ,
+F
2
1
g4 (x1 , x2 , x3 ) = F14 (x2 , x3 , x1 ) + F20 (x1 , x2 , x3 ),
12

g5 (x1 , x2 , x3 ) = F18 (x1 , x2 , x3 ) h (x2 ) + h (x3 ) ,
19 (x1 , x2 , x3 ),
g6 (x1 , x2 , x3 ) = F
1
g7 (x1 , x2 , x3 ) = F17 (x1 , x2 , x3 ),
2
21 (x1 , x2 , x3 ) 4h (x2 ),
g8 (x1 , x2 , x3 ) = F
1
g9 (x1 , x2 , x3 ) = F14 (x1 , x2 , x3 ) + 2h (x2 ),
(4.4)
2

where h (x) dh
dx and the Fi are given in terms of the BGZV Fi by Eq. (2.16). Again we
assume that the BGZV form factors have been appropriately symmetrized to reflect the
symmetries of the basis operators on which they act.
The BGZV form factors themselves are known algebraic expressions which involve the
following integrals [13,15]:
1
f (x) =
d e(1)x ,
(4.5)
and

F (x1 , x2 , x3 ) =
d 3 (1 1 2 3 )e(1 2 x3 +2 3 x1 +3 1 x2 ) .
(4.6)
0
The second-order form factors are easily simplified. Using the BGZV expressions,
1
f4 (x) = f (x),
2
f5 (x) =

1
f (x) 1 ,
2x
(4.7)
293
we have

1
1
1
1
+
+
g0 (x) =
f (x).
3x x 2
2x x 2
(4.8)
Using the small x expansion of f (x), viz.

1
1 3
1
x + ,
f (x) = 1 x + x 2
(4.9)
6
60
840
we find
1
1
x + .
g0 (x) =
(4.10)
15 140
The absence of inverse powers of x confirms that this is indeed a local form factor.
The expansions for the third-order form factors Fi (x1 , x2 , x3 ) are extremely complicated. They can be read off from the results given in Ref. [13] and we will not quote them
in full here. Instead, we focus on a special case of particular interest. In most applications, it
is reasonable to assume that the derivatives of the gravitational curvatures are governed by
= O(3c /L3 ).
the same scale as the curvatures themselves, i.e., if mR2 = O(2c /L2 ) then DR
m3
In that case, terms with derivatives of curvatures are suppressed by the same parameter
governing the expansion in increasing powers of curvature, and should be neglected. We
would then only be interested in the form factors with the first argument set to zero, i.e.,
gn (0, x2, x3 ).
i (0, x2 , x3 )
We have calculated the corresponding form factors Fi (0, x2, x3 ) and F

appearing in Eq. (4.4), identifying the new form factor F0 (0, x2 , x3 ) required for the
transformation to the local basis. The resulting lengthy expressions are collected in
Appendix A.
These results must pass three crucial consistency checks. First, the transformation from
0 (x1 , x2 , x3 ) as described in
the non-local basis to the local one via the introduction of F
Eq. (2.16) must work, i.e., the 1/x1 singularities in each of the form factors F8,18,19,21
0 . Second, all the resulting form factors must be
must be removed by the same function F
entirely local, in the sense that they admit Taylor expansions with no inverse powers of
x1,2,3 . Finally, the final form factors g0 (0) and gn (0, 0, 0), n = 1, 2, 3 must reproduce the
SchwingerDe Witt coefficients in the DrummondHathrell effective action. In algebraic
terms, each of these tests is highly non-trivial, but as can be verified from the explicit
results in the appendix, each is indeed satisfied.
The x1,2,3 0 limit of the form factors in the effective action can be found from the
results quoted in Appendix A, using Gn (0, 0, 0) = 12 gn (0, 0, 0). From the definitions
(4.4) we find:
G0 (0) =
1
,
30
1
,
360
1
,
G6 (0, 0, 0) =
2520
5
,
G9 (0, 0, 0) =
1008
G3 (0, 0, 0) =
1
,
144
1
G4 (0, 0, 0) =
,
3360
1
G7 (0, 0, 0) =
,
720
G1 (0, 0, 0) =
13
,
360
1
G5 (0, 0, 0) =
,
315
2
G8 (0, 0, 0) =
,
315
G2 (0, 0, 0) =
(4.11)
294
and we confirm that G0 (0), G1,2,3(0, 0, 0) correctly reproduce the coefficients d, a, b, c,

respectively, in the DrummondHathrell action (1.2).
5. Numerical analysis of the form factors

In order to get a better feeling for the behaviour of the form factors, we have evaluated
them numerically. We present some of the results in graphical form in this section. The
general features of the different form factors are quite similar, so here we just give
the analysis for the form factors relevant for Ricci flat spaces, i.e., g3 (0, x2 , x3 ) and
g9 (0, x2, x3 ).
Collecting formulae from Appendix A and substituting into Eq. (4.4) we find

1
1
g3 (0, x2 , x3 ) = F (0, x2, x3 )
4 + 1 + (x2 + x3 ) (x2 + x3 )
2
4

1 3x2 x3 1
2(x2 + x3 )2
+ f (x2 )
+ (x2 x3 ) 1
2
2x2
4
2

1 3x3 x2 1
2(x2 + x3 )2
+ f (x3 )
+ (x3 x2 ) 1
2
2x3
4
2

1 x3 x2
1 1
1
1
1 +
+
+ x2 + x3 +
+
+
2
2 x2 x3
12 x2 x3

1 1
1
+
(5.1)
+ 2 ,
2
4 x2
x3
and

1
2 2
2
1
2
g9 (0, x2 , x3 ) = F (0, x2, x3 ) [4 + x2 + x3 ] + f (x2 ) 2 x2 x3 2 3
2
2
2x2
x2

2
1
1
+ f (x3 ) 2 x22 x32 + f (x2 ) 2 +
2
2x2
x2
1
2
2
+ 2 + 3.
(5.2)
2 3x2 x2
The behaviour of these functions is illustrated in the following plots of g3 (0, x2 , x3 )

1
5
and g9 (0, 0, 0) = 504
, as
and g9 (0, x2 , x3 ). The values at the origin are g3 (0, 0, 0) = 180
already quoted in Eq. (4.11). Notice that along the diagonals x2 = x3 , both functions tend
asymptotically to zero. This is consistent with the vanishing of the BGZV form factors
as as can be seen from the definitions in Eqs. (4.5) and (4.6). However, if one
argument is set to zero, then the functions may tend to a finite limit. It is easy to check
that this behaviour is entirely determined by the h(x) and h (x) factors in Eq. (4.4). These
1
1
limits are g3 (0, , 0) = g3 (0, 0, ) = 60
, g9 (0, , 0) = 0 and g9 (0, 0, ) = 140
as can
be seen in Figs. 14.
This completes our numerical discussion of the form factors. We confirm that they
are regular at xi = 0 as required for their role in the local effective action. Their values
295
Fig. 1. 3D plots of g3 (0, x2 , x3 ) over different ranges.
Fig. 2. Graphs of g3 (0, x2 , x3 ) along the x2 or x3 axes (left) and the diagonal x2 = x3 (right).
Fig. 3. 3D plots of g9 (0, x2 , x3 ) over different ranges.
at xi = 0 match the well-known results for the DrummondHathrell effective action.

They are smooth functions with well-understood asymptotic behaviour. The local effective
action (4.2) for photongravity interactions induced by vacuum polarization is therefore
under complete algebraic and numerical control to all orders in the derivative expansion.
The precise behaviour of the form factors may now be used to address a variety of
physical applications, notably the important issue of dispersion for photon propagation
in background gravitational fields [12].
296
Fig. 4. Graphs of g9 (0, x2 , x3 ) along the x2 axis (top left), the diagonal x2 = x3 (top right) and the x3 axis
(lower).
Acknowledgements
This research is supported in part by PPARC grant PPA/G/O/2000/00448.
Appendix A
In this appendix, we quote explicit formulae for the third-order form factors appearing
in the heat kernel in the limit discussed in Section 4.
These are expressed in terms of the two basic integrals
1
f (x) =
d e(1)x ,
(A.1)
1
F (0, x2 , x3 ) =
x2 x3
1
d

1 (1)x3
e(1)x2 .
e
(A.2)
We will at times use the further limit

1
F (0, x, 0) =
x
1
d
0
1
1
1 2
1
1 3
1 e(1)x = x +
x
x + .
2 24
360
6720
(A.3)
297
The final expressions also involve

F (0, x2 , x3 )
F (x1 , x2 , x3 )
,
x1
x1 =0

(A.4)
which has a small x expansion

F (0, x, 0) =
1
1 2
1
+
x
x + .
24 360
6720
(A.5)
We also define 2 = (x2 x3 )2 .

It is convenient to recall here the definition of h(x) from Eqs. (3.7), (4.8), viz.

1
1
1
1
1
1
+ 2 f (x) = +
x + .
h(x) = 2 +
6x 2x
4x 2x
30 280
(A.6)
Beginning with the form factors which are not changed in the transformation to the local
basis, we need the following contributions to g1 (0, x2, x3 ). Note that these have been
appropriately symmetrized.
1
F1 (0, x2 , x3 ) = F (0, x2, x3 ),
3

(A.7)

2
2
x2 x3 (x2 + x3 )
2
2

4
1
f (x2 ) f (x3 ) 2 x2 x3 (x2 x3 ) + (x2 + x3 ),
(A.8)
2
2

1
1 2
F6 (x2 , x3 , 0) = F (0, x2 , x3 )
x2 + 4x2 x3 + x32 (x2 + x3 )
62
2

1
2
f (x2 ) f (x3 )
(x2 x3 ) x2 + 6x2x3 + x32
422
1
+
(A.9)
(x2 + x3 ),
22

1
x2 x3 x22 + 4x2 x3 + x32
F7 (0, x2 , x3 ) = F (0, x2 , x3 )
2
32

2
1
2
(x
+
x
)
x
+
34x
x
+
x
+
2
3
2 3
2
3
322

1
1 2 2
x2 + 10x2x3 + x32 2 (x2 + x3 )x2 x3
+ F (0, x2 , x3 )
2
2 2
2

1 f (x2 ) 1 1 4
+
x2 x2 + 44x23x3 10x22x32 36x2 x33 + x34
3
2
x2
2

+ x2 x3

1 f (x3 ) 1 1 4
+
x3 x2 36x3x23 10x32x22 + 44x33 x2 + x34
3
2
x3
2

(A.10)
+ x2 x3 .
F3 (x2 , x3 , 0) = F (0, x2 , x3 )
298
For g4 (0, x2 , x3 ) and g9 (0, x2 , x3 ), we require the following:

F14 (x2 , x3 , 0) = F14 (0, x2 , x3 )

2
1 8
= F (0, x2 , x3 ) (x2 + x3 ) + F (0, x2 , x3 )
2
2 2

4 2

+ f (x2 ) f (x3 ) 2 x2 x32 ,
(A.11)
2

1
F20 (0, x2 , x3 ) = F (0, x2 , x3 ) 2 (x2 + x3 ) x22 + 4x2x3 + x32
32

2
2
2
17x
+
38x
x
+
17x
2 3
2
3
322

1 24
1
F (0, x2 , x3 )
(x2 + x3 ) + 2 (x2 + x3 )2
2 22
2

2
f (x2 ) 3 (x2 x3 )(x2 + x3 ) x22 + 4x2 x3 + x32 + x2 x3
32

f (x2 ) 1 1
(x2 x3 ) 21x23 + 45x22x3 + 15x2 x32 x33
3
x2
2

(A.12)
+ x2 x3 .
For g3 (0, x2 , x3 ) we need

1
(x2 x3 ),
F3 (0, x2 , x3 ) = f (x2 ) f (x3 )
(A.13)
22
while g7 (0, x2 , x3 ) involves
1
4
F17 (0, x2 , x3 ) = F (0, x2 , x3 ) + f (x2 ) + f (x3 ) .
(A.14)
2
2
We have explicitly checked that these form factors are indeed local by evaluating the limits
Fi (0, x, 0) and verifying the cancellation of all the terms with inverse powers of x.
The remaining form factors involve the transformation (2.16) between the original nonlocal BGZV form factors and the new local ones. The first step is to identify the new form
0 (0, x2 , x3 ). We find this from any of the identities in Eq. (2.16). For example
factor F
(remembering that we understand F18 to be symmetrized on x2 , x3 and using the BGZV
expression [13]),
0 (0, x2 , x3 ) = 1 lim x1 F18 (x1 , x2 , x3 ).
F
(A.15)
4 x1 0
0 (0, x2, x3 ), verifying the consistency of our
All the identities produce the same result for F
method. We find

1 1
1 4

+ f (x2 ) 1
(3x2 x3 )
F0 (0, x2 , x3 ) = F (0, x2 , x3 )
2 2
22 x2

1 1
f (x3 ) 1
(A.16)
(x2 3x3 ).
22 x3
299
The remaining form factors are then:

18 (0, x2 , x3 )
F
20
= F (0, x2 , x3 ) 2 (x2 + x3 )2
2

1 80
16
+ F (0, x2 , x3 )
(x2 + x3 ) + F (0, x2 , x3 )
2
2 2
2

f (x2 ) 1 1
3
2
2
3
+ 2
(x2 x3 ) 17x2 + 45x2 x3 + 19x2 x3 x3 + x2 x3 ,
x2
23
(A.17)

F19 (0, x2 , x3 )

8
= F (0, x2 , x3 ) 2 x22 + 3x2 x3 + x32
2

1 40
8
F (0, x2 , x3 )
(x2 + x3 ) F (0, x2, x3 )
2
2 2
2

f (x2 ) 1
+
x2

8
3 x22 (x2 x3 )(3x2 + 7x3)
2

4
12
3
2 2
2
4
+ 4 x2 (x2 + x3 ) x2 x2 x3 + 6x2 x3 4x2 x3 + x3 + x2 x3 ,
2
(A.18)
21 (0, x2 , x3 )
F
16
= F (0, x2, x3 ) 2 (x2 + x3 )(x2 + 4x3)
2

1 64
32
+ F (0, x2 , x3 )
(x2 + 4x3 ) + F (0, x2 , x3 )
2
2 2
2

3
f (x2 ) 1
8
+
(x2 x3 ) 9x2 + 21x22x3 + 11x2x32 x33
3
x2
2

48
2 x2 (x2 + x3 )
2

48
2
f (x3 ) 1
32
2
+
x3 (x2 x3 ) x2 6x2 x3 5x3 2 x3 (x2 + x3 ) ,
x3
23
2
(A.19)
and finally
8 (0, x2 , x3 )
F
40
= F (0, x2 , x3 ) 2 x2 x3 (x2 + x3 )
2

1 160
8
F (0, x2 , x3 )
x2 x3 (x2 + x3 )F (0, x2 , x3 )
2
2 2
2
300

f (x2 ) 1
x2

1
3 (x2 x3 ) 17x24 + 86x23x3 + 72x22x32 22x2x33 + 7x34
2

6
+ 2 (x2 + x3 ) 3x22 + 2x2 x3 x32 + x2 x3 .
(A.20)
2
0,8,18,19,21
Again, using the results above, we have explicitly checked that the form factors F
all have regular small x limits, as required for the local operator basis. The required
numerical cancellations are complicated and provide a very stringent test of the algebraic
forms quoted.
Their remaining finite values when all of x1,2,3 0 reproduce the coefficients in
the DrummondHathrell action. To find these, we need the small x expansions of f (x),
F (0, x, 0) and F (0, x, 0) given in Eqs. (4.9), (A.3), (A.5). We find:

0 (0, 0, 0) = lim 4 F (0, x, 0) 1 + f (x) 1 3 + 1 = 1 ,
F
x0
2
x
2x 12x
72
x2
(A.21)
and similarly

1
F1 (0, 0, 0) = ,
6
F7 (0, 0, 0) = 0,
1
,
180
1
F20 (0, 0, 0) =
,
7560
F17 (0, 0, 0) =
1
,
12
8 (0, 0, 0) = 1 ,
F
180
1

F18 (0, 0, 0) =
,
1260
21 (0, 0, 0) = 1 .
F
630
F3 (0, 0, 0) =
F6 (0, 0, 0) = 0,
1
,
180
19 (0, 0, 0) = 1 ,
F
1260
F14 (0, 0, 0) =
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
I.T. Drummond, S. Hathrell, Phys. Rev. D 22 (1980) 343.

J.S. Schwinger, Phys. Rev. 82 (1951) 664.
B.S. De Witt, Dynamical Theory of Groups and Fields, Gordon and Breach, New York, 1965.
S. Minakshisundaram, J. Indian Math. Soc. 17 (1953) 158.
R.T. Seely, Proc. Sympos. Pure Math. 10 (1967) 288.
P.B. Gilkey, J. Differential Geom. 10 (1975) 601.
I.B. Khriplovich, Phys. Lett. B 346 (1995) 251.
G.M. Shore, Nucl. Phys. B 460 (1996) 379.
A.D. Dolgov, I.D. Novikov, Phys. Lett. B 442 (1998) 82.
G.W. Gibbons, C.A.R. Herdeiro, Phys. Rev. D 63 (2001) 064006.
G.M. Shore, Nucl. Phys. B 605 (2001) 455.
G.M. Shore, gr-qc/0203034.
A.O. Barvinsky, Yu.V. Gusev, V.V. Zhytnikov, G.A. Vilkovisky, Print-93-0274, Manitoba, 1993.
A.O. Barvinsky, G.A. Vilkovisky, Nucl. Phys. B 282 (1987) 163.
I.G. Avramidi, Nucl. Phys. B 355 (1991) 712.
(A.22)

Superbeams plus neutrino factory: the golden path

to leptonic CP violation
J. Burguet-Castell a , M.B. Gavela b , J.J. Gmez-Cadenas a,c ,
P. Hernndez c,1 , O. Mena b
a Departamento de Fsica Atmica y Nuclear and IFIC, Universidad de Valencia, Valencia, Spain
b Departamento de Fsica Terica, Universidad Autnoma de Madrid, Madrid, Spain
c CERN, 1211 Geneva 23, Switzerland
Abstract
Superbeams (SB) and neutrino factories (NF) are not alternative facilities for exploring neutrino
oscillation physics, but successive steps. The correct strategy is to contemplate the combination
of their expected physics results. We show its important potential on the disappearance of fake
degenerate solutions in the simultaneous measurement of 13 and leptonic CP violation. Intrinsic,
sign(m213 ) and 23 degeneracies are shown to be extensively eliminated when the results from one
NF baseline and a SB facility are combined. A key point is the different average neutrino energy
and baseline of the facilities. For values of 13 near its present limit, the short NF baseline, e.g.,
L = 732 km, becomes, after such a combination, a very interesting distance. For smaller 13 , an
intermediate NF baseline of O(3000 km) is still required.
PACS: 12.15.Ff; 14.60.P
1. Introduction
Recent data [1] strongly favour the large mixing angle solution (LMA-MSW) [2] to the
solar neutrino deficit [3]. This, if confirmed, is very good news as regards the prospects
of discovering leptonic CP violation. The measurement of the angle 13 and the CP-odd
E-mail addresses: gavela@delta.ft.uam.es (M.B. Gavela), gomez@hal.ific.uv.es (J.J. Gmez-Cadenas),
pilar.hernandez@cern.ch (P. Hernndez), mena@delta.ft.uam.es (O. Mena).
1 On leave from Departamento de Fsica Terica, Universidad de Valencia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 7 2 - 6
302
J. Burguet-Castell et al. / Nuclear Physics B 646 (2002) 301320
phase would be possible at a superbeam (SB) facility [46] or/and a neutrino factory
(NF) [7], mainly through e and e oscillations, provided 13 is not too small.
How small is one of the points to be quantified below.
The development of a neutrino factory requires, by design, the essentials of a superbeam
facility as an intermediate step. Although the ultimate precision and discovery goals in
neutrino oscillation physics may only be attained with a neutrino factory from muon
storage rings, those for free superbeam results can already lead to significant progress in
central physics issues such as those mentioned above.
Superbeams and neutrino factory are thus not alternative options (see [8] for a recent
comparison of their reach), but successive steps. In this perspective, the correct analysis
strategy is to contemplate the combination of their expected physics results. We will take
such a step here and show its important impact, in particular on the disappearance of fake
solutions in the simultaneous measurement of 13 and .
It was shown in Ref. [9] that there exists generically, at a given (anti)neutrino energy and
, ) that gives the same oscillation probabilities
fixed baseline, a second value of the set (13
for neutrinos and antineutrinos as the true value chosen by nature. In what follows, these
fake solutions will be dubbed intrinsic degeneracies.
More recently it has been pointed out [10,11] that other fake solutions might appear
from unresolved degeneracies in two other oscillation parameters:
the sign of m213 ,
23 , upon the exchange 23 /2 23 for 23 = /4.
It is not expected that these degeneracies will be resolved before the time of the SB/NF
operation. However, the subleading transitions e , from which the parameters 13
and can be measured, are sensitive to these discrete ambiguities. A complete analysis of
the sensitivity to the set (13 , ) should therefore assume that sign(m213) can be either
positive or negative and 23 > or < /4. If a wrong choice of these possibilities cannot fit
the data, the ambiguities will be resolved, else they will generically give rise to new fake
solutions for the parameters 13 and .
Strategies to eliminate some of the fake solutions have previously been discussed,
advocating the combination of different baselines [9], an improved experimental technique
allowing the measurement of the neutrino energy with good precision [8,12], the
supplementary detection of e channels [13] and a cluster of detectors at a superbeam
facility located at different off-axis angles, so as to have different E
[14].
We consider the three types of degeneracies as they would appear in the analysis
from the data taken at a neutrino factory and at an associated superbeam. We then show
the potential of combining their results. For the NF we consider the experimental setup
presented in [9,15,16]. The parent energy is 50 GeV and the reference baselines
considered will be 732 and 2810 km. As the SB facility we consider the design proposed
for the CERN SPL [17], which could be the initial step of a NF based at CERN. The
average energy is E
= 0.25 GeV and the baseline is L = 130 km (CERNFrjus). The
fluxes and detector systematics have been discussed in [6]. In the detailed computations
of this paper we will not consider other types of superbeams, such as JHF [18], nor the
303
case of beta-beams [19], although we will also discuss the potential of these alternative
experimental setups in this context.
Whenever it is not specified otherwise, we take the following central values for the oscillation parameters: sin 212 m212 = 104 eV2 with m212 > 0, |m213| = 3 103 eV2
and sin 223 = 1. Some implications of making them vary within their currently allowed
ranges will be discussed as well. The errors expected in the knowledge of these parameters,
as well as the error in the matter profile of the Earth will not be included. We previously
studied their impact in Ref. [9] and they are not expected to change the conclusions
significantly.
2. Degenerate solutions
In this study we will consider in detail the measurement of the subleading probabilities
Pe ( e ) in the NF complex, through the detection of wrong-sign muons, and
P e ( e ) in the SB facility, through the detection of electrons/positrons.
For the energies and baselines under discussion, the oscillation probabilities in vacuum
are accurately given by (for more details see [16]):

2
m212L 2
2
2
2 m13 L
2
2
+ c23 sin 212
Pe ( e ) = s23 sin 213 sin
4E
4E

2
2
2
m13 L
m13L m12L
sin
,
+ J cos
(1)
4E
4E
4E
where J is defined as
J cos 13 sin 213 sin 223 sin 212.
(2)
The vacuum approximation should be excellent as regards the SB scenario with a baseline
of hundreds of kilometers, while in practice it also gives a good indication for the results at
the short (732 km) and intermediate (20003000 km) baselines of a NF. Of course all the
fits below include the exact formulae for the probabilities, including matter effects.
As in Ref. [9], we will denote the three terms in Eq. (1), atmospheric, solar and
atm
sol and P inter , respectively. When is relatively large or |m2 |
interference, by P(
13
12
) , P
( )
small, the probability is dominated by the atmospheric term. We will refer to this situation
as the atmospheric regime. Conversely, when 13 is very small or |m212| large (with
atm
respect to E
/L and |m213|), the solar term dominates P sol P(
) . This is the solar
regime. The interference term, which contains the information on the CP-violating phase
, can never dominate since
inter
atm
sol
P

(3)
( ) P( ) + P .
It is precisely in the transition between the two regimes where the interference term
becomes larger in relative terms. There the corresponding CP-odd asymmetry becomes
maximal and independent of the precise value of the two small parameters: m212 and 13
[9]. As an indication, for the solar parameters sin 212 m212 = 104 eV2 , the transition
is at 13 1 for the NF setups considered here and 13 2 for the SPL-SB facility.
304
3. Intrinsic degeneracies
In a previous work [9] we uncovered, at fixed neutrino energy and baseline, the existence
of degenerate solutions in the (13 , ) plane for fixed values of the oscillation probabilities
e (e ) ( ). If (13 , ) are the values chosen by nature, the conditions

, ) = P
Pe (13
e (13 , )
(4)
,
, ) = P
P e (13
e (13 , )
, ). Using the approximate formulae of
can be generically satisfied by another set (13
Eq. (1), it is easy to find the expression for these intrinsic degeneracies deep in the
atmospheric and solar regimes. In Ref. [9] the general results including matter effects were
presented. Here we just recall the solutions in vacuum, which are simpler and accurately
describe the new situation considered here, that of the superbeams.
For 13 sufficiently large and in the vacuum approximation, apart from the true solution,
= , there is a fake one at
= and 13
13
,

13

m213L
m212L
cot 23 cot
13 + cos sin 212
.
4E
4E
(5)
Note that for the values = 90 , 90 , the two solutions degenerate into one. Typically
cot(m213 L/4E) has on average opposite signs for the proposed SB and the NF setups,2
for m213 = 0.003 eV2 (see Table 1):
When 13 0 and in the vacuum approximation, the intrinsic degeneracy is independent3 of :

m2 L
if cot 4E13 > 0, then
m2 L
if cot 4E13 < 0, then 0

m213 L
m212L

.
cot
13 sin 212
(6)
cot
23

4E
4E
This solution was named 13 = 0-mimicking solution and occurs because there is a
for which there is an exact cancellation of the atmospheric and interference
value of 13
terms in both the neutrino and antineutrino probabilities simultaneously, with sin = 0.
Fig. 1 shows the results of measuring (13 , ) at the SPL-SB facility, for 13 = 8 and the
central values of = 180, 90 , 90, 180 . The intrinsic degeneracies clearly appear
and are well described by Eq. (5). The details of the analysis can be found in Ref. [6]. We
simply stress here that the analysis is based on the total number of electron/positron events,
so we do not assume that the neutrino energy can be reconstructed. Our results about the
NF can be found in Ref. [9], where it was shown that these intrinsic degeneracies also
2 Clearly the parameters for these setups are not fixed yet and might be modified conveniently in the final
designs.
3 All throughout this paper we will only consider those fake solutions which lie inside the experimentally
allowed range for 13 .
305
Table 1
SB-SPL
JHF-off-axis
NF@732
NF@2810
-beam
E
(GeV)
L (km)
cot(m213 L/4E)
0.25
0.7
30
30
0.35
130
295
732
2810
130
0.43
0.03
+10.7
+2.68
+0.17
Fig. 1. Fits to given true (nature) solutions and their intrinsic degenerate solutions at a SB facility. The 68.5%,
90% and 99% contours are depicted, for four central values of = 90 , 0 , 90 , 180 and for 13 = 8 .
could not be resolved in a single baseline, in spite of the spectral information, although
they did get eliminated when two NF baselines (intermediate and long) were combined.4
A comparison of the NF and SPL-SB fits shows that the displacement of the fake solution
with respect to the true one is opposite for the two facilities.
In order to understand the intermediate region between the solar and atmospheric
regimes, as well as the influence of matter effects, we have determined numerically all the
possible physical solutions to Eq. (4), using the approximate formulae for the probabilities
including matter effects [16]. L and E are fixed to the average values for the different
and are shown in Fig. 2 as a function of ,
facilities. The results for the shift 13
13
13
for two values of = 0 , 90 and for the different experimental setups. In the whole range
4 The authors of [8] did not find intrinsic degeneracies in their simulations by assuming a very optimistic lower
cut in the momentum of the muon.
306
(left) and (right) versus , for the intrinsic fake solution, for fixed values of = 0 (up) and
Fig. 2. 13
13
13
= 90 (down).
of parameters we find two solutions, as expected by periodicity in , since one solution is

warranted: the true one.
The most important point to note in Eqs. (5) and (6) and in Fig. 2 is that the position

(measured in 13
13 or ) of the degenerate solution is very different in the NF, the
5
SPL-SB and JHF setups. As a result, we expect that any combination of the results of

two of these three facilities could in principle exclude the fake solutions. The 13
13
of the fake solution depends strongly on the baseline and the neutrino energy through the
ratio L/E, so the combination of the results of two experiments with a different value for
this ratio should be able to resolve these degeneracies, within their range of sensitivity.
Even more important is that, for small 13 , may differ by 180 if the two facilities have
opposite sign for cot(m213L/4E), see Eq. (6). For the NF setups, this sign is clearly
positive, since the measurement of CP violation requires, because of the large matter
effects, a baseline considerably shorter than that corresponding to the maximum of the
atmospheric oscillation (in vacuum), where the cotangent changes sign. In the SB scenario
on the other hand, because of the smaller E
, matter effects are small at the maximum
of the atmospheric oscillation, which then becomes the optimal baseline for CP violation
5 For the JHF scenario the shift in is minimal because for the reference parameters E
= 0.7 GeV and
13
L = 295 km, cot(m213 L/4E) 0.
307
versus |m2 | for the intrinsic fake solution in the atmospheric regime and = 0 .
Fig. 3. 13
13
13
studies. It is then not very difficult to ensure that cot(m213L/4E) be dominantly negative
in this case,6 which results in an optimal complementarity of the two facilities in resolving
degeneracies.
Clearly the position of the fake solution is very sensitive to the atmospheric |m213|.
In matter we expect a milder dependence, especially if matter effects become dominant.
In Fig. 3 we show the separation in 13 of the intrinsic degenerate solution at = 0
in the atmospheric regime as a function of |m213|. Although in general the separation
becomes smaller for smaller |m213|, it is sizeable in the whole allowed range. The relative
difference between the results for the NF facility and the SPL superbeam option is always
largest, although the differences between the two superbeams and that between the NF and
, which is related to that of
JHF are also very large. Note also that the sign of 13
13
cot(m213 L/4E), is positive in all the domain for the NF baselines and negative in most of
the domain for SPL-SB scenario, which implies that the difference in between the two
facilities is 180 for small 13 . For JHF, it is negative only for |m213 | 0.003 eV2 .
Concerning the dependence on the solar parameters, it enters only through the

combination sin 212 (m212 L/4E). In general 13
13 is linear in this quantity, so
degenerate solutions become closer with smaller m212 and also closer to the true solution.
Note however that in the solar regime does not depend on the solar parameters and that it
differs by 180 in the two facilities, and this separation will remain when m212 is lowered.
6 Note however that the neutrino beams are generally broad so it is necessary for this argument to hold that
most of the events have a parent neutrino energy giving the appropriate sign. The results of the fits indicate that
this is the case in the two facilities (NF and SPL-SB) that are considered in detail.
308
Turning to the variation of the solar parameters while in the atmospheric regime, we will
now argue that, if the two facilities that are combined have opposite sign(cot(m213 L/4E)),
the effect of lowering m212 is not dramatic either in the resolution of degeneracies. The
statistical error on the measurement of 13 and is mainly independent of the solar parameters (it is dominated by the atmospheric term), which means that at some point when
m212 is lowered, the degenerate solutions of the two facilities will merge, since the error
remains constant while the separation of the solutions gets smaller. However, because of
, the solutions of the two facilities will merge only when
the opposite sign of 13
13
they merge with the true solution in 13 . If this happens, it would therefore not bias the
measurement of 13 and .
We have performed a detailed combined analysis of the NF results [9,16] and those
from the SPL-SB [6] facility. The combination with the optimal NF baseline L = 2810 km
is sufficient to get rid of all the fake solutions, as shown in Fig. 4 (left). Note that indeed
the disappearance of the fake solutions takes place even in the solar regime!
One very interesting exercise is to reconsider the combination of the SPL-SB and the NF
results at a shorter baseline of L = 732 km. As explained in Refs. [9,16], the degenerate
solution is not so relevant to this NF baseline when considered alone, because there the
sensitivity to CP violation is so poor that there exists a continuum of almost degenerate
solutions, which makes the determination of impossible with the wrong-sign muon
signals. The combination of the results from this NF baseline with those from the SPLSB facility is summarized in the tantalizing plots in Fig. 4 (right). Not only do the fake
solutions corresponding to the intrinsic degeneracies in the superbeam disappear, but the
accuracy in the determination of the true solution becomes competitive with that obtained
in the combination with the optimal baseline for large values of 13 . At small values of 13
the latter still gives better results, as expected.
4. sign(m213 ) degeneracy
Assume nature has chosen a given sign for m213 , while the data analysis is performed
assuming the opposite sign. Let us call Pe ( e ) (13, ) the oscillation probability with
, ), at fixed E and
the sign of m213 reversed. We may then get new fake solutions (13
L, if the equations

, ) = P
Pe (13
e (13 , )
(7)
,
, ) = P
Pe (13
e (13 , )
have solutions in the allowed physical range.

It turns out that there are generically two fake sign solutions to Eq. (7). It is very easy
to find them in the vacuum approximation, as they mirror the two solutions (true and fake)
obtained in the analysis of the intrinsic degeneracies. It can be seen in Eq. (1) that a change
in the sign of m213 can be traded in vacuum by the substitution [10,20],
implying then for Eq. (7)

Pe ( e ) (13
, ) Pe ( e ) (13
, ),
(8)
309
Fig. 4. Fits combining the results from the SPL-SB facility and a NF baseline at L = 2810 km (left) or L = 732
km (right). The true values illustrated correspond to = 90 , 0 , 90 , 180 and 13 = 8 (top) or 13 = 0.6
(bottom). Notice that the fake intrinsic solutions have completely disappeared in the combinations.
in the vacuum approximation. Consequently, the solutions in vacuum can be obtained from
those present for the intrinsic case, upon the substitution . One of them mirrors
the true (nature) solution and will be called below solution I, given in vacuum by
,

13 .
13
(9)
The fact that it is approximately E- and L-independent suggests that it will be hard
to eliminate it by exploiting the L-, E-dependence of different facilities, as indeed is
310
confirmed by the fits below. Fortunately, this fake solution does not interfere significantly
with the determination of 13 or CP-violation (i.e., sin ).
The second fake sign solution, which we will call solution II, can be read in vacuum
from Eqs. (5) and (6), upon the mentioned exchange. It is strongly L- and
E-dependent. Both solutions I and II can be nicely seen in the numerical analysis for the
SPL-SB in Fig. 5 (left), for 13 = 8 and positive sign(m213 ). All the plots below also
correspond to a true positive sign, which turns out to be the most pessimistic situation.
Matter effects are obviously very important in resolving fake sign solutions [21]: the
task should thus be easier at large 13 and large enough NF baselines, where matter effects
are largest.7 In fact it is easy to prove that no solutions can remain for large enough 13 .
This can be seen in Fig. 6, which show the fake sign solutions as they result from solving
numerically Eq. (7) (using the approximate probabilities with matter effects included [16])
for the different experiments. For small 13 the two solutions I and II exist in all cases,
while for large 13 they degenerate and disappear because of matter effects. One should
keep in mind, though, that even if no fake solution exists, there might be approximate ones
that will show up in a measurement with finite errors.
Our fits including realistic background errors and efficiencies confirm the above
expectations, at each given facility. To be more precise, we have found no fake sign solution
for values of 13 > 2 , when considering just one NF baseline of L = 2810 km (or longer),
while for 2 > 13 > 1 they do appear but get eliminated when the data are combined
with those from the SPL-SB. At L = 732 km some fake sign solutions remain close to
the present experimental limit for 13 , as shown in Fig. 5 (right). It should be noticed that,
again, in the combination of these latter data with those from the SPL-SB facility, all fake
sign solutions disappear for large 13 4 , and the sign of m213 could thus be determined
from it.
Fig. 6 also illustrate that solution I is more facility-independent than solution II, as
argued above. The solutions that survive in the combinations for small 13 are indeed of
type I, as shown in Fig. 7.8
In conclusion, in the LMA-SMW regime, the sign of m213 can be determined from data
at an intermediate or long NF baseline alone for 13 well inside the atmospheric regime.
For the larger values of 13 , the combination of data from the SB facility and a L = 732 km
NF baseline also results in no fake sign solutions.
With lowering 13 (13 > 1 for our central parameters), the sign can still be determined
through the combination of SB and NF data at the intermediate or long distance.
Finally, for the range 13 < 1 , the sign cannot be determined, but the combination of
data from the SB facility and an intermediate (or long) NF baseline is still important to
reduce the fake solutions to those of type I, which do not interfere significantly with the
determination of 13 and .
7 We have checked that for the sign degeneracies (and only for these), the combined results from a SB facility
and a NF baseline at L = 7332 km are competitive or even superior to those obtained in the combination with an
intermediate NF baseline.
8 In the same exercise, but with the opposite sign of m2 , which leads to larger statistics, solution II
13
disappears completely.
311
Fig. 5. Fits for central values 13 = 8 and = 90 , 0 , 90 , 180 for the SPL-SB (left) and NF at L = 732 km
(right). Natures sign for m223 is assumed to be positive, while the fits have been performed with the opposite
sign. All fake solutions disappear when the two sets of data are combined.
(left) and (right) for the sign degeneracies as functions of for fixed values of = 0 (up)
Fig. 6. 13
13
13
and = 90 (down). Solutions I and II are described in the text. Natures sign for m223 is assumed to be positive,
while the fits have been performed with the opposite sign.
312
Fig. 7. Fits resulting in fake sign solutions, for central values 13 = 0.6 and = 90 , 0 , 90 , or 180 . The
nature sign for m213 is positive, while the fits have been performed with the opposite sign. The results from an
NF baseline at L = 2810 km can be appreciated on the left, while their combination with data from the SPL-SB
facility can be seen on the right.
Concerning the dependence on the solar parameters, we do not expect that the
conclusions will change very much with lower sin 212 m212 . The argument for solutions
of type II parallels that given in the previous section for the intrinsic fake solution, while the
existence and position of the type I solutions is pretty insensitive to the solar parameters.
5. 23 /2 23 degeneracy
The present atmospheric data indicate that 23 is close to maximal, although not
necessarily 45 . Super-Kamiokande results [22] give 90% C.L.-allowed parameter regions
for sin2 223 > 0.88, translating into the allowed range 35 < 23 < 55 . Therefore even if
the value of sin2 223 is determined with great accuracy in disappearance measurements,
there may remain a discrete ambiguity under the interchange 23 /2 23 . If this 23
ambiguity is not cleared up by the time of the NF operation, supplementary fake solutions
[11] may appear when extracting 13 and , when the wrong choice of octant is taken for
23 . Fake solutions follow from solving the system of equations, for fixed L and E :

, ) = P
Pe (13
e (13 , )
(10)
,

, ) = P
P e (13
e (13 , )
where Pe denotes the oscillation probabilities upon the exchange 23 /2 23 .

It turns out that, within the allowed range for the parameters, there are generically
two solutions to these equations. They should converge towards the true solution and its
intrinsic degeneracy, in the limit 23 /4. We will thus denote again solution I that
which mirrors natures choice and solution II that which mirrors the intrinsic degeneracy.
Because of this parenthood, solution I is a priori expected to present generically less L- and
E-dependence than solution II, and be thus more difficult to eliminate in the combination.
313
It is easy and simple to obtain the analytical form of the fake degeneracies in the vacuum
approximation, in which, from Eq. (1) we get

m213L
m212 L

2

2
+ s23
, ) = c23
sin2 213
sin2
sin2 212 sin2
Pe ( e ) (13
4E
4E

2
2
2
m13 L
m13 L m12 L
sin
+ J cos
(11)
.
4E
4E
4E
Let us consider in turn the atmospheric and solar regimes. For large 13 , fake 23
solutions are given by
sin cot 23 sin ,

13
tan 23 13
+
sin 212(m212L/4E)
2 sin(m213 L/4E)

m213 L
m213 L
cos
tan 23 cos
.
4E
4E
(12)
This system describes two solutions. For one of them (I) the L- and E-dependent terms
= , = in this limit. The
in Eq. (12) tend to cancel for 23 /4, resulting in 13
13
other solution (II) coincides in this limit with that for the intrinsic degeneracy, Eq. (5), as

expected. For both fake 23 solutions, deep in the atmospheric regime the shift 13
13
is positive (negative) for 23 > (<) /4. Note also that, from Eq. (12), no fake solutions
are expected for | cot 23 sin | > 1. In the plots of Figs. 8 and 10, we show the solutions
to Eq. (10), including matter effects, for 23 at the two extremes of the 90% C.L.-allowed
interval. Note that for large 13 there is one solution (I) that is more facility-independent
than the other, although the E-, L-dependence is sizeable for both solutions (see, for
instance, the curves for = 90 in Fig. 10) when 23 is so far from maximal.
We have performed fits with the wrong choice of octant for 23 and central values of 23
at the limit of the currently allowed domains. The results confirm the expectations above
and indicate a situation close to that for the fake sign degeneracies, albeit slightly more
difficult. For instance, at the L = 2810 km baseline of the NF alone, still some fake 23
solutions remain down to 13 > 2 , but again they all disappear when combined with the
SPL-SB data. As an illustration, in Fig. 9 we show the results for 23 = 35 and 13 = 4 ,
at the SPL-SB facility (left) and the L = 2810 km NF baseline (right). The same exercise,
but for an L = 732 km baseline of the NF, results in the elimination of the 23 degeneracies
only for 13 8 .
Let us now turn to the study of the solar regime. For 13 0 , there are again two fake
solutions if the following condition is met:
tan2 23 <
1
2
sin (m213L/4E)
(13)
Otherwise no solution exists. This is important for the larger possible values of 23 and
well reflected in Fig. 10, which show the exact solutions for 23 = 55 . Indeed no fake 23
degeneracies appear in the SB facilities in this case, for 13 in the solar regime.
314
(left) and (right) for the fake solution as functions of , for = 35 , for fixed values
Fig. 8. 13
13
23
13
23
of = 0 (up) and = 90 (down).
Fig. 9. Fake solutions due to 23 degeneracies for SPL-SB results (left) and a L = 2810 km NF baseline (right),
for 23 = 35 , 13 = 4 and = 90 , 0 , 90 , 180 . The combination of the results from both experiments
resolves the degeneracies.
315
(left) and (right) for the intrinsic fake solution as a function of , for = 55 , for fixed
Fig. 10. 13
13
13
23
values of = 0 (up) and = 90 (down).
For 13 0 , Eq. (10) can be easily solved to first order in 23 tan 23 1. Solution
I becomes in this limit:

m2 L
if cos 223 cot 4E13 > 0, then 0
m2 L
if cos 223 cot 4E13 < 0, then

m213L
m212L

13
(14)
sin 212
csc
23
.
4E
2E
Similarly, solution II for 13 0 is given by:

m2 L
if cot 4E13 > 0, then
m2 L
if cot 4E13 < 0, then 0

m213 L
m212L m213L

cot

13
sin 212
cot
,
23

4E
4E
2E
(15)
where the sign corresponds to the sign(cot(m213 L/4E)). The intrinsic degeneracy,
Eq. (6), is recovered for 23 = 45 . Note that, in the solar regime both fake 23 solutions
have a sizeable L-, E-dependence, when 23 is far from maximal. These two solutions can
316
Fig. 11. As Fig. 7 but for the case of 23 degeneracies.
be seen in Figs. 8 and 10 for small 13 . Only for the NF setups do solutions I and II remain
on the same curve in the solar and atmospheric regimes. In the case of the SPL and JHF
facilities, they are mixed.
Fig. 11 show the fits for 13 = 0.6 , for a NF at L = 2810 km (left) as well as the
same combined with the results from the SPL-SB facility (right): only one annoying fake
solution remains in the latter, which results from the merging of solution I for SB and
solution II for the NF, owing to the finite resolution. See Eqs. (14) and (15).
In general we have found that the NF and SPL-SB combination brings an enormous
improvement to the solution of these fake degeneracies, particularly for large 13 . The
conclusions are rather parallel to those for the fake sign(m213 ) solutions, with the caveat
that for the 23 ambiguities, solution I, which is harder to resolve, is not that close to
satisfying sin = sin , and it is thus potentially more harmful to the measurement of CP
violation.
As regards the dependence on the solar parameters, the arguments of the previous two
sections can be repeated for solutions I and II, when 23 is close to maximal. When 23 is
farther from /4, the situation is more confusing since both solutions have a dependence
on the solar parameters and a detailed exploration of the whole LMA parameter space is
necessary.
6. The silver channels

One possibility that can help very much to remove degeneracies further is to measure
also the e and e transition probabilities. The relevance of these silver
channels in reducing intrinsic degeneracies was studied in Ref. [13], in the atmospheric
regime. Consider the approximate oscillation probabilities [13,16,23] in vacuum for e
( e ):

m213L
m212 L 2
2
2
+ s23 sin 212
4E
4E

2
2
2
m13L
m13 L m12 L
sin
.
J cos
4E
4E
4E
317
2
Pe ( e ) = c23
sin2 213 sin2
(16)
They differ from those in Eq. (1) by the interchange 23 /2 23 and by a change in
the sign of the interference term.
For the intrinsic degeneracies in the atmospheric regime, it follows that the sign of

13
13 will be opposite to that for the golden e ( e ) channels given in
Eq. (5). In the solar regime, the intrinsic solutions in these silver channels will thus be
identical to Eq. (6) upon exchanging = 0 and , and the combination of the golden and
silver channels remains a promising option.
Let us now turn to the fake 23 solutions. When considering only e and
e oscillations, the location of the fake solutions related to the 23 ambiguity, in
the atmospheric regime, is:
sin tan 23 sin ,

13
cot 23 13
sin 212
m212L/4E
2 sin(m213L/4E)

m213 L
m213 L

cos
cot 23 cos
.
4E
4E
(17)
at large would have the opposite sign to that in Eq. (12).

Thus the shift 13
13
13
In the solar regime, on the other hand, solution I for the appearance measurement is
the same as that in Eq. (14), while solution II is different, namely:

m2 L
if cot 4E13 > 0, then 0
m2 L
if cot 4E13 < 0, then

m213 L
m212L m213L

13 sin 212
(18)
.
cot 4E 23 cot 2E
4E
The condition for the existence of solutions in the solar regime is also different:
cot2 23 <
1
2
sin (m213L/4E)
(19)
A detailed analysis for a realistic experimental setup will be done elsewhere [24], but we
expect that the combination of the two appearance measurements: e and e
for both polarities can help to resolve the dangerous solution I associated with the 23
ambiguity, for 13 in the atmospheric regime.
Finally, we recall that the disappearance measurements (e.g., ) should also be
helpful in reducing these ambiguities for large 23 . Obviously, if the angle 23 will turn out
to be close to maximal (as the best-fit point now indicates), the 23 degeneracies will be of
very little relevance.
318
As for the removal of the fake sign(m213 ) degeneracies, the silver channels will also
help, for qualitatively the same reason as in the combination of facilities with opposite
value of cot(m213L/4E). For maximal 23 , the solution of type I in the silver channel is
the same in vacuum as that in the golden channel, and it is thus not expected to disappear
in the combination of the two appearance measurements. The solution of type II, instead,
has an opposite displacement in 13 in the atmospheric regime and a difference of 180 in
the phase in the solar one.
7. Conclusions
The extraction of a given set of nature values (13, ) from the detection of neutrino
oscillations through the golden channels e ( e ) results generically in that the
true solution may come out accompanied by fake ones, which might interfere severely
with the measurement of CP violation. One of the fake solutions comes from the intrinsic
correlation between and 13 . The others come from the discrete ambiguities: (sign m213 )
and sign(cos 223).
We have shown the enormous potential of combining the data from a superbeam facility
and a neutrino factory, to eliminate these degeneracies. Because of the sizeable matter
effects, neutrino factory baselines that are optimal to measure CP violation (as well as
shorter ones), imply a considerably smaller ratio L/E
than in the proposed superbeam
facilities. It turns out that the location of the fake solutions is very sensitive to this quantity,
hence the potential of combining the results from both type of facilities.
We have shown that the fake solutions associated with the sign and 23 ambiguities
can be grouped in two sets: those closer to natures values (solutions of type I) and those
related to the intrinsic fake solution (solutions II). Generically solutions I show a milder
L/E-dependence and are thus more difficult to eliminate through this strategy. We have
studied all fake solutions both analytically and through simulations, including realistic
background errors and efficiencies for a magnetized iron detector at a neutrino factory, and
a water Cerenkov one at the proposed SPL superbeam facility.
The fits have been performed assuming the LMA-MSW solar solution, with sin2 212
m212 = 104 eV2 . For 13 near its present limit, the combination of the SPL-SB data and
those from a short NF baseline, i.e., L = 732 km, is sufficient to resolve all of them and
deliver a clean measurement of 13 and leptonic CP violation. With lowering 13 but still
in the atmospheric regime, although the same setup often produces interesting results, it
is necessary to consider an intermediate NF baseline, i.e., L = 2810 km, together with the
superbeam. In particular the sign of m213 can be measured from the combination of their
data down to 13 > 1 .
For values of 0.5 < 13 < 1 most degeneracies still disappear in the combined data
from the SPL-SB facility and the L = 2810 km NF baseline, but some fake solutions
remain, mainly of type I. While those associated to the sign(m213L/4E) ambiguity bias
only slightly the extraction of the true 13 and values, those related to 23 would remain
a problem, if 23 were far from maximal.
A simultaneous error on the assumed sign(m213 ) and 23 octant, of course, gives
rise to additional combined fake solutions: we have checked, for our central values of
319
the oscillation parameters, that those get resolved when the corresponding individual
degeneracies get resolved. Besides, although we have not done a systematic exploration of
the presently allowed range for the atmospheric and solar parameters, we have argued that
we expect conclusions similar to those obtained in this work, for lower values of m213
and m212 , to the extent that the sign of cot(m213 L/4E) remains opposite in the two
facilities.
It has previously been pointed out [13] that a supplementary measurement of the silver
channels, i.e., e ( e ), could help in removing the intrinsic degeneracy. We
have also discussed in this paper the expected impact of such measurements on resolving
the fake sign and 23 degeneracies. Although a detailed analysis will be done elsewhere,
we expect a big improvement in eliminating in particular the dangerous fake solutions
associated with the 23 octant ambiguity.
Superbeams and Neutrino Factory are two successive steps in the same path towards the
discovery of leptonic CP violation: a golden path, not so much for its budgetary cost, but
for the solid and shining perspective offered by the combination of their physics results.
Acknowledgements
We are indebted to Andrea Donini and Stefano Rigolin for useful discussions and
comments. The work of M.B. Gavela and J.J. Gmez-Cadenas was partially supported by
CICYT projects FPA 2000-0980 (GCPA640), and SPA2001-1910-C03-02, respectively.
That of O. Mena and P. Hernndez was partially supported by AEN-99/0692.
References
[1] SNO Collaboration, nucl-ex/0204008;
SNO Collaboration, nucl-ex/0204009.
[2] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369;
L. Wolfenstein, Phys. Rev. D 20 (1979) 2634;
S.P. Mikheyev, A.Yu. Smirnov, Sov. J. Nucl. Phys. 42 (1986) 913.
[3] For recent analyzes see:
V. Barger, et al., Phys. Lett. B 537 (2002) 186, hep-ph/0204253;
J. Bachall, M.C. Gonzlez-Garca, C. Pea-Garay, JHEP 0207 (2002) 054, hep-ph/0204314;
A. Strumia, C. Cattadori, Phys. Lett. B 541 (2002) 327, hep-ph/0205261;
G.L. Fogli, et al., hep-ph/0208026.
[4] B. Richter, hep-ph/0008222;
H. Minakata, H. Nunokawa, Phys. Lett. B 495 (2000) 369, hep-ph/0004114;
K. Dick, et al., Nucl. Phys. B 598 (2001) 543, hep-ph/0008016;
V. Barger, et al., Phys. Rev. D 63 (2001) 113011, hep-ph/0012017.
[5] M. Mezzetto, CERN-NUFACT-NOTE-60.
[6] J.J. Gmez-Cadenas, et al., hep-ph/0105297.
[7] S. Geer, Phys. Rev. D 57 (1998) 6989;
S. Geer, Phys. Rev. D 59 (1998) 039903, Erratum, hep-ph/9712290;
A. De Rjula, M.B. Gavela, P. Hernndez, Nucl. Phys. B 547 (1999) 21, hep-ph/9811390;
For a comprehensive set of further references and a recent review, see:
O. Yasuda, hep-ph/0111172;
J.J. Gmez-Cadenas, D. Harris, FERMILAB-PUB-02-044-T, Annu. Rev. Mod. Phys., in press.
320
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
P. Huber, M. Lindner, W. Winter, hep-ph/0204352.

J. Burguet-Castell, et al., Nucl. Phys. B 608 (2001) 301, hep-ph/0103258.
H. Minakata, H. Nunokawa, JHEP 0110 (2001) 1, hep-ph/0108085.
V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023, hep-ph/0112119.
M. Freund, P. Huber, M. Lindner, Nucl. Phys. B 615 (2001) 331, hep-ph/0105071.
A. Donini, D. Meloni, P. Migliozzi, hep-ph/0206034.
V. Barger, D. Marfatia, K. Whisnant, hep-ph/0206038.
A. Cervera, F. Dydak, J.J. Gmez-Cadenas, Nucl. Instrum. Methods A 451 (2000) 123.
A. Cervera, et al., Nucl. Phys. B 579 (2000) 17;
A. Cervera, et al., Nucl. Phys. B 593 (2001) 731, Erratum.
A. Blondel, M. Donega, S. Gilardoni, CERN-NUFACT-NOTE-078.
Y. Itow, et al., hep-ex/0106019.
P. Zucchelli, hep-ex/0107006.
H. Minakata, H. Nunokawa, S. Parke, hep-ph/0204171.
V. Barger, et al., Phys. Lett. B 485 (2000) 379.
T. Toshito, hep-ex/0105023.
M. Freund, Phys. Rev. D 64 (2001) 053003;
K. Kimura, A. Takamura, H. Yokomakura, hep-ph/0203099;
K. Kimura, A. Takamura, H. Yokomakura, hep-ph/0205295.
A. Donini, et al., private communication.

The silver channel at the neutrino factory

A. Donini a , D. Meloni b , P. Migliozzi c
a INFN, Sezione di Roma I and Dip. Fisica, Universit di Roma La Sapienza, P. le A. Moro 2,
I-00185 Rome, Italy

b Dip. Fisica, Universit di Roma La Sapienza and INFN, Sezione di Roma I, P. le A. Moro 2,
I-00185 Rome, Italy

c INFN, Sezione di Napoli, Complesso Universitario di Monte SantAngelo, Via Cintia ed. G,
I-80126 Naples, Italy

Abstract
We notice that looking for e at the same time as e oscillations could significantly
help to reduce the errors in the leptonic CP-violating phase measurement. We show how the
e (golden) and e (silver) transitions observed at an OPERA-like 2 kton leademulsion detector at L = 732 km, in combination with the e transitions observed at a 40 kton
magnetized iron detector with a baseline of L = 3000 km, strongly reduce the so-called (13 , )
ambiguity. We also show how a moderate increase in the OPERA-like detector mass (4 kton instead
of 2 kton) completely eliminates the clone regions even for small values of 13 .
1. Introduction
The present atmospheric [16] and solar [714] neutrino data are strongly supporting
the hypothesis of neutrino oscillations [1518] and can be easily accommodated in a three
family mixing scenario.
Let the PontecorvoMakiNakagawaSakata UPMNS matrix be the leptonic analogue
of the hadronic CabibboKobayashiMaskawa (CKM) mixing matrix in its most conven-
E-mail addresses: andrea.donini@roma1.infn.it (A. Donini), davide.meloni@roma1.infn.it (D. Meloni),

pasquale.migliozzi@cern.ch (P. Migliozzi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 9 4 - 5
322
A. Donini et al. / Nuclear Physics B 646 (2002) 321349
tional parametrization [19]:

1
0
0
c13
UPMNS = 0 c23 s23
0
s13 ei
0 s23 c23
0
1
0
s13 ei
0
c13
c12
s12
0
s12
c12
0

0
0 ,
1
with the short-form notation sij sin ij , cij cos ij . Oscillation experiments are
sensitive to the two neutrino mass differences m212 , m223 and to the four parameters
in the mixing matrix: three angles and the Dirac CP-violating phase, .
In particular, data on atmospheric neutrinos are interpreted as oscillations of muon
neutrinos into neutrinos that are not e s, with a mass gap that we denote by m223 . The
corresponding mixing angle is close to maximal, sin2 223 > 0.8, and |m223| is in the
range 1.8 to 4 103 eV2 [20].
The recent SNO results for solar neutrinos [1214] favour the LMA-MSW [21] solution
of the solar neutrino deficit with e oscillations into active ( , ) neutrino states. The
corresponding squared mass difference, that in this parametrization should be identified
with m212 , is 105 104 eV2 . Comprehensive analyses of the solar neutrino data,
however, do not exclude the LOW-MSW solution [2225], with m212 107 eV2 . In
both cases, the corresponding mixing angle (12 ) is large (albeit not maximal).
Finally, the LSND data [26,27] would indicate a e oscillation with a third,
very distinct, neutrino mass difference: m2LSND 0.36 eV2 . The LSND evidence in
favour of neutrino oscillation has not been confirmed by other experiments so far [28]; the
MiniBooNE experiment [29] will be able to do it in the near future [30]. In the absence
of an independent confirmation of the LSND evidence, we restrict ourselves to the three
neutrino mixing scenario (the impact of a Neutrino Factory in the case of four neutrino
mixing has been discussed in full detail in [3133]).
These oscillation signals will be confirmed in ongoing and planned atmospheric and
solar neutrino experiments, as well as in long baseline ones, with the latter being free of
model-dependent estimations of neutrino fluxes. There is a strong case for going further
in the fundamental quest of the neutrino masses and mixing angles, as a necessary step
to unravel the fundamental new scale(s) behind neutrino oscillations. In particular, it is
possible that in ten years from now no information whatsoever will be at hand regarding
the 13 angle (the key between the atmospheric and solar neutrino realms, for which the
present bound is sin2 (213 ) 1 101 , [34]) and the leptonic CP violating phase .
An experimental set-up with the ambitious goal of precision measurement of the whole
three-neutrino mixing parameter space is under study. This experimental programme
consists of the development of a Neutrino Factory (high-energy muons decaying in the
straight section of a storage ring and producing a very pure and intense neutrino beam,
[35,36]) and of suitably optimized detectors located far away from the neutrino source.
The effort to prepare such very long baseline neutrino experiments will require a time
period covering this and the beginning of the following decade. It is therefore of interest
to look for the optimal conceivable factory-detectors combination. One of its main goals
would be the discovery of leptonic CP violation and, possibly, its study [3740]. Previous
analyses [4143] on the foreseeable outcome of experiments at a Neutrino Factory have
shown that the determination of the two still unknown parameters in the three-neutrino
mixing matrix, 13 and , will be possible (if the LMA-MSW solution of the solar neutrino
323
deficit is confirmed). The most sensitive method to study these topics is to measure the
transition probabilities involving e and e , in particular, e (e ) ( ). This is what is
called the golden measurement at the neutrino factory. Such a facility is indeed unique
in providing high energy and intense e (e ) beams. Since these beams contain no ( ),
the transitions of interest can be measured by searching for wrong-sign muons: negative
(positive) muons appearing in a massive detector with good muon charge identification
capabilities [40].
An incredible amount of work has been devoted to this topic in the last few years: we
refer the interested reader to [4454] and to the references therein for an overview of the
status-of-the-art in all its different aspects; we address to [5561] and references therein
for a comparison of the physics reach of a conventional (super)beam and of a Neutrino
Factory; eventually, we point out that in [62] the idea of a e beam originating from decay (the so-called -beam), was advanced: it appears that the physics reach of such a
beam is complementary to that of a conventional superbeam [63].
E ) for neutrinos at a fixed
In [64] it has been noticed that the probability P (13, ,
defines
energy (and for a given baseline) computed for a given theoretical input pair (13, )
a continuous equiprobability curve in the (13 , ) plane. Therefore, for a fixed energy,
a continuum of solutions reproduce the input probability. A second equiprobability curve
is defined in this plane by the probability for antineutrinos at the same energy and with
E ). The two equiprobability curves have, quite
the same input parameters, P (13 , ,
the second at a different point
generally, two intersection points: the first of them at (13 , ),
(13, ). It is the intersection of equiprobability curves from the neutrinos and antineutrinos
that resolves the continuum degeneracy of solutions in the (13, ) plane, restricting the
and (13 , ).
This second
allowed values for 13 and to the two regions around (13, )
intersection, however, introduces an ambiguity in the measurement of the physical values
of 13 and . Different proposals have been suggested to solve this ambiguity: in [64] the
ambiguity is solved by fitting at two different baselines at the same time; another possibility
is an increase in the energy resolution of the detector [46,50,51]; see also [65].
New degeneracies have later been noticed [59,66], resulting from our ignorance of
the sign of the m2atm squared mass difference (by the time the Neutrino Factory will
be operational) and from the approximate [23 , /2 23 ] symmetry for the atmospheric
angle.
In the first part of the paper we describe how the (13, ) ambiguity arises in e
oscillation due to the equiprobability curves in the (13 , ) plane at fixed neutrino energy.
We then extend our analysis showing how the same phenomenon can be observed in a
real experiment: equal-number-of-events (ENE) curves for any given neutrino energy bin
appears and their intersections in the (13, ) plane explain how and where do clone
regions arise. Our analysis is afterwards compared with the results of simulations with a
realistic magnetized iron detector (studied in [67]).
We then propose to reduce the continuum degeneracy and to resolve the last ambiguity
using two baselines with detectors of different design. We notice that muons proceeding
from decay when s are produced via a e transition show a different (13 , )
correlation from those coming from e (first considered in [40]). By using a near
lead-emulsion detector, capable of the -decay vertex recognition, we can therefore use the
complementarity of the information from e and from e to solve the (13, )
324
ambiguity. We find that the combination of the near emulsion detector and of a massive
magnetized iron detector at L = 3000 km could indeed help to achieve a good resolution
in the (13 , ) plane.
In this paper we restrict ourselves to the (13 , ) ambiguity, by fixing 23 = 45 and by
choosing a given sign for m2atm (in the hypothesis that more information on the three
neutrino spectrum will be available by the time the Neutrino Factory will be operational).
However, we found that the core of our results do not depend on the sign of m2atm : indeed,
we have been carrying on simulations with the opposite sign, with similar results (i.e., we
still observe how clone regions disappear due to the two-detector types combination;
notice, however, that the location of the clone regions and all the details of the simulation
do depend on the sign of m2atm ). We remind that, if 13 is not extremely small, the
combined measurement of e and e transitions could significantly help
in solving the [23, /2 23 ] ambiguity [66]. We understand that solving the three
ambiguities at the same time will need the combination of different measurements and
we plan to explore this possibility in a forthcoming paper.
In Section 2 we present our analysis of the e equiprobability curves in the (13, )
plane; in Section 3 we introduce the corresponding equal-number-of-events (ENE) curves;
in Section 4 we present a similar analysis for the e oscillation probability and
study the impact of silver wrong-sign muon events; in Section 5 we show our results
for the combination of a near (L = 732 km) OPERA-like detector and of a L = 3000 km
magnetized iron detector; in Section 6 we eventually draw our conclusions. In Appendix A
a perturbative expansion in of the formulae of Section 2 is presented; in Appendix B
we report some useful formulae for CC-interaction and decay.
2. e equiprobability curves in the (13 , ) plane
We consider the e transition, first. This channel has been shown to be the optimal
one to measure simultaneously 13 and at the Neutrino Factory in the context of threefamily mixing, through the appearance of wrong-sign muons in the detector, [40]. It
therefore deserves the nickname of golden channel.
Following Eq. (1) of [64] we get for the transition probability at second order in
perturbation theory in 13 , /atm , /A and L (see also [6870]),

atm L

2
+ Z, (1)
Pe (13 , ) = X sin (213) + Y cos(13 ) sin(213) cos
2
where refers to neutrinos and antineutrinos, respectively, and
2 2 B L
X = sin2 (23 ) Batm

sin
2 ,

atm
AL
B L
(2)
Y = sin(212 ) sin(223) A
B sin 2 sin 2 ,
2
Z = cos2 (23 ) sin2 (212) A sin2 AL
2 ,
with Z = Z+ = Z . In these formulae, A = 2 GF ne (expressed in eV2 /GeV) and

B = |A atm | (with referring to neutrinos and antineutrinos, respectively). Finally,
atm = m2atm /2E and = m2 /2E .
325
The parameters 13 and are the physical parameters that must be reconstructed by
fitting the experimental data with the theoretical formula for oscillations in matter. In what
follows, the other parameters have been considered as fixed quantities, supposed to be
known by the time when the Neutrino Factory will be operational. In particular, in the
solar sector we fixed 12 = 33 and m2 = 1.0 104 eV2 [2225], corresponding to
the LMA region of the solar neutrino problem (accordingly to the recent SNO results
[13,14]); in the atmospheric sector, 23 = 45 and m2atm = 2.9 103 eV2 [20] (notice
that for 23 = 45 the [23 , /2 23 ] ambiguity [66] is absent). Finally, we considered
a fixed value for the matter parameter, A = 1.1 104 eV2 /GeV for L < 4000 km and
A = 1.5 104 eV2 /GeV for L > 4000 km, obtained by using the average matter density
alongside the path for the chosen distance computed with the Preliminary Earth Model
[71]. For simplicity, we have not included errors on these parameters.1
Eq. (1) leads to an equiprobability curve in the plane (13, ) for neutrinos and
antineutrinos of a given energy:

= Pe
(13 , )
(13 , ).
Pe
We can solve Eq. (3) for :

Pe X sin2 (213) Z
atm L
cos
=
.
2
Y cos(13 ) sin(213)
It is useful to introduce the following functions:
f (13 , 13 ) = sin2 (213)sin2 (213 ) ,

cos 13 sin(213 )
g( , ) =
13 13
cos 13 sin(213 )
cos 13 sin(213 ) ,
(3)
(4)
(5)
with the obvious limit f (13 , 13 ) = 0 and g(13 , 13) = 1. Eq. (4) can then be written as:

atm L
atm L
= R f (13 , 13 ) + cos
g(13 , 13 ).
cos
(6)
2
2
Eq. (6) is particularly illuminating: it describes a family of two branches curves in the plane
(13, ) for the neutrinos and a second family of two branches curves for the antineutrinos.
The dependence on the neutrino energy resides in atm and in the ratio R = X /Y ,
whereas the dependence on the angle is factorized in the two 13 -dependent functions f
and g.
It is helpful to introduce the parameter :
13 = 13 + ,
constrained by the bound

cos atm L 1.

2
(7)
1 It has been shown in [64] that the inclusion of the foreseeable uncertainties on these parameters does not
modify the results on the 13 and measurements in a relevant manner.
326
Fig. 1. Allowed region in as a function of 13 for E = 38 GeV at three different distances: L = 732, 3000
and 7332 km.
on the neutrino energy

The allowed region for depend on the input parameters (13 , ),
and on the baseline. In Fig. 1, we compute the allowed values of as a function of 13
(13 [0 , 13]) at three different baselines, L = 732, 3000 and 7332 km, for = 0 and
E = 38 GeV, by numerically solving Eq. (7).
It may be noticed that, for (almost) every value of 13 in the considered range, two
different regions of allowed values for exist. The first region corresponds to 0,
whereas the second corresponds to large negative values for . In this region, 13 is
negative: this region is therefore unphysical when the sign of the mass differences and of
the various angles are defined in an appropriate way [52]. We concentrate hereafter on the
tiny region around 0.
+ (the upper row) and P (the
In Fig. 2 we present the equiprobability curves for Pe
e
lower row) in the (, ) plane, at L = 732, 3000 and 7332 km for different values of
the neutrino energy in the range E [5, 50] GeV. The input values are 13 = 5 and
= 60 . In the upper row (neutrinos), it can be seen that all the equiprobability curves
However, notice that the
intersect in = 0 , = 60 (namely, 13 = 13 and = ).
equiprobability curve for a given neutrino energy intersects the curves corresponding to
a different neutrino energy in a second point, at positive and negative . This second
intersection depends on the energies of the two curves. In the upper branch of the neutrino
equiprobability curves, no second intersection is observed, for these particular values of the
input parameters. The results of Fig. 2 may be understood with the help of a perturbative
expansion of Eqs. (5) and (6) in terms of powers of (always possible in the allowed
region, for 13 large enough). Details on this expansion can be found in Appendix A.
We can draw some conclusion from what observed in Fig. 2 and from the previous
considerations on the energy dependence of the equiprobability curves. In particular, it is
to be expected that by fitting experimental data for neutrinos only it should be quite difficult
with good accuracy. We expect, instead, that
to determine the physical parameters (13, )
the fitting procedure will identify a low 2 region whenever the family of equiprobability
curves are not well separated, within the experimental energy resolution. In particular, at
short distance (L = 732 km) it is to be expected a good determination of 13 (notice that
is generally less than 2 ) and no determination whatsoever of the CP-violating phase .

At the intermediate distance, L = 3000 km, the equiprobability curves for neutrinos (for
this particular set of input parameters) do not depend strongly on the energy in the upper
327
Fig. 2. Equiprobability curves in the (, ) plane, for 13 = 5 , = 60 , E [5, 50] GeV and L = 732, 3000
and 7332 km. The upper row represents equiprobability curves for neutrinos, the lower row for antineutrinos. The
dashed line is E = 5 GeV, the solid line is E = 45 GeV; the dotted lines lie in between these two.
Fig. 3. Same as in Fig. 2, but with neutrinos and antineutrinos equiprobability curves superimposed.
branch, whereas a larger separation can be seen in the lower branch. Therefore, we expect
a low 2 region alongside the upper branch of the equiprobability curves spanning from
to the
around the single point corresponding to the physical parameters (at = 0, = )
(diluted) region where the curves show the second intersection (that by periodicity in the
axis happens to be in the lower branch).
Finally, at large distance (L = 7332 km) we expect a low 2 region in the region
0.5 and 100 (notice that the small spread in the variable for this baseline
is in agreement with Fig. 1).
A great improvement in the reconstruction of the physical parameters from the
experimental data is achievable using at the same time neutrino and antineutrino data. This
can be seen in Fig. 3, where the equiprobability curves for neutrinos and antineutrinos (for
the same input parameters as in Fig. 2) have been superimposed. At short distance, the two
family of equiprobability curves overlap for any value of , and no improvement is to be
expected. However, at the intermediate distance the equiprobability curves for neutrinos
and antineutrinos overlap only in the vicinity of the physical point = 0, = and in
328
the region of the second intersection, whereas in the intermediate region they are quite
well separated, both in the upper and lower branch. We expect, in this case, that the fitting
procedure of the whole set of neutrino and antineutrino data will identify two separate low
2 region, around the physical point and around the region where all the curves show the
second intersection. This second allowed region in the parameter space was first observed
in [64] and subsequently confirmed in [46,50,51]. Finally, at large distance we expect no
significant improvement with respect to the previous case.
Notice that these considerations can be drawn by looking at the equiprobability curves
for neutrinos and antineutrinos, only. We will see in the following section how the
theoretical expectation is indeed reproduced in the experimental data.
3. Number of wrong-sign muons in the detector

but the number of
The experimental information is not the transition probability Pe
muons with charge opposite to that of the muons circulating in the storage ring, that in
the following will be often called golden muons. The events are then grouped in bins
of energy, with the size of the energy bin depending on the energy resolution E of the
considered detector. In general,
=
Ni (13 , )
Ei+E
dE ( ) (E)Pe
(E, 13 , )
de ( e ) (E)
dE
(8)
Ei
E is the
is the number of wrong-sign muons in the ith energy bin for the input pair (13, );
2
neutrino (antineutrino) energy. The charged current neutrino and antineutrino interaction
rates can be computed using the approximate expressions for the neutrinonucleon cross
sections on an isoscalar target,
E
m2 ,
GeV
E
m2 .
N 0.34 1042
(9)
GeV
In the laboratory frame the neutrino fluxes, boosted along the muon momentum vector, are
given by:
N 0.67 1042
24n 4 2
d 2 e ,e
=
E y (1 cos )
dy d
L2 m6

2 y(1 cos ) P m2 2E
2 y(1 cos ) . (10)
m2 2E

)2 , E
is the parent muon energy, y = E /E
, n is the number
Here, = 1 (m /E
of useful muons per year obtained from the storage ring and is the angle between the
2 The neutrino energy can be reconstructed if the considered detector has a hadronic calorimeter capable to
measure the energy of the hadronic shower (Ehadr ) in the N CC interactions with good precision.
329
beam axis and the direction pointing towards the detector. In what follows, the fluxes
have been integrated in the forward direction with an angular divergence (taken to be
constant) 0.1 mr. The effects of the beam divergence and the QED one-loop radiative
corrections to the neutrino fluxes have been properly taken into account in [72]. The overall
correction to the neutrino flux has been shown to be of O(0.1%).
In the same approximations as for Eq. (1), we get for the number of events per bin:

Ni = IXi sin2 (213 ) + IYi c cos + IYi s sin cos 13 sin(213) + IZi ,
+
+
+

(11)
N i + = IXi sin2 (213 ) + I i c cos I i s sin cos 13 sin(213) + IZi ,
Y
Y
where we introduced a short-form notation for the following integrals:
Ei +E
d ( e ) (E)
i
dE ( ) edE
X (E),
IX = Ei
d
(E)
E
+E
L
( e )
IYi c = E i
,
dE ( ) edE
Y (E) cos atm
2
i
d
(E)
E
+E
L
( e )
Ii s = i
dE ( ) edE
Y (E) sin atm
,
Ei
2
Y
i Ei +E
de ( e ) (E)
dE ( )
Z(E).
IZ = Ei
dE
(12)
we can draw a continuous

For a fixed energy bin and fixed input parameters (13 , ),
curve of equal number of events in the (, ) plane,
Ni (13 , ) = Ni (13 , ),
(13)
as it was the case for the transition probability, Eq. (3).

We therefore get an implicit equation in ,
F () = G(13, 13 , ),
(14)
where
Ii
s
Y
F
()
=
cos
sin ,
IY c
i
IX
G(13 , 13 , ) = i f (13 , 13 ) + F ()g(

13 , 13 )
(15)
IY c
and f (13 , 13) and g(13 , 13 ) are the -dependent functions introduced in Eq. (5).
Solving for ,

= F 1 G 13, 13 ,
(16)
we get equal-number-of-events curves (ENE) in the (, ) plane, see Fig. 4.
The problem arises in the reconstruction of the physical parameters from a data set
consisting of some given number of events per bin, for a given number of bins (depending
on the specific detector energy resolution). As it was the case for the equiprobability curves
in the previous section, all the ENE curves intersect in the physical point ( = 0, = )
and any given couple of curves intersect in a second point in the same region as in Fig. 2. As
it can be seen in Fig. 4, the second intersection differs when considering different couples
330
Fig. 4. Equal-number-of-events (ENE) curves in the (, ) plane for neutrinos, for 13 = 5 , = 60 ,
E [5, 50] GeV and L = 732, 3000 and 7332 km. The dashed line represents the first energy bin,
E [0, 10] GeV, the solid line the last energy bin, E [40, 50] GeV; the dotted lines lie in between these
two.
of curves, but lies always in a restricted area of the (, ) plane, the specific location of
see Eqs. (A.2) and (A.3). The 2
this region depending on the input parameters (13, ),
analysis of the data will therefore identify two allowed regions: the physical one (around
and the clone solution, spanning all the area where a second
the physical value, 13 , )
intersection between any two ENE curves occurs. This is the source of the ambiguity
pointed out in [64].
In the remaining of this section, we apply the analysis in energy bins of [40,64]. Let
be the total number of wrong-sign muons detected when the factory is run in polarity
Ni,p
p = + , , grouped in energy bins specified by the index i, and three possible distances,
= 1, 2, 3 (corresponding to L = 732 km, L = 3000 km and L = 7332 km, respectively).
In order to simulate a typical experimental situation we generate a set of data ni,p as
follows: for a given value of the oscillation parameters, the expected number of events,
, is computed; taking into account backgrounds and detection efficiencies per bin, b
Ni,p
i,p
, we then perform a Gaussian (or Poissonian for N 10 events) smearing to
and 3i,p
i,p
mimic the statistical uncertainty:
ni,p = Smear Ni,p

(17)
.
3i,p + bi,p
Finally, data are fitted to the theoretical expectation as a function of the neutrino
parameters under study, using a Gaussian 2 minimization:
2
ni,p Ni,p
2 =
(18)
,
ni,p
p
i
where ni,p is the statistical error for ni,p (errors on background and efficiencies are
neglected) or a Poissonian 2 minimization:

ni,p
ni,p Ni,p
ni,p log
2 = 2
(19)
Ni,p
p
i
whenever events are Poisson-distributed around the theoretical values (see [61] and
references therein). We verified that the fitting of theoretical numbers to the smeared
(experimental) ones is able to reproduce the values of the input parameters (the best
fit always lies within a restricted region around 13, ).
331
Fig. 5. 68.5, 90 and 99% C.L. contours resulting from a 2 fit of 13 and , for 13 = 1 and = 0 (up)
and = 90 (down). For each considered input parameters couple, the two rows represent: from left to right,
L = 732, 3000 and 7332 km; the upper row is Ni + , i = 1, . . . , 5; the lower row is Ni + and Ni , i = 1, . . . , 5.
The following reference set-up has been considered: neutrino beams resulting from
= 50 GeV muon
the decay of 2 1020+ s and s per year in a straight section of an E
accumulator. An experiment with a realistic 40 kton detector of magnetized iron and five
years of data taking for each polarity is envisaged. Detailed estimates of the corresponding
expected backgrounds and efficiencies have been included in the analysis, following [67].
Notice that this set-up is exactly the same of [40,64].
In the first row of Figs. 5, 6, we present the results of the fit to five bins of data
for decaying muons of one single polarity, . The energy binning of the detector is
332
Fig. 6. 68.5, 90 and 99% C.L. contours resulting from a 2 fit of 13 and , for 13 = 5 and = 0 (up)
and = 90 (down). For each considered input parameters couple, the two rows represent: from left to right,
L = 732, 3000 and 7332 km; the upper row is N i + , i = 1, . . . , 5; the lower row is N i + and N i , i = 1, . . . , 5.
E = 10 GeV. In all cases we observe the pattern depicted in the previous section, with
In the second row
a good determination of 13 and an extremely poor determination of .
we fit to five bins of data for decaying muons of both polarities. The results follow again
the theoretical analysis of this and of the previous section and are in perfect agreement
with what presented in [64]. In particular, notice how at the intermediate distance it is
possible now to reconstruct with an error of tens of degrees in two separate regions of the
parameter space.
333
4. The e equiprobability and ENE curves

We present in this section the possibility to use a different channel, namely the e
oscillation probability, to improve the reconstruction of the physical parameters (13, )

in combination with the results for the e transition described in [40,64] and in the
previous section.
The e oscillation probability at second order in perturbation theory in 13 ,
/atm , /A and L is:

= X
Pe
(13 , )
sin2 (213)

atm L
Y cos(13 ) sin(213) cos

+ Z ,
2
where refers to neutrinos and antineutrinos, respectively, and
2 2 B L
X = cos2 (23) Batm

sin
2 ,

atm
AL
B L
Y = sin(212 ) sin(223) A
B sin 2 sin 2 ,

2
Z = sin2 (23 ) sin2 (212) A sin2 AL
2 ,
(20)
(21)
= Z . Notice that X and Z differs from the corresponding coefficients

with Z = Z+
for the e transition for the cos 23 sin 23 exchange, only. The Y term is identical
for the two channels, but it appears with an opposite sign. This sign difference in the
Y -term is crucial, as it determines a different shape in the (, ) plane for the two sets of
equiprobability curves.
In Fig. 7, we superimposed the equiprobability curves for the e and e
oscillations at a fixed distance, L = 732 km, with input parameters 13 = 5 and = 60 ,
for different values of the energy, E [5, 50] GeV. The effect of the different sign in
front of the Y -term in Eqs. (1) and (20) can be seen in the opposite shape in the (13, )
plane of the e curves with respect to the e ones. Notice that all the curves
Fig. 7. Equiprobability curves in the (, ) plane, for 13 = 5 , = 60 , E [5, 50] GeV and L = 732 km for
the e and e oscillation (neutrinos on the left, antineutrinos on the right).
334
and that now three wouldof both families met in the physical point, 13 = 13 , = ,
be clone regions (i.e., the spread regions where the intersections of any given couple of
equiprobability curves lie) can be seen.
As a final comment we signal that, if 13 is not extremely small (in such a way
terms dominate over the Y
that the X and X
terms in Eqs. (1) and (20)), the
combined measurement of e and e transitions could in principle solve the
[23, /2 23 ] ambiguity.
To follow the line of reasoning adopted for the e channel, we should now discuss
how the e channel can be used in a realistic experiment. A number of modifications
with respect to the case of the golden channel should be taken into account.
First, the approximate expressions for the neutrinonucleon cross section on an isoscalar
target, Eq. (9), are no longer appropriate in the case of a CC interaction inside the
detector. In this case we used the reported values for the N cross-section [73] that
have been applied in the CHORUS and OPERA experiment to compute the expected
number of CC events. The considered cross-section includes mass effects in the DIS
region following [74] (see Appendix B for details), as well as the elastic and quasi-elastic
contributions to the cross section.
Second, the will decay in flight into a muon of the same charge and two neutrinos,
with a branching ratio BR( ) 0.17 [19]. This silver wrong-sign muon is
the experimental signal we are looking for, to be identified and to be separated from
the golden wrong-sign muons originated from CC interactions.3 The first tool to
distinguish the two sets of wrong-sign muons is their different energy distribution (see
Appendix B for details on the differential decay rate). It has been shown in [67] that in the
magnetized iron detector considered in the previous section, muons from decay cannot
be distinguished from the main background represented by muons from charmed mesons
decay by means of kinematical cuts. In order to take advantage of this channel, we should
therefore use a different kind of detector: for this reason we concentrate in the remaining of
the paper on a lead-emulsion detector, where the observation of the decay vertex allows
to distinguish golden and silver wrong-sign muons, and the latter from the charmed
mesons decay background. We must mention that the e oscillations were previously
considered in [48] for a liquid argon detector, using kinematical cuts to identify s. It
could be of interest to explore further the possibility of using silver muon events in such
a detector to reduce or eliminate the (13, ) ambiguity.
In what follows, we consider an OPERA-like detector with a mass of 2 kton and
spectrometers capable of muon charge identification (see the OPERA proposal for details,
[75]) located at L = 732 km down the neutrino source (obviously, in the back of our mind
we are thinking of the CNGS set-up). The results for both the golden and the silver
channel at the near emulsion detector will be combined with results for the golden
channel obtained with the magnetized iron detector located at the optimal distance for
the measurement of leptonic CP violation, L = 3000 km.
3 We adopt the nick-name of silver muon events due to the lesser statistical significance with respect to
golden ones.
335
In this paper, we will first restrict ourselves to an ideal OPERA-like detector with perfect
efficiency and no background. Afterwards, we take into account the realistic estimates
of the energy-dependent reconstruction efficiency and of the most relevant backgrounds.4
Eventually, we will consider how an increase in the detector mass or an improvement on
the signal/noise ratio affects our results.
Schematically, starting from a positive charged muon in the storage ring, silver muons
are obtained by the following chain:
+
e ,
+ ,
e ,
whereas golden muons come from:
+
e ,
+ ,
e .
If we group the events in bins of the final muon energy E , with the size of the energy
bin E depending on the energy resolution of the considered detector, the number of
and for a parent muon
golden muons in the ith energy bin for the input pair (13, )
is:
energy E

) Ei +E
d ( ) (E , E )
de ( e ) (E , E
g
=

N (13 , )
Pe
(E , 13, )
dE
dE
Ei
(22)
(remember that
is the oscillation probability for neutrinos and antineutrinos, respectively, see Section 2), whereas the number of silver muons in the ith energy bin is:

dN (E , E ) d ( ) (E , E )
s
N (13 , ) = BR( )
dE
dE
) Ei +E
de ( e ) (E , E

Pe
(E , 13 , )
.
dE
Ei
(23)
In both equations, stands for a convolution integral on the intermediate energy: for
example,

dN (E , E ) d ( ) (E , E )
BR( )
dE
dE
P
gives the number of muons in the ith bin in the final muon energy E , for a given neutrino
energy E (see Appendix B for details). In Fig. 8 we present a direct comparison of
4 A dedicated careful analysis of the silver muons reconstruction efficiency and of the background as a
function of the neutrino energy for this specific detector is currently under progress, [76]. A key issue is the
maximum affordable amount of charge discrimination, both for the emulsion and the magnetized iron detector.
Estimates can be found in [67] for the magnetized iron detector and in Fig. 86 of [75] for the emulsion detector.
336
Fig. 8. Comparison of the number of golden muons (lightest bars) and silver muons (darkest bars), for a
= 50 GeV and input parameters 13 = 5 , = 60 and a near detector (L = 732 km)
parent muon energy E
with a mass of 2 kton and perfect reconstruction efficiency for both channels.
golden and silver muons grouped in five energy bins with E = 10 GeV, for a parent
= 50 GeV with input parameters 13 = 5 , = 60 . We consider here
muon energy E
a near detector (L = 732 km) with a mass of 2 kton and perfect reconstruction efficiency
for both channels. Notice that in Fig. 8 we have not included quasi-elastic and resonance
contributions to the N cross section that could enhance production at low neutrino
energy.
g
The total number of events for these parameters are N 700 and Ns 30. The
strong reduction in the number of silver muons with respect to the golden muons with
the same input parameters depends on the suppression due to the BR( ) branching
ratio and to the different N DIS cross section for muons and taus.
Following the same procedure used to get the golden muons ENE curves presented in
Fig. 4 we can compute ENE curves for silver muons. These curves are reported in Fig. 9
in the case of L = 732 km.
In Fig. 10 we superimpose golden and silver ENE curves for L = 732 km and
13 = 5 , = 60 . Notice how, as it was expected from the equiprobability curves
analysis, the two sets of curves have opposite concavity in the (, ) plane. As for the
equiprobability curves, all lines met in the physical point. Therefore, a combined 2
analysis of the two sets of data should present a well defined global minimum around
the physical region, whereas the local minima situated in the three clone regions are
considerably raised with respect to what presented in the previous section where only the
golden muon signal was considered.
5. Combination of golden and silver muon events

We follow the analysis in energy bins outlined in the previous section and in [40,
g
64]: we produce a theoretical data set (N , Ns ) for fixed input parameters 13, and
337
Fig. 9. Equal-number-of-event-curves in the (, ) plane, for E [5, 50] GeV and L = 732 km in the case of
e oscillation (neutrinos on the left, antineutrinos on the right), for 13 = 5 and = 60 .
Fig. 10. Superposition of the equal-number-of-events curves for the transition e (light lines) and e
(dark lines), for L = 732 km and 13 = 5 , = 60 .
g
then we smear it as in Eq. (17) to obtain an experimental data set (n , ns ). Finally,

experimental data are fitted and 68.5, 90 and 99% C.L. contours in the (13, ) plane are
drawn.
In Fig. 11 we present the results of this analysis comparing two different possibilities:
in the upper row we combine two realistic magnetized iron detectors at L = 732 and
L = 3000 km; in the lower row we combine an ideal OPERA-like detector at L = 732 km
and a realistic magnetized iron detector at L = 3000 km.
First, we present our results for the combination of the two iron detectors (where only
g
golden muons, N , can be used). We draw in each figure the contours for different input
parameters: three values for 13 = 1 , 6 and 11 and three values for the phase = 90 ,
are shown:
0 and 90 . In each figure, therefore, fits to nine input parameter pairs (13 , )
this has to be compared with Figs. 5 and 6 where in each plot the results of a fit to one
single input parameter pair was presented.
338
On the left, only five years of data taking for + circulating in the storage ring are
considered. Notice that for any given input pair 13 is always reconstructed within a
2 error; on the contrary, roughly any value for the CP-violating phase is allowed. The
situation is drastically improved on the right, where five years of data taking for each muon
polarization are considered. In this case, the phase is reconstructed with a precision of
tens of degrees for all values of the input parameters. Notice, however, how some clone
region is still present at 90% C.L. (e.g., for 13 = 1 , = 90 the small region around
13 = 2 , = 150 ; for 13 = 1 , = 0 the small region around 13 = 2.5 , = 150 ;
see also Fig. 5). These results can be easily understood in terms of the theoretical analysis
of the equiprobability and equal-number-of-events curves for the e channel of
Sections 2 and 3.
We present now results for the combination of a near emulsion detector (with both
g
golden and silver muons, N and Ns ) and a not-so-far iron detector (with golden
muons, only). On the left, again only five years of data taking for + circulating in the
storage ring are considered. Notice that a significant reduction in the reconstruction errors
on the phase is already achieved. On the right, we simply add to the first five years of data
taking for the + polarity further five years for the opposite polarity in the iron detector,
only. We have not included a further five year operational time for the emulsion detector to
take into account the mass decrease due to the brick removal in the first five year period.5
Notice, however, that a quite relevant improvement with respect to the one-polarity twodetector types case (lower left) is achieved in the reconstruction error. More important,
an improvement with respect to the two-polarities one-detector type (upper right) can
also be observed. In particular, the clone regions have completely disappeared, due to
the combination of silver and golden muons with a different -dependent oscillation
probability. The effect of the inclusion of silver muons can be seen in Fig. 12, where we
present the two-polarities two-detector types combination (Fig. 11, lower right) compared
with the same combination but with only a golden muon signal. In this figure we can
clearly see that clone regions are still present and that the near emulsion detector is
too small to compete with a 40 kton iron detector located at the same distance down the
neutrino source.
We also studied the effect of a change in the longest baseline, moving the iron detector
from L = 3000 km to L = 2000 km, with no significant improvement. This was to be
expected, since the former distance, L = 3000 km has been found to be the optimal one to
measure the CP-violating phase for the considered set-up, [4050].
We remind here that in [64] it was shown that the optimal combination of two
magnetized iron detectors is achieved for L = 3000 km and L = 7332 km: in that case,
no clone region is present (see Fig. 6 of [64]). We prefer however to compare our results
for the OPERA-like near detector and magnetized iron not-so-far detector option with the
two magnetized iron detectors combination at the same distances.
Up to this moment, we considered an ideal lead-emulsion detector with identification
efficiency equal to 1 and no backgrounds. We should consider instead a real detector with
5 This is a quite conservative assumption, being the expected number of bricks to be removed looking for
silver muons smaller than in the case of oscillations considered in the OPERA proposal [75].
339
Fig. 11. 68.5, 90 and 99% C.L. contours resulting from a 2 fit of 13 and , for 13 = 1 , 6 and 11 , and
= 90 , 0 and 90 , for the combination of two iron detectors (upper row) or one iron detector and one
g
g
g
emulsion detector (lower row). Upper row: iron detector at L = 732 and 3000 km: (a) N+ ; (b) N+ and N ;
g
Lower row: iron detector at L = 3000 km and emulsion detector at L = 732 km: (a) N+ +Ns + ; (b) N+ +Ns +
g
and N .
= 90 , 0 and 90 , for the combination of an iron detector at L = 3000 km and an emulsion detector at
L = 732 km: (left) both silver and golden muon events are taken into account (same figure as in Fig. 11, lower
right); (right) only golden muon events are taken into account. Five years of data taking for both polarities in
the distant detector and only five years in the + polarity in the near detector have been considered.
a refined analysis of efficiencies and background for the two wrong-sign muons signals,
e and e . A detailed simulation of the expected performance at an OPERAlike detector is under way [76]. However, a preliminary study has been done following the
outline of the OPERA proposal [75] and of the recent progress report [77].
We first consider golden muons. In this case, it seems reasonable to consider a
90% average efficiency in the neutrino energy range E [5, 50] GeV. The most relevant
background to this channel are right-sign muons with a wrong charge assignment: with
no improvement with respect to the present OPERA proposal, a level of 103 of charge
340
misidentification is achieved (a level of 106 was envisaged for the magnetized iron
detector, see [67]). On the other side, the dominant background at the magnetized iron
detector coming from charmed mesons decay with no right-sign muon identification is
here under control by looking for the decay vertex.
Regarding silver muons, we have an unavoidable important reduction in the average
efficiency, mainly due to the detector design: the decay vertex cannot be observed if
it occurs inside the lead plates. An average efficiency for silver muons of 25% only is
achieved in the neutrino energy range E [5, 50] GeV [75]. In the case of silver muon
events, the dominant background is the standard charmed mesons decay with no rightsign muon identification. Sub-dominant backgrounds coming from and K mesons
decay can be kept at a reasonable level by imposing kinematical cuts as described in [75].
As an illustrative example, in Fig. 13 we present the charmed mesons decay background
and the silver muon signal for 13 = 5 and = 90 as a function of the neutrino energy
(for 1 GeV energy bins). Notice how the silver muons signal peaks at lower energy with
respect to the background, thus allowing us to use the e oscillation to improve the
(13, ) reconstruction.
Eventually, we present in Fig. 14 the results for the combination of an iron detector at
L = 3000 km and of an emulsion detector at L = 732 km with reasonable estimates for
the reconstruction efficiencies for golden and silver muons and for the corresponding
backgrounds. We can see that the results obtained with the two-detector types combination
strongly resembles those obtained with the reference two iron detectors set-up (Fig. 11,
upper right). In particular, we notice a general improvement at the 68% C.L. but we still
observe clone regions for small values of 13 .
Although the results of Fig. 14 do not indicate a strong improvement with respect to
the reference two iron detectors set-up, two comments are in order: first, we have been
considering up to now an OPERA-like detector with the characteristics of the detector
described in [75] which is currently under construction and should be operational by year
2006. We therefore study the effects on the physics reach of foreseeable improvements
in the following. Secondly, it seems to us that the combination of two different detector
types with different characteristics and systematics could be of the utmost importance in
Fig. 13. Charmed mesons decay background and silver muons signal (shaded) for 13 = 5 and = 90 in the
OPERA-like emulsion detector as a function of the reconstructed neutrino energy (for 1 GeV energy bins).
341
= 90 , 0 and 90 , for the combination of an iron detector at L = 3000 km (with golden muon events) and
an emulsion detector at L = 732 km (with golden and silver muon events). Realistic estimates for efficiencies
and backgrounds at both the iron and the emulsion detector have been taken into account. Five years of data taking
for both polarities in the distant detector and only five years in the + polarity in the near detector have been
considered.
the project of a facility for precision measurements of the leptonic mixing matrix such as
the Neutrino Factory.
In Table 1 we schematically present which are the main advantages of an increase in
the emulsion detector mass, or of a reduction of the main backgrounds, considering the
presence or absence of clone regions and the maximal error in that can be achieved.
The main background for silver muon events Bsilver (originating from charmed mesons
decay) is normalized to the value that can be deduced on the basis of the OPERA proposal.
In this case we do not consider that significant improvements can be achieved and we
restrict ourselves to a 10 to 20% reduction. We believe that, on the other hand, the main
background for golden muon events Bgolden (originating from charge misidentification)
could be significantly reduced: we therefore present results for a 103 (the predicted level
in the OPERA proposal) and for a 104 charge misidentification. Finally, having in mind
that in the last years the scanning power increased by about a factor 10 every two years,
we consider as realistic the assumption that, by the time the Neutrino factory will become
operational, we will be able to scan a larger number of events, say a factor two. Therefore,
we studied the physical reach of the Neutrino factory-detectors set-up as a function of the
lead-emulsion mass, considering a linear increase in the mass from 2 to 4 kton.
The outcome of this analysis is the following:
Clone regions disappear increasing the emulsion mass. For a 4 kton detector no
ambiguity is present. An improvement in the charge identification (i.e., a reduction
of golden muons main background) implies generally a decrease in the statistical
significance of clone regions.
When no clone regions are observed, a reduction of golden muons main
background induces a shift in the 99% C.L. allowed region for towards higher values.
In the optimal case we observe a Gaussian-distributed region with = 20 .
342
Table 1
Effects of a mass increase or of a silver (Bsilver ) or golden (Bgolden ) main background reduction for the
emulsion detector, for 13 = 1 and = 90 . Whenever a clone region is present, no overall error on is given
Bsilver
Bgolden
Clone regions
min , max at 99% C.L.
1.0
0.9
0.8
1.0
0.9
0.8
103
103
103
104
104
104
At 68% C.L.
At 68% C.L.
At 68% C.L.
At 68% C.L.
At 90% C.L.
At 90% C.L.
1.0
0.9
0.8
1.0
0.9
0.8
103
103
103
104
104
104
At 99% C.L.
At 99% C.L.
At 99% C.L.
No
No
No
50 100
50 110
50 110
1.0
0.9
0.8
1.0
0.9
0.8
103
103
103
104
104
104
No
No
No
No
No
No
55 105
55 105
55 105
60 110
60 110
70 110
Mass (kton)
A 10 to 20% reduction of the silver muons main background (charmed mesons

decay) does not seem to improve significantly the previous results.
We remind that this table should be interpreted as an indication of which kind of
improvement in the emulsion detector helps more to improve the physics reach of the
envisaged two detector types combination.
6. Conclusions
It was previously shown [40] that the golden wrong-sign muon signal at the Neutrino
Factory can be extremely useful to measure simultaneously two of the parameters of the
PMNS leptonic mixing matrix, 13 and . In [64] was first noticed that degenerate regions
in the (13, ) parameter space occur in many cases, severely reducing the Neutrino Factory
sensitivity to the CP-violating phase .
In the first part of the paper we describe how the (13, ) ambiguity arises in e
oscillation due to the equiprobability curves in the (13 , ) plane at fixed neutrino energy.
We then extend our analysis showing how the same phenomenon can be observed in a
real experiment: equal-number-of-events (ENE) curves for any given neutrino energy bin
appears and their intersections in the (13, ) plane explain how and where do clone
regions arise. Our analysis is subsequently compared with the results of simulations with a
realistic magnetized iron detector, already studied in previous papers.
343
In the second part of the paper, we propose the use of e oscillation to solve the
(13, ) ambiguity problem. We apply the arguments presented in the case of e and
we show how e actually give a complementary information that can be used to
improve the measurement of the phase . However, to take full advantage of this silver
channel we should use a detector of a different kind to distinguish muons originated from
decay from the so-called golden muons. In the last part of the paper we therefore perform
simulations using the combination of a realistic magnetized iron detector at L = 3000 km
and of an OPERA-like lead-emulsion detector located at L = 732 km from the Neutrino
Factory. The decay vertex recognition in the emulsion detector allow us to separate
silver from golden muon events and to strongly reduce the effect of the charmed
mesons decay background.
We first use an ideal lead-emulsion detector with the mass currently envisaged by
the OPERA Collaboration (2 kton), perfect reconstruction efficiency and no background,
showing how the clone regions disappear for any value of 13 1 . Eventually, we
consider a realistic estimate of the reconstruction efficiency and of the main backgrounds
both for golden and silver muon events at the emulsion detector, following the outline
of the OPERA proposal [75]. In this case, clone regions for 13 1 do appear and our
results are similar to those that we obtain for a combination of two realistic magnetized iron
detectors with L = 732 and L = 3000 baselines. However, we stress here that the OPERAlike 2 kton lead-emulsion detector is under construction and should be operational by mid
2006 to exploit the CNGS beam. We can therefore study how possible improvements of the
detector (from 2006 till the time the Neutrino factory will become operational) affect the
results of our analysis. In particular, we remind that a moderate scaling in the lead-emulsion
detector mass could indeed be seriously taken into account (due to the rapid increase in the
emulsion scanning power). The outcome of this study is that an increase in the emulsion
mass from 2 to 4 kton eliminates completely the clone regions for 13 1 . On the other
hand, a moderate increase in the main background rejection does not seem to improve in a
significant way the previous results.
As a final comment we believe that the two-detector types combination should be
investigated further, even in the case of results comparable to those of a two-baselines
magnetized iron detector combination, since having detectors with different characteristics
and systematics can be extremely helpful to ameliorate the measurement.
Acknowledgements
We are deeply indebted with Paolo Lipari and Maurizio Lusignoli for exhaustive discussions on many different aspects of this paper. We acknowledge useful discussions with
A. Cervera, L. Di Lella, M.B. Gavela, J.J. Gomez-Cadenas, P. Hernandez, L. Ludovici,
Carlos Pea-Garay, S. Rigolin, G. Rosa, P. Strolin, D. Zardetto and P. Zucchelli.
344
Appendix A. Perturbative expansion in
If 13 is large enough, we can expand Eqs. (5) and (6) in powers of for any value of
in the allowed region of Fig. 1.
At first order in ,

f (13 , 13 ) f(13 ) + O 2 4 cos(213 ) + O 2 ,
sin(13 )

2
(A.1)
13 )1
g(13 , 13) 1 g(
13 ) + O 2 1 3 cos(2
+
O
,
sin(2 )
13
where, for 13 [0 , 13 ], f and g are positive functions of 13 .

For neutrinos we get

L
4R+ f(13 ) + cos( atm
13 )
atm L
atm L
2 )g(

+ =
(A.2)
L
2
2
| sin( atm
2 )|
and for antineutrinos we get

L
4R f(13) + cos( + atm
13 )
atm L
atm L
2 )g(
.

+
=
L
2
2
| sin( + atm
2 )|
(A.3)
Notice that from Eq. (5), for = 0, we get = or = atm L. For = 0, we

get the two branches of the equiprobability curves.
If we turn on the correction at first order, the results of Fig. 2 are easily understood
with the following argument: consider two different equiprobability curves for neutrinos,
for two different values of the neutrino energy E1 and E2 , E2 E1 . The intersections
between the two curves are defined by
+ (E1 ) = + (E2 ).
Solving for we get two solutions:

= 0,
(E1 )atm (E2 )
= L atm
H (E2 )H (E1 ) ,
(A.4)
where
H (E) =
13 )
4R+ (E)f(13 ) + cos( + atm2(E)L )g(
.
| sin( + atm (E)L )|
2
On the other hand, the second intersection gets an O( )

For = 0, we get = .
correction:
= + Latm (E1 ) + H (E1).
(A.5)
Being H (E) and (for the considered parameters and baselines) the second solution for
both positive quantities, the second intersection between any two equiprobability curves
for neutrinos will be displaced towards negative values of and positive values of .
This is precisely what observed in the upper row of Fig. 2. For the considered set of
input parameters, baselines and neutrino energy range the same argument applies to the
antineutrino equiprobability curves also, as it can be seen in the lower row of Fig. 2.
345
Appendix B. Formulae for decay and N cross section

Grouping events in bins of the final muon energy E , with the size of the energy bin
E depending on the energy resolution of the considered detector, the number of silver
and for a parent muon energy
muon events in the ith energy bin for the input pair (13, )
is:
E
= BR( )
Ns (13 , )

dN (E , E ) d ( ) (E , E )
dE
dE

Pe
(E , 13, )
)
de ( e ) (E , E
dE
Ei +E
,
Ei
where stands for a convolution integral on the intermediate energy.

Several comments are in order:
)/dE has been given in Eq. (10).
The differential neutrino flux de ( e ) (E , E
The e oscillation probability has been given in Eq. (20).
The N cross section. The (anti)neutrinonucleon differential cross sections on an
isoscalar target, defined as the average of the differential cross sections on proton and
neutron, can be divided in three components: the elastic, the quasi-elastic and the DIS
cross section. We refer to [78] for a detailed discussion on the different contributions
to these components. The DIS cross-section can be expressed in terms of the isoscalar
structure functions Fi (x) (i = 1, . . . , 5):

m2l
m2
d 2 ( ) G2F ME
Mxy
y xy +
=
F1 (x) + 1 y
l 2 F2 (x)
dx dy
2ME
2E
4E

m2l
m4l
m2l
y
xy 1
+
F3 (x) + xy
F4 (x)
y
2
4ME
2ME 4M 2 E2

m2l
F5 (x) ,
(B.1)
2ME
where M is the nucleon mass and ml the charged lepton mass; x = Q2 /2ME y is
the DIS Bjorken variable and y = 1 El /E depends on the fraction of neutrino
energy carried away by the charged lepton. Notice that the x, y variables defined in
this appendix should be interpreted as local variables: outside the appendix the same
letters may represent different objects (we are simply following the standard notation
for the DIS cross section).
Making use of the CallanGross relation, 2xF1 (x) = F2 (x), and of the Albright
Jarlskog relations [74], F4 (x) = 0, xF5 (x) = F2 (x), Eq. (B.1) depends on two
independent structure functions, only. Using the quark parton model (see [79] and
346
references therein, for example) the isoscalar structure functions F2 (x) and xF3 (x)
are:

F2 (x) = x (u + d + 2s) + (u + d + 2c ) ,
F (x) = x (u + d + 2c) + (u + d + 2s ),

2

(B.2)
(x) = x (u + d + 2s) (u
+ 2c ) ,
xF
+
d

xF3 (x) = x (u + d + 2c) (u + d + 2s ) .

Eventually, the structure functions, given in terms of the parton distribution at a definite
value of Q2 [80], must be evolved according to the DGLAP equations to compute the
N cross section for a given neutrino energy.
In Fig. 15 we show the dependence of the d/dy on y for a fixed value of E (in
). Notice the strong suppression of the antineutrino cross section
this case, E = E
with respect to the neutrino one for high and intermediate energy (y 0). The
integration limits for the (x, y) variables are reported in Fig. 16 for different neutrino
energies [81].
The decay rate. The muon distribution is given, in the rest frame, by the following
expression:

d 2 N
1
= f0 (x , R) P f1 (x , R) cos ,
(B.3)

dx d cos
2
where R = m /m , P is the average polarization along the direction in the
laboratory frame, x = 2E /m (not to be confused with the Bjorken x variable
defined for the cross section) and is the angle between the muon momentum vector
and the spin direction. The two f -functions are:

f0 (x , R) = 2 x 2 4R 2 4R 2 + 3 1 + R 2 x 2x 2 ,
(B.4)
2

2
2

f1 (x , R) = 2 x 4R 1 + 3R 2x
(B.5)
and they reduce to the standard f0 (x ) and f1 (x ) functions for the and e flux in the
R 0 limit [82]. The boost in the laboratory frame is given by the following relations:
.
Fig. 15. Neutrino and antineutrinonucleon differential cross section as a function of y for E = E
347
Fig. 16. Kinematical bounds in the x, y plane for different neutrino energies [81].
1/2
x 2
+ x 4R 2
cos ,
2
2
1/2

cos
x + x 2 4R 2

cos =
,

2
2
1/2
[(x + (x 4R ) cos ) 4(1 2 )R 2 ]1/2
y =
where y = E /E and is the angle boosted in the lab frame.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Y. Fukuda, et al., Kamiokande Collaboration, Phys. Lett. B 335 (1994) 237.

R. Becker-Szendy, et al., Nucl. Phys. Proc. Suppl. 38 (1995) 331.
Y. Fukuda, et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 82 (1999) 2644.
W.W. Allison, et al., Soudan-2 Collaboration, Phys. Lett. B 449 (1999) 137.
M. Ambrosio, et al., MACRO Collaboration, Phys. Lett. B 434 (1998) 451.
T. Toshito, SuperKamiokande Collaboration, hep-ex/0105023.
B.T. Cleveland, et al., Astrophys. J. 496 (1998) 505.
Y. Fukuda, et al., Kamiokande Collaboration, Phys. Rev. Lett. 77 (1996) 1683.
W. Hampel, et al., GALLEX Collaboration, Phys. Lett. B 447 (1999) 127.
J.N. Abdurashitov, et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801.
Y. Suzuki, SuperKamiokande Collaboration, Nucl. Phys. Proc. Suppl. 77 (1999) 35.
Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 87 (2001) 071301.
Q.R. Ahmad, et al., SNO Collaboration, nucl-ex/0204008.
Q.R. Ahmad, et al., SNO Collaboration, nucl-ex/0204009.
B. Pontecorvo, Sov. Phys. JETP 6 (1957) 429, Zh. Eksp. Teor. Fiz. 33 (1957) 549 (in Russian).
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
B. Pontecorvo, Sov. Phys. JETP 26 (1968) 984.
(B.6)
348
[18] V.N. Gribov, B. Pontecorvo, Phys. Lett. B 28 (1969) 493.

[19] Particle Data Group, Eur. Phys. J. C 15 (2000) 1.
[20] G.L. Fogli, E. Lisi, A. Marrone, Phys. Rev. D 65 (2002) 073028;
G.L. Fogli, E. Lisi, A. Marrone, Phys. Rev. D 64 (2001) 093005.
[21] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369;
L. Wolfenstein, Phys. Rev. D 20 (1979) 2634;
S.P. Mikheyev, A.Yu. Smirnov, Sov. J. Nucl. Phys. 42 (1986) 913.
[22] V. Barger, D. Marfatia, K. Whisnant, B.P. Wood, hep-ph/0204253.
[23] J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pea-Garay, hep-ph/0204314.
[24] A. Bandyopadhyay, S. Choubey, S. Goswami, D.P. Roy, hep-ph/0204286.
[25] P.C. Holanda, A.Y. Smirnov, hep-ph/0205241.
[26] C. Athanassopoulos, et al., LSND Collaboration, Phys. Rev. Lett. 81 (1998) 1774.
[27] A. Aguilar, et al., LSND Collaboration, Phys. Rev. D 64 (2001) 112007.
[28] J. Kleinfeller, KARMEN Collaboration, Nucl. Phys. Proc. Suppl. 87 (2000) 281.
[29] E. Church, et al., BooNe Collaboration, nucl-ex/9706011.
[30] M. Sorel, hep-ph/0205207.
[31] A. Donini, M.B. Gavela, P. Hernandez, S. Rigolin, Nucl. Phys. 574 (2000) 23;
A. Donini, M.B. Gavela, P. Hernandez, S. Rigolin, Nucl. Instrum. Methods A 451 (2000) 58, hepph/0007283.
[32] A. Donini, D. Meloni, Eur. Phys. J. C 22 (2001) 179, hep-ph/0105163;
A. Donini, D. Meloni, hep-ph/0107274.
[33] A. Donini, M. Lusignoli, D. Meloni, Nucl. Phys. B 624 (2002) 405;
D. Meloni, hep-ph/0204351.
[34] M. Apollonio, et al., Phys. Lett. B 338 (1998) 383.
[35] S. Geer, Phys. Rev. D 57 (1998) 6989;
S. Geer, Phys. Rev. D 59 (1998) 039903, Erratum.
[36] A. De Rujula, M.B. Gavela, P. Hernandez, Nucl. Phys. B 547 (1999) 21.
[37] K. Dick, M. Freund, M. Lindner, A. Romanino, Nucl. Phys. B 562 (1999) 29.
[38] V. Barger, S. Geer, K. Whisnant, Phys. Rev. D 61 (2000) 053004.
[39] A. Bueno, M. Campanelli, A. Rubbia, Nucl. Phys. B 573 (2000) 27.
[40] A. Cervera, et al., Nucl. Phys. B 579 (2000) 17;
A. Cervera, et al., Nucl. Phys. B 593 (2001) 731, Erratum.
[41] C. Albright, et al., hep-ex/0008064.
[42] A. Blondel, et al., Nucl. Instrum. Methods A 451 (2000) 102.
[43] T. Adams, et al., in: R. Davidson, C. Quigg (Eds.), Proc. of the APS/DPF/DPB Summer Study on the Future
of Particle Physics (Snowmass 2001), hep-ph/0111030.
[44] M. Koike, J. Sato, Phys. Rev. D 61 (2000) 073012;
M. Koike, J. Sato, Phys. Rev. D 62 (2000) 079903, Erratum;
T. Ota, J. Sato, Phys. Rev. D 63 (2001) 093004;
M. Koike, T. Ota, J. Sato, Phys. Rev. D 65 (2002) 053015;
T. Ota, J. Sato, Y. Kuno, Phys. Lett. B 520 (2001) 289.
[45] M. Freund, M. Lindner, S.T. Petcov, A. Romanino, Nucl. Instrum. Methods A 451 (2000) 18;
M. Freund, P. Huber, M. Lindner, Nucl. Phys. B 585 (2000) 105.
[46] M. Freund, P. Huber, M. Lindner, Nucl. Phys. B 615 (2001) 331.
[47] V. Barger, S. Geer, R. Raja, K. Whisnant, Phys. Rev. D 62 (2000) 073002;
V. Barger, S. Geer, R. Raja, K. Whisnant, Phys. Lett. B 485 (2000) 379;
V. Barger, S. Geer, R. Raja, K. Whisnant, Phys. Rev. D 63 (2001) 033002.
[48] A. Bueno, M. Campanelli, A. Rubbia, Nucl. Phys. B 589 (2000) 577.
[49] M. Campanelli, A. Bueno, A. Rubbia, Nucl. Instrum. Methods A 451 (2000) 207.
[50] A. Rubbia, hep-ph/0106088.
[51] A. Bueno, M. Campanelli, S. Navas-Concha, A. Rubbia, Nucl. Phys. B 631 (2002) 239.
[52] P. Lipari, Phys. Rev. D 64 (2001) 033002.
[53] J. Pinney, O. Yasuda, Phys. Rev. D 64 (2001) 093008;
O. Yasuda, hep-ph/0203273.
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
349
J.J. Gomez-Cadenas, D.A. Harris, FERMILAB-PUB-02-044-T.

V.D. Barger, S. Geer, R. Raja, K. Whisnant, Phys. Rev. D 63 (2001) 113011.
V. Barger, et al., hep-ph/0103052.
J.J. Gomez-Cadenas, et al., CERN working group on Super Beams Collaboration, hep-ph/0105297.
H. Minakata, H. Nunokawa, Phys. Lett. B 495 (2000) 369.
H. Minakata, H. Nunokawa, JHEP 0110 (2001) 001.
K. Dick, M. Freund, P. Huber, M. Lindner, Nucl. Phys. B 598 (2001) 543.
P. Huber, M. Lindner, W. Winter, hep-ph/0204352.
P. Zucchelli, hep-ex/0107006.
M. Campanelli, hep-ex/0204004.
J. Burguet-Castell, M.B. Gavela, J.J. Gomez-Cadenas, P. Hernandez, O. Mena, Nucl. Phys. B 608 (2001)
301;
J. Burguet-Castell, O. Mena, hep-ph/0108109.
T. Kajita, H. Minakata, H. Nunokawa, Phys. Lett. B 528 (2002) 245.
V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023.
A. Cervera, F. Dydak, J. Gomez Cadenas, Nucl. Instrum. Methods A 451 (2000) 123.
M. Freund, Phys. Rev. D 64 (2001) 053003.
K. Kimura, A. Takamura, H. Yokomakura, hep-ph/0203099;
K. Kimura, A. Takamura, H. Yokomakura, hep-ph/0205295.
H. Minakata, H. Nunokawa, S. Parke, hep-ph/0204171.
A.M. Dziewonski, D.L. Anderson, Phys. Earth Planet. Int. 25 (1981) 297.
A. Broncano, O. Mena, hep-ph/0203052.
See http://vassilo.home.cern.ch/vassilo/NGS/sig_opera.dat.
C.H. Albright, C. Jarlskog, Nucl. Phys. B 84 (1975) 467.
M. Guler, et al., OPERA Collaboration, CERN-SPSC-2000-028.
D. Autiero, et al., in preparation.
M. Guler, et al., OPERA Collaboration, CERN-SPSC-2001-025.
P. Lipari, M. Lusignoli, D. Meloni, D. Zardetto, in preparation.
E.A. Paschos, J.Y. Yu, Phys. Rev. D 65 (2002) 033002.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 23 (2002) 73.
P. Zucchelli, Ph.D. Thesis, unpublished.
T.K. Gaisser, Cosmic Rays and Particle Physics, Cambridge Univ. Press, Cambridge, 1990.
Nuclear Physics B 646 [PM] (2002) 353497

TFT construction of RCFT correlators I:

partition functions
Jrgen Fuchs a , Ingo Runkel b , Christoph Schweigert b
a Institutionen fr fysik, Universitetsgatan 5, S-65188 Karlstad, Sweden
b LPTHE, Universit Paris VI, 4 place Jussieu, F-75252 Paris cedex 05, France
Received 1 May 2002; received in revised form 2 August 2002; accepted 14 August 2002
Abstract
We formulate rational conformal field theory in terms of a symmetric special Frobenius algebra
A and its representations. A is an algebra in the modular tensor category of MooreSeiberg data of
the underlying chiral CFT. The multiplication on A corresponds to the OPE of boundary fields for
a single boundary condition. General boundary conditions are A-modules, and (generalised) defect
lines are AA-bimodules.
The relation with three-dimensional TFT is used to express CFT data, like structure constants or
torus and annulus coefficients, as invariants of links in three-manifolds. We compute explicitly the
ordinary and twisted partition functions on the torus and the annulus partition functions. We prove
that they satisfy consistency conditions, like modular invariance and NIM-rep properties.
We suggest that our results can be interpreted in terms of non-commutative geometry over the
modular tensor category of MooreSeiberg data.
1. Introduction and summary

The use of two-dimensional conformal field theory in string theory, statistical mechanics
and condensed matter physics has often focussed on issues related to bulk fields on closed
world sheets. But among the multitude of applications of CFT there are also many
like the study of percolation probabilities, of defects in condensed matter systems and
of string perturbation theory in D-brane backgroundsthat require an understanding of
CFT on world sheets with boundary, and in particular of conformally invariant boundary
conditions. These aspects have been investigated intensively over the last few years.
E-mail address: schweige@lpthe.jussieu.fr (C. Schweigert).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 4 4 - 7
354
J. Fuchs et al. / Nuclear Physics B 646 [PM] (2002) 353497
Apart from its important physical applications, the study of boundary conditions is also
considerably contributing to increase our structural insight in conformal field theory.
Further progress can be expected to result from the analysis of defect lines, a subject that
so far has attracted comparatively moderate attention.
In the present paper we make transparent the behavior of rational conformal field
theories on arbitrary (orientable) world sheets, including boundaries and defect lines. This
is achieved by combining tools from topological field theory with concepts from noncommutative algebra, making ample use of two basic facts:
The MooreSeiberg data of a rational chiral CFT give rise to a topological field theory
in three dimensions, and thereby to invariants of links in three-manifolds.
The MooreSeiberg data give rise to a modular tensor category C. One can set up
algebra and representation theory in this category C in very much in the same way as
it is usually done in the categories of vector spaces or of super-vector spaces.
A modular tensor category is actually nothing else than a basis-independent formulation
of the MooreSeiberg data. An important motivation to adopt this framework is the
observation that there exist several rather different mathematical formalisations of the
physical intuition of a chiral conformal field theory, i.e., of the chiral algebra, its space
of physical states and of the properties of chiral vertex operators and conformal blocks
associated to these states. Two prominent examples of such formalisms are the one based
on local algebras of observables on the circle, and hence nets of subfactors, and the
one of vertex algebras. Both frameworks involve quite intricate mathematical structures.
Accordingly, in both settings the explicit treatment of even modestly complicated models
proves to be difficult.
A major problem is to work out the representation theory of the vertex algebras,
respectively to find the (physically relevant) representations of the local algebras of
observables. As a consequence, there have been various attempts to extract the relevant
part of the information about the representation category of the chiral algebra and to
encode it in simpler structures. These attempts have been particularly successful for
rational theories, for which the representation category is semisimple and there are only
finitely many inequivalent irreducible representations. In the present paper we require
the chiral algebra to be rational. However, we do not insist on choosing the maximally
extended chiral algebra. This allows us to deal also with symmetry breaking boundary
conditions, as well as theories for which the left- and right-moving chiral algebras are
different.
The attempts to formalise aspects of the representation theory of rational conformal field
theories have lead, among other results, to new algebraic notions, like truncated quantum
groups (see, e.g., [15]), weak Hopf algebras [69] and double triangle algebras [10,11].
A more direct approach is to formalise the properties of the representation category itself.
This gives rise to the notion of a modular tensor category [12], which we will explain in
355
detail in Section 2, and of module categories [13]. Schematically:

Formalisations of the chiral algebra Structures capturing the representation category
Nets of subfactors
Vertex algebras
Truncated quantum groups

Weak Hopf algebras
Double triangle algebras
Modular tensor categories
By making use of double triangle algebras and weak Hopf algebras, aspects of rational
CFT have been analyzed in [1417]. The present paper develops an approach to rational
CFT that is based on modular categories.
The MooreSeiberg data capture the chiral aspects of rational conformal field theory.
To arrive at a full conformal field theory with local correlation functions, additional input
is required. This can already be seen in the example of a free boson. Here one can choose
u (1) as the chiral algebra, which has a unitary irreducible highest weight representation
for every charge q R. There are many consistent CFTs associated to these chiral data,
for example those describing a free boson compactified on a circle of some given radius.
These models possess (modulo T-duality) in particular different modular invariant torus
partition functions. So the first structure that comes to mind as additional information for
the construction of a full conformal field theory is the choice of a modular invariant for
the bulk theory. However, this proves to be too naive, as one knows of many examples
of modular invariant bilinear combinations of characters that do not arise as the partition
function of any consistent CFT at all [1821].
It is therefore a crucial insight [22] that complete information on how to construct
a full CFT from given chiral data is contained in the structure of a symmetric special
Frobenius algebra in C. (We will explain in detail below what is meant by an algebra A
in C. It is an object, with specific properties, in the category C, and thereby corresponds
to some representation of the chiral algebrathe algebra object A must in particular not
be confused with the chiral algebra V itself.) Already at this point a significant advantage
of tensor categories becomes apparent: once one accepts the idea of doing algebra and
representation theory in the setting of general tensor categories rather than the category
of complex vector spaces, one can directly use standard algebraic and representation
theoretic concepts, like, in the case at hand, the notion of a Frobenius algebra. Indeed,
all mathematical concepts that are needed in the approach to CFT that is developed here
can already be found in standard textbooks on associative algebras [23,24] and category
theory [25,26].
We can show that every symmetric special Frobenius algebra object in the modular tensor category of a chiral CFT leads to a full CFT that is consistent on all orientable world
sheets; Morita equivalent algebras yield the same CFT. Conversely, we establish that every
unitary rational full CFTprovided only that it possesses one boundary condition preserving the chiral algebra1 at alldetermines uniquely a (Morita class of) symmetric special
1 Since the chiral algebra is not required to be maximally extended, the boundary condition is still allowed to
break part of the bulk symmetry. For brevity, in this paper we will sometimes refer to boundary conditions that
preserve the chiral algebra as conformal boundary conditions or also just as boundary conditions.
356
Frobenius algebra(s). The Frobenius algebra in question is actually nothing else than the
algebra of boundary fields associated to one given boundary condition of the theory. It is
associative due to the associativity of the operator product of boundary fields; the nondegenerate bilinear invariant form that turns it into a Frobenius algebra expresses the nondegeneracy of the two-point functions of boundary fields on the disk. Our results can thus
be briefly summarised by saying that we are able to construct the correlation functions of
a unitary rational conformal field theory starting from just one of its boundary conditions.
In fact, every boundary condition of a full CFT gives rise to a symmetric special
Frobenius algebra object, and all such algebra objects are Morita equivalent and hence
lead to one and the same CFT. Moreover, when a given full CFT can be constructed from
a Frobenius algebra A, then any of its boundary conditions gives back an algebra in the
Morita class of A. (On the other hand, we cannot, as yet, exclude the possibility that there
exists a boundary condition M of some CFT C such that the CFT reconstructed from the
Frobenius algebra A that arises from M does not coincide with the original CFT C.)
Apart from its conceptual aspects, the formalism presented in this paper is also
of considerable practical and computational value. The main point is that structure
constantsOPE constants as well as coefficients of the torus and annulus partition
functionsare given as link invariants in three-manifolds. Computing the value of an
invariant is straightforward once we have gathered three ingredients: the MooreSeiberg
data (i.e., the fusing and braiding matrices), the structure constants for the multiplication
of the algebra object A, and the representation matrices describing the action of A on its
irreducible modules.
We will treat the MooreSeiberg data as given. It is of course a non-trivial problem
to obtain these data from a chiral CFT; here, however, we are concerned with building a
full CFT given all the chiral information. Finding an algebra and a multiplication involves
solving a nonlinear associativity constraint. It turns out that this constraint is equivalent
to the sewing constraint for boundary structure constants of a single boundary condition.
Solving this nonlinear equation is not easy, but still much simpler than finding a solution
to the full set of nonlinear constraints involving all boundary structure constants as well
as the bulk-boundary couplings and the bulk structure constants. Finally, finding the
representations of A is a linear problem.
There is a concept that allows us to systematically construct examples of symmetric
special Frobenius algebras: simple currents [27], that is, the simple objects of C with
quantum dimension one. As we will explain elsewhere, for algebra objects that contain only
simple currents as simple subobjects the associativity constraints reduce to a cohomology
problem for abelian groups that can be solved explicitly. Algebras built from simple
currents describe modular invariants of D-type. Often, in particular for all WZW models,
they provide representatives for almost all Morita classes of algebras. It is, however,
a virtue of the formalism developed in this paper that it treats exceptional modular
invariants, including those of automorphism type, on the same footing as simple current
modular invariants. (The structure of full conformal field theories having an exceptional
modular invariant is therefore not really exceptional.)
An important aspect of our construction is that it is possible to prove that the resulting
structure constants of the CFT solve all sewing constraints. From a computational point of
view, the present formalism thus allows us to generate a solution to the full set of sewing
357
Table 1
Physical concepts
Algebraic notions
Boundary conditions
Defect lines
Boundary fields iMN
Bulk fields ij
A-modules
AA-bimodules
HomA (M i, N )

HomA|A (A j ) , (A )+

HomA|A (B1 j ) , (B2 )+
B ,B2
Disorder fields ij 1
constraints from a solution to a small subset of these constraints. This way one can also
check the consistency of boundary conditions that have been proposed in the literature.
Modular tensor categories possess in particular a braiding which accounts for the
braid group statistics of two-dimensional field theories. Thus there is a natural notion
of commutativity with respect to this braiding. It is therefore worth emphasising that
the Frobenius algebras A we consider are not necessarily (braided-) commutative. As
a consequence, our construction constitutes a natural generalisation of non-commutative
algebra to modular tensor categories. This allows us to summarise our results in the
following dictionary between physical concepts in CFT and notions in the theory of
associative algebras (Table 1).
Let us explain the entries of this table in some detail. Boundary conditions will be shown
to be in correspondence with (left) modules of A. In particular, simple modules correspond
to elementary boundary conditions, while direct sums of simple modules indicate the
presence of ChanPaton multiplicities. If M is a left A-module, then for any object i
of C, M i is a left A-module, too. As a consequence, it makes sense to consider left
A-module morphisms from M i to another module N , i.e., morphisms from M i to
N that intertwine the action of A on the two objects. For each such morphism there is
a boundary field changing the boundary condition from M to N and carrying the chiral
label i.
For non-commutative algebras, it is natural to consider not only left (or right) modules,
but also bimodules, i.e., objects that carry an action of A both from the left and from the
right, such that both actions commute. We show that bimodules correspond to (generalised)
defect lines. The trivial defect linei.e., no defect at allis A itself, and the tensor product
(over A) of bimodules is fusion of defect lines.
Given a bimodule B1 , one can endow the object B1 i for any object i of C with the
structure of a bimodule in two different ways: the left action of A is just the one inherited
from the left action of A on B1 , while a right action of A can be defined by using either the
braiding of i and A or the inverse of this braiding. We denote the two resulting bimodules
by (B1 i)+ and (B1 i) .
A particular bimodule is A itself. The degeneracy of a bulk field with chiral labels i
and j is again given by a space of morphismsthe space HomA|A ((A j ) , (A )+ )
of bimodule morphisms. This suggests the following re-interpretation of bulk fields: they
change the trivial defect A to itself. It is therefore natural to generalise bulk fields and
consider fields with chiral labels i, j that change the defect line of type corresponding to
the bimodule B1 to a defect line of some other type B2 . The degeneracy of these disorder
fields is described by the space HomA|A ((B1 j ) , (B2 )+ ) of bimodule morphisms.
358
For each such type of fields there are partition functions that count the corresponding
states. For boundary fields these are linear, for bulk fields bilinear combinations of
characters with non-negative integral coefficients. This way, every full rational conformal
field theory gives rise to a collection of combinatorial dataessentially the dimensions of
the morphism spaces introduced in the table above. Clearly, these data must satisfy various
consistency constraints, both among each other and with the underlying category C, in
particular with the fusion rules of C. Concrete instances of such consistency conditions
have been obtained from a variety of arguments, see, e.g., [14,21,2836]. In particular,
the annulus partition functions provide non-negative integral matrix representations (NIMreps) of the fusion rules of C, while the partition functions of defect line changing operators
give rise to NIM-reps of the double fusion rules.
One important result of the present paper is a rigorous proof of these relations. As
a word of warning, let us point out that, by themselves, the problems of classifying
modular invariants or NIM-reps are not physical problems. (Still, the classification of such
combinatorial data can be a useful auxiliary task.) Indeed, as already mentioned, they tend
to possess solutions that do not describe the partition functions of any conformal field
theory (see, e.g., [1821]). In contrast, in our approach, the partition functions arise as
special cases of correlation functions. Indeed, in a forthcoming publication our approach
will be extended to general amplitudes (compare [22]), and it will be shown that the
system of amplitudes, with arbitrary insertions and on arbitrary world sheets, satisfies all
factorisation and locality constraints, and that they are invariant under the relevant mapping
class groups. This result guarantees that only physical solutions occur in our approach.
A brief outline of the paper is as follows. We start in Section 2 with a review of
some facts about modular tensor categories and topological field theory. In Sections 3
and 4 we investigate symmetric special Frobenius algebra objects and their representation
theory, respectively, and show how these structures arise in conformal field theory. In the
remainder of the paper these tools are employed to deduce various properties of torus and
annulus partition functions and to study defect lines. As an illustration how our approach
works in practice, two examples accompany our development of the general theory: the
free boson compactified at a radius of rational square, and the E7 modular invariant of the
su(2) WZW model.
Some of our results have been announced in [22,37].
2. Modular tensor categories and chiral CFT
2.1. Modular tensor categories
As already pointed out in the introduction, the framework we are going to use is the
one of modular tensor categories [12,26,38,39]. Let us explain in some detail what these
structures are and why it is natural and appropriate to work in this setting.
359
In conformal field theory, modular tensor categories arise in the form of representation
categories of rational vertex operator algebras [4042],2 which in turn constitute a concrete
mathematical realisation of the physical concepts of a chiral algebra and its primary fields.
It has been demonstrated by Moore and Seiberg [46,47] that the basic properties of a
rational chiral conformal field theory can be encoded in a small collection of datalike
braiding and fusing matrices and the modular S-matrixand relations among themlike
the pentagon and hexagon identities. One must be aware, however, of the fact that the
usual presentation of those data implicitly involves various non-canonical basis choices.
As a consequence, the fusing matrices, for instance, enjoy a large gauge freedom, whereas
only their gauge-invariant part has a physical meaning. Posing the MooreSeiberg data in
a basis-free setting leads rather directly to the concept of a modular tensor category. As an
additional benefit, this formulation supplies us with a powerful graphical calculus.
In the sequel we start out by reviewing details of the mathematical machinery that is
required to understand modular tensor categories; only afterwards we return to the origin
of these structures in rational conformal field theory. A category C consists of two types of
data: A class Obj (C) of objects and a family of morphism sets Hom(U, V ), one for each
pair U, V Obj (C). In the situation of our interest, the objects are the representations of
the chiral algebra V of the CFT (a rational vertex operator algebra), and the morphisms are
the intertwiners between V-representations.
Morphisms can be composed when the relevant objects match, i.e., the composition g
f of f Hom(U, V ) and g Hom(Y, Z) exists if Y = V . This operation of composition
is associative, and for every object U the endomorphism space End(U ) Hom(U, U )
contains a distinguished element, the identity morphism idU , satisfying g idU = g for
all g Hom(U, V ) and idU f = f for all f Hom(Y, U ). The categories C of our
interest are complete with respect to direct sums (this can always be assumed without loss
of generality) and come enriched with quite a bit of additional structure; we introduce this
structure in three steps.
First, C is a semisimple Abelian strict tensor category with the complex numbers as
ground ring.
Let us explain the various qualifications appearing in this statement. Abelianness [25,
Chapter VIII] means that there is a zero object 0 and the morphisms possess various
natural properties familiar from vector spaces. Concretely, every morphism set is an abelian
group, and composition of morphisms is bilinear; every morphism has a kernel and a
cokernel (they are defined by a universal property); every monomorphism is the kernel of
its cokernel, and every epimorphism is the cokernel of its kernel; finally, every morphism f
can be written as the composition f = h g of a monomorphism h and an epimorphism g.
In a tensor category (see, e.g., [25, Chapter VII] or [26, Chapter XI]) there is a tensor
product , both of objects and of morphisms. The tensor product on objects has a unit
element, which is denoted by 1; for f Hom(U, Y ) and g Hom(V , Z), the tensor
2 It is still a conjecture that the representation category of every rational VOA is modular. There is no general
proof, but the property has been established for several important classes of VOAs, compare, e.g., [4345],
and it is expected that possible exceptions should better be accounted for by an appropriate refinement of the
qualification rational.
360
product morphism f g is an element of Hom(U V , Y Z). The endomorphisms

End(1) of the tensor unit 1 form a commutative ring k, called the ground ring, and every
morphism set is a k-module. The operations of composition and of forming the tensor
product of morphisms are bilinear and compatible in an obvious manner. In the present
context we require that the ground ring is the field of complex numbers, k = C; then the
morphism sets are complex vector spaces.
In any tensor category there is a family of isomorphisms between U (Y Z) and (U
Y ) Z, with U, Y, Z any triple of objects, and families of isomorphisms between U 1
and U as well as between 1 U and U , for any object U . They are called associativity and
unit constraints, respectively, and are subject to the so-called pentagon identity (assuring
that any two possibilities of bracketing multiple tensor products are related via the
associativity constraints) and triangle identities (compatibility between associativity and
unit constraints). A tensor category is called strict if all these isomorphisms are identities,
so that the tensor product of objects is associative and 1 U = U = U 1. By the
coherence theorems [25, Chapter VII.2], there is no loss of generality in imposing this
strictness property. On the other hand, when dealing with certain other structures below,
we will often have to be careful not to mix up equality and isomorphy of objects.
Finally, the meaning of semisimplicity is analogous as for other algebraic structures.
A simple (or irreducible) object U of an abelian tensor category is an object whose
endomorphisms are given by the ground ring, End(U ) = k idU , i.e., End(U ) = C idU for
the categories considered here;3 in particular, the tensor unit 1 is automatically simple.
A semisimple category is then characterised by the property that every object is the direct
sum of finitely many simple objects.
Semisimplicity of a tensor category C implies in particular dominance of C, which means
that there exists a family {Ui }iI of simple objects with the following property: for any
V , W Obj (C) every morphism f Hom(V , W ) can be decomposed into a finite sum

f=
(2.1)
gr hr
r
with
hr Hom(V , Ui )
and gr Hom(Ui , W )
(2.2)
for suitable members Ui = Ui (r) (possibly with repetitions) of this family.

Since in the categories we are considering, the morphism sets are vector spaces (over C),
it is convenient to introduce a shorthand notation for their dimension:
dim Hom(X, Y ) =: X, Y
(2.3)
for X, Y Obj (C). As a consequence of semisimplicity we have X, Y = Y, X.

3 In a general category, this property rather characterises an absolutely simple object, while simplicity of
an object means that it does not possess a non-trivial proper subobject. In semisimple categories, absolutely
simple implies simple, and in any Abelian category over an algebraically closed ground field the two notions are
equivalent.
361
A convenient way to visualise morphisms in an Abelian tensor category is via graphs in

which lines stand for identity morphisms; thus idU and f Hom(U, V ) are depicted as
(2.4)
We follow the convention that such pictures are read from bottom to top. Because
of End(1) = C we have id1 = 1 C; accordingly, lines labelled by the tensor unit 1
can and will be omitted, so that in the pictorial description morphisms in Hom(1, U )
or Hom(U, 1) emerge from and disappear into nothing, respectively. Composition of
morphisms amounts to concatenation of lines, while the tensor product corresponds to
juxtaposition:
(2.5)
Second, C is a ribbon category, that is [26, Chapter XIV.3], a strict tensor category
supplemented with three additional ingredients: a duality, a braiding, and a twist.
A (right-)duality on a tensor4 category C associates to every U Obj (C) another object
U Obj (C) and morphisms

dU Hom U U, 1 ,
bU Hom 1, U U ,
(2.6)
and to every morphism f Hom(U, Y ) the morphism

f := (dY idU ) (idY f idU ) (idY bU ) Hom Y , U .
(2.7)
is called the object (right-)dual to U , and

the morphism (right-)dual to f ;
the duality morphisms dU and bU are also known as the evaluation and co-evaluation
morphisms, respectively. A braiding on a tensor category C allows one to exchange
objects; it consists of a family of isomorphisms cU,V Hom(U V , V U ), one for
each pair U, V Obj (C). Finally, a twist is a family of isomorphisms U , one for each
U Obj (C). Graphically, the braiding and twist and their inverses and the duality will be
4 Often the existence of a duality is included in the definition of the term tensor category. What we refer to
as a tensor category here is then called a monoidal category.
362
depicted as follows:
(2.8)
Actually we should think of these morphisms as ribbons rather than linesthis is the
reason for the terminology. For example, the twist , braiding c and duality morphism
b are drawn as
(2.9)
In the sequel the interpretation of graphs with lines as ribbon graphs will be implicit.
The duality, braiding and twist are subject to a number of consistency conditions, which
precisely guarantee that the visualisation via ribbons is appropriate, so that in particular
the graphs obtained by their composition share the properties of the correspondingly glued
ribbons. More concretely, one has to impose duality identities, functoriality and tensoriality
of the braiding, functoriality of the twist, and compatibility of the twist with duality and
363
with braiding. In the notation of (2.8) these properties look as follows:
(2.10)
(2.11)
In a ribbon category there is automatically also a left duality; it is defined on objects by
U := U and left duality morphisms b and d ; the latter, as well as left-dual morphisms
U
U
364
f ,
are given by
(2.12)
One can check that this left duality coincides with the right duality not only on objects, but
also on morphisms, i.e., f = f ; categories with a coinciding left and right duality are
called sovereign. It follows, e.g., that the double dual (U ) of an object U is isomorphic
(though in general not equal) to U . In fact, a natural isomorphism between (U ) and U
can be obtained with the help of the twist (see, e.g., [39, Chapter 2.2]): for any object U
we have a morphism U := U1 U Hom(U, (U ) ), with

1
,U .
U := (idU dU ) idU c(U
) ,U (bU id(U ) ) Hom U
(2.13)
The properties of the twist U are precisely such that these morphisms are tensorial, i.e.,
satisfy V W = V W , preserve the unit, 1 = id1 , and are compatible with the duality
in the sense that U = ((U ) )1 .
Further, once we have two dualities, we can also define left and right traces of
endomorphisms, via
(2.14)
Both traces are cyclic,
trl,r (g f ) = trl,r (f g)
(2.15)
and obey
trl,r (f g) = trl,r (f )trl,r (g).
(2.16)
In the case at hand, where the left duality is constructed from the right duality by (2.12),
the two notions of trace coincide; thus the category C is spherical [48]. The trace of the
365
identity morphism is known as the quantum dimension of an object,

dim(U ) := tr(idU ).
(2.17)
The quantum dimension is additive under direct sums and multiplicative under tensor
products.
For any self-dual object U of a sovereign tensor category and any isomorphism f in the
space Hom(U, U ) one introduces the endomorphism VU Hom(U , U ) as ([36], see
also [4951])
(2.18)
One can show [36] that VU is in fact an automorphism, and that for simple self-dual U it
does not depend on f and satisfies
VU = U idU
(2.19)
U {1}.
(2.20)
with
The sign U is called the FrobeniusSchur indicator of the object U . In agreement with the
terminology in the representation theory of groups and Lie algebras, self-dual objects U
with U = 1 are called real (or orthogonal), while those with U = 1 are called pseudoreal (or symplectic, or quaternionic).
Third, C is a modular tensor category (or, briefly, modular category), that is, a
semisimple Abelian ribbon category with ground field5 C that has only a finite number
of isomorphism classes of simple objects and a non-degenerate s-matrix.
To explain the latter property, first note that in a modular tensor category the family of
simple objects that appears in the definition of dominance (see formula (2.2)) can be chosen
such that it contains precisely one object out of each isomorphism class of simple objects
and hence in particular is finite, |I| < . Also, the dual U of any member U of the
, and as representative of the class of the tensor
family is isomorphic to another member U
= U iff U is self-dual; in particular, 1 = 1. From
unit 1 we choose 1 itself. Note that U
now on we denote the elements of the family by Ui with i {0, 1, 2, . . ., |I| 1} and set
U0 = 1.
5 In the original definition of a modular tensor category [12], semisimplicity is replaced by dominance and
instead of abelianness also only a weaker property is imposed, nor is it required that the ground ring is C. But in
the context of rational conformal field theory modular tensor categories appear naturally in this more restricted
form.
366
One then defines an (|I| |I|)-matrix s = (si,j )i,j I by

si,j := tr(cUi ,Uj cUj ,Ui ),
(2.21)
or pictorially,
(2.22)
(Here and in the sequel we often simplify notation by writing the label i in place of Ui .)
The final requirement to be imposed on C in order that it is a modular category is that
this square matrix s is non-degenerate. This latter property provides in fact an explanation
of the qualification modular: when combined with the other axioms, it implies that the
matrices s and t = diag(Ui ) (when multiplied with overall constants that are expressible
through the Uj and the quantum dimensions dim(Uj )) generate a projective representation
of the modular group SL(2, Z).
The tensor product of objects induces on the set of isomorphism classes of C the structure
of a commutative and associative ring over the integers, called the Grothendieck ring K0 (C)
of C. (Conversely, the tensor category C may be thought of as a categorification of the ring
K0 (C) [52].) A distinguished basis of this ring is given by the isomorphism classes of the
objects Ui with i I. In this basis, the structure constants are the non-negative integers
dim Hom(Ui Uj , Uk ).
k is an involution on the finite set I, which induces an involution
The mapping Uk U
k . With this convention, U is isomorphic to Uk , for

k k on the set of labels by Uk = U
k
every k I; let us then fix an isomorphism

k Hom Uk , Uk
(2.23)
for each k I, and for each pair k, k with k = k perform the choice in such a way that the
number pk defined by
(2.24)
is equal to 1. Using sovereignty of C, one can then show that for these values of k we have
consistently pk = 1, too. In contrast, for the self-dual objects Uk = Uk in the family, we
are not free to choose the number pk ; rather, it follows directly from its definition in (2.24)
that pk coincides with the FrobeniusSchur indicator (2.20) of Uk :
pk = k
for Uk
= Uk .
(2.25)
367
A particularly simple example of a modular tensor category, which can serve as a guide to
the general theory, is the category V of finite-dimensional vector spaces over the complex
numbers. The category V has a single isomorphism class of simple objectsthe class of
the one-dimensional vector space Cand has trivial twist and braiding. In conformal field
theory, this category arises for meromorphic models, i.e., models with a single primary
field, such as the E8 WZW theory at level 1.
2.2. Fusing and braiding matrices
Let us now explain the meaning of the various properties of a modular category
in the conformal field theory context. The (simple) objects of C are the (irreducible)
representations of the chiral algebra V, and the morphisms of C are V-intertwiners.
The tensor product is the (fusion) tensor product of V-representations, with the tensor
unit given by the vacuum representation (identity field and its descendants). Thus the
isomorphism classes of simple objects correspond to the primary chiral vertex operators,
and the Grothendieck ring of C is the fusion ring of the conformal field theory.
The duality in C encodes the existence of conjugate V-representations, and the twist is
determined by the fractional part of the conformal weight:
U = exp(2iU ) idU
(2.26)
for simple objects U . The braiding of C accounts for the presence of braid group statistics
(see, e.g., [5356]) in two dimensions, and the matrix s coincides, up to normalisation, with
the modular S-matrix of the CFT that implements the modular transformation 1/
on the characters of primary fields:
si,j =
Si,j
.
S0,0
(2.27)
(Conversely, S0,0 and thereby S is recovered from the data of the modular tensor category
by requiring S = S0,0 s to be unitary.) In terms of s, the quantum dimensions are
dim(Ui ) = si,0 =
Si,0
.
S0,0
(2.28)
All the axioms of C can be understood as formalisations of properties of primary chiral

vertex operators in rational CFT. Often such properties are presented in a form where
explicit basis choices in the three-point coupling spaces have been made. To make contact
with such a formulation we fix6 once and for all bases {(i,j
)k } in the coupling spaces
(i,j )k
Hom(Ui Uj , Uk ), as well as dual bases {
} in Hom(Uk , Ui Uj ). We depict the
6 Our strategy is to keep these bases as general as possible. More specific basis choices can be interesting for
purposes different from ours, e.g., for an efficient numerical encoding of the defining data of a modular category
as discussed in [57,58].
368
basis morphisms as follows:
(2.29)
Duality of the bases means that
(2.30)
By the dominance property of C we also have the completeness relation
(2.31)
When all three labels i, j, k are generic, no basis in Hom(Uk , Ui Uj ) or Hom(Ui
Uj , Uk ) is distinguished. In contrast, when one of the labels equals 0, i.e., when one of the
simple objects involved is the tensor unit 1, then it is natural to make the choice
(i,0)i
= (0,i)i
= idUi =
(i,0)i
(0,i)i
(2.32)
(which is possible due to strictness of the tensor category C). Here we have used the
symbol in order to indicate that the coupling label can take only a single value. In the
sequel, for notational simplicity we suppress such unique labels, both in the formulas and
369
in the pictures. Thus, e.g., the pictorial form of the relation (2.32) is
(2.33)
k Uk , 1) are one-dimensional, the morphisms

Since the spaces Hom(U
and (0,k)k
(k,k)0
are proportional to the respective combinations of dualities and s (as defined in (2.23)),
i.e., there are numbers k and k such that
(2.34)
The normalisation condition (2.30) implies that the constants of proportionality are related
by
k k = (dim Uk )1 .
(2.35)
Recall now that C is a strict tensor category, i.e., that the tensor product of objects
is strictly associative. Nevertheless, once we have chosen bases as above, there are
two distinct distinguished bases for the
morphism space Hom(Ui Uj Uk , Ul ),
corresponding
to
its
two
decompositions
q Hom(Ui Uj , Uq ) Hom(Uq Uk , Ul )

and p Hom(Uj Uk , Up ) Hom(Ui Up , Ul ), respectively. The coefficients of the
basis transformation between the two are known as the fusing matrices, F-matrices, or 6j symbols of C. We denote them as follows:
(2.36)
370
By composing with the morphism dual to the one on the right-hand side, we arrive at the
formula
(2.37)
for the F-matrices. When any of the labels i, j, k equals 0, then the left-hand side of formula
(2.37) degenerates (if non-zero) to the left-hand side of (2.30), leading to
(0 j k) l
Fl,j
= , ,
(i 0 k) l
Fk,i = , ,
(i j 0) l
Fj , l = , .
(2.38)
For the morphisms dual to those appearing in (2.36), there is an analogous relation
(2.39)
It is convenient to introduce a separate symbol G for the inverse of F. It is defined as
371
(2.40)
Combining the braiding morphisms with the basis choice (2.29) provides the braiding
matrices R:
(2.41)
1
The number that is obtained when the braiding ci,j (over-braiding) is replaced by cj,i
(i j )k
(under-braiding) is denoted by R
+ (i j )k
R
.
(i j )k
, and instead of R
one also often writes
One easily checks that
(i j )k (j i)k
R R
= ,
(2.42)
and, using the compatibility of twist and braiding and functoriality of the twist (see (2.11)
and (2.10)), that
(i j )k (j i)k
R R
k
.
i j ,
(2.43)
Here the complex number k is defined by Uk =: k idUk , i.e., specifies the twist of the
simple object Uk . We can express k in terms of special matrix elements of F and R. To
372
this end one rewrites the twist morphism of the object Uk as
(2.44)
and then computes the constant by which the right-hand side differs from idUk by first
using the identity (2.34), then (2.41) and finally (2.37) (with p = q = 0, i = k and j = k).
The result is
(k k k) k (k k)0
k = dim(Uk )F0 0
(2.45)
The FrobeniusSchur indicator of a self-dual simple object Uk is encoded in the F matrix

as well. To see this take formula (2.37) with p = q = 0, i = k = l and j = k for a not
necessarily self-dual Uk . Then apply (2.34) and (2.35) to the left-hand side and use (2.24)
to cancel the morphisms k . The resulting relation reads
k k) k
= pk k /k .
dim(Uk )F(k
00
(2.46)
Finally specialise to k = k and employ (2.25) to arrive at

k k) k
.
k = dim(Uk )F(k
00
(2.47)
Also note that combining (2.45) (or rather, the analogous equation obtained when using
the inverse braiding) and (2.46) one has, for any simple Uk ,
R(k k) 0 = pk k1 k /k .
(2.48)
For comparison with the literature, we note that our convention for the F- and R-matrices
is related to the one of [47] by

j k
(j k l) i
Fp q
=
Fp,q
jki
and R(i j )k =
(2.49)
i l
(compare formula (2.4) and Example 2.8 in [47]). Note that often also the composite
quantities B F + are used, see, e.g., formula (3.3) in [47].
2.3. Three-manifolds and ribbon graphs
To determine a correlation function of a rational CFT, we specify it as a particular
element in the relevant space of conformal blocks. A very convenient characterisation
of conformal blocks is via ribbon graphs in three-manifolds. In this formulation, the
coefficients in the expansion of a CFT correlator in terms of a chosen basis of conformal
blocks are obtained as invariants of closed three-manifolds with embedded ribbon graphs.
373
To explain this construction we need to introduce the concepts of a ribbon graph and of
a three-dimensional topological field theory. (For more details see, e.g., [12,26,39,5967];
this quick introduction follows Section 2 of [51].)
A ribbon graph consists of an oriented three-manifold M, possibly with boundaries,
together with embedded ribbons and coupons. A ribbon is an oriented rectangle, say
[1/10, 1/10] [0, 1], together with an orientation for its core {0} [0, 1]. The two
subsets [1/10, 1/10] {0} and [1/10, 1/10] {1} are the ends of the ribbon. A coupon
is an oriented rectangle with two preferred opposite edges, called top and bottom. The
embeddings of ribbons and coupons into M are demanded to be injective. A ribbon minus
its ends does not intersect any other ribbon, nor any coupon, nor the boundary of M.
A coupon does not intersect any other coupon nor the boundary of M. The ends of a
ribbon must either lie on one of the preferred edges of some coupon or on M. For ribbons
ending on a coupon, the orientation of the ribbon and of the coupon must agree.
Choosing an orientation for the ribbons and coupons is equivalent to choosing a
preferred sidehenceforth called the white sidewhich in the drawings will usually
face the reader. The opposite (black) side is drawn in a darker shade, as has already been
done in figure (2.9) above. We use open arrows to indicate the orientation of the ribbons
core.
Each constituent ribbon of a ribbon graph is labelled (sometimes called colored) by a
(not necessarily simple) object of the modular category C, and each coupon is labelled by
a morphism of C. Which space the morphism belongs to depends on the ribbons ending
on the coupon. If a ribbon labelled by U is incoming, the relevant object is U when the
ribbon is attached to the bottom of the coupon, and U when it is attached to its top. If it
is outgoing, then the convention for the object is the other way round. As an illustration,
the coupon in
(2.50)
is labelled by an element Hom(U V , X Y Z).
Consider now a ribbon graph in S 3 . We can assign an element in Hom(1, 1), i.e.,
a complex number, to it as follows: regard S 3 as R3 {}, with R3 parametrised by
Cartesian coordinates (x, y, z). Deform the ribbon graph such that all its coupons are
rectangles in the (xy)-plane, with white sides facing upwards, and such that the top and
bottom edges are parallel to the x-axis, with the bottom edge having a smaller y-coordinate
than the top edge. Deform the ribbons to lie in a small neighbourhood of the (xy)-plane.
In short, the ribbon graph must be arranged in such a manner that the bends, twists and
crossings of all ribbons can be expressed as dualities, twists and braidings as appearing
in (2.8) and (2.9). The element in Hom(1, 1) is then obtained by reading the graph from
y = to y = + and interpreting it as a concatenation of morphisms in C. Let us
374
display an example:
(2.51)
One of the non-trivial results following from the defining relations of a modular category
is that different ways to translate a ribbon graph into a morphism give rise to one and the
same value for the ribbon graph. In other words, the value for the ribbon graph is invariant
under the various local moves that transform those different descriptions into each other.
All pictures in Sections 2.54.5 directly stand for morphisms in C. The first genuine
ribbon graphs will occur in Section 5.1. We will usually simplify the pictures involving
ribbon graphs by replacing all ribbons by lines. When doing so, it is understood that the
ribbon lies in the plane of the paper, with the white side facing up (this convention is
known as blackboard framing). Also note that, strictly speaking, the definition of a ribbon
graph given above forbids annulus-shaped ribbons. Whenever such a ribbon occurs below
it is understood to be of the form displayed in (2.51), with f the identity morphism in
Hom(U, U ).
2.4. Topological field theory
So far we have used the modular category C only to assign numbers to ribbon graphs
in S 3 ; for this purpose it is actually sufficient that C is a ribbon category. That C is even
modular implies the highly non-trivial result that it gives rise to a full three-dimensional
topological field theory. By definition, a three-dimensional TFT is a pair (Z, H) of
assignments that associate algebraic structures to geometric dataextended surfaces and
cobordismsand satisfy various properties, to be outlined below.
By an extended surface we mean an oriented closed compact two-manifold X with a
finite number of disjoint oriented arcs (the remnants, at the topological level, of the local
coordinates that one must choose around the insertion points of the world sheet) labelled by
pairs (U, ) with U Obj (C) and {1}, and with a lagrangian subspace L(X) of the
first homology group H1 (X, R). The opposite X of an extended surface is obtained from
X by reversing the orientation of X, reversing the orientation of all arcs and replacing
by . To make explicit the insertions U, V , . . . of an extended surface we will sometimes
write (U, V , . . . ; X) instead of X. This notation assumes that all signs are +1, but does
not encode the positions and orientations of the arcs; when the sign for some object is
1, then we write (U, ) instead of U .
The first datum, H, of the TFT assigns to each extended surface X a finite-dimensional
vector space H(X). This space is constructed from appropriate tensor products of
375
morphism spaces of C; one has

H() = C.
(2.52)
The CFT interpretation of H(X) is as the space of conformal blocks on X. Thus we have,
for example, dim H(k; S 2 ) = k,0 , in agreement with the fact that the space of one-point
blocks on the sphere is zero-dimensional except for the insertion of a vacuum. Similarly,
dim H(k, (;, ); S 2)= k,; , i.e., the two-point blocks on the sphere vanish unless the two
fields are dual to each other. More generally,

dim H i, j, (k, ); S 2 = Nijk
(2.53)
(recall that Nijk = dim Hom(Ui Uj , Uk ) = dim Hom(Ui Uj Uk , 1)). Each homeomorphism f : X Y of extended surfaces induces a linear map f< : H(X) H(Y).
Consider now a manifold M with ribbon graph whose boundary can be written as a
disjoint union M = 1 M 2 M. Turn 1 M and 2 M into extended surfaces by taking
as arcs the ends of ribbons, with orientation induced by the ribbons. In addition fix a
Lagrangian subspace (this is part of the defining data of an extended surface; it is not
determined by M and the embedded ribbon graph). When a ribbon ending on M is
labelled by U , then the corresponding arc is labelled by (U, +1) if the core of the ribbon
points away from the surface, and by (U, 1) otherwise. Denote the extended surface 2 M
by + M and 1 M by M. Then the triple (M, M, + M) is called a cobordism from
M to + M.
The second datum, Z, of the TFT assigns a linear map
Z(M, M, + M) : H( M) H(+ M)
(2.54)
to every cobordism. Let us mention two of its properties. The first concerns the
normalisation of Z. Let X be an extended surface and M the manifold X [0, 1] with
embedded ribbon graph given by straight ribbons connecting the arcs in X {0} to the arcs
in X {1}, with cores oriented from 0 to 1. Then
Z(M, X, X) = idH(X).
(2.55)
The second property is functoriality. Let M1 and M2 be two three-manifolds with ribbon
graphs and let f : + M1 M2 be a homeomorphism of extended surfaces, and let M
be the manifold obtained from glueing M1 to M2 using f . Then
Z(M, M1 , + M2 ) = m Z(M2 , M2 , + M2 ) f< Z(M1 , M1 , + M1 ),

(2.56)
where m is an integer (see the review in [51] for details) and = S0,0 j I j1 dim(Uj )2 ,
which is called the charge of the modular category C.
In our application we will not need the properties of Z in their most general form. But
we will use the following special cases and consequences: for the manifolds S 2 S 1 and
S 3 without ribbon graph we have

Z S 2 S 1 ; ; = 1 and Z S 3 ; ; = S0,0 .
(2.57)
376
More generally, for S 3 with embedded ribbon graph, the number assigned by Z is S0,0
times the number obtained by translating the ribbon graph to a morphism of C as described
in the previous section.
Functoriality of Z implies that the invariant of a ribbon graph in any closed threemanifold can be related to an invariant in S 3 by the use of surgery along links. This will be
used in Sections 5.2 and 5.7 to relate invariants of S 2 S 1 and S 3 .
We will also make intensive use of a trace formula that is obtained as follows: for X an
extended surface, consider the three-manifold N = X [0, 1] with embedded ribbon graph
such that N = + N = X as extended surfaces. Let M be the closed three-manifold with
ribbon graph obtained from N by identifying (x, 0) with (x, 1) for all x X. Then
Z(M, , ) = trH(X) Z(N, X, X).
(2.58)
This trace formula, too, is a consequence of the functoriality of Z.

2.5. The case Nij k {0, 1}
In this section we specialise the general treatment above to the case that the dimensions
Nij k of all coupling spaces Hom(Ui Uj , Uk ) are either 0 or 1. This greatly simplifies
both notation and calculation. It may be thought of as a meta-example; in particular, it
encompasses both concrete examples that we study in this paper.
The main notational simplification is that the multiplicity indices disappear, so that the
fusing and braiding matrices take the form
(i j k) ;
Fp q
and R(i j )k ,
(2.59)
respectively. (But we must still choose bases in the morphism spaces, and the form of
the fusing and braiding matrices does depend on this choice.) The quantum dimensions
dim(Uk ) (which in a modular category are positive real numbers) can be obtained from F
and R, by taking the absolute value of (2.45) (since k is a phase). The twist eigenvalues k
follow from (2.45) as well, and finally the FrobeniusSchur indicators are given by (2.47).
(In practice, it is often easier to obtain dim(Uk ) and k by some other means, though.)
To reconstruct the S-matrix from F and R, as well as for later use, the following
identities prove to be useful:
377
(2.60)
as well as
(i j k) ;
Gp q
R(j k)q R(i q); (k j i) ;

Fp q
.
R(i j )p R(p k);
(2.61)
The first equation, for instance, follows by composing with the three-point coupling dual
to the right-hand side and noticing that the left-hand side thereby becomes a special case
(q = 0) of the graph on the left-hand side of (2.37) that gives the general F-matrix element.
The last equation relating the F-matrix to its inverse can be seen by following the sequence
of moves indicated below (this is nothing but the hexagon identity):
(2.62)
The S-matrix can then be expressed through the F- and R-matrices by the moves in the
following figure:
378
(2.63)
Using now the two relations in (2.60) as well as (2.43), this results in
Si,j = S0,0 dim(Ui ) dim(Uj )
k ( j i) i ( j i) i
G
Fk 0
.
i j 0 k
(2.64)
kI
Recall that we chose the basis in the spaces of three-point couplings in such a way that
an F-matrix is equal to one (see (2.38)) when one of the ingoing objects is the tensor
unit 1. It is convenient to have a similar behavior when the outgoing object is 1. The
(i j k) 0
= pi pj pk :
following lemma implies that we can always make choices such that F k
Lemma 2.1. If dim Hom(X Y, Z) {0, 1} for all simple objects X, Y, Z Obj (C), then
there is a choice of basis in the spaces of three-point couplings such that
(i j k) ;
Fp q
=1
(2.65)
whenever the F-matrix element is allowed to be non-zero by the fusion rules and one or
more of the i, j, k are 0, and that
(i j k) 0
Fij k F k
= pi pj pk .
(2.66)
Proof. Recall the conventions (2.33) and (2.34) that we have made already. For the sake
of this proof, we will specialise the choice of basis further such that k = k . By definition
of F we have
(2.67)
379
Then we have the relation
(2.68)
which when iterated results in
Fkij Fj ki Fij k = pi pj pk .
(2.69)
When the three coupling spaces that form an orbit upon iteration, i.e., Hom(i j, k),
Hom(k i, ) and Hom(j k, ), are mutually distinct, then we can link the choice of
basis in the latter two to the choice for the first in such a manner that Fkij = Fj ki = pi pj pk .
Relation (2.69) then implies that Fij k = pi pj pk as well.
That the three coupling spaces are not mutually distinct can happen only if i = j = k.
Let us abbreviate from here on Fiii by Fi . Define the two numbers Bi and Ri via
(2.70)
The number Bi exists because the space Hom(Ui Ui Ui , 1) has dimension one and Ri
is just R(i i) , compare formula (2.41).
Next note the relations
Fi Fi Fi = pi ,
Fi = pi i Bi Ri ,
Bi Bi = 1/i ,
Ri Ri = 1/i .
(2.71)
The second relation, for instance, is obtained by the moves (we omit the obvious labels)
380
(2.72)
together with formula (2.48) for
R(i )0 .
Using the relations (2.71), it follows that
Fi Fi = (pi i Bi Ri )2 = 1,
(2.73)
and thus pi = Fi Fi Fi = Fi , which proves the lemma.
2.5.1. Example: Free boson

The first example is the modular category associated to the chiral data of a free boson
field compactified on a circle of rational radius squared. The generic chiral algebra of this
theory is u (1). If the compactification radius R fulfills (in the normalisation we choose)
R 2 = p/q with p, q coprime, then u (1) can be extended by a pair of Virasoro primary
fields of weight N = pq.
The modular category for the MooreSeiberg data of this theory can be presented as
follows (see, e.g., [68]). There is a simple object Uk for each integer k. The simple objects
Uk and Uk+2N are isomorphic, so that the category has 2N isomorphism classes of simple
objects; we label them as I = {0, 1, . . . , 2N 1}. (Note that we should not ignore these
for instance,
in line with
isomorphisms and pretend that Uk and Uk+2N are equal. This is,
the fact that the operator product of two fields of u(1) charge j/ 2N and k/ 2N in the
standard range, i.e., j, k I, does not contain any field with charge in the standard range
when j + k 2N .) The conformal weight of the primary field labelled by k I, whose
non-integral part determines the twist of the object Uk , is
1
k2
for k N,
(k) = 4N
(2.74)
1
2 for k > N.
(2N
k)
4N
The representation k of the
extended chiral algebra contains u (1)-representations
of charges q(k) = (k + 2Nm)/ 2N for every integer m (in the convention (k) =
(q(k))2 /2). From the conformal weights we read off that
k = e2i(k) = eik
2 /2N
(2.75)
To give the braiding and fusing matrices, it is useful to introduce the function [] : Z I
that associates to n the element of I with which it coincides modulo 2N , i.e., [n] =
n 2N (n) with (n) (n [n])/2N the unique integer such that n 2N (n) is in I.
Then the fusion rules, which furnish the Abelian group Z2N , read
[i] E [j ] = [i + j ],
(2.76)
and with our conventions for the coupling spaces the non-zero braiding and fusing matrices
are given by
(i j k) [i+j +k]
F[j +k] [i+j ]
= (1)(i+k+1){j (i+j +k)+(j +k)( (i+j )+ (j +k))},
R(k ;)[k+;] = (1)(k+;) (k+;)eik;/(2N) .
381
(2.77)
Two points should be noted: first, had we chosen another way to represent the isomorphism
classes of simple objects, these formulas would look different. In particular, they depend
on our specific choice for the set I, and indeed are not invariant under the shift n
n + 2N . Second, by redefining the bases of the coupling spaces Hom(j k, [j + k])
by a factor (1)j (k+1) (j +k), the Fs could be made to look much simpler, namely
(i j k) [i+j +k]
F[j +k] [i+j ] = (1) (j +k)i , which is the form used in [69]. This gauge does not, however,
(i j k) 0
= pi pj pk (which we use in simplifying the explicit expressions in the

fulfill F k
examples treated), whereas (2.77) does.
2.5.2. Example: sl(2) WZW models
The chiral algebra of the sl(2) WZW model at level k is the affine Lie algebra sl(2)k .
It has k + 1 isomorphism classes of integrable highest weight modules, which we label
by their highest weights (twice the spin): (0), (1), . . . , (k). The corresponding conformal
weights read
(n) =
n(n + 2)
,
4(k + 2)
(2.78)
and the quantum dimensions are

(n + 1)
dim(n) = S0,n /S0,0 = sin

sin
.
k+2
k+2
(2.79)
The fusion rules read

min(m+n,2kmn)
(m) E (n) =
(p),
(2.80)
p=|mn|
where p is increased in steps of 2.

The concrete expressions for the braiding and fusing matrices depend on the chosen
normalisation of the chiral vertex operators. We will use the form in which they naturally
arise as quantum group 6j -symbols, as given, e.g., in [70,71]:
R(rs)t = (1)(r+st )/2ei((t )(r)(s)),
s t) u
F(r
=
pq
with
a
d
b
c
e
f
:= (1)
t/2
r/2
s/2 p/2
u/2 q/2
(2.81)
q

q
a+bcd2e
[2e + 1][2f + 1]
1/2
(a, b, e)(a, c, f )(c, e, d)(d, b, f )

(1)z [z + 1]! [z a b e]![z a c f ]![z b d f ]!
382
[z d c e]![a + b + c + d z]![a + d + e + f z]!

1
[b + c + e + f z]!
and
(a, b, c) :=
[a + b + c]![a b + c]![a + b c]!/[a + b + c + 1]! .
The symbols [n] and [n]! stand for q-numbers and q-factorials, respectively, i.e.,
n
n

sin k+2
[n] = ,
[m],
[0]! = 1.
[n]! =
sin k+2
m=1
(2.82)
(2.83)
(2.84)
(The fusion rules ensure that the numbers that appear as arguments of [] are always
integral.) The range of the summation is such that the arguments are non-negative, i.e.,
z runs over all integers (in steps of 1) from max(a + b + e, a + c + f, b + d + f, c + d + e)
to min(a + b + c + d, a + d + e + f, b + c + e + f ).
It is worth pointing out that, while these expressions solve the pentagon and
hexagon identities, to find the normalisation of WZW conformal blocks such that their
transformation behaviour under fusion and braiding is described by the matrices (2.81) is a
non-trivial task. In the sequel, however, we will mainly be interested in partition functions;
these are independent of the normalisation of the chiral vertex operators.
3. Frobenius objects and algebras of open string states
3.1. Algebra objects
In the construction of correlation functions a central role will be played by objects with
additional structure in the modular category of the chiral CFT, namely so-called algebra
objects. It is quite natural to study such extra structures for objects of a tensor category;
for some category theoretic background, see [25, Chapter VI] and [72, Chapters 2, 3]. In
short, an algebra in a tensor category C is an object A Obj (C) together with a product,
i.e., a morphism between A A and A, that is associative and has a unit. More concretely:
Definition 3.1. An algebra object, or simply an algebra, in a tensor category C is a triple
(A, m, ), where A is an object of C, m Hom(A A, A) and Hom(1, A), such that
the multiplication morphism m and the unit morphism fulfill
m (m idA ) = m (idA m) and m ( idA ) = idA = m (idA ).
(3.1)
Our pictorial notation for m and and the resulting pictures expressing formulas (3.1) are
as follows:
383
(3.2)
Later on, we will often suppress the label A on the algebra lines when this is
unambiguous, e.g., when the presence of morphisms depicted by small circlesproducts
or unitsindicates that we are dealing with morphisms involving only A (and possibly 1).
To call A an algebra is appropriate because in the particular case that C is the category
of vector spaces over C (or some other field), the prescription reduces to the conventional
notion of an algebra. In every tensor category the tensor unit 1 provides a trivial example
of an algebra; its product is m = id1 id11 id1 id1 , and its unit is = id1 . A class of
less trivial examples, present in any tensor category with duality, is given by objects of the
form A = U U ; in these cases one can take
m = idU dU idU
and = bU .
(3.3)
However, it turns out that many interesting algebras are not of this special form.
Consider now an algebra A in a semisimple tensor category C. The object A is a finite
A } in the embedding
direct sum of simple subobjects Ui . We fix once and for all bases {i
i
spaces Hom(Ui , A), as well as dual bases {A } in Hom(A, Ui ). We draw these morphisms
as
(3.4)
In accordance with the notation (2.3), the dimensions of these coupling spaces will be
abbreviated by
i, A := dim Hom(Ui , A) = dim Hom(A, Ui ).
(3.5)

k,A
. The unit morphism is non-zero, and

Thus as an object we have A =
kI Uk
hence
0, A = dim Hom(1, A) 1.
(3.6)
A } of Hom(U , A) and { i }
We may express the product m with respect to the bases {i
i
A
of Hom(A, Ui ), for i = a, b, c, as compared to the basis {(a,b)c } of Hom(Ua Ub , Uc )

that we introduced in (2.29). This way we characterise m by the collection of numbers
384
c ;
ma,b that are defined as
(3.7)
Upon use of formula (2.36) one finds that in terms of these numbers
associativity property of m is expressed as
f,A
f ;
ma,b mfd;
,c
c ;
ma,b ,
the
Nbc Nae
e,A

e; K
d; (a b c) d
mb,c ma,e
F eK,f .
(3.8)
eI =1 K=1 =1
=1
This shows that the concept of associativity we are dealing with is most natural indeed:
when expressing the associativity property in terms of bases, it involves precisely the
F-matrices, a behavior that is familiar from the representation theory of algebras over C,
like, e.g., universal enveloping algebras of simple Lie algebras.
3.2. From boundary conditions to algebras
We will now argue that each boundary condition of a rational CFT determines an algebra
in the modular category for that CFT. To explain this relationship, let us start on the CFT
side, i.e., with some boundary condition M.
For definiteness take the world sheet geometry to be the upper half plane. Then the
boundary is the real line R. Let M be a boundary condition that preserves the chiral algebra
V. We do not assume, however, that V is maximally extended, so M may break some of
the symmetries of the model; but we do assume that the theory is still rational with respect
to V.
The space H of states that corresponds to boundary fields living on a boundary segment
with boundary condition Mfirst introduced in [73]is often referred to as the state
space on the upper half plane with boundary condition M or, in string terminology, as
the space of open string states for open strings stretching from one D-brane to itself. H is
organised in modules of one copy of the chiral algebra V, rather than in modules of V V
as for the bulk:

H=
(3.9)
na Ua .
aI
The sum runs over (representatives for the isomorphism classes of) irreducible highest
weight modules Ua of V, and na Z0 specifies their multiplicities. In the CFTs we
consider, the representation theory of V gives rise to a modular category C whose simple
objects are the irreducible V-modules. The notation Ua for these modules is chosen so as
385
to conform with our convention for the representatives for isomorphism classes of simple
objects in C, see Section 2.1.
There is a one-to-one correspondence between states in H and fields that live on the
boundary M. Let us denote primary boundary fields (i.e., those that correspond to highest
weight states) on M as a (x). Here a labels the V-representation, while is a multiplicity
index that runs from 1 to na . Since is a boundary field, its argument x takes values in R
only.
In simple cases, for instance in minimal models, the operator product expansion of
boundary fields takes the form
c
Ca,b (x y)c a b
a (x)b (y) =
c,

c (y) + terms with descendants .
(3.10)
c
This defines the boundary operator product coefficients (or structure constants) Ca,b of
primary boundary fields on M.
In the general case two related additional features must be taken into account: first, there
may be more than one independent way for representations a and b to fuse to c. Second,
for a given coupling of a and b to c, the primary field of c does not necessarily appear7
on the right-hand side of (3.10). The dimension of the space of couplings is the fusion rule
Nabc .
To obtain a formulation of the OPE that covers the generic situation we regard a
coupling of a and b to c as a prescription on how to associate to every vector v in the
highest weight module Ua of the chiral algebra a linear map from Ub to Uc . A basis in the
space of couplings is then a collection of Nabc maps Vab c, : Ua Hom(Ub , Uc )[[z, z1 ]]
zc a b , = 1, 2, . . . , Nabc ; the basis elements are known as chiral vertex operators.8
Further, we denote by {vdD }D an orthonormal basis of L0 -eigenvectors in Ud , with vd0 vd
D the corresponding descendant field of the primary
a highest weight vector, and by d
0
d d . Then the OPE reads
c
a (x)b (y) =
Nab
n

c ;
Ca,b (x y)c a b
cI =1 =1

C
C
vcC Vab c, (va ; z = 1)|vb (x y)(vc )c c
(y).
(3.11)
Note that here we consider only special cases of boundary operator products, since we
are interested in a single boundary condition M. To be fully general, we must also deal
MM % that change the boundary condition, from M to M % , and
with boundary fields d
MM % (x) M % M %% (y) expanded in boundary fields MM %% (y).
hence operator products a
c
b
7 In fact, with a suitable basis choice the primary field appears for at most one of the couplings.
8 Hom(U , U )[[z, z1 ]] denotes the space of Laurent series with values in the space Hom(U , U ). To make
b c
b c
the concept of chiral vertex operators precise, one should work in the vertex operator algebra setting, where z is
a formal variable, so that one is dealing with formal Laurent series.
386
Correspondingly, the operator product coefficients acquire three more labels M, M % , M %% .

As we will discuss in Section 4.4, boundary changing fields play a natural role in the
categorical setup as well; they occur in the CFT interpretation of representations of algebra
objects.
The sewing constraint [74] that arises from the factorisation of the correlator of four
boundary fields on a disk with boundary condition M then looks as follows:
nf
f ;
Ca,b Cfd;
,c
Nbc Nae
ne

e; K
(a b c) d
d;
Cb,c Ca,e
F eK,f .
(3.12)
eI =1 K=1 =1
=1
It proves to be convenient to describe the boundary structure constants with the help of
the concepts of the category CA of A-modules and of a module category, which will be
described in Section 4.1. They correspond to the generalised 6j -symbols (1) F (defined via
formula (4.9)) of CA regarded as a module category, and the relation (3.12) is nothing but
the corresponding generalisation of the pentagon identity. The identification between the
boundary structure constants C and the quantities (1)F also appears in the formalism used
in [14], where it is obtained via the relation with weak Hopf algebras; our formulas (3.11)
and (3.12) correspond to (4.11) and (4.29) of [14] for a single boundary condition. In the
Cardy case the (1)F specialise to the ordinary F-matrices (this was first observed for the
A-series of Virasoro minimal models in [75] and established in general in [14,76]); the
sewing constraint (3.12) is then just the pentagon identity for F.
From the data provided by the decomposition (3.9) of the state space and by formula
(3.11) for the structure constants we can obtain an algebra A in C, which we call the algebra
of open string states. This is achieved as follows. As an object in C, the algebra of open
string states coincides with H as given in formula (3.9), i.e.,

na Ua .
A
(3.13)
=
aI
Now let us choose a basis in Hom(Ua , A) as described in Section 3.1, and define a
c ;
multiplication m on A in this basis via (3.7), by demanding that the coefficients ma,b
are just the structure constants:
c ;
c ;
ma,b := Ca,b .
(3.14)
With this assignment the consistency condition (3.12) fulfilled by the structure constants of
the boundary fields on M coincides with the associativity condition (3.8) for the algebra A.
The unit of A is the identity field on M. We conclude that indeed the fields living on a
boundary with boundary condition M that preserves V provide us with an algebra object
A in C.
The algebra objects arising from boundary conditions enjoy an additional important
property which derives from the non-degeneracy of the two-point correlation functions.
Non-degeneracy means that for any field on the M-boundary there exists at least one9
9 We cannot formulate this condition in terms of primary fields alone, for the same reason that we had to adopt
the more complicated form (3.11) of the OPE. E.g., in WZW models generically it takes a (horizontal) descendant
to get a non-zero two-point function with a primary field.
% such that

(x) % (y) = 0.
387
(3.15)
If there were a field in the theory that had zero two-point function with all other fields, it
would decouple from the theory, i.e., every correlator involving that field would vanish.
To evaluate the two-point functions one can use the OPE (3.11) together with the onepoint functions (x) of boundary fields. The latter can be non-zero only for boundary
fields of weight zero, but (as is natural in view of the connection of our results with those in
two-dimensional lattice TFT) we allow for the possibility that there can be more of those
than just the identity field. A superposition of two elementary boundaries, for example,
has at least two fields of weight zero. In the case at hand, the one-point functions are
determined once we know them for the primary boundary fields a on M.
In the category theoretic setting, the collection of one-point functions of boundary fields
on M will give rise to a morphism Hom(A, 1). The non-degeneracy (3.15) translates
into the non-degeneracy of the composition m. That is, the matrix G(a) defined by
(3.16)
is invertible as a matrix in the multiplicity labels , .
In fact, the one-point functions a (x) of primary boundary fields on the upper half
plane (or equivalently, on the unit disk) are themselves of a special form. First of all, by
conformal invariance they can be non-zero only for a = 0. A more interesting result follows
from the sewing constraint for the boundary one point functions on an annulus.
Consider a cylinder of length L and circumference T with boundary condition M at both
ends. On one of the boundaries insert a boundary field 0 of weight zero. The cylinder can
be conformally mapped to an annulus of inner radius q and outer radius 1, and alternatively
to a half-annulus (with the two half-circular boundaries identified) in the upper half plane,
with the two ends stretching from 1 to q and from q to 1, where
q := exp(2L/T )
and q := exp(T /L).
(3.17)
This results in the equation

M(0 )q L0 +L0 c/12 |M = trH q L0 c/24 0 ,
(3.18)
where |M(0 ) denotes the boundary state for the M-boundary with an insertion of the
boundary field 0 and |M the boundary state without field insertion. On the right-hand
side of (3.18), the boundary field 0 is interpreted as an operator on the boundary state
space H.
Let us now assume that the CFT under consideration does not possess any state of
negative conformal weight (as is in particular the case for all unitary theories) and that
388
it has a unique vacuum (i.e., state |0 of weight zero) in the bulk, which we take to be
normalised as 0|0 = 1. Then in the limit of infinite length L only the vacuum propagates
on the left-hand side of (3.18), so that

M(0 )q L0 +L0 c/12 |M

= q c/12 1 + O(q ) M(0 )|0 0|M for L
(3.19)
with > 0. Finding the L limit of the right-hand side of (3.18) requires slightly
D denote a basis
more work. In accordance with the notation introduced in (3.11), let |d
D
C
D
of H, and d | a dual basis, i.e., c |d = c,d , C,D . Then
trH q L0 c/24 0
d
n

D
D L0 c/24
q
d
=
0 d
dI =1 D desc.

d
d |0 |d d (q)
=
C0,d
Sd,c c q 2
d,
=q
c/12
1 + O(q )
d, cI
d
C0,d
Sd,0
for L .
(3.20)
d,
In the first step, the trace is written out in terms of the bases. The crucial step is the second
one. Since 0 is a field of weight zero, it behaves just like the identity field as far as
commutation with elements of the chiral algebra is concerned, and this in turn implies that
D
D L c/24
D
= q (d )c/24 d |0 |d ,
0 d
d q 0
(3.21)
so that the sum over descendants D gives rise to the character d . The third step involves
a modular transformation of this character, upon which finally the limit as L can be
taken, leaving only the contribution of the vacuum character.10
Comparing the results (3.19) and (3.20) we conclude that the one-point functions of
weight zero boundary fields on the unit disk are given by
d
0 = const
n

d
Sd,0 C0,d
.
(3.22)
dI =1
Note that here the sum runs over all primary boundary fields on the M-boundary, not only
over those of zero weight. When translated back into the language of algebra objects, this
relation reads (using the basis (3.4))
A
0
= const%
d,A
d
m0,d
dim(Ud ).
(3.23)
dI =1
10 Here we assume that of the representations U of the chiral algebra V, the vacuum representation U is the
i
0
only one which contains a state of weight zero.
389
The additional properties (3.16) and (3.23) of the multiplication m mean thatprovided
that the CFT under consideration has a unique bulk vacuumthe algebra object A can
be turned into what will be called a symmetric special Frobenius algebra. It is the aim
of the next section to explain these notions. Before entering that discussion, let us make
some short remarks on the situation that Z00 > 1 for the torus partition function, taking
Z00 = 2 as an example. In this case there exist two distinct bulk fields of weight zero. All
correlators are independent of the insertion points of these weight zero fields, thus they
form a topological subsector of the theory. In particular, their operator product algebra
is commutative. As a consequence there is a projector basis {P1 , P2 } for the weight zero
fields, in which their operator products take the form P1 P1 = P1 , P2 P2 = P2 and P1 P2 = 0.
The identity field is then given by 1 = P1 + P2 . The projector fields P1 , P2 act on the space
of fields via the OPE, and thereby decompose the space of fields into eigenspaces. Let us
denote by 1 (z, z ) bulk fields for which P1 1 (z, z ) = 1 (z, z ), and analogously for 2 .
Mixed correlators then vanish:

1 (z, z )2 (w, w)
= P1 1 (z, z ) 2 (w, w)

= 0.
= 1 (z, z ) P1 2 (w, w)
(3.24)
We conclude that a CFT with Z00 = 2 should be interpreted as a superposition of two
CFTs each of which has Z00 = 1. By a superposition of two conformal field theories,
CFT1 and CFT2 , with the same central charge we mean the theory in which fields are pairs
= (1 , 2 ) of fields in the two constituent CFTs and correlation functions are sums

(z, z )(w, w)
= 1 (z, z )1 (w, w)
(3.25)
1 + 2 (z, z )2 (w, w)
2.
Requiring that a CFT cannot be written as a superposition of this type has a counterpart in
the underlying algebra object, which in this case must be indecomposable in the sense of
Definition 9(ii) of [13].
3.3. Frobenius algebras
The construction of correlation functions that will be discussed in Sections 5.3 and 5.8
(and further in forthcoming papers) uses a symmetric special Frobenius algebra object as
an input. In the present section we explain the qualifiers special, symmetric, and Frobenius,
and we show that algebras with these properties are natural from the CFT point of view.
This is the content of Theorem 3.6.
From a computational point of view, the construction of such an algebra object requires
the solution of the boundary factorisation constraint (3.12) for boundary fields living on
one single boundaryrather than a simultaneous solution of all constraints involving
all boundary conditions as well as boundary changing fields, which form a much larger
system. It turns out that the construction of the algebra is the only place where a nonlinear constraint ever must be solved. Finding the other boundary conditions and structure
constants then amounts to solving systems of linear equations only.
The notions of co-algebra (which is the notion dual to that of an algebra), Frobenius
algebra, special and symmetric given below are straightforward extensions to the category
setting of the corresponding concepts in algebras over C, see, e.g., [23].
390
Definition 3.2. A co-algebra A in a tensor category C is an object with a coassociative

coproduct and a counit , i.e., morphisms Hom(A, A A) and Hom(A, 1) that
satisfy
( idA ) = (idA )
and ( idA ) = idA = (idA ) .
(3.26)
The pictures for these morphisms and their properties are obtained by turning upside-down
those for an algebra:
(3.27)
We will be interested in objects that possess both an algebra and a co-algebra structure
and for which these structures are interrelated in a special way. This is encoded in the
definition of a Frobenius algebra (see [77,78] and references therein). Note that provided
the tensor category has a braiding (a condition that is not needed in the Frobenius case)
one may combine the algebra and co-algebra structure also into the one of a bialgebra;
this amounts to a different compatibility condition between product and coproduct. Bi- or
Hopf algebras in tensor categories do not occur in our discussion of boundary conditions,
though they play an important role in other contexts (see, e.g., [7983]), and we refrain
from describing those structures here.
Definition 3.3. A Frobenius algebra in a tensor category C is an object that is both an
algebra and a co-algebra and for which the product and coproduct are related by
(idA m) ( idA ) = m = (m idA ) (idA ).
(3.28)
In pictures,
(3.29)
391
To be able to discuss additional properties of the algebras that come from boundary
conditions, let us further introduce the following notions.
Definition 3.4.
(i) A special algebra in a tensor category C is an object that is both an algebra and a
co-algebra such that
= 1 id1
and m = A idA
(3.30)
for non-zero numbers 1 and A . In pictures,
(3.31)
(ii) A symmetric algebra in a sovereign tensor category C is an algebra object (A, m, )
together with a morphism Hom(A, 1) such that the two morphisms 1 , 2
Hom(A, A ) defined as

1 := ( m) idA (idA bA ) and

2 := idA ( m) (bA idA )
(3.32)
are equal. In pictorial notation the morphisms 1 and 2 are given by
(3.33)
(iii) A commutative algebra in a braided tensor category C is an algebra object (A, m, )
for which m cA,A = m.
(iv) A haploid algebra in a tensor category is an algebra object A for which dim Hom(1, A)
= 1.
Remark 3.5.
(i) In the study of algebras over C [23], symmetric algebras constitute an interesting
subclass of Frobenius algebras. They contain for instance all group algebras. Also,
392
the property of being symmetric is considerably weaker than being commutative

(Abelian). For algebras in modular tensor categories, commutativity (with respect
to the braiding c) and triviality of the twist (i.e., A = idA ) together imply that A is
symmetric.
(ii) Algebras possessing the same representation theory are called Morita equivalent (see,
e.g., [13,22,78,84,85]). It turns out that neither haploidity nor commutativity are
invariant under Morita equivalence. As will be discussed in more detail elsewhere, the
algebra objects U U described below Definition 3.1 turn out to be Morita equivalent
for different (not necessarily simple) U . But while for U = 1 the resulting algebra is
certainly haploid and commutative, for a general object U this need not be the case.
(iii) In practice, haploidity is a quite useful property, because a haploid algebra is
automatically symmetric, as follows from Corollary 3.10. Furthermore it can be shown
[13, Section 3.3] that every Morita class of algebras that cannot be decomposed nontrivially in a direct sum has at least one haploid representative. (For the notion of direct
sum of algebras see Proposition 3.21.)
(iv) The algebras A of our interest are not necessarily (braided-)commutative. However,
commutativity is still an interesting property, because in CFT terms it immediately
allows for an interpretation of A as an extension of the chiral algebra (see Section 5.5).
Note, however, that according to point (ii) above, a non-commutative algebra can still
correspond to an extension. Commutative algebras A in a braided tensor category have
also been studied under the name quantum subgroups [11,86]. This terminology has
its origin in the fact [87] that commutative algebras in the representation category
of a group G are in one-to-one correspondence with the algebras of functions on
homogeneous spaces G/H , and thereby with subgroups H of G.
In Section 3.2 we have seen that algebra objects are a natural structure to consider from
the CFT point of view, as they can be obtained from boundary conditions. In the rest of
this section we show that:
Theorem 3.6.
(i) If an algebra can be endowed with the structure of a symmetric special Frobenius
algebra, then this structure is unique up to a normalisation constant C .
(ii) Every algebra coming from a boundary condition that preserves the chiral algebra
V of a (rational, unitary) CFT with unique bulk vacuum can be endowed with the
structure of a symmetric special Frobenius algebra.
Note that the results of this paper rely only on the existence of a symmetric special
Frobenius algebra. Part (ii) of Theorem 3.6 guarantees the existence of such an algebra in
a large class of theories. It does not exclude, however, the existence of such an algebra
in theories not belonging to this class, in particular in certain non-unitary theories (e.g.,
non-unitary minimal models).
The proof of the theorem will require several lemmata.
Lemma 3.7. Let (A, m, ) be an algebra and let Hom(A, 1). Then the following holds.
393
(i) If there exists a Frobenius structure on A with counit , then it is unique.

(ii) There exists a Frobenius structure on A if and only if 1 as defined in (3.32) is
invertible.
Proof.
(i) We must show that the coproduct is unique. If there exists any coproduct
satisfying the Frobenius property, then 1 as defined in (3.32) is invertible, with (leftand right-)inverse 11 given by

11 = [dA idA ] idA ( ) .
(3.34)
Analogously, 2 is invertible as well. In pictures, the inverses are
(3.35)
Moreover, application of the Frobenius property also shows that we can express,
conversely, the coproduct through the product and the morphism (3.34):

= (idA m) idA 11 idA (bA idA )
(3.36)
(a picture can be found below in (3.42)). Now for given m and , 1 is uniquely determined
by (3.32). Since in addition the morphism 11 is uniquely determined by its property of
being inverse to 1 , it follows that is unique as well.
(ii) We have already seen in the proof of part (i) that A being Frobenius implies that the
morphism 1 is invertible, with inverse given by (3.34). Conversely, let us now assume
invertibility of 1 . One quickly checks that 2 as given in (3.32) can be expressed through
1 as11
2 = [idA dA ] [idA 1 idA ] [bA idA ].
Consequently it is invertible as well, with inverse given by

21 = [idA dA ] idA 11 idA [bA idA ].
(3.37)
(3.38)
It follows that we can define a candidate coproduct by formula (3.36), and yet another
candidate % by the corresponding formula with 21 , i.e.,

% := (m idA ) idA 21 idA (idA bA ).
(3.39)
11 Note that does not, in general, coincide with the morphism that is dual to . Indeed, is an
2
1
1
1
element of Hom(A , A ), and while A is certainly isomorphic to A there is no reason why the two objects
should be equal.
394
Note that because of the relation (3.38), we can also write

= (idA m) 21 idA idA (bA idA )
(3.40)
and
% = (m idA ) (idA idA 11 ) (idA bA ),
(3.41)
respectively. Pictorially,
and
(3.42)
However, the two morphisms and % defined this way actually coincide. To see this, we
compose the morphisms as given by (3.40) and (3.41), respectively, with 2 idA . In the
case of this yields immediately the morphism (idA m) (bA idA ), while in the case
of % the same result follows with the help of associativity:
395
(3.43)
The coassociativity property then follows from the fact that associativity of m implies
equality of ( idA ) % and (idA % ) :
(3.44)
That the counit properties (idA ) = idA and ( idA ) % = idA are satisfied follows
directly from the definition of 1 and 2 . Finally, the Frobenius property (3.29) follows
again easily from associativity of m, using for showing the first equality, and % for the
396
second:
and
(3.45)
Definition 3.8. For A an algebra in a sovereign tensor category, the morphisms < and
in Hom(A, 1) are defined as
(3.46)
Lemma 3.9. Let A be an algebra in a sovereign tensor category. Then we have:
(i) A together with := < is a symmetric algebra, for any C. The same holds with
= .
(ii) If A is a symmetric Frobenius algebra, then < = .
397
Proof.
(i) According to Definition 3.4(ii) we must show that 2 = 1 . This is verified as
follows. We have
(3.47)
Here the first step is the definition, the second uses associativity, and the third identity
follows by replacing the left-dual of the morphism that is enclosed in a dashed box by its
right-dual, which is allowed because we are working in a sovereign category. That the righthand side of (3.47) equals 1 follows by using once more associativity and the definition
of = < .
(ii) Since A is a Frobenius algebra, the morphisms 1,2 are invertible. And since A is
symmetric, we have 1 = 2 . Thus in particular the equalities 1 21 = 2 11 =
idA and 11 2 = 21 1 = idA hold. Pictorially, we have
(3.48)
398
as well as the mirror images of these identities. Consider now the transformations
(3.49)
where in the first step we used (3.48) to insert unit and counit, while the second step is an
application of the Frobenius and unit properties. Analogously one shows a mirror version
of (3.49); combining the two identities then establishes the lemma.
Corollary 3.10. If A is a haploid algebra, then it is symmetric for any choice of
Hom(A, 1).
Proof. By definition, for haploid algebras the morphism space Hom(A, 1) has dimension
one. Furthermore, < = 0 since < = dim(A). Thus < forms a basis of Hom(A, 1) and
any choice of is proportional to < .
As an immediate consequence of the relation (3.48), for every symmetric special
Frobenius algebra the normalisations 1 and A in (3.31) obey
1 A = dim(A).
(3.50)
In the present section we keep these normalisations explicitly. But from Section 4 onwards
we will simplify the presentation by assuming (without loss of generality) that the
coproduct is normalised such that A = 1, and hence 1 = dim(A).
Lemma 3.11. For a symmetric Frobenius algebra A the following statements are
equivalent:
(i) A is special.
(ii) The counit obeys12 = < for some non-zero number .
Proof.
(i) (ii): Composing the specialness property m = A idA (where A is non-zero)
with yields = A1 m , which is the last expression in (3.49). Backtracking the
steps in (3.49) thus gives = A1 < .
(ii) (i): By composition with the unit we get
= < = dim(A) = 0,
12 Recall from Lemma 3.9 that = for symmetric Frobenius algebras.
<
(3.51)
399
where in the last step the unit property is used to obtain the trace of idA . Second, consider
the moves
(3.52)
In the first step a counit is inserted, the second step combines the Frobenius property
and coassociativity, the third step is again backtracking the steps (3.49), and finally the
assumption = < is inserted and the counit property is used.
Lemma 3.12. For any algebra (A, m, ) the following two statements are equivalent:
(i) There exist Hom(A, 1) and Hom(A, A A) such that (A, m, , , ) is a
symmetric special Frobenius algebra.
(ii) The morphism 1% Hom(A, A ), defined as in (3.32), but with < (as defined in
(3.46)) in place of , is invertible.
(The same holds when taking instead of < .)
Proof.
(i) (ii): When A is symmetric special Frobenius, we have = < (by Lemma 3.11),
and hence 1 = 1% , with a non-zero number . Further, since A is Frobenius, the
morphism 1 is invertible, and hence so is 1% .
(ii) (i): We must find morphisms and such that (A, m, , , ) is a symmetric
special Frobenius algebra. For we choose := < . From Lemma 3.9(i) it follows that A
is symmetric. Since by assumption 1 is invertible, there exists a that turns A into a
Frobenius algebra; this morphism is the one given in (3.42). Thus (A, m, , , ) is a
symmetric Frobenius algebra. But because of = < , Lemma 3.11 implies that A is special,
too.
Remark 3.13. Let us add two comments on the case where C is the category of complex
vector spaces.
(i) Suppose A is an algebra over C with basis {bi }. Then the morphism < is the
trace of the
the sense that, with the product given by
multiplication operator, in
bi bj = k mijk bk , it satisfies < (bi ) = k mkik . For example, when A is the algebra
of functions over some finite set, then a basis is given by the delta functions, for which
400
the product reads bi bj = i,j bj , and we obtain < (bi ) = 1. Thus in this case < is an
integral with a measure that weighs every point evenly.
(ii) It is easy to construct algebras (A, m, ) such that 1 is not invertible when setting
= < . Take, e.g., A = C2 with a basis {e, n} such that e acts as the unit element while
n n = 0. Then < (e) = 2 and < (n) = 0, and as a consequence 1 (n) = 0.
Proof of Theorem 3.6.
(i) Given an algebra object (A, m, ), suppose there exist and such that
(A, m, , , ) is a symmetric special Frobenius algebra. By Lemma 3.11 we have = <
for some C . But < is already fixed in terms of the multiplication m and thus is fixed
up to a normalisation constant in terms of (A, m, ). By Lemma 3.7 the coproduct is
uniquely fixed in terms of (A, m, , ) and by Lemma 3.9, (A, m, , ) is symmetric.
(ii) In Section 3.2 it was shown that (A, m, ) with A given in terms of the CFT
data by (3.13) and multiplication m defined in terms of boundary structure constants by
(3.14) is an algebra object. The counit was defined in terms of boundary one-point
functions on the upper half plane and has the property (3.23). Note that the calculation
leading to (3.23) is based on the assumption that there are no states of negative conformal
weight; this assumption is fulfilled in every unitary CFT. The invertibility of the matrix
G(a) in (3.16) is equivalent to the statement that the morphism 1 in (3.33) is invertible.
Furthermore, comparing (3.23) to , as defined in (3.46) and expressed in a basis, shows
that = const% . The constant is non-zero since 1 is invertible. Thus Lemma 3.12 is
applicable, since 1% = (const% )1 1 is invertible as well.
3.4. The associated topological algebra
An algebra object A in a tensor category C defines automatically also an algebra over C,
to be denoted as Atop and called the topological algebra associated to A. The complex
algebra Atop is defined as follows. As a vector space, Atop is the morphism space
Atop := Hom(1, A).
(3.53)
The multiplication mtop and unit top (regarded as a map C Atop ) are given by
mtop ( ) := m ( ),
top (1) := ,
(3.54)
where , Atop .
In the CFT interpretation introduced in Section 3.2, Atop is the algebra of boundary
fields (on a fixed boundary condition) that transform in the vacuum representation of the
chiral algebra. The fields of weight zero in this set form a topological subsector of the
boundary theory, hence the name Atop . The presence of such a subsector fits well with the
description [77,8892] of two-dimensional lattice TFTs via symmetric special Frobenius
algebras in the category of complex vector spaces.
In the sequel we will deduce various somewhat technical results, which will be
instrumental later on. The reader not interested in the proofs of these results may proceed
directly to Section 3.5.
Let us choose bases for Atop and its dual in such a way that
bi Hom(1, A),
bj Hom(A, 1),
b j b i = i .
(3.55)
401
Lemma 3.14. Let A be a symmetric special Frobenius algebra in C and Atop its associated
topological algebra. Setting
top () :=

b i b j bi bj ,
and top() :=
(3.56)
i,j
for all Atop turns Atop into a symmetric Frobenius algebra over C.
Proof. The statement follows by direct computation. The calculations all boil down to the
observation that a graph with Atop -lines, products and coproducts without loops and with
arbitrary basis elements bi and bj inserted at the external lines is equal to the same graph
with A-lines, products and coproducts.
Note, however, that with top defined as above, Atop is not necessarily special. Indeed,
the definition of specialness contains graphs with loops and the above argument no
longer applies. For example, let u and v be two non-isomorphic simple objects in C
and let A = (u v) (u v) , with product as defined in formula (3.3), and = <
as counit. A basis of Atop is given by the duality morphisms {bu , bv } of simple objects
u, v. One verifies that (with x and y standing for any of u, v) mtop(bx , by ) = x,y bx and
top(bx ) = dim(u v) dim(x). However, by the same reasoning as in Remark 3.13 one
shows that <,top (bx ) 1. It follows that for dim(u) = dim(v), top as defined by (3.56) is
not proportional to <,top , and hence by Lemma 3.11 Atop cannot be special.
Definition 3.15. Let A be an algebra in C and Atop the associated topological algebra. The
relative center of Atop with respect to A is the subspace

centA (Atop) := Atop | m ( idA ) = m (idA ) .
(3.57)
Elements in centA (Atop) in particular commute with all elements of Atop . Thus the
relative center is a subalgebra of the ordinary center cent(Atop) of Atop (defined in the
usual way for algebras over C). Also note that the unit always lies in centA (Atop). Thus
we have the inclusions
{ | C} centA (Atop ) cent(Atop ) Atop.
(3.58)
The subset centA turns out to be useful in the description of the torus partition function.
Indeed, we will see that Z00 = dim(centA (Atop)). The proof will be given in Section 5.3;
it relies on the following lemma.
402
Lemma 3.16. For any symmetric special Frobenius algebra A with A = 1 one has
(3.59)
Proof. We have the following equivalences:
(3.60)
The first of these equivalences is obtained by composing the first equality on the left with
21 on the left and with 11 on the right-hand side (these inverses exist by the Frobenius
property, and they are equal by the symmetry property) and then using the Frobenius
property, while the second equivalence follows upon composition of the middle identity
with a duality morphism. The lemma is now established if the equality on the right-hand
side of (3.59) can be shown to be equivalent to the one on the right-hand side of (3.60). But
this is just a special case, obtained by setting X = 1, of Lemma 3.17.
In the following considerations the braiding between the algebra A and arbitrary objects
X plays a role.
403
Lemma 3.17. Let A be a special Frobenius algebra with A = 1. For any object X and any
morphism Hom(X, A) we have
(3.61)
as well as the analogous equivalence in which all braidings are replaced by inverse
braidings.
Proof. With the braidings chosen as in picture (3.61), the two directions are shown as
follows:
: This is seen immediately by composing both sides of the equality on the left-hand
side of (3.61) with m and then using specialness of A.
: Starting from the right-hand side of (3.61), and using the Frobenius and (co-)
associativity properties, we have
(3.62)
404
If we now substitute again the equality on the right-hand side of (3.61) in the last
expression, we arrive at the equality on the left-hand side of (3.61). For the inverse
braidings the equivalence is shown in the same way.
3.5. Sums, products, and the opposite algebra
Given two (symmetric special Frobenius) algebras A and B in a ribbon category, we

can define algebra structures on A B and A B. It is also possible to twist the product
of A by the braiding, giving rise to an algebra Aop . In Section 5.3 we will encounter a
physical interpretation of these operations: denoting the torus partition function obtained
kl := Z(A) , we have the matrix equations
from an algebra A as Z(A), and defining Z(A)
kl

Z(B)

Z(A B) = Z(A) + Z(B), Z(A B) = Z(A)
and Z(Aop ) = Z(A)t .
These expressions suggest the interpretation that A B describes a superposition of two
CFTsin the sense introduced before formula (3.25)while A B defines a (in general
non-commutative) product of two CFTs that are associated to the same chiral data. Note
that this product is different from the usual product CFT1 CFT2 , where the new stress
tensor is given by T1 + T2 and thus the central charges add. The CFT resulting from A B
has the same stress tensor and central charge as the ones resulting from A and B.
In the rest of this section we will make precise the notions of A B, A B and Aop
and prove some properties of these algebras.
Proposition 3.18 (Opposite algebra).

(i) A = (A, m, ) is an algebra if and only if Aop := (A, m (cA,A )1 , ) is an algebra.13
(ii) A = (A, m, , , ) is a symmetric special Frobenius algebra if and only if

Aop := A, m (cA,A )1 , , cA,A ,
(3.63)
is a symmetric special Frobenius algebra.

Proof. The statement (i) follows by a straightforward application of Definition 3.1.
Similarly, for obtaining (ii) one checks easily from the Definitions in 3.2, 3.3 and 3.4
that the respective properties of A and Aop follow from each other. For example, that A
13 Note that we take the inverse braiding in this definition. It is this choice that is needed to prove
Propositions 4.6 and 5.3. The first of these links the definition of Aop to that of the tensor product, while the
second links that of the tensor product to that of the graph for the torus partition function. In this sense the
conventions implicit in the graph (5.30) fix the convention for Aop .
405
symmetric implies Aop symmetric is seen as follows:
(3.64)
The first morphism is 1 for Aop , expressed via the multiplication m of A. Symmetry of
A enters in the third equality.
Note that even though Aop is equal to A as an object in C, it has a different
multiplication and is therefore, in general, not isomorphic to A. In particular, this remark
still applies when A and Aop are symmetricrecall that the symmetry property, introduced
in Definition 3.4(ii), does not refer in any way to the braiding.
Corollary 3.19.
(i) For any algebra (A, m, ) and any n Z also A(n) := (A, m (cA,A )n , ) is an
algebra.
(ii) For any symmetric special Frobenius algebra (A, m, , , ) and any n Z also

A(n) := A, m (cA,A )n , , (cA,A )n ,
(3.65)
is a symmetric special Frobenius algebra.
In particular we have A = A(0) and Aop = A(1) .
Proposition 3.20.
(i) The twist A is an intertwiner between the algebras A(n) and A(n+2) , in the sense that
A (n) = (n+2)
with
m(n)
= m (cA,A
)n
and A m(n) = m(n+2) (A A ),

and
(n)
= the product and unit of
A(n) .
(3.66)
406
(ii) If A(n) is symmetric special Frobenius then in addition

(n) = (n+2) A
and (A A ) (n) = (n+2) A .
(3.67)
Proof. By direct computation. One only has to use some of the defining properties of
a modular tensor category, namely (see (2.10), (2.11)) that V f = f U , for any
f Hom(U, V ) as well as U V = cV U cU V (U V ). For instance,
A m(n) = m(n) AA = m(n) (cA,A )2 (A A ) = m(n+2) (A A ). (3.68)
By definition of the direct sum A B of two objects A and B, there exist morphisms
XC Hom(C, A B)
and YC Hom(A B, C)
(3.69)
and XA YA + XB YB = idAB
(3.70)
for C {A, B} such that

YC XD = C,D idC
for C, D {A, B}. When A and B are algebras, then these morphisms can be used to
endow A B with the structure of an algebra, too:
Proposition 3.21 (Direct sum of algebras).
(i) The triple (A B, mAB , AB ) with

XC mC (YC YC )
mAB :=
and AB :=
C=A,B
XC C (3.71)
C=A,B
furnishes an algebra structure on the object A B.

(ii) When A, B are symmetric special Frobenius and normalised such that A A =
dim(A) and B B = dim(B), then A B is symmetric special Frobenius, with counit
and coproduct given by

AB =
C YC and AB =
(XC XC ) C YC . (3.72)
C=A,B
C=A,B
Proof. The proof proceeds again by direct computation. To check, for example, that A B
is special, one computes

mAB AB =
XC mC (YC YC )
C=A,B

C

(XD XD ) D YD
D=A,B
XC mC C YC =

C
XC idC YC = idAB .
(3.73)
407
In the second equality we insert the orthogonality relation in (3.70), while the third step
uses that A and B are special, with constants A and B both equal to 1. The last step is
just the completeness in (3.70).
Proposition 3.22 (Tensor product of algebras).
(i) For any two algebras A and B the triple (A B, mAB , AB ) with

mAB := (mA mB ) idA (cA,B )1 idB
and AB := A B
(3.74)
furnishes an algebra structure on the object A B.
(ii) If A, B are symmetric special Frobenius, then also A B is symmetric special
Frobenius, with counit and coproduct given by
AB = A B
and AB = (idA cA,B idB ) (A B ).
(3.75)
Proof. In this case, too, one checks all the properties by direct computation. To give an
example, the associativity of A B follows from
(3.76)
Remark 3.23.
(i) In the definition (3.74) one could have used the braiding cB,A instead of (cA,B )1 . It
is easy to check that this alternative multiplication on the object A B coincides with
the multiplication on (Aop Bop )(1) given above. We will not introduce a separate
symbol for this way of defining a multiplication on A B.
(ii) For any three algebras A, B, C there are two a priori distinct ways to construct an
algebra structure on the object A B C by the tensor product of algebras, namely
408
(A B) C and A (B C). As C is a strict tensor category, the two combinations

are identical as objects in C. One can quickly convince oneself from the definitions
in Proposition 3.22 that the unit and multiplication morphism (as well as the counit
and comultiplication, if they exist) are identical in both cases as well. Thus the tensor
product of algebras defined in Proposition 3.22 is associative.
(iii) As will be discussed in a forthcoming paper, the operations of taking sums, products
and opposites of algebras are compatible with Morita equivalence. Denote by [A] the
Morita class of A, and let A% be another representative of [A], and similarly for [B]
and B, B % . Then
[A B] = [A% B % ],
[A B] = [A% B % ] and [Aop ] = [A%op]. (3.77)
As a consequence, the operations on algebras can be lifted to the level of CFT. We will
return to this issue in Section 5.3.
It is also worth mentioning that the resulting ring structure on Morita classes of symmetric
special Frobenius algebras is reminiscent of Brauer groups for (finite-dimensional, central,
semisimple) algebras over a field, compare, e.g., Chapter 4 of [24]. Indeed, consider the
tensor category Vect(k) of vector spaces over some field k, which we do not assume to
be algebraically closed. Then the module categories over Vect(k)or, equivalently, the
Morita classes of associative algebras in Vect(k)are classified [13] by division algebras
over k, which amounts to compute the Brauer group of all finite extensions of k. Quite
generally, the problem of classifying full CFTs based on a modular tensor category C
can therefore be expected to amount to the following two tasks: find all extensions of C,
and compute the (suitably defined) Brauer group of each extended theory. Here we call
a modular tensor category D an extension of C iff there exists a haploid commutative
symmetric special Frobenius algebra A in C such that D is equivalent to the modular tensor
category of local A-modules (see Section 4.1).
3.6. The case Nij k {0, 1} and dim Hom(Uk , A) {0, 1}
To continue our meta example, we will further limit ourselves to the case where each
simple subobject a in the algebra A occurs with multiplicity one. In this case we can omit
the index labelling the basis of possible embeddings of a in (3.4), rendering the notation
less heavy.
An algebra object is now described by a collection of pairwise non-isomorphic simple
objects {1, Ua1 , . . . , Uan }. The multiplication on A can be expressed by constants mab c ,
as in formula (3.7), and a comultiplication similarly by constants cab . The associativity
condition (3.8) takes the form

(a b c) d
mab f mf c d =
(3.78)
mbc e mae d Fe f
.
eA
Here we use the symbol to indicate the relation is a simple subobject of. Further,
we choose the basis vector in Hom(1, A) to be given by the unit of A; then the dual basis
vector in Hom(A, 1) is fixed to be the multiple /1 of the counit . Thus in particular
409
composing with the counit and unit, respectively, gives
(3.79)
Now the multiplication already determines the comultiplication. Let us make this
relation explicit. The normalisation (3.79) of the unit implies that ma0 a = 1 = m0a a .
Since A is haploid, the two morphisms 1 and 2 in (3.33) are automatically equal. For
them to be isomorphisms we need ma a 0 = 0 for all a A. To determine the inverse of
1 = 2 , we express it in a basis as
0
with a = maa
pa a dim(A),
(3.80)
where pa and a are the numbers introduced in (2.24) and (2.34), respectively. It follows
that the map 1 is then just given by
(3.81)
410
Using the expression (3.40) for the coproduct and the identities (2.60), one finds
(3.82)
With the help of the relations (2.34), (2.35), (2.46) and (2.60), it then follows that
(a a c) c
ab
c =
b F
mac
1
b0
.
0 (a a a) a
dim(A) maa
(3.83)
00

Different algebra objects in the Z2N free boson modular category correspond to
different compactification radii. (This will be made more explicit in Section 5.6.1, when
we come to computing the torus partition function.) There is one algebra object associated
to every subgroup of Z2N that contains only even elements, in the sense that there is one
algebra object for every divisor r of N . For such a subgroup, 2r with 0 r < N is its
minimal element. We denote as A2r the corresponding object
A2r :=
N/r1

[2nr].
(3.84)
n=0
The multiplication on A2r is given by

m[a][b] [a+b] = 1
for all a, b, c A.
(3.85)
411
One checks that all non-zero F-matrix elements (2.77) with only even labels are equal to 1.
As a consequence, with (3.85) the associativity condition (3.78) is satisfied trivially, as
1 = 1.
The algebras presented above do not exhaust all algebra objects of the free boson theory,
but they do give all haploid ones. This follows from a general treatment of algebra objects
containing only simple currents as simple subobjects, which will be presented in a separate
paper. For now we only remark that simple current theory [27,9395] tells us that to obtain
a modular invariant, a simple current J must be [96] in the so-called effective center, i.e.,
the product of its order (the smallest natural number ; such that J ;
= 1) and its conformal
weight must be an integer. In the case under study the simple current orbits are generated by
objects r with r a divisor of 2N , i.e., by simple currents of order 2N/r. Modulo integers,
the conformal weight of r is r 2 /4N and thus it is in the effective center iff r 2Z.
3.6.2. Example: E7 modular invariant
According to [13,32], for the E7 type modular invariant of the su(2) WZW model at
level 16, the associated algebra object A should be expected to have the form
A = (0) (8) (16).
(3.86)
This object can indeed be turned into an algebra [13,32], and we will show in the sequel
how this can be established with our methods. But of course, at this point we do not yet
know whether the modular invariant of the CFT associated to the algebra (3.86) is of E7
typethis will be verified in Section 5.6.2.
Note that J := (16) is a simple current for su(2)16 and that f := (8) is its fixed point.
Let us therefore investigate, more generally, algebra objects of the form
A = 1 f J,
(3.87)
for which the fusion rules of the simple subobjects are

J J
= 1,
J f
=f
and f f
= 1 f J .
(3.88)
Whereas in the free boson one would have easily guessed the product structure on A, in
the E7 case at this point we do have to find a solution to polynomial equations. However,
we will see (though not in all detail in the present paper) that all the rest of the calculations,
down to the structure constants, then reduces to solving systems of linear equations. Let
us write the relevant equations explicitly as relations that the numbers mab c must satisfy
in order to be the components of a valid multiplication morphism. First, according to
Lemma 3.6(ii) the Frobenius property imposes the restriction maa 1 = 0; we can therefore
choose the basis elements in Hom(f, A) and Hom(J, A) such that
mff 1 = mJ J 1 = 1.
(3.89)
The system (3.78) of polynomial equations encoding associativity is finite, and it is not
difficult to write out all equations when A is of the particular form (3.87). This system of
conditions is necessary and sufficient for the numbers mab c to provide an algebra structure.
After some manipulations, one deduces that these conditions imply
mff J = mf J f = mJf f ,
(J f f ) J
(mff J ) = F1 f
412
mff f
2
(f f f ) f 1
(f f f ) f
(J f f ) J (f f f ) f
1 F1 1
.
= Ff 1
F1 f
FJ 1
(3.90)
A priori this fixes the algebra structure up to sign choices only. But in fact the convention
(3.89) still permits a sign flip both in Hom(f, A) and in Hom(J, A). As a result, the choice
of sign allowed by (3.90) for the numbers mff f and mff J can be absorbed into a change
of basis, and hence (3.90) determines the algebra structure uniquely. Thus, if it exists, the
algebra structure on A is unique up to isomorphism.
(f f f ) f
Note that in (3.90) it is assumed that Ff 1
= 0 (which requires in particular that
f f E f ). This holds true for su(2)16 , but it need not hold in general. (If not, then the
calculation will proceed differently from what is reported in the sequel; we do not discuss
this case here.) In the same spirit we will also assume that mff f = 0. Again this may not
be satisfied in some examples, but it does hold for su(2)16 . With these two assumptions,
the remaining associativity constraints are equivalent to
J = f = 1,
(J f J ) f
Ff f
(f J f ) J
= Ff f
= 1,
(J f f ) f
(f J f ) f
(f f J ) f
(f f f ) J
Ff f
= Ff f
= Ff f
= Ff f
= 1,
(J f f ) J (J J f ) f
(J f f ) J
(f J J ) f
F1 f
Ff 1
= 1,
F1 f
= F1 f
,

2 (f f f ) f
2 (f f f ) f
(f f f ) f
F1 k
+ mff J FJ k
+ mff f Ff k
2 (f f f ) f
2 (f f f ) f

(f f f ) f
F1 k
+ mff J FJ k
+ mff f Ff k
for k I \ {1, J, f } and Nffk = 1.
(J J f ) f
Ff 1
= mff

k 2
(f f J ) J
= Ff 1
for k {J, f },
=0
(3.91)
Owing to the pentagon identity fulfilled by the fusing matrices, not all of these
requirements are independent. At first sight it might look surprising that the pentagon
identity for morphisms from f J J f to f implies that J has FrobeniusSchur
indicator J = 1 (recall formula (2.47)). However, on general grounds [97], for self-dual
f and J the fact that NJf f is 1 already implies that J f = f .
But not all of the relations (3.91) follow from the pentagon, as one can easily convince
oneself by finding explicit counter examples. So A = 1 f J can only be endowed
with an algebra structure in rather special cases, as befits the fact that it should describe an
exceptional modular invariant. We verified numerically that the relations (3.91) hold true
for su(2)k with k = 16. In fact they also hold for k = 8 for which they yield the D-invariant
(see Section 5.6.2), but not for various other k we tested. For k = 4 the object 1 f J
possesses an algebra structure, too (it yields just the A-invariant), but with mff f = 0 so
that the associativity constraints look different from (3.91).
4. Representation theory and boundary conditions

4.1. Representations and modules
When studying an ordinary algebra over the complex numbers (or any other number
field), a key concept is that of a representation of the algebra, together with the closely
related notion of a module, i.e., the vector space on which a representation acts. Moreover,
413
the modules of an algebra (or analogously, of a group, a Lie algebra, a quantum group or
similar algebraic structures) are the objects of a category, with morphisms given by algebra
(or group, Lie algebra, . . . ) intertwiners.
It is not difficult to extend these notions in such a way that they apply to algebra objects
in arbitrary tensor categories C. The basic notion is that of an A-module:
Definition 4.1. For A an algebra in a tensor category C, a (left) A-module in C is a pair
N = (N , ) of two data: an object N of C and a morphism of C that specifies the action of
N ). Further, must satisfy
A on N the representation morphism N Hom(A N,
the representation properties
(m idN ) = (idA )
and ( idN ) = idN .
(4.1)
Pictorially:
(4.2)
What is introduced here is called a left module because the action of A on N is
from the left. Analogously one can define right A-modules, on which A acts from the
right, i.e., for which the representation morphism is an element of Hom(N A, N).
And there is also the notion of a (left or right) comodule over a co-algebra, involving a
(or Hom(N , N A), respectively) that satisfies relations
morphism in Hom(N , A N)
corresponding to the pictures (4.2) turned upside-down, with coproduct and counit in
place of product and unit. Note that not every object U Obj (C) needs to underlie
some A-module, and that an object of C can be an A-module in several inequivalent
ways.
M Obj (C) that both carry the
Among the morphisms f between two objects N,
structure of an A-module, those that intertwine the action of A play a special role.
Here the notion of intertwining the A-action is analogous as for modules over ordinary
algebras:
414
Definition 4.2. For (left) A-modules N, M, an A-intertwiner is a morphism f between N

and M satisfying f N = M (idA f ), i.e.,
(4.3)
Given an algebra A in C, taking the A-modules in C as objects and the subspaces

M)
| f N = M (idA f )
HomA (N, M) := f Hom(N,
(4.4)
of the C-morphisms that intertwine the A-action as morphisms results in another category,
called the category of (left) A-modules and denoted by CA . Similarly as for morphisms in
C (see formula (2.3)), we will use the shorthand notation
dim HomA (M, N) =: M, NA
(4.5)
for any two A-modules M, N Obj (CA ).

Typical representation theoretic tools, like induced modules and reciprocity theorems,
generalise to the category theoretic setting (see, e.g., [13,36,87,98]) and allow one to work
out the representation theory in concrete examples. In particular, one shows that CA inherits
various properties of C. For instance, when C is modular and hence semisimple and when
A is special Frobenius, then the category CA is semisimple. On the other hand, that C is
modular does not imply that CA is modular. In fact, it does not even imply that CA is a
tensor category.
However, a sufficient condition for CA to be tensor is then that A is a commutative
algebra and has trivial twist, i.e., A = idA . The irreducible modules over such an algebra
A in a modular tensor category fall into two different classes, the local and the solitonic
modules. Local modules M can be characterised [87] by the fact that their twist is a
morphism in HomA (M, M). If M is simple it follows that M is a multiple of the identity,
i.e., all irreducible subobjects of M have the same conformal weight modulo integers. Let
us denote by (CA )loc the full subcategory of CA whose objects are the local A-modules. One
can check that the twist and braiding on C induce a twist and braiding on (CA )loc . Moreover,
if C is a modular category, then (CA )loc is modular as well [87].14 In CFT terms, (CA )loc
is the modular category for the chiral algebra V extended by the (primary) chiral fields
that correspond to the simple subobjects of the algebra A. Local modules correspond to
boundary conditions that preserve the extended chiral algebra, whereas solitonic modules
break the extended symmetry (while still preserving V) [101103].
14 The corresponding subcategory of C is not modular, but only pre-modular. It can be shown [99,100] that
such a category possesses a unique modularisation; this modularisation is precisely (CA )loc .
415
A simple object in the category CA is called a simple A-module. Similarly as in

Section 2.1 (see before formula (2.22)), let us introduce a set of definite representatives
for the isomorphism classes of simple A-modules; we denote them by M and the
corresponding label set by
J = {}.
(4.6)
As we will see later on, in the case of interest to us there are only finitely many isomorphism
classes of simple A-modules, i.e., J is a finite set.
That the category CA of A-modules is in general not a tensor category even when the
category C is modular, actually does not come as a surprise. The crucial property of CA is
that it is a so-called module category over C, and tensoriality is not a natural ingredient of
that structure. A module category [13,52,79,104] M over a tensor category C is a category
together with an exact bifunctor15
: M C M
(4.7)
that obeys generalised unit and associativity constraints: there are natural isomorphisms
from (M V ) W to M (V W ) and from M 1 to M, where M is an object of M
while V and W are objects of C.
In the case at hand, where M = CA is the category of left A-modules, the structure of a
module category amounts to the observation that for any left A-module M and any object
V of C, also M V has a natural structure of left A-module. Now recall that the tensor
product bifunctor of C is exact, implying that the Grothendieck group K0 (C) inherits from
the tensor product on C the structure of a ring, the fusion ring. It follows from the exactness
of the bifunctor (4.7) that the Grothendieck group K0 (M) carries the structure of a right
module over the ring K0 (C). Thus the notion of a module category over a tensor category
is a categorification of the algebraic notion of a module over a ring, in the same sense as
the structure of a tensor category categorifies the structure of a ring.
It can be shown [13] that every module category over a tensor category C is equivalent
to the category of left modules for some associative algebra A in C. Moreover, the
analysis (see Sections 3.2 and 4.4) of the OPE of boundary fields, including those that
change boundary conditions, shows that any unitary rational conformal field theory that
possesses at least one boundary condition preserving the chiral algebra gives rise to a
module category over the category of MooreSeiberg data. This provides another way to
understand the emergence of algebra objects which are central to our approach. (Indeed,
the reconstruction theorem of [13] allows one to recover not just one algebra in C, but
rather a family of Morita equivalent algebras.)
We are now also in a position to explain why the algebras of open string states
corresponding to different boundary conditions are Morita equivalent. The object M of the
module category M that describes a boundary condition is not necessarily simple, i.e., we
can allow for a superposition of elementary boundary conditions. According to Theorem 1
in [13], each such object M gives rise to an algebra AM in the underlying tensor category C,
15 It is conventional to denote this functor by the same symbol as the tensor product bifunctor of C.
416
which is constructed using the so-called internal Hom:

AM := Hom(M, M).
(4.8)
In the same theorem it is shown that the category of AM -modules is equivalent to the
module category M. Thus in particular, any two objects M1 and M2 of the module
category, i.e., any two boundary conditions, give rise to algebras AM1 and AM2 with
equivalent categories of left modules. By Definition 10 of [13], these two algebras of
open string states are therefore Morita equivalent. In fact, the two interpolating bimodules
can be given explicitly in terms of internal Homs as well, namely as Hom(M1 , M2 ) and
Hom(M2 , M1 ).
In any module category M, one can consider the morphism spaces HomM (M V , N)
for any pair M, N of objects of M and any object V of C. When A is a special Frobenius
algebra, as in the case of our interest, then semisimplicity of C implies that M = CA is
semisimple as well. One can then generalise the arguments given in Section 3.2 and choose
definite bases of the spaces HomM (M V , N) with M, N simple modules and V a simple
object, and then write down the matrix elements of the isomorphism

HomM (Mi Up , Ml ) HomC (Uj Uk , Up )
pI
= HomM (Mi Uj Uk , Ml )

(Mq Uk , Ml ) Hom(Mi Uj , Mq )
=
(4.9)
qJ
in these bases. Using the same graphs as in (2.36), but replacing the simple objects Ui , Ul
and Uq of C by simple objects Mi , Ml and Mq in the module category, one arrives at a
generalisation of the 6j -symbols (F-matrices). These symbols, often denoted by (1) F , are
labelled by three simple objects of M and three simple objects of C (as well as one basis
morphism of the type (2.29) and three other basis morphisms involving also modules).
These mixed 6j -symbols express the associativity of the bifunctor (4.7). They appear
naturally in the theory of weak Hopf algebras [8]. As already mentioned after formula
(3.12), in CFT they arise as structure constants for the operator products of boundary
fields.
Below it will be convenient to use the following construction of A-intertwiners:
M % ), with M, M % (left) A-modules, the A-averaged
Definition 4.3. For any Hom(M,
morphism is

:= M % idA ( M ) ( ) idM .
(4.10)
417
Pictorially:
(4.11)
Lemma 4.4.
M % ), where M, M % are
(i) When A is a Frobenius algebra, then for every Hom(M,
A-modules, the A-averaged morphism is an A-intertwiner,
HomA (M, M % ).
(4.12)
M % ), then = .
(ii) When A is special Frobenius and HomA (M,
(iii) When A is a symmetric special Frobenius algebra, then the trace of an endomorphism
does not change under averaging,
tr = tr .
(4.13)
Proof.
(i) The statement follows by a simple application of (both parts of) the Frobenius
property of A.
(ii) Start by using the fact that is an A-intertwiner, so as to move the representation
morphism M past to get another M % . Next employ the representation property for M % .
The A-lines can then be removed completely by first using specialness of A (recall the
convention A = 1) and the unit property for M % .
(iii) Start with the left trace, as defined in (2.14) (recall that since C is modular and
hence spherical, trr = trl =: tr). Owing to the cyclicity property of the trace we can pull
the upper A-line around until the two representation morphisms are directly composed.
Now use symmetry of A as in (3.35) and the first representation property (4.1). Finally
employ specialness (3.31) of A and the second representation property in (4.1). Upon these
manipulations the A-lines have completely disappeared.
We will also need the notion of an AB-bimodule. This is the following structure:
Definition 4.5. For A and B algebra objects in a tensor category C, an AB-bimodule
A , B ), consisting of an object M Obj (C) and two morphisms
M is a triple (M,
A
M)
and B Hom(M B, M),
such that (M,
A ) is a left A-module,
Hom(A M,
418
B ) is a right B-module, and such that the actions A and B commute, i.e.,
(M,

A idA B = B A idB .
(4.14)
When we want to make explicit the algebras over which M is a bimodule, we write it
as AMB . If C is a braided tensor category, then bimodules can equivalently be thought of
as left A Bop -modules, see Remark 12 in [13].
This is an important point, because it implies that we do not need to develop new
techniques to find all bimodules, once we know how to deal with left modules. In particular
the methods of induced modules described in Section 4.3 can be applied.
Let us formulate the correspondence between bimodules and left modules more
precisely.
Proposition 4.6. For A, B algebras in a braided tensor category, the mapping (see picture
(4.19))
f:

A , B M,
A idA B idA c1
M,
M,B
(4.15)
takes AB-bimodules to left A Bop -modules. f is invertible, with inverse g given by

g:

ABop (idA Bop id ), ABop (A c ) .
ABop M,
M,
M
M,B
(4.16)
Proof. To check that g f = id and f g = id is straightforward. Now suppose we

are given an AB-bimodule M. Substituting the prescription (4.15), the representation
property for M can be rewritten as in the first equality in
419
(4.17)
This equality, in turn, directly implies the second equality in (4.17), which precisely spells
out the representation property for a left (A Bop )-module. (Note that in the second
equality, the multiplication morphisms are those of A and B, not of Bop .)
Conversely, let us suppose that M is a left (A Bop )-module, and verify that g(M)
provides an AB-bimodule. It is easily checked that ABop (idA Bop idM )
does provide a left representation of A. That ABop (A cM,B
) provides a right
representation of B and that the two actions commute amounts to verifying that
and
420
(4.18)
Both of these properties follow from the representation property of left (A Bop )-modules
as given in the second equality in (4.17).
Remark 4.7.
(i) Proposition 4.6 provides an isomorphism f that takes AB-bimodules to left (A
B (1) )-modules. In an analogous manner one constructs an isomorphism f that takes
AB-bimodules to left (B (1) A)-modules. In pictures, the two isomorphisms are
given by
(4.19)
(ii) It can be shown that for any Frobenius algebra A in a tensor category C, the category
A CA of AA-bimodules is again a tensor category. This is in remarkable contrast to the
category CA of left A-modules, which in general cannot can be equipped with a tensor
product. Moreover, when C is sovereign and A is symmetric, then A CA is equipped
with a duality. However, in general the bimodule category does not have a braiding,
and therefore it does not have a twist either.
421
4.2. Representation functions

M
for the morphism spaces16
For any given A-module M we choose bases b(i,)
and dual bases b(j,) for Hom(M,
Uj ) such that
Hom(Ui , M)
M
(j,)
bM
M
b(i,)
= i idUi .
(4.20)
Pictorially:
(4.21)
Dominance in the category C implies the completeness property
(i,)
M
b(i,)
bM
= idM ,
(4.22)
iI
i.e.,
(4.23)
and g Hom(M,
Y)
Next let us associate to each pair of morphisms f Hom(X, M)
the morphism
(f, g) := g M (idA f ) Hom(A X, Y ).
(4.24)
We can then introduce the following notion:
16 The label i I refers to the chosen representatives U of the isomorphism classes of simple objects in C,
i
see Section 2.1.
422
Definition 4.8.
(i) For any A-module M the morphism
(4.25)
M , b (j,) basis elements in Hom(U , M)
Uj ) as
(with i, j I and b(i,)
i and Hom(M,
M
described above) is called a representation function of M.
(ii) For every j I, the j th character of M is the sum
j
M :=
(j,)
M(j,) Hom(A Uj , Uj )
(4.26)
of representation functions, and the j th dimension morphism Dj of M is

j
Dj (M) := M ( idUj ) End(Uj ).
(4.27)
This diction is sensible because in the special case C = Vect(C) and A the group algebra
C[G] of a group G, the quantities (4.25) are indeed ordinary representation functions
(M (g))ba . (Also, a module possesses then only a single character, since Vect(C) has only
a single isomorphism class of simple objects.) Besides the representation property
M (g)ba M (h)cb = M (gh)ca ,
(4.28)
ordinary representation functions for simple modules M, M % also obey the orthogonality
relation
b
1
1
%.
d b
M (g) a M % (g)dc =
a c M,M
|G|
dim(
M)
gG
(4.29)
In our context, the representation property amounts to the identity

(k, )
M(i,) (m idUi ) =

(j,)

(k, )
(j,)
M,(j,) idA M,(i)
(4.30)
423
in Hom(A A Ui , Uk ), i.e.,
(4.31)
(Use completeness (4.23) and the representation property (4.1) to see this equality.) The
generalisation of orthogonality is less direct. We find:
Lemma 4.9. The representation functions (4.25) for simple modules M , M % (, % J )
of a special Frobenius algebra A satisfy

(i,)
(k,)
M % (j,) idA M (l, ) ( ) idUl
dim(Ui )
(i,) (k,)
, % (j,)
(l, ) idUl .
dim(M )
(4.32)
Pictorially:
(4.33)
Proof. Since Uk and Ul are simple objects (and we have restricted our attention to only a
single representative of each isomorphism class), the morphism on the left-hand side can
be non-zero only for k = l. For the same reason we also need i = j , and we will tacitly
assume this in the sequel. Similarly, noting that the left-hand side involves an A-averaged
M %
(j,)
bM
morphism (see Definition 4.3)namely the one associated to = b(i,)
Hom(M , M % )and that by Lemma 4.4(i) is an A-intertwiner, we can use the fact
424
that M and M % are simple modules, and non-isomorphic for = % , to conclude that
i of id , and hence the
M = M % =: M. Moreover, must then be a multiple
M
morphism on the left-hand side of (4.32) is equal to
i (l,)
M
W =
bM b(l,
).
(4.34)
i we take the trace of (4.34) and sum over l I and over

To determine the coefficients
and Hom(M,
Ul )). Using the cyclicity of the
= (i.e., over dual bases in Hom(Ul , M)
trace and then the completeness property (4.23) yields

i
(4.35)
tr W =
dim(M).
lI
On the other hand, when applying the trace and summation on the left-hand side of (4.32),
and then again using the cyclicity and completeness, we get

(4.36)
tr W = tr = tr = dim(Ui ),
lI
where in the second step one invokes Lemma 4.4(iii) and in the last step once more cyclicity
of the trace, as well as the orthogonality (4.21) is used. Comparison of the two results yields
i = dim(U )/ dim(M)
and thereby establishes the validity of (4.32).
Corollary 4.10. The characters (4.26) of the simple A-modules M , M % satisfy

dim(Ui )
i
i
Ui , M idUi ,
idA M
( ) idUi = , %
M
%
dim(M )
(4.37)
and the dimension morphisms (4.27) obey17

Di (M ) = Ui , M idUi .
(4.38)
Proof. Formula (4.37) follows immediately from (4.33) by setting j = k = l = i, = ,

= and summing over and . Eq. (4.38) is a direct consequence of the unit property
of and the orthogonality (4.21).
Note that in the category of vector spaces, evaluating the character of a module at the
unit element yields the dimension of the module. In the situation studied here, Di (M)
tells us with which multiplicity the simple object Ui occurs in M, and we obtain the
(quantum) dimension of M as

=
dim(Ui ).
dim(M)
(4.39)
tr Di (M) =
Ui , M
iI
iI
17 Recall the shorthand X, Y = dim Hom(X, Y ) introduced in (2.3).
425
4.3. Induced modules

Later on the A-modules will label the boundary conditions of the CFT that A defines.
Given an algebra A in C, we would therefore like to find all A-modules obtainable from
objects in C. A powerful method to find A-modules is the one of induced modules, which
we recall briefly in this section. The main result is that if A is special Frobenius, then every
A-module is a submodule of an induced module. The extension of the method of induced
representations to a general category theoretic setting was developed in [36,87].
Definition 4.11. For U Obj (C), the induced (left) module IndA (U ) is the A-module that
is equal to A U as an object in C, and has representation morphism m idU , i.e.,
IndA (U ) := (A U, m idU ).
(4.40)
Pictorially:
(4.41)
That the morphism (4.40) satisfies the representation properties (4.1) follows directly
A (U ) = (U
from the associativity of m and the unit property of . Analogously, Ind
A, idU m) is a right A-module, called an induced right module. For the tensor unit 1
A (1). Moreover, the product
one obtains this way just the algebra itself, IndA (1) = A = Ind
of A is a morphism between A A and A as AA-bimodules, m HomA|A (A A, A),
while the coproduct is in HomA|A (A, A A) if and only if A is Frobenius.
Proposition 4.12 (Reciprocity). For A a Frobenius algebra in a tensor category C,
U Obj (C) and M an A-module, there are natural isomorphisms

=

Hom(U, M)
HomA IndA (U ), M and

=
U )
HomA M, IndA (U ) ,
Hom(M,
(4.42)
and (idA ( M )) (( )
given by M (idA ) for Hom(U, M)
idM ) for Hom(M, U ). In particular,

= IndA (U ), M
U = M, IndA (U ) .
U, M
(4.43)
and M,
A
A
For a proof see, e.g., [36], Propositions 4.7 and 4.11.18
18 Via these maps one actually gets functors C C (induction) and C C (restriction). Induction is a
A
A
right-adjoint functor to restriction, and for modular C both functors are exact, i.e., carry exact sequences of
426
Proposition 4.13. If A is a special Frobenius algebra in a semisimple tensor category C,

then every A-module is a submodule of an induced module, and the category CA of Amodules is semisimple.
For a proof see, e.g., [36], Lemma 5.24 and Proposition 5.25.
That CA is semisimple means that every A-module can be written as the direct sum of
finitely many simple A-modules. Furthermore, every simple module already appears as a
submodule of an induced module IndA (U ) with U a simple object of C. For C a modular
tensor category, which is the case under study, there are only finitely many isomorphism
classes of simple objects, and as a consequence there are then also only finitely many
isomorphism classes of simple A-modules. Thus as announced, the index set J (4.6) is
finite.
One important aspect of Proposition 4.12 is that it allows us to learn already quite a
lot about the structure of simple A-modules from the fusion ring of C alone. In simple
caseslike for the free boson and for the algebra giving the E7 modular invariant, which
are treated in Section 4.5the reciprocity relation already determines how many simple
A-modules there are and into which simple objects of C they decompose as elements of
Obj (C).
Even when U is a simple object of C, the induced module IndA (U ) will in general not
be a simple A-module. Indeed, we obtain:
Corollary 4.14.
(i) The module IndA (Uj ) (j I) decomposes into simple modules M ( J ) according
to

IndA (Uj )
(4.44)
Uj , M M .
=
J
(ii) For every simple object U of C the sum rule

dim(A) =

1
dim(M )U, M
dim(U )
(4.45)
holds.
Proof.
(i) Formula (4.44) follows immediately by applying the reciprocity relation (4.43) to the
simple A-modules M .
(ii) Since dimensions are constant on isomorphism classes of objects, we can restrict
our attention to U = Uj for j I. For these objects the result follows by taking the trace
of both sides of the relation (4.44).
morphisms to exact sequences [36]. Moreover, even when CA is a tensor category, the induction functor is not a
tensor functor. In contrast, there are functors from C to the bimodule category A CA (so-called -induction, see
Section 5.4) which are tensor functors.
427

In the vector space case the relation analogous to (4.45) reads dim(A) = (dim(V ))2 .
Thus when this formula is extended to general tensor categories, the two factors of dim(V )
get generalised in two distinct ways, one of them to dim(M )/ dim(Ui ), the other to
these two factors take the
Ui , M . Let us also mention that for simple current algebras,
form dim(M )/ dim(Ui )= |G|/ si ui and Ui , M = si /ui , respectively, where G is

the simple current group, si is the order of the stabiliser of Ui (the subgroup of G of simple
currents that leave i fixed), and ui the order of the so-called untwisted stabiliser (for its
definition see [105,106]).
We are now also in a position to make the following assertion that will be useful later
on (see Section 5.7):
Theorem 4.15. Let X, Y be simple objects of C and A a symmetric special Frobenius
algebra in C. Then the prescription (4.24) furnishes a natural isomorphism

Hom(X, M ) Hom(M , Y ) Hom(A X, Y ).
:
(4.46)
J
Proof. Recall that the direct sum in (4.46) is over one representative M out of each
isomorphism class of simple A-modules. In the following for simplicity we will just write
in place of M . To show injectivity, we apply to a basis of the space on the left-hand
side as in (4.25). Suppose that
(j,) (j,)
(i,) (i,) = 0
(4.47)
J i,j I ,
(j)
for some complex numbers (i) . Upon composition with another representation function
(j % , % )
(j % , % )
% (i % , % ) and averaging over A, the left-hand side yields % (i % , % ) by Lemma 4.9. Hence all
(j)
the numbers (i) are zero.

Surjectivity is checked as follows. Using the additivity of the ith dimension morphism
(4.27), it follows from Corollary 4.14 that

Uj , M Di (M ) =
Uj , M M , Ui idUi
Di IndA (Uj ) =
J
dim

Hom(Uj , M ) Hom(M , Ui ) idUi .
(4.48)
On the other hand, using directly formula (4.38) we get

Di IndA (Uj ) = Ui , A Uj idUi dim Hom(A Uj , Ui ) idUi .
(4.49)
Comparison of (4.49) and (4.48) establishes surjectivity of the map (4.46) for the case that
X = Uj and Y = Ui . By additivity this extends to the case of arbitrary objects X and Y
(using also that C is semisimple).
In the special case that A is a group algebra in Vect(C), the theorem amounts to the
statement that for complex semisimple algebras the representation functions span the dual
space of the algebra.
428
An interesting consequence of the relation (4.33) is that it allows us to invert the

isomorphism (4.46) explicitly. We find:
Lemma 4.16. The coefficients , in the expansion

(j,)
=
,
M (i,)
(4.50)
J ,
of a morphism Hom(A Ui , Uj ) with respect to the representation functions (4.25)

read

(i,)
dim(M )
M
idA bM M ( ) b(j,)
.
idj =
,
(4.51)
dim(Ui )
(j,)
Proof. Since the morphisms M (i,) form a basis of Hom(A Ui , Uj ), there exists
a unique set of numbers , with property (4.50). Their values can be extracted by
composing the right-hand side of (4.50) with a representation function and integrating
over A. In this way one arrives at the expression (4.51) for , . Pictorially,
(4.52)
4.4. From boundary conditions to representations

In Section 3.2 we have seen how a boundary condition (preserving the chiral algebra V)
in a rational CFT gives rise to a symmetric special Frobenius algebra A in the modular
tensor category of the CFT (the representation category of V). Let us henceforth also label
the boundary condition from which this algebra arises by A. Recall that the OPE of two
boundary fields living on the A-boundary furnishes the multiplication of the algebra A,
and the sewing constraint for four boundary fields on a disk assures its associativity.
The notion of an A-module has a direct interpretation in CFT terms as well. Consider
a stretch of boundary for which at some point y the boundary condition changes from A
to another boundary condition, say M. Then at the point y a boundary changing field
AM (y) must be inserted. The OPE of any field (x) that lives on the boundary A with
the boundary changing field AM (y) is forced to be of the form

(x y)k kAM (y),
(x) AM (y) =
429
(4.53)
where kAM is a suitable collection of boundary changing fields. The boundary OPE thus
defines a mapping

fields
boundary changing
boundary changing
.
(4.54)
living on A
fields A M
fields A M
This is the analogue of the representation morphism M as defined in formula (4.1).
To make this qualitative description precise, let us write out the sewing constraint
and the representation property in a basis and check whether they agree. We start with
the general sewing constraint for four boundary changing fields on a disk. For two (not
MN (x) denote a boundary changing
necessarily distinct) boundary conditions M, N , let a
19
field that is primary with respect to the chiral algebra V and carries the representation
a of V. The index counts its multiplicity. For three boundary conditions M, N, K, let
(MNK)c ;
MN (x) NK (y),
Ca,b
be the boundary structure constants appearing in the OPE a
b
defined by a direct generalisation of the OPE (3.11). Then the general sewing constraint
reads [14,74,76]
nfKM
(MNK)f ;
Ca,b
(MKL)d;
Cf ,c
=1
L
neN
Nbc Nae

(NKL)e;K
Cb,c
(a b c) d
(MNL)d;
Ca,e
F eK,f .
(4.55)
=1 eI K=1 =1
Here nfKM gives the multiplicity of the boundary changing field fMK
, and similarly for
N = A N.
L
neN . In fact, these are nothing but the annulus coefficients (see Section 5.8), niM
iM
Let us now write out the representation property (4.1) in a basis. To do so define the
numbers
(4.56)
19 The boundary fields living on a given boundary M are just those fields that leave the boundary condition
MM (x).
invariant. Thus in the present notation they are written as a
430
In terms of these the representation property reads
f,
M
f ; M (d);
ma,b (f )(c )
eI
=1
Nbc Nae
M
e,

M (e);K M (d); (a b c) d
(b)(c ) (a)(e) F eK,f .
(4.57)
=1 K=1 =1
Thus we see that upon setting

(AAA)c ;
Ca,b
c ;
= ma,b
(AAM)j ;
and Ca,i
M (j );
= (a)(i) ,
(4.58)
formula (4.57) becomes the special case M = N = K = A, L = M of (4.55). As an object

in C, the module defined by the boundary condition M is given by
M
M
=
niA

Ui ,
M
= niA
i.e., i, M
.
(4.59)
iI =1
Thus the module M consists of all representations of the chiral algebra V that occur as
boundary changing fields A M, counted with multiplicities.
Together with the results of Section 3.2 we have now learned the following. Fix an
arbitrary boundary condition A; this defines an algebra object. Moreover, every boundary
condition (including A itself) defines a module over this algebra. In other words, in our
approach boundary conditions are labelled by modules of the algebra object A. It also turns
out that elementary boundary conditions correspond to simple A-modules, see Section 5.8.
Direct sums of simple modules, in contrast, correspond to ChanPaton multiplicities,
which play an important role in type I string theories (for a recent review see [107]). It
will be shown in a forthcoming paper that with the help of the data provided by the algebra
A one can indeed define structure constants that solve the sewing constraint (4.55) in full
generality (not merely in the special cases (3.14) and (4.58)).
There is no intrinsic meaning to the statement that boundary conditions in a CFT
correspond to left rather than to right A-modules. This is just a convention we chose,
and everything could as well be formulated in terms of right modules. Given a left
M ) we can construct a right A-module (M , f (M )), where
A-module M = (M,
f (M ) Hom(M A, M ) is given by
f (M ) = (dM idM ) (idM M idM ) (idM idA bM ).
(4.60)
M ) (M , f (M )) is invertible and that

It is easy to check that the mapping (M,
f (M ) is a right representation morphism (as defined below (4.2)). The right module
M ).
(M , f (M )) describes the same boundary condition as the left module (M,
The notion of an AB-bimodule, given in Definition 4.5, also has a CFT interpretation,
which we mention briefly here (it will be studied in more detail elsewhere). For A = B, the
bimodules correspond to generalised disorder lines, a notion introduced in [34] and studied
further in [16,108]. For A = B, bimodules describe tensionless interfaces between two
CFTs. The CFTs on each side of the defect are the ones obtained from the algebra objects
A and B, respectively. Thus they contain the same chiral algebra V and in particular have
the same Virasoro central charge. These defects are special cases of those considered in
431
[109,110], where they are treated as boundary states in the product20 of the two CFTs (this
is known as the folding trick, see, e.g., [111,112]).
In this section we study the representation theory of A under the simplifying
assumptions that Nij k {0, 1} for all i, j, k I and that dim Hom(k, A) {0, 1} for all
k I. In contrast, we do not restrict our attention to such modules which contain simple
objects only with multiplicity zero or one.
M (j );
The notation (a)(i) introduced in Eq. (4.56) to describe the representation morphism
M j
in a basis now reduces to a , i , i.e.,
(4.61)
When expressed in terms of these numbers, the representation properties (4.1) also take an
easier form than in formula (4.57):
M j M k (a b i) j
M j
M j
0 , i = i,j , and
(4.62)
a , k b , i Fk c
= cA mab c c , i .
(k )M
Solving this polynomial equation is not easy. But fortunately we can invoke Proposition 4.13, which states that every irreducible module is a submodule of an induced module.
Ai ;
Therefore, instead of attacking (4.62) directly, we start by working out the matrix a , k
of an induced module IndA (Ui ). We choose the bases
(4.63)
20 This product must not be confused with the CFT associated to the tensor product of two algebras, see the
discussion at the beginning of Section 3.5
432
in Hom(Uk , A Ui ) and Hom(A Ui , Uk ); they are dual to each other in the sense of
(4.21). The basis vectors are labelled by those a A that occur in the fusion Uk U . The
set is complete, since dim Hom(Uk , A Ui ) = dim Hom(Uk U , A).
To find the representation matrix (4.61) in this basis, one must evaluate
(4.64)
Using the definition of G (2.40), this leads to the simple formula
(a b i) l
lc
c
aAi
, kb = mab Gc k
(4.65)
In case the induced module IndA (Ui ) is a simple module, then we are already done. If it is
not, then we must find a projector decomposition of the space HomA (IndA (Ui ), IndA (Ui ))
to extract the simple submodules. This may require some work, but it amounts to solving
linear equations only.
Suppose now that a simple module M that is not itself an induced module occurs as a
submodule in
IndA (Up )
= IndA (Ui1 ) IndA (Uin ) M,
(4.66)
with all induced modules IndA (Uim ) being simple.21 Given (4.66), we can extract the
representation matrix M with a little extra work from our knowledge of the induced
representations. To this end, we choose a special basis in the spaces Hom(Uk , A Up )
together with a dual basis in Hom(A Up , Uk ). We do so in two steps. First, define the
21 This is of course not the generic situation. In particular, it may well happen that two unknown simple modules
M and N always occur as a pair M N , in which case one cannot escape the full projector calculation. But the
special situation studied here is, e.g., realised in the E7 example that we will investigate below.
433
morphisms
and
(4.67)
These are indeed dual to each other,
Aj
Ai
Yk,cd Xk,ab
= i,j a,c b,d idUk .
(4.68)
434
This follows from the fact that dim HomA (IndA (Ui ), IndA (Uj )) = i,j when IndA (Ui ) and
IndA (Uj ) are simple A-modules. In particular, the morphism on the left-hand side of
(4.69)
must be a scalar multiple of idAi . Composing both sides of this relation with unit and
counit, one determines bc = b,c / dim(A). Another important property of the morphisms
X, Y is that they allow us to build elements in HomA (IndA (Up ), IndA (Up )):

HomA IndA (Up ), IndA (Up )
(4.70)
The morphisms X are linearly independent, but do not yet form a basis for all of
Hom(Uk , A Up ), just because the module M is still missing. So we must find elements
M
Xk
Hom(Uk , A Up )
M
and Yk
Hom(A Up , Uk )
(4.71)
such that we have a basis in each of these spaces, and such that the Y s are still dual to the
Xs. Using this basis, we can apply dominance in the form

Ai
Ai
M
M
Xk,ab
Yk,ab
+
Xk
Yk
.
idAUp =
(4.72)
i,kI a,bA
kI
435
Now since both the left-hand side and the first term of the right-hand side of this formula
are in HomA , the last term is in HomA as well. To finally obtain the representation matrix
M , express XM and Y M in a basis as
and
(4.73)
The relation
(4.74)
M
then determines
to be

M
M ;
(a b p) ;
M
a , k =
Xk
Y
m c Gc k
.
b ; c ab
(4.75)
b,c
(Recall that the label p refers to the induced module IndA (Up ) form which the simple
submodule M has been extracted.)
436

For the free boson, all induced modules turn out to be simple. (Thus the second half of
the arguments above is not needed.) To see this, recall the definition of the algebra object
A2r in formula (3.84). The dimensions of the morphism spaces HomA are then given by

dim HomA IndA2r [n] , IndA2r [m]

1 if n m = 0 mod 2r,
= dim Hom [n], A2r [m] =
(4.76)
0 otherwise.
In particular, indeed every induced module is simple. But not all of them are distinct.
Rather, we have

IndA2r [n]
(4.77)
= IndA2r [n + 2r]
for any n (and no further isomorphisms). A set of representatives for each equivalence class
of simple modules is thus given by

Mk = IndA2r [k] for k = 0, 1, . . . , 2r 1.
(4.78)
The multiplication can be read off the formula for induced modules (4.65):
Mn [k+a]
[a+kn] ([a] [kn] [n]) [k+a]
[a]
G[a+kn] [k]
=1
, [k] = m[a][kn]
(4.79)
(and 0 else). In deriving this formula one uses that a A is even and that k n 0 is a
multiple of 2r and thus is even, too. Note that the 11-matrix F in (2.77) is just a sign and
hence equals its own inverse, G = F.
For the E7 modular invariant we need to determine the representation theory of the
algebra object A = (0) (8) (16) in the modular category formed by the integrable
highest weight representations of su(2)16 .
As a first step we work out the embedding structure of the induced modules. The
following table gives dim HomA (IndA (i), IndA (j )) for i, j = 0, 1, . . . , 8. (For easier
reading, in the table zero entries are indicated by a dot.)
0
1
2
3
4
5
6
7
8
2 3
4 5
6 7

1
1

1
1
1
1

1
2
2
1
1
1
(4.80)
All other cases follow from the table via IndA (r)
= IndA (16 r), an isomorphism due to
the simple current (16). From the diagonal entries of (4.80) we see that the four induced
437
modules
IndA (0),
IndA (1),
IndA (2),
IndA (3)
(4.81)
are simple, while the other induced modules decompose as

IndA (4)
= P Q,
IndA (5)
= IndA (3) R,
IndA (7)
= IndA (1) IndA (3),
IndA (6)
= IndA (2) Q,
IndA (8)
= IndA (0) IndA (2) P ,
(4.82)
with simple A-modules P , Q, R that are not isomorphic to any induced module. As objects
in C, these simple modules decompose into simple objects as
P
= (4) (8) (12),
R
= (5) (11).
Q
= (4) (6) (10) (12),
(4.83)
In agreement with the discussion above we can extract P from IndA (8), Q from IndA (6)
and R from IndA (5). Let us start with IndA (8). Its decomposition into simple objects of C
reads
A (8)
=
(0)
(2)
(8)
(16) [from A (0)]
(6) (8) (10)
(14)
[from A (2)]
(4)
(8)
(12)
[from P ]
(4.84)
One verifies that to complete the bases in the spaces Hom((4), A (8)), Hom((8), A (8))
and Hom((12), A (8)) one may choose the elements
(4.85)
438
P X P = id ,
where the normalisation constant is defined implicitly by the condition Y(8)
(8)
(8)
and the dual basis elements are
(4.86)
P . The calculation that this indeed extends XA(0)
The only non-trivial case is X(8)
(8)
A(2)
and X(8)
(in the conventions defined in (4.67)) to a basis of Hom((8), A (8))
A(0)
A(2)
P amounts to the observation that

and Y(8)
with dual basis given by Y(8) , Y(8)
dim HomA (IndA (4), IndA (r)) = 0 for r = 0, 2.
The same procedure works to extract Q from IndA (6), giving the additional basis
elements
(4.87)
439
with duals defined in the same manner as in (4.86) and 1 , 2 defined implicitly by the
Q
Q
Q
Q
X(6)
= id(6) and Y(10)
X(10)
= id(10) .
conditions Y(6)
For the module R this short-cut does not work, and we need to solve a linear system to
complete the basis and its dual. But since both morphism spaces (embedding (5) or (11)
into A (5)) are two-dimensional, we can easily write down (one choice for) the missing
vectors:
(4.88)
where, in the notation (4.73) from above,
A(3)
A(3)
,
v = Y(5)
,
u = Y(5)
(8)
A(3)
A(3) (0)
x = X(5)
,
y = X(5)
.
(8)
(0)
(4.89)
R and Y R can be determined in the same way.

The morphisms X(11)
(11)
We have now gathered the necessary information to give the representation matrix of the
simple A-modules P , Q, R via Eq. (4.75). As an additional check of the result, we verified
numerically for a random sample of cases that these representation matrices indeed solve
the non-linear constraint (4.62) for su(2)16 .
5. Partition functions
Before proceeding, let us recall a few aspects of our general philosophy. The
construction we want to implement can be summarised as
chiral data + symm. special Frobenius algebra full CFT
(5.1)
By chiral data we mean the representation theory of the chiral algebra and the conformal
blocks of a rational conformal field theory. The corresponding modular category contains
440
strictly less information than the chiral data. Roughly speaking, the category encodes only
the monodromies of the blocks, but not their functional dependence on the insertion points
and the moduli of the world sheet or the information which state of a given representation
of the chiral algebra is inserted.22 But it turns out that much important information about
a CFT, like its field content, its boundary conditions and defect lines, the OPE structure
constants, and the consistency of these data with factorisation, can be discussed entirely at
the level of the modular category. It is a strength of the present approach that, in a sense, in
dealing with the modular category C one forgets exactly the right amount of complexity
of the chiral data to render those problems tractable.
In order to select which full CFT is to be constructed from a given set of chiral data,
additional input is required. In our approach this input is a symmetric special Frobenius
algebra object A in C. The full CFT that one obtains this waythe one indicated on the
right-hand side of (5.1)will be specified in terms of its correlation functions. To this end,
the correlators on an arbitrary world sheet X are expressed as specific elements in the vector
spaces of conformal blocks on the complex double
X of X. Such an element is described
by a ribbon graph in a three-manifold MX , called the connecting three-manifold whose
boundary MX is equal to the double
X. Which three-manifold MX is to be used, and
which ribbon graph in MX , was established for the Cardy case in [51,113] (for all possible
world sheets), and described in the general case in [22] (with restriction to orientable world
sheets).
The idea to study a chiral CFT as the edge system of a (topological) theory on a threedimensional manifold is quite natural and has been put forward in various guises, see for
instance [114116]. It finds a physical realisation in the CFT description of quantum Hall
fluids in the scaling limit (see [117] for a review). By exploiting chiral CFT on the complex
double
X, this relationship can be used to study full CFT on the world sheet X.
In the following we present this construction in the form that it takes for orientable
world sheets without field insertions. Afterwards we further specialise to the cases where
the world sheet is a torus or an annulus.
5.1. The connecting manifold and the ribbon graph
Having fixed a modular category C and a symmetric special Frobenius algebra A in C,
to construct the correlator on a world sheet without field insertions we must in addition
provide the following data:
22 More precisely, the modular tensor category encodes the information about the conformal blocks as abstract
vector bundles V over the moduli space (which for m insertion points labelled by - := (i1 , i2 , . . . , im ) I m is
the moduli space Mg,m of complex curves of genus g with m marked points). There are actually two different
formulations of TFT: the topological modular functor works with vector spaces and provides a representation of
the mapping class group. The complex modular functor gives a local system over the moduli space Mm,g . The
two descriptions are related by a RiemannHilbert problem, so that they contain equivalent information [39]. The
full chiral data provide much more: they specify an embedding of V- in the trivial bundle over Mg,m whose fiber
is the algebraic dual of the tensor product Hi1 Hi2 Him of the vector spaces underlying the irreducible
representations of the chiral algebra. This information allows to obtain the values of conformal blocks on all
chiral states.
441
An orientable world sheet X, possibly with boundaries;

an orientation on X;
for each component of the boundary X an A-module that specifies the boundary
condition.
Here an important point is that orientability of the world sheet is not sufficient. Rather, to
determine the correlator uniquely we need to select an actual orientation. For the correlator
to be defined unambiguously without specifying the orientation, more structure than just
a symmetric special Frobenius algebra is needed. This is linked to the fact that there exist
CFTs which cannot be defined on non-orientable world sheets while still preserving the
chiral algebra V, most notably those whose torus partition function is not symmetric,
Zij = Zj i . Since also such modular invariants can be described in terms of a Frobenius
algebra, some extra structure is needed to ensure that the theory is consistent also on
non-orientable world sheets. It turns out that the relevant extra structure is given by a
conjugation on A, which furnishes an isomorphism between A and Aop and may be thought
of as providing a square root of the twist. Not every symmetric special Frobenius algebra
possesses such an extra structure and for those who do, the conjugation need not be unique.
Moreover, the number of conjugations is not Morita invariant. We will return to this issue
in a future publication.
The second important aspect of the prescription is that boundary conditions are labelled
by A-modules. We will prove below that simple A-modules provide a complete set of
elementary boundary conditions. This implies in particular that the number of elementary
boundary conditions is the same as the number of Ishibashi states in the (dual of the) bulk
state space. A general A-module is a direct sum of simple modules and corresponds to a
superposition of elementary boundary conditions.
All these boundary conditions preserve the chiral algebra V whose representation
category is the modular category C. Recall that V is not necessarily the maximally extended
chiral algebrain the most extreme case, it is just the Virasoro algebra. As a consequence,
our formalism covers symmetry breaking boundary conditions as well, as long as the
subalgebra of the chiral algebra that is preserved by all boundary conditions under study is
still rational.
The complex double
X of an orientable world sheet X can be obtained by taking
two disjoint copies of X, with opposite orientation, and gluing them together along the
boundary:

X := X {1, 1} / with (x, 1) (x, 1) iff x X.
(5.2)
The connecting three-manifold MX is then the following natural interval bundle over X
[113]:

MX := X [1, 1] / ;
(5.3)
the equivalence relation now identifies the values t and t in all intervals {x} [1, 1]
that lie above a boundary point of X, i.e.,
(x, t) (x, t) for all x X and all t [1, 1].
(5.4)
442
One quickly checks that in this manner one indeed achieves MX =

X. Moreover, the
world sheet X is naturally embedded in MX , via the map
X MX ,
x (x, 0).
(5.5)
Thereby X is a retract of MX , and conversely the connecting manifold MX can be regarded

as a fattening of the world sheet (and hence in particular its construction does not introduce
any additional homotopical information). In the sequel we will always think of the world
sheet X as being embedded in MX in this fashion. Via the embedding (5.5) each boundary
component of X coincides with a circular line of singular points of MX that results from
the fixed points under the identification (5.4).
We now describe how the ribbon graph in MX is constructed. This construction involves
several choices. As we will show afterwards, the invariant associated to the graph is
independent of all these choices.
First choose a triangulation of the world sheet Xchoice #1. To be precise, the
faces of the triangulation are allowed to possess arbitrarily many edges, while all the
vertices are required to be trivalent (so strictly speaking, this is the dual of a genuine
triangulation).23
All ribbons lie in the two-dimensional submanifold X of MX . Furthermore, all
ribbons (regarded as embedded two-manifolds) are to be oriented in such a way that their
orientation coincides with the one induced from Xin short, the white side of each
ribbon faces up. The boundary of X is taken to be oriented such that upon an orientation
preserving map of a patch of X that contains a boundary segment to the upper half plane
the orientation of the image of the segment agrees with that of the real axis.
For every boundary component of X, labelled by an A-module M, place an annular
M-ribbon in X MX , along the circular line of singular points of MX that corresponds to
the boundary component. The orientation of the core of the ribbon must be taken to agree
with the orientation of the boundary component.
On those edges of the triangulation that are not part of the boundary of X, place Aribbons that are directed away from the vertices. In the middle of each edge these are to
be joined by the morphism 2 Hom(A, A ) that we defined in (3.32) and display once
again on the left-hand side of figure (5.6). As indicated on the right-hand side of (5.6), this
23 It is in fact sufficient to place an edge along each boundary component and along a basis of non-contractible
cycles of the two-manifold (resolving any four-point vertices that result from this prescription into two three-point
vertices). This amounts to the rule that the triangulation needs to be just fine enough such that any further edge
that is added can be removed by repeated use of the fusion and bubble moves given in (5.10).
443
can be done in two distinct ways; pick one of them on each edgethis is choice #2.
(5.6)
At every trivalent vertex in the interior of X, join the three outgoing A-lines with the
morphism 11 Hom(A , A A), see the left part of figure (5.7). This morphism
can be inserted in three different wayschoice #3.
At every trivalent vertex on a boundary component of X that is labelled by M, put
A M),
with M the
the morphism (idA M ) ( idM ) ( idM ) Hom(M,
representation morphism of M, as in the right part of figure (5.7).
(5.7)
This construction will be illustrated for the torus and the annulus in Sections 5.3 and 5.8,
respectively.
According to the discussion in Section 2.4, the invariant of the ribbon graph in MX
constructed as above is a vector in the space of conformal zero-point blocks on MX =
X,
which is the value of the correlator on X. We proceed to show that this element is
independent of the three choices made in the construction. We will call two ribbon
graphs in a three-manifold equivalent if they possess the same invariant for any modular
category C. In this terminology, what we are going to show is that different choices always
lead to equivalent ribbon graphs.
444
Choice #2two ways to insert 2 : the two fragments of the ribbon graph can be
transformed into each other as follows (all ribbons are A-ribbons):
(5.8)
The first equality is just a deformation of the ribbon graph, while the second equality
amounts to the fact that the algebra A is symmetric, as in (3.33).
Choice #3coupling three A-ribbons: it is sufficient to show that a 120 clockwise
rotation of the encircled vertex that appears on the left-hand side of (5.7) does produce
equivalent ribbon graphs, i.e., that (all ribbons in the figure are A-ribbons)
(5.9)
The left-hand side of this equation can be transformed into the right-hand side by first
using that A is symmetric, i.e., that 11 = 21 for the morphisms (3.35), on the lower
coproduct, and then coassociativity as in (3.27).
Choice #1triangulation of the world sheet: any two triangulations of a Riemann
surface can be transformed into each other via the so-called fusion and bubble moves (see,
e.g., [66,89,92]), which look as follows:
(5.10)
445
Two triangulations related by one of these moves give rise to equivalent ribbon graphs
if the following transformations can be performed with the help of the properties of the
algebra A and the module M: in the interior,
(5.11)
and on the boundary,
(5.12)
(Here the parts enclosed by dashed lines stand for the corresponding parts of figures (5.6)
and (5.7).) A slightly lengthy but straightforward calculation, using only the various
properties of A and M, i.e., the defining axioms of a symmetric special Frobenius algebra
and its representations, shows that this is indeed the case and thus proves independence of
the triangulation.
In the remainder of this section we will mainly deal with the following three-manifolds:
S 2 S 1 , D S 1 and S 3 , with D denoting a disk. The pictorial representation that we
will use for ribbon graphs in these manifolds is illustrated in the following figure, for the
example of the Hopf link:
446
(5.13)
S1
In the first two pictures, the vertical direction corresponds to the factor, and accordingly
top and bottom are to be identified. The first picture stands for the complex number given
by the invariant of the relevant ribbon graph (here the Hopf link) in S 2 S 1 ; the second
picture denotes an element in the vector space H(, T) where T is a two-torus. The third
picture again stands for the complex number that is given by the invariant associated to the
relevant ribbon graph, but now multiplied by the factor 1/S0,0 .
The latter convention avoids a proliferation of factors S0,0 . Indeed, recall (see formula
(2.57) and the subsequent text) that the invariant of a ribbon graph embedded in S 3 is
given by S0,0 times the complex number obtained from translating the ribbon graph into
a morphism in Hom(1, 1). The convention allows us to replace ribbon graphs in S 3 by
morphisms without the need to introduce explicit factors of S0,0 .
Let us also remark that on orientable world sheets, independence from the triangulation
is precisely what is needed for topological invariance in two-dimensional lattice topological theory. Indeed, triangulation independence of our results follows by the same arguments
[89] as in two-dimensional lattice TFT. Furthermore, the construction in [89] shows that
two-dimensional lattice TFTs are in one-to-one correspondence with symmetric special
Frobenius algebras in the category of complex vector spaces. Our construction of conformal field theory amplitudes can therefore be understood [22] as a natural generalisation of
lattice TFTs from the category of finite-dimensional complex vector spaces to more general
modular tensor categories.
It is also worth mentioning that [118] what is needed in order for triangulations with
the same number of faces to yield the same results, are only the properties of A to be
a symmetric Frobenius algebra. The property of A to be also special, on the other hand,
allows one to reduce the number of faces in a triangulation by the bubble move (5.10). The
latter issue can be expected to be much more subtle for irrational conformal field theories.
5.2. Zero-point blocks on the torus
The torus and annulus partition functions can both be expressed in terms of conformal
blocks on the torus. Let us therefore have a closer look at the space of torus blocks from
the TFT point of view.
Let T be the extended surface given by the oriented torus S 1 S 1 , without any
embedded arcs (and with a choice of Lagrangian subspace, to be detailed below). As
discussed in Section 2.4, the TFT assigns to T a vector space H(; T) (the symbol
makes explicit that the extended surface T does not carry any arc). The space H(; T)
has dimension |I|. A distinguished basis can be obtained in the following way: let M1 be
a solid torus with a k-ribbon running along its non-contractible cycle. Choose + M1 = T
and M1 = , and take the Lagrangian subspace in the first homology of T to be spanned
by the cycle in T that is contractible within M1 . The basis vectors |k ; T are then obtained
447
by applying the map Z(M1 , , T) to the number 1 C = H():

|k ; T = Z(M1 , , T)1 H(; T).
(5.14)
In pictures,
(5.15)
Interpreting H(; T) as the space of conformal blocks on the torus, these basis elements
correspond to the characters k of the irreducible highest weight modules of the chiral
algebra V. Note that one should actually think of the character as a one-point block on
the torus, with insertion of the vacuum representation. The character thus depends on
the modulus as well as on a field insertion (v; z) with v V (owing to translation
invariance there is no dependence on z). The Virasoro specialised characters are obtained
when choosing v to be the vacuum vector, corresponding to = 1, upon which only the
-dependence is left. When V is larger than the VOA associated to the Virasoro algebra,
the specialised characters need not be linearly independent. For example, k ( ) = k ( )
even when k = k.
Reversing the orientation of T takes the modulus to . In terms of specialised
characters the correspondence is
|k ; T k ( )

and |k ; T k .
(5.16)
In order to expand a general element |; T H(; T) in terms of this basis, we
also
need the dual basis vectors k ; T| in H(; T) . Then we can write |; T =
kI k ; T|; T|k ; T. Let M2 be a solid torus with a k-ribbon running along its noncontractible cycle. Take its orientation such that M2 = T. We choose + M2 = and
M2 = T (recall from Section 2.4 that M2 is defined to have reversed orientation).
Then Z(M2 , T, ) is a linear function from H(; T) to C, and we have
k ; T| = Z(M2 , T, ) H(; T) .
(5.17)
448
In pictures,
(5.18)
To verify that this is indeed dual to (5.15), use the identity map to glue M1 to M2 . Then
functoriality (2.56) implies k ; T|; ; T = Z(M2 , T, ) Z(M1 , , T) = Z(M, , ),
where M is the three-manifold S 2 S 1 , with two ribbons as in the following figure:
(5.19)
To see how the delta symbol arises, use relation (2.58) to rewrite the number Z(M, , )
as a trace over H(k, (;, ); S 2 ). Property (2.55) implies that the trace is taken over the
identity operator. This leads to k ; T|; ; T = dim H(k, (;, ); S 2). This dimension, in
turn, equals k,; , as discussed in Section 2.4.
By using surgery on the relation (5.19) one can relate the Hopf link (2.22) to the matrix
(Si,j ) of modular transformations of characters. To see this first recall [59] that any link in
449
S 2 S 1 is related to an equivalent link in S 3 via surgery; in the present situation,
(5.20)
Here X is an arbitrary (not necessarily simple) object and a morphism in Hom(X, X).
Concretely, this result is obtained by cutting out a small solid torus containing the i-ribbon
on the left-hand side of (5.20) and gluing it back after an S-transformation, yielding the
three-manifold S 3 with ribbon graph as on the right-hand side.
Upon setting i = ;, X = Uk and = idUk , (5.20) turns into

S;,j sj,k ,
k,; = S0,0
(5.21)
j
which by (S 2 )i,j = Ci,j i, implies that

si,j = Si,j /S0,0 .
(5.22)
5.3. The torus partition function

Applied to the torus partition function, the general construction in Section 5.1 proceeds
as follows. The world sheet X is a torus T, thus the double
X is the disconnected sum
T (T) of two copies of the torus, and the connecting manifold is MT = T [1, 1].
Next we pick a triangulation of the world sheet, and then convert this triangulation to a
ribbon graph embedded in MT . The left-hand side of the following figure displays the
triangulation we choose. The ribbon graph in MT obtained by inserting the elements (5.6)
and (5.7) is displayed on the right-hand side (the extension in the direction of the interval
[1, 1] is suppressed, all ribbons are labelled by A):
(5.23)
450
This graph can be simplified a lot using the relations for multiplication and comultiplication
of A; this way we arrive at
(5.24)
It is this ribbon graph which we will work with in the rest of this section.
Often we will need a move that reverses the orientation of (the core of) an A-ribbon to
see that two given ribbon graphs are equivalent. The following equality on segments of a
ribbon graph is a direct consequence of the Frobenius property of A:
(5.25)
Using this identity, one can reverse the orientation of the core of an A-ribbon. This is
achieved as follows: first replace a stretch of A-ribbon with the right-hand side of (5.25);
then move the multiplication and comultiplication appearing there to the respective ends of
the A-ribbon. Whenever both ends of the ribbon end on (the coupon for) a multiplication
or comultiplication morphism, one can then use the Frobenius or associativity properties
of A to remove the unit and counit.
It was already shown in Section 5.1 that the invariant (5.24) is independent of the
triangulation we start from. Further, invariance under the action of the relative modular
group amounts to the usual modular invariance of the torus partition function. In more
detail, consider the two transformations
U:
/( + 1)
and T :
+ 1
(5.26)
of the complex upper half-plane. The map U corresponds to the change of basis in the
lattice that takes {, 1} to {, 1 + }. It is expressible in terms of T and S : 1/ as
U = T ST . These changes of the fundamental region of the torus modify the graph (5.24)
as follows:
451
(5.27)
In agreement with our general arguments about triangulation independence, both resulting
graphs can be transformed back to (5.24) by using properties of A. Explicitly, for the
change of fundamental domain induced by T we have:
(5.28)
The ribbon graph (5.24) for Z describes an element in H(; T) H(; T). It can be
expanded in a standard basis of conformal blocks (i.e., characters) as
Z=
Zij |i , T |j , T.
(5.29)
i,j I
Via the correspondence (5.16), this tells us that in terms of specialised characters of the

chiral algebra the CFT partition function reads Z = i,j I Zij i ( )j ( ) .
452
To obtain the coefficients Zij we glue the dual basis elements for |i , T and |j , T
to the two boundaries of Z. This yields the ribbon graph
(5.30)
in S 2 S 1 .
Theorem 5.1. The numbers Zij given by the invariant of the ribbon graph (5.30) enjoy the
following properties:
(i)
[Z, S] = 0
and [Z, T ] = 0.
(ii)
Zij Z0 .
(5.32)
(iii)
Z00 = dim centA (Atop ).
(5.33)
(5.31)
(As usual, we denote the matrices that implement the modular transformations S and
T on the space of characters by S and T , respectively.) Before proving these claims, it is
useful to introduce, for any object X, a specific morphism PX Hom(A X, A X):
(5.34)
Lemma 5.2. PX is a projector:
PX PX = PX .
(5.35)
453
Proof. It is straightforward to establish the sequence
(5.36)
of equalities for PX PX . By first using the Frobenius property and then the specialness of
A one concludes that the morphism on the right-hand side is equal to PX .
Proof of Proposition 5.1. Property (i) has already been derived: Z is invariant under T
and U , and hence under T and S.
To obtain (ii), cut the picture (5.30) along an S 2 to arrive at a ribbon graph in S 2 [0, 1],
which defines a linear map
Pij :

H i, A, j ; S 2 H i, A, j ; S 2 .
(5.37)
The coefficients Zij are then recovered as

Zij = trH(i,A,j ;S 2 ) Pij .
(5.38)
The morphism described by Pij is nothing but idUi Pj with Pj PUj as defined in
(5.34). Since Pj is a projector, it follows in particular that Pij is a projector as well,
Pij Pij = Pij .
(5.39)
Now the trace of a projector equals the dimension of its image. Hence
trH(i,A,j ;S 2 ) Pij Z0 ,
(5.40)
which establishes (5.32).

To show (iii) we write Z00 = trH P00 = trH P0 with H H(A; S 2 ). Then we use
A k ) (the
dominance to insert a basis in the A-line, leading to Z00 = k, trH (P0 k
A
2
basis morphisms are those defined in (3.4)). The space H(k; S ) of blocks is non-zero only
454
for k = 0; thus we obtain
(5.41)
with the linear map
Ptop :
Atop Atop
P0 .
(5.42)
The second equality of (5.41) is just a translation in S 2 S 1 , while in the last step the
summation over is recognised as a trace in Atop . We conclude that Z00 = dim Im(Ptop ).
Noting the identity
(5.43)
we can now use Lemma 3.16 to conclude that Ptop = is equivalent to centA (Atop).
It follows that dim Im(Ptop ) = dim centA (Atop), thus proving formula (5.33).
Let us denote the torus partition function (5.30) by Z(A)ij so as to make its dependence
on the algebra object A explicit. Denote further by C the charge conjugation matrix
Ck; = k,;
and set

ij :=
Z(A)
(5.44)
Cik Z(A)kj .
kI
Also recall from Section 3.5 the definition of the opposite algebra Aop and of the direct
sum A B and product A B of algebras.
455
Proposition 5.3. The following relations are valid as matrix equations:

(i)
B) = Z(A)

Z(A
+ Z(B),
(ii)
B) = Z(A)
Z(B),

Z(A
(iii)
op ) = Z(A)
t,
Z(A
or equivalently, Z(Aop ) = Z(A)t .
(5.45)
Proof. (i) The definition of Z(A) gives immediately
= Z(A)ij + Z(B)ij
(5.46)
in S 2 S 1 . Multiplying by C from the left yields relation (5.45(i)).

(ii) For ribbon graphs in a solid torus D S 1 the relation
(5.47)
holds. Here the right-hand side is expanded in a basis of zero-point blocks on the torus, as
discussed in Section 5.2. That the coefficients are precisely given by Z(A)kj can be seen
by gluing a D S 1 containing a single i-ribbon to the left- and right-hand sides of (5.47).
On the left-hand side one thereby obtains the graph (5.30) for the torus partition function,
while the right-hand side gives ik . Next note that the ribbon graph for Z(A B)ij can be
456
deformed as
(5.48)
Now we can apply surgery to cut out a solid torus containing the B- and j -ribbons, and
then use (5.47) to obtain

= Z(A)CZ(B) ij .
This implies (ii).
(5.49)
457
(iii) The following relations are valid:
= Z(A)j i .
(5.50)
Here in the first step the definition of the multiplication on Aop in terms of A is inserted.
The second step uses the fact that in the horizontal direction the manifold has the topology
of a two-sphere S 2 . The A-ribbon is deformed around the two-sphere to wrap around the
i-ribbon (this step is best checked by visualising it with actual ribbons, rather than lines,
which facilitates keeping track of the twists). Finally we move the j -ribbon to the left side
of the graph and use Frobenius and associativity properties to change the A-ribbon so as to
match the one appearing in the defining relation (5.30).
op ) = C Z(A)
t C and
Thus we have Z(Aop ) = Z(A)t . This, in turn, is equivalent to Z(A
t

op ) = Z(A)
.
hence (because of [Z, C] = 0) to Z(A
Remark 5.4.
(i) Suppose A and B are algebras such that Z(A)00 = Z(B)00 = 1. Then Z(A B)00 = 2.
Accordingly the CFT should be interpreted as a superposition of two CFTs with
Z00 = 1, see the brief discussion at the end of Section 3.2. Unsurprisingly, it is a
superposition of the CFTs associated to A and B, as is confirmed on the level of
partition functions by Proposition 5.3(i).
(ii) It is obvious that for two matrices Z1 and Z2 that commute with S and T , the product
Z1 CZ2 also commutes with S and T . It is less obvious that this matrix appears as the
torus partition function of any consistent CFT. However, when both Z1 and Z2 are
458
obtained from algebra objects, then according to the result above Z1 CZ2 is obtained
from an algebra object as well, and is therefore indeed realised as the torus partition
function of a CFT. (If the tensor product A B is isomorphic to a direct sum of
algebras, the resulting theory can be interpreted as a superposition of several CFTs
with Z00 = 1, as discussed in point (i).)
(iii) In terms of Morita classes of algebras, the tensor product of CFTs amounts to a
product given by [A] [B] := [A B]. As an example, consider the WZW theories for
su(2)k and u(1)2N . For su(2)k , denote by [A] the Morita class of all algebras giving
rise to an A-series modular invariant, and similarly for [D] and for [E6], [E7] and [E8].
The A-invariant occurs at all levels k; its product with itself is [A] [A] = [A]. In all
other cases, too, the multiplication is commutative and [A] acts as unit. Further, one
computes
k 4Z:
k 4Z + 2:
k = 10:
k = 16:
k = 28:
[D] [D] = 2[D],

[D] [D] = [A],
[D] [E6 ] = [E6 ],
[D] [E7 ] = 2[E7],
[D] [E8 ] = 2[E8],
[E6 ] [E6 ] = 2[E6 ],

[E7 ] [E7 ] = [D] + [E7 ],
[E8 ] [E8 ] = 4[E8 ].
(5.51)
For u(1)2N the situation is more involved. We find

[A2r ] [A2s ] = n[A2t ]
(5.52)
with integers n and t determined as follows: for x = r, s set gx := gcd(x, N/x) and
x := N/(xgx ). Let ax be any integer obeying ax x 1 mod (x/gx ) (the result does
not depend on the particular choice of ax ). Set further G := gcd(N/gr , N/gs ). Then n
and t are given by
n = gcd(gr , gs ),
t=

nN
gcd gr gs (ar r as s )/n, G .
Ggr gs
(5.53)
One can verify that, as expected, [A2N ] is the identity element, and further [A2r ]
[A2 ] = [A2N/r ] as well as [A2r ] [A2r ] = gcd(r, N/r)[A2 lcm(r,N/r) ].
Note that the matrices Z(A2r ) are simple current modular invariants. Products of
simple current modular invariants have been considered in [119].
As will be explained in more detail elsewhere, on orientable world sheets bulk fields of
the CFT determined by A are triples (i, j, ), where Ui and Uj are simple objects and
is an element of the space Hom(A Uj , Ui ) of three-point couplings. However, it turns
out that not all triples (i, j, ) are allowed. Instead has to be local, in the sense defined
below.
Definition 5.5. Let A be a symmetric special Frobenius algebra and X, Y objects in C.
A morphism in the space Hom(A X, Y ) is called local iff PX = . Local morphisms
in Hom(A X, Y ) are denoted by

Homloc (A X, Y ) := Hom(A X, Y ) | PX = .
(5.54)
(The morphism PX is defined in (5.34).)
459
Consider a morphism Hom(A Uj , U ). Via PUj , the morphism PUj

induces a projector p on Hom(A Uj , U ). This follows immediately from (5.36). Denote
ij
by { } an eigenbasis of p, i.e.,
ij
ij
basis of Hom(A Uj , U ) s.t. ij
(5.55)
PUj = with {0, 1}.
ij
ij
Denote by { } Hom(U , A Uj ) a basis dual to { }, that is (using also dominance)

ij
ij
ij
(5.56)
= , id and
ij

= idAUj .
iI
ij
ij
ij
ij
One verifies that PUj = iff PUj = , and that it is zero otherwise.
We also fix the bases { } of Hom(A Uj , Ui ) and { } of Hom(Ui , A Uj ) via
ij
ij := ij
ij
1
and ij := ij
.
ij
(5.57)
ij
One quickly checks that the bases { } and { } are dual to each other, similar to (5.56),
and that
ij local
ij
local.
(5.58)
Unless mentioned otherwise, from here on, whenever summing over a basis of
morphisms in Hom(A Ui , Uj ), Hom(A Ui , Uj ), Hom(Uj , A Ui ) or Hom(Uj , A
Ui ), it is understood that the basis is chosen in the manner described above.
It was stated above that bulk fields are labelled by elements in Homloc (A Uj , Ui ).
This is consistent with the following observation.
Lemma 5.6. The dimension of the subspace Homloc (A Uj , Ui ) of local morphisms in
Hom(A Uj , Ui ) is equal to Zij :

dim Homloc A Uj , Ui = Zij .
(5.59)
Proof. Using dominance and the fact that H(i, (m, );S 2 ) is non-zero only for i = m, we
can rewrite Zij from (5.30) as
(5.60)
Here the sum over runs over the basis introduced in (5.57). The last ribbon graph in
ij
(5.60) can be seen to contain the element PUj , thus the sum can be restricted to a
460
basis of Homloc (A Uj , Ui ). By definition of the basis we can replace PUj by

ij
ij
ij
ij
in the last graph of (5.60). But now and its dual cancel to idUi and the resulting
ribbon graph takes the constant value 1. The sum is thus equal to the number of local basis
elements in Hom(A Uj , Ui ).
5.4. Bulk fields and -induced bimodules
We now further investigate the space of bulk fields. This allow us in particular to show
that our prescription for the modular invariant torus partition function coincides with the
one obtained by different methods in [32], and at the same time give a deeper understanding
of the space Homloc (A X, Y ) of local morphisms. Bimodules over A are a crucial
ingredient in this analysis, so we start with a few comments on the category A CA of AAbimodules. In contrast to the category of left A-modules, A CA is naturally endowed with
the structure of a tensor category, with tensor product MA N defined to be the tensor
product over A. The tensor unit of A CA is A itself. We denote the morphism spaces of A CA
by HomA|A ( , ).
The following prescription defines two tensor functors () from C to A CA . For every
object V of C the induced left module IndA (V ) = (A V , m idV ) can be endowed
with two different structures of a right A-module: in the first case the right action r
Hom(A V A, A) is
r(+) := (m idV ) (idA cV ,A ),
(5.61)
and in the second case,

r() := (m idV ) idA (cA,V )1 .
(5.62)
Since A is associative, both of these right actions commute with the left action of A on
A V . We denote the two induced bimodules obtained this way by () (V ). (In the
introduction we have used the notation (A V ) instead, because it adapts easier to the
general situation, where one has (B V ) with V an object of C and B an arbitrary AAbimodule.) Both functors () act on morphisms f Hom(V , W ) in the same way as the
induction functor of left modules:
() (f ) := idA f Hom(A V , A W ).
(5.63)
()
have first appeared in the theory of subfactors [120],

The two tensor functors
where they play a crucial role (see, e.g., [121125]) and have been termed -induction.
The category-theoretic reformulation presented here was obtained in [13]. For those cases
where A describes the embedding of a subfactor into a factor, it has been shown [32] that
the matrix with entries

(BEK)
:= dim HomA|A () (Uk ), (+) (Ul )
Zk,l
(5.64)
(k, l I) commutes with the matrices that describe the action of the modular group and
hence provides a modular invariant.
We now show that the matrix (5.64) is related to the modular invariant partition function
discussed in Section 5.3 by
(BEK)
= Zl,k
Zk,l
.
(5.65)
461
Along with the construction of general correlators outlined in [22], this shows in particular
that Z (BEK) is realised as the torus partition function of a physical conformal field theory,
a property that does not follow from the subfactor considerations.
To derive the relation (5.65) we make use of Lemma 5.6, in view of which we need
to relate the space Homloc (A X, Y ) of local morphisms to morphisms of -induced
bimodules. This is done as follows.
Proposition 5.7. For any symmetric special Frobenius algebra A in a braided tensor
category the linear map

f : Homloc (A X, Y ) HomA|A (X), + (Y ) ,
(idA ) ( idX )
(5.66)
is an isomorphism.
Its (left and right) inverse is given by g : ( idY ) for HomA|A ( (X),
+
(Y )).
Proof. We first show that f maps into the correct space. That f is a linear map from
Homloc (A X, Y ) to Hom( (X), + (Y )) is obvious. What we still must check is that
f () is in fact even in HomA|A , i.e., a morphism of AA-bimodules; this is done as
follows.
It is a direct consequence of the Frobenius property of A that for every Homloc (A
X, Y ) the image f () is a morphism of left A-modules. To see that it is a morphism of
right A-modules as well, we consider the equalities
(5.67)
In the first equality one uses the Frobenius property of A and the locality property of to
insert a projector PX (as defined in formula (5.34)). The second step summarises a number
of moves that take the A-ribbon around the projector; these employ the Frobenius and
associativity properties of A. The third step uses again that is local, which allows one
462
to leave out the projector, and that A is symmetric Frobenius to take the A-ribbon past the
comultiplication. Notice that on A X the action of A from the right is defined via the
braiding, while on A Y it is defined using the inverse of the braiding.
Next we establish that g maps into the correct space, too. By definition, g is a linear
map from HomA|A ( (X), + (Y )) to Hom(A X, Y ). It remains to show that g is local.
For every HomA|A ( (X), + (Y )) we have the equalities
(5.68)
In the first step the definition of g( ) is inserted, and it is used that A is special.
The second step follows because is a morphism of right A-modules. The third step
employs symmetry of A, and the last one that is a left A-morphism. Comparing the
left and right sides of (5.68), we see that g( ) = g( ) PX , which means that indeed
g( ) Homloc (A X, Y ).
Finally we show that f and g are each others inverses. That g f = id on Homloc (A
X, Y ) is an immediate consequence of the defining property of the counit. To see that
also f (g( )) = for all HomA|A ( (X), + (Y )) one uses that A is Frobenius and
that is in particular a left A-morphism. These properties imply that f is injective and
surjective.
5.5. The extended left and right chiral algebras
Let us recall that the chiral algebra V, which corresponds to a subsector of the
holomorphic fields of a CFT, is the vertex operator algebra whose representation theory
gives rise to the modular category C. The full, maximally extended, chiral algebra of the
CFT can be larger than V. It can also be different for the left and right (holomorphic and
anti-holomorphic) parts of the CFT. Let us denote these maximally extended left and right
chiral algebras by A; and Ar , respectively. Their V-representation content can be read off
the torus partition function Z as

Zj 0 Uj and Ar
Z0i Ui .
A;
(5.69)
=
=
j I
iI
These left and right chiral algebras also turn out to possess a nice interpretation in terms of
the algebra object A. To this end let us introduce the notion of the left and right center of
an algebra [13].
463
Definition 5.8. The left center C; (A) of an algebra A is the maximal subobject of A such
that
m cA,A (; idA ) = m (; idA ),
(5.70)
with ; Hom(C; (A), A) is the embedding morphism of C; (A) into A.

The right center Cr (A) of an algebra A is the maximal subobject of A such that the
equality m cA,A (idA r ) = m (idA r ) holds. In pictures,
(5.71)
As already mentioned, two-dimensional conformal field theories that are even topological are included in our construction by specialising to the tensor category of complex vector
spaces. In this specific context, the idea has been put forward [126128] that the boundary
fields yield a non-commutative Frobenius algebrathe algebra of open string statesand
that its center describes an algebra of bulk fields (closed string states). We will now see
how the situation looks like for general conformal field theories.
An important property [13] of the left and right centers of an algebra A is that they
are invariant under Morita equivalence. This is consistent with the fact that the CFT only
depends on the Morita class of the algebra A, so that physical quantities extracted from the
algebra A must be invariant under Morita equivalence, too. Furthermore, the left and right
chiral algebras are precisely given by the left and right centers of A (see Claim 5 in [13]),
i.e.:
Proposition 5.9. As objects in C,
A;
= C; (A) and Ar
= Cr (A).
(5.72)
The proof of Proposition 5.9 will fill the remainder of this section. Let us start by giving
an alternative definition of the left and right centers:
Definition 5.10. The left center centA,left of an algebra A is the operation that assigns to
any object X of C the subspace of all elements in Hom(X, A) such that the multiplication
becomes commutative (with respect to the braiding cA,A ), i.e.,

centA,left (X) := Hom(X, A) | m cA,A ( idA ) = m ( idA ) . (5.73)
Analogously, the right center centA,right of an algebra A is defined by

centA,right(X) := Hom(X, A) | m cA,A (idA ) = m (idA ) . (5.74)
464
The relation to the previous definition is

C; (A)
dim centA,left (Ui ) Ui ,
=
(5.75)
iI
and similarly for Cr (A). Thus in this alternative formulation the relevant subobject of A
is characterised via the components of its embedding morphism. There is also a close
relationship with the relative center of Atop as introduced in Definition 3.15:
centA (Atop) = centA,left (1) = centA,right(1).
(5.76)
Lemma 5.11. The morphism Hom(Ui , A) is in centA,right(Ui ) iff
(5.77)
An analogous statement holds for the left center when the inverse braiding is used in (5.77).
Proof. To see the relation for the right center, consider the equivalence
(5.78)
465
The first of these equalities is equivalent to the left equality in (3.61); indeed, it can be
obtained by composing id (3.61) with ( m) idA . Further, Lemma 3.17 implies that
this first equality is, in turn, equivalent to the relation (5.77). To get the equivalence asserted
in (5.78), one uses the Frobenius property several times, on both sides of the first equality,
as well as the fact that A is symmetric, i.e., relation (3.48).
For the left center the proof proceeds analogously.
To proceed, we introduce, for all i, j I, the isomorphism Lij : Hom(A Ui , Uj )
Hom(Ui Uj , A) via
(5.79)
Lemma 5.12. For each i I, the map Li0 defines an isomorphism between centA,right(Ui )
and Homloc (A Ui , 1).
Proof. Consider the moves
(5.80)
The first equality is the definition of being local. In the equivalence we composed both
sides with ( ) idUi . The right-hand side of the resulting equation can be rewritten
as sketched above by using that A is symmetric and Frobenius. The resulting equality is
equivalent, by Lemma 5.11 to Li0 () being in the right center.
Lemma 5.13. For each j I, the map L0j defines an isomorphism between
centA,left (Uj , A) and the subspace of local elements in Hom(A 1, Uj ).
466
Proof. This is shown similarly to the previous lemma. The manipulations look as follows:
(5.81)
The left equality is again the definition of being local, while by Lemma 5.11 the right
equality is equivalent to being in the left center.
Proof of Proposition 5.9. According to Lemma 5.6, the dimension of the subspace of
local elements in Hom(A Uj , Ui ) equals Zij . Thus by the Lemmata 5.12 and 5.13 it
follows that

Z0i = dim centA,right(Ui ) and Zj 0 = dim centA,left (Uj ) .
(5.82)
As a consequence, the definition of A; and Ar in Eq. (5.69) does coincide with
Definition 5.8 of C; (A) and Cr (A).
In this section we express the ribbon graph (5.30) giving the torus partition function in
terms of the MooreSeiberg data of the modular category C. For notational simplicity we
concentrate on the situation that both Nij k {0, 1} for all i, j, k I and dim Hom(Uk , A)
{0, 1} for all k I.
467
The ribbon graphs defined by
(5.83)
can be related to each other by inserting bases as described in Sections 3.1 and
3.6, using the compatibility constraint between braiding and twist, and accounting for
dim H(i, j ; S 2 ) = j, . This way we simplify the expression for the ribbon graph (5.30)
to

mbc a acb b 1 =
mbc a acb b R(c b)a 2
Zij =
a,b,cA
a,b,cA kI
a,b,cA kI
k
mbc a acb R(c b)a 3 .
j
(5.84)
Via a fusion and an inverse fusion operation one then arrives at a graph that can be further
simplified by using the definition of the dual coupling. The final result is

(c b j )
k (c b j )
mbc a acb
Ga k
R(c b)a Fk a
.
Zij =
(5.85)
j
a,b,cA
kI
(Recall that the coproduct can also be expressed in terms of the product using (3.83).) Note
that in this formulation, it is not obvious why Zij should be real, let alone a non-negative
468
integer. This illustrates the power of the graphical approach, which readily supplies a proof
of non-negativity and integrality.
The expression (5.85) can now be used to compute the torus partition function
associated to the algebra A2r as defined in (3.84) and (3.85). We also need the Z2N free
boson modular data given in (2.77). Substituting these into formula (5.85), one notes that
the simple fusion rules make three of the four sums disappear, and we end up with

r
xy a
exp 2i
Z[x][y] (A2r ) = [x+y]A
N
2 2N
aA
= x+y,0 mod 2r xy,0 mod 2N/r .
(5.86)
Let us have a look at the two extremal cases. For r = N we have A2r = A2N = [0]. This
leads to
Z[x][y] (A2N ) = x+y,2N .
(5.87)
(Recall that we chose representatives of simple objects such that 0 x, y < 2N .) The
partition function (5.87) is just the charge conjugation modular invariant. On the other
extreme, for r = 1 we deal with A2 = [0] [2] [2N 2] and find
Z[x][y] (A2 ) = x,y ,
(5.88)
which pairs V-representations of the same (minimal) u(1) charge; this is precisely the
T-dual of the partition function (5.87).
More generally, one easily verifies that
Z[x][2Ny] (A2r ) = Z[x][y] (A2N/r ),
(5.89)
which means that the two algebras A2r and A2N/r produce T-dual theories.
It is only now that we can verify that the title of this example indeed deserves its
name. We combine the definition of the modular category for su(2)k in Section 2.5.2 at
level k = 16, the explicit form
(3.90) for the product of A, and the formula (5.85) for
the coefficients Zij in Z = i,j I Zij i j . We can then check numerically that we get
precisely the E7 modular invariant:
)
Z = |0 + 16 |2 + |4 + 12 |2 + |6 + 10 |2 + |8 |2 + 8 (2 + 14
+ (2 + 14 )8 .
(5.90)
Incidentally, one also checks that A of the form

A = (0) (k/2) (k),
(5.91)
i.e., precisely containing the unit, the simple current and its fixed point (see (3.90)), is an
algebra also at the levels k = 4 and k = 8. In the first case it is Morita equivalent to 1, since
469
(recall from Section 3.6.2 that the multiplication on 1 f J is unique)

(0) (2) (4)
= (2) (2),
(5.92)
and accordingly gives rise to the A-series modular invariant, while in the second case one
obtains the D-series modular invariant.
5.7. One-point blocks on the torus
Similarly as for the zero-point blocks on the torus, we may study the behavior of torus
blocks with one A-insertion, i.e., of elements of H(A; T), under an S-transformation. In
analogy with (5.20) we fix our conventions by defining the matrix LA = (LA
i,j ) via
(5.93)
Here X is an arbitrary (not necessarily simple) object, Hom(A X, X), Ui , Uj are
simple objects, denotes a basis vector in Hom(Ui , A Ui ), and runs over the basis in
Hom(Uj , A Uj ). That the ribbon graphs on the two sides are indeed linearly related
and thus that the matrix LA existsfollows from the general result that the TFT provides
us with a projective representation of the mapping class group.
Let us examine some properties of LA .
Lemma 5.14. The matrix LA is invertible, with inverse
LA given by
(5.94)
470
Proof. For X = Uk and = , the left-hand side of (5.93) reads
(5.95)
while the ribbon graph on the right-hand side of (5.93) reduces to
(5.93) becomes

A
i,k , =
LA
i,j Lj,k .

LA
j,k /S0,0 ,
so that
(5.96)
j I
A as its right-inverse. Since LA is a square matrix, this is also a leftThus LA has L

inverse.
Note that the relation (5.21) for the zero-point blocks is recovered as a special case of
A is just the inverse of the S-matrix,
(5.96), by setting A = 1 and = = id. In this case L
1

i.e., Lj,k = S0,0 sj,k with s as defined in (2.22).
A ) and
A ), where
In the next section we will heavily use matrices S A = (S,i
S A = (
Si,
J (corresponding to a simple A-module M ) and i I (corresponding to a simple
object Ui ), and Hom(A Ui , Ui ) is local. We start by defining
(5.97)
Note that the row and column labels of
S A take their values in two rather different index
sets. (Thus, while being similar to the S-matrix for the category of local A-modules that is
used in [87], it is not quite the same matrix. The notion of a local module as defined in [87]
refers to algebras A that are commutative and have trivial twist, neither of which properties
is imposed here.)
471
The matrix S A is defined to implement an S-transformation on the torus one-point

blocks, analogously to LA in (5.93):
(5.98)
Here M is a simple A-module, and the second sum extends over the dual basis {}
of
the morphism space Hom(Ui , A Ui ). When X = M and = is a representation
morphism, then the ribbon graph on the right-hand side of (5.98) looks like (5.97), except
for the fact that in (5.97) is restricted to be local. This constraint does, however, not pose
any problem:
Lemma 5.15. The relation (5.98), with X = M and = , remains true when the sum
on the right-hand side is restricted to local couplings .
Proof. We have the following identities of ribbon graphs in S 3 :
(5.99)
This means that one may always dress the coupling on the right-hand side of (5.98) by the
projector Pi . This, in turn, implies that the sum does not change its value when restricted
to local couplings.
The following two results imply that, when restricted to a basis of local couplings in
Hom(A Ui , Ui ), S A is invertible.
472
Proposition 5.16. The matrix S A is a left-inverse for

S A . That is, for two simple A-modules
%
M , M one has

A A
S,i
Si, % = , % .
(5.100)
iI local
Proof. Consider the defining relation (5.98) for X = M % , where M % is a simple A-module
and = % the representation morphism. By Lemma 5.15, the range of summation on the
right-hand side of (5.98) can be restricted to local , thus resulting in the left-hand side of
(5.100). On the other hand, defining the graphs
(5.101)
the left-hand side of (5.98) becomes

D, % := 1 =
2 =
a,bI ,
aI ,
473

1
1
3 =
4 . (5.102)
dim(Ua )
dim(Ua )
aI ,
Here in the first step bases of the type (4.21) for the morphisms between modules
and their subobjects are inserted. The second step uses dominance and the fact that
dim(H(m; S 2 )) = m,0 for the one-point blocks on the sphere, implying that only the tensor
unit 1 survives in the intermediate channel. In the third step the ribbon graph is shifted
along the S 1 direction and is shown from a slightly different angle.
One can now use relation (4.33), which applies because M and M % are simple
A-modules. It follows that
(5.103)
where in the last equality it is used that by the completeness property (4.22) the remaining
sum over a and simply amounts to a factor of dim(M ).
Proposition 5.17. The matrix
S A has a right-inverse. That is, there exists a matrix
= (,j ), with J , j I and local in Hom(A Uj , Uj ), such that

A

,j = i,j , .
(5.104)
Si,
J
A are each others inverses:

Proof. According to Lemma 5.14 the matrices LA and L

A

(5.105)
LA
i,j Lj,k = i,k , .
j I
Note that the -summation runs over a basis of the full coupling space Hom(A Uj , Uj ).
Let us switch from this basis {}, which consists of eigenvectors of Pj , to the basis (4.25)
j
of Hom(A Uj , Uj ) that is labelled by ( ) (recall also Theorem 4.15). Let D,
denote the (invertible) matrix that implements this change of basis, and define
1
D j , LA
and
K(k, ),k :=
j,k
i,(j, ) :=
K
j
A
L
i,j D, .
(5.106)
are inverse to each other, too:

The matrices K and K

i,(j, ) K(j, ),k = i,k , .
K
j I J ,
(5.107)
474
is
The pictorial representation of K
(5.108)
We now impose the condition that is local. We are then allowed to insert a projector Pi
in the graph (5.108). This leads to the following identities:
(5.109)
In the last picture we can use the orthogonality relation (4.33) of representation functions
(in (4.33), take k = l, take the trace on both sides and sum over = ). As a consequence,
for local we have the equality
i,(j, ) = ,
K
dim(Uj )
dim(M )
A

.
Si,
(5.110)
Using this relation, (5.107) takes the form

j I J ,

dim(Uj ) A
A

Si,
,k = i,k ,
Si, K(j, ),k =
dim(M )
J
(5.111)
475
with
,k :=
dim(Uj )
K(j, ),k .
dim(M )
j I
(5.112)
Thus the matrix defined this way, with restricted to be local, is right-inverse to
SA.
We conclude that
S A has both a left- and a right-inverse. Thus the two inverses coincide,
A
A

i.e., S and S are square matrices and inverse to each other. This provides us with a key
information for the next section: for M , M % simple A-modules, Ui , Uj simple objects
and , local basis vectors in the corresponding coupling spaces, we have

A A
A
A

(5.113)
S,i
S,j
= i,j , .
Si, % = , % and
Si,
iI local
For now let us just note one astonishing consequence of this result.
Theorem 5.18. Let A be a symmetric special Frobenius algebra in the modular tensor
category C. Then

|isom. classes of simple left A-modules| = tr CZ(A) .
(5.114)
Proof. Consider the space of local couplings in Hom(A Uk , Uk ), or equivalently, in
Hom(A Uk , Uk ). By Lemma 5.6, its dimension is equal to Zk,k . Thus a basis B for all

local couplings in kI Hom(A Uk , Uk ) has kI Zk,k elements. Moreover, since S A
is a square matrix, the number of isomorphism classes of simple modules must be the same
as the number of basis vectors in B.
Remark 5.19.
(i) In CFT terms, this result tells us that for the CFT based on A the number of
elementary boundary conditions (corresponding to simple A-modules)
is the same

Z
.
as the number of Ishibashi states, whose total number is
kI k k This is the
completeness condition for boundary conditions [14,29]. The completeness condition
enters in the proof [14] that the annulus coefficients of a CFT furnish a NIM-rep of
the fusion rules. Theorem 5.18 implies that CFTs constructed from algebra objects
fulfill this condition, and hence their annulus coefficients must provide a NIM-rep. In
Theorem 5.20 we will see explicitly that this is indeed the case.
(ii) By combining Theorem 5.18 with Propositions 4.6 and 5.3 we arrive at the following
relation for the category of AB-bimodules:
|isom. classes of simple AB-bimodules|

= tr CZ(A Bop ) = tr CZ(A)CZ(B)t = tr Z(A)Z(B)t ,
(5.115)
where in the last step we used that the charge conjugation matrix commutes with a
modular invariant matrix Z.
476
Recall the comment in the end of Section 4.4 that AB-bimodules describe tensionless
interfaces between the CFTs associated to the algebras A and B. In the special case A = B
Eq. (5.115) reproduces the relation for the number of generalised defect lines found in [34].
5.8. Annulus partition functions
Boundary conditions correspond to modules of a symmetric special Frobenius algebra
A, and elementary boundary conditions are simple A-modules. In this section we work
out the partition function AMN of an annulus with boundary conditions M and N . Let
us for now take M and N to be simple A-modules. The world sheet X for the annulus
amplitude is an annulus. Its complex double
X is a torus T = S 1 S 1 , and the connecting
1
three-manifold MX is a full torus D S , where D is a disk.
According to the prescription in Section 5.1, the first step in the construction of the
partition function is to specify a triangulation of the world sheet X; let us choose the one
given in the picture on the left-hand side of
(5.116)
(Here top and bottom are to be identified, which is indicated by the dashed lines.) Then
we substitute the elements (5.6), (5.7) to convert the triangulation into a ribbon graph;
simplifying the result, we arrive at the graph in the middle of (5.116). Using the Frobenius
property of A and the representation property of M , this graph can in turn be rewritten
as on the right-hand side. It follows that the A-ribbon that runs around the annulus can be
removed. Then the ribbon graph describing the annulus amplitude finally becomes
(5.117)
This graph determines an element AMN in the space H(; T) of zero-point conformal
blocks on the torus. To obtain the annulus coefficients we must expand AMN in a basis of
blocks on the torus, i.e., in terms of characters (5.15):

AkMN |k ; T.
AM N =
477
(5.118)
kI
Just like for the torus partition function, we can isolate the coefficient AkMN by gluing the
basis dual to |k ; T to the boundary of the three-manifold on both sides of the equation.
We are then left with
(5.119)
We now establish that the quantities defined by formula (5.117) indeed possess the
properties that befit a partition function on an annulus.
Theorem 5.20. Let A be a symmetric special Frobenius algebra, M M and N M %
(, % J ) simple A-modules, and Uk (k I) a simple object. Then the numbers AkMN
given by the invariant of the ribbon graph (5.119) have the following properties:
(i) Non-negativity:
AkMN Z0 .
(5.120)
(ii) Uniqueness of the vacuum:

A0MN = M,N .
(5.121)
(iii) Exchange of boundary conditions:

AkMN = AkNM .
(iv) Consistency with the bulk spectrum:

Skp A
A

AkMN =
SN,p
.
S
S0p p,M
(5.122)
(5.123)
pI local
(v) NIM-rep of the fusion rules:

Ai Aj =
Nij k Ak ,
kI
where the Ak are the matrices with entries AkMN .
(5.124)
478
Proof. To show (i) one cuts the three-manifold (5.119) along an S 2 to obtain a ribbon
graph in S 2 [1, 1]. This defines a linear map

), M;
S2 ,
), M;
S 2 H k, (N,
Qk : H k, (N,
(5.125)
from which the annulus coefficients are recovered via

AkMN = trH(k,(N,),
2 ) Qk .
M;S
(5.126)
Next one considers the identities
(5.127)
Here the first step uses the representation property, while the second step results from
repeated application of the associativity and Frobenius property of A. Using also that A
is special, as well as the defining property of the unit, one concludes that the map Qk is a
projector,
Qk Qk = Qk ,
(5.128)
and therefore
trH(k,(N,),
2 ) Z0 .
M;S
(5.129)
(ii): For k = 0, AkMN is given by the graph 1 in picture (5.101). The statement of
(ii) is therefore equivalent to the result (5.103) that was obtained in the course of proving
Proposition 5.16.
479
To see (iii) it is sufficient to rotate the ribbon graph (5.119) according to
(5.130)
The second equality holds because the two three-manifolds with embedded ribbon graphs
are related by an (invertible) orientation preserving homeomorphism and thus have the
same topological invariant. The second ribbon graph in (5.130) equals AkNM , as can be
seen by using that A is symmetric and by inserting a pair k k1 to replace the k-ribbon
by a k-ribbon with reversed orientation of the core.
For the proof of (iv) we use the surgery relation (5.20) with i = k and X = N M,
and with corresponding to an A-ribbon suspended between N and M as in (5.119).
This yields
480
(5.131)
Here the first step is the relation (5.20). In the second step dominance is used to insert
a basis for Hom(A Up , Up ) (in the summation over intermediate objects Um , only Up
survives, as is seen by reading the picture from left to right, which involves a morphism
from 1 to Um Up to 1, implying that m = p). The index runs over the eigenbasis in
Hom(A Up , Up ), and denotes the dual basis. In the last step, dominance is used once
more, and again only a single intermediate channel contributes, this time the tensor unit 1.
Moreover, the sum over can be restricted to only local basis vectors of Hom(AUp , Up ),
by the same reasoning that was used in the proof of Lemma 5.15, i.e., by similar moves as
in (5.99).
The second ribbon graph in the last line of (5.131) is nothing but the quotient
A

Sp,M
/S0,0 , as defined in (5.97). Consider now the result (5.131) for the special case k = 0.
Then according to (5.121) we must get M,N . Using also that S0,p / dim(Up ) = S0,0 , it
follows that the first ribbon graph in the last line describes the matrix that is inverse to
SA .
A
Thus it equals SN,p /S0,0 , with S A as defined in (5.98). In pictures,
(5.132)
This completes the proof of part (iv).
Finally, property (v) is a direct consequence of (iv). Indeed, substituting (5.123) into the
right-hand side of (5.124), we get
AiMR Aj RN =
RJ

A
SR,p
RJ p,

p, q,

Sp,i A
Sq,j A
A

SN,q
Sp,M
S
Sp,0
Sq,0 q,R
q,
A
SN,q
Sp,i
Sq,j A

p,q ,
S
Sp,0
Sq,0 p,M
1
Sp,i Sp,j Sp,k
p, kI
Sp,0
A
SN,p
Sp,k A

S
Sp,0 p,M
Nij k AkMN ,
481
(5.133)
kI
which establishes that we are indeed dealing with a NIM-rep.

Remark 5.21.
(i) For A = 1, i.e., for the charge conjugation modular invariant Zij = i, , one recovers
[51] the original result [73] that in this situation the annulus coefficients coincide with
the fusion rules.
(ii) The matrices Ak commute and are normal (since Akt = Ak , see formula (5.122))
and hence can be simultaneously diagonalised, with eigenvalues Sk,a /S0,a . Requiring
consistency between the open and closed string channel in the annulus amplitude leads
to the condition that the multiplicity m(a) of the set {Sk,a /S0,a | k I} of eigenvalues
must be equal to Za a . This condition has been thoroughly investigated in [21], where
many examples of modular invariants without associated NIM-rep as well as NIM-reps
without associated modular invariant were found.
Since the number of basis elements in the space Homloc (A Up , Up ) of local
couplings is Zpp (see Lemma 5.6), part (iv) of Theorem 5.20 implies that for any
CFT obtained from an algebra object, the annulus coefficients AkMN are consistent
with the modular invariant Zij in the sense of Definition 4 of [21].
(iii) In [35], polynomial equations and trace identities for the (integral) coefficients appearing in the torus, cylinder, Mbius and Klein bottle amplitudes have been presented.
Those identities of [35] that involve Zij and AkMN follow from completeness of the
boundary conditions, and are thus automatically satisfied by CFTs constructed from
algebra objects. For example, one derives the trace identity [35]
AkMM =

M

pI
pI local
A
SM,p
Sk,p
Sk,p A

Sp,M =
S0,p
S0,p
pI local

Sk,p
1 Sk,p
Zpp
=
Sp,i Zij Sp,j
=
Nik Zij . (5.134)
S0,p
S0,p
i,j,pI
i,j,pI
Here we used the facts that S A and

S A are inverse to each other and that Z commutes
with S.
Boundary fields will be labelled by (M, N, k, ), where M, N are A-modules, Uk is a
simple object and an element of HomA (M Uk , N). That this prescription is consistent
with the annulus amplitude AkMN follows from
Proposition 5.22. Let Uk (k I) be a simple object and M, N two (not necessarily simple)
A-modules of a symmetric special Frobenius algebra A. Then

AkMN = dim HomA (M Uk , N) ,
(5.135)
where AkMN is the invariant associated to the ribbon graph (5.119).
482
Before proving the proposition, it is helpful to introduce some additional notation.

N
Define the linear map QN

kM acting on Hom(M Uk , N) via QkM () := , where
denotes the A-averaged morphism as introduced in formula (4.11). By combining parts (i)
and (ii) of Lemma 4.4 it follows that QN
kM is a projector and that
Im QN
kM = HomA (M Uk , N) Hom(M Uk , N).
{kMN }
(5.136)
QN
kM ,
Let us choose an eigenbasis

of
i.e.
kMN
kMN
Hom(M Uk , N ) such that QN
= kMN
kM
with {0, 1}.
(5.137)
kMN
kMN
Further, we fix a basis { } in Hom(N , M Uk ) that is dual to { } in the sense
that

tr kMN kMN = , .
(5.138)
Then we can write
(5.139)
where the first trace is a trace over the vector space Hom(M Uk , N), while the second
trace is a trace in the category theoretic sense of (2.14).
Proof of Proposition 5.22. Consider the two ribbon graphs
(5.140)
The three-manifolds in which these ribbon graphs are embedded are both solid three-balls,
but with opposite orientation, B. In the drawing, the disk bounded by the circle indicates
the boundary S 2 of B. Define elements

), M,
k; S 2 ,
v := Z 1 , , S 2 1 H (N,

1

k; S 2 .
v := S0,0 dim(N ) Z 2 , S 2 , H (N , ), M,
(5.141)
483
That indeed v (v ) = , can be seen as follows:
(5.142)
In the first equality the concatenation of v and v is expressed as a ribbon graph in
S 3 obtained by glueing the two three-balls along their boundary. Recall the convention
described below (5.13), introducing a factor S0,0 for the invariant associated to a ribbon
graph in S 3 .
Having established that the vectors v and v constitute dual bases of H((N , ), M,
), M,
k; S 2 ) , we can rewrite the invariant (5.119) as in (5.126) and
k; S 2 ) and H((N,
express the trace as (we can move the k-ribbon in (5.119) from the left side of the module
ribbons to the right side)
AkMN = trH((N,),
2 ) Qk =
M,k;S
v (Qk v )
(5.143)
To see the last equality, we note that the ribbon graph above can be transformed into the one
shown in figure (5.139) (using that the algebra A is symmetric). Eq. (5.136) now implies
the proposition.
This is the last part of our meta example (as far as the present paper is concerned). We
will illustrate how to compute the invariant associated to the annulus partition function
484
(5.119). This can be done by the following series of transformations:
(5.144)
485
Here the usual calculational devices are used: In the first step bases for the morphisms
involving the simple A-modules M and N are inserted. The second step uses dominance
twice, together with dim(H((i, ), m; S 2 )) = m,i and dim(H(m; S 2 )) = m,0 , which
imply that the sums over intermediate simple objects each reduce to a single term. The
third step consists again in the insertion of a basis, and in the fourth step one substitutes
the definition (4.61) for .
To arrive at our final formula, we also use an inverse fusion move on the last graph
above, together with relation (2.60). The result is
AkMN =

aA i,j I ,
(a j k) i (a a i) i
Fi 0
,
aa
aN, i
i a , j 0 Gj i
M j
(5.145)
and a basis of Hom(Uj , N ). Here the coproduct

where labels a basis of Hom(Ui , M)
aa
0 can also be expressed through the multiplication using (3.83).

Substituting the free boson modular data (2.77), the A2r -algebra (3.84), (3.85) and the
expressions (4.79) for the representation matrices into the general formula (5.145), we
obtain
Mn
A[k]M
=
m
r
[j +km]A [j n]A = n+k,m mod 2r .
N
(5.146)
j I
Note that when the two boundary conditions are equal, then the result no longer depends
on m = n. Thus each of the 2r distinct elementary boundary conditions has the same field
content.
As seen in Section 4.5.2, the algebra object A that gives the E7 modular invariant of
the su(2)16 WZW model has seven isomorphism classes of simple modules. We label
representatives for these classes as
M1 = A,
M3 = IndA (2),
M5 = P ,
M2 = IndA (1),
M4 = IndA (3),
M6 = Q,
M7 = R.
(5.147)
(Also recall from Section 4.5.2 that the latter three are not induced modules.) We can
now numerically evaluate formula (5.145) for the annulus coefficients AkM N . To obtain
the representation matrices we proceed as described in Section 4.5.2 and use formula
(4.75). One then directly verifies that the numbers AkM N are non-negative integers, satisfy
A0M N = M,N , and furnish a NIM-rep of the fusion rules.
486
To make contact with the classification of boundary conditions in [129,130] we also

present the matrix A(1) :
A(1) =
1
1
1
1

1
1
1
1

1 .
1
1
(5.148)
(Rows and columns are ordered according to the labelling (5.147), and zero entries have
been replaced by dots to improve readability.) In agreement with the results of [129,130],
this is indeed the adjacency matrix of the Dynkin diagram of the Lie algebra E7 .
5.10. Defect lines and double fusion algebra
In this section we consider the partition function of a torus with two defect lines
inserteda setup studied in [34]. In particular we recover the property that the coefficients
of such partition functions furnish a NIM-rep of the double fusion algebra, a structure
also seen in [11,123].
As already mentioned at the end of Section 4.4 and in Remark 5.19, in our framework
defect lines are described (and labelled) by AA-bimodules. Consider the situation where
the world sheet is a two-torus T without field insertions, but with two defect lines X, Y
running parallel and with opposite orientation. The setup is shown in the following picture,
together with the triangulation we will use (compare also (5.23)):
(5.149)
X|Y
Proceeding similarly to Section 5.3 we can express the coefficients Zkl

partition function
Z X|Y =

k,lI
X|Y
Zkl |k ; T |l ; T
in the twisted
(5.150)
487
as the invariant of a ribbon graph. We obtain
(5.151)
Before proceeding to prove some properties of these numbers it is useful to slightly change
the notation for annulus coefficients, so as to make the dependence on the algebra explicit.
Thus for a symmetric special Frobenius algebra B and for left B-modules M, N , let
A(B)kM N denote the number defined in (5.119), but with all algebra ribbons labelled by B.
Theorem 5.23. Let A be a symmetric special Frobenius algebra, let Uk , Ul be simple
= A(1) A. The
objects and X, Y be AA-bimodules. Abbreviate B = A A(1) and B
X|Y
numbers Zkl have the following properties:
X|Y
Z0 .
(5.152)
Zkl
A|A
= Z(A)kl .
(5.153)
X|Y
Zk l
Y |X
= Zkl .
(5.154)
(i)
Zkl
(ii)
(iii)
(iv)
(v)
X|Y
X|Y
f (Y ) .
Zk0 = A(B)kf (X) f (Y ) and Z0l = A(B)
l f (X)
X|R R|Y
X|Y
Zk0 Z0l
or, as matrix equation, Zkl = Zk0 Z0l .
Zkl =
(5.155)
(5.156)
(vi)
(vii)
As matrix equations, [Zl0 , Zk0 ] = [Z0l , Z0k ] = [Z0l , Zk0 ] = 0.

As matrix equation, Zij Zkl =
Nikr Njsl Zrs .
(5.157)
(5.158)
r,sI
In (ii), Z(A)kl denotes the coefficients of the untwisted torus partition function (5.30). In
X|Y
(v)(vii) Zkl is understood as a matrix with entries (Zkl )X,Y = Zkl . The sum in (v) is
over (representatives of isomorphism classes of ) simple AA-bimodules R. The notation
A(n) was defined in (3.65). In (iv), f and f are the isomorphisms defined in (4.19) taking
AA-bimodules to left A A(1) - and A(1) A-modules, respectively.
Proof. Property (i) follows along the same lines as the proof of Theorem 5.1(ii). That is,
the invariant (5.151) can be rewritten as the trace of a projector. The projector is again
obtained by cutting the three-manifold in (5.151) along a horizontal S 2 ; the projector
property follows from the representation property of the bimodules X, Y and the properties
of A.
488
To see (ii), note that the ribbon graph resulting from (5.151) when replacing X and Y by A
is almost identical to the graph (5.30), except for an additional A-ribbon running vertically.
The latter can be removed by transformations similar to (5.116) with M, N set to A (that
the geometry in that situation is actually a cylinder, rather than a torus, does not play a role
in the calculation).
The proof of (iii) uses the same argument as the proof of Theorem 5.20(iii): the ribbon
X|Y
Y |X
graph for Zk l is that of Zkl turned upside down (this is a rotation, preserving the
and l
orientation of the three-manifold) and with the k- and l-ribbons replaced by kribbons with opposite orientation of their cores.
To show (iv) we draw the ribbon graphs for the annulus coefficients appearing in
(5.155), inserting the relation (4.19) between AA-bimodules and A A(1) - (respectively
left A(1) A-) modules and the definition of the algebras A(1) . The resulting graphs are:
f (Y )
A(A A(1) )k f (X)
f(Y )
A(A(1) A)l f(X)
(5.159)
(Here all representation morphisms are those of AA-bimodules, and all comultiplications
f (Y )
, the ribbon graph
are given by the coproduct of A.) In the case of A(A A(1) )f (X)
f(Y )
, the required
is obviously equal to the graph (5.151) with l = 0. For A(A(1) A)f(X)
moves are slightly more complicated; they are again best visualised by using actual ribbons.
The main step is to recall that the horizontal direction is an S 2 , which allows us to
move one of the horizontal A-ribbons around the S 2 in such a way that in the pictorial
representation the A-ribbons now seem to wrap around the l-ribbon. Once this step is
X|Y
performed, equality with Z0l is easily established.
Proof of (v): Let B = A A(1) . Given a simple object Uk and a left B-module
M, the tensor product M Uk is again a left B-module (compare the discussion of
489
module categories in Section 4.1). By Proposition 4.13 the category of left B-modules
is semisimple, and hence we can decompose M Uk in terms of simple left B-modules as

M Uk
(5.160)
M Uk , Si B Si ,
=
iIB
as an isomorphism of left B-modules. Here the sum is over representatives Si of

isomorphism classes of simple B-modules. The multiplicities M Uk , SB have been
computed in Proposition 5.22 to be given by annulus coefficients:
M Uk , SB = A(B)kM S .
(5.161)
We can apply this relation to (5.151) by understanding the AA-bimodule Y as a left Bmodule f (Y ). This results in
(5.162)
The first step makes use of relations (5.160) and (5.161); the sum is over representatives of
isomorphism classes of simple AA-bimodules, which are taken to simple left B-modules
via f . The second step inserts definition (5.151) for the ribbon graph and the previous
result (5.155).
That the first two commutators in part (vi) vanish follows directly from (iv) together
with the NIM-rep property (5.124). The last equality is obtained by combining (iii) and
(v). Indeed, by (v) we have
X|R R|Y
X|Y
Zkl =
(5.163)
Zk0 Z0l
R
as well as, using (iii) and 0 = 0,

Y |R R|X X|R R|Y
Y |X
Zk0
Z0l Zk0 .
Zk l =
Z0l =
R
(5.164)
Employing (iii) once more we see that (5.163) and (5.164) are equal, thereby establishing
(vi).
Finally, (vii) follows by a short calculation from (iv), (v) and (vi). With B = A A(1)
= A(1) A we have
and B
(Zij Zkl )X|Y = (Zi0 Z0j Zk0 Z0l )X|Y = (Zi0 Zk0 Z0j Z0l )X|Y
490
X|R
R|S
S|T
T |Y
Zi0 Zk0 Z0j Z0l
R,S,T
f (T ) A(B)
f (Y )
A(B)i f (X) f (R)A(B)k f (R) f (S) A(B)
j f (S)
l f (T )
R,S,T

r,sI
r,sI
f (Y )
Nikr A(B)r f (X) f (S) Njsl A(B)
s f (S)
Nikr Njsl
X|S
S|Y
Zr0 Z0s .
(5.165)
In the next-to-last step it is used that the annulus coefficients furnish a NIM-rep, see
Theorem 5.20.
Remark 5.24.
(i) In [34], the double NIM-rep property of the twisted partition functions was proven
under the assumption that there is a complete set of defect lines. Now according to
formula (5.115), the number of (elementary) defect lines present in a CFT constructed
from an algebra object A is given by tr[Z(A)Z(A)t]. This is precisely the number
needed for completeness in [34]; thus their arguments apply, in agreement with point
(vii) of the theorem.
X|Y
(ii) The properties of Zkl derived above can already be found explicitly or implicitly
in [16,34], with the exception of (iv). This point, together with (v), has a curious
interpretation: the twisted torus partition functions of the CFT associated to an algebra
A can be expressed in terms of the annulus coefficients of (in general) different
CFTs possessing the same chiral datathe full CFTs associated to the tensor product
algebras A A(1) and A(1) A.
That there exists a relation between defect lines in a CFT and boundary conditions
in a product CFT bears some similarity with the folding trick mentioned at the
end of Section 4.4. However, the folding trick uses boundary states in a CFT of
twice the central charge of the CFT whose defects are described, whereas the annulus
coefficients in point (iv) are those of a CFT with the same central charge. The physical
interpretation of this observation remains to be clarified.
(iii) The structure of a double NIM-rep can in fact be generalised further. So far we have
established that annulus coefficients associate a single NIM-rep to left A-modules and
that defect lines associate a double NIM-rep to left A1 A2 -modules, with suitable
algebras A1 and A2 . We may view the corresponding coefficients schematically as
follows:
(5.166)
491
What is shown are horizontal sections of ribbon graphs in S 2 S 1 . More specifically,

the figures are cross sections of the graphs (5.119) and (5.151), respectively; vertical
ribbons for simple objects and left modules are indicated by filled circles, while the
lines symbolise A-ribbons.
X|Y
Analogously, for any pair of left (A1 An )-modules X, Y , the numbers Zi1 ,...,in
defined by
(5.167)
furnish an n-fold NIM-rep of the fusion rules, i.e.,

X|R
R|Y
X|Y
Zi1 ,...,in Zj1 ,...,jn =
Ni1kj11 Ninkjnn Zk1 ,...,kn .
R
(5.168)
k1 ,...,kn I
Applications of this structure in string theory remain to be clarified. It is, however,

tempting to conjecture that they appear in situations where the world sheet is no longer
a smooth manifold. Such world sheets play an important role in the description of string
junctions and, more generally, in string networks (see, e.g., [131]).
6. Epilogue: non-commutative geometry in tensor categories

Our results relate rational conformal field theory to the theory of non-commutative
algebras and their representations in modular tensor categories. Now a convenient way
to think about non-commutative algebras in the tensor category of vector spaces is noncommutative geometry. It is therefore tempting to relate conformal field theory to a version
of non-commutative geometry over a modular tensor category.
Consider a compact topological manifold M with measure . On the commutative
of continuous functions a counit : A C is provided by the
algebra A = C 0 (M)
$
integral, (f ) = d f . This way, A becomes a symmetric Frobenius algebra (though
not necessarily a special one). In this sense, the Frobenius algebra A in C that we used to
describe conformal field theory can be regarded as non-commutative measure theory in the
tensor category C.
The algebra A itself provides all information needed to analyze correlation functions on
arbitrary closed oriented world sheets. The study of boundary conditions and defect lines
requires in addition the study of A-modules and AA-bimodules. They should be thought
of as non-commutative vector bundles. As will be discussed in a future publication, our
formalism can be extended to unorientable world sheets as well. This requires the choice
492
Table 2
World sheet
Algebraic data
NC geometry over C
Closed oriented
Boundaries
Defect lines
Orientifolds
...
symm. special Frobenius object A

A-modules
AA-bimodules
generalised -structure on A
...
NC measure theory
NC vector bundles
real NC geometry
...
of a conjugation on A, i.e., the category-theoretic analogue of a conjugation or, in other

words, of a *-structure. This can be interpreted as the choice of a real structure in the
non-commutative geometry over the tensor category C.
We summarise these ideas in Table 2.
Obviously, in all three columns more entries must be added to arrive at a complete
picture. For example, cyclic cohomology, the non-commutative analogue of de Rham
cohomology, should play an important role for describing deformations of conformal field
theories. Indeed, in the case of complex algebras it is known that the second degree of
cyclic cohomology controls deformations of associative algebras with bilinear invariant
form (see, e.g., [118]). In any case, already on the basis of the presently available evidence
it is reasonable to expect that viewing conformal field theory as non-commutative geometry
over a tensor category can serve as a fruitful guiding principle in the future.
The emergence of non-commutative structures in conformal field theory does not come
as a surprise. There exist families of CFTs which are known to yield, in a certain limit,
non-commutative field theories [132]. The present results are, however, much closer to a
different proposal [133] according to which every single conformal field theory gives rise
to a non-commutative geometry. These ideas have been formulated within the category of
vector spaces. Our results indicate that we can realise them by lifting all relevant structures,
algebraic and geometric, to more general tensor categories.
Acknowledgements
We are grateful to Jrg Frhlich and Albert Schwarz for discussions and encouragement,
and to Kevin Graham, Viktor Ostrik and Bodo Pareigis for correspondence. Jrgen Fuchs
thanks the LPTHE for hospitality. Ingo Runkel is supported by the EU grant HPMF-CT2000-00747.
References
[1] G. Mack, V. Schomerus, Quasi Hopf quantum symmetry in quantum theory, Nucl. Phys. B 370 (1992) 185.
[2] P. Vecsernys, On the quantum symmetry of the chiral Ising model, Nucl. Phys. B 415 (1994) 557.
[3] K. Szlachnyi, Chiral decomposition as a source of quantum symmetry in the Ising model, Rev. Math.
Phys. 6 (1994) 649.
[4] J. Fuchs, The quantum symmetry of rational field theories, Theor. Math. Phys. 98 (1994) 266.
[5] J. Fuchs, A.Ch. Ganchev, P. Vecsernys, Rational Hopf algebras: polynomial equations, gauge fixing, and
low-dimensional examples, Int. J. Mod. Phys. A 10 (1995) 3431, hep-th/9402153.
493
[6] K. Szlachnyi, Weak Hopf algebras, in: S. Doplicher (Ed.), Operator Theory and Quantum Field Theory,
International Press, Cambridge, 1997, p. 46.
[7] G. Bhm, F. Nill, K. Szlachnyi, Weak Hopf algebras I: integral theory and C -structure, J. Algebra 221
(1999) 385.
[8] G. Bhm, K. Szlachnyi, Weak Hopf algebras II: representation theory, dimensions and the Markov trace,
J. Algebra 233 (2000) 156.
[9] P.I. Etingof, D. Nikshych, V. Ostrik, On fusion categories, math.QA/0203060.
[10] A. Ocneanu, Paths on Coxeter diagrams: from Platonic solids and singularities to minimal models and
subfactors, (notes by S. Goto), in: R. Bhat, G.A. Elliott, P.A. Fillmore (Eds.), Lectures on Operator Theory,
Fields Institute Monographs, Vol. 13, American Mathematical Society, Providence, 1999, p. 243.
[11] A. Ocneanu, The classification of subgroups of quantum SU (N ), http://www.cpt.univ-mrs.fr/~coque/
Bariloche2000/Bariloche2000/Ocneanu.ps.
[12] V.G. Turaev, Quantum Invariants of Knots and 3-Manifolds, de Gruyter, New York, 1994.
[13] V. Ostrik, Module categories, weak Hopf algebras and modular invariants, math.QA/0111139.
[14] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Boundary conditions in rational conformal field
theories, Nucl. Phys. B 579 (2000) 707.
[15] V.B. Petkova, J.-B. Zuber, The many faces of Ocneanu cells, Nucl. Phys. B 603 (2001) 449, hepth/0101151.
[16] V.B. Petkova, J.-B. Zuber, Conformal boundary conditions and what they teach us, in: Z. Horvth,
L. Palla (Eds.), Non-Perturbative QFT Methods and their Applications, World Scientific, 2001, p. 1, hepth/0103007.
[17] V.B. Petkova, J.-B. Zuber, Quantum field theories, graphs and quantum algebras, hep-th/0108236.
[18] D. Verstegen, New exceptional modular invariant partition functions for simple KacMoody algebras, Nucl.
Phys. B 346 (1990) 349.
[19] A.N. Schellekens, S. Yankielowicz, Field identification fixed points in the coset construction, Nucl. Phys.
B 334 (1990) 67, Appendix B.
[20] J. Fuchs, A.N. Schellekens, C. Schweigert, Galois modular invariants of WZW models, Nucl. Phys. B 437
(1995) 667.
[21] T. Gannon, Boundary conformal field theory and fusion ring representations, Nucl. Phys. B 627 (2002)
506.
[22] J. Fuchs, I. Runkel, C. Schweigert, Conformal correlation functions, Frobenius algebras and triangulations,
Nucl. Phys. B 624 (2002) 452.
[23] C.W. Curtis, I. Reiner, Representation Theory of Finite Groups and Associative Algebras, Wiley, New
York, 1962.
[24] B. Farb, R.K. Dennis, Noncommutative Algebra, Springer, New York, 1993.
[25] S. MacLane, Categories for the Working Mathematician, Springer, New York, 1971.
[26] C. Kassel, Quantum Groups, Springer, New York, 1995.
[27] A.N. Schellekens, S. Yankielowicz, Simple currents, modular invariants, and fixed points, Int. J. Mod. Phys.
A 5 (1990) 2903.
[28] P.E. Dorey, Partition functions, intertwiners and the Coxeter element, Int. J. Mod. Phys. A 8 (1993) 193,
hep-th/9205040.
[29] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Completeness conditions for boundary operators in 2D conformal
field theory, Phys. Lett. B 381 (1996) 97.
[30] A. Sagnotti, Ya.S. Stanev, Open descendants in conformal field theory, Fortschr. Phys. 44 (1996) 585.
[31] J. Bckenhauer, D.E. Evans, Modular invariants graphs, and -induction for nets of subfactors III,
Commun. Math. Phys. 205 (1999) 183.
[32] J. Bckenhauer, D.E. Evans, Y. Kawahigashi, Chiral structure of modular invariants for subfactors,
[33] V.B. Petkova, J.-B. Zuber, BCFT: from the boundary to the bulk, Proc. HEP-tmr2000 (2000) 038.
[34] V.B. Petkova, J.-B. Zuber, Generalized twisted partition functions, Phys. Lett. B 504 (2001) 157.
[35] A.N. Schellekens, Ya.S. Stanev, Trace formulas for annuli, J. High Energy Phys. 0112 (2001) 012.
[36] J. Fuchs, C. Schweigert, Category theory for conformal boundary conditions, math.CT/0106050, to appear
in Fields Institute Commun.
494
[37] J. Fuchs, I. Runkel, C. Schweigert, Conformal boundary conditions and 3D topological field theory, in:
A. Cappelli, G. Mussardo (Eds.), Statistical Field Theories, Kluwer, 2002, p. 185, hep-th/0110158.
[38] C. Kassel, M. Rosso, V. Turaev, Quantum Groups and Knot Invariants, Soc. Math. de France, Paris, 1997.
[39] B. Bakalov, A.A. Kirillov, Lectures on Tensor Categories and Modular Functors, American Mathematical
Society, Providence, 2000, http://www.math.sunysb.edu/~kirillov/tensor/tensor.html.
[40] I.B. Frenkel, J. Lepowsky, A. Meurman, Vertex Operator Algebras and the Monster, Academic Press, New
York, 1988.
[41] I.B. Frenkel, Y. Huang, J. Lepowsky, On axiomatic approaches to vertex operator algebras and modules,
Memoirs Am. Math. Soc. 104 (1993) 1.
[42] V.G. Kac, Vertex Algebras for Beginners, American Mathematical Society, Providence, 1996.
[43] Y.-Z. Huang, J. Lepowsky, Intertwining operator algebras and vertex tensor categories for affine Lie
algebras, Duke Math. J. 99 (1999) 113.
[44] D. Kazhdan, G. Lusztig, Tensor categories arising from affine Lie algebras I, J. Am. Math. Soc. 6 (1993)
905;
D. Kazhdan, G. Lusztig, Tensor categories arising from affine Lie algebras II, J. Am. Math. Soc. 6 (1993)
949;
D. Kazhdan, G. Lusztig, Tensor categories arising from affine Lie algebras III, J. Am. Math. Soc. 7 (1993)
335;
D. Kazhdan, G. Lusztig, Tensor categories arising from affine Lie algebras IV, J. Am. Math. Soc. 7 (1993)
9383.
[45] M. Finkelberg, An equivalence of fusion categories, Geom. Funct. Anal. 6 (1996) 249.
[46] G. Moore, N. Seiberg, Classical and quantum conformal field theory, Commun. Math. Phys. 123 (1989)
177.
[47] G. Moore, N. Seiberg, Lectures on RCFT, in: H.C. Lee (Ed.), Physics, Geometry, and Topology, Plenum,
New York, 1990, p. 263.
[48] J.W. Barrett, B.W. Westbury, Invariants of piecewise-linear 3-manifolds, Trans. Am. Math. Soc. 348 (1996)
3997.
[49] P. Bantay, The FrobeniusSchur indicator in conformal field theory, Phys. Lett. B 394 (1997) 87.
[50] J. Fuchs, A.Ch. Ganchev, K. Szlachnyi, P. Vecsernys, S4 -symmetry of 6j -symbols and FrobeniusSchur
indicators in rigid monoidal C -categories, J. Math. Phys. 40 (1999) 408.
[51] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, Correlation functions and boundary conditions in RCFT
and three-dimensional topology, Compos. Math. 131 (2002) 189.
[52] J. Bernstein, Tensor categories, q-alg/9501032.
[53] J. Frhlich, Statistics of fields, the YangBaxter equation and the theory of knots and links, in: G. t Hooft,
A. Jaffe, G. Mack, P.K. Mitter, R. Stora (Eds.), Nonperturbative Quantum Field Theory, Plenum, New
York, 1988, p. 71.
[54] B. Schroer, New methods and results in conformal QFT2 and the string idea, in: G. t Hooft, A. Jaffe,
G. Mack, P.K. Mitter, R. Stora (Eds.), Nonperturbative Quantum Field Theory, Plenum, New York, 1988,
p. 529.
[55] K. Fredenhagen, K.-H. Rehren, B. Schroer, Superselection sectors with braid group statistics and exchange
algebras I: general theory, Commun. Math. Phys. 125 (1989) 201.
[56] R. Longo, Index of subfactors and statistics of quantum fields I, Commun. Math. Phys. 126 (1989) 217.
[57] I. Bobtcheva, Numerical generation of semisimple tortile categories coming from quantum groups, PhD
thesis, Virginia Polytechnic Institute, 1996.
[58] I. Bobtcheva, F. Quinn, Numerical presentations of tortile categories, Contemp. Math. 233 (1999) 45.
[59] E. Witten, Quantum field theory and the Jones polynomial, Commun. Math. Phys. 121 (1989) 351.
[60] J. Frhlich, C. King, Two-dimensional conformal field theory and three-dimensional topology, Int. J. Mod.
Phys. A 4 (1989) 5321.
[61] A. Joyal, R. Street, The geometry of tensor calculus I, Adv. Math. 88 (1991) 55.
[62] D. Birmingham, M. Blau, M. Rakowski, G. Thompson, Topological field theory, Phys. Rep. 209 (1991)
129.
[63] K. Walker, On Wittens 3-manifold invariants, http://www.xmission.com/~kwalker/math/, preprint, 1991.
[64] B. Durhuus, H.P. Jakobsen, R. Nest, Topological quantum field theories from generalized 6j -symbols, Rev.
Math. Phys. 5 (1993) 1.
495
[65] M. Karowski, R. Schrader, State sum invariants of three-manifolds: a combinatorial approach to topological
quantum field theories, J. Geom. Phys. 11 (1993) 181.
[66] S.-w. Chung, M. Fukuma, A. Shapere, Structure of topological lattice field theories in three dimensions,
Int. J. Mod. Phys. A 9 (1994) 1305.
[67] V. Kodiyalam, V.S. Sunder, Topological Quantum Field Theories from Subfactors, Chapman & Hall, Boca
Raton, 2001.
[68] J. Frhlich, T. Kerler, Quantum Groups, Quantum Categories and Quantum Field Theory, in: Springer
Lecture Notes in Mathematics, Vol. 1542, Springer, Berlin, 1993.
[69] I. Brunner, V. Schomerus, On superpotentials for D-branes in Gepner models, J. High Energy Phys. 0010
(2000) 016.
[70] A.N. Kirillov, N.Yu. Reshetikhin, Representations of the algebra Uq (sl(2)), q-orthogonal polynomials
and invariants of links, in: V.G. Kac (Ed.), Infinite-dimensional Lie Algebras and Groups, World Scientific,
Singapore, 1989, p. 285.
[71] B.-y. Hou, K.-j. Shi, P. Wang, R.-H. Yue, The crossing matrices of WZW SU(2) model and minimal models
with the quantum 6j symbols, Nucl. Phys. B 345 (1990) 659.
[72] B. Pareigis, Quantum groups and noncommutative geometry, available at http://www.mathematik.
uni-muenchen.de/~pareigis/pa_schft.html.
[73] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[74] D.C. Lewellen, Sewing constraints for conformal field theories on surfaces with boundaries, Nucl. Phys.
B 372 (1992) 654.
[75] I. Runkel, Boundary structure constants for the A-series Virasoro minimal models, Nucl. Phys. B 549
(1999) 563.
[76] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, The geometry of WZW branes, J. Geom. Phys. 34 (2000)
162.
[77] N.P. Strickland, K(N )-local duality for finite groups and groupoids, Topology 39 (2000) 733.
[78] M. Mger, From subfactors to categories and topology I: Frobenius algebras in and Morita equivalence of
tensor categories, math.CT/0111204.
[79] B. Pareigis, Non-additive ring and module theory I: general theory of monoids, Publ. Math. Debrecen 24
(1977) 189.
[80] S. Majid, Braided groups, J. Pure Appl. Alg. 86 (1993) 187.
[81] V.V. Lyubashenko, Tangles and Hopf algebras in braided categories, J. Pure Appl. Alg. 98 (1995) 245.
[82] B. Pareigis, On Lie algebras in braided categories, in: Quantum Groups and Quantum Spaces, in: Banach
Center Publications, Vol. 40, Warsaw, 1998, p. 139.
[83] T. Kerler, V.V. Lyubashenko, Non-Semisimple Topological Quantum Field Theories for 3-Manifolds with
Corners, in: Springer Lecture Notes in Mathematics, Vol. 1765, Springer, New York, 2001.
[84] B. Pareigis, Non-Additive Ring and Module Theory IV: the Brauer Group of a Symmetric Monoidal
Category, in: Springer Lecture Notes in Mathematics, Vol. 549, Springer, New York, 1976.
[85] B. Pareigis, Non-additive ring and module theory III: Morita theorems, Publ. Math. Debrecen 25 (1978)
177.
[86] A. Wassermann, Quantum subgroups and vertex algebras, available at http://www.msri.org/publications/
ln/msri/2000/subfactors/wassermann/1/.
[87] A.A. Kirillov, V. Ostrik, On q-analog of McKay correspondence and ADE classification of
sl(2) conformal
field theories, math.QA/0101219.
[88] C. Bachas, P.M.S. Petropoulos, Topological models on the lattice and a remark on string theory cloning,
[89] M. Fukuma, S. Hosono, H. Kawai, Lattice topological field theory in two dimensions, Commun. Math.
Phys. 161 (1994) 157.
[90] B. Durhuus, T. Jonsson, Classification and construction of unitary topological field theories in two
dimensions, J. Math. Phys. 35 (1994) 5306.
[91] L. Abrams, Two-dimensional topological quantum field theories and Frobenius algebras, J. Knot Theory
Ramif. 5 (1996) 569.
[92] V. Karimipour, A. Mostafazadeh, Lattice topological field theory on non-orientable surfaces, J. Math.
Phys. 38 (1997) 49.
496
[93] A.N. Schellekens, S. Yankielowicz, Extended chiral algebras and modular invariant partition functions,
Nucl. Phys. B 327 (1989) 673.
[94] K. Intriligator, Bonus symmetry in conformal field theory, Nucl. Phys. B 332 (1990) 541.
[95] A.N. Schellekens, Introduction to conformal field theory, Fortschr. Phys. 44 (1996) 605.
[96] A.N. Schellekens, S. Yankielowicz, Modular invariants from simple currents: an explicit proof, Phys. Lett.
B 227 (1989) 387.
[97] J. Fuchs, I. Runkel, C. Schweigert, A reason for fusion rules to be even, J. Phys. A 35 (2002) L255.
[98] A.A. Kirillov, Modular categories and orbifold models, math.QA/0104242.
[99] A. Bruguires, Catgories prmodulaires, modularisations et invariants des varits de dimension 3, Math.
Ann. 316 (2000) 215.
[100] M. Mger, Galois theory for braided tensor categories and the modular closure, Adv. Math. 150 (2000)
151.
[101] J. Fuchs, C. Schweigert, Symmetry breaking boundaries I: general theory, Nucl. Phys. B 558 (1999) 419.
[102] L. Birke, J. Fuchs, C. Schweigert, Symmetry breaking boundary conditions and WZW orbifolds, Adv.
Theor. Math. Phys. 3 (1999) 671.
[103] J. Fuchs, C. Schweigert, Symmetry breaking boundaries II: more structures, examples, Nucl. Phys. B 568
(2000) 543.
[104] L. Crane, I. Frenkel, Four-dimensional topological field theories, Hopf categories, and the canonical bases,
J. Math. Phys. 35 (1994) 5136.
[105] J. Fuchs, A.N. Schellekens, C. Schweigert, A matrix S for all simple current extensions, Nucl. Phys. B 473
(1996) 323.
[106] P. Bantay, The untwisted stabilizer in simple current extensions, Phys. Lett. B 396 (1997) 183.
[107] C. Angelantonj, A. Sagnotti, Open strings, hep-th/0204089.
[108] R. Coquereaux, G. Schieber, Twisted partition functions for ADE boundary conformal field theories and
Ocneanu algebras of quantum symmetries, J. Geom. Phys. 42 (2002) 216, hep-th/0107001.
[109] C. Bachas, J. de Boer, R. Dijkgraaf, H. Ooguri, Permeable conformal walls and holography, J. High Energy
Phys. 0206 (2002) 027, hep-th/0111210.
[110] T. Quella, V. Schomerus, Symmetry breaking boundary states and defect lines, J. High Energy Phys. 0206
(2002) 028, hep-th/0203161.
[111] M. Oshikawa, I. Affleck, Boundary conformal field theory approach to the critical two-dimensional Ising
model with a defect line, Nucl. Phys. B 495 (1997) 533.
[112] A. LeClair, A.W.W. Ludwig, Minimal models with integrable local defects, Nucl. Phys. B 549 (1999) 546.
[113] G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, Conformal boundary conditions and three-dimensional
topological field theory, Phys. Rev. Lett. 84 (2000) 1659.
[114] G. Moore, N. Seiberg, Taming the conformal zoo, Phys. Lett. B 220 (1989) 422.
[115] P. Horava, ChernSimons gauge theory on orbifolds: open strings from three dimensions, J. Geom. Phys. 21
(1996) 1.
[116] P.C. Ferreira, I.I. Kogan, Open and unoriented strings from topological membrane I, prolegomena, J. High
Energy Phys. 0106 (2001) 056.
[117] J. Frhlich, B. Pedrini, C. Schweigert, J. Walcher, Universality in quantum Hall systems: coset construction
of incompressible states, J. Stat. Phys. 103 (2001) 527.
[118] M. Penkava, A.S. Schwarz, A algebras and the cohomology of moduli spaces, Trans. Am. Math. Soc. 169
(1995) 91.
[119] M. Kreuzer, A.N. Schellekens, Simple currents versus orbifolds with discrete torsiona complete
classification, Nucl. Phys. B 411 (1994) 97.
[120] R. Longo, K.-H. Rehren, Nets of subfactors, Rev. Math. Phys. 7 (1995) 567.
[121] F. Xu, New braided endomorphisms from conformal inclusions, Commun. Math. Phys. 192 (1998) 349.
[122] J. Bckenhauer, D.E. Evans, Modular invariants, graphs, and -induction for nets of subfactors, Commun.
Math. Phys. 197 (1998) 361.
[123] J. Bckenhauer, D.E. Evans, Y. Kawahigashi, On -induction, chiral generators and modular invariants for
subfactors, Commun. Math. Phys. 208 (1999) 429.
[124] J. Bckenhauer, D.E. Evans, Modular invariants from subfactors, math.OA/0006114.
[125] J. Bckenhauer, D.E. Evans, Y. Kawahigashi, LongoRehren subfactors arising from -induction, Publ.
RIMS 37 (2001) 1.
497
[126] G. Moore, Some comments on branes, G-flux, and K-theory, Int. J. Mod. Phys. A 16 (2001) 936.
[127] G. Moore, D-branes, RR-fields and K-theory IIV, available at http://online.itp.ucsb.edu/online/mp01/
moore1.
[128] C.I. Lazaroiu, On the structure of open-closed topological field theory in two dimensions, Nucl. Phys.
B 603 (2001) 497.
[129] R.E. Behrend, P.A. Pearce, J.-B. Zuber, Integrable boundaries, conformal boundary conditions and ADE
fusion rules, J. Phys. A 31 (1998) L763.
[130] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, On the classification of bulk and boundary conformal
field theories, Phys. Lett. B 444 (1998) 163.
[131] A. Sen, String network, J. High Energy Phys. 9803 (1998) 005.
[132] N. Seiberg, E. Witten, String theory and noncommutative geometry, J. High Energy Phys. 9909 (1999)
032.
[133] J. Frhlich, K. Gawedzki, Conformal field theory and geometry of strings, in: J. Feldman, et al. (Eds.),
Mathematical Quantum Theory I: Field Theory and Many Body Theory, American Mathematical Society,
Providence, 1994, p. 57.

Disc instantons in linear sigma models

Suresh Govindarajan a , T. Jayaraman b , Tapobrata Sarkar c
a Department of Physics, Indian Institute of Technology, Madras, Chennai 600 036, India
b The Institute of Mathematical Sciences, Chennai 600 113, India
c Department of Theoretical Physics, Tata Institute of Fundamental Research, Homi Bhabha Road,
Mumbai 400 005, India

Received 26 March 2002; accepted 4 October 2002
Abstract
We construct a linear sigma model for open-strings ending on special Lagrangian cycles of a
CalabiYau manifold. We illustrate the construction for the cases considered by Aganagic and Vafa
(AV). This leads naturally to concrete models for the moduli space of open-string instantons. These
instanton moduli spaces can be seen to be intimately related to certain auxiliary boundary toric
varieties. By considering the relevant GelfandKapranovZelevinsky (GKZ) differential equations
of the boundary toric variety, we obtain the contributions to the worldvolume superpotential on the
A-branes from open-string instantons. By using an ansatz due to Aganagic, Klemm and Vafa (AKV),
we obtain the relevant change of variables from the linear sigma model to the non-linear sigma model
variablesthe open-string mirror map. Using this mirror map, we obtain results in agreement with
those of AV and AKV for the counting of holomorphic disc instantons.
1. Introduction
D-branes have come to play a fundamental role in our understanding of string theory,
especially for the case CalabiYau (CY) compactifications. Subsequent to the initial
description of D-branes wrapping supersymmetric cycles of CY manifolds [1], it was
realised that A- and B-type D-branes1 are closely related to the objects that are relevant to
E-mail addresses: suresh@chaos.iitm.ernet.in (S. Govindarajan), jayaram@imsc.ernet.in (T. Jayaraman),
tapo@theory.tifr.res.in (T. Sarkar).
1 The terminology arising from the topological theory that support these branes, in accordance with the openstring world sheet boundary conditions that preserve some amount of supersymmetry in either case. We shall
refer to the D-branes as A-branes/B-branes indicating the appropriate boundary condition.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 9 0 1 - X
S. Govindarajan et al. / Nuclear Physics B 646 [PM] (2002) 498523
499
Kontsevichs homological mirror symmetry proposal [2,3]. The fact that mirror symmetry,
generally speaking, exchanges A-branes and B-branes has begun to play a crucial role in
understanding several aspects of stringy geometry on CalabiYau manifolds.
One of the applications of mirror symmetry in type II compactifications has been the
computation of closed string instanton corrections to various physical quantities. With the
inclusion of D-brane sectors for strings moving on CalabiYau backgrounds, one has, in
addition, to consider a generalisation of mirror symmetry that also includes the open string
sectors. In particular, one may expect that mirror symmetry in the open and closed string
sectors taken together is relevant to understanding the contribution of open-string instanton
corrections.
In contrast to the case of purely closed strings on CY, it has been pointed out that various
quantities in the world-volume theories of both A- and B-type D-branes acquire open-string
instanton corrections. In general it is expected that the F-terms in the world-volume theory
have open-string instanton corrections for A-branes whereas the D-terms are the ones that
acquire instanton corrections for B-branes [4,5]. Various general considerations relevant to
open-string instanton corrections in the case of compact CalabiYau manifolds as well as
non-compact CalabiYau manifolds have been studied earlier [6,7].
However, the explicit calculation of open-string instanton corrections to the worldvolume superpotential for A-branes using mirror symmetry have first appeared recently
in the important work of Aganagic and Vafa [8] and Aganagic, Klemm and Vafa
[9]. The authors in these papers have computed the world-volume superpotential for
some non-compact A-branes in non-compact CalabiYau manifolds (in particular, those
that can be described without a world-sheet superpotential), by appealing to mirror
symmetry and using the holomorphic ChernSimons action on the B-brane side to reduce
the problem to a more tractable computation. This computation uses and extends the
techniques first described in [10,11], whereby the mirror transformation, in the presence
of D-branes, can be explicitly implemented as a world-sheet duality transformation
involving the fields of a Gauged Linear Sigma Model (GLSM) that describes these
D-brane configurations. Remarkably, the authors of [8,9] showed that the (instanton
generated) A-brane superpotential thus computed conformed exactly to the strong
integrality predictions of such quantities from completely different considerations of large
N duality [12].
We briefly recall here the manner in which this integrality predictions were first made.
In [13] it was conjectured that the large N limit of ChernSimons theory on S 3 is dual
to A-type closed topological string theory on the resolved conifold, with a particular
identification between the parameters of the two theories. This conjecture was verified
at the level of the partition function on both sides. The conjecture was extended to
the observables of the ChernSimons theory, namely, the knot invariants in [12]. By
identifying the extra D-branes (associated with the knots on the ChernSimons side) on
the resolved conifold, non-trivial predictions were made in [12] regarding the structure
of the topological string amplitudes, using ideas of [15]. (Knot theoretic aspects of the
conjecture were verified in [14].) Further, from the fact that topological disc amplitudes
compute superpotential terms in N = 1 d = 4 gauge theories, a general structure for the
A-type superpotential was proposed, which had the integrality properties that we referred
to above.
500
It is of interest to ask the question whether the A-brane world-volume superpotential,

together with the open-string instanton corrections, can be directly computed without
explicit recourse to mirror symmetry, particularly in the framework of the GLSM [16].
In this paper, we will take the first steps in this direction. We will show that the description
of the A-branes in the linear sigma model framework suggests a natural and concrete
description of these (partially compactified) open-string instanton moduli spaces. These
instanton moduli spaces will be seen to be intimately related to some auxiliary toric
varieties that we shall refer to as boundary toric varieties. For these boundary toric varieties
we will write down the appropriate GelfandKapranovZelevinsky (GKZ) differential
equations, by an extension of the methods developed for the closed string case, particularly
for non-compact CY manifolds by [17,18]. We will then show how the relevant solutions
of the GKZ equations describe the world-volume superpotential for the A-branes in the
examples considered by [8,9]. We will also show how to obtain the analogue of the openstring mirror map in this context, using in part, an ansatz due to Aganagic, Klemm and
Vafa.
This paper is organised as follows. In Section 2, we describe some basic facts
about the Gauged Linear Sigma Model. After describing the construction of A- and
B-type superspace, we discuss the boundary conditions of [8] that describes A-branes.
In Section 3, we discuss the GLSM for A-branes and the boundary conditions therein.
In Section 4, we discuss the topological aspects of the A-model, before moving on, in
Section 5, to the open string instanton moduli space. After discussing briefly certain issues
of stability, we introduce the boundary toric variety. In Section 6, we consider aspects of
local mirror symmetry and the GKZ system of equations for our boundary toric variety. We
discuss how to obtain the bulk and boundary periods that are relevant for the computation
of the A-brane superpotential in the mirror. Section 7 deals with some examples in order
to illustrate the methods. We conclude with some observations in Section 8.
2. Background
The supersymmetry generators for (2, 2) supersymmetry can be written as (in the
notation of [16])

susy = Q + Q
+ + Q
.
= Q+ + + Q + Q
(2.1)
In superspace with coordinates (x m , , ), the supersymmetry generators have the

following representation
Q =
+ i (0 1 ),

= i (0 1 ).
Q

(2.2)
On the boundary, one set of linear combinations of this supersymmetry is preserved.

Let us denote the
Let us denote the unbroken supersymmetry generators by Q and Q.
0
Then, the supersymmetry generators
boundary in superspace by the coordinates (x , , ).
501
can be given the following representation

= i 0 .
Q
(2.3)

We will now work out the precise relationship between the boundary coordinates/
supersymmetry generators. We will deal with the A- and B-type cases separately.
Q=
+ i 0 ,
A-type superspace Under A-type supersymmetry, let
A 2 = 2 + ,
where we use the subscript A to denote that we are considering boundaries that preserve
A-type supersymmetry. Then, on using
A
A A QA A Q

+ Q+ + + Q + Q
+ ,
= + Q
we obtain
QA =
(2.4)

Q+ + Q
and A 2 + = 2 .
B-type superspace In a similar fashion one chooses B

Q+ + Q
Q+ + Q
B = B
B
2
2

B QB B QB ,
2 = 2 + .
(2.5)
from which we identify

QB =
and B 2 + = 2 .
Q+ + Q
2.1. Boundary multiplets

the superderivatives are
In boundary superspace with coordinates (x 0 , , ),
D=
0,
i
= + i 0 .
D

(2.6)
We will deal with two kinds of boundary multiplets: a boundary chiral multiplet and an
unconstrained complex multiplet.
A boundary chiral multiplet satisfying the constraint
= 0,
DL
which has the following component expansion
0 l.
L = l + i i
502
An unconstrained complex superfield U with component expansion

U = u + + (D
+ iR),
where
(2.7)
and R are real fields. Under supersymmetry, the transformations are
u = + ,
= (i0 u + D
+ iR),
= (i0 u + D
+ iR),
(D
+ iR) = i0 i 0 .
(2.8)
Now consider a complex superfield with the following gauge invariance

U U + L,
(2.9)
.
where L is a chiral superfield. A gauge-invariant observable is given by DU
The invariance trivially follows from the chirality of L. has the following superfield
expansion
= + (D
+ iR + i0 u) i
0 .
= DU
(2.10)
We will identify the real part of U with the bosonic u discussed earlier. Keeping this in
mind, we can partially gauge-fix the imaginary part of U as follows
Im(U ) = (R).
(2.11)
Note that the gauge choice is algebraic and hence we will call it the WessZumino (WZ)
gauge. This leaves us with the following gauge invariance which acts as
Re(u) = Re(l),
R = 0 Re(l).
(2.12)
In the WZ gauge, the field U has the following superfield expansion

U = u + + (D
+ iR),
(2.13)
where u is now a real field. Thus, if R = 0, it would be a real superfield.

The supersymmetry and gauge transformations which preserve the WZ gauge are
u = + + l,
= (i0 u + D
+ iR),
R = 0 l,
D
= i0 i 0 ,
(2.14)
where l is now a real parameter. Notice that R is invariant under supersymmetry and
transforms as 0 l under gauge transformations and thus behaves as a gauge field.
2.2. Decomposition of bulk multiplets: A-type
First, consider a chiral superfield which has the following component expansion:
= + 2 + + + 2 + 2 + F i + + (D0 + D1 )
i (D0 D1 ) + i + 2 Q i + 2 Q

i 2 + (D0 D1 )+ + 2 Q

+ i 2 + + (D0 + D1 ) + 2 Q + .
(2.15)
In order to obtain the A-type superfield, we need to set

obtain
503
2 = 2 . Thus, we

| + = = + + (F + iD1 ),
(2.16)
where the prime indicates an A-type superfield. This is a complex unconstrained superfield.
Next, let us consider a twisted chiral superfield which has the following component
expansion
= + i 2 + + i 2 + 2 + (D iv01 )
i (0 1 ) i + + (0 + 1 ) + 2 + (0 1 ) +

+ 2 + + (0 + 1 ) + + 02 12 .
(2.17)
We define the A-type superfield
as in the case of the chiral superfield. We obtain
0

| + = = + i ( + + ) i
(2.18)
which is the superfield expansion for an A-type chiral superfield. In addition, one obtains
a Fermi chiral superfield whose lowest component is ( + ) satisfying
= 21
,
D
(2.19)
where one treats 1

as an A-type chiral superfield with lowest component 1 . The
component expansion for is
= ( + ) + i 2 (D iv01 ) 2 1

+ i0 ( + ) + 2i1 ( + + ) .
(2.20)
Something interesting happens if D iv01 is set to zero on the boundary (part of which we
know is required by the vanishing of ordinary variations). Then, using the bulk equations
of motion for the s, one can see that becomes an anti-chiral boundary superfield.
2.3. The AV boundary conditions
Aganagic and Vafa [8] have constructed a class of special Lagrangian (sL) cycles in toric
varieties by adapting the construction of Harvey and Lawson in Cn . We shall describe this
construction in the language of the GLSM. Let i (i = 1, . . . , n) be the chiral superfields
with U (1) charges Qai which appear in the gauged linear sigma model associated with the
toric variety. The toric variety is obtained by imposing the (bulk) D-term constraint given
by

Da
= ra
Qai |i |2 = 0,
2
e
(2.21)
where a = 1, . . . , h1,1 . After taking into account the gauge invariances involved, one ends
up with some weighted projective space (actually, a line bundle on that space) of dimension
(n h1,1 ). This is the
standard procedure called symplectic reduction. The CalabiYau
condition is given by i Qai = 0.
504
Lagrangian submanifolds in Cn are specified by the following D-term like conditions

(2.22)
qi |i |2 = c , = 1, . . . , r.
i
Note, however, that there is no associated local U (1) invariance as occurs in symplectic
reduction. One further imposes (n r) additional conditions on linear combinations of the
phases Im log i of the fields i

vi Im log i = 0, = 1, . . . , (n r).
(2.23)
i
The (q , v ) specify a Lagrangian submanifold of Cn (with the standard symplectic

structure) if

qi vi = 0.
i
In order for this Lagrangian submanifold in Cn to descend to the toric variety, we choose
h1,1 of the conditions given above to be the D-term conditions associated with the toric

variety. The conditions i Qai vi = 0 imply that the conditions on the phases are invariant
under U (1) gauge transformations. Further, it follows that the submanifold is special
Lagrangian if

qi = 0.
i
This is clearly compatible with the CalabiYau condition on the bulk U (1) charges.
For generic values of c , the sL submanifolds have a boundary. One can obtain sL
submanifolds without boundary by a doubling procedure. However, there exist non-generic
situations (such as the case when one of the c vanish) where this doubling is not necessary.
In other words, when we take an sL submanifold without boundary for generic values of c
to the non-generic value, the sL manifold splits into two.2 We shall refer to one component
as an half-brane following AV [8]. In the toric description, half-branes lie on the edges of
toric skeletons.
In order to be more concrete, let us consider a specific example furnished by the noncompact CalabiYau manifold given by the total space of the line-bundle O(3) over
P2 . This non-compact CalabiYau manifold is described in the GLSM by four fields with
U (1) charge Q = (3, 1, 1, 1) and the boundary conditions of the special Lagrangian
submanifold are given by q 1 = (1, 1, 0, 0), q 2 = (1, 0, 1, 0) and v = (1, 1, 1, 1). For
generic values of c , the topology of the sL submanifolds (after doubling) is R S 1 S 1 .
However, for suitable choices of c (see below), one obtains half-branes which are
topologically C S 1 .
Following the discussion in AKV [9], there are three phases (half-branes) of interest:
Phase I given by c2 = 0 and 0 < c1 < r, Phase II given by c1 = 0 and 0 < c2 < r and
Phase III is given by c1 = c2 and c1 < 0. The relevant phase in Fig. 1.
2 Or a suitable power of two depending on the number of doublings involved.
505
Fig. 1. Phases (half-branes) for O(3) on P2 .
In Phase I, for large values of c1 and r when quantum corrections are small, c1 measures
the size of a holomorphic disc ending on the half-brane. This is the real part of a complex
modulus, w 1 , whose imaginary part corresponds to turning on a Wilson line wrapping
the S 1 of the half-brane. w 2 is the other modulus associated with the other S 1 which
appears for generic values of c vanishes on the half-brane. These values suffer quantum
corrections which become relevant for small values of c1 and r. We will denote the
quantum corrected quantities by unhatted quantities. However, w2 will not be independent
and will be determined as a function of w1 on the half-brane.
2.4. The AV/AKV computation of the A-brane superpotential
As we mentioned in the introduction, the essential idea of [8,9] is to compute the
superpotential for the A-branes described in the last subsection by using mirror symmetry
to calculate this in the B-model. Here, we will briefly review their arguments. We will be
interested in the case of a space-filling D6-brane wrapping a sL cycle of a non-compact
CY three-fold.
It was shown in [12] that the amplitude of interest, the (instanton generated) disc
amplitude in the A-model, is of the form
W=

dm,
k
n=0 k,
m

n2
q nk s nm ,
(2.24)
where s = ew , w is the (complexified) volume of the holomorphic disc with boundary

a
ending on a particular one-cycle on the sL submanifold and q a = et , t a being the closed
string Khler class. m
are the winding numbers associated with the holomorphic disc, i.e.,
an element of H1 of the sL submanifold and k is an element of H2 of the CY. The integer n
relates to the multiple coverings of these. Further, this prediction involves the highly nontrivial integers dm,
k , that count the number of primitive disc instantons in this homology
506
class. These have the interpretation of counting the degeneracy of domain wall D4-branes
ending on the space-filling D6-brane [12], which wrap a 2-cycle in the CY (whose Khler
and m
class is given by k)
denotes the wrapping number around the boundary.
Since a direct computation of the superpotential in the A-model involves summing up
open-string instantons, AV and AKV use the mirror map of [10] to map this computation
to one involving holomorphic ChernSimons theory in the mirror B-model. For the case
when the B-brane is of complex dimension one, and wraps a Riemann surface C in the
mirror CY, the dimensional reduction of the ChernSimons theory on C was computed
in [8] in terms of the holomorphic three-form of the ChernSimons theory restricted to
the brane and two scalar fields (denoted by u and v parametrising normal deformations
of the B-brane in the mirror CY.3 The final result for the superpotential is given by the
AbelJacobi map associated with the one-form v du
u
W (u) =
v(u) du,
(2.25)
where u can be chosen to be the location of the other half-brane (which converts the
half-brane into a full one, can be taken to infinity). v is fixed on the B-model side by the
condition that (u, v) be points on the Riemann surface C. However, from the A-model
viewpoint, it suffices to note that
v = u W (u)
(2.26)
which is obtained from the AbelJacobi map.

There are a couple of issues that one has to address before one can translate these
results in the B-model to the A-model side. First, there is an SL(2, Z) ambiguity in the
parametrisation of the B-model variables given by

u
a b
u
.
(2.27)
v
c d
v
Thus, the superpotential given above is also dependent on the choice of u and v. Second,
the variables which appear in the B-model side are not the same as the ones that appear in
the A-model. The change of variables that relates the two is the open-string mirror map.
However, in the limit where quantum corrections are suppressed in the A-model, i.e.,
where the volumes of the holomorphic disc and the CY are large, one can relate the two
sets of variables. However, the condition that Re(u) be the volume of the holomorphic disc
and v 0 in this limit, breaks the SL(2, Z) down to the subgroup given by

u
1 n
u
.
(2.28)
v
0 1
v
Thus, imposing the classical limit leaves us with an ambiguity parametrised by the
integer n. It turns out that for cases where the A-model is dual to large N ChernSimons
theory [12,13,15], this integer reflects a framing ambiguity in the ChernSimons theory.
3 These are w 1 and w 2 , respectively, in the notation of the previous subsection. For the convenience of the
reader, we shall use the convention of AKV in this subsection.
507
The framing ambiguity has recently been studied in the context of the Wilson loop
variables of ChernSimons theory in [23], and evidence has been found that the above
mentioned duality is more appropriate in the context of U (N) ChernSimons theories
rather than the SU(N) ones originally considered in the proposal of [12]. Initial works
on Knot invariants were insensitive to this because the computations were performed in
the standard framing, where the U (N) and SU(N) Knot invariants are the same. However,
hints of this had already appeared in [24]. In [23], explicit computations of certain U (N)
Wilson loop operators have been carried out, and the results have been shown to conform
to the integrality predictions of [12].
Aside from the ambiguity mentioned above, one has to obtain the open-string mirror
map in order to obtain the A-model results. It has been argued in AKV that the appropriate
variables are given by [9]
u = u + 7(
q ),
(2.29)
W
,
v =
(2.30)
u
where u receives corrections only from closed-string instantons. u is computed by
identifying it with the tension of the domain wall which is mirror to the D4-brane in the
A-model. The variables u and v are conjugate fields of the holomorphic ChernSimons
theory. The second equation corresponds to the condition that v remains conjugate to u in
the quantum theory.
As we will see in the next section, the B-model variables, i.e., the unhatted variables,
naturally appear in the GLSM that we construct to describe these half-branes.
3. Linear sigma models for A-branes

3.1. The basic boundary condition
The basic A-type boundary conditions that one considers in a GLSM with n chiral
superfields and r U (1) vector multiplets are

Da
= ra
Qai |i |2 = 0, a = 1, . . . , r.
(3.1)
2
e
i
The bosonic boundary conditions (in the matter sector) are completed by the following
(n r) conditions

vib Im log i = 0, b = 1, . . . , (n r),
(3.2)
i
where the Lagrangian condition implies that the vs satisfy

vib Qai = 0.
i
This condition permits us to identify, in the toric description, the vib with the integral
generators of the one-dimensional cones in the fan 7, the combinatoric data associated
with the toric variety.
508
The boundary conditions on the vector multiplet are given by the superfield condition
a
= 0.
(3.3)
The easy way to understand this is to notice that in the topological A-model, the field is
proportional to the Khler form whose restriction on a Lagrangian submanifold is zero.
The vanishing of the boundary terms in the ordinary variation is obtained by the conditions
a
v01
= 0.
(3.4)
The above set of boundary conditions form a consistent set of boundary conditions in the
GLSM which also have a proper NLSM limit in the sense used in [19,20]. Let us denote
by |B0 the boundary state (in the GLSM) that corresponds to these boundary conditions.
This reference state plays a role similar to the D6-brane in the context of B-branes in
the GLSM [21,22]. It is not hard to see that it corresponds to the D6-brane in the mirror
CalabiYau manifold. It is also rigid, i.e., there are no moduli associated with this sL
submanifold.
3.2. The AV boundary conditions in the GLSM
The boundary conditions that AV choose can be implemented in the GLSM in a simple
fashion.4 In the matter sector, it corresponds to simply replacing some of the vs by
boundary conditions of the form5

qi |i |2 = c .
i
In the GLSM, we will realise these boundary conditions as low-energy conditions

using complex boundary supermultiplets U . The corresponding boundary state can be
represented by the path-integral

|BAV = [DU ] exp(S1 + S2 + S3 )|B0 ,
(3.5)
where
S1 =

0 2
dx d

i
i
qi

U + U
2
qj
Im log j
(3.6)
This action is invariant under the gauge-invariances

U U + L ,
Im log i
Im log i
+ qi

L + L
,
2
(3.7)
(3.8)
4 A similar construction was pursued in [25] though it differs from ours in some details.
5 The angles dual these are given by q i Im log , where q i are the dual basis vectors (in Cn ) to the q , i.e.,
i
i

they satisfy: i qi qi
=
, i qi Qai = 0 and i qi vi = 0.
509
where L are chiral superfields. This gauge invariances enable us to identify the real parts
of the lowest components of the superfields U and with coordinates on an appropriate
three-cycle. The action S1 is also invariant under bulk gauge transformations. On using the
fact that D a are generators of the bulk gauge invariance, one obtains that
D a |BAV = 0,
given that S2 and S3 (given below) are also gauge-invariant.

The kinetic energy for DU

1

S2 = 2 dx 0 d 2
2eb

1
1
2 1
0
2
= 2 dx i 0 + D + (R + 0 u ) ,
2
2
eb
(3.9)
where we have introduced the coupling constant eb on the boundary. This has mass
dimension half.
We can now introduce the boundary analogue of the F.I. term:

0
dx d
S3 = Re w

(R + 0 u ) ,
= dx c D
(3.10)
2
is a complex parameter.
where w = c + i 2
3.3. Taking eb
In the limit eb , the fields in the multiplet impose the constraints
D
eb2

i
= c
qi
qi |i |2 = 0,

i i + i +i = 0,

qi i +i + i i = 0,

qi i Fi + i Fi + i
qi i D1 i i D1 i = 0.
(3.11)
The last three equations are the supersymmetric variation of the first equation.
The term is a topological term as one expects. Thus, we can see that the real part of
R behaves like the bulk D term while the imaginary part should behave like v01 . In order
for this correspondence to work, we should require that (after Wick rotation to Euclidean
space and x 0 becomes a circle variable which we will call x 2 with periodicity 2 )

dx 2
(R + 2 u ) = m ,
2
where m are some integers.
510
4. The topological version of the GLSM

In order to discuss the topological aspects of the A-model, we need to Wick rotate:
x 0 ix 2 (Hence one has 0 i2 , R iR and v01 iv12 ). Let us choose to twist
is the BRST operator. For the closed string case, the localisation
the model so that Q
equations are [16] (a = 1, . . . , h1,1 ).
(D1 + iD2 ) = 0,
Fi = 0,
a
D a + v12
= 0,
a = 0.
(4.1)
Let na be the instanton numbers defined by

1
a
na
d 2 x v12
.
2
(4.2)
The action is then equal to

Sbulk = 2it a na ,
a
2
(4.3)
where =
In the open-string sector, one obtains
ta
+ ir a .
D
+ (R + 2 u ) = 0,
= 0
in addition to the boundary condition

vi Im log i = 0.
(4.4)
(4.5)
The winding numbers are defined by

1
m =
(R + 2 u ).
2
(4.6)
The contribution of the boundary terms to the action is

Sbdry = 2is m .
(4.7)
Thus, the total action in the topological theory is given by

S = Sbulk + Sbdry = 2it a na 2is m .
4.1. Observables in the topological theory
The gauge-invariant BRST closed observables in the closed string sector are a . These
have ghost number two. In the open-string sector, the observables in the NLSM limit should
be in one-to-one correspondence with closed forms on the Lagrangian submanifold. In the
AV construction, we assume that a subset of closed one-forms is given by du , where we
consider u to be (compact) coordinates (angles) on the Lagrangian submanifold. Thus,
in the topological model, the corresponding observables are (which are naively BRST
511
exact). Thus one has
Bulk
Boundary
Observable
Ghost No.
2
1
5. The open-string instanton moduli space

The open-string instanton moduli space can be defined to be the moduli space of
the localisation equations that we derived in the previous section. We assume that the
worldsheet is a disc. The relevant equations are:
(D1 + iD2 )i = 0,
D
a
+ v12
= 0,
(5.1)
(5.2)
+ (R + 2 u )|D = 0,
(5.3)
vi
(5.4)
Im log i |D = 0.
The first two equations are bulk equations while the last two equations are on the boundary
of the disc. Thus, we are interested in the moduli space of these equations modulo bulk as
well as boundary (real) gauge invariances. Further, the instanton moduli spaces are fixed
by the instanton numbers na as well as the winding numbers m .
The first equation is invariant under complex gauge transformations. As in the bulk case,
we shall consider the last two equations as gauge-fixing of the imaginary part of the gauge
transformation. Thus, we shall be interested in the moduli space of the equations
(D1 + iD2 )i = 0
(5.5)
modulo complex gauge invariances. Under suitable circumstances, i.e., when one is dealing
with a stable situation, this moduli space is identical to the moduli space of the full set
of equations including (5.2) and (5.3) modulo real gauge invariance. In a suitable gauge
function on the disc. In the
where vza = 0, this equation states that i is a holomorphic
closed string case, this implies that i are sections of O(ni a Qai na ) in the sector with
instanton number na .
5.1. Stability
In order to obtain the conditions under which stability is obtained (see, for instance,
[26]), we shall first consider the bulk equation (5.2). Let us choose the worldsheet be a disc
D. Integrating (5.2) over the disc we obtain

2na + e2 Qai |i |2 = e2 r a vol(D).
(5.6)
D
512
Let us consider the case when e2 r a 0 (for some fixed a) and all fields i whose U (1)
charge is negative are set to zero. One then obtains the following inequality
2na
(5.7)
.
e2 r a
It is not hard to see that for fixed na 0, this condition can always be satisfied for large
enough e2 r a . Thus stability is guaranteed in this case. A similar argument occurs for the
boundary equation (5.3) leads to the condition
vol(D)
vol(D)
2m
,
eb2 c
(5.8)
where stability is again guaranteed for large enough eb2 c for m 0. The conditions
involving the phases (Eq. (5.4)) can always be imposed and hence do not play a role in
considerations of stability.
Let us now consider the situation where the disc is large, i.e., can be approximated
by the complex plane. Then, one has the following interpretation for the bulk instanton
numbers and the boundary winding numbers. Consider ni Qai na + qi m . Then, one has
(as solutions of the localisation equations):
ni < 0.
(1) i = 0
whenever
ni
(2) i = j =0
i,j zj otherwise,
where z is a (complex) coordinate on the disc. The winding numbers (instanton numbers)
fix the leading power of z (number of zeros on the disc). Let us denote this space by Yna ,m .
Thus, the moduli space is given by the space of i,j for all ni > 0 modulo gauge
invariances. This is fixed by:
(1) the bulk gauge invariance is fixed by a D-term of the form
n

i

a
2
Qi
|i,j | = r a
i
j =0
and the associated U (1) gauge invariance;

(2) the boundary condition on the disc, Eq. (5.3), becomes a constraint of the form
n

i

2
qi
|i,j | = c ;
i
j =0
(3) the boundary conditions in (5.4) can be considered to arise from the gauge fixing of
the U (1) gauge invariance
i,j eivi i,j .

In this case, the moduli space is the variety given by
Mna ,m = (Yna ,m Fna ,m )/T ,
(5.9)
513
where T acts as
a
for C ,
(5.10)
qi
for R ,
(5.11)
for real .
(5.12)
i,j Qi i,j ,
i,j i,j ,
i,j e
ivi
i,j ,
Fna ,m is the set of points in Yna ,m for which the D-term constraints given above are not
satisfied. Note that in the zero instanton sector (i.e., na = 0, m = 0), one obtains the sL
submanifold as expected.
5.2. The boundary toric variety
Suppose, we treat the D-term-like boundary conditions which are specifiedby qi on
par with the bulk D-terms specified by Qai . The special Lagrangian condition i qi = 0
now becomes the CalabiYau condition on a toric-variety which we shall call the boundary
toric variety. The boundary moduli now become bulk moduli on the toric variety. One can
see that the spaces Yna ,m and Mna ,m obtained in the previous subsection play a similar
role for the boundary toric variety. A well known method of computing mirror periods
of a CalabiYau hypersurface is by the PicardFuchs system of differential equations. By
treating the bulk and the boundary D-term constraints on the same footing, one might
hope to compute the period integrals of the boundary toric variety in a similar fashion.
This should, in principle, directly give the necessary variables for the computation of
the worldvolume superpotential for A-branes. In the following sections, we will proceed
with such a computation, and show that the boundary toric variety indeed reproduces the
structure of the A-model amplitude predicted in [12].
6. Summing up open-string instantons

6.1. The GKZ equation for the bulk GLSM
As discussed earlier, the toric data is seen in the GLSM construction through the U (1)
charges, Qai (a = 1, . . . , r), of the chiral superfields and vib (b = 1, . . . , (n r)) which are
the integral generators of one-dimensional cones in the fan 7 of the toric variety. Consider
the differential operators constructed from this data
Qai
Qai
LQa =
(6.1)
,
ai
ai
a
a
Qi >0
Zb =
n

i=0
vib ai
Qi <0
b
ai
(6.2)
with = (1, 0, . . . , 0) and ai are coordinates on Cn . The two operators given above
define a consistent system of differential equations, due to Gelfand, Kapranov and
514
Zelevinsky, the GKZ hypergeometric system with exponent
= 0,
LQa (a)
Zb (a)
= 0.
(6.3)
It is known that the periods (associated with three-cycles) of the mirror CalabiYau
manifold are solutions of a PicardFuchs differential equation which naturally arises in
the special geometry associated with vector multiplets with N = 2 supersymmetry in four
dimensions. It turns out that for the CY manifolds which arise as hypersurfaces in toric
varieties, the GKZ equations described above are closely related to these PicardFuchs
equations. In the cases that we will be interested in, the above system of equations provide
the PicardFuchs differential equations. In general, however, one may need to supplement
the above system by additional differential operators to convert the GKZ into the relevant
PicardFuchs equation.
The second set of differential equations, i.e., those given by the Zb , are solved for by a
change of variables ai za given by

a
za = (ai )Qi ,
i
and the change (a)
=
L
1
a0 (z).
Further, it is useful to rewrite the differential operators
in terms of logarithmic derivatives a = za z a .

Let us assume that za 0 is the large-volume limit. There are two classes of solutions
(periods) of the GKZ equation which are of interest (see [17] for the case of compact CY
manifolds and [18] for the non-compact case [local mirror symmetry]).
Qa
(1) Those with logarithmic behaviour as z 0. These are important in providing the
change of variables which form the mirror map. This also relates the GLSM variables
with the NLSM (large-volume) variables. This is clearly seen in the work of Morrison
and Plesser [27].
(2) Those that behave as (log z)2 as z 0. Using special geometry, these are related to
the first derivatives of the prepotential. This prepotential encodes the corrections due
to closed-string instantons to the classical one obtained from the triple intersection
numbers of two-cycles on the CY manifold.
6.2. The GKZ system for the LSM with boundary
We will now show that we can sensibly fit the case of LSM with boundary into this
framework and write down the corresponding GKZ equations. We will explicitly show this
for all the cases that have been considered by AV and AKV. Loosely speaking, the AV
boundary conditions correspond to replacing some of the vs with qs. This suggests that
the GKZ obtained by the set of charge vectors (Qai , qi ) might be relevant for obtaining the
open-string instanton effects just as the bulk GKZ for the closed-string case. Consider the
following operators
a
a
Qi
Qi
LQa =
(6.4)
,
ai
ai
a
a
Qi >0
Qi <0
Lq
qi
qi
=
,
ai
ai
qi >0
Z =
n

515
(6.5)
qi <0
vi ai
i=0
.
ai
(6.6)
This is the GKZ which is naturally associated with the boundary toric variety. As before,
we will solve the differential equations given by Z by a change of variables
za =
(ai )Qi ,
s =
(ai )qi ,
i
where s are related to boundary moduli just as the za are related to bulk moduli. It is
useful to separate the remaining differential operators into those that arise from the Qai
(we will call these Lbulk ) and those that arise from the qi (we will call these Lbdry ). Note
that Lbulk is not quite the differential operator which appears in the closed-string GKZit
also involves the boundary variables.
6.3. Identifying the solutions of interest
We will restrict our attention to the cases of the half-branes with topology C S 1 and let
s = exp(w)
be the modulus associated with the S 1 . The limit where quantum corrections
are small is given by z, s 0.
By construction, the bulk periods are solutions of the differential equations involving
Lbulk . In particular, the solutions that provide the closed-string mirror map are solutions
with asymptotic behaviour log z as z, s 0. The AKV ansatz plays a similar role in
obtaining the open-string mirror map, i.e., we look for solutions of Lbulk with the property
w = log s + 7(z).
It is not hard to see that the above ansatz is not compatible with Lbdry since 7 is not a
function of s.
The solution associated with the superpotential is obtained by an ansatz of the form
w W (w)
=
m)zk s m ,
a(k,

k,m
where we require that the zero instanton contribution vanish, i.e., a(0, 0) = 0 and the only
m) are as given by the considerations of Section 5. To be more precise,
non-vanishing a(k,

we set a(k, m) = 0 whenever the corresponding instanton moduli space does not exist.
Unlike the solution associated with the open-string mirror map we require that this solves
the full boundary GKZ differential equation.
516
7. Examples
7.1. O(3) on P2
7.1.1. Bulk periods
The GKZ equation is given by
3

z + z(3z + 2)(3z + 1)3z = 0.
(7.1)
The basic solutions which we consider are fixed by the behaviour near z = 0: (i) the
constant solution; (ii) the log z solution and (iii) the (log z)2 solution. Using the Frobenius
method, one obtains
0 = 1,
1 = log z +
(7.2)
()n
n=1
zn
3n!
,
(n!)3 n
(7.3)
3n

1
1 zn
1 2
2
n 3n!
+
,
2 = 1 + C +
()
2
n
j n
(n!)3
(7.4)
j =n+1
n=1

n 3n! zn
where C =
n=1 () (n!)3 n .
The mirror map relates the large-volume coordinate t to 1 . Thus, we obtain
t = 1 /0 = log z

3n! zn
()n
(n!)3 n
(7.5)
n=1
which has the right behaviour z = et as z 0 and the terms in the summation are
instanton corrections. The inverse relation is of the form
z = q + 6q 2 + 9q 3 + 56q 4 300q 5 + 3942q 6 48412q 7 + ,
where q
(7.6)
et .
7.1.2. Phase I
The GKZ equation for the boundary toric variety is given by
2

z (z + s ) + z(3z + s + 2)(3z + s + 1)(3z + s ) = 0,

(z + s ) + s(3z + s ) = 0.
(7.7)
(7.8)
The first equation represents Lbulk and the second represents Lbdry for this example. Since
c2 = 0 in this phase, we have switched-off the coordinate associated with c2 .
The boundary flat coordinate u satisfies the first of the two equations given above. For
z, s 0, one requires that u = log s. It can be argued that the general solution should
take the form
u = log s + 7(z),
517
i.e., the boundary flat coordinate differs from the GLSM coordinate only by closed moduli.
One can then easily prove that
1
u = log s (t t ) + i
3
is a solution of the first of the two equations (i.e., Lbulk ) that make up the boundary GKZ
equations.
Defining s = eu , one obtains the following

s = s 1 + 2q q 2 + 20q 3 177q 4 + 1980q 5 24023q 6

(7.9)
+ 97684q 7 + .
This equation along with the earlier equation for z, gives us the required change of variables
from z, s to the flat coordinates s and q.
We are now in a position to compute the boundary superpotential. We require that it
be a solution v of both equations of the boundary GKZ. It should satisfy the following
conditions: (a) v should vanish and be regular at z = s = 0. (b) By consider the instantons
in the GLSM, one obtains the vector kQ + mq 1 = (3k m, k + m, k, k). In Phase (i), we
need both 1 and 3 to be non-vanishing. This is achieved if k 0 and k + m 0. This
suggests the following ansatz for this solution:
v =
a(k, m)zk s m .
k=0 m=k
It may seem that this solution cannot be regular at z = s = 0. But, since we require that
c1 < r, this requires the limit be take such that z/s 0. Thus zk s m vanishes for z = s = 0
as long as k + m 0.
The solution is given by
v =

k=0 m=k
(1)k+m
(3k + m)
zk s m .
(k + m + 1) (k + 1)2
(7.10)
Putting in the expressions for the flat coordinates, by making the substitutions as in (7.6)
and (7.9), and comparing with the expression

m

n

dk,m q k sm .
v =
(7.11)
mdk,m log 1 q k s m =
n
k,m
k,m,n
Table 1 lists the degeneracies dk,m of disc instantons.

7.1.3. Phase III
The charges relevant for this phase are: Q = (3, 1, 1, 1) and q = q 2 = (1, 0, 1, 0).
The relevant GKZ equations are
2

z (z r ) + z(3z r + 2)(3z r + 1)(3z r ) = 0,

(3z r ) + r(z r ) = 0,
(7.12)
518
Table 1
Degeneracies for Phase I of O(3) on P2
m\k
5
4
3
2
1
1
2
3
4
5
6
7
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
0
0
0
1
2
5
7
9
12
15
19
23
0
0
1
4
12
40
61
93
140
206
296
416
0
2
10
32
104
399
648
1070
1750
2821
4450
6868
5
28
102
326
1085
4524
7661
13257
22955
39315
66213
109367
where the boundary modulus is represented by r in this phase. In this case, as before, the
boundary flat coordinate v has a general solution
1
v = log r + (t t ) + i
3
and defining r = ev , we have the expansion

r = r 1 2q + 5q 2 32q 3 + 286q 4 3038q 5 + 35870q 6 + .
(7.13)
(7.14)
The allowed instanton numbers in the GLSM are given by looking at kQ + mq =

(m 3k, k, k m, k). Since 0 = 0 in this phase, we need m 3k 0 in addition to
k 0 which suggests an ansatz of the form (for the superpotential)
u
=

a(k, m)zk r m .
(7.15)
k=0 m=3k
As can be easily seen, the GKZ equations (7.12) are solved with the choice of a(k, m) as
a(k, m) = ()k+m
(m k)
.
(1 3k + m) (k + 1)2
(7.16)
The degeneracies that are obtained from these can be seen to be consistent with those
obtained by AKV (see Table 2).
7.2. O(1) + O(1) on P1
The boundary toric variety is characterised by the charges Q = (1, 1, 1, 1), and
q = (0, 0, 1, 1). Let us consider the phase where c1 = 0 and c2 > 0 (Phase II of AV).
The relevant GKZ equations can easily be seen to be

2
z z(z + s )(z s ) = 0,

(s z ) + s(s + z ) = 0.
(7.17)
519
Table 2
Degeneracies for Phase III of O(3) on P2
m\k
..
.
3
4
5
6
7
8
9
10
11
12
13
14
..
.
..
.
0
0
0
0
0
0
0
0
0
0
0
0
..
.
1
1
1
1
1
1
1
1
1
1
1
1
3
4
5
7
9
12
15
19
23
28
33
39
..
.
18
20
26
36
52
76
111
161
230
323
446
605
..
.
153
160
196
260
365
528
784
1176
1777
2678
4018
5968
..
.
1560
1595
1875
2403
3254
4578
6627
9800
14720
22347
34141
52289
..
.
Where, as before, z = et and we define s = ev . In this example, there are no corrections

to the open string moduli coming from closed string ones, and one can simply write
v = v + i,
u = u + i
(7.18)
so that s = s. In this case, the allowed instanton numbers are given by the vector
kQ + mq = (k, k, k + m, k m), and since 3 = 0, we require m > k, which suggests
an ansatz of the form
u =
a(k, m)zk s m .
(7.19)
k=0 m=k
The appropriate a(k, m) are easy to find, and, defining z = et , the solution of the GKZ
equations can be seen to be
u =

k=0 m=k
(1)k+m
(k + m)
zk s m .
(m k + 1) (k + 1)2
(7.20)
We now reproduce some of the degeneracies dk,m for this example in Table 3. The
numbers agree with those in [8] up to an overall sign.
7.3. Degenerate limit of P1 P1
In this subsection, we will calculate the superpotential for another example, that of
a non-compact CalabiYau, containing P1 P1 . We will consider the theory with one
of the closed string moduli goes to infinity, and will consider the phase where the
relevant charges that define the other closed string modulus and Lagrangian submanifold
are Q = (2, 1, 1, 0, 0), q = (1, 0, 0, 1, 0). The GKZ equations for the boundary toric
520
Table 3
Degeneracies for Phase II of O(1) + O(1) on P1
m\k
..
.
11
12
13
14
15
16
17
18
19
20
21
22
..
.
..
.
0
0
0
0
0
0
0
0
0
0
0
0
..
.
1
1
1
1
1
1
1
1
1
1
1
1
30
36
42
49
56
64
72
81
90
100
110
121
..
.
390
556
770
1040
1375
1785
2280
2871
3570
4389
5341
6440
..
.
variety defined by these charges are

2
z z(2z + s + 1)(2z + s ) = 0,

s + s(2z + s ) = 0,
2730
4690
7700
12152
18564
27552
39900
56520
78540
107255
144210
191180
..
.
11466
24024
47124
87516
155195
264537
435708
696388
1083852
1647455
2451570
3579030
..
.
(7.21)
where we have defined z = et and s = ev . The mirror map dictates that the boundary flat
coordinate in this case is given by
s = (1 + q)s,
(7.22)
where s = ev , and similarly for the other open string modulus. q = et is defined through
z=
q
.
(1 + q)2
(7.23)
The allowed instanton numbers are determined from the vector kQ + mq = (2k
m, k, k, m) and since 1 , 2 , 4 are non-vanishing, we require that k, m 0. Writing an
ansatz for the solution to the GKZ equations,
u =
a(k, m)zk s m .
(7.24)
k=0,m=0
The coefficients are easily determined to be

a(k, m) = ()m
(2k + m)
.
(m + 1) (k + 1)2
(7.25)
Substituting in (7.24), with the summation excluding (0, 0), and, substituting for z and s
in terms of q and s , we recover the integers dk,m which are consistent with the results of
AKV (see Table 4).
521
Table 4
Degeneracies for the degeneration of P1 P1
m\k
..
.
11
12
13
14
15
16
17
18
19
20
21
22
..
.
..
.
0
0
0
0
0
0
0
0
0
0
0
0
..
.
1
1
1
1
1
1
1
1
1
1
1
1
36
42
49
56
64
72
81
90
100
110
121
132
..
.
676
920
1225
1600
2055
2601
3249
4011
4900
5929
7112
8464
..
.
8281
12936
19600
28896
41616
58680
81225
110550
148225
196020
256036
330616
..
.
74529
132496
226576
374544
600935
938961
1432809
2140369
3136441
4516475
6400900
8940100
..
.
8. Conclusion
In this paper, we have constructed a GLSM with boundary which corresponds to
D-branes associated with the class of special Lagrangian cycles proposed by Aganagic
and Vafa. The considerations of the topological field theory associated with the GLSM
naturally leads to the associated boundary toric variety. We have seen that the GKZ
associated with the boundary toric variety provides us with the open-string mirror map
and enables us to verify the integrality conjectures for the degeneracies associated with
open-string instantons. We also obtain natural models for the open-string instanton moduli
spaces. We have however not directly computed the degeneracies in the topological
modelthis involves a generalisation to the GLSM with boundary, of the methods used by
Morrison and Plesser for the closed-string case [27]. The methods used in this paper clearly
have a generalisation to cases involving compact CY manifolds, i.e., those that include a
bulk superpotential in the GLSM. However, there might be cases where the moduli are
related to non-toric deformations for which special techniques are necessary.
The framing ambiguity is quite similar to what may happen in the case of a conifold
singularity, in say the quintic. In that case, one has an S 3 which shrinks to zero size. Its
dual three-cycle is a T 3 . The monodromy action around the conifold singularity generates
the same subgroup of SL(2, Z) that we saw in the framing ambiguity. The volume of the
T 3 thus suffers from a similar ambiguity in the sense that one can always add an integer
times the S 3 volume (which clearly vanishes at the conifold singularity) to this volume.
However, in the case of the quintic, this ambiguity can be fixed by global considerations.
In a similar fashion, we expect the framing ambiguity to be associated with a monodromy
of the GKZ solutions and that in the compact case, this ambiguity should be lifted by global
considerations.
A derivation of the GKZ equations that we use can be obtained in a fashion similar to
the closed-string case. By extending the HoriVafa [10] mirror map, one can represent the
522
periods on the B-model side as a functional integral over some dual fields with appropriate
superpotentials corresponding to all the D-terms, bulk and boundary, that appear on the
A-model side. Using this representation, it is possible to derive the appropriate GKZ in a
manner similar to that of Hori and Vafa for the closed-string periods. From a somewhat
different view point, one notices that the periods can be represented as an integral of
the holomorphic three-form over the three-cycle and then one may obtain the differential
equation satisfied by the periods. The GKZ equations that we use can be derived by
replacing the integral over the holomorphic three-form by the holomorphic ChernSimons
action.
The AV boundary conditions that we have implemented in the GLSM correspond to
two-branes on the B-model side given by intersections of hyperplanes. It is of interest to
extend our methods to construct the GLSM for the A-model mirror to D-branes associated
with vector bundles [22].
While this manuscript was being readied for publication, an interesting paper [28]
appeared which also proposes that the superpotential on half-branes are solutions to a
differential equation of the GKZ type. The differential equation proposed there involves
the addition of new variables (interpreted there as periods of a dual CY four-fold via an
open-closed duality) and thus is not identical to ours. However, it is not clear (to us) as to
how one may identify the solution associated with the superpotentialthis was done in our
case by using the topological model that we discussed in Section 4. The open-string mirror
map as we proposed in this paper is a solution of the bulk GKZ alone. The addition of
new variables (ai in the notation of Section 6) can be used to make the mirror map solve
the boundary GKZ as well. This has a nice interpretation in terms of a M-theory lift
where a disc instanton becomes an S 3 as discussed in [9]. However, the identification of
the solution corresponding to the superpotential is not obvious to us.
Acknowledgements
One of us (S.G.) would like to thank the Theory Division, CERN and two of us (S.G.
and T.J.) the Abdus Salam ICTP for hospitality while this work was being completed.
References
[1] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477 (1996) 407,
hep-th/9606112.
[2] M. Kontsevich, Homological algebra of mirror symmetry, alg-geom/9411018.
[3] C. Vafa, Extending mirror conjecture to CalabiYau with bundles, hep-th/9804131.
[4] I. Brunner, M.R. Douglas, A. Lawrence, C. Rmelsberger, D-branes on the quintic, JHEP 0008 (2000) 015,
hep-th/9906200.
[5] M.R. Douglas, Topics in D-geometry, Class. Quantum Grav. 17 (2000) 1057, hep-th/9910170.
[6] S. Kachru, S. Katz, A.E. Lawrence, J. McGreevy, Mirror symmetry for open strings, Phys. Rev. D 62 (2000)
126005, hep-th/0006047;
S. Kachru, S. Katz, A.E. Lawrence, J. McGreevy, Open string instantons and superpotentials, Phys. Rev.
D 62 (2000) 026001, hep-th/9912151;
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
523
S. Kachru, J. McGreevy, Supersymmetric three-cycles and (super)symmetry breaking, Phys. Rev. D 61

(2000) 026001, hep-th/9908135.
C.I. Lazaroiu, Instanton amplitudes in openclosed topological string theory, hep-th/0011257.
M. Aganagic, C. Vafa, Mirror symmetry, D-branes and counting holomorphic discs, hep-th/0012041.
M. Aganagic, A. Klemm, C. Vafa, Disk instantons, mirror symmetry and the duality web, hep-th/0105045.
K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
H. Ooguri, C. Vafa, Knot invariants and topological strings, Nucl. Phys. B 577 (2000) 419, hep-th/9912123.
R. Gopakumar, C. Vafa, On the gauge theory/geometry correspondence, hep-th/9811131.
J.M. Labastida, M. Marino, Polynomial invariants for torus knots and topological strings, Commun. Math.
Phys. 217 (2001) 423, hep-th/0004196;
P. Ramadevi, T. Sarkar, On link invariants and topological string amplitudes, Nucl. Phys. B 600 (2001) 487,
hep-th/0009188.
R. Gopakumar, C. Vafa, M-theory and topological strings I, hep-th/9809187;
R. Gopakumar, C. Vafa, M-theory and topological strings II, hep-th/9812127.
E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/9301042.
S. Hosono, A. Klemm, S. Theisen, S.T. Yau, Mirror symmetry, mirror map and applications to CalabiYau
hypersurfaces, Commun. Math. Phys. 167 (1995) 301, hep-th/9308122.
T.M. Chiang, A. Klemm, S.T. Yau, E. Zaslow, Local mirror symmetry: calculations and interpretations, Adv.
Theor. Math. Phys. 3 (1999) 495, hep-th/9903053.
S. Govindarajan, T. Jayaraman, On the LandauGinzburg description of boundary CFTs and special
Lagrangian submanifolds, JHEP 0007 (2000) 016, hep-th/0003242.
S. Govindarajan, T. Jayaraman, T. Sarkar, On D-branes from gauged linear sigma models, Nucl. Phys. B 593
(2001) 155182, hep-th/0007075.
S. Govindarajan, T. Jayaraman, D-branes, exceptional sheaves and quivers on CalabiYau manifolds: From
Mukai to McKay, Nucl. Phys. B 600 (2001) 457, hep-th/0010196.
S. Govindarajan, T. Jayaraman, Boundary fermions, coherent sheaves and D-branes on CalabiYau
manifolds, hep-th/0104126.
M. Marino, C. Vafa, Framed knots at large N , hep-th/0108064.
J.M.F. Labastida, M. Marino, C. Vafa, Knots, links and branes at large N , JHEP 0011 (2000) 007, hepth/0010102.
K. Hori, Linear models of supersymmetric D-branes, hep-th/0012179.
S.B. Bradlow, Vortices in holomorphic line bundles over closed Kahler manifolds, Commun. Math.
Phys. 135 (1990) 1.
D.R. Morrison, M.R. Plesser, Summing the instantons: quantum cohomology and mirror symmetry in toric
varieties, Nucl. Phys. 440 (1995) 279, hep-th/9412236.
P. Mayr, N = 1 mirror symmetry and openclosed duality, hep-th/0108229.

Triple intersections and geometric transitions

Julie D. Blum
Physics Department, University of Texas at Austin, Austin, TX 78712, USA
Abstract
The local neighborhood of a triple intersection of fivebranes in type IIA string theory is shown to
be equivalent to type IIB string theory on a noncompact CalabiYau fourfold. The phases and the
effective theory of the intersection are analyzed in detail.
1. Introduction
Many supersymmetric gauge theories are obtained as a sector of string theory in a
particular geometric background. Varying the moduli of the string geometry may allow
passage between different backgrounds and open a window onto the strong coupling
regions of the gauge theory. Some of the phases are better described by open strings while
others are described by closed strings with nonzero fluxes of background fields.
The conifold is a noncompact CalabiYau (CY) threefold described geometrically as a
cone over an S2 S3 base. The transitions between the geometries where an S2 is blown up
(O(1) + O(1) P1 ) and an S3 is blown up (T S3 ) have been studied [1]. Recently,
the transitions between open strings in one of the conifold geometries and fluxes in the
other have been studied as a means to describe phases of N = 1 gauge theories [2]. The
conifold has a T-dual description as a pair of orthogonally intersecting NeveuSchwarz
(NS) fivebranes on R3 [3]. The conifold is the limit of a degenerating T2 T2 fibration
over the complex plane, and one performs T-dualities on two cycles, one from each torus, to
obtain the NS fivebrane description. Our aim in the present paper is to understand better the
triple intersection of NS fivebranes on a string by analogy with the double intersection. The
triple intersection, while sharing features of the double intersection, is more complicated
in many respects. Our motivation was to provide evidence for conjectural bound states
E-mail address: julie@zippy.ph.utexas.edu (J.D. Blum).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 9 6 - 9
J.D. Blum / Nuclear Physics B 646 [PM] (2002) 524539
525
associated to this intersection [4]. After T-duality the triple intersection has a geometric
description, and one can describe the phases through geometric transitions. Work is in
progress to determine the spectrum of BPS states and possibly the metric.
In section two we will discuss the supergravity solution for triple intersections
of fivebranes in eleven-dimensional supergravity. The solution will be reduced to ten
dimensions along one of the transverse directions and shown to be equivalent to the
stringy cosmic string construction of [5]. Finally, a T-duality along three directions
(generally, mirror symmetry for a CY threefold) transforms the solution to a CY fourfold
in type IIB string theory. In section three we present a linear sigma model describing
the local neighborhood of the singularity of the CY fourfold. The model turns out to be
O(1, 1) + O(1, 1) P1 P1 . We show that this model has flops to isomorphic
phases as well as transitions to CY fourfolds with various other cycles. In section four
we show that the perturbative modes of the supersymmetric spacetime two-dimensional
theory of a Dirichlet fivebrane in the P1 P1 phase match with those in the dual fivebrane
picture. The dual fivebrane is a manifold partly constructed from a fivebrane. Anomaly
cancellation for the fivebrane guarantees consistency in the CY fourfold framework. We
conjecture that a fivebrane instanton is the mechanism for resolving the singularity. A local
mirror for the CY fourfold is given. The other phases are compared and also seem to
agree. The analogy with the conifold leads to various conjectures discussed in this section.
Our results indicate that chiral anomalies only occur in the P1 P1 phase where the
fivebranes are unseparated and that the theories are effectively four-dimensional in the
other phases.
2. Supergravity solution of intersecting fivebranes

We start with the following configuration of fivebranes preserving one-eighth of the
supersymmetry in eleven-dimensional supergravity:
0
1 2 3 4 5 6 7 8 9 10

The metric and equations that result from solving the supersymmetry constraints for this
configuration have been determined in [69]. The metric can be written in the following
form:

2
2
2
ds11
= H 1/3 dx 0 + dx 1 + 2H 1/3gmn dzm dzn + H 2/3 dx dx ,
g = Hf f.
(2.1)
In the above f is an arbitrary holomorphic function of the zm , z1 = x 2 +ix 3 , z2 = x 4 +ix 5 ,

z3 = x 6 + ix 7 , and {8, 9, 10}. Note that g is the determinant of the 3 3 matrix gmn .
The threefold metric gmn is required by supersymmetry to be Kahler. The field strength of
526
the three-form gauge field takes the following form:

Fn
= in ln H ,
Fm = im ln H ,
Fmn
= i gmn H 2/3 .
(2.2)
The equation of motion for the gauge field yields

,
2m n H + 2 gmn = Jmsource
n
(2.3)
Jmsource
n
is the magnetic source for the fivebranes. These equations are interesting
where
to analyze directly in eleven-dimensional supergravity since they resemble a generalized
system of monopoles. For our present purposes, however, we need to reduce to type IIA
string theory by taking x 10 to be a hidden compact dimension. The metric becomes
2
2
ds10
= dx 0 + dx 1 + gmn dzm dzn + H (dz d z ),
where z = x 8 + ix 9 , H = e2 =
g
f f
(2.4)
( is the dilaton). The field strengths are:
1
m H = m gzz ,
2
1
Hnz
z = n H = n gzz ,
2
Hmnz
= z gmn ,
Hmzz =
Hmn
z = z gmn ,
(2.5)
and the equation of motion is

m n H + 2z z gmn = Jmsource
.
n
(2.6)
Defining the Kahler form J = igmn dzm dzn leads to the following equation for the
Kahler parameters, = B + iJ ,
z = 0.
(2.7)
This is precisely the mirror of the stringy cosmic string construction where one has z = 0
for the complex structure deformation parameters of a Kahler threefold. Note that the
complex fourfold total space fibered over the z plane with fiber the Kahler threefold X3 is
not Kahler since z Jmn = 0. The would be Kahler form J total of the fourfold satisfies the
equation
i
= Jmsource
(2.8)
2 n
since the fivebranes contribute to the stress-energy tensor. The equations of motion can be
solved by taking gmn to be a CY metric dependent on z and piecing together neighborhoods
3 in type IIB, we
covering the CY in which m n H = 0. By replacing X3 by its mirror X
obtain the stringy cosmic string construction of generally noncompact CY fourfolds [10].
There are constraints on the number of strings, and it remains unclear whether a Ricci
d 2 J total
zzmn
527
3 (in particular ones where the complex structure

flat fourfold can be constructed for all X
degenerates at infinity).
In this paper we will restrict the analysis to the simplest case of a T2 T2 T2 fibration
(the coordinates of this fibration are z1 , z2 , and z3 ) over the z plane in which each fiber
degenerates at one or more points in the z plane. Performing
a T-duality along one cycle
of a torus exchanges the Kahler parameter, = B + i G, of the torus (B is the NS twoform, G is the determinant of the T2 metric) with the complex structure parameter of
the torus. T-dualizing on one cycle of each torus yields a T2 T2 T2 fibration over the
z plane in which the parameters are holomorphic functions of z, and the parameters
are independent of z. To avoid questions of global consistency due to orbifold points on
the tori, we will study the local neighborhood of the degeneration where the three fibers
are noncompact. The T2 T2 case corresponds to the conifold while a T2 fibration that
degenerates at one point corresponds to a KaluzaKlein fivebrane and is nonsingular. There
is additional freedom in the above solution to add various fluxes of background fields. The
supergravity analysis is not usually valid in the limit where the number of fivebranes is
small and the sources are delta functions, but this analysis has provided insight into the
nature of the singularity. In the next section we will determine a noncompact CY fourfold
corresponding to this degeneration.
3. Geometry of the triple intersection

The equations describing the local neighborhood of the degenerate fibers in the triple
T2 fibration can be written in the form
z = a 1 a 2 = b 1 b 2 = c1 c2 .
(3.1)
The singularity can be expressed as the intersection of two quadrics in C6 :

a1 a2 b1 b2 = 0,
(3.2)
b1 b2 c1 c2 = 0.
(3.3)
Each equation by itself describes a conifold so the two equations represent the intersection
of two conifolds. Doing a small resolution of each conifold (blowing up a P1 ) yields
four phaseseach conifold has two flops. There is one large resolution describing the
deformations of intersecting S3 s. Additionally, there are six phases in which one conifold
is blown up and the other is deformed. We will present each of these phases in detail and
show that there are transitions in which the holomorphic four-form is preserved.
3.1. The small resolution
The small resolution is described by the following four equations for eight complex
variables
a1 = z1 b1 ,
b1 = z2 c1 ,
b2 = z1 a2 ,
c2 = z2 b2 .
(3.4)
528
Fig. 1. Toric diagram of the small resolution.
The flops are generated by a1 a2 and c1 c2 . Actually, there are many possible
deformations of the small resolution when one takes into account deformations of equation
(3.3). Fixing (3.2) by a linear transformation, there is still an SO(4, C) invariance.
Eq. (3.3) then has 21 12 1 = 8 complex deformations where we have subtracted
an overall scale. This counting agrees with the fivebrane side. All of these deformations
are nonnormalizable since the fivebrane is noncompact. We conjecture that there are no
deformations of the small resolution preserving supersymmetry. In section four we will find
evidence for this conjecture by obtaining a fivebrane manifold which looks very dissimilar
to this small resolution but produces an equivalent theory.
This resolution is also described by a gauged linear sigma model with two U (1)s and
the following charges for six chiral fields xi under the two U (1)s:
l1 = (1, 1, 1, 1, 0, 0),
l2 = (0, 0, 1, 1, 1, 1).
(3.5)
We can partly visualize the resolution in the toric diagram (Fig. 1) that projects four
dimensions onto a plane. Generically, the diagram depicts a T4 fibration that shrinks to a
T3 on three-dimensional boundaries enclosed by three lines, a T2 on two-plane boundaries,
an S1 on lines in the toric base, and a point at the intersection of four lines.
Here, t1 and t2 are the Kahler parameters for the two P1 s. Note that there are other
two-planes not labeled in the diagram.
3.2. The large resolution
To describe the large resolution we pick a nongeneric deformation of the moduli
space that can be understood fairly easily and then argue that if we deform along a path
that does not pass through singular regions, topological features of the manifold should
remain invariant. Because the CY fourfold is noncompact, most of the parameters of the
deformation are not normalizable. The nongeneric deformation that we choose is
a1 a2 b1 b2 = ,
b 1 b 2 c1 c2 =
(3.6)
529
with , real and > > 0. The base of this manifold can be depicted schematically in
Fig. 2 as a four-torus fibration.
One can describe the space as T S1 fibered over T S3 where the T S1 degenerates
away from the base S3 . The two independent S3 s are noncontractible with sizes
determined by and . The manifold is simply connected. Each S3 intersects the other two
S3 s along a T2 . There is a third S3 with size parameter + that intersects a1 = a2 = 0
and c1 = c2 = 0 but not b1 = b2 = 0 or anywhere else on Fig. 2. The S3 s are not
isolated since the T S1 direction is a zero mode direction. There are also two independent
noncontractible four-cycles with topology S3 S1 . The S3 S1 is homologous to a fourcycle where the S1 shrinks over a portion of the S3 at loci where the T S1 fiber degenerates,
but there is no flat direction preserving the area of the S3 and connecting the two fourcycles. Another picture of the base as a three-torus fibration is presented in Fig.3 with
Fig. 2. Base of the large resolution.
Fig. 3. Base of the large resolution ( real, i imaginary).
530
real and i. (Angles are not drawn accurately.) In the diagram k = i

. The
2
2
+
three edges are noncontractible four-cycles of topology S3 S1 bounding an open fivechain.

In the type IIA picture the and deformations correspond to the resolution in which
two of the three intersecting fivebranes move independently away from the intersection.
The two independent differences of position are determined by and .
To deform the moduli space in a generic way we choose coordinates such that the form
of the first equation is preserved. Writing the second equation as
xi Gij xj =
(3.7)
with x1 = a1 , x2 = a2 , x3 = b1 , x4 = b2 , x5 = c1 , x6 = c2 , and complex, one can analyze

the condition of transversality. For every symmetric, nonzero matrix G, there will be a
possible nontransverse intersection of the hypersurfaces if det(Gij aij j ) = 0 where
1 = 2 = 3 = 4 = 1 and 5 = 6 = 0. Nontransversality implies that a = / for
one of the roots a. Obviously, this condition will not be a generic one. One can show
that transversality implies that the holomorphic four-form has no zeroes or poles. The
deformation should preserve the structure of three three-cycles that are topologically S3
(two are independent) intersecting as described above. In terms of the original fivebranes,
this deformation should correspond to varying the holomorphic four cycle on which one
of the fivebranes is wrapped without compromising transversality. From this point of view
there are nine complex parameters determining the deformation. At a first glance there are
twenty-one complex parameters deforming the second equation where we fix . However,
there is an SO(4, C) group of symmetries preserving the undeformed equation so that we
are left with nine parameters in agreement with the fivebrane count. There are also the
modes and which control the sizes of the S3 s and will turn out not to be normalizable.
3.3. The mixed resolution
We are calling the mixed resolution the CY fourfold obtained by doing a small
resolution on one of the two conifolds and a large resolution on the other. There are six
isomorphic phases and we will describe one. The equations are as follows:
a1 = zb1 ,
b2 = za2 ,
b1 b2 c1 c2 = .
(3.8)
The base of this CY fourfold is a noncontractible five-cycle containing an S2 and an S3 .

We can visualize the base of the manifold in Fig. 4 where we take real and positive.
The horizontal edges are S2 S1 with the S1 contracting at the opposite edge while the
angled edges are S3 . Since the minimal S2 is isolated at b1 = 0 inside the five-cycle, the
S2 S1 three-cycle, although homologous to the S2 at c1 = 0, is not connected to it by a
flat direction that preserves the radius of the S2 . This cycle is not supersymmetric since it
is not Lagrangian.
The normal directions are b1 b 2 and c1 + c2 constrained by b1 b2 c1 c2 = . There is
a zero mode direction for the S2 along b1 = b2 = 0 and for S3 inside the five-cycle. In this
531
Fig. 4. Base of the mixed resolution.
phase the modes associated with the deformation are massive because the four-cycle of
topology S3 S1 shrinks to an S3 at the poles of the S2 .
In terms of the fivebranes of type IIA, the mixed resolution corresponds roughly to
moving one of the three fivebranes a distance from the intersection determined by while
resolving the remaining two fivebranes into a smooth fivebrane wrapped on a Riemann
surface parametrized by the Kahler modulus of the two-sphere. The flop corresponds to
going from a positive imaginary to a negative imaginary value of this modulus. This
description is rough because we expect that an exact description will involve one connected
fivebrane rather than two discrete pieces.
3.4. Geometric transitions
To show that there are possible transitions between all of the phases that are described
above, we note that there is a common set of coordinates for the small resolution, the
nongeneric large resolution, and the mixed resolution in which the holomorphic four-form
is expressed as
=
da1 db1 db2 dc2

.
a 1 c2
(3.9)
The transversality of the intersections for the various resolutions implies that has
no zeroes or poles. The large resolution generic takes another form in general but
can be continuously deformed to the above without encountering zeroes or poles so
that supersymmetry is preserved. The existence of ensures that the holonomy is no
larger group than SU(4). The mixed resolution can be described as a manifold with base
O(1) + O(1) P1 and fiber T S1 where the fiber degenerates along b1 b2 = (see
(3.8)). The large resolution can be described as a T S1 fibration over T S3 that degenerates
along b1 = 0 and b2 = 0. The base in each case has a nonvanishing and covariantly constant
holomorphic three-form. The pullback of this form to the CY fourfold must have a zero
or pole. Otherwise, we could obtain a covariantly constant one-form by wedging with the
antiholomorphic four-form and taking the Hodge star dual. This is impossible since the
manifold is simply connected. The holonomy in all cases is, thus, SU(4). The dynamics of
the possible transitions will be discussed in the next section.
532
4. Effective theory of the triple intersection

In this section we argue that there is a transition of the triply intersecting fivebrane
in type IIA through a conjectured fivebrane instanton to a smooth fivebrane wrapped
on a P2 four-cycle in the noncompact CY fourfold O(4) P3 dual to a Dirichlet
fivebrane wrapped on the P1 P1 four-cycle of the small resolution in type IIB. The
analogous transition in type IIB for the doubly intersecting fivebrane does not occur
because a threebrane instanton mediating a transition to the O(3) P2 would not be
supersymmetric.
The mixed and large resolutions are similar to the S2 and S3 resolutions of the conifold.
There are extra cycles with sizes determined by , and the conifold is retrieved in the
limit . On the fivebrane side one has two possible types of resolution of the doubly
intersecting fivebrane T-dual to the conifold with an additional fivebrane separated from
the intersection resolution by a distance determined by . The double intersection alone is
regained as . For the conifold one has conjectured transitions of the geometry in
type IIB in which a large number, Nc , of fivebranes wrapped on S2 is dual to H field flux
on S3 , and the S3 resolution describes a confined phase of the N = 1 supersymmetric fourdimensional SU(Nc ) gauge theory. In the intersecting fivebrane picture (type IIA) one has
Nc fourbranes joining the two fivebranes corresponding to the fivebranes wrapped on S2
and a smooth fivebrane wrapped on a Riemann surface with RamondRamond (RR) flux in
the phase corresponding to the S3 resolution. (The intersecting fivebranes of type IIA and
the conifold of type IIB are related by one T-duality here.) In our case the supersymmetry
breaking is inversely proportional to , and it may be possible to understand the resulting
transitions as a perturbative deformation of the N = 1 transitions in the parameter 1 .
One would expect that the extra fivebrane corresponds to a massive BPS string in the fourdimensional gauge theory. As 0 this string becomes tensionless. To have more than
a qualitative understanding, the metric needs to be determined, possibly as a perturbative
deformation of the conifold metric in the parameter 1 .
4.1. The small resolution
The effective two-dimensional theory for the triple intersection of fivebranes in type
IIA has a chiral (0, 4) supersymmetry. Similarly, type IIB on a CY fourfold has (0, 4)
supersymmetry. We require that the spatial dimension of the two-dimensional theory be a
circle so that fivebrane deformations in the fivebrane theory are normalizable and to avoid
supergravity anomalies in the CY fourfold theory as discussed below. To determine the zero
modes of the triple intersection, we first consider the case of a smooth fivebrane wrapping
a four-cycle on a six torus as discussed in [11]. This analysis is mostly inapplicable when
the six-torus is decompactified, and the intersection is localized. We do not have a rigorous
derivation of the zero mode spectrum but argue that the self-intersecting fivebrane can be
resolved to a smooth fivebrane wrapped on a P2 face of the manifold P3 . Requiring that the
first Chern class of the total space vanish, we obtain O(4) P3 . This resolution reduces
the supersymmetry to (0, 2). The moduli space of P2 s inside P3 is P3 . We illustrate the
fivebrane resolution of the singularity as a tetrahedron in Fig. 5. The fivebrane is wrapped
on the shaded face.
533
Fig. 5. Small fivebrane resolution-toric diagram of P3 .
Fig. 6. Instanton transition from triple intersection to smooth fivebrane.
If we exclude deformations of the fivebrane that intersect the point labeled x, the moduli
space of deformations is C3 . When the fivebrane reaches x, there is a flop transition
but no singularity. The moduli space of the fivebrane inside P3 encompasses all of the
flopped phases of the CY fourfold with no singularities. If we were to remove a P2 face
on which the fivebrane is wrapped and wrap a fivebrane on the three other faces, we
would return to the singularity of the triply intersecting fivebrane in C3 . The resolution
is a compactification in which the fivebrane flops to the compactifying face. Since this
manifold allows for fivebrane instantons, it is natural to conjecture that this transition is
mediated by a fivebrane instanton in P3 . The transition is depicted in Fig. 6.
Additional arguments for selecting the above manifold are that the broken translational
invariance due to the intersection provides six real modes on R6 and one translational
mode for the compactified dimension of eleven-dimensional supergravity. Translations on
the complex plane acquire a mass because of the conical deficit. As we have seen in the
supergravity analysis, the manifold with the fivebrane is not Kahler so there is no scale
modulus, and the P3 cannot shrink. From the supergravity equations we also see that the
scale of the P3 should be set by the number of fivebranes which in our case is one. The
generator of the second homology class inside the P2 of the fivebrane has genus zero and
positive self-intersection and is therefore self-dual. The deformations of the fivebrane yield
six nonchiral scalar modes from the moduli of P2 inside P3 and one nonchiral scalar mode
from the compactified dimension of eleven-dimensional supergravity. The reduction of the
B field of the fivebrane (with self-dual curvature) on the two-cycle of P2 yields a right
moving scalar. Supersymmetry pairs right moving scalars with fermions. All together we
534
have the following worldsheet theory for scalar bosons (B) and fermions (F ):
NLB = 7,
NRB = 8,
NLF = 0,
NRF = 8,
cL = 7,
cR = 12,
(4.1)
where R is for right movers and L for left movers and c is the central charge.
Now compare this result with the CY fourfold theory in the O(1, 1)+O(1, 1)
P1 P1 phase. In two dimensions all moduli of the vacua must be dynamical variables. Supersymmetry constrains the number of left moving matter scalars to be a multiple of four.
Reduction of the Kahler form and the two B fields on the P1 s yields six nonchiral scalars.
The RR scalar provides another nonchiral scalar. Additionally, there is one antiself-dual
four-cycle which provides a left moving scalar. Since the four-cycle is isolated, its selfintersection must be negative. Supersymmetry pairs the left movers with fermions. In the
compact case the number of right moving fermions is determined by index theorems to
be proportional to the number of three-cycles [12]. We assume that if the small resolution
is a limit of a compact CY fourfold with three-cycles that these modes are nonnormalizable in this limit. The dilaton goes into the supergravity multiplet which includes four
right moving gravitinos and four left moving spin one-half fermions. In order to match
the supersymmetry of the effective theory of the fivebrane resolution which is broken to
(0, 2) by the CY fourfold, we need to wrap a Dirichlet fivebrane around the four-cycle. The
fivebrane also adds a U (1) field to match the U (1) from the membrane potential reduced
on the P1 of P2 . The moduli of the four-cycle (Kahler parameters) are zero modes in the
two-dimensional effective theory of the fivebrane. Additionally, the RR fields couple to the
fivebrane giving the same zero modes as for the CY fourfold. The scalar from the antiselfdual four-cycle is charged under the U (1) of the fivebrane. The zero mode spectrum of the
matter multiplets agrees with that found for the triple intersection except that left and right
movers are exchanged. These zero modes would be precisely those of the heterotic string
on a circle if seventeen right movers were not missing.
The conformal anomaly of this theory is the reduction of the fivebrane anomaly which
is cancelled by bulk counterterms as has been shown in great detail by many authors. On
the other hand, type IIB on a compact CY fourfold is not anomalous as shown in [12].
Accordingly, the appropriate counterterms are not present for type IIB on a compact CY
fourfold. The T-duality relation ensures that these terms are present to cancel the anomaly
for our noncompact case. From the discussion we see that a compact CY fourfold cannot
be constructed easily out of intersecting fivebranes because the conformal anomaly of the
(0, 4) theory generally requires a bulk counterterm. The supergravity multiplet in type IIA
is not anomalous, but there is a two-dimensional anomaly for this multiplet in type IIB
theory. The T-dual counterterms will not cancel this anomaly. Consistency requires that
either the supergravity multiplet be decoupled or that there are further corrections to type
IIB on a noncompact CY fourfold. As we mentioned above the two-dimensional theory is
compactified on a circle so there is no supergravity anomaly. Threebrane instantons may
smooth the flop transitions.
Note that the case of four triply intersecting fivebranes can be resolved by fivebrane
instantons to a smooth fivebrane wrapped around a K3 surface embedded in P3 . Counting
535
deformations and two-cycles we obtain 72 right moving scalars and fermions and 88 left
movers. The amount of supersymmetry makes this theory likely equivalent to the heterotic
string on the product of a two-torus and a noncompact CY threefold.
4.2. Local mirror symmetry
The resolution of the triple intersection singularity by a fivebrane instanton reduces the
supersymmetry to (0, 2). There is another resolution that preserves (0, 4) supersymmetry
and should be equivalent to the CY fourfold in the small resolution phase by local mirror
symmetry and T-duality. Our discussion here has many gaps as we are not able to fully
analyze this system. Despite these gaps this presentation may be useful for further analysis
of this system. The techniques of local mirror symmetry applied to a gauged linear sigma
model are discussed in [13]. Starting with the linear sigma model (3.5), the mirror CY
fourfold is obtained from the following equations:
z = 1 + eu + ev + ew + evut1 + evwt2 ,
z = xy.
(4.2)
The first equation is a noncompact holomorphic surface fibered over the z plane that
degenerates at two points in the z plane for finite u, v, and w and at z = 1 for u, v, w
. The second equation is a C fibration that degenerates at z = 0. The total space has
no singularities for generic t1 and t2 . At each degeneration of the holomorphic surface
for finite u, v, and w, we expect but cannot prove that an S2 shrinks. We can form two
four-cycles of topology S4 by connecting z = 0 with the two points where the holomorphic
surface degenerates. A linear combination of these two cycles as well as a cycle joining
z = 0 and z = 1 is expected to be mirror to the four-cycle, P1 P1 , and antiself-dual.
Mirror symmetry requires that this four-cycle be rigid as there should not be a compact
two-cycle. It is not clear to me how this works. The mirror of the zero-cycle is expected to
be noncompact.
The two S4 s are combinations of cycles with Hodge type (1, 3) and (3, 1), and their
deformations correspond to complex structure deformations. Generically, the deformations
of middle-dimensional Lagrangian cycles are not normalizable but are logarithmically
divergent. In two dimensions massless scalars have logarithmically divergent two-point
functions so that this type of divergence is acceptable here. Each S4 yields three nonchiral
scalars in the effective two-dimensional theory from the complex structure deformations
and the reduction of the RR four-form. The (2, 2) cycle mirror to P1 P1 gives a left
moving scalar, and the RR scalar gives a nonchiral scalar. Supersymmetry generates left
moving fermionic partners. With a lot of assumptions, the zero mode spectrum is the same
as what we previously found.
By T-dualizing on the circle of the C fibration, we obtain a type IIA fivebrane wrapped
on the noncompact surface at z = 0
1 + eu + ev + ew + evut1 + evwt2 = 0.
(4.3)
536
Introducing another coordinate and taking the coordinates to be projective, we can write
this equation as
xyz + y 2 z + wyz + yz2 + 1 wxz + 2 wxy = 0.
(4.4)
If either or both 1 and 2 diverge, the singularity of the triply intersecting fivebrane is
recovered. Otherwise, there are two isolated singularities when three of the coordinates are
zero which are on the boundary of (4.3). The boundary is the intersection of wxyz = 0 and
(4.4). We will not in this paper be able to determine the deformations and two-cycles of
the noncompact four-cycle to analyze the effective two-dimensional theory.
4.3. The mixed resolution
The phase where we separate one fivebrane from the other two corresponds to the mixed
resolution. The effective theory at the intersection of two fivebranes is four-dimensional,
and the matter content in this phase is nonchiral. There are four nonchiral translational
modes moving the intersection of the two fivebranes in R4 as well as a supersymmetric
completion of fermions. The supersymmetry breaking by the third fivebrane occurs at a
distance from the intersection and only affects fields that interact with the bulk. The zero
modes restricted to the intersection are unaffected by this breaking and are nonchiral.
Going to the CY fourfold, we argue that the effective theory is four-dimensional and
nonchiral. The S2 has four flat, normal spacetime directions, T S1 S1 R. The T S1
fibration degenerates, and the supersymmetry is broken in half at a distance of order ||1/2
from the S2 . Zero modes which come from reducing the theory on S2 are not affected by
this breaking since they do not interact with the bulk. We obtain a left-right symmetric
combination of four scalar bosons and four fermions as well as a U (1) gauge field from
the S3 . The zero mode scalars come from reducing the Kahler form, two B fields, and RR
four-form on S2 . In the four dimensions of spacetime normal to the S2 (T S1 S1 R),
the reduced RR four-form is dual to a scalar. For the effective four-dimensional theory we
expect these modes to be nondynamical. If we compactified this theory on T S1 , we
would have a two-dimensional theory with the following zero mode spectrum:
NLB = 4,
NRB = 4,
NLF = 4,
NRF = 4,
cL = 6,
cR = 6.
(4.5)
We would actually have a continuous family of these theories labeled by which was what
we found before we realized that the effective theories are four-dimensional, and the above
modes are nondynamical.
The modes related to the complex parameter are massive because of the four-cycle
that shrinks at the poles of the S2 . Unless there is a five-form flux on the five-cycle, the
four-form gauge field is pure gauge. The situation here is analogous to the magnetic field
for a Dirac monopole outside of a two-sphere of radius r in R3 . If there is a magnetic field,
its flux is quantized. The total energy of the field outside of the two-sphere is g 2 /r where g
is the magnetic charge unit. Our case is more complicated because there are two scales, the
areas of the S2 and the S3 , and there are three normal directions. One of these directions is
537
compact. There is also the zero mode direction for the minimal S2 which potentially causes
a divergence in the energy. The minimal S2 is a set of measure zero in the five-cycle, and
to estimate the energy we assume there is no divergence. A rough estimate of the energy
of the five-form field is to divide the volume of a minimal eight-ball surrounding the fivecycle by the area squared of the five-cycle. Without knowing the metric and taking the
square of the radius to be the mean square radius of the S2 and the S3 , ||1/2 R, we obtain
an energy scaling as 1/||. Assuming that this energy should be independent of the Kahler
parameter uniquely determines that the energy scale as 1/||. The zero modes of the Kahler
parameter, the expectation value of the RR B field on the S2 , and that of the RR four-form
form a massive supermultiplet in the presence of five-form flux. Note that the mass diverges
at the origin of the mixed resolution so that dynamical transitions into this phase when the
S3 shrinks are not possible. On the fivebrane side this flux should correspond to a magnetic
field. Another argument to support the above hand waving is the following. If
what remains is the small resolution of the conifold with no flux, and this hypermultiplet
should be massless. On the other hand, taking the Kahler parameter to infinity leaves the
S3 resolution of the conifold, and this hypermultiplet should be massive. There is another
massive supermultiplet with mass proportional to ||3/2 from the threebrane wrapping S3
as in the conifold [14]. The four massive modes are the S2 S1 directions and the global
U (1) charge mode.
4.4. The large resolution
The phase where all three fivebranes are separated corresponds to the large resolution.
There are no normalizable translational zero modes in this phase. The effective theory
for two fivebranes joined by a fourbrane is a four-dimensional theory with a decoupled
U (1) parametrized by the separation of the fivebranes in ten-dimensional spacetime and
their difference in position in the compactified eleventh dimension. The two independent
separations of the fivebranes and differences in positions along the compactified eleventh
dimension will not be normalizable modes. One way to see this is that these modes correspond to complex parameter deformations of S3 inside T S3 which are logarithmically
divergent.
On the CY fourfold side we will explain why the theory is trivial at low energies.
There is a subtlety concerning the deformations of the four-cycles. The four-cycles are
topologically S3 S1 s, and the parameter governing the size of the S3 is distinct from that
determining the size of S1 . Deforming an S3 inside T S3 would yield a logarithmically
divergent four-dimensional mode and integrating over the zero mode T S1 direction would
generate an additional divergence for a two-dimensional mode. We obtain no normalizable
scalar modes from the four-cycles in this phase. To understand this phase we need to
remember that each S3 has a cylindrical zero mode direction along T S1 . In the four
dimensions of spacetime S1 R T S1 remaining after reduction on S3 , the RR scalar, the
two B fields, and the dilaton are on an equal footing since scalars and two-forms are dual.
Again, one would expect these modes to be nondynamical in four dimensions. The same
argument about supersymmetry breaking in the mixed resolution applies here. We should
obtain no chiral fermions in this phase. Compactifying the theory on T S1 would yield a
trivial two-dimensional theory at low energies with no conformal anomaly. We summarize
538
this below:
NLB = 0,
NRB = 0,
NLF = 0,
NRF = 0,
cL = 0,
cR = 0.
(4.6)
As in the mixed resolution the puzzle that we appear to have a family of two-dimensional
vacua parametrized by and is resolved by realizing that the effective theory is
really four-dimensional here. There are two massive supermultiplets from threebranes
wrapping the independent S3 s. The zero modes are the global U (1) and T S1 S1 . These
supermultiplets become massless at the singularity.
The theories we have found are not strictly two-dimensional because of the bulk
counterterms needed for anomaly cancellation and the zero mode directions. The mixed
and large resolutions are effectively four-dimensional and nonchiral. Any dynamical
transition from the small resolution into the other phases seems unlikely. Wrapping a
large number of fivebranes on P1 P1 would yield a large N two-dimensional gauge
theory without dynamics. On the other hand the mixed and large resolutions are analogous
to the S2 and S3 resolutions of the conifold. It is natural to conjecture that there is a
duality relating fivebranes wrapped on S2 T S1 S1 R in the mixed resolution with
threeform flux on one of the S3 s in the large resolution since the effective theories are
four-dimensional. By taking the radius of the other S3 to infinity, one retrieves the two
resolutions of the conifold. It might be interesting to explore further the equations for the
fivebrane following from local mirror symmetry to have a purely geometric understanding
of the fivebrane manifold in the mixed and large phases. Determining whether there is a
compact manifold with these transitions would be interesting. Also, can one understand a
triple intersection of fivebranes in type IIB similarly as giving an effective two-dimensional
theory of two (2, 2) supersymmetric vector multiplets in type IIA? The triple intersection
theory is related to string theory black holes. In type IIA the number of intersecting
fivebranes is limited by the conical deficit angle, but the number in eleven-dimensional
supergravity is unlimited. The various phases should still be present as the eleventh
dimension opens up. There are many additional questions including nonperturbative terms
in these theories left for further work.
Acknowledgements
I wish to thank J. Distler, W. Fischler, T. Husain, and A. Iqbal for many helpful
discussions. This material is based upon work supported in part by the National Science
Foundation under Grant No. 0071512.
References
[1] P. Candelas, X. de la Ossa, Nucl. Phys. B 341 (1990) 383.
[2] I. Klebanov, M. Strassler, JHEP 0008 (2000) 052, hep-th/0007191;
C. Vafa, J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
M. Bershadsky, V. Sadov, C. Vafa, Nucl. Phys. B 463 (1996) 420, hep-th/9511222.

J. Blum, JHEP 0001 (2000) 006, hep-th/9907101.
B. Greene, A. Shapere, C. Vafa, S.-T. Yau, Nucl. Phys. B 337 (1990) 1.
G. Papadopoulos, P. Townsend, Phys. Lett. B 380 (1996) 273, hep-th/9603087.
A. Tseytlin, Phys. Lett. B 381 (1996) 73, hep-th/9604035.
A. Fayyazuddin, D. Smith, JHEP 9904 (1999) 030, hep-th/9902210.
H. Cho, M. Emam, D. Kastor, J. Traschen, Phys. Rev. D 63 (2001) 064003, hep-th/0009062.
P. Green, T. Hbsch, Int. J. Mod. Phys. A 9 (1994) 3203, hep-th/9303158.
G. Cardoso, B. de Wit, T. Mohaupt, Class. Quantum Grav. 17 (2000) 1007, hep-th/9910179.
K. Dasgupta, S. Mukhi, Phys. Lett. B 398 (1997) 285, hep-th/9612188.
S. Katz, P. Mayr, C. Vafa, Adv. Theor. Math. Phys. 1 (1998) 53, hep-th/9706110;
K. Hori, C. Vafa, hep-th/0002222;
K. Hori, A. Iqbal, C. Vafa, hep-th/0005247.
[14] A. Strominger, Nucl. Phys. B 451 (1995) 96, hep-th/9504090.
539

CUMULATIVE AUTHOR INDEX B641B646
Abdalla, E.
Agashe, K.
Ahn, C.
Akhmedov, E.Kh.
Alekseev, A.Yu.
Amico, L.
Anastasiou, C.
Andersson, B.
Arias, O.
Arutyunov, G.
Asano, M.
B644 (2002) 201

B643 (2002) 172
B646 (2002) 257
B643 (2002) 339
B646 (2002) 127
B644 (2002) 409
B646 (2002) 220
B646 (2002) 102
B643 (2002) 187
B643 (2002) 49
B644 (2002) 151
Babichev, E.
Bagnoud, M.
Bajnok, Z.
Bastianelli, F.
Beneke, M.
Berglund, P.
Bianchi, M.
Bieliavsky, P.
Bietenholz, W.
Bilal, A.
Blanchard, P.
Blum, J.D.
Blumenhagen, R.
Bolognesi, S.
Bouttier, J.
Bouttier, J.
Brandhuber, A.
Brignole, A.
Brodsky, S.J.
Bruckmann, F.
Buchmller, W.
Burguet-Castell, J.
B645 (2002) 134

B641 (2002) 61
B644 (2002) 509
B642 (2002) 372
B643 (2002) 431
B641 (2002) 351
B646 (2002) 69
B645 (2002) 349
B644 (2002) 223
B641 (2002) 61
B644 (2002) 495
B646 (2002) 524
B641 (2002) 235
B645 (2002) 337
B641 (2002) 519
B645 (2002) 477
B641 (2002) 351
B643 (2002) 79
B642 (2002) 344
B645 (2002) 105
B643 (2002) 367
B646 (2002) 301
Cao, J.-P.
Cardenas, R.
Carlevaro, L.
Casali, A.
B644 (2002) 476

B643 (2002) 187
B641 (2002) 61
B644 (2002) 201
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 9 5 9 - 8
Chapovsky, A.P.
Choudhury, D.
Corley, S.
Cremades, D.
Cristofano, G.
Cuadros-Melgar, B.
Curio, G.
Cvetic, M.
Cvetic, M.
Czakon, M.
B643 (2002) 431

B646 (2002) 3
B641 (2002) 131
B643 (2002) 93
B641 (2002) 547
B644 (2002) 201
B643 (2002) 131
B642 (2002) 139
B644 (2002) 65
B642 (2002) 157
Dabholkar, A.
DAmbrosio, G.
Das, P.
De Forcrand, P.
Degrassi, G.
de la Ossa, X.
Delgado, A.
Derkachov, S..
Dermek, R.
Di Bari, P.
Di Clemente, V.
Diehl, M.
Di Francesco, P.
Di Francesco, P.
Dijkgraaf, R.
Dijkgraaf, R.
Di Lorenzo, A.
Dittmaier, S.
Di Vecchia, P.
Dokuchaev, V.
Dolgushev, V.A.
Donini, A.
Doyon, B.
B641 (2002) 223

B645 (2002) 155
B644 (2002) 395
B642 (2002) 290
B643 (2002) 79
B644 (2002) 170
B643 (2002) 172
B645 (2002) 237
B641 (2002) 327
B643 (2002) 367
B646 (2002) 24
B643 (2002) 431
B641 (2002) 519
B645 (2002) 477
B644 (2002) 3
B644 (2002) 21
B644 (2002) 409
B642 (2002) 307
B646 (2002) 43
B645 (2002) 134
B645 (2002) 457
B646 (2002) 321
B644 (2002) 451
Eden, B.
Ellis, J.
B646 (2002) 69
B643 (2002) 229
Faraggi, A.E.
Faraggi, A.E.
B641 (2002) 93
B641 (2002) 111
542
Feldmann, Th.
Fjelstad, J.
Florea, B.
Foerster, A.
Forte, S.
Fortunato, S.
Fradkin, E.
Fredenhagen, S.
Fuchs, J.
B643 (2002) 431

B641 (2002) 376
B644 (2002) 170
B642 (2002) 501
B643 (2002) 477
B644 (2002) 495
B642 (2002) 483
B646 (2002) 127
B646 (2002) 353
Gaberdiel, M.R.
Gaillard, M.K.
Gandolfo, D.
Garavuso, R.
Garland, L.W.
Gavela, M.B.
Gehrmann, T.
Ghilencea, D.M.
Ghosh, D.K.
Giaquinta, G.
Giedt, J.
Gies, H.
Giudice, G.F.
Glover, E.W.N.
Gluza, J.
Gmez-Cadenas, J.J.
Gonzlez, J.
Govindarajan, S.
Gracey, J.A.
Graham, N.
Groot Nibbelink, S.
Guan, X.-W.
Guitter, E.
Guitter, E.
B645 (2002) 403

B643 (2002) 201
B644 (2002) 495
B641 (2002) 111
B642 (2002) 227
B646 (2002) 301
B642 (2002) 227
B641 (2002) 35
B646 (2002) 3
B644 (2002) 409
B643 (2002) 201
B646 (2002) 158
B645 (2002) 155
B642 (2002) 227
B642 (2002) 157
B646 (2002) 301
B642 (2002) 407
B646 (2002) 498
B644 (2002) 433
B645 (2002) 49
B641 (2002) 35
B642 (2002) 501
B641 (2002) 519
B645 (2002) 477
Hall, L.J.
Hamberg, R.
Hands, S.J.
Hasenfratz, P.
Hatsuda, M.
Hauswirth, S.
Hejczyk, J.
Henkel, M.
Hernndez, P.
Higashijima, K.
Hikami, K.
Hoffmann, L.
Holland, K.
Huber, P.
Hwang, D.S.
Hwang, S.
B645 (2002) 85
B644 (2002) 403
B645 (2002) 321
B643 (2002) 280
B644 (2002) 40
B643 (2002) 280
B642 (2002) 157
B641 (2002) 405
B646 (2002) 301
B645 (2002) 438
B644 (2002) 409
B641 (2002) 188
B643 (2002) 280
B645 (2002) 3
B642 (2002) 344
B641 (2002) 376
Ibez, L.E.
Igarashi, H.
B643 (2002) 93
B644 (2002) 383
Isaev, A.P.
Isidori, G.
Isidro, J.M.
Itoh, T.
Itoyama, H.
B645 (2002) 457

B645 (2002) 155
B641 (2002) 111
B646 (2002) 257
B644 (2002) 248
Jaffe, R.L.
Jahn, O.
Jakbek, A.
Jansen, K.
Janssen, B.
Jayaraman, T.
Jegerlehner, F.
Jejjala, V.
Jrg, T.
B645 (2002) 49
B642 (2002) 357
B645 (2002) 217
B643 (2002) 517
B643 (2002) 399
B646 (2002) 498
B641 (2002) 285
B642 (2002) 483
B643 (2002) 280
Kalmykov, M.Yu.
Kamimura, K.
Karabali, D.
Kawamoto, N.
Kerler, W.
Khemani, V.
Khveshchenko, D.V.
Kiem, Y.
Kiem, Y.
Kim, N.
Kim, S.-S.
Kim, Y.
Kimura, T.
King, S.F.
Kitanine, N.
Kitanine, N.
Kitazawa, Y.
Kogut, J.B.
Kogut, J.B.
Konishi, K.
Korchemsky, G.P.
Krs, B.
Kotanski, J.
Koukoutsakis, A.
Krause, A.
Kristjansen, C.
B641 (2002) 285

B644 (2002) 40
B641 (2002) 533
B644 (2002) 533
B646 (2002) 201
B645 (2002) 49
B642 (2002) 515
B641 (2002) 256
B642 (2002) 389
B643 (2002) 31
B641 (2002) 256
B642 (2002) 389
B645 (2002) 438
B646 (2002) 24
B641 (2002) 487
B642 (2002) 433
B642 (2002) 210
B642 (2002) 181
B645 (2002) 321
B645 (2002) 337
B645 (2002) 237
B641 (2002) 235
B645 (2002) 237
B642 (2002) 227
B643 (2002) 131
B643 (2002) 3
Langacker, P.
Langfeld, K.
Latorre, J.J.
Lee, S.
Leigh, R.G.
Lerda, A.
Levman, G.
Lima-Santos, A.
Lindner, M.
Links, J.
Lozano, Y.
B642 (2002) 139

B646 (2002) 158
B643 (2002) 477
B642 (2002) 389
B642 (2002) 483
B646 (2002) 43
B642 (2002) 3
B644 (2002) 568
B645 (2002) 3
B642 (2002) 501
B643 (2002) 399
L, H.
L, H.
Lukyanov, S.
Lunardini, C.
Lunin, O.
Lst, D.
Lyakhovich, S.L.
B642 (2002) 173

B644 (2002) 65
B644 (2002) 451
B643 (2002) 339
B642 (2002) 91
B641 (2002) 235
B645 (2002) 457
Ma, E.
Magnea, L.
Mahanta, U.
Maiella, G.
Maillet, J.M.
Maillet, J.M.
Maltoni, M.
Manashov, A.N.
Mnsson, T.
Marchesano, F.
Marotta, V.
Mathur, S.D.
Matsuura, T.
Maxwell, C.J.
Meissner, K.A.
Melnikov, K.
Meloni, D.
Mena, O.
Merlatti, P.
Mesref, L.
Meziane, A.
Migliozzi, P.
Mirjalili, A.
Moch, S.
Mohanty, S.
Moyaerts, L.
Mueller, A.H.
Murayama, H.
B644 (2002) 290

B643 (2002) 477
B644 (2002) 395
B641 (2002) 547
B641 (2002) 487
B642 (2002) 433
B643 (2002) 321
B645 (2002) 237
B641 (2002) 376
B643 (2002) 93
B641 (2002) 547
B642 (2002) 91
B644 (2002) 403
B645 (2002) 298
B645 (2002) 217
B646 (2002) 220
B646 (2002) 321
B646 (2002) 301
B646 (2002) 43
B641 (2002) 188
B641 (2002) 188
B646 (2002) 321
B645 (2002) 298
B646 (2002) 181
B646 (2002) 102
B646 (2002) 158
B643 (2002) 501
B645 (2002) 85
Nair, V.P.
Nakamura, S.
Nakatsu, T.
Nandi, S.
Niccoli, G.
Niedermayer, F.
Nitta, M.
Nomura, Y.
Notari, A.
B641 (2002) 533

B644 (2002) 248
B642 (2002) 13
B641 (2002) 327
B641 (2002) 547
B643 (2002) 280
B645 (2002) 438
B645 (2002) 85
B644 (2002) 371
Ohlsson, T.
Okuda, T.
Okuyama, K.
Olechowski, M.
Olive, D.I.
Olum, K.D.
Ooguri, H.
B643 (2002) 247

B641 (2002) 393
B644 (2002) 383
B645 (2002) 217
B645 (2002) 403
B645 (2002) 367
B641 (2002) 3
543
Osterloh, A.
Ott, T.
B644 (2002) 409

B641 (2002) 235
Palla, L.
Palumbo, F.
Palumbo, F.
Park, J.
Parvizi, S.
Pawlowski, J.M.
Penati, S.
Petkou, A.C.
Philipsen, O.
Piccione, A.
Pilaftsis, A.
Plefka, J.
Plefka, J.
Plmacher, M.
Ponsot, B.
Pope, C.N.
B644 (2002) 509

B643 (2002) 391
B645 (2002) 309
B642 (2002) 389
B641 (2002) 223
B642 (2002) 357
B643 (2002) 49
B643 (2002) 49
B642 (2002) 290
B643 (2002) 477
B644 (2002) 263
B643 (2002) 3
B643 (2002) 31
B643 (2002) 367
B642 (2002) 114
B644 (2002) 65
Quandt, M.
Quella, T.
Quiros, I.
B645 (2002) 49
B646 (2002) 127
B643 (2002) 187
Raby, S.
Raidal, M.
Ramgoolam, S.
Randjbar-Daemi, S.
Rayner, D.A.J.
Remiddi, E.
Rey, S.-J.
Riotto, A.
Rooman, M.
Rossi, G.
Roth, M.
Roy, D.P.
Roy, S.
Rhl, W.
Runkel, I.
B641 (2002) 327

B643 (2002) 229
B641 (2002) 131
B645 (2002) 188
B646 (2002) 24
B642 (2002) 227
B641 (2002) 256
B644 (2002) 371
B645 (2002) 349
B646 (2002) 69
B642 (2002) 307
B644 (2002) 290
B646 (2002) 3
B641 (2002) 188
B646 (2002) 353
Sakaguchi, M.
Santambrogio, A.
Sarkar, T.
Sato, H.-T.
Scandurra, M.
Schmidt, I.
Schomerus, V.
Schweigert, C.
Schwetz, T.
Seidl, G.
Sekino, Y.
Semenoff, G.W.
Sezgin, E.
Shaposhnikov, M.
B644 (2002) 40
B643 (2002) 49
B646 (2002) 498
B641 (2002) 256
B645 (2002) 49
B642 (2002) 344
B646 (2002) 127
B646 (2002) 353
B643 (2002) 321
B643 (2002) 247
B644 (2002) 151
B643 (2002) 3
B644 (2002) 303
B645 (2002) 188
544
Sharapov, A.A.
Shiu, G.
Shore, G.M.
Siemens, X.
Sinclair, D.K.
Skarke, H.
Slavich, P.
Slavnov, N.A.
Slavnov, N.A.
Smirnov, A.Yu.
Sderberg, F.
Sokatchev, E.
Sommer, R.
Spindel, Ph.
Stanev, Y.S.
Staudacher, M.
Strouthos, C.G.
Strumia, A.
Sugiyama, K.
Sugiyama, K.
Sundell, P.
Sundrum, R.
Suyama, T.
Suzuki, H.
B645 (2002) 457

B642 (2002) 139
B646 (2002) 281
B645 (2002) 367
B642 (2002) 181
B644 (2002) 170
B643 (2002) 79
B641 (2002) 487
B642 (2002) 433
B643 (2002) 339
B646 (2002) 102
B643 (2002) 49
B643 (2002) 517
B645 (2002) 349
B646 (2002) 69
B643 (2002) 3
B645 (2002) 321
B645 (2002) 155
B644 (2002) 113
B644 (2002) 128
B644 (2002) 303
B643 (2002) 172
B641 (2002) 341
B644 (2002) 383
Takcs, G.
Takcs, G.
Terras, V.
Terras, V.
Trtola, M.A.
Toublan, D.
B642 (2002) 456

B644 (2002) 509
B641 (2002) 487
B642 (2002) 433
B643 (2002) 321
B642 (2002) 181
Vafa, C.
Vafa, C.
Vafa, C.
Valle, J.W.F.
van Baal, P.
van Neerven, W.L.
van Tonder, A.
van Tonder, A.
Veretin, O.
Vermaseren, J.A.M.
Vives, O.
Vogt, A.
B641 (2002) 3
B644 (2002) 3
B644 (2002) 21
B643 (2002) 321
B645 (2002) 105
B644 (2002) 403
B645 (2002) 371
B645 (2002) 387
B641 (2002) 285
B646 (2002) 181
B641 (2002) 93
B646 (2002) 181
Wang, Y.-P.
Watts, G.M.T.
Wei, Z.-T.
Weigel, H.
West, P.C.
Winter, W.
Wosiek, J.
B644 (2002) 476

B642 (2002) 456
B642 (2002) 263
B645 (2002) 49
B645 (2002) 403
B645 (2002) 3
B644 (2002) 85
Yang, M.-Z.
Yoshida, K.
Yoshida, K.
Yotsuji, K.
Yue, R.-H.
B642 (2002) 263

B644 (2002) 113
B644 (2002) 128
B644 (2002) 533
B644 (2002) 476
Zarembo, K.
Zhou, H.-Q.
Zirotti, A.
Zupnik, B.
Zwirner, F.
B643 (2002) 157

B642 (2002) 501
B642 (2002) 372
B644 (2002) 405
B643 (2002) 79

Nucl - Phys.B v.646

Uploaded by

Copyright:

Available Formats

Nucl - Phys.B v.646

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.646

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 646 (2002) 323

Signals of anomaly mediated supersymmetry

Received 10 September 2002; accepted 8 October 2002

E-mail addresses: debchou@mri.ernet.in (D. Choudhury), dilip.ghosh@cern.ch (D.K. Ghosh),

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

3. Model parameters and constraints

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

scale variable), respectively, we have, for the gaugino () masses

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

4. Signal and backgrounds

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

Fig. 1. Schematic diagram for the signal process e e 1 e + + + p

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

B (450, 55.0, 30)

4.3. Signal strength and the parameter space

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

is the corresponding momentum.

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79.

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

E. Katz, Y. Shadmi, Y. Shirman, J. High Energy Phys. 08 (1999) 015.

D. Choudhury et al. / Nuclear Physics B 646 (2002) 323

Nuclear Physics B 646 (2002) 2442

Supersymmetric Higgs bosons in a 5D

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

warped extra dimensions [11] and a (de)constructed model [12].

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

3 ), MSSM gauge sector (

scenario [14], but

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

2.1. Top/stop KK mass spectra

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

Minimal mixing (t  1) is equivalent to a very small extra dimension (MC

In the limit that electroweak symmetry remains unbroken (i.e.,

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

We use dimensional regularization and zeta-function regularization techniques [18] to

9MC4 cos[2nmt [hu ]/MC ] 135 [5] 4

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

2.3. Reliability and perturbativity

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

m2H = m2A0 + m2W .

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

changes in MC ) to find an upper limit on the compactification scale MC

but these provide a much weaker constraint on our model.

V. Di Clemente et al. / Nuclear Physics B 646 (2002) 2442

Minimal mixing (t 1) is equivalent to a very small extra dimension (MC

H (3) = dA(2) + !ij k F ij Ak Aij Akm Am .

with 0 , 0 2 and 0 4 . The corresponding left-invariant 1-forms